diff options
94 files changed, 17375 insertions, 512 deletions
diff --git a/src/Makefile.am b/src/Makefile.am index 08feb29ae40..641707ed122 100644 --- a/src/Makefile.am +++ b/src/Makefile.am @@ -88,6 +88,7 @@ include vlib-api.am include vnet.am include vpp.am include vpp-api-test.am +include uri.am SUBDIRS += plugins diff --git a/src/plugins/ioam/export-common/ioam_export.h b/src/plugins/ioam/export-common/ioam_export.h index e84dab0b398..dd48a93b36b 100644 --- a/src/plugins/ioam/export-common/ioam_export.h +++ b/src/plugins/ioam/export-common/ioam_export.h @@ -21,7 +21,7 @@ #include <vnet/ip/ip4_packet.h> #include <vnet/ip/ip6_packet.h> #include <vnet/ip/ip6_hop_by_hop.h> -#include <vnet/ip/udp.h> +#include <vnet/udp/udp.h> #include <vnet/flow/ipfix_packet.h> #include <vppinfra/pool.h> diff --git a/src/plugins/ioam/ipfixcollector/ipfixcollector.c b/src/plugins/ioam/ipfixcollector/ipfixcollector.c index 4ae47edcb3c..71b934ecd93 100644 --- a/src/plugins/ioam/ipfixcollector/ipfixcollector.c +++ b/src/plugins/ioam/ipfixcollector/ipfixcollector.c @@ -15,7 +15,7 @@ #include <vnet/ip/ip.h> #include <vnet/plugin/plugin.h> -#include <vnet/ip/udp.h> +#include <vnet/udp/udp.h> #include <ioam/ipfixcollector/ipfixcollector.h> ipfix_collector_main_t ipfix_collector_main; diff --git a/src/plugins/ioam/lib-vxlan-gpe/ioam_transit.c b/src/plugins/ioam/lib-vxlan-gpe/ioam_transit.c index b42c357c79b..f334c98321a 100644 --- a/src/plugins/ioam/lib-vxlan-gpe/ioam_transit.c +++ b/src/plugins/ioam/lib-vxlan-gpe/ioam_transit.c @@ -16,7 +16,7 @@ #include <vppinfra/hash.h> #include <vnet/vnet.h> #include <vnet/ip/ip.h> -#include <vnet/ip/udp.h> +#include <vnet/udp/udp.h> #include <vnet/ethernet/ethernet.h> #include <vnet/vxlan-gpe/vxlan_gpe.h> #include <ioam/lib-vxlan-gpe/vxlan_gpe_ioam_packet.h> diff --git a/src/plugins/snat/in2out.c b/src/plugins/snat/in2out.c index e30c913c02d..b4b7793dae0 100644 --- a/src/plugins/snat/in2out.c +++ b/src/plugins/snat/in2out.c @@ -689,12 +689,12 @@ snat_hairpinning (snat_main_t *sm, ip4_header_t, dst_address); ip0->checksum = ip_csum_fold (sum0); - old_dst_port0 = tcp0->ports.dst; + old_dst_port0 = tcp0->dst; if (PREDICT_TRUE(new_dst_port0 != old_dst_port0)) { if (PREDICT_TRUE(proto0 == SNAT_PROTOCOL_TCP)) { - tcp0->ports.dst = new_dst_port0; + tcp0->dst = new_dst_port0; sum0 = tcp0->checksum; sum0 = ip_csum_update (sum0, old_dst_addr0, new_dst_addr0, ip4_header_t, dst_address); @@ -872,9 +872,9 @@ snat_in2out_node_fn_inline (vlib_main_t * vm, if (PREDICT_TRUE(proto0 == SNAT_PROTOCOL_TCP)) { - old_port0 = tcp0->ports.src; - tcp0->ports.src = s0->out2in.port; - new_port0 = tcp0->ports.src; + old_port0 = tcp0->src_port; + tcp0->src_port = s0->out2in.port; + new_port0 = tcp0->src_port; sum0 = tcp0->checksum; sum0 = ip_csum_update (sum0, old_addr0, new_addr0, @@ -1012,9 +1012,9 @@ snat_in2out_node_fn_inline (vlib_main_t * vm, if (PREDICT_TRUE(proto1 == SNAT_PROTOCOL_TCP)) { - old_port1 = tcp1->ports.src; - tcp1->ports.src = s1->out2in.port; - new_port1 = tcp1->ports.src; + old_port1 = tcp1->src_port; + tcp1->src_port = s1->out2in.port; + new_port1 = tcp1->src_port; sum1 = tcp1->checksum; sum1 = ip_csum_update (sum1, old_addr1, new_addr1, @@ -1188,9 +1188,9 @@ snat_in2out_node_fn_inline (vlib_main_t * vm, if (PREDICT_TRUE(proto0 == SNAT_PROTOCOL_TCP)) { - old_port0 = tcp0->ports.src; - tcp0->ports.src = s0->out2in.port; - new_port0 = tcp0->ports.src; + old_port0 = tcp0->src_port; + tcp0->src_port = s0->out2in.port; + new_port0 = tcp0->src_port; sum0 = tcp0->checksum; sum0 = ip_csum_update (sum0, old_addr0, new_addr0, @@ -1667,8 +1667,8 @@ snat_in2out_fast_static_map_fn (vlib_main_t * vm, { if (PREDICT_TRUE(proto0 == SNAT_PROTOCOL_TCP)) { - old_port0 = tcp0->ports.src; - tcp0->ports.src = new_port0; + old_port0 = tcp0->src_port; + tcp0->src_port = new_port0; sum0 = tcp0->checksum; sum0 = ip_csum_update (sum0, old_addr0, new_addr0, diff --git a/src/plugins/snat/out2in.c b/src/plugins/snat/out2in.c index 328f5ba4c5e..3bfc0aa31e0 100644 --- a/src/plugins/snat/out2in.c +++ b/src/plugins/snat/out2in.c @@ -19,7 +19,7 @@ #include <vnet/handoff.h> #include <vnet/ip/ip.h> -#include <vnet/ip/udp.h> +#include <vnet/udp/udp.h> #include <vnet/ethernet/ethernet.h> #include <vnet/fib/ip4_fib.h> #include <snat/snat.h> @@ -602,9 +602,9 @@ snat_out2in_node_fn (vlib_main_t * vm, if (PREDICT_TRUE(proto0 == SNAT_PROTOCOL_TCP)) { - old_port0 = tcp0->ports.dst; - tcp0->ports.dst = s0->in2out.port; - new_port0 = tcp0->ports.dst; + old_port0 = tcp0->dst_port; + tcp0->dst_port = s0->in2out.port; + new_port0 = tcp0->dst_port; sum0 = tcp0->checksum; sum0 = ip_csum_update (sum0, old_addr0, new_addr0, @@ -737,9 +737,9 @@ snat_out2in_node_fn (vlib_main_t * vm, if (PREDICT_TRUE(proto1 == SNAT_PROTOCOL_TCP)) { - old_port1 = tcp1->ports.dst; - tcp1->ports.dst = s1->in2out.port; - new_port1 = tcp1->ports.dst; + old_port1 = tcp1->dst_port; + tcp1->dst_port = s1->in2out.port; + new_port1 = tcp1->dst_port; sum1 = tcp1->checksum; sum1 = ip_csum_update (sum1, old_addr1, new_addr1, @@ -907,9 +907,9 @@ snat_out2in_node_fn (vlib_main_t * vm, if (PREDICT_TRUE(proto0 == SNAT_PROTOCOL_TCP)) { - old_port0 = tcp0->ports.dst; - tcp0->ports.dst = s0->in2out.port; - new_port0 = tcp0->ports.dst; + old_port0 = tcp0->dst_port; + tcp0->dst_port = s0->in2out.port; + new_port0 = tcp0->dst_port; sum0 = tcp0->checksum; sum0 = ip_csum_update (sum0, old_addr0, new_addr0, @@ -1369,8 +1369,8 @@ snat_out2in_fast_node_fn (vlib_main_t * vm, { if (PREDICT_TRUE(proto0 == SNAT_PROTOCOL_TCP)) { - old_port0 = tcp0->ports.dst; - tcp0->ports.dst = new_port0; + old_port0 = tcp0->dst_port; + tcp0->dst_port = new_port0; sum0 = tcp0->checksum; sum0 = ip_csum_update (sum0, old_addr0, new_addr0, diff --git a/src/scripts/vnet/tcp b/src/scripts/vnet/tcp index a2ee8b2dacc..b9c23c3a019 100644 --- a/src/scripts/vnet/tcp +++ b/src/scripts/vnet/tcp @@ -1,16 +1,18 @@ +loop create +set int ip address loop0 192.168.1.1/8 +set int state loop0 up + packet-generator new { name x - limit 1 + limit 2048 node ip4-input - size 64-64 + size 100-100 + interface loop0 no-recycle data { - TCP: 1.2.3.4 -> 5.6.7.8 - TCP: 1234 -> 5678 + TCP: 192.168.1.2 -> 192.168.1.1 + TCP: 32415 -> 80 + SYN incrementing 100 } } - -tr add pg-input 100 -ip route 5.6.7.8/32 via local -ip route 1.2.3.4/32 via local diff --git a/src/scripts/vnet/udp b/src/scripts/vnet/udp new file mode 100644 index 00000000000..7dda1eec511 --- /dev/null +++ b/src/scripts/vnet/udp @@ -0,0 +1,19 @@ +loop create +set int ip address loop0 192.168.1.1/8 +set int state loop0 up + +packet-generator new { + name udp + limit 512 + rate 1e4 + node ip4-input + size 100-100 + interface loop0 + no-recycle + data { + UDP: 192.168.1.2 - 192.168.2.255 -> 192.168.1.1 + UDP: 4321 -> 1234 + length 72 + incrementing 100 + } +} diff --git a/src/scripts/vnet/uri/tcp-setup.sh b/src/scripts/vnet/uri/tcp-setup.sh new file mode 100755 index 00000000000..e0b01588fe9 --- /dev/null +++ b/src/scripts/vnet/uri/tcp-setup.sh @@ -0,0 +1,39 @@ +#!/usr/bin/env bash + +function topo_setup +{ + ip netns add vppns1 + ip link add veth_vpp1 type veth peer name vpp1 + ip link set dev vpp1 up + ip link set dev veth_vpp1 up netns vppns1 + + ip netns exec vppns1 \ + bash -c " + ip link set dev lo up + ip addr add 6.0.1.2/24 dev veth_vpp1 + " + + ethtool --offload vpp1 rx off tx off + ip netns exec vppns1 ethtool --offload veth_vpp1 rx off tx off + +} + +function topo_clean +{ + ip link del dev veth_vpp1 &> /dev/null + ip netns del vppns1 &> /dev/null +} + +if [ "$1" == "clean" ] ; then + topo_clean + exit 0 +else + topo_setup +fi + +# to test connectivity do: +# sudo ip netns exec vppns1 telnet 6.0.1.1 1234 +# to push traffic to the server +# dd if=/dev/zero bs=1024K count=512 | nc 6.0.1.1 +# to listen for incoming connection from vpp +# nc -l 1234 diff --git a/src/scripts/vnet/uri/tcp_server b/src/scripts/vnet/uri/tcp_server new file mode 100644 index 00000000000..7f5a86defad --- /dev/null +++ b/src/scripts/vnet/uri/tcp_server @@ -0,0 +1,4 @@ +create host-interface name vpp1 +set int state host-vpp1 up +set int ip address host-vpp1 6.0.1.1/24 +trace add af-packet-input 10 diff --git a/src/scripts/vnet/uri/udp b/src/scripts/vnet/uri/udp new file mode 100644 index 00000000000..ca13b83cd3a --- /dev/null +++ b/src/scripts/vnet/uri/udp @@ -0,0 +1,19 @@ +loop create +set int ip address loop0 10.0.0.1/32 +set int state loop0 up + +packet-generator new { + name udp + limit 512 + rate 1e4 + node ip4-input + size 100-100 + interface loop0 + no-recycle + data { + UDP: 192.168.1.2 - 192.168.2.255 -> 192.168.1.1 + UDP: 4321 -> 1234 + length 72 + incrementing 100 + } +} diff --git a/src/svm.am b/src/svm.am index 2cd385bd037..442eba8e6a9 100644 --- a/src/svm.am +++ b/src/svm.am @@ -13,13 +13,14 @@ bin_PROGRAMS += svmtool svmdbtool -nobase_include_HEADERS += svm/svm.h svm/ssvm.h svm/svmdb.h +nobase_include_HEADERS += svm/svm.h svm/ssvm.h svm/svmdb.h \ + svm/svm_fifo.h svm/svm_fifo_segment.h lib_LTLIBRARIES += libsvm.la libsvmdb.la +libsvm_la_SOURCES = svm/svm.c svm/ssvm.c svm/svm_fifo.c svm/svm_fifo_segment.c libsvm_la_LIBADD = libvppinfra.la -lrt -lpthread libsvm_la_DEPENDENCIES = libvppinfra.la -libsvm_la_SOURCES = svm/svm.c svm/ssvm.c svmtool_SOURCES = svm/svmtool.c svmtool_LDADD = libsvm.la libvppinfra.la -lpthread -lrt @@ -31,4 +32,9 @@ libsvmdb_la_SOURCES = svm/svmdb.c svmdbtool_SOURCES = svm/svmdbtool.c svmdbtool_LDADD = libsvmdb.la libsvm.la libvppinfra.la -lpthread -lrt +noinst_PROGRAMS += test_svm_fifo1 +test_svm_fifo1_SOURCES = svm/test_svm_fifo1.c +test_svm_fifo1_LDADD = libsvm.la libvppinfra.la -lpthread -lrt +test_svm_fifo1_LDFLAGS = -static + # vi:syntax=automake diff --git a/src/svm/ssvm.c b/src/svm/ssvm.c index 6f409eb68b6..6cda1f279e8 100644 --- a/src/svm/ssvm.c +++ b/src/svm/ssvm.c @@ -169,6 +169,22 @@ re_map_it: return 0; } +void +ssvm_delete (ssvm_private_t * ssvm) +{ + u8 *fn; + + fn = format (0, "/dev/shm/%s%c", ssvm->name, 0); + + /* Throw away the backing file */ + if (unlink ((char *) fn) < 0) + clib_unix_warning ("unlink segment '%s'", ssvm->name); + + munmap ((void *) ssvm->requested_va, ssvm->ssvm_size); + vec_free (fn); +} + + /* * fd.io coding-style-patch-verification: ON * diff --git a/src/svm/ssvm.h b/src/svm/ssvm.h index 9e61b9a0827..bccfc164bef 100644 --- a/src/svm/ssvm.h +++ b/src/svm/ssvm.h @@ -38,7 +38,10 @@ #include <vppinfra/pool.h> #include <vppinfra/format.h> -#define MMAP_PAGESIZE (4<<10) +#ifndef MMAP_PAGESIZE +#define MMAP_PAGESIZE (clib_mem_get_page_size()) +#endif + #define SSVM_N_OPAQUE 7 typedef struct @@ -125,12 +128,12 @@ ssvm_pop_heap (void *oldheap) } #define foreach_ssvm_api_error \ -_(NO_NAME, "No shared segment name", -10) \ -_(NO_SIZE, "Size not set (master)", -11) \ -_(CREATE_FAILURE, "Create failed", -12) \ -_(SET_SIZE, "Set size failed", -13) \ -_(MMAP, "mmap failed", -14) \ -_(SLAVE_TIMEOUT, "Slave map timeout", -15) +_(NO_NAME, "No shared segment name", -100) \ +_(NO_SIZE, "Size not set (master)", -101) \ +_(CREATE_FAILURE, "Create failed", -102) \ +_(SET_SIZE, "Set size failed", -103) \ +_(MMAP, "mmap failed", -104) \ +_(SLAVE_TIMEOUT, "Slave map timeout", -105) typedef enum { @@ -143,6 +146,7 @@ typedef enum int ssvm_master_init (ssvm_private_t * ssvm, u32 master_index); int ssvm_slave_init (ssvm_private_t * ssvm, int timeout_in_seconds); +void ssvm_delete (ssvm_private_t * ssvm); #endif /* __included_ssvm_h__ */ diff --git a/src/svm/svm_fifo.c b/src/svm/svm_fifo.c new file mode 100644 index 00000000000..11f901931be --- /dev/null +++ b/src/svm/svm_fifo.c @@ -0,0 +1,568 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "svm_fifo.h" + +/** create an svm fifo, in the current heap. Fails vs blow up the process */ +svm_fifo_t * +svm_fifo_create (u32 data_size_in_bytes) +{ + svm_fifo_t *f; + pthread_mutexattr_t attr; + pthread_condattr_t cattr; + + f = clib_mem_alloc_aligned_or_null (sizeof (*f) + data_size_in_bytes, + CLIB_CACHE_LINE_BYTES); + if (f == 0) + return 0; + + memset (f, 0, sizeof (*f) + data_size_in_bytes); + f->nitems = data_size_in_bytes; + f->ooos_list_head = OOO_SEGMENT_INVALID_INDEX; + + memset (&attr, 0, sizeof (attr)); + memset (&cattr, 0, sizeof (cattr)); + + if (pthread_mutexattr_init (&attr)) + clib_unix_warning ("mutexattr_init"); + if (pthread_mutexattr_setpshared (&attr, PTHREAD_PROCESS_SHARED)) + clib_unix_warning ("pthread_mutexattr_setpshared"); + if (pthread_mutex_init (&f->mutex, &attr)) + clib_unix_warning ("mutex_init"); + if (pthread_mutexattr_destroy (&attr)) + clib_unix_warning ("mutexattr_destroy"); + if (pthread_condattr_init (&cattr)) + clib_unix_warning ("condattr_init"); + if (pthread_condattr_setpshared (&cattr, PTHREAD_PROCESS_SHARED)) + clib_unix_warning ("condattr_setpshared"); + if (pthread_cond_init (&f->condvar, &cattr)) + clib_unix_warning ("cond_init1"); + if (pthread_condattr_destroy (&cattr)) + clib_unix_warning ("cond_init2"); + + return (f); +} + +always_inline ooo_segment_t * +ooo_segment_new (svm_fifo_t * f, u32 start, u32 length) +{ + ooo_segment_t *s; + + pool_get (f->ooo_segments, s); + + s->fifo_position = start; + s->length = length; + + s->prev = s->next = OOO_SEGMENT_INVALID_INDEX; + + return s; +} + +always_inline void +ooo_segment_del (svm_fifo_t * f, u32 index) +{ + ooo_segment_t *cur, *prev = 0, *next = 0; + cur = pool_elt_at_index (f->ooo_segments, index); + + if (cur->next != OOO_SEGMENT_INVALID_INDEX) + { + next = pool_elt_at_index (f->ooo_segments, cur->next); + next->prev = cur->prev; + } + + if (cur->prev != OOO_SEGMENT_INVALID_INDEX) + { + prev = pool_elt_at_index (f->ooo_segments, cur->prev); + prev->next = cur->next; + } + else + { + f->ooos_list_head = cur->next; + } + + pool_put (f->ooo_segments, cur); +} + +/** + * Add segment to fifo's out-of-order segment list. Takes care of merging + * adjacent segments and removing overlapping ones. + */ +static void +ooo_segment_add (svm_fifo_t * f, u32 offset, u32 length) +{ + ooo_segment_t *s, *new_s, *prev, *next, *it; + u32 new_index, position, end_offset, s_sof, s_eof, s_index; + + position = (f->tail + offset) % f->nitems; + end_offset = offset + length; + + if (f->ooos_list_head == OOO_SEGMENT_INVALID_INDEX) + { + s = ooo_segment_new (f, position, length); + f->ooos_list_head = s - f->ooo_segments; + f->ooos_newest = f->ooos_list_head; + return; + } + + /* Find first segment that starts after new segment */ + s = pool_elt_at_index (f->ooo_segments, f->ooos_list_head); + while (s->next != OOO_SEGMENT_INVALID_INDEX + && ooo_segment_offset (f, s) <= offset) + s = pool_elt_at_index (f->ooo_segments, s->next); + + s_index = s - f->ooo_segments; + s_sof = ooo_segment_offset (f, s); + s_eof = ooo_segment_end_offset (f, s); + + /* No overlap, add before current segment */ + if (end_offset < s_sof) + { + new_s = ooo_segment_new (f, position, length); + new_index = new_s - f->ooo_segments; + + /* Pool might've moved, get segment again */ + s = pool_elt_at_index (f->ooo_segments, s_index); + + if (s->prev != OOO_SEGMENT_INVALID_INDEX) + { + new_s->prev = s->prev; + + prev = pool_elt_at_index (f->ooo_segments, new_s->prev); + prev->next = new_index; + } + else + { + /* New head */ + f->ooos_list_head = new_index; + } + + new_s->next = s - f->ooo_segments; + s->prev = new_index; + f->ooos_newest = new_index; + return; + } + /* No overlap, add after current segment */ + else if (s_eof < offset) + { + new_s = ooo_segment_new (f, position, length); + new_index = new_s - f->ooo_segments; + + /* Pool might've moved, get segment again */ + s = pool_elt_at_index (f->ooo_segments, s_index); + + if (s->next != OOO_SEGMENT_INVALID_INDEX) + { + new_s->next = s->next; + + next = pool_elt_at_index (f->ooo_segments, new_s->next); + next->prev = new_index; + } + + new_s->prev = s - f->ooo_segments; + s->next = new_index; + f->ooos_newest = new_index; + + return; + } + + /* + * Merge needed + */ + + /* Merge at head */ + if (offset <= s_sof) + { + /* If we have a previous, check if we overlap */ + if (s->prev != OOO_SEGMENT_INVALID_INDEX) + { + prev = pool_elt_at_index (f->ooo_segments, s->prev); + + /* New segment merges prev and current. Remove previous and + * update position of current. */ + if (ooo_segment_end_offset (f, prev) >= offset) + { + s->fifo_position = prev->fifo_position; + s->length = s_eof - ooo_segment_offset (f, prev); + ooo_segment_del (f, s->prev); + } + } + else + { + s->fifo_position = position; + s->length = s_eof - ooo_segment_offset (f, s); + } + + /* The new segment's tail may cover multiple smaller ones */ + if (s_eof < end_offset) + { + /* Remove segments completely covered */ + it = (s->next != OOO_SEGMENT_INVALID_INDEX) ? + pool_elt_at_index (f->ooo_segments, s->next) : 0; + while (it && ooo_segment_end_offset (f, it) < end_offset) + { + next = (it->next != OOO_SEGMENT_INVALID_INDEX) ? + pool_elt_at_index (f->ooo_segments, it->next) : 0; + ooo_segment_del (f, it - f->ooo_segments); + it = next; + } + + /* Update length. Segment's start might have changed. */ + s->length = end_offset - ooo_segment_offset (f, s); + + /* If partial overlap with last, merge */ + if (it && ooo_segment_offset (f, it) < end_offset) + { + s->length += + it->length - (ooo_segment_offset (f, it) - end_offset); + ooo_segment_del (f, it - f->ooo_segments); + } + } + } + /* Last but overlapping previous */ + else if (s_eof <= end_offset) + { + s->length = end_offset - ooo_segment_offset (f, s); + } + /* New segment completely covered by current one */ + else + { + /* Do Nothing */ + } + + /* Most recently updated segment */ + f->ooos_newest = s - f->ooo_segments; +} + +/** + * Removes segments that can now be enqueued because the fifo's tail has + * advanced. Returns the number of bytes added to tail. + */ +static int +ooo_segment_try_collect (svm_fifo_t * f, u32 n_bytes_enqueued) +{ + ooo_segment_t *s; + u32 index, bytes = 0, diff; + + s = pool_elt_at_index (f->ooo_segments, f->ooos_list_head); + + /* If last tail update overlaps one/multiple ooo segments, remove them */ + diff = (f->nitems + f->tail - s->fifo_position) % f->nitems; + while (0 < diff && diff < n_bytes_enqueued) + { + /* Segment end is beyond the tail. Advance tail and be done */ + if (diff < s->length) + { + f->tail += s->length - diff; + f->tail %= f->nitems; + break; + } + /* If we have next go on */ + else if (s->next != OOO_SEGMENT_INVALID_INDEX) + { + index = s - f->ooo_segments; + s = pool_elt_at_index (f->ooo_segments, s->next); + diff = (f->nitems + f->tail - s->fifo_position) % f->nitems; + ooo_segment_del (f, index); + } + /* End of search */ + else + { + break; + } + } + + /* If tail is adjacent to an ooo segment, 'consume' it */ + if (diff == 0) + { + bytes = ((f->nitems - f->cursize) >= s->length) ? s->length : + f->nitems - f->cursize; + + f->tail += bytes; + f->tail %= f->nitems; + + ooo_segment_del (f, s - f->ooo_segments); + } + + return bytes; +} + +static int +svm_fifo_enqueue_internal (svm_fifo_t * f, + int pid, u32 max_bytes, u8 * copy_from_here) +{ + u32 total_copy_bytes, first_copy_bytes, second_copy_bytes; + u32 cursize, nitems; + + if (PREDICT_FALSE (f->cursize == f->nitems)) + return -2; /* fifo stuffed */ + + /* read cursize, which can only decrease while we're working */ + cursize = f->cursize; + nitems = f->nitems; + + /* Number of bytes we're going to copy */ + total_copy_bytes = (nitems - cursize) < max_bytes ? + (nitems - cursize) : max_bytes; + + if (PREDICT_TRUE (copy_from_here != 0)) + { + /* Number of bytes in first copy segment */ + first_copy_bytes = ((nitems - f->tail) < total_copy_bytes) + ? (nitems - f->tail) : total_copy_bytes; + + clib_memcpy (&f->data[f->tail], copy_from_here, first_copy_bytes); + f->tail += first_copy_bytes; + f->tail = (f->tail == nitems) ? 0 : f->tail; + + /* Number of bytes in second copy segment, if any */ + second_copy_bytes = total_copy_bytes - first_copy_bytes; + if (second_copy_bytes) + { + clib_memcpy (&f->data[f->tail], copy_from_here + first_copy_bytes, + second_copy_bytes); + f->tail += second_copy_bytes; + f->tail = (f->tail == nitems) ? 0 : f->tail; + } + } + else + { + /* Account for a zero-copy enqueue done elsewhere */ + ASSERT (max_bytes <= (nitems - cursize)); + f->tail += max_bytes; + f->tail = f->tail % nitems; + total_copy_bytes = max_bytes; + } + + /* Any out-of-order segments to collect? */ + if (PREDICT_FALSE (f->ooos_list_head != OOO_SEGMENT_INVALID_INDEX)) + total_copy_bytes += ooo_segment_try_collect (f, total_copy_bytes); + + /* Atomically increase the queue length */ + __sync_fetch_and_add (&f->cursize, total_copy_bytes); + + return (total_copy_bytes); +} + +int +svm_fifo_enqueue_nowait (svm_fifo_t * f, + int pid, u32 max_bytes, u8 * copy_from_here) +{ + return svm_fifo_enqueue_internal (f, pid, max_bytes, copy_from_here); +} + +/** Enqueue a future segment. + * Two choices: either copies the entire segment, or copies nothing + * Returns 0 of the entire segment was copied + * Returns -1 if none of the segment was copied due to lack of space + */ + +static int +svm_fifo_enqueue_with_offset_internal2 (svm_fifo_t * f, + int pid, + u32 offset, + u32 required_bytes, + u8 * copy_from_here) +{ + u32 total_copy_bytes, first_copy_bytes, second_copy_bytes; + u32 cursize, nitems; + u32 tail_plus_offset; + + ASSERT (offset > 0); + + /* read cursize, which can only decrease while we're working */ + cursize = f->cursize; + nitems = f->nitems; + + /* Will this request fit? */ + if ((required_bytes + offset) > (nitems - cursize)) + return -1; + + ooo_segment_add (f, offset, required_bytes); + + /* Number of bytes we're going to copy */ + total_copy_bytes = required_bytes; + tail_plus_offset = (f->tail + offset) % nitems; + + /* Number of bytes in first copy segment */ + first_copy_bytes = ((nitems - tail_plus_offset) < total_copy_bytes) + ? (nitems - tail_plus_offset) : total_copy_bytes; + + clib_memcpy (&f->data[tail_plus_offset], copy_from_here, first_copy_bytes); + + /* Number of bytes in second copy segment, if any */ + second_copy_bytes = total_copy_bytes - first_copy_bytes; + if (second_copy_bytes) + { + tail_plus_offset += first_copy_bytes; + tail_plus_offset %= nitems; + + ASSERT (tail_plus_offset == 0); + + clib_memcpy (&f->data[tail_plus_offset], + copy_from_here + first_copy_bytes, second_copy_bytes); + } + + return (0); +} + + +int +svm_fifo_enqueue_with_offset (svm_fifo_t * f, + int pid, + u32 offset, + u32 required_bytes, u8 * copy_from_here) +{ + return svm_fifo_enqueue_with_offset_internal2 + (f, pid, offset, required_bytes, copy_from_here); +} + + +static int +svm_fifo_dequeue_internal2 (svm_fifo_t * f, + int pid, u32 max_bytes, u8 * copy_here) +{ + u32 total_copy_bytes, first_copy_bytes, second_copy_bytes; + u32 cursize, nitems; + + if (PREDICT_FALSE (f->cursize == 0)) + return -2; /* nothing in the fifo */ + + /* read cursize, which can only increase while we're working */ + cursize = f->cursize; + nitems = f->nitems; + + /* Number of bytes we're going to copy */ + total_copy_bytes = (cursize < max_bytes) ? cursize : max_bytes; + + if (PREDICT_TRUE (copy_here != 0)) + { + /* Number of bytes in first copy segment */ + first_copy_bytes = ((nitems - f->head) < total_copy_bytes) + ? (nitems - f->head) : total_copy_bytes; + clib_memcpy (copy_here, &f->data[f->head], first_copy_bytes); + f->head += first_copy_bytes; + f->head = (f->head == nitems) ? 0 : f->head; + + /* Number of bytes in second copy segment, if any */ + second_copy_bytes = total_copy_bytes - first_copy_bytes; + if (second_copy_bytes) + { + clib_memcpy (copy_here + first_copy_bytes, + &f->data[f->head], second_copy_bytes); + f->head += second_copy_bytes; + f->head = (f->head == nitems) ? 0 : f->head; + } + } + else + { + /* Account for a zero-copy dequeue done elsewhere */ + ASSERT (max_bytes <= cursize); + f->head += max_bytes; + f->head = f->head % nitems; + cursize -= max_bytes; + total_copy_bytes = max_bytes; + } + + __sync_fetch_and_sub (&f->cursize, total_copy_bytes); + + return (total_copy_bytes); +} + +int +svm_fifo_dequeue_nowait (svm_fifo_t * f, + int pid, u32 max_bytes, u8 * copy_here) +{ + return svm_fifo_dequeue_internal2 (f, pid, max_bytes, copy_here); +} + +int +svm_fifo_peek (svm_fifo_t * f, int pid, u32 offset, u32 max_bytes, + u8 * copy_here) +{ + u32 total_copy_bytes, first_copy_bytes, second_copy_bytes; + u32 cursize, nitems; + + if (PREDICT_FALSE (f->cursize == 0)) + return -2; /* nothing in the fifo */ + + /* read cursize, which can only increase while we're working */ + cursize = f->cursize; + nitems = f->nitems; + + /* Number of bytes we're going to copy */ + total_copy_bytes = (cursize < max_bytes) ? cursize : max_bytes; + + if (PREDICT_TRUE (copy_here != 0)) + { + /* Number of bytes in first copy segment */ + first_copy_bytes = + ((nitems - f->head) < total_copy_bytes) ? + (nitems - f->head) : total_copy_bytes; + clib_memcpy (copy_here, &f->data[f->head], first_copy_bytes); + + /* Number of bytes in second copy segment, if any */ + second_copy_bytes = total_copy_bytes - first_copy_bytes; + if (second_copy_bytes) + { + clib_memcpy (copy_here + first_copy_bytes, &f->data[0], + second_copy_bytes); + } + } + return total_copy_bytes; +} + +int +svm_fifo_dequeue_drop (svm_fifo_t * f, int pid, u32 max_bytes) +{ + u32 total_drop_bytes, first_drop_bytes, second_drop_bytes; + u32 cursize, nitems; + + if (PREDICT_FALSE (f->cursize == 0)) + return -2; /* nothing in the fifo */ + + /* read cursize, which can only increase while we're working */ + cursize = f->cursize; + nitems = f->nitems; + + /* Number of bytes we're going to drop */ + total_drop_bytes = (cursize < max_bytes) ? cursize : max_bytes; + + /* Number of bytes in first copy segment */ + first_drop_bytes = + ((nitems - f->head) < total_drop_bytes) ? + (nitems - f->head) : total_drop_bytes; + f->head += first_drop_bytes; + f->head = (f->head == nitems) ? 0 : f->head; + + /* Number of bytes in second drop segment, if any */ + second_drop_bytes = total_drop_bytes - first_drop_bytes; + if (second_drop_bytes) + { + f->head += second_drop_bytes; + f->head = (f->head == nitems) ? 0 : f->head; + } + + __sync_fetch_and_sub (&f->cursize, total_drop_bytes); + + return total_drop_bytes; +} + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/svm/svm_fifo.h b/src/svm/svm_fifo.h new file mode 100644 index 00000000000..70624b74ed1 --- /dev/null +++ b/src/svm/svm_fifo.h @@ -0,0 +1,157 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#ifndef __included_ssvm_fifo_h__ +#define __included_ssvm_fifo_h__ + +#include <vppinfra/clib.h> +#include <vppinfra/vec.h> +#include <vppinfra/mheap.h> +#include <vppinfra/heap.h> +#include <vppinfra/pool.h> +#include <vppinfra/format.h> +#include <pthread.h> + +typedef enum +{ + SVM_FIFO_TAG_NOT_HELD = 0, + SVM_FIFO_TAG_DEQUEUE, + SVM_FIFO_TAG_ENQUEUE, +} svm_lock_tag_t; + +/** Out-of-order segment */ +typedef struct +{ + u32 next; /**< Next linked-list element pool index */ + u32 prev; /**< Previous linked-list element pool index */ + + u32 fifo_position; /**< Start of segment, normalized*/ + u32 length; /**< Length of segment */ +} ooo_segment_t; + +#define OOO_SEGMENT_INVALID_INDEX ((u32)~0) + +typedef struct +{ + pthread_mutex_t mutex; /* 8 bytes */ + pthread_cond_t condvar; /* 8 bytes */ + u32 owner_pid; + svm_lock_tag_t tag; + volatile u32 cursize; + u32 nitems; + + /* Backpointers */ + u32 server_session_index; + u32 client_session_index; + u8 server_thread_index; + u8 client_thread_index; + CLIB_CACHE_LINE_ALIGN_MARK (end_shared); + u32 head; + CLIB_CACHE_LINE_ALIGN_MARK (end_consumer); + + /* producer */ + u32 tail; + + ooo_segment_t *ooo_segments; /**< Pool of ooo segments */ + u32 ooos_list_head; /**< Head of out-of-order linked-list */ + u32 ooos_newest; /**< Last segment to have been updated */ + + CLIB_CACHE_LINE_ALIGN_MARK (data); +} svm_fifo_t; + +static inline int +svm_fifo_lock (svm_fifo_t * f, u32 pid, u32 tag, int nowait) +{ + if (PREDICT_TRUE (nowait == 0)) + pthread_mutex_lock (&f->mutex); + else + { + if (pthread_mutex_trylock (&f->mutex)) + return -1; + } + f->owner_pid = pid; + f->tag = tag; + return 0; +} + +static inline void +svm_fifo_unlock (svm_fifo_t * f) +{ + f->owner_pid = 0; + f->tag = 0; + CLIB_MEMORY_BARRIER (); + pthread_mutex_unlock (&f->mutex); +} + +static inline u32 +svm_fifo_max_dequeue (svm_fifo_t * f) +{ + return f->cursize; +} + +static inline u32 +svm_fifo_max_enqueue (svm_fifo_t * f) +{ + return f->nitems - f->cursize; +} + +static inline u8 +svm_fifo_has_ooo_data (svm_fifo_t * f) +{ + return f->ooos_list_head != OOO_SEGMENT_INVALID_INDEX; +} + +svm_fifo_t *svm_fifo_create (u32 data_size_in_bytes); + +int svm_fifo_enqueue_nowait (svm_fifo_t * f, int pid, u32 max_bytes, + u8 * copy_from_here); + +int svm_fifo_enqueue_with_offset (svm_fifo_t * f, int pid, + u32 offset, u32 required_bytes, + u8 * copy_from_here); + +int svm_fifo_dequeue_nowait (svm_fifo_t * f, int pid, u32 max_bytes, + u8 * copy_here); + +int svm_fifo_peek (svm_fifo_t * f, int pid, u32 offset, u32 max_bytes, + u8 * copy_here); +int svm_fifo_dequeue_drop (svm_fifo_t * f, int pid, u32 max_bytes); + +always_inline ooo_segment_t * +svm_fifo_newest_ooo_segment (svm_fifo_t * f) +{ + return f->ooo_segments + f->ooos_newest; +} + +always_inline u32 +ooo_segment_offset (svm_fifo_t * f, ooo_segment_t * s) +{ + return ((f->nitems + s->fifo_position - f->tail) % f->nitems); +} + +always_inline u32 +ooo_segment_end_offset (svm_fifo_t * f, ooo_segment_t * s) +{ + return ((f->nitems + s->fifo_position + s->length - f->tail) % f->nitems); +} + +#endif /* __included_ssvm_fifo_h__ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/svm/svm_fifo_segment.c b/src/svm/svm_fifo_segment.c new file mode 100644 index 00000000000..acabb3bd95c --- /dev/null +++ b/src/svm/svm_fifo_segment.c @@ -0,0 +1,193 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <svm/svm_fifo_segment.h> + +svm_fifo_segment_main_t svm_fifo_segment_main; + +/** (master) create an svm fifo segment */ +int +svm_fifo_segment_create (svm_fifo_segment_create_args_t * a) +{ + int rv; + svm_fifo_segment_private_t *s; + svm_fifo_segment_main_t *sm = &svm_fifo_segment_main; + ssvm_shared_header_t *sh; + svm_fifo_segment_header_t *fsh; + void *oldheap; + + /* Allocate a fresh segment */ + pool_get (sm->segments, s); + memset (s, 0, sizeof (*s)); + + s->ssvm.ssvm_size = a->segment_size; + s->ssvm.i_am_master = 1; + s->ssvm.my_pid = getpid (); + s->ssvm.name = (u8 *) a->segment_name; + s->ssvm.requested_va = sm->next_baseva; + + rv = ssvm_master_init (&s->ssvm, s - sm->segments); + + if (rv) + { + _vec_len (s) = vec_len (s) - 1; + return (rv); + } + + /* Note; requested_va updated due to seg base addr randomization */ + sm->next_baseva = s->ssvm.requested_va + a->segment_size; + + sh = s->ssvm.sh; + oldheap = ssvm_push_heap (sh); + + /* Set up svm_fifo_segment shared header */ + fsh = clib_mem_alloc (sizeof (*fsh)); + memset (fsh, 0, sizeof (*fsh)); + sh->opaque[0] = fsh; + s->h = fsh; + fsh->segment_name = format (0, "%s%c", a->segment_name, 0); + + /* Avoid vec_add1(...) failure when adding a fifo, etc. */ + vec_validate (fsh->fifos, 64); + _vec_len (fsh->fifos) = 0; + + ssvm_pop_heap (oldheap); + + sh->ready = 1; + a->new_segment_index = s - sm->segments; + return (0); +} + +/** (slave) attach to an svm fifo segment */ +int +svm_fifo_segment_attach (svm_fifo_segment_create_args_t * a) +{ + int rv; + svm_fifo_segment_private_t *s; + svm_fifo_segment_main_t *sm = &svm_fifo_segment_main; + ssvm_shared_header_t *sh; + svm_fifo_segment_header_t *fsh; + + /* Allocate a fresh segment */ + pool_get (sm->segments, s); + + memset (s, 0, sizeof (*s)); + + s->ssvm.ssvm_size = a->segment_size; + s->ssvm.my_pid = getpid (); + s->ssvm.name = (u8 *) a->segment_name; + s->ssvm.requested_va = sm->next_baseva; + + rv = ssvm_slave_init (&s->ssvm, sm->timeout_in_seconds); + + if (rv) + { + _vec_len (s) = vec_len (s) - 1; + return (rv); + } + + /* Fish the segment header */ + sh = s->ssvm.sh; + fsh = (svm_fifo_segment_header_t *) sh->opaque[0]; + s->h = fsh; + + a->new_segment_index = s - sm->segments; + return (0); +} + +void +svm_fifo_segment_delete (svm_fifo_segment_private_t * s) +{ + svm_fifo_segment_main_t *sm = &svm_fifo_segment_main; + ssvm_delete (&s->ssvm); + pool_put (sm->segments, s); +} + +svm_fifo_t * +svm_fifo_segment_alloc_fifo (svm_fifo_segment_private_t * s, + u32 data_size_in_bytes) +{ + ssvm_shared_header_t *sh; + svm_fifo_segment_header_t *fsh; + svm_fifo_t *f; + void *oldheap; + + sh = s->ssvm.sh; + fsh = (svm_fifo_segment_header_t *) sh->opaque[0]; + oldheap = ssvm_push_heap (sh); + + /* Note: this can fail, in which case: create another segment */ + f = svm_fifo_create (data_size_in_bytes); + if (f == 0) + { + ssvm_pop_heap (oldheap); + return (0); + } + + vec_add1 (fsh->fifos, f); + + ssvm_pop_heap (oldheap); + return (f); +} + +void +svm_fifo_segment_free_fifo (svm_fifo_segment_private_t * s, svm_fifo_t * f) +{ + ssvm_shared_header_t *sh; + svm_fifo_segment_header_t *fsh; + void *oldheap; + int i; + + sh = s->ssvm.sh; + fsh = (svm_fifo_segment_header_t *) sh->opaque[0]; + oldheap = ssvm_push_heap (sh); + + for (i = 0; i < vec_len (fsh->fifos); i++) + { + if (fsh->fifos[i] == f) + { + vec_delete (fsh->fifos, 1, i); + goto found; + } + } + clib_warning ("fifo 0x%llx not found in fifo table...", f); + +found: + clib_mem_free (f); + ssvm_pop_heap (oldheap); +} + +void +svm_fifo_segment_init (u64 baseva, u32 timeout_in_seconds) +{ + svm_fifo_segment_main_t *sm = &svm_fifo_segment_main; + + sm->next_baseva = baseva; + sm->timeout_in_seconds = timeout_in_seconds; +} + +u32 +svm_fifo_segment_index (svm_fifo_segment_private_t * s) +{ + return s - svm_fifo_segment_main.segments; +} + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/svm/svm_fifo_segment.h b/src/svm/svm_fifo_segment.h new file mode 100644 index 00000000000..793fa7c8a0e --- /dev/null +++ b/src/svm/svm_fifo_segment.h @@ -0,0 +1,89 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#ifndef __included_ssvm_fifo_segment_h__ +#define __included_ssvm_fifo_segment_h__ + +#include "svm_fifo.h" +#include "ssvm.h" + +typedef struct +{ + volatile svm_fifo_t **fifos; + u8 *segment_name; +} svm_fifo_segment_header_t; + +typedef struct +{ + ssvm_private_t ssvm; + svm_fifo_segment_header_t *h; +} svm_fifo_segment_private_t; + +typedef struct +{ + /** pool of segments */ + svm_fifo_segment_private_t *segments; + /* Where to put the next one */ + u64 next_baseva; + u32 timeout_in_seconds; +} svm_fifo_segment_main_t; + +extern svm_fifo_segment_main_t svm_fifo_segment_main; + +typedef struct +{ + char *segment_name; + u32 segment_size; + u32 new_segment_index; +} svm_fifo_segment_create_args_t; + +static inline svm_fifo_segment_private_t * +svm_fifo_get_segment (u32 segment_index) +{ + svm_fifo_segment_main_t *ssm = &svm_fifo_segment_main; + return vec_elt_at_index (ssm->segments, segment_index); +} + +#define foreach_ssvm_fifo_segment_api_error \ +_(OUT_OF_SPACE, "Out of space in segment", -200) + +typedef enum +{ +#define _(n,s,c) SSVM_FIFO_SEGMENT_API_ERROR_##n = c, + foreach_ssvm_fifo_segment_api_error +#undef _ +} ssvm_fifo_segment_api_error_enum_t; + +int svm_fifo_segment_create (svm_fifo_segment_create_args_t * a); +int svm_fifo_segment_attach (svm_fifo_segment_create_args_t * a); +void svm_fifo_segment_delete (svm_fifo_segment_private_t * s); + +svm_fifo_t *svm_fifo_segment_alloc_fifo (svm_fifo_segment_private_t * s, + u32 data_size_in_bytes); +void svm_fifo_segment_free_fifo (svm_fifo_segment_private_t * s, + svm_fifo_t * f); + +void svm_fifo_segment_init (u64 baseva, u32 timeout_in_seconds); + +u32 svm_fifo_segment_index (svm_fifo_segment_private_t * s); + +#endif /* __included_ssvm_fifo_segment_h__ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/svm/test_svm_fifo1.c b/src/svm/test_svm_fifo1.c new file mode 100644 index 00000000000..355653df42a --- /dev/null +++ b/src/svm/test_svm_fifo1.c @@ -0,0 +1,361 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "svm_fifo_segment.h" + +clib_error_t * +hello_world (int verbose) +{ + svm_fifo_segment_create_args_t _a, *a = &_a; + svm_fifo_segment_private_t *sp; + svm_fifo_t *f; + int rv; + u8 *test_data; + u8 *retrieved_data = 0; + clib_error_t *error = 0; + int pid = getpid (); + + memset (a, 0, sizeof (*a)); + + a->segment_name = "fifo-test1"; + a->segment_size = 256 << 10; + + rv = svm_fifo_segment_create (a); + + if (rv) + return clib_error_return (0, "svm_fifo_segment_create returned %d", rv); + + sp = svm_fifo_get_segment (a->new_segment_index); + + f = svm_fifo_segment_alloc_fifo (sp, 4096); + + if (f == 0) + return clib_error_return (0, "svm_fifo_segment_alloc_fifo failed"); + + test_data = format (0, "Hello world%c", 0); + vec_validate (retrieved_data, vec_len (test_data) - 1); + + while (svm_fifo_max_enqueue (f) >= vec_len (test_data)) + svm_fifo_enqueue_nowait (f, pid, vec_len (test_data), test_data); + + while (svm_fifo_max_dequeue (f) >= vec_len (test_data)) + svm_fifo_dequeue_nowait (f, pid, vec_len (retrieved_data), + retrieved_data); + + while (svm_fifo_max_enqueue (f) >= vec_len (test_data)) + svm_fifo_enqueue_nowait (f, pid, vec_len (test_data), test_data); + + while (svm_fifo_max_dequeue (f) >= vec_len (test_data)) + svm_fifo_dequeue_nowait (f, pid, vec_len (retrieved_data), + retrieved_data); + + if (!memcmp (retrieved_data, test_data, vec_len (test_data))) + error = clib_error_return (0, "data test OK, got '%s'", retrieved_data); + else + error = clib_error_return (0, "data test FAIL!"); + + svm_fifo_segment_free_fifo (sp, f); + + return error; +} + +clib_error_t * +master (int verbose) +{ + svm_fifo_segment_create_args_t _a, *a = &_a; + svm_fifo_segment_private_t *sp; + svm_fifo_t *f; + int rv; + u8 *test_data; + u8 *retrieved_data = 0; + int i; + int pid = getpid (); + + memset (a, 0, sizeof (*a)); + + a->segment_name = "fifo-test1"; + a->segment_size = 256 << 10; + + rv = svm_fifo_segment_create (a); + + if (rv) + return clib_error_return (0, "svm_fifo_segment_create returned %d", rv); + + sp = svm_fifo_get_segment (a->new_segment_index); + + f = svm_fifo_segment_alloc_fifo (sp, 4096); + + if (f == 0) + return clib_error_return (0, "svm_fifo_segment_alloc_fifo failed"); + + test_data = format (0, "Hello world%c", 0); + vec_validate (retrieved_data, vec_len (test_data) - 1); + + for (i = 0; i < 1000; i++) + svm_fifo_enqueue_nowait (f, pid, vec_len (test_data), test_data); + + return clib_error_return (0, "master (enqueue) done"); +} + +clib_error_t * +mempig (int verbose) +{ + svm_fifo_segment_create_args_t _a, *a = &_a; + svm_fifo_segment_private_t *sp; + svm_fifo_t *f; + svm_fifo_t **flist = 0; + int rv; + int i; + + memset (a, 0, sizeof (*a)); + + a->segment_name = "fifo-test1"; + a->segment_size = 256 << 10; + + rv = svm_fifo_segment_create (a); + + if (rv) + return clib_error_return (0, "svm_fifo_segment_create returned %d", rv); + + sp = svm_fifo_get_segment (a->new_segment_index); + + for (i = 0; i < 1000; i++) + { + f = svm_fifo_segment_alloc_fifo (sp, 4096); + if (f == 0) + break; + vec_add1 (flist, f); + } + + fformat (stdout, "Try #1: created %d fifos...\n", vec_len (flist)); + for (i = 0; i < vec_len (flist); i++) + { + f = flist[i]; + svm_fifo_segment_free_fifo (sp, f); + } + + _vec_len (flist) = 0; + + for (i = 0; i < 1000; i++) + { + f = svm_fifo_segment_alloc_fifo (sp, 4096); + if (f == 0) + break; + vec_add1 (flist, f); + } + + fformat (stdout, "Try #2: created %d fifos...\n", vec_len (flist)); + for (i = 0; i < vec_len (flist); i++) + { + f = flist[i]; + svm_fifo_segment_free_fifo (sp, f); + } + + return 0; +} + +clib_error_t * +offset (int verbose) +{ + svm_fifo_segment_create_args_t _a, *a = &_a; + svm_fifo_segment_private_t *sp; + svm_fifo_t *f; + int rv; + u32 *test_data = 0; + u32 *recovered_data = 0; + int i; + int pid = getpid (); + + memset (a, 0, sizeof (*a)); + + a->segment_name = "fifo-test1"; + a->segment_size = 256 << 10; + + rv = svm_fifo_segment_create (a); + + if (rv) + return clib_error_return (0, "svm_fifo_segment_create returned %d", rv); + + sp = svm_fifo_get_segment (a->new_segment_index); + + f = svm_fifo_segment_alloc_fifo (sp, 200 << 10); + + if (f == 0) + return clib_error_return (0, "svm_fifo_segment_alloc_fifo failed"); + + for (i = 0; i < (3 * 1024); i++) + vec_add1 (test_data, i); + + /* Enqueue the first 1024 u32's */ + svm_fifo_enqueue_nowait (f, pid, 4096 /* bytes to enqueue */ , + (u8 *) test_data); + + /* Enqueue the third 1024 u32's 2048 ahead of the current tail */ + svm_fifo_enqueue_with_offset (f, pid, 4096, 4096, (u8 *) & test_data[2048]); + + /* Enqueue the second 1024 u32's at the current tail */ + svm_fifo_enqueue_nowait (f, pid, 4096 /* bytes to enqueue */ , + (u8 *) & test_data[1024]); + + vec_validate (recovered_data, (3 * 1024) - 1); + + svm_fifo_dequeue_nowait (f, pid, 3 * 4096, (u8 *) recovered_data); + + for (i = 0; i < (3 * 1024); i++) + { + if (recovered_data[i] != test_data[i]) + { + clib_warning ("[%d] expected %d recovered %d", i, + test_data[i], recovered_data[i]); + return clib_error_return (0, "offset test FAILED"); + } + } + + return clib_error_return (0, "offset test OK"); +} + +clib_error_t * +slave (int verbose) +{ + svm_fifo_segment_create_args_t _a, *a = &_a; + svm_fifo_segment_private_t *sp; + svm_fifo_segment_header_t *fsh; + svm_fifo_t *f; + ssvm_shared_header_t *sh; + int rv; + u8 *test_data; + u8 *retrieved_data = 0; + int pid = getpid (); + int i; + + memset (a, 0, sizeof (*a)); + + a->segment_name = "fifo-test1"; + + rv = svm_fifo_segment_attach (a); + + if (rv) + return clib_error_return (0, "svm_fifo_segment_attach returned %d", rv); + + sp = svm_fifo_get_segment (a->new_segment_index); + sh = sp->ssvm.sh; + fsh = (svm_fifo_segment_header_t *) sh->opaque[0]; + + /* might wanna wait.. */ + f = (svm_fifo_t *) fsh->fifos[0]; + + /* Lazy bastards united */ + test_data = format (0, "Hello world%c", 0); + vec_validate (retrieved_data, vec_len (test_data) - 1); + + for (i = 0; i < 1000; i++) + { + svm_fifo_dequeue_nowait (f, pid, vec_len (retrieved_data), + retrieved_data); + if (memcmp (retrieved_data, test_data, vec_len (retrieved_data))) + return clib_error_return (0, "retrieved data incorrect, '%s'", + retrieved_data); + } + + return clib_error_return (0, "slave (dequeue) done"); +} + + +int +test_ssvm_fifo1 (unformat_input_t * input) +{ + clib_error_t *error = 0; + int verbose = 0; + int test_id = 0; + + svm_fifo_segment_init (0x200000000ULL, 20); + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "verbose %d", &verbose)) + ; + else if (unformat (input, "verbose")) + verbose = 1; + else if (unformat (input, "master")) + test_id = 1; + else if (unformat (input, "slave")) + test_id = 2; + else if (unformat (input, "mempig")) + test_id = 3; + else if (unformat (input, "offset")) + test_id = 4; + else + { + error = clib_error_create ("unknown input `%U'\n", + format_unformat_error, input); + goto out; + } + } + + switch (test_id) + { + case 0: + error = hello_world (verbose); + break; + + case 1: + error = master (verbose); + break; + + case 2: + error = slave (verbose); + break; + + case 3: + error = mempig (verbose); + break; + + case 4: + error = offset (verbose); + break; + + default: + error = clib_error_return (0, "test id %d unknown", test_id); + break; + } + +out: + if (error) + clib_error_report (error); + + return 0; +} + + + +int +main (int argc, char *argv[]) +{ + unformat_input_t i; + int r; + + unformat_init_command_line (&i, argv); + r = test_ssvm_fifo1 (&i); + unformat_free (&i); + return r; +} + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/uri.am b/src/uri.am new file mode 100644 index 00000000000..8cdd77c6a8c --- /dev/null +++ b/src/uri.am @@ -0,0 +1,22 @@ +# Copyright (c) 2016 Cisco and/or its affiliates. +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at: +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +noinst_PROGRAMS += uri_udp_test2 uri_tcp_test + +uri_udp_test2_SOURCES = uri/uri_udp_test2.c +uri_udp_test2_LDADD = libvlibmemoryclient.la libvlibapi.la libsvm.la \ + libvppinfra.la -lpthread -lm -lrt + +uri_tcp_test_SOURCES = uri/uri_tcp_test.c +uri_tcp_test_LDADD = libvlibmemoryclient.la libvlibapi.la libsvm.la \ + libvppinfra.la -lpthread -lm -lrt diff --git a/src/uri/uri_tcp_test.c b/src/uri/uri_tcp_test.c new file mode 100644 index 00000000000..ed5a37d813b --- /dev/null +++ b/src/uri/uri_tcp_test.c @@ -0,0 +1,916 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <stdio.h> +#include <signal.h> +#include <vlib/vlib.h> +#include <vnet/vnet.h> +#include <svm/svm_fifo_segment.h> +#include <vlibmemory/api.h> +#include <vpp/api/vpe_msg_enum.h> + +#include "../vnet/session/application_interface.h" + +#define vl_typedefs /* define message structures */ +#include <vpp/api/vpe_all_api_h.h> +#undef vl_typedefs + +/* declare message handlers for each api */ + +#define vl_endianfun /* define message structures */ +#include <vpp/api/vpe_all_api_h.h> +#undef vl_endianfun + +/* instantiate all the print functions we know about */ +#define vl_print(handle, ...) +#define vl_printfun +#include <vpp/api/vpe_all_api_h.h> +#undef vl_printfun + +/* Satisfy external references when not linking with -lvlib */ +vlib_main_t vlib_global_main; +vlib_main_t **vlib_mains; + +typedef struct +{ + svm_fifo_t * server_rx_fifo; + svm_fifo_t * server_tx_fifo; + + u32 vpp_session_index; + u32 vpp_session_thread; +} session_t; + +typedef enum +{ + STATE_START, + STATE_READY, + STATE_DISCONNECTING, + STATE_FAILED +} connection_state_t; + +typedef struct +{ + /* vpe input queue */ + unix_shared_memory_queue_t *vl_input_queue; + + /* API client handle */ + u32 my_client_index; + + /* The URI we're playing with */ + u8 * uri; + + /* Session pool */ + session_t * sessions; + + /* Hash table for disconnect processing */ + uword * session_index_by_vpp_handles; + + /* intermediate rx buffer */ + u8 * rx_buf; + + /* URI for slave's connect */ + u8 * connect_uri; + + u32 connected_session_index; + + int i_am_master; + + /* drop all packets */ + int drop_packets; + + /* Our event queue */ + unix_shared_memory_queue_t * our_event_queue; + + /* $$$ single thread only for the moment */ + unix_shared_memory_queue_t * vpp_event_queue; + + pid_t my_pid; + + /* For deadman timers */ + clib_time_t clib_time; + + /* State of the connection, shared between msg RX thread and main thread */ + volatile connection_state_t state; + + /* Signal variables */ + volatile int time_to_stop; + volatile int time_to_print_stats; + + u32 configured_segment_size; + + /* VNET_API_ERROR_FOO -> "Foo" hash table */ + uword * error_string_by_error_number; + + /* convenience */ + svm_fifo_segment_main_t * segment_main; + + u8 *connect_test_data; +} uri_tcp_test_main_t; + +uri_tcp_test_main_t uri_tcp_test_main; + +#if CLIB_DEBUG > 0 +#define NITER 10000 +#else +#define NITER 4000000 +#endif + +int +wait_for_state_change (uri_tcp_test_main_t * utm, connection_state_t state) +{ +#if CLIB_DEBUG > 0 +#define TIMEOUT 600.0 +#else +#define TIMEOUT 600.0 +#endif + + f64 timeout = clib_time_now (&utm->clib_time) + TIMEOUT; + + while (clib_time_now (&utm->clib_time) < timeout) + { + if (utm->state == state) + return 0; + if (utm->state == STATE_FAILED) + return -1; + } + clib_warning ("timeout waiting for STATE_READY"); + return -1; +} + +static void +init_error_string_table (uri_tcp_test_main_t * utm) +{ + utm->error_string_by_error_number = hash_create (0, sizeof (uword)); + +#define _(n,v,s) hash_set (utm->error_string_by_error_number, -v, s); + foreach_vnet_api_error; +#undef _ + + hash_set (utm->error_string_by_error_number, 99, "Misc"); +} + +static void +stop_signal (int signum) +{ + uri_tcp_test_main_t *um = &uri_tcp_test_main; + + um->time_to_stop = 1; +} + +static void +stats_signal (int signum) +{ + uri_tcp_test_main_t *um = &uri_tcp_test_main; + + um->time_to_print_stats = 1; +} + +static clib_error_t * +setup_signal_handlers (void) +{ + signal (SIGINT, stats_signal); + signal (SIGQUIT, stop_signal); + signal (SIGTERM, stop_signal); + + return 0; +} + +void +vlib_cli_output (struct vlib_main_t *vm, char *fmt, ...) +{ + clib_warning ("BUG"); +} + +int +connect_to_vpp (char *name) +{ + uri_tcp_test_main_t *utm = &uri_tcp_test_main; + api_main_t *am = &api_main; + + if (vl_client_connect_to_vlib ("/vpe-api", name, 32) < 0) + return -1; + + utm->vl_input_queue = am->shmem_hdr->vl_input_queue; + utm->my_client_index = am->my_client_index; + + return 0; +} + +static void +vl_api_map_another_segment_t_handler (vl_api_map_another_segment_t *mp) +{ + svm_fifo_segment_create_args_t _a, *a = &_a; + int rv; + + a->segment_name = (char *) mp->segment_name; + a->segment_size = mp->segment_size; + /* Attach to the segment vpp created */ + rv = svm_fifo_segment_attach (a); + if (rv) + { + clib_warning ("svm_fifo_segment_attach ('%s') failed", + mp->segment_name); + return; + } + clib_warning ("Mapped new segment '%s' size %d", mp->segment_name, + mp->segment_size); +} + +static void +vl_api_disconnect_session_t_handler (vl_api_disconnect_session_t * mp) +{ + uri_tcp_test_main_t *utm = &uri_tcp_test_main; + session_t * session; + vl_api_disconnect_session_reply_t * rmp; + uword * p; + int rv = 0; + u64 key; + + key = (((u64)mp->session_thread_index) << 32) | (u64)mp->session_index; + + p = hash_get (utm->session_index_by_vpp_handles, key); + + if (p) + { + session = pool_elt_at_index (utm->sessions, p[0]); + hash_unset (utm->session_index_by_vpp_handles, key); + pool_put (utm->sessions, session); + } + else + { + clib_warning ("couldn't find session key %llx", key); + rv = -11; + } + + rmp = vl_msg_api_alloc (sizeof (*rmp)); + memset (rmp, 0, sizeof (*rmp)); + + rmp->_vl_msg_id = ntohs (VL_API_DISCONNECT_SESSION_REPLY); + rmp->retval = rv; + rmp->session_index = mp->session_index; + rmp->session_thread_index = mp->session_thread_index; + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *)&rmp); +} + +static void +vl_api_reset_session_t_handler (vl_api_reset_session_t * mp) +{ + uri_tcp_test_main_t *utm = &uri_tcp_test_main; + session_t * session; + vl_api_reset_session_reply_t * rmp; + uword * p; + int rv = 0; + u64 key; + + key = (((u64)mp->session_thread_index) << 32) | (u64)mp->session_index; + + p = hash_get(utm->session_index_by_vpp_handles, key); + + if (p) + { + session = pool_elt_at_index(utm->sessions, p[0]); + hash_unset(utm->session_index_by_vpp_handles, key); + pool_put(utm->sessions, session); + } + else + { + clib_warning("couldn't find session key %llx", key); + rv = -11; + } + + rmp = vl_msg_api_alloc (sizeof (*rmp)); + memset (rmp, 0, sizeof (*rmp)); + rmp->_vl_msg_id = ntohs (VL_API_DISCONNECT_SESSION_REPLY); + rmp->retval = rv; + rmp->session_index = mp->session_index; + rmp->session_thread_index = mp->session_thread_index; + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *)&rmp); +} + +void +handle_fifo_event_connect_rx (uri_tcp_test_main_t *utm, session_fifo_event_t * e) +{ + svm_fifo_t * rx_fifo; + int n_read, bytes; + + rx_fifo = e->fifo; + + bytes = e->enqueue_length; + do + { + n_read = svm_fifo_dequeue_nowait (rx_fifo, 0, vec_len(utm->rx_buf), + utm->rx_buf); + if (n_read > 0) + bytes -= n_read; + } + while (n_read < 0 || bytes > 0); + + // bytes_to_read = svm_fifo_max_dequeue (rx_fifo); + // + // bytes_to_read = vec_len(utm->rx_buf) > bytes_to_read ? + // bytes_to_read : vec_len(utm->rx_buf); + // + // buffer_offset = 0; + // while (bytes_to_read > 0) + // { + // rv = svm_fifo_dequeue_nowait2 (rx_fifo, mypid, + // bytes_to_read, + // utm->rx_buf + buffer_offset); + // if (rv > 0) + // { + // bytes_to_read -= rv; + // buffer_offset += rv; + // bytes_received += rv; + // } + // } + + + // while (bytes_received < bytes_sent) + // { + // rv = svm_fifo_dequeue_nowait2 (rx_fifo, mypid, + // vec_len (utm->rx_buf), + // utm->rx_buf); + // if (rv > 0) + // { + //#if CLIB_DEBUG > 0 + // int j; + // for (j = 0; j < rv; j++) + // { + // if (utm->rx_buf[j] != ((bytes_received + j) & 0xff)) + // { + // clib_warning ("error at byte %lld, 0x%x not 0x%x", + // bytes_received + j, + // utm->rx_buf[j], + // ((bytes_received + j )&0xff)); + // } + // } + //#endif + // bytes_received += (u64) rv; + // } + // } +} + +void +handle_connect_event_queue (uri_tcp_test_main_t * utm) +{ + session_fifo_event_t _e, *e = &_e;; + + unix_shared_memory_queue_sub (utm->our_event_queue, (u8 *) e, 0 /* nowait */); + switch (e->event_type) + { + case FIFO_EVENT_SERVER_RX: + handle_fifo_event_connect_rx (utm, e); + break; + + case FIFO_EVENT_SERVER_EXIT: + return; + + default: + clib_warning("unknown event type %d", e->event_type); + break; + } +} + +void +uri_tcp_connect_send (uri_tcp_test_main_t *utm) +{ + u8 *test_data = utm->connect_test_data; + u64 bytes_sent = 0; + int rv; + int mypid = getpid(); + session_t * session; + svm_fifo_t *tx_fifo; + int buffer_offset, bytes_to_send = 0; + session_fifo_event_t evt; + static int serial_number = 0; + int i; + u32 max_chunk = 64 << 10, write; + + session = pool_elt_at_index (utm->sessions, utm->connected_session_index); + tx_fifo = session->server_tx_fifo; + + vec_validate (utm->rx_buf, vec_len (test_data) - 1); + + for (i = 0; i < 10; i++) + { + bytes_to_send = vec_len (test_data); + buffer_offset = 0; + while (bytes_to_send > 0) + { + write = bytes_to_send > max_chunk ? max_chunk : bytes_to_send; + rv = svm_fifo_enqueue_nowait (tx_fifo, mypid, write, + test_data + buffer_offset); + + if (rv > 0) + { + bytes_to_send -= rv; + buffer_offset += rv; + bytes_sent += rv; + + /* Fabricate TX event, send to vpp */ + evt.fifo = tx_fifo; + evt.event_type = FIFO_EVENT_SERVER_TX; + /* $$$$ for event logging */ + evt.enqueue_length = rv; + evt.event_id = serial_number++; + + unix_shared_memory_queue_add (utm->vpp_event_queue, (u8 *) &evt, + 0 /* do wait for mutex */); + } + } + } +} + +static void +uri_tcp_client_test (uri_tcp_test_main_t * utm) +{ + vl_api_connect_uri_t * cmp; + vl_api_disconnect_session_t *dmp; + session_t *connected_session; + int i; + + cmp = vl_msg_api_alloc (sizeof (*cmp)); + memset (cmp, 0, sizeof (*cmp)); + + cmp->_vl_msg_id = ntohs (VL_API_CONNECT_URI); + cmp->client_index = utm->my_client_index; + cmp->context = ntohl(0xfeedface); + memcpy (cmp->uri, utm->connect_uri, vec_len (utm->connect_uri)); + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *)&cmp); + + if (wait_for_state_change (utm, STATE_READY)) + { + return; + } + + /* Init test data */ + vec_validate (utm->connect_test_data, 64 * 1024 - 1); + for (i = 0; i < vec_len (utm->connect_test_data); i++) + utm->connect_test_data[i] = i & 0xff; + + /* Start reader thread */ + /* handle_connect_event_queue (utm); */ + + /* Start send */ + uri_tcp_connect_send (utm); + + /* Disconnect */ + connected_session = pool_elt_at_index(utm->sessions, + utm->connected_session_index); + dmp = vl_msg_api_alloc (sizeof (*dmp)); + memset (dmp, 0, sizeof (*dmp)); + dmp->_vl_msg_id = ntohs (VL_API_DISCONNECT_SESSION); + dmp->client_index = utm->my_client_index; + dmp->session_index = connected_session->vpp_session_index; + dmp->session_thread_index = connected_session->vpp_session_thread; + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *)&dmp); +} + +void +handle_fifo_event_server_rx (uri_tcp_test_main_t *utm, session_fifo_event_t * e) +{ + svm_fifo_t * rx_fifo, * tx_fifo; + int n_read; + + session_fifo_event_t evt; + unix_shared_memory_queue_t *q; + int rv, bytes; + + rx_fifo = e->fifo; + tx_fifo = utm->sessions[rx_fifo->client_session_index].server_tx_fifo; + + bytes = e->enqueue_length; + do + { + n_read = svm_fifo_dequeue_nowait (rx_fifo, 0, vec_len(utm->rx_buf), + utm->rx_buf); + + /* Reflect if a non-drop session */ + if (!utm->drop_packets && n_read > 0) + { + do + { + rv = svm_fifo_enqueue_nowait (tx_fifo, 0, n_read, utm->rx_buf); + } + while (rv == -2); + + /* Fabricate TX event, send to vpp */ + evt.fifo = tx_fifo; + evt.event_type = FIFO_EVENT_SERVER_TX; + /* $$$$ for event logging */ + evt.enqueue_length = n_read; + evt.event_id = e->event_id; + q = utm->vpp_event_queue; + unix_shared_memory_queue_add (q, (u8 *) &evt, 0 /* do wait for mutex */); + } + + if (n_read > 0) + bytes -= n_read; + } + while (n_read < 0 || bytes > 0); +} + +void +handle_event_queue (uri_tcp_test_main_t * utm) +{ + session_fifo_event_t _e, *e = &_e;; + + while (1) + { + unix_shared_memory_queue_sub (utm->our_event_queue, (u8 *)e, + 0 /* nowait */); + switch (e->event_type) + { + case FIFO_EVENT_SERVER_RX: + handle_fifo_event_server_rx (utm, e); + break; + + case FIFO_EVENT_SERVER_EXIT: + return; + + default: + clib_warning ("unknown event type %d", e->event_type); + break; + } + if (PREDICT_FALSE(utm->time_to_stop == 1)) + break; + if (PREDICT_FALSE(utm->time_to_print_stats == 1)) + { + utm->time_to_print_stats = 0; + fformat(stdout, "%d connections\n", pool_elts (utm->sessions)); + } + } +} + +static void +vl_api_bind_uri_reply_t_handler (vl_api_bind_uri_reply_t * mp) +{ + uri_tcp_test_main_t *utm = &uri_tcp_test_main; + svm_fifo_segment_create_args_t _a, *a = &_a; + int rv; + + if (mp->retval) + { + clib_warning("bind failed: %d", mp->retval); + return; + } + + if (mp->segment_name_length == 0) + { + clib_warning("segment_name_length zero"); + return; + } + + a->segment_name = (char *) mp->segment_name; + a->segment_size = mp->segment_size; + + ASSERT(mp->server_event_queue_address); + + /* Attach to the segment vpp created */ + rv = svm_fifo_segment_attach (a); + if (rv) + { + clib_warning("svm_fifo_segment_attach ('%s') failed", mp->segment_name); + return; + } + + utm->our_event_queue = + (unix_shared_memory_queue_t *) mp->server_event_queue_address; + + utm->state = STATE_READY; +} + +static void +vl_api_connect_uri_reply_t_handler (vl_api_connect_uri_reply_t * mp) +{ + uri_tcp_test_main_t *utm = &uri_tcp_test_main; + svm_fifo_segment_create_args_t _a, *a = &_a; + session_t *session; + u32 session_index; + svm_fifo_t *rx_fifo, *tx_fifo; + int rv; + + if (mp->retval) + { + clib_warning ("connection failed with code: %d", mp->retval); + utm->state = STATE_FAILED; + return; + } + /* + * Attatch to segment + */ + + if (mp->segment_name_length == 0) + { + clib_warning ("segment_name_length zero"); + utm->state = STATE_FAILED; + return; + } + + a->segment_name = (char *) mp->segment_name; + a->segment_size = mp->segment_size; + + ASSERT(mp->client_event_queue_address); + + /* Attach to the segment vpp created */ + rv = svm_fifo_segment_attach (a); + if (rv) + { + clib_warning ("svm_fifo_segment_attach ('%s') failed", + mp->segment_name); + return; + } + + /* + * Save the queues + */ + + utm->our_event_queue = (unix_shared_memory_queue_t *) + mp->client_event_queue_address; + + utm->vpp_event_queue = (unix_shared_memory_queue_t *) + mp->vpp_event_queue_address; + + /* + * Setup session + */ + + pool_get (utm->sessions, session); + session_index = session - utm->sessions; + + rx_fifo = (svm_fifo_t *)mp->server_rx_fifo; + rx_fifo->client_session_index = session_index; + tx_fifo = (svm_fifo_t *)mp->server_tx_fifo; + tx_fifo->client_session_index = session_index; + + session->server_rx_fifo = rx_fifo; + session->server_tx_fifo = tx_fifo; + session->vpp_session_index = mp->session_index; + session->vpp_session_thread = mp->session_thread_index; + + /* Save handle */ + utm->connected_session_index = session_index; + + utm->state = STATE_READY; +} + +void +uri_tcp_bind (uri_tcp_test_main_t *utm) +{ + vl_api_bind_uri_t * bmp; + u32 fifo_size = 3 << 20; + bmp = vl_msg_api_alloc (sizeof (*bmp)); + memset (bmp, 0, sizeof (*bmp)); + + bmp->_vl_msg_id = ntohs (VL_API_BIND_URI); + bmp->client_index = utm->my_client_index; + bmp->context = ntohl(0xfeedface); + bmp->initial_segment_size = 256<<20; /* size of initial segment */ + bmp->options[SESSION_OPTIONS_FLAGS] = + SESSION_OPTIONS_FLAGS_USE_FIFO | SESSION_OPTIONS_FLAGS_ADD_SEGMENT; + bmp->options[SESSION_OPTIONS_RX_FIFO_SIZE] = fifo_size; + bmp->options[SESSION_OPTIONS_TX_FIFO_SIZE] = fifo_size; + bmp->options[SESSION_OPTIONS_ADD_SEGMENT_SIZE] = 128<<20; + memcpy (bmp->uri, utm->uri, vec_len (utm->uri)); + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *)&bmp); +} + +static void +vl_api_unbind_uri_reply_t_handler (vl_api_unbind_uri_reply_t *mp) +{ + uri_tcp_test_main_t *utm = &uri_tcp_test_main; + + if (mp->retval != 0) + clib_warning ("returned %d", ntohl(mp->retval)); + + utm->state = STATE_START; +} + +void +uri_tcp_unbind (uri_tcp_test_main_t *utm) +{ + vl_api_unbind_uri_t * ump; + + ump = vl_msg_api_alloc (sizeof (*ump)); + memset (ump, 0, sizeof (*ump)); + + ump->_vl_msg_id = ntohs (VL_API_UNBIND_URI); + ump->client_index = utm->my_client_index; + memcpy (ump->uri, utm->uri, vec_len (utm->uri)); + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *)&ump); +} + +static void +vl_api_accept_session_t_handler (vl_api_accept_session_t * mp) +{ + uri_tcp_test_main_t *utm = &uri_tcp_test_main; + vl_api_accept_session_reply_t *rmp; + svm_fifo_t * rx_fifo, * tx_fifo; + session_t * session; + static f64 start_time; + u64 key; + u32 session_index; + + if (start_time == 0.0) + start_time = clib_time_now (&utm->clib_time); + + utm->vpp_event_queue = (unix_shared_memory_queue_t *) + mp->vpp_event_queue_address; + + /* Allocate local session and set it up */ + pool_get (utm->sessions, session); + session_index = session - utm->sessions; + + rx_fifo = (svm_fifo_t *)mp->server_rx_fifo; + rx_fifo->client_session_index = session_index; + tx_fifo = (svm_fifo_t *)mp->server_tx_fifo; + tx_fifo->client_session_index = session_index; + + session->server_rx_fifo = rx_fifo; + session->server_tx_fifo = tx_fifo; + + /* Add it to lookup table */ + key = (((u64)mp->session_thread_index) << 32) | (u64)mp->session_index; + hash_set (utm->session_index_by_vpp_handles, key, session_index); + + utm->state = STATE_READY; + + /* Stats printing */ + if (pool_elts (utm->sessions) && (pool_elts(utm->sessions) % 20000) == 0) + { + f64 now = clib_time_now (&utm->clib_time); + fformat (stdout, "%d active sessions in %.2f seconds, %.2f/sec...\n", + pool_elts(utm->sessions), now - start_time, + (f64)pool_elts(utm->sessions) / (now - start_time)); + } + + /* Send accept reply to vpp */ + rmp = vl_msg_api_alloc (sizeof (*rmp)); + memset (rmp, 0, sizeof (*rmp)); + rmp->_vl_msg_id = ntohs (VL_API_ACCEPT_SESSION_REPLY); + rmp->session_type = mp->session_type; + rmp->session_index = mp->session_index; + rmp->session_thread_index = mp->session_thread_index; + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *)&rmp); +} + +void +uri_tcp_server_test (uri_tcp_test_main_t * utm) +{ + + /* Bind to uri */ + uri_tcp_bind (utm); + + if (wait_for_state_change (utm, STATE_READY)) + { + clib_warning ("timeout waiting for STATE_READY"); + return; + } + + /* Enter handle event loop */ + handle_event_queue (utm); + + /* Cleanup */ + uri_tcp_unbind (utm); + + if (wait_for_state_change (utm, STATE_START)) + { + clib_warning ("timeout waiting for STATE_START"); + return; + } + + fformat (stdout, "Test complete...\n"); +} + +#define foreach_uri_msg \ +_(BIND_URI_REPLY, bind_uri_reply) \ +_(UNBIND_URI_REPLY, unbind_uri_reply) \ +_(ACCEPT_SESSION, accept_session) \ +_(CONNECT_URI_REPLY, connect_uri_reply) \ +_(DISCONNECT_SESSION, disconnect_session) \ +_(RESET_SESSION, reset_session) \ +_(MAP_ANOTHER_SEGMENT, map_another_segment) + +void +uri_api_hookup (uri_tcp_test_main_t * utm) +{ +#define _(N,n) \ + vl_msg_api_set_handlers(VL_API_##N, #n, \ + vl_api_##n##_t_handler, \ + vl_noop_handler, \ + vl_api_##n##_t_endian, \ + vl_api_##n##_t_print, \ + sizeof(vl_api_##n##_t), 1); + foreach_uri_msg; +#undef _ +} + +int +main (int argc, char **argv) +{ + uri_tcp_test_main_t *utm = &uri_tcp_test_main; + unformat_input_t _argv, *a = &_argv; + u8 *chroot_prefix; + u8 *heap; + u8 * bind_name = (u8 *) "tcp://0.0.0.0/1234"; + u32 tmp; + mheap_t *h; + session_t * session; + int i; + int i_am_master = 1, drop_packets = 0; + + clib_mem_init (0, 256 << 20); + + heap = clib_mem_get_per_cpu_heap (); + h = mheap_header (heap); + + /* make the main heap thread-safe */ + h->flags |= MHEAP_FLAG_THREAD_SAFE; + + vec_validate (utm->rx_buf, 65536); + + utm->session_index_by_vpp_handles = + hash_create (0, sizeof(uword)); + + utm->my_pid = getpid(); + utm->configured_segment_size = 1<<20; + + clib_time_init (&utm->clib_time); + init_error_string_table (utm); + svm_fifo_segment_init(0x200000000ULL, 20); + unformat_init_command_line (a, argv); + + while (unformat_check_input (a) != UNFORMAT_END_OF_INPUT) + { + if (unformat (a, "chroot prefix %s", &chroot_prefix)) + { + vl_set_memory_root_path ((char *) chroot_prefix); + } + else if (unformat (a, "uri %s", &bind_name)) + ; + else if (unformat (a, "segment-size %dM", &tmp)) + utm->configured_segment_size = tmp<<20; + else if (unformat (a, "segment-size %dG", &tmp)) + utm->configured_segment_size = tmp<<30; + else if (unformat (a, "master")) + i_am_master = 1; + else if (unformat (a, "slave")) + i_am_master = 0; + else if (unformat (a, "drop")) + drop_packets = 1; + else + { + fformat (stderr, "%s: usage [master|slave]\n"); + exit (1); + } + } + + utm->uri = format (0, "%s%c", bind_name, 0); + utm->i_am_master = i_am_master; + utm->segment_main = &svm_fifo_segment_main; + utm->drop_packets = drop_packets; + + utm->connect_uri = format (0, "tcp://6.0.1.2/1234%c", 0); + + setup_signal_handlers(); + uri_api_hookup (utm); + + if (connect_to_vpp (i_am_master? "uri_tcp_server":"uri_tcp_client") < 0) + { + svm_region_exit (); + fformat (stderr, "Couldn't connect to vpe, exiting...\n"); + exit (1); + } + + if (i_am_master == 0) + { + uri_tcp_client_test (utm); + exit (0); + } + + /* $$$$ hack preallocation */ + for (i = 0; i < 200000; i++) + { + pool_get (utm->sessions, session); + memset (session, 0, sizeof (*session)); + } + for (i = 0; i < 200000; i++) + pool_put_index (utm->sessions, i); + + uri_tcp_server_test (utm); + + vl_client_disconnect_from_vlib (); + exit (0); +} diff --git a/src/uri/uri_udp_test.c b/src/uri/uri_udp_test.c new file mode 100644 index 00000000000..6f5284c9338 --- /dev/null +++ b/src/uri/uri_udp_test.c @@ -0,0 +1,553 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <stdio.h> +#include <setjmp.h> +#include <signal.h> +#include <vppinfra/clib.h> +#include <vppinfra/format.h> +#include <vppinfra/error.h> +#include <vppinfra/time.h> +#include <vppinfra/macros.h> +#include <vnet/vnet.h> +#include <vlib/vlib.h> +#include <vlib/unix/unix.h> +#include <vlibapi/api.h> +#include <vlibmemory/api.h> +#include <vpp-api/vpe_msg_enum.h> +#include <svm_fifo_segment.h> + +#include <vnet/uri/uri.h> + +#define vl_typedefs /* define message structures */ +#include <vpp-api/vpe_all_api_h.h> +#undef vl_typedefs + +/* declare message handlers for each api */ + +#define vl_endianfun /* define message structures */ +#include <vpp-api/vpe_all_api_h.h> +#undef vl_endianfun + +/* instantiate all the print functions we know about */ +#define vl_print(handle, ...) +#define vl_printfun +#include <vpp-api/vpe_all_api_h.h> +#undef vl_printfun + +/* Satisfy external references when not linking with -lvlib */ +vlib_main_t vlib_global_main; +vlib_main_t **vlib_mains; + +typedef enum +{ + STATE_START, + STATE_READY, + STATE_DISCONNECTING, +} connection_state_t; + +typedef struct +{ + svm_fifo_t *server_rx_fifo; + svm_fifo_t *server_tx_fifo; +} session_t; + +typedef struct +{ + /* vpe input queue */ + unix_shared_memory_queue_t *vl_input_queue; + + /* API client handle */ + u32 my_client_index; + + /* The URI we're playing with */ + u8 *uri; + + /* Session pool */ + session_t *sessions; + + /* Hash table for disconnect processing */ + uword *session_index_by_vpp_handles; + + /* fifo segment */ + svm_fifo_segment_private_t *seg; + + /* intermediate rx buffer */ + u8 *rx_buf; + + /* Our event queue */ + unix_shared_memory_queue_t *our_event_queue; + + /* $$$ single thread only for the moment */ + unix_shared_memory_queue_t *vpp_event_queue; + + /* For deadman timers */ + clib_time_t clib_time; + + /* State of the connection, shared between msg RX thread and main thread */ + volatile connection_state_t state; + + volatile int time_to_stop; + volatile int time_to_print_stats; + + /* VNET_API_ERROR_FOO -> "Foo" hash table */ + uword *error_string_by_error_number; +} uri_udp_test_main_t; + +#if CLIB_DEBUG > 0 +#define NITER 1000 +#else +#define NITER 1000000 +#endif + +uri_udp_test_main_t uri_udp_test_main; + +static void +stop_signal (int signum) +{ + uri_udp_test_main_t *um = &uri_udp_test_main; + + um->time_to_stop = 1; +} + +static void +stats_signal (int signum) +{ + uri_udp_test_main_t *um = &uri_udp_test_main; + + um->time_to_print_stats = 1; +} + +static clib_error_t * +setup_signal_handlers (void) +{ + signal (SIGINT, stats_signal); + signal (SIGQUIT, stop_signal); + signal (SIGTERM, stop_signal); + + return 0; +} + +u8 * +format_api_error (u8 * s, va_list * args) +{ + uri_udp_test_main_t *utm = va_arg (*args, uri_udp_test_main_t *); + i32 error = va_arg (*args, u32); + uword *p; + + p = hash_get (utm->error_string_by_error_number, -error); + + if (p) + s = format (s, "%s", p[0]); + else + s = format (s, "%d", error); + return s; +} + +int +wait_for_state_change (uri_udp_test_main_t * utm, connection_state_t state) +{ + f64 timeout = clib_time_now (&utm->clib_time) + 5.0; + + while (clib_time_now (&utm->clib_time) < timeout) + { + if (utm->state == state) + return 0; + } + return -1; +} + +static void +vl_api_bind_uri_reply_t_handler (vl_api_bind_uri_reply_t * mp) +{ + uri_udp_test_main_t *utm = &uri_udp_test_main; + svm_fifo_segment_create_args_t _a, *a = &_a; + int rv; + + if (mp->segment_name_length == 0) + { + clib_warning ("segment_name_length zero"); + return; + } + + a->segment_name = (char *) mp->segment_name; + + /* Attach to the segment vpp created */ + rv = svm_fifo_segment_attach (a); + if (rv) + { + clib_warning ("sm_fifo_segment_create ('%s') failed", mp->segment_name); + return; + } + + utm->our_event_queue = (unix_shared_memory_queue_t *) + mp->server_event_queue_address; + + utm->state = STATE_READY; +} + +static void +vl_api_unbind_uri_reply_t_handler (vl_api_unbind_uri_reply_t * mp) +{ + uri_udp_test_main_t *utm = &uri_udp_test_main; + + if (mp->retval != 0) + clib_warning ("returned %d", ntohl (mp->retval)); + + utm->state = STATE_START; +} + +static void +vl_api_accept_session_t_handler (vl_api_accept_session_t * mp) +{ + uri_udp_test_main_t *utm = &uri_udp_test_main; + vl_api_accept_session_reply_t *rmp; + svm_fifo_t *rx_fifo, *tx_fifo; + session_t *session; + static f64 start_time; + u64 key; + + if (start_time == 0.0) + start_time = clib_time_now (&utm->clib_time); + + utm->vpp_event_queue = (unix_shared_memory_queue_t *) + mp->vpp_event_queue_address; + + pool_get (utm->sessions, session); + + rx_fifo = (svm_fifo_t *) mp->server_rx_fifo; + rx_fifo->client_session_index = session - utm->sessions; + tx_fifo = (svm_fifo_t *) mp->server_tx_fifo; + tx_fifo->client_session_index = session - utm->sessions; + + session->server_rx_fifo = rx_fifo; + session->server_tx_fifo = tx_fifo; + + key = (((u64) mp->session_thread_index) << 32) | (u64) mp->session_index; + + hash_set (utm->session_index_by_vpp_handles, key, session - utm->sessions); + + utm->state = STATE_READY; + + if (pool_elts (utm->sessions) && (pool_elts (utm->sessions) % 20000) == 0) + { + f64 now = clib_time_now (&utm->clib_time); + fformat (stdout, "%d active sessions in %.2f seconds, %.2f/sec...\n", + pool_elts (utm->sessions), now - start_time, + (f64) pool_elts (utm->sessions) / (now - start_time)); + } + + rmp = vl_msg_api_alloc (sizeof (*rmp)); + memset (rmp, 0, sizeof (*rmp)); + rmp->_vl_msg_id = ntohs (VL_API_ACCEPT_SESSION_REPLY); + rmp->session_type = mp->session_type; + rmp->session_index = mp->session_index; + rmp->session_thread_index = mp->session_thread_index; + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *) & rmp); +} + +static void +vl_api_disconnect_session_t_handler (vl_api_disconnect_session_t * mp) +{ + uri_udp_test_main_t *utm = &uri_udp_test_main; + session_t *session; + vl_api_disconnect_session_reply_t *rmp; + uword *p; + int rv = 0; + u64 key; + + key = (((u64) mp->session_thread_index) << 32) | (u64) mp->session_index; + + p = hash_get (utm->session_index_by_vpp_handles, key); + + if (p) + { + session = pool_elt_at_index (utm->sessions, p[0]); + hash_unset (utm->session_index_by_vpp_handles, key); + pool_put (utm->sessions, session); + } + else + { + clib_warning ("couldn't find session key %llx", key); + rv = -11; + } + + rmp = vl_msg_api_alloc (sizeof (*rmp)); + memset (rmp, 0, sizeof (*rmp)); + rmp->_vl_msg_id = ntohs (VL_API_DISCONNECT_SESSION_REPLY); + rmp->retval = rv; + rmp->session_index = mp->session_index; + rmp->session_thread_index = mp->session_thread_index; + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *) & rmp); +} + +#define foreach_uri_msg \ +_(BIND_URI_REPLY, bind_uri_reply) \ +_(UNBIND_URI_REPLY, unbind_uri_reply) \ +_(ACCEPT_SESSION, accept_session) \ +_(DISCONNECT_SESSION, disconnect_session) + +void +uri_api_hookup (uri_udp_test_main_t * utm) +{ +#define _(N,n) \ + vl_msg_api_set_handlers(VL_API_##N, #n, \ + vl_api_##n##_t_handler, \ + vl_noop_handler, \ + vl_api_##n##_t_endian, \ + vl_api_##n##_t_print, \ + sizeof(vl_api_##n##_t), 1); + foreach_uri_msg; +#undef _ + +} + + +int +connect_to_vpp (char *name) +{ + uri_udp_test_main_t *utm = &uri_udp_test_main; + api_main_t *am = &api_main; + + if (vl_client_connect_to_vlib ("/vpe-api", name, 32) < 0) + return -1; + + utm->vl_input_queue = am->shmem_hdr->vl_input_queue; + utm->my_client_index = am->my_client_index; + + return 0; +} + +void +vlib_cli_output (struct vlib_main_t *vm, char *fmt, ...) +{ + clib_warning ("BUG"); +} + +static void +init_error_string_table (uri_udp_test_main_t * utm) +{ + utm->error_string_by_error_number = hash_create (0, sizeof (uword)); + +#define _(n,v,s) hash_set (utm->error_string_by_error_number, -v, s); + foreach_vnet_api_error; +#undef _ + + hash_set (utm->error_string_by_error_number, 99, "Misc"); +} + +void +handle_fifo_event_server_rx (uri_udp_test_main_t * utm, + session_fifo_event_t * e) +{ + svm_fifo_t *rx_fifo, *tx_fifo; + int nbytes; + + session_fifo_event_t evt; + unix_shared_memory_queue_t *q; + int rv; + + rx_fifo = e->fifo; + tx_fifo = utm->sessions[rx_fifo->client_session_index].server_tx_fifo; + + do + { + nbytes = svm_fifo_dequeue_nowait (rx_fifo, 0, + vec_len (utm->rx_buf), utm->rx_buf); + } + while (nbytes <= 0); + do + { + rv = svm_fifo_enqueue_nowait (tx_fifo, 0, nbytes, utm->rx_buf); + } + while (rv == -2); + + /* Fabricate TX event, send to vpp */ + evt.fifo = tx_fifo; + evt.event_type = FIFO_EVENT_SERVER_TX; + /* $$$$ for event logging */ + evt.enqueue_length = nbytes; + evt.event_id = e->event_id; + q = utm->vpp_event_queue; + unix_shared_memory_queue_add (q, (u8 *) & evt, 0 /* do wait for mutex */ ); +} + +void +handle_event_queue (uri_udp_test_main_t * utm) +{ + session_fifo_event_t _e, *e = &_e;; + + while (1) + { + unix_shared_memory_queue_sub (utm->our_event_queue, (u8 *) e, + 0 /* nowait */ ); + switch (e->event_type) + { + case FIFO_EVENT_SERVER_RX: + handle_fifo_event_server_rx (utm, e); + break; + + case FIFO_EVENT_SERVER_EXIT: + return; + + default: + clib_warning ("unknown event type %d", e->event_type); + break; + } + if (PREDICT_FALSE (utm->time_to_stop == 1)) + break; + if (PREDICT_FALSE (utm->time_to_print_stats == 1)) + { + utm->time_to_print_stats = 0; + fformat (stdout, "%d connections\n", pool_elts (utm->sessions)); + } + } +} + +void +uri_udp_test (uri_udp_test_main_t * utm) +{ + vl_api_bind_uri_t *bmp; + vl_api_unbind_uri_t *ump; + + bmp = vl_msg_api_alloc (sizeof (*bmp)); + memset (bmp, 0, sizeof (*bmp)); + + bmp->_vl_msg_id = ntohs (VL_API_BIND_URI); + bmp->client_index = utm->my_client_index; + bmp->context = ntohl (0xfeedface); + bmp->segment_size = 2 << 30; + memcpy (bmp->uri, utm->uri, vec_len (utm->uri)); + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *) & bmp); + + if (wait_for_state_change (utm, STATE_READY)) + { + clib_warning ("timeout waiting for STATE_READY"); + return; + } + + handle_event_queue (utm); + + ump = vl_msg_api_alloc (sizeof (*ump)); + memset (ump, 0, sizeof (*ump)); + + ump->_vl_msg_id = ntohs (VL_API_UNBIND_URI); + ump->client_index = utm->my_client_index; + memcpy (ump->uri, utm->uri, vec_len (utm->uri)); + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *) & ump); + + if (wait_for_state_change (utm, STATE_START)) + { + clib_warning ("timeout waiting for STATE_START"); + return; + } + + fformat (stdout, "Test complete...\n"); +} + +int +main (int argc, char **argv) +{ + uri_udp_test_main_t *utm = &uri_udp_test_main; + unformat_input_t _argv, *a = &_argv; + u8 *chroot_prefix; + u8 *heap; + u8 *bind_name = (u8 *) "udp4:1234"; + mheap_t *h; + session_t *session; + int i; + + clib_mem_init (0, 256 << 20); + + heap = clib_mem_get_per_cpu_heap (); + h = mheap_header (heap); + + /* make the main heap thread-safe */ + h->flags |= MHEAP_FLAG_THREAD_SAFE; + + vec_validate (utm->rx_buf, 8192); + + utm->session_index_by_vpp_handles = hash_create (0, sizeof (uword)); + + clib_time_init (&utm->clib_time); + init_error_string_table (utm); + svm_fifo_segment_init (0x200000000ULL, 20); + unformat_init_command_line (a, argv); + + while (unformat_check_input (a) != UNFORMAT_END_OF_INPUT) + { + if (unformat (a, "chroot prefix %s", &chroot_prefix)) + { + vl_set_memory_root_path ((char *) chroot_prefix); + } + else if (unformat (a, "uri %s", &bind_name)) + ; + else + { + fformat (stderr, "%s: usage [master|slave]\n"); + exit (1); + } + } + + utm->uri = format (0, "%s%c", bind_name, 0); + + setup_signal_handlers (); + + uri_api_hookup (utm); + + if (connect_to_vpp ("uri_udp_test") < 0) + { + svm_region_exit (); + fformat (stderr, "Couldn't connect to vpe, exiting...\n"); + exit (1); + } + + /* $$$$ hack preallocation */ + for (i = 0; i < 200000; i++) + { + pool_get (utm->sessions, session); + memset (session, 0, sizeof (*session)); + } + for (i = 0; i < 200000; i++) + pool_put_index (utm->sessions, i); + + uri_udp_test (utm); + + vl_client_disconnect_from_vlib (); + exit (0); +} + +#undef vl_api_version +#define vl_api_version(n,v) static u32 vpe_api_version = v; +#include <vpp-api/vpe.api.h> +#undef vl_api_version + +void +vl_client_add_api_signatures (vl_api_memclnt_create_t * mp) +{ + /* + * Send the main API signature in slot 0. This bit of code must + * match the checks in ../vpe/api/api.c: vl_msg_api_version_check(). + */ + mp->api_versions[0] = clib_host_to_net_u32 (vpe_api_version); +} + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/uri/uri_udp_test2.c b/src/uri/uri_udp_test2.c new file mode 100644 index 00000000000..ddfffaa6ff4 --- /dev/null +++ b/src/uri/uri_udp_test2.c @@ -0,0 +1,954 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <stdio.h> +#include <setjmp.h> +#include <signal.h> +#include <vppinfra/clib.h> +#include <vppinfra/format.h> +#include <vppinfra/error.h> +#include <vppinfra/time.h> +#include <vppinfra/macros.h> +#include <vnet/vnet.h> +#include <vlib/vlib.h> +#include <vlib/unix/unix.h> +#include <vlibapi/api.h> +#include <vlibmemory/api.h> +#include <vpp/api/vpe_msg_enum.h> +#include <svm/svm_fifo_segment.h> +#include <pthread.h> + +#include "../vnet/session/application_interface.h" + +#define vl_typedefs /* define message structures */ +#include <vpp/api/vpe_all_api_h.h> +#undef vl_typedefs + +/* declare message handlers for each api */ + +#define vl_endianfun /* define message structures */ +#include <vpp/api/vpe_all_api_h.h> +#undef vl_endianfun + +/* instantiate all the print functions we know about */ +#define vl_print(handle, ...) +#define vl_printfun +#include <vpp/api/vpe_all_api_h.h> +#undef vl_printfun + +/* Satisfy external references when not linking with -lvlib */ +vlib_main_t vlib_global_main; +vlib_main_t **vlib_mains; + +typedef enum +{ + STATE_START, + STATE_READY, + STATE_DISCONNECTING, +} connection_state_t; + +typedef struct +{ + svm_fifo_t *server_rx_fifo; + svm_fifo_t *server_tx_fifo; +} session_t; + +typedef struct +{ + /* vpe input queue */ + unix_shared_memory_queue_t *vl_input_queue; + + /* API client handle */ + u32 my_client_index; + + /* The URI we're playing with */ + u8 *uri; + + /* Session pool */ + session_t *sessions; + + /* Hash table for disconnect processing */ + uword *session_index_by_vpp_handles; + + /* fifo segment */ + svm_fifo_segment_private_t *seg; + + /* intermediate rx buffer */ + u8 *rx_buf; + + /* URI for connect */ + u8 *connect_uri; + + int i_am_master; + + /* Our event queue */ + unix_shared_memory_queue_t *our_event_queue; + + /* $$$ single thread only for the moment */ + unix_shared_memory_queue_t *vpp_event_queue; + + /* $$$$ hack: cut-through session index */ + volatile u32 cut_through_session_index; + + /* unique segment name counter */ + u32 unique_segment_index; + + pid_t my_pid; + + /* pthread handle */ + pthread_t cut_through_thread_handle; + + /* For deadman timers */ + clib_time_t clib_time; + + /* State of the connection, shared between msg RX thread and main thread */ + volatile connection_state_t state; + + volatile int time_to_stop; + volatile int time_to_print_stats; + + u32 configured_segment_size; + + /* VNET_API_ERROR_FOO -> "Foo" hash table */ + uword *error_string_by_error_number; + + /* convenience */ + svm_fifo_segment_main_t *segment_main; + +} uri_udp_test_main_t; + +#if CLIB_DEBUG > 0 +#define NITER 10000 +#else +#define NITER 4000000 +#endif + +uri_udp_test_main_t uri_udp_test_main; + +static void +stop_signal (int signum) +{ + uri_udp_test_main_t *um = &uri_udp_test_main; + + um->time_to_stop = 1; +} + +static void +stats_signal (int signum) +{ + uri_udp_test_main_t *um = &uri_udp_test_main; + + um->time_to_print_stats = 1; +} + +static clib_error_t * +setup_signal_handlers (void) +{ + signal (SIGINT, stats_signal); + signal (SIGQUIT, stop_signal); + signal (SIGTERM, stop_signal); + + return 0; +} + +u8 * +format_api_error (u8 * s, va_list * args) +{ + uri_udp_test_main_t *utm = va_arg (*args, uri_udp_test_main_t *); + i32 error = va_arg (*args, u32); + uword *p; + + p = hash_get (utm->error_string_by_error_number, -error); + + if (p) + s = format (s, "%s", p[0]); + else + s = format (s, "%d", error); + return s; +} + +int +wait_for_state_change (uri_udp_test_main_t * utm, connection_state_t state) +{ +#if CLIB_DEBUG > 0 +#define TIMEOUT 600.0 +#else +#define TIMEOUT 600.0 +#endif + + f64 timeout = clib_time_now (&utm->clib_time) + TIMEOUT; + + while (clib_time_now (&utm->clib_time) < timeout) + { + if (utm->state == state) + return 0; + } + return -1; +} + +u64 server_bytes_received, server_bytes_sent; + +static void * +cut_through_thread_fn (void *arg) +{ + session_t *s; + svm_fifo_t *rx_fifo; + svm_fifo_t *tx_fifo; + u8 *my_copy_buffer = 0; + uri_udp_test_main_t *utm = &uri_udp_test_main; + i32 actual_transfer; + int rv; + u32 buffer_offset; + + while (utm->cut_through_session_index == ~0) + ; + + s = pool_elt_at_index (utm->sessions, utm->cut_through_session_index); + + rx_fifo = s->server_rx_fifo; + tx_fifo = s->server_tx_fifo; + + vec_validate (my_copy_buffer, 64 * 1024 - 1); + + while (true) + { + /* We read from the tx fifo and write to the rx fifo */ + do + { + actual_transfer = svm_fifo_dequeue_nowait (tx_fifo, 0, + vec_len (my_copy_buffer), + my_copy_buffer); + } + while (actual_transfer <= 0); + + server_bytes_received += actual_transfer; + + buffer_offset = 0; + while (actual_transfer > 0) + { + rv = svm_fifo_enqueue_nowait (rx_fifo, 0, actual_transfer, + my_copy_buffer + buffer_offset); + if (rv > 0) + { + actual_transfer -= rv; + buffer_offset += rv; + server_bytes_sent += rv; + } + + } + if (PREDICT_FALSE (utm->time_to_stop)) + break; + } + + pthread_exit (0); +} + +static void +uri_udp_slave_test (uri_udp_test_main_t * utm) +{ + vl_api_connect_uri_t *cmp; + int i; + u8 *test_data = 0; + u64 bytes_received = 0, bytes_sent = 0; + i32 bytes_to_read; + int rv; + int mypid = getpid (); + f64 before, after, delta, bytes_per_second; + session_t *session; + svm_fifo_t *rx_fifo, *tx_fifo; + int buffer_offset, bytes_to_send = 0; + + vec_validate (test_data, 64 * 1024 - 1); + for (i = 0; i < vec_len (test_data); i++) + test_data[i] = i & 0xff; + + cmp = vl_msg_api_alloc (sizeof (*cmp)); + memset (cmp, 0, sizeof (*cmp)); + + cmp->_vl_msg_id = ntohs (VL_API_CONNECT_URI); + cmp->client_index = utm->my_client_index; + cmp->context = ntohl (0xfeedface); + memcpy (cmp->uri, utm->connect_uri, vec_len (utm->connect_uri)); + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *) & cmp); + + if (wait_for_state_change (utm, STATE_READY)) + { + clib_warning ("timeout waiting for STATE_READY"); + return; + } + + session = pool_elt_at_index (utm->sessions, utm->cut_through_session_index); + rx_fifo = session->server_rx_fifo; + tx_fifo = session->server_tx_fifo; + + before = clib_time_now (&utm->clib_time); + + vec_validate (utm->rx_buf, vec_len (test_data) - 1); + + for (i = 0; i < NITER; i++) + { + bytes_to_send = vec_len (test_data); + buffer_offset = 0; + while (bytes_to_send > 0) + { + rv = svm_fifo_enqueue_nowait (tx_fifo, mypid, + bytes_to_send, + test_data + buffer_offset); + + if (rv > 0) + { + bytes_to_send -= rv; + buffer_offset += rv; + bytes_sent += rv; + } + } + + bytes_to_read = svm_fifo_max_dequeue (rx_fifo); + + bytes_to_read = vec_len (utm->rx_buf) > bytes_to_read ? + bytes_to_read : vec_len (utm->rx_buf); + + buffer_offset = 0; + while (bytes_to_read > 0) + { + rv = svm_fifo_dequeue_nowait (rx_fifo, mypid, + bytes_to_read, + utm->rx_buf + buffer_offset); + if (rv > 0) + { + bytes_to_read -= rv; + buffer_offset += rv; + bytes_received += rv; + } + } + } + while (bytes_received < bytes_sent) + { + rv = svm_fifo_dequeue_nowait (rx_fifo, mypid, + vec_len (utm->rx_buf), utm->rx_buf); + if (rv > 0) + { +#if CLIB_DEBUG > 0 + int j; + for (j = 0; j < rv; j++) + { + if (utm->rx_buf[j] != ((bytes_received + j) & 0xff)) + { + clib_warning ("error at byte %lld, 0x%x not 0x%x", + bytes_received + j, + utm->rx_buf[j], + ((bytes_received + j) & 0xff)); + } + } +#endif + bytes_received += (u64) rv; + } + } + + after = clib_time_now (&utm->clib_time); + delta = after - before; + bytes_per_second = 0.0; + + if (delta > 0.0) + bytes_per_second = (f64) bytes_received / delta; + + fformat (stdout, + "Done: %lld recv bytes in %.2f seconds, %.2f bytes/sec...\n\n", + bytes_received, delta, bytes_per_second); + fformat (stdout, + "Done: %lld sent bytes in %.2f seconds, %.2f bytes/sec...\n\n", + bytes_sent, delta, bytes_per_second); + fformat (stdout, + "client -> server -> client round trip: %.2f Gbit/sec \n\n", + (bytes_per_second * 8.0) / 1e9); +} + +static void +vl_api_bind_uri_reply_t_handler (vl_api_bind_uri_reply_t * mp) +{ + uri_udp_test_main_t *utm = &uri_udp_test_main; + svm_fifo_segment_create_args_t _a, *a = &_a; + int rv; + + if (mp->segment_name_length == 0) + { + clib_warning ("segment_name_length zero"); + return; + } + + a->segment_name = (char *) mp->segment_name; + a->segment_size = mp->segment_size; + + ASSERT (mp->server_event_queue_address); + + /* Attach to the segment vpp created */ + rv = svm_fifo_segment_attach (a); + if (rv) + { + clib_warning ("svm_fifo_segment_attach ('%s') failed", + mp->segment_name); + return; + } + + utm->our_event_queue = (unix_shared_memory_queue_t *) + mp->server_event_queue_address; + + utm->state = STATE_READY; +} + +static void +vl_api_map_another_segment_t_handler (vl_api_map_another_segment_t * mp) +{ + svm_fifo_segment_create_args_t _a, *a = &_a; + int rv; + + a->segment_name = (char *) mp->segment_name; + a->segment_size = mp->segment_size; + /* Attach to the segment vpp created */ + rv = svm_fifo_segment_attach (a); + if (rv) + { + clib_warning ("svm_fifo_segment_attach ('%s') failed", + mp->segment_name); + return; + } + clib_warning ("Mapped new segment '%s' size %d", mp->segment_name, + mp->segment_size); +} + +static void +vl_api_connect_uri_t_handler (vl_api_connect_uri_t * mp) +{ + u32 segment_index; + uri_udp_test_main_t *utm = &uri_udp_test_main; + svm_fifo_segment_main_t *sm = &svm_fifo_segment_main; + svm_fifo_segment_create_args_t _a, *a = &_a; + svm_fifo_segment_private_t *seg; + unix_shared_memory_queue_t *client_q; + vl_api_connect_uri_reply_t *rmp; + session_t *session; + int rv = 0; + + /* Create the segment */ + a->segment_name = (char *) format (0, "%d:segment%d%c", utm->my_pid, + utm->unique_segment_index++, 0); + a->segment_size = utm->configured_segment_size; + + rv = svm_fifo_segment_create (a); + if (rv) + { + clib_warning ("sm_fifo_segment_create ('%s') failed", a->segment_name); + rv = VNET_API_ERROR_URI_FIFO_CREATE_FAILED; + goto send_reply; + } + + vec_add2 (utm->seg, seg, 1); + + segment_index = vec_len (sm->segments) - 1; + + memcpy (seg, sm->segments + segment_index, sizeof (utm->seg[0])); + + pool_get (utm->sessions, session); + + /* + * By construction the master's idea of the rx fifo ends up in + * fsh->fifos[0], and the master's idea of the tx fifo ends up in + * fsh->fifos[1]. + */ + session->server_rx_fifo = svm_fifo_segment_alloc_fifo (utm->seg, + 128 * 1024); + ASSERT (session->server_rx_fifo); + + session->server_tx_fifo = svm_fifo_segment_alloc_fifo (utm->seg, + 128 * 1024); + ASSERT (session->server_tx_fifo); + + session->server_rx_fifo->server_session_index = session - utm->sessions; + session->server_tx_fifo->server_session_index = session - utm->sessions; + utm->cut_through_session_index = session - utm->sessions; + + rv = pthread_create (&utm->cut_through_thread_handle, + NULL /*attr */ , cut_through_thread_fn, 0); + if (rv) + { + clib_warning ("pthread_create returned %d", rv); + rv = VNET_API_ERROR_SYSCALL_ERROR_1; + } + +send_reply: + rmp = vl_msg_api_alloc (sizeof (*rmp)); + memset (rmp, 0, sizeof (*rmp)); + + rmp->_vl_msg_id = ntohs (VL_API_CONNECT_URI_REPLY); + rmp->context = mp->context; + rmp->retval = ntohl (rv); + rmp->segment_name_length = vec_len (a->segment_name); + memcpy (rmp->segment_name, a->segment_name, vec_len (a->segment_name)); + + vec_free (a->segment_name); + + client_q = (unix_shared_memory_queue_t *) mp->client_queue_address; + vl_msg_api_send_shmem (client_q, (u8 *) & rmp); +} + +static void +vl_api_unbind_uri_reply_t_handler (vl_api_unbind_uri_reply_t * mp) +{ + uri_udp_test_main_t *utm = &uri_udp_test_main; + + if (mp->retval != 0) + clib_warning ("returned %d", ntohl (mp->retval)); + + utm->state = STATE_START; +} + +static void +vl_api_accept_session_t_handler (vl_api_accept_session_t * mp) +{ + uri_udp_test_main_t *utm = &uri_udp_test_main; + vl_api_accept_session_reply_t *rmp; + svm_fifo_t *rx_fifo, *tx_fifo; + session_t *session; + static f64 start_time; + u64 key; + + if (start_time == 0.0) + start_time = clib_time_now (&utm->clib_time); + + utm->vpp_event_queue = (unix_shared_memory_queue_t *) + mp->vpp_event_queue_address; + + pool_get (utm->sessions, session); + + rx_fifo = (svm_fifo_t *) mp->server_rx_fifo; + rx_fifo->client_session_index = session - utm->sessions; + tx_fifo = (svm_fifo_t *) mp->server_tx_fifo; + tx_fifo->client_session_index = session - utm->sessions; + + session->server_rx_fifo = rx_fifo; + session->server_tx_fifo = tx_fifo; + + key = (((u64) mp->session_thread_index) << 32) | (u64) mp->session_index; + + hash_set (utm->session_index_by_vpp_handles, key, session - utm->sessions); + + utm->state = STATE_READY; + + if (pool_elts (utm->sessions) && (pool_elts (utm->sessions) % 20000) == 0) + { + f64 now = clib_time_now (&utm->clib_time); + fformat (stdout, "%d active sessions in %.2f seconds, %.2f/sec...\n", + pool_elts (utm->sessions), now - start_time, + (f64) pool_elts (utm->sessions) / (now - start_time)); + } + + rmp = vl_msg_api_alloc (sizeof (*rmp)); + memset (rmp, 0, sizeof (*rmp)); + rmp->_vl_msg_id = ntohs (VL_API_ACCEPT_SESSION_REPLY); + rmp->session_type = mp->session_type; + rmp->session_index = mp->session_index; + rmp->session_thread_index = mp->session_thread_index; + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *) & rmp); +} + +static void +vl_api_disconnect_session_t_handler (vl_api_disconnect_session_t * mp) +{ + uri_udp_test_main_t *utm = &uri_udp_test_main; + session_t *session; + vl_api_disconnect_session_reply_t *rmp; + uword *p; + int rv = 0; + u64 key; + + key = (((u64) mp->session_thread_index) << 32) | (u64) mp->session_index; + + p = hash_get (utm->session_index_by_vpp_handles, key); + + if (p) + { + session = pool_elt_at_index (utm->sessions, p[0]); + hash_unset (utm->session_index_by_vpp_handles, key); + pool_put (utm->sessions, session); + } + else + { + clib_warning ("couldn't find session key %llx", key); + rv = -11; + } + + rmp = vl_msg_api_alloc (sizeof (*rmp)); + memset (rmp, 0, sizeof (*rmp)); + rmp->_vl_msg_id = ntohs (VL_API_DISCONNECT_SESSION_REPLY); + rmp->retval = rv; + rmp->session_index = mp->session_index; + rmp->session_thread_index = mp->session_thread_index; + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *) & rmp); +} + +static void +vl_api_connect_uri_reply_t_handler (vl_api_connect_uri_reply_t * mp) +{ + svm_fifo_segment_main_t *sm = &svm_fifo_segment_main; + uri_udp_test_main_t *utm = &uri_udp_test_main; + svm_fifo_segment_create_args_t _a, *a = &_a; + ssvm_shared_header_t *sh; + svm_fifo_segment_private_t *seg; + svm_fifo_segment_header_t *fsh; + session_t *session; + u32 segment_index; + int rv; + + ASSERT (utm->i_am_master == 0); + + if (mp->segment_name_length == 0) + { + clib_warning ("segment_name_length zero"); + return; + } + + memset (a, 0, sizeof (*a)); + + a->segment_name = (char *) mp->segment_name; + + sleep (1); + + rv = svm_fifo_segment_attach (a); + if (rv) + { + clib_warning ("sm_fifo_segment_create ('%v') failed", mp->segment_name); + return; + } + + segment_index = vec_len (sm->segments) - 1; + + vec_add2 (utm->seg, seg, 1); + + memcpy (seg, sm->segments + segment_index, sizeof (*seg)); + sh = seg->ssvm.sh; + fsh = (svm_fifo_segment_header_t *) sh->opaque[0]; + + while (vec_len (fsh->fifos) < 2) + sleep (1); + + pool_get (utm->sessions, session); + utm->cut_through_session_index = session - utm->sessions; + + session->server_rx_fifo = (svm_fifo_t *) fsh->fifos[0]; + ASSERT (session->server_rx_fifo); + session->server_tx_fifo = (svm_fifo_t *) fsh->fifos[1]; + ASSERT (session->server_tx_fifo); + + /* security: could unlink /dev/shm/<mp->segment_name> here, maybe */ + + utm->state = STATE_READY; +} + +#define foreach_uri_msg \ +_(BIND_URI_REPLY, bind_uri_reply) \ +_(CONNECT_URI, connect_uri) \ +_(CONNECT_URI_REPLY, connect_uri_reply) \ +_(UNBIND_URI_REPLY, unbind_uri_reply) \ +_(ACCEPT_SESSION, accept_session) \ +_(DISCONNECT_SESSION, disconnect_session) \ +_(MAP_ANOTHER_SEGMENT, map_another_segment) + +void +uri_api_hookup (uri_udp_test_main_t * utm) +{ +#define _(N,n) \ + vl_msg_api_set_handlers(VL_API_##N, #n, \ + vl_api_##n##_t_handler, \ + vl_noop_handler, \ + vl_api_##n##_t_endian, \ + vl_api_##n##_t_print, \ + sizeof(vl_api_##n##_t), 1); + foreach_uri_msg; +#undef _ + +} + + +int +connect_to_vpp (char *name) +{ + uri_udp_test_main_t *utm = &uri_udp_test_main; + api_main_t *am = &api_main; + + if (vl_client_connect_to_vlib ("/vpe-api", name, 32) < 0) + return -1; + + utm->vl_input_queue = am->shmem_hdr->vl_input_queue; + utm->my_client_index = am->my_client_index; + + return 0; +} + +void +vlib_cli_output (struct vlib_main_t *vm, char *fmt, ...) +{ + clib_warning ("BUG"); +} + +static void +init_error_string_table (uri_udp_test_main_t * utm) +{ + utm->error_string_by_error_number = hash_create (0, sizeof (uword)); + +#define _(n,v,s) hash_set (utm->error_string_by_error_number, -v, s); + foreach_vnet_api_error; +#undef _ + + hash_set (utm->error_string_by_error_number, 99, "Misc"); +} + +void +handle_fifo_event_server_rx (uri_udp_test_main_t * utm, + session_fifo_event_t * e) +{ + svm_fifo_t *rx_fifo, *tx_fifo; + int nbytes; + + session_fifo_event_t evt; + unix_shared_memory_queue_t *q; + int rv; + + rx_fifo = e->fifo; + tx_fifo = utm->sessions[rx_fifo->client_session_index].server_tx_fifo; + + do + { + nbytes = svm_fifo_dequeue_nowait (rx_fifo, 0, + vec_len (utm->rx_buf), utm->rx_buf); + } + while (nbytes <= 0); + do + { + rv = svm_fifo_enqueue_nowait (tx_fifo, 0, nbytes, utm->rx_buf); + } + while (rv == -2); + + /* Fabricate TX event, send to vpp */ + evt.fifo = tx_fifo; + evt.event_type = FIFO_EVENT_SERVER_TX; + /* $$$$ for event logging */ + evt.enqueue_length = nbytes; + evt.event_id = e->event_id; + q = utm->vpp_event_queue; + unix_shared_memory_queue_add (q, (u8 *) & evt, 0 /* do wait for mutex */ ); +} + +void +handle_event_queue (uri_udp_test_main_t * utm) +{ + session_fifo_event_t _e, *e = &_e;; + + while (1) + { + unix_shared_memory_queue_sub (utm->our_event_queue, (u8 *) e, + 0 /* nowait */ ); + switch (e->event_type) + { + case FIFO_EVENT_SERVER_RX: + handle_fifo_event_server_rx (utm, e); + break; + + case FIFO_EVENT_SERVER_EXIT: + return; + + default: + clib_warning ("unknown event type %d", e->event_type); + break; + } + if (PREDICT_FALSE (utm->time_to_stop == 1)) + break; + if (PREDICT_FALSE (utm->time_to_print_stats == 1)) + { + utm->time_to_print_stats = 0; + fformat (stdout, "%d connections\n", pool_elts (utm->sessions)); + } + } +} + +void +uri_udp_test (uri_udp_test_main_t * utm) +{ + vl_api_bind_uri_t *bmp; + vl_api_unbind_uri_t *ump; + + bmp = vl_msg_api_alloc (sizeof (*bmp)); + memset (bmp, 0, sizeof (*bmp)); + + bmp->_vl_msg_id = ntohs (VL_API_BIND_URI); + bmp->client_index = utm->my_client_index; + bmp->context = ntohl (0xfeedface); + bmp->initial_segment_size = 256 << 20; /* size of initial segment */ + bmp->options[SESSION_OPTIONS_FLAGS] = + SESSION_OPTIONS_FLAGS_USE_FIFO | SESSION_OPTIONS_FLAGS_ADD_SEGMENT; + bmp->options[SESSION_OPTIONS_RX_FIFO_SIZE] = 16 << 10; + bmp->options[SESSION_OPTIONS_TX_FIFO_SIZE] = 16 << 10; + bmp->options[SESSION_OPTIONS_ADD_SEGMENT_SIZE] = 128 << 20; + memcpy (bmp->uri, utm->uri, vec_len (utm->uri)); + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *) & bmp); + + if (wait_for_state_change (utm, STATE_READY)) + { + clib_warning ("timeout waiting for STATE_READY"); + return; + } + + handle_event_queue (utm); + + ump = vl_msg_api_alloc (sizeof (*ump)); + memset (ump, 0, sizeof (*ump)); + + ump->_vl_msg_id = ntohs (VL_API_UNBIND_URI); + ump->client_index = utm->my_client_index; + memcpy (ump->uri, utm->uri, vec_len (utm->uri)); + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *) & ump); + + if (wait_for_state_change (utm, STATE_START)) + { + clib_warning ("timeout waiting for STATE_START"); + return; + } + + fformat (stdout, "Test complete...\n"); +} + +int +main (int argc, char **argv) +{ + uri_udp_test_main_t *utm = &uri_udp_test_main; + unformat_input_t _argv, *a = &_argv; + u8 *chroot_prefix; + u8 *heap; + u8 *bind_name = (u8 *) "udp://0.0.0.0/1234"; + u32 tmp; + mheap_t *h; + session_t *session; + int i; + int i_am_master = 1; + + clib_mem_init (0, 256 << 20); + + heap = clib_mem_get_per_cpu_heap (); + h = mheap_header (heap); + + /* make the main heap thread-safe */ + h->flags |= MHEAP_FLAG_THREAD_SAFE; + + vec_validate (utm->rx_buf, 8192); + + utm->session_index_by_vpp_handles = hash_create (0, sizeof (uword)); + + utm->my_pid = getpid (); + utm->configured_segment_size = 1 << 20; + + clib_time_init (&utm->clib_time); + init_error_string_table (utm); + svm_fifo_segment_init (0x200000000ULL, 20); + unformat_init_command_line (a, argv); + + while (unformat_check_input (a) != UNFORMAT_END_OF_INPUT) + { + if (unformat (a, "chroot prefix %s", &chroot_prefix)) + { + vl_set_memory_root_path ((char *) chroot_prefix); + } + else if (unformat (a, "uri %s", &bind_name)) + ; + else if (unformat (a, "segment-size %dM", &tmp)) + utm->configured_segment_size = tmp << 20; + else if (unformat (a, "segment-size %dG", &tmp)) + utm->configured_segment_size = tmp << 30; + else if (unformat (a, "master")) + i_am_master = 1; + else if (unformat (a, "slave")) + i_am_master = 0; + else + { + fformat (stderr, "%s: usage [master|slave]\n"); + exit (1); + } + } + + utm->cut_through_session_index = ~0; + utm->uri = format (0, "%s%c", bind_name, 0); + utm->i_am_master = i_am_master; + utm->segment_main = &svm_fifo_segment_main; + + utm->connect_uri = format (0, "udp://10.0.0.1/1234%c", 0); + + setup_signal_handlers (); + + uri_api_hookup (utm); + + if (connect_to_vpp (i_am_master ? "uri_udp_master" : "uri_udp_slave") < 0) + { + svm_region_exit (); + fformat (stderr, "Couldn't connect to vpe, exiting...\n"); + exit (1); + } + + if (i_am_master == 0) + { + uri_udp_slave_test (utm); + exit (0); + } + + /* $$$$ hack preallocation */ + for (i = 0; i < 200000; i++) + { + pool_get (utm->sessions, session); + memset (session, 0, sizeof (*session)); + } + for (i = 0; i < 200000; i++) + pool_put_index (utm->sessions, i); + + uri_udp_test (utm); + + vl_client_disconnect_from_vlib (); + exit (0); +} + +#undef vl_api_version +#define vl_api_version(n,v) static u32 vpe_api_version = v; +#include <vpp/api/vpe.api.h> +#undef vl_api_version + +void +vl_client_add_api_signatures (vl_api_memclnt_create_t * mp) +{ + /* + * Send the main API signature in slot 0. This bit of code must + * match the checks in ../vpe/api/api.c: vl_msg_api_version_check(). + */ + mp->api_versions[0] = clib_host_to_net_u32 (vpe_api_version); +} + +u32 +vl (void *p) +{ + return vec_len (p); +} + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/uri/uritest.c b/src/uri/uritest.c new file mode 100644 index 00000000000..edcdb3ad61c --- /dev/null +++ b/src/uri/uritest.c @@ -0,0 +1,484 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <stdio.h> +#include <setjmp.h> +#include <vppinfra/clib.h> +#include <vppinfra/format.h> +#include <vppinfra/error.h> +#include <vppinfra/time.h> +#include <vppinfra/macros.h> +#include <vnet/vnet.h> +#include <vlib/vlib.h> +#include <vlib/unix/unix.h> +#include <vlibapi/api.h> +#include <vlibmemory/api.h> +#include <vpp-api/vpe_msg_enum.h> +#include <svm_fifo_segment.h> + +#define vl_typedefs /* define message structures */ +#include <vpp-api/vpe_all_api_h.h> +#undef vl_typedefs + +/* declare message handlers for each api */ + +#define vl_endianfun /* define message structures */ +#include <vpp-api/vpe_all_api_h.h> +#undef vl_endianfun + +/* instantiate all the print functions we know about */ +#define vl_print(handle, ...) +#define vl_printfun +#include <vpp-api/vpe_all_api_h.h> +#undef vl_printfun + +typedef enum +{ + STATE_START, + STATE_READY, + STATE_DISCONNECTING, +} connection_state_t; + +typedef struct +{ + /* vpe input queue */ + unix_shared_memory_queue_t *vl_input_queue; + + /* API client handle */ + u32 my_client_index; + + /* role */ + int i_am_master; + + /* The URI we're playing with */ + u8 *uri; + + /* fifo segment */ + svm_fifo_segment_private_t *seg; + + svm_fifo_t *rx_fifo; + svm_fifo_t *tx_fifo; + + /* For deadman timers */ + clib_time_t clib_time; + + /* State of the connection, shared between msg RX thread and main thread */ + volatile connection_state_t state; + + /* VNET_API_ERROR_FOO -> "Foo" hash table */ + uword *error_string_by_error_number; +} uritest_main_t; + +#if CLIB_DEBUG > 0 +#define NITER 1000 +#else +#define NITER 1000000 +#endif + +uritest_main_t uritest_main; + +u8 * +format_api_error (u8 * s, va_list * args) +{ + uritest_main_t *utm = va_arg (*args, uritest_main_t *); + i32 error = va_arg (*args, u32); + uword *p; + + p = hash_get (utm->error_string_by_error_number, -error); + + if (p) + s = format (s, "%s", p[0]); + else + s = format (s, "%d", error); + return s; +} + +int +wait_for_state_change (uritest_main_t * utm, connection_state_t state) +{ + f64 timeout = clib_time_now (&utm->clib_time) + 1.0; + + while (clib_time_now (&utm->clib_time) < timeout) + { + if (utm->state == state) + return 0; + } + return -1; +} + +static void +vl_api_bind_uri_reply_t_handler (vl_api_bind_uri_reply_t * mp) +{ + uritest_main_t *utm = &uritest_main; + svm_fifo_segment_create_args_t _a, *a = &_a; + int rv; + + ASSERT (utm->i_am_master); + + if (mp->segment_name_length == 0) + { + clib_warning ("segment_name_length zero"); + return; + } + + a->segment_name = (char *) mp->segment_name; + a->segment_size = mp->segment_size; + + /* Create the segment */ + rv = svm_fifo_segment_create (a); + if (rv) + { + clib_warning ("sm_fifo_segment_create ('%s') failed", mp->segment_name); + return; + } + + vec_validate (utm->seg, 0); + + memcpy (utm->seg, a->rv, sizeof (*utm->seg)); + + /* + * By construction the master's idea of the rx fifo ends up in + * fsh->fifos[0], and the master's idea of the tx fifo ends up in + * fsh->fifos[1]. + */ + utm->rx_fifo = svm_fifo_segment_alloc_fifo (utm->seg, 10240); + ASSERT (utm->rx_fifo); + + utm->tx_fifo = svm_fifo_segment_alloc_fifo (utm->seg, 10240); + ASSERT (utm->tx_fifo); + + utm->state = STATE_READY; +} + +static void +vl_api_connect_uri_reply_t_handler (vl_api_connect_uri_reply_t * mp) +{ + uritest_main_t *utm = &uritest_main; + svm_fifo_segment_create_args_t _a, *a = &_a; + ssvm_shared_header_t *sh; + svm_fifo_segment_header_t *fsh; + int rv; + + ASSERT (utm->i_am_master == 0); + + if (mp->segment_name_length == 0) + { + clib_warning ("segment_name_length zero"); + return; + } + + memset (a, 0, sizeof (*a)); + + a->segment_name = (char *) mp->segment_name; + + rv = svm_fifo_segment_attach (a); + if (rv) + { + clib_warning ("sm_fifo_segment_create ('%s') failed", mp->segment_name); + return; + } + + vec_validate (utm->seg, 0); + + memcpy (utm->seg, a->rv, sizeof (*utm->seg)); + sh = utm->seg->ssvm.sh; + fsh = (svm_fifo_segment_header_t *) sh->opaque[0]; + + while (vec_len (fsh->fifos) < 2) + sleep (1); + + utm->rx_fifo = (svm_fifo_t *) fsh->fifos[1]; + ASSERT (utm->rx_fifo); + utm->tx_fifo = (svm_fifo_t *) fsh->fifos[0]; + ASSERT (utm->tx_fifo); + + /* security: could unlink /dev/shm/<mp->segment_name> here, maybe */ + + utm->state = STATE_READY; +} + +static void +vl_api_unbind_uri_reply_t_handler (vl_api_unbind_uri_reply_t * mp) +{ + uritest_main_t *utm = &uritest_main; + + if (mp->retval != 0) + clib_warning ("returned %d", ntohl (mp->retval)); + + utm->state = STATE_START; +} + +#define foreach_uri_msg \ +_(BIND_URI_REPLY, bind_uri_reply) \ +_(CONNECT_URI_REPLY, connect_uri_reply) \ +_(UNBIND_URI_REPLY, unbind_uri_reply) + +void +uri_api_hookup (uritest_main_t * utm) +{ +#define _(N,n) \ + vl_msg_api_set_handlers(VL_API_##N, #n, \ + vl_api_##n##_t_handler, \ + vl_noop_handler, \ + vl_api_##n##_t_endian, \ + vl_api_##n##_t_print, \ + sizeof(vl_api_##n##_t), 1); + foreach_uri_msg; +#undef _ + +} + + +int +connect_to_vpp (char *name) +{ + uritest_main_t *utm = &uritest_main; + api_main_t *am = &api_main; + + if (vl_client_connect_to_vlib ("/vpe-api", name, 32) < 0) + return -1; + + utm->vl_input_queue = am->shmem_hdr->vl_input_queue; + utm->my_client_index = am->my_client_index; + + return 0; +} + +void +vlib_cli_output (struct vlib_main_t *vm, char *fmt, ...) +{ + clib_warning ("BUG"); +} + +static void +init_error_string_table (uritest_main_t * utm) +{ + utm->error_string_by_error_number = hash_create (0, sizeof (uword)); + +#define _(n,v,s) hash_set (utm->error_string_by_error_number, -v, s); + foreach_vnet_api_error; +#undef _ + + hash_set (utm->error_string_by_error_number, 99, "Misc"); +} + +void +uritest_master (uritest_main_t * utm) +{ + vl_api_bind_uri_t *bmp; + vl_api_unbind_uri_t *ump; + int i; + u8 *test_data = 0; + u8 *reply = 0; + u32 reply_len; + int mypid = getpid (); + + for (i = 0; i < 2048; i++) + vec_add1 (test_data, 'a' + (i % 32)); + + bmp = vl_msg_api_alloc (sizeof (*bmp)); + memset (bmp, 0, sizeof (*bmp)); + + bmp->_vl_msg_id = ntohs (VL_API_BIND_URI); + bmp->client_index = utm->my_client_index; + bmp->context = ntohl (0xfeedface); + bmp->segment_size = 256 << 10; + memcpy (bmp->uri, utm->uri, vec_len (utm->uri)); + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *) & bmp); + + if (wait_for_state_change (utm, STATE_READY)) + { + clib_warning ("timeout waiting for STATE_READY"); + return; + } + + for (i = 0; i < NITER; i++) + svm_fifo_enqueue (utm->tx_fifo, mypid, vec_len (test_data), test_data); + + vec_validate (reply, 0); + + reply_len = svm_fifo_dequeue (utm->rx_fifo, mypid, vec_len (reply), reply); + + if (reply_len != 1) + clib_warning ("reply length %d", reply_len); + + if (reply[0] == 1) + fformat (stdout, "Test OK..."); + + ump = vl_msg_api_alloc (sizeof (*ump)); + memset (ump, 0, sizeof (*ump)); + + ump->_vl_msg_id = ntohs (VL_API_UNBIND_URI); + ump->client_index = utm->my_client_index; + memcpy (ump->uri, utm->uri, vec_len (utm->uri)); + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *) & ump); + + if (wait_for_state_change (utm, STATE_START)) + { + clib_warning ("timeout waiting for STATE_READY"); + return; + } + + fformat (stdout, "Master done...\n"); +} + +void +uritest_slave (uritest_main_t * utm) +{ + vl_api_connect_uri_t *cmp; + int i, j; + u8 *test_data = 0; + u8 *reply = 0; + u32 bytes_received = 0; + u32 actual_bytes; + int mypid = getpid (); + u8 ok; + f64 before, after, delta, bytes_per_second; + + vec_validate (test_data, 4095); + + cmp = vl_msg_api_alloc (sizeof (*cmp)); + memset (cmp, 0, sizeof (*cmp)); + + cmp->_vl_msg_id = ntohs (VL_API_CONNECT_URI); + cmp->client_index = utm->my_client_index; + cmp->context = ntohl (0xfeedface); + memcpy (cmp->uri, utm->uri, vec_len (utm->uri)); + vl_msg_api_send_shmem (utm->vl_input_queue, (u8 *) & cmp); + + if (wait_for_state_change (utm, STATE_READY)) + { + clib_warning ("timeout waiting for STATE_READY"); + return; + } + + ok = 1; + before = clib_time_now (&utm->clib_time); + for (i = 0; i < NITER; i++) + { + actual_bytes = svm_fifo_dequeue (utm->rx_fifo, mypid, + vec_len (test_data), test_data); + j = 0; + while (j < actual_bytes) + { + if (test_data[j] != ('a' + (bytes_received % 32))) + ok = 0; + bytes_received++; + j++; + } + if (bytes_received == NITER * 2048) + break; + } + + vec_add1 (reply, ok); + + svm_fifo_enqueue (utm->tx_fifo, mypid, vec_len (reply), reply); + after = clib_time_now (&utm->clib_time); + delta = after - before; + bytes_per_second = 0.0; + + if (delta > 0.0) + bytes_per_second = (f64) bytes_received / delta; + + fformat (stdout, + "Slave done, %d bytes in %.2f seconds, %.2f bytes/sec...\n", + bytes_received, delta, bytes_per_second); +} + +int +main (int argc, char **argv) +{ + uritest_main_t *utm = &uritest_main; + unformat_input_t _argv, *a = &_argv; + u8 *chroot_prefix; + u8 *heap; + char *bind_name = "fifo:uritest"; + mheap_t *h; + int i_am_master = 0; + + clib_mem_init (0, 128 << 20); + + heap = clib_mem_get_per_cpu_heap (); + h = mheap_header (heap); + + /* make the main heap thread-safe */ + h->flags |= MHEAP_FLAG_THREAD_SAFE; + + clib_time_init (&utm->clib_time); + init_error_string_table (utm); + svm_fifo_segment_init (0x200000000ULL, 20); + unformat_init_command_line (a, argv); + + utm->uri = format (0, "%s%c", bind_name, 0); + + while (unformat_check_input (a) != UNFORMAT_END_OF_INPUT) + { + if (unformat (a, "master")) + i_am_master = 1; + else if (unformat (a, "slave")) + i_am_master = 0; + else if (unformat (a, "chroot prefix %s", &chroot_prefix)) + { + vl_set_memory_root_path ((char *) chroot_prefix); + } + else + { + fformat (stderr, "%s: usage [master|slave]\n"); + exit (1); + } + } + + uri_api_hookup (utm); + + if (connect_to_vpp (i_am_master ? "uritest_master" : "uritest_slave") < 0) + { + svm_region_exit (); + fformat (stderr, "Couldn't connect to vpe, exiting...\n"); + exit (1); + } + + utm->i_am_master = i_am_master; + + if (i_am_master) + uritest_master (utm); + else + uritest_slave (utm); + + vl_client_disconnect_from_vlib (); + exit (0); +} + +#undef vl_api_version +#define vl_api_version(n,v) static u32 vpe_api_version = v; +#include <vpp-api/vpe.api.h> +#undef vl_api_version + +void +vl_client_add_api_signatures (vl_api_memclnt_create_t * mp) +{ + /* + * Send the main API signature in slot 0. This bit of code must + * match the checks in ../vpe/api/api.c: vl_msg_api_version_check(). + */ + mp->api_versions[0] = clib_host_to_net_u32 (vpe_api_version); +} + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vlib/buffer.c b/src/vlib/buffer.c index 4f5eb09da91..9f26bec7b56 100644 --- a/src/vlib/buffer.c +++ b/src/vlib/buffer.c @@ -360,7 +360,7 @@ vlib_buffer_create_free_list_helper (vlib_main_t * vm, memset (f, 0, sizeof (f[0])); f->index = f - bm->buffer_free_list_pool; f->n_data_bytes = vlib_buffer_round_size (n_data_bytes); - f->min_n_buffers_each_physmem_alloc = 16; + f->min_n_buffers_each_physmem_alloc = VLIB_FRAME_SIZE; f->name = clib_mem_is_heap_object (name) ? name : format (0, "%s", name); /* Setup free buffer template. */ diff --git a/src/vlib/buffer.h b/src/vlib/buffer.h index 1f723f3b677..69c8c7ccb09 100644 --- a/src/vlib/buffer.h +++ b/src/vlib/buffer.h @@ -240,6 +240,74 @@ vlib_get_buffer_opaque2 (vlib_buffer_t * b) return (void *) b->opaque2; } +/** \brief Get pointer to the end of buffer's data + * @param b pointer to the buffer + * @return pointer to tail of packet's data + */ +always_inline u8 * +vlib_buffer_get_tail (vlib_buffer_t * b) +{ + return b->data + b->current_data + b->current_length; +} + +/** \brief Append uninitialized data to buffer + * @param b pointer to the buffer + * @param size number of uninitialized bytes + * @return pointer to beginning of uninitialized data + */ +always_inline void * +vlib_buffer_put_uninit (vlib_buffer_t * b, u8 size) +{ + void *p = vlib_buffer_get_tail (b); + /* XXX make sure there's enough space */ + b->current_length += size; + return p; +} + +/** \brief Prepend uninitialized data to buffer + * @param b pointer to the buffer + * @param size number of uninitialized bytes + * @return pointer to beginning of uninitialized data + */ +always_inline void * +vlib_buffer_push_uninit (vlib_buffer_t * b, u8 size) +{ + ASSERT (b->current_data + VLIB_BUFFER_PRE_DATA_SIZE >= size); + b->current_data -= size; + b->current_length += size; + + return vlib_buffer_get_current (b); +} + +/** \brief Make head room, typically for packet headers + * @param b pointer to the buffer + * @param size number of head room bytes + * @return pointer to start of buffer (current data) + */ +always_inline void * +vlib_buffer_make_headroom (vlib_buffer_t * b, u8 size) +{ + ASSERT (b->current_data + VLIB_BUFFER_PRE_DATA_SIZE >= size); + b->current_data += size; + return vlib_buffer_get_current (b); +} + +/** \brief Retrieve bytes from buffer head + * @param b pointer to the buffer + * @param size number of bytes to pull + * @return pointer to start of buffer (current data) + */ +always_inline void * +vlib_buffer_pull (vlib_buffer_t * b, u8 size) +{ + if (b->current_length + VLIB_BUFFER_PRE_DATA_SIZE < size) + return 0; + + void *data = vlib_buffer_get_current (b); + vlib_buffer_advance (b, size); + return data; +} + /* Forward declaration. */ struct vlib_main_t; diff --git a/src/vlibmemory/unix_shared_memory_queue.c b/src/vlibmemory/unix_shared_memory_queue.c index 25d28910706..e86edec3c65 100644 --- a/src/vlibmemory/unix_shared_memory_queue.c +++ b/src/vlibmemory/unix_shared_memory_queue.c @@ -33,18 +33,13 @@ * nels = number of elements on the queue * elsize = element size, presumably 4 and cacheline-size will * be popular choices. - * coid = consumer coid, from ChannelCreate * pid = consumer pid - * pulse_code = pulse code consumer expects - * pulse_value = pulse value consumer expects - * consumer_prio = consumer's priority, so pulses won't change - * the consumer's priority. * * The idea is to call this function in the queue consumer, * and e-mail the queue pointer to the producer(s). * - * The spp process / main thread allocates one of these - * at startup; its main input queue. The spp main input queue + * The vpp process / main thread allocates one of these + * at startup; its main input queue. The vpp main input queue * has a pointer to it in the shared memory segment header. * * You probably want to be on an svm data heap before calling this @@ -70,7 +65,7 @@ unix_shared_memory_queue_init (int nels, q->signal_when_queue_non_empty = signal_when_queue_non_empty; memset (&attr, 0, sizeof (attr)); - memset (&cattr, 0, sizeof (attr)); + memset (&cattr, 0, sizeof (cattr)); if (pthread_mutexattr_init (&attr)) clib_unix_warning ("mutexattr_init"); @@ -277,6 +272,7 @@ unix_shared_memory_queue_sub (unix_shared_memory_queue_t * q, clib_memcpy (elem, headp, q->elsize); q->head++; + /* $$$$ JFC shouldn't this be == 0? */ if (q->cursize == q->maxsize) need_broadcast = 1; diff --git a/src/vlibmemory/unix_shared_memory_queue.h b/src/vlibmemory/unix_shared_memory_queue.h index f758f17c856..13800065beb 100644 --- a/src/vlibmemory/unix_shared_memory_queue.h +++ b/src/vlibmemory/unix_shared_memory_queue.h @@ -29,7 +29,7 @@ typedef struct _unix_shared_memory_queue pthread_cond_t condvar; /* 8 bytes */ int head; int tail; - int cursize; + volatile int cursize; int maxsize; int elsize; int consumer_pid; diff --git a/src/vnet.am b/src/vnet.am index 64484e18486..923f61d8a0b 100644 --- a/src/vnet.am +++ b/src/vnet.am @@ -324,11 +324,7 @@ libvnet_la_SOURCES += \ vnet/ip/ip_input_acl.c \ vnet/ip/lookup.c \ vnet/ip/ping.c \ - vnet/ip/punt.c \ - vnet/ip/udp_format.c \ - vnet/ip/udp_init.c \ - vnet/ip/udp_local.c \ - vnet/ip/udp_pg.c + vnet/ip/punt.c nobase_include_HEADERS += \ vnet/ip/format.h \ @@ -354,11 +350,7 @@ nobase_include_HEADERS += \ vnet/ip/ports.def \ vnet/ip/protocols.def \ vnet/ip/punt_error.def \ - vnet/ip/punt.h \ - vnet/ip/tcp_packet.h \ - vnet/ip/udp_error.def \ - vnet/ip/udp.h \ - vnet/ip/udp_packet.h + vnet/ip/punt.h API_FILES += vnet/ip/ip.api @@ -473,6 +465,38 @@ test_map_LDADD = libvnet.la libvppinfra.la libvlib.la \ test_map_LDFLAGS = -static endif +######################################## +# Layer 4 protocol: tcp +######################################## +libvnet_la_SOURCES += \ + vnet/tcp/tcp_format.c \ + vnet/tcp/tcp_pg.c \ + vnet/tcp/tcp_syn_filter4.c \ + vnet/tcp/tcp_output.c \ + vnet/tcp/tcp_input.c \ + vnet/tcp/tcp_newreno.c \ + vnet/tcp/tcp.c + +nobase_include_HEADERS += \ + vnet/tcp/tcp_packet.h \ + vnet/tcp/tcp_timer.h \ + vnet/tcp/tcp.h + +######################################## +# Layer 4 protocol: udp +######################################## +libvnet_la_SOURCES += \ + vnet/udp/udp.c \ + vnet/udp/udp_input.c \ + vnet/udp/builtin_server.c \ + vnet/udp/udp_format.c \ + vnet/udp/udp_local.c \ + vnet/udp/udp_pg.c + +nobase_include_HEADERS += \ + vnet/udp/udp_error.def \ + vnet/udp/udp.h \ + vnet/udp/udp_packet.h ######################################## # Tunnel protocol: gre @@ -834,6 +858,28 @@ nobase_include_HEADERS += \ vnet/devices/ssvm/ssvm_eth.h ######################################## +# session managmeent +######################################## + +libvnet_la_SOURCES += \ + vnet/session/session.c \ + vnet/session/node.c \ + vnet/session/transport.c \ + vnet/session/application.c \ + vnet/session/session_cli.c \ + vnet/session/hashes.c \ + vnet/session/application_interface.c \ + vnet/session/session_api.c + +nobase_include_HEADERS += \ + vnet/session/session.h \ + vnet/session/application.h \ + vnet/session/transport.h \ + vnet/session/application_interface.h + +API_FILES += vnet/session/session.api + +######################################## # Linux packet interface ######################################## diff --git a/src/vnet/api_errno.h b/src/vnet/api_errno.h index 8680ef7ca63..861a5767af2 100644 --- a/src/vnet/api_errno.h +++ b/src/vnet/api_errno.h @@ -91,14 +91,19 @@ _(INVALID_ADDRESS_FAMILY, -97, "Invalid address family") \ _(INVALID_SUB_SW_IF_INDEX, -98, "Invalid sub-interface sw_if_index") \ _(TABLE_TOO_BIG, -99, "Table too big") \ _(CANNOT_ENABLE_DISABLE_FEATURE, -100, "Cannot enable/disable feature") \ -_(BFD_EEXIST, -101, "Duplicate BFD object") \ -_(BFD_ENOENT, -102, "No such BFD object") \ -_(BFD_EINUSE, -103, "BFD object in use") \ -_(BFD_NOTSUPP, -104, "BFD feature not supported") \ -_(LISP_RLOC_LOCAL, -105, "RLOC address is local") \ -_(BFD_EAGAIN, -106, "BFD object cannot be manipulated at this time") \ -_(INVALID_GPE_MODE, -107, "Invalid GPE mode") \ -_(LISP_GPE_ENTRIES_PRESENT, -108, "LISP GPE entries are present") +_(BFD_EEXIST, -101, "Duplicate BFD object") \ +_(BFD_ENOENT, -102, "No such BFD object") \ +_(BFD_EINUSE, -103, "BFD object in use") \ +_(BFD_NOTSUPP, -104, "BFD feature not supported") \ +_(ADDRESS_IN_USE, -105, "Address in use") \ +_(ADDRESS_NOT_IN_USE, -106, "Address not in use") \ +_(QUEUE_FULL, -107, "Queue full") \ +_(UNKNOWN_URI_TYPE, -108, "Unknown URI type") \ +_(URI_FIFO_CREATE_FAILED, -109, "URI FIFO segment create failed") \ +_(LISP_RLOC_LOCAL, -110, "RLOC address is local") \ +_(BFD_EAGAIN, -111, "BFD object cannot be manipulated at this time") \ +_(INVALID_GPE_MODE, -112, "Invalid GPE mode") \ +_(LISP_GPE_ENTRIES_PRESENT, -113, "LISP GPE entries are present") typedef enum { diff --git a/src/vnet/bfd/bfd_udp.c b/src/vnet/bfd/bfd_udp.c index 146faad6bfa..cf05089b430 100644 --- a/src/vnet/bfd/bfd_udp.c +++ b/src/vnet/bfd/bfd_udp.c @@ -18,12 +18,12 @@ #include <vlib/buffer.h> #include <vnet/ip/format.h> #include <vnet/ethernet/packet.h> -#include <vnet/ip/udp_packet.h> +#include <vnet/udp/udp_packet.h> +#include <vnet/udp/udp.h> #include <vnet/ip/lookup.h> #include <vnet/ip/icmp46_packet.h> #include <vnet/ip/ip4.h> #include <vnet/ip/ip6.h> -#include <vnet/ip/udp.h> #include <vnet/ip/ip6_packet.h> #include <vnet/adj/adj.h> #include <vnet/adj/adj_nbr.h> diff --git a/src/vnet/buffer.h b/src/vnet/buffer.h index f1cc637179a..3de01f2afa1 100644 --- a/src/vnet/buffer.h +++ b/src/vnet/buffer.h @@ -277,6 +277,16 @@ typedef struct u16 buffer_advance; } device_input_feat; + /* TCP */ + struct + { + u32 connection_index; + u32 seq_number; + u32 seq_end; + u32 ack_number; + u8 flags; + } tcp; + u32 unused[6]; }; } vnet_buffer_opaque_t; diff --git a/src/vnet/classify/vnet_classify.c b/src/vnet/classify/vnet_classify.c index 6093e2acb19..b651a1f1eb3 100644 --- a/src/vnet/classify/vnet_classify.c +++ b/src/vnet/classify/vnet_classify.c @@ -695,8 +695,8 @@ int vnet_classify_add_del_table (vnet_classify_main_t * cm, } #define foreach_tcp_proto_field \ -_(src_port) \ -_(dst_port) +_(src) \ +_(dst) #define foreach_udp_proto_field \ _(src_port) \ diff --git a/src/vnet/dhcp/dhcp_proxy.h b/src/vnet/dhcp/dhcp_proxy.h index c0d79c41af6..4586d8838af 100644 --- a/src/vnet/dhcp/dhcp_proxy.h +++ b/src/vnet/dhcp/dhcp_proxy.h @@ -26,7 +26,7 @@ #include <vnet/ip/ip4_packet.h> #include <vnet/pg/pg.h> #include <vnet/ip/format.h> -#include <vnet/ip/udp.h> +#include <vnet/udp/udp.h> typedef enum { #define dhcp_proxy_error(n,s) DHCP_PROXY_ERROR_##n, diff --git a/src/vnet/flow/flow_report.h b/src/vnet/flow/flow_report.h index 4e764377dc8..e8ed3818534 100644 --- a/src/vnet/flow/flow_report.h +++ b/src/vnet/flow/flow_report.h @@ -23,7 +23,7 @@ #include <vnet/ip/ip_packet.h> #include <vnet/ip/ip4_packet.h> #include <vnet/ip/ip6_packet.h> -#include <vnet/ip/udp.h> +#include <vnet/udp/udp.h> #include <vlib/cli.h> #include <vppinfra/error.h> #include <vppinfra/hash.h> diff --git a/src/vnet/ip/ip.h b/src/vnet/ip/ip.h index 02a1a9636ee..70b4ccd8dd7 100644 --- a/src/vnet/ip/ip.h +++ b/src/vnet/ip/ip.h @@ -50,8 +50,8 @@ #include <vnet/ip/ip_packet.h> #include <vnet/ip/lookup.h> -#include <vnet/ip/tcp_packet.h> -#include <vnet/ip/udp_packet.h> +#include <vnet/tcp/tcp_packet.h> +#include <vnet/udp/udp_packet.h> #include <vnet/ip/icmp46_packet.h> #include <vnet/ip/ip4.h> diff --git a/src/vnet/ip/ip4.h b/src/vnet/ip/ip4.h index b184fbaef42..4e075d0fb5c 100644 --- a/src/vnet/ip/ip4.h +++ b/src/vnet/ip/ip4.h @@ -309,8 +309,8 @@ ip4_compute_flow_hash (const ip4_header_t * ip, b = (flow_hash_config & IP_FLOW_HASH_REVERSE_SRC_DST) ? t1 : t2; b ^= (flow_hash_config & IP_FLOW_HASH_PROTO) ? ip->protocol : 0; - t1 = is_tcp_udp ? tcp->ports.src : 0; - t2 = is_tcp_udp ? tcp->ports.dst : 0; + t1 = is_tcp_udp ? tcp->src : 0; + t2 = is_tcp_udp ? tcp->dst : 0; t1 = (flow_hash_config & IP_FLOW_HASH_SRC_PORT) ? t1 : 0; t2 = (flow_hash_config & IP_FLOW_HASH_DST_PORT) ? t2 : 0; @@ -334,6 +334,44 @@ u8 *format_ip4_forward_next_trace (u8 * s, va_list * args); u32 ip4_tcp_udp_validate_checksum (vlib_main_t * vm, vlib_buffer_t * p0); +#define IP_DF 0x4000 /* don't fragment */ + +/** + * Push IPv4 header to buffer + * + * This does not support fragmentation. + * + * @param vm - vlib_main + * @param b - buffer to write the header to + * @param src - source IP + * @param dst - destination IP + * @param prot - payload proto + * + * @return - pointer to start of IP header + */ +always_inline void * +vlib_buffer_push_ip4 (vlib_main_t * vm, vlib_buffer_t * b, + ip4_address_t * src, ip4_address_t * dst, int proto) +{ + ip4_header_t *ih; + + /* make some room */ + ih = vlib_buffer_push_uninit (b, sizeof (ip4_header_t)); + + ih->ip_version_and_header_length = 0x45; + ih->tos = 0; + ih->length = clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b)); + + /* No fragments */ + ih->flags_and_fragment_offset = clib_host_to_net_u16 (IP_DF); + ih->ttl = 255; + ih->protocol = proto; + ih->src_address.as_u32 = src->as_u32; + ih->dst_address.as_u32 = dst->as_u32; + + ih->checksum = ip4_header_checksum (ih); + return ih; +} #endif /* included_ip_ip4_h */ /* diff --git a/src/vnet/ip/ip4_forward.c b/src/vnet/ip/ip4_forward.c index 8081b34b3dc..66d91ab697d 100644 --- a/src/vnet/ip/ip4_forward.c +++ b/src/vnet/ip/ip4_forward.c @@ -1478,8 +1478,18 @@ ip4_tcp_udp_validate_checksum (vlib_main_t * vm, vlib_buffer_t * p0) return p0->flags; } -static uword -ip4_local (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) +/* *INDENT-OFF* */ +VNET_FEATURE_ARC_INIT (ip4_local) = +{ + .arc_name = "ip4-local", + .start_nodes = VNET_FEATURES ("ip4-local"), +}; +/* *INDENT-ON* */ + +static inline uword +ip4_local_inline (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * frame, int head_of_feature_arc) { ip4_main_t *im = &ip4_main; ip_lookup_main_t *lm = &im->lookup_main; @@ -1487,6 +1497,7 @@ ip4_local (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) u32 *from, *to_next, n_left_from, n_left_to_next; vlib_node_runtime_t *error_node = vlib_node_get_runtime (vm, ip4_input_node.index); + u8 arc_index = vnet_feat_arc_ip4_local.feature_arc_index; from = vlib_frame_vector_args (frame); n_left_from = frame->n_vectors; @@ -1513,7 +1524,7 @@ ip4_local (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) i32 len_diff0, len_diff1; u8 error0, is_udp0, is_tcp_udp0, good_tcp_udp0, proto0; u8 error1, is_udp1, is_tcp_udp1, good_tcp_udp1, proto1; - u8 enqueue_code; + u32 sw_if_index0, sw_if_index1; pi0 = to_next[0] = from[0]; pi1 = to_next[1] = from[1]; @@ -1522,6 +1533,8 @@ ip4_local (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) to_next += 2; n_left_to_next -= 2; + next0 = next1 = IP_LOCAL_NEXT_DROP; + p0 = vlib_get_buffer (vm, pi0); p1 = vlib_get_buffer (vm, pi1); @@ -1531,14 +1544,18 @@ ip4_local (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) vnet_buffer (p0)->ip.start_of_ip_header = p0->current_data; vnet_buffer (p1)->ip.start_of_ip_header = p1->current_data; - fib_index0 = vec_elt (im->fib_index_by_sw_if_index, - vnet_buffer (p0)->sw_if_index[VLIB_RX]); + sw_if_index0 = vnet_buffer (p0)->sw_if_index[VLIB_RX]; + sw_if_index1 = vnet_buffer (p1)->sw_if_index[VLIB_RX]; + + fib_index0 = vec_elt (im->fib_index_by_sw_if_index, sw_if_index0); + fib_index1 = vec_elt (im->fib_index_by_sw_if_index, sw_if_index1); + + fib_index0 = vec_elt (im->fib_index_by_sw_if_index, sw_if_index0); fib_index0 = (vnet_buffer (p0)->sw_if_index[VLIB_TX] == (u32) ~ 0) ? fib_index0 : vnet_buffer (p0)->sw_if_index[VLIB_TX]; - fib_index1 = vec_elt (im->fib_index_by_sw_if_index, - vnet_buffer (p1)->sw_if_index[VLIB_RX]); + fib_index1 = vec_elt (im->fib_index_by_sw_if_index, sw_if_index1); fib_index1 = (vnet_buffer (p1)->sw_if_index[VLIB_TX] == (u32) ~ 0) ? fib_index1 : vnet_buffer (p1)->sw_if_index[VLIB_TX]; @@ -1557,6 +1574,13 @@ ip4_local (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) until support of IP frag reassembly is implemented */ proto0 = ip4_is_fragment (ip0) ? 0xfe : ip0->protocol; proto1 = ip4_is_fragment (ip1) ? 0xfe : ip1->protocol; + + if (head_of_feature_arc == 0) + { + error0 = error1 = IP4_ERROR_UNKNOWN_PROTOCOL; + goto skip_checks; + } + is_udp0 = proto0 == IP_PROTOCOL_UDP; is_udp1 = proto1 == IP_PROTOCOL_UDP; is_tcp_udp0 = is_udp0 || proto0 == IP_PROTOCOL_TCP; @@ -1686,6 +1710,7 @@ ip4_local (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) next0 = lm->local_next_by_ip_protocol[proto0]; next1 = lm->local_next_by_ip_protocol[proto1]; + skip_checks: next0 = error0 != IP4_ERROR_UNKNOWN_PROTOCOL ? IP_LOCAL_NEXT_DROP : next0; next1 = @@ -1694,44 +1719,17 @@ ip4_local (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) p0->error = error0 ? error_node->errors[error0] : 0; p1->error = error1 ? error_node->errors[error1] : 0; - enqueue_code = (next0 != next_index) + 2 * (next1 != next_index); - - if (PREDICT_FALSE (enqueue_code != 0)) + if (head_of_feature_arc) { - switch (enqueue_code) - { - case 1: - /* A B A */ - to_next[-2] = pi1; - to_next -= 1; - n_left_to_next += 1; - vlib_set_next_frame_buffer (vm, node, next0, pi0); - break; - - case 2: - /* A A B */ - to_next -= 1; - n_left_to_next += 1; - vlib_set_next_frame_buffer (vm, node, next1, pi1); - break; - - case 3: - /* A B B or A B C */ - to_next -= 2; - n_left_to_next += 2; - vlib_set_next_frame_buffer (vm, node, next0, pi0); - vlib_set_next_frame_buffer (vm, node, next1, pi1); - if (next0 == next1) - { - vlib_put_next_frame (vm, node, next_index, - n_left_to_next); - next_index = next1; - vlib_get_next_frame (vm, node, next_index, to_next, - n_left_to_next); - } - break; - } + if (PREDICT_TRUE (error0 == (u8) IP4_ERROR_UNKNOWN_PROTOCOL)) + vnet_feature_arc_start (arc_index, sw_if_index0, &next0, p0); + if (PREDICT_TRUE (error1 == (u8) IP4_ERROR_UNKNOWN_PROTOCOL)) + vnet_feature_arc_start (arc_index, sw_if_index1, &next1, p1); } + + vlib_validate_buffer_enqueue_x2 (vm, node, next_index, to_next, + n_left_to_next, pi0, pi1, + next0, next1); } while (n_left_from > 0 && n_left_to_next > 0) @@ -1746,6 +1744,7 @@ ip4_local (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) u8 error0, is_udp0, is_tcp_udp0, good_tcp_udp0, proto0; load_balance_t *lb0; const dpo_id_t *dpo0; + u32 sw_if_index0; pi0 = to_next[0] = from[0]; from += 1; @@ -1753,14 +1752,18 @@ ip4_local (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) to_next += 1; n_left_to_next -= 1; + next0 = IP_LOCAL_NEXT_DROP; + p0 = vlib_get_buffer (vm, pi0); ip0 = vlib_buffer_get_current (p0); vnet_buffer (p0)->ip.start_of_ip_header = p0->current_data; - fib_index0 = vec_elt (im->fib_index_by_sw_if_index, - vnet_buffer (p0)->sw_if_index[VLIB_RX]); + sw_if_index0 = vnet_buffer (p0)->sw_if_index[VLIB_RX]; + + fib_index0 = vec_elt (im->fib_index_by_sw_if_index, sw_if_index0); + fib_index0 = (vnet_buffer (p0)->sw_if_index[VLIB_TX] == (u32) ~ 0) ? fib_index0 : vnet_buffer (p0)->sw_if_index[VLIB_TX]; @@ -1775,6 +1778,13 @@ ip4_local (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) /* Treat IP frag packets as "experimental" protocol for now until support of IP frag reassembly is implemented */ proto0 = ip4_is_fragment (ip0) ? 0xfe : ip0->protocol; + + if (head_of_feature_arc == 0) + { + error0 = IP4_ERROR_UNKNOWN_PROTOCOL; + goto skip_check; + } + is_udp0 = proto0 == IP_PROTOCOL_UDP; is_tcp_udp0 = is_udp0 || proto0 == IP_PROTOCOL_TCP; @@ -1847,6 +1857,8 @@ ip4_local (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) ip0->dst_address.as_u32 != 0xFFFFFFFF) ? IP4_ERROR_SRC_LOOKUP_MISS : error0); + skip_check: + next0 = lm->local_next_by_ip_protocol[proto0]; next0 = @@ -1854,18 +1866,15 @@ ip4_local (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) p0->error = error0 ? error_node->errors[error0] : 0; - if (PREDICT_FALSE (next0 != next_index)) + if (head_of_feature_arc) { - n_left_to_next += 1; - vlib_put_next_frame (vm, node, next_index, n_left_to_next); - - next_index = next0; - vlib_get_next_frame (vm, node, next_index, to_next, - n_left_to_next); - to_next[0] = pi0; - to_next += 1; - n_left_to_next -= 1; + if (PREDICT_TRUE (error0 == (u8) IP4_ERROR_UNKNOWN_PROTOCOL)) + vnet_feature_arc_start (arc_index, sw_if_index0, &next0, p0); } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, pi0, next0); + } vlib_put_next_frame (vm, node, next_index, n_left_to_next); @@ -1874,21 +1883,57 @@ ip4_local (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) return frame->n_vectors; } +static uword +ip4_local (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) +{ + return ip4_local_inline (vm, node, frame, 1 /* head of feature arc */ ); +} + +/* *INDENT-OFF* */ VLIB_REGISTER_NODE (ip4_local_node) = { - .function = ip4_local,.name = "ip4-local",.vector_size = - sizeof (u32),.format_trace = - format_ip4_forward_next_trace,.n_next_nodes = - IP_LOCAL_N_NEXT,.next_nodes = + .function = ip4_local, + .name = "ip4-local", + .vector_size = sizeof (u32), + .format_trace = format_ip4_forward_next_trace, + .n_next_nodes = IP_LOCAL_N_NEXT, + .next_nodes = { - [IP_LOCAL_NEXT_DROP] = "error-drop", - [IP_LOCAL_NEXT_PUNT] = "error-punt", - [IP_LOCAL_NEXT_UDP_LOOKUP] = "ip4-udp-lookup", - [IP_LOCAL_NEXT_ICMP] = "ip4-icmp-input",} -,}; + [IP_LOCAL_NEXT_DROP] = "error-drop", + [IP_LOCAL_NEXT_PUNT] = "error-punt", + [IP_LOCAL_NEXT_UDP_LOOKUP] = "ip4-udp-lookup", + [IP_LOCAL_NEXT_ICMP] = "ip4-icmp-input",}, +}; +/* *INDENT-ON* */ VLIB_NODE_FUNCTION_MULTIARCH (ip4_local_node, ip4_local); +static uword +ip4_local_end_of_arc (vlib_main_t * vm, + vlib_node_runtime_t * node, vlib_frame_t * frame) +{ + return ip4_local_inline (vm, node, frame, 0 /* head of feature arc */ ); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (ip4_local_end_of_arc_node,static) = { + .function = ip4_local_end_of_arc, + .name = "ip4-local-end-of-arc", + .vector_size = sizeof (u32), + + .format_trace = format_ip4_forward_next_trace, + .sibling_of = "ip4-local", +}; + +VLIB_NODE_FUNCTION_MULTIARCH (ip4_local_end_of_arc_node, ip4_local_end_of_arc) + +VNET_FEATURE_INIT (ip4_local_end_of_arc, static) = { + .arc_name = "ip4-local", + .node_name = "ip4-local-end-of-arc", + .runs_before = 0, /* not before any other features */ +}; +/* *INDENT-ON* */ + void ip4_register_protocol (u32 protocol, u32 node_index) { diff --git a/src/vnet/ip/ip4_packet.h b/src/vnet/ip/ip4_packet.h index 8da788b411f..b2c1fcd49c1 100644 --- a/src/vnet/ip/ip4_packet.h +++ b/src/vnet/ip/ip4_packet.h @@ -41,7 +41,7 @@ #define included_ip4_packet_h #include <vnet/ip/ip_packet.h> /* for ip_csum_t */ -#include <vnet/ip/tcp_packet.h> /* for tcp_header_t */ +#include <vnet/tcp/tcp_packet.h> /* for tcp_header_t */ #include <vppinfra/byte_order.h> /* for clib_net_to_host_u16 */ /* IP4 address which can be accessed either as 4 bytes @@ -342,10 +342,10 @@ ip4_tcp_reply_x1 (ip4_header_t * ip0, tcp_header_t * tcp0) ip0->src_address.data_u32 = dst0; ip0->dst_address.data_u32 = src0; - src0 = tcp0->ports.src; - dst0 = tcp0->ports.dst; - tcp0->ports.src = dst0; - tcp0->ports.dst = src0; + src0 = tcp0->src; + dst0 = tcp0->dst; + tcp0->src = dst0; + tcp0->dst = src0; } always_inline void @@ -363,14 +363,14 @@ ip4_tcp_reply_x2 (ip4_header_t * ip0, ip4_header_t * ip1, ip0->dst_address.data_u32 = src0; ip1->dst_address.data_u32 = src1; - src0 = tcp0->ports.src; - src1 = tcp1->ports.src; - dst0 = tcp0->ports.dst; - dst1 = tcp1->ports.dst; - tcp0->ports.src = dst0; - tcp1->ports.src = dst1; - tcp0->ports.dst = src0; - tcp1->ports.dst = src1; + src0 = tcp0->src; + src1 = tcp1->src; + dst0 = tcp0->dst; + dst1 = tcp1->dst; + tcp0->src = dst0; + tcp1->src = dst1; + tcp0->dst = src0; + tcp1->dst = src1; } #endif /* included_ip4_packet_h */ diff --git a/src/vnet/ip/ip6.h b/src/vnet/ip/ip6.h index 5456f0f2ed6..2615fbfab02 100644 --- a/src/vnet/ip/ip6.h +++ b/src/vnet/ip/ip6.h @@ -461,8 +461,8 @@ ip6_compute_flow_hash (const ip6_header_t * ip, b = (flow_hash_config & IP_FLOW_HASH_REVERSE_SRC_DST) ? t1 : t2; b ^= (flow_hash_config & IP_FLOW_HASH_PROTO) ? ip->protocol : 0; - t1 = is_tcp_udp ? tcp->ports.src : 0; - t2 = is_tcp_udp ? tcp->ports.dst : 0; + t1 = is_tcp_udp ? tcp->src : 0; + t2 = is_tcp_udp ? tcp->dst : 0; t1 = (flow_hash_config & IP_FLOW_HASH_SRC_PORT) ? t1 : 0; t2 = (flow_hash_config & IP_FLOW_HASH_DST_PORT) ? t2 : 0; @@ -497,6 +497,46 @@ int ip6_hbh_register_option (u8 option, int ip6_hbh_unregister_option (u8 option); void ip6_hbh_set_next_override (uword next); +/** + * Push IPv6 header to buffer + * + * @param vm - vlib_main + * @param b - buffer to write the header to + * @param src - source IP + * @param dst - destination IP + * @param prot - payload proto + * + * @return - pointer to start of IP header + */ +always_inline void * +vlib_buffer_push_ip6 (vlib_main_t * vm, vlib_buffer_t * b, + ip6_address_t * src, ip6_address_t * dst, int proto) +{ + ip6_header_t *ip6h; + u16 payload_length; + + /* make some room */ + ip6h = vlib_buffer_push_uninit (b, sizeof (ip6_header_t)); + + ip6h->ip_version_traffic_class_and_flow_label = + clib_host_to_net_u32 (0x6 << 28); + + /* calculate ip6 payload length */ + payload_length = vlib_buffer_length_in_chain (vm, b); + payload_length -= sizeof (*ip6h); + + ip6h->payload_length = clib_host_to_net_u16 (payload_length); + + ip6h->hop_limit = 0xff; + ip6h->protocol = proto; + clib_memcpy (ip6h->src_address.as_u8, src->as_u8, + sizeof (ip6h->src_address)); + clib_memcpy (ip6h->dst_address.as_u8, dst->as_u8, + sizeof (ip6h->src_address)); + + return ip6h; +} + #endif /* included_ip_ip6_h */ /* diff --git a/src/vnet/ip/ip6_packet.h b/src/vnet/ip/ip6_packet.h index 1e551c8b67d..4fd14b960e4 100644 --- a/src/vnet/ip/ip6_packet.h +++ b/src/vnet/ip/ip6_packet.h @@ -40,7 +40,7 @@ #ifndef included_ip6_packet_h #define included_ip6_packet_h -#include <vnet/ip/tcp_packet.h> +#include <vnet/tcp/tcp_packet.h> #include <vnet/ip/ip4_packet.h> typedef union @@ -373,10 +373,10 @@ ip6_tcp_reply_x1 (ip6_header_t * ip0, tcp_header_t * tcp0) { u16 src0, dst0; - src0 = tcp0->ports.src; - dst0 = tcp0->ports.dst; - tcp0->ports.src = dst0; - tcp0->ports.dst = src0; + src0 = tcp0->src; + dst0 = tcp0->dst; + tcp0->src = dst0; + tcp0->dst = src0; } } @@ -400,14 +400,14 @@ ip6_tcp_reply_x2 (ip6_header_t * ip0, ip6_header_t * ip1, { u16 src0, dst0, src1, dst1; - src0 = tcp0->ports.src; - src1 = tcp1->ports.src; - dst0 = tcp0->ports.dst; - dst1 = tcp1->ports.dst; - tcp0->ports.src = dst0; - tcp1->ports.src = dst1; - tcp0->ports.dst = src0; - tcp1->ports.dst = src1; + src0 = tcp0->src; + src1 = tcp1->src; + dst0 = tcp0->dst; + dst1 = tcp1->dst; + tcp0->src = dst0; + tcp1->src = dst1; + tcp0->dst = src0; + tcp1->dst = src1; } } diff --git a/src/vnet/ip/punt.c b/src/vnet/ip/punt.c index 9c735128a3b..48558401292 100644 --- a/src/vnet/ip/punt.c +++ b/src/vnet/ip/punt.c @@ -23,7 +23,7 @@ */ #include <vlib/vlib.h> #include <vnet/pg/pg.h> -#include <vnet/ip/udp.h> +#include <vnet/udp/udp.h> #include <vnet/ip/punt.h> #define foreach_punt_next \ diff --git a/src/vnet/ip/tcp_packet.h b/src/vnet/ip/tcp_packet.h deleted file mode 100644 index 93f73e010aa..00000000000 --- a/src/vnet/ip/tcp_packet.h +++ /dev/null @@ -1,141 +0,0 @@ -/* - * Copyright (c) 2015 Cisco and/or its affiliates. - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at: - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ -/* - * ip4/tcp_packet.h: TCP packet format (see RFC 793) - * - * Copyright (c) 2008 Eliot Dresselhaus - * - * Permission is hereby granted, free of charge, to any person obtaining - * a copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sublicense, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice shall be - * included in all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, - * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND - * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE - * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION - * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION - * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - */ - -#ifndef included_tcp_packet_h -#define included_tcp_packet_h - -/* TCP flags bit 0 first. */ -#define foreach_tcp_flag \ - _ (FIN) \ - _ (SYN) \ - _ (RST) \ - _ (PSH) \ - _ (ACK) \ - _ (URG) \ - _ (ECE) \ - _ (CWR) - -enum -{ -#define _(f) TCP_FLAG_BIT_##f, - foreach_tcp_flag -#undef _ - TCP_N_FLAG_BITS, - -#define _(f) TCP_FLAG_##f = 1 << TCP_FLAG_BIT_##f, - foreach_tcp_flag -#undef _ -}; - -typedef struct -{ - /* Source and destination port. */ - union - { - union - { - struct - { - u16 src, dst; - }; - u32 src_and_dst; - } ports; - struct - { - u16 src_port, dst_port; - }; - }; - - /* Sequence and acknowledgment number. */ - u32 seq_number, ack_number; - - /* Size of TCP header in 32-bit units plus 4 reserved bits. */ - u8 tcp_header_u32s_and_reserved; - - /* see foreach_tcp_flag for enumation of tcp flags. */ - u8 flags; - - /* Current window advertised by sender. - This is the number of bytes sender is willing to receive - right now. */ - u16 window; - - /* Checksum of TCP pseudo header and data. */ - u16 checksum; - - u16 urgent_pointer; -} tcp_header_t; - -always_inline int -tcp_header_bytes (tcp_header_t * t) -{ - return (t->tcp_header_u32s_and_reserved >> 4) * sizeof (u32); -} - -/* TCP options. */ -typedef enum tcp_option_type -{ - TCP_OPTION_END = 0, - TCP_OPTION_NOP = 1, - TCP_OPTION_MSS = 2, - TCP_OPTION_WINDOW_SCALE = 3, - TCP_OPTION_SACK_PERMITTED = 4, - TCP_OPTION_SACK_BLOCK = 5, - TCP_OPTION_TIME_STAMP = 8, -} tcp_option_type_t; - -/* All except NOP and END have 1 byte length field. */ -typedef struct -{ - tcp_option_type_t type:8; - - /* Length of this option in bytes. */ - u8 length; -} tcp_option_with_length_t; - -#endif /* included_tcp_packet_h */ - - -/* - * fd.io coding-style-patch-verification: ON - * - * Local Variables: - * eval: (c-set-style "gnu") - * End: - */ diff --git a/src/vnet/ip/udp_init.c b/src/vnet/ip/udp_init.c deleted file mode 100644 index 1241ca4ab32..00000000000 --- a/src/vnet/ip/udp_init.c +++ /dev/null @@ -1,71 +0,0 @@ -/* - * Copyright (c) 2015 Cisco and/or its affiliates. - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at: - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ -/* - * ip/udp_init.c: udp initialization - * - * Copyright (c) 2008 Eliot Dresselhaus - * - * Permission is hereby granted, free of charge, to any person obtaining - * a copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sublicense, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice shall be - * included in all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, - * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND - * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE - * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION - * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION - * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - */ - -#include <vnet/ip/ip.h> - -clib_error_t * -udp_init (vlib_main_t * vm) -{ - ip_main_t *im = &ip_main; - ip_protocol_info_t *pi; - clib_error_t *error; - - error = vlib_call_init_function (vm, ip_main_init); - - if (!error) - { - pi = ip_get_protocol_info (im, IP_PROTOCOL_UDP); - if (pi == 0) - return clib_error_return (0, "UDP protocol info AWOL"); - pi->format_header = format_udp_header; - pi->unformat_pg_edit = unformat_pg_udp_header; - } - - return 0; -} - -VLIB_INIT_FUNCTION (udp_init); - -/* - * fd.io coding-style-patch-verification: ON - * - * Local Variables: - * eval: (c-set-style "gnu") - * End: - */ diff --git a/src/vnet/ipsec/ikev2.c b/src/vnet/ipsec/ikev2.c index 092093346fd..2c1074d814b 100644 --- a/src/vnet/ipsec/ikev2.c +++ b/src/vnet/ipsec/ikev2.c @@ -17,7 +17,7 @@ #include <vnet/vnet.h> #include <vnet/pg/pg.h> #include <vppinfra/error.h> -#include <vnet/ip/udp.h> +#include <vnet/udp/udp.h> #include <vnet/ipsec/ipsec.h> #include <vnet/ipsec/ikev2.h> #include <vnet/ipsec/ikev2_priv.h> diff --git a/src/vnet/ipsec/ikev2_cli.c b/src/vnet/ipsec/ikev2_cli.c index 5c88d8d4681..05ed4e6067f 100644 --- a/src/vnet/ipsec/ikev2_cli.c +++ b/src/vnet/ipsec/ikev2_cli.c @@ -16,7 +16,7 @@ #include <vnet/vnet.h> #include <vnet/pg/pg.h> #include <vppinfra/error.h> -#include <vnet/ip/udp.h> +#include <vnet/udp/udp.h> #include <vnet/ipsec/ikev2.h> #include <vnet/ipsec/ikev2_priv.h> diff --git a/src/vnet/ipsec/ikev2_crypto.c b/src/vnet/ipsec/ikev2_crypto.c index c201d3eb41f..ca56158f898 100644 --- a/src/vnet/ipsec/ikev2_crypto.c +++ b/src/vnet/ipsec/ikev2_crypto.c @@ -17,7 +17,7 @@ #include <vnet/vnet.h> #include <vnet/pg/pg.h> #include <vppinfra/error.h> -#include <vnet/ip/udp.h> +#include <vnet/udp/udp.h> #include <vnet/ipsec/ikev2.h> #include <vnet/ipsec/ikev2_priv.h> #include <openssl/obj_mac.h> diff --git a/src/vnet/lisp-cp/packets.c b/src/vnet/lisp-cp/packets.c index 3a4f421b02a..f24024f1500 100644 --- a/src/vnet/lisp-cp/packets.c +++ b/src/vnet/lisp-cp/packets.c @@ -15,7 +15,7 @@ #include <vnet/lisp-cp/packets.h> #include <vnet/lisp-cp/lisp_cp_messages.h> -#include <vnet/ip/udp_packet.h> +#include <vnet/udp/udp_packet.h> /* Returns IP ID for the packet */ /* static u16 ip_id = 0; @@ -142,61 +142,6 @@ pkt_push_udp (vlib_main_t * vm, vlib_buffer_t * b, u16 sp, u16 dp) } void * -pkt_push_ipv4 (vlib_main_t * vm, vlib_buffer_t * b, ip4_address_t * src, - ip4_address_t * dst, int proto) -{ - ip4_header_t *ih; - - /* make some room */ - ih = vlib_buffer_push_uninit (b, sizeof (ip4_header_t)); - - ih->ip_version_and_header_length = 0x45; - ih->tos = 0; - ih->length = clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b)); - - /* iph->fragment_id = clib_host_to_net_u16(get_IP_ID ()); */ - - /* TODO: decide if we allow fragments in case of control */ - ih->flags_and_fragment_offset = clib_host_to_net_u16 (IP_DF); - ih->ttl = 255; - ih->protocol = proto; - ih->src_address.as_u32 = src->as_u32; - ih->dst_address.as_u32 = dst->as_u32; - - ih->checksum = ip4_header_checksum (ih); - return ih; -} - -void * -pkt_push_ipv6 (vlib_main_t * vm, vlib_buffer_t * b, ip6_address_t * src, - ip6_address_t * dst, int proto) -{ - ip6_header_t *ip6h; - u16 payload_length; - - /* make some room */ - ip6h = vlib_buffer_push_uninit (b, sizeof (ip6_header_t)); - - ip6h->ip_version_traffic_class_and_flow_label = - clib_host_to_net_u32 (0x6 << 28); - - /* calculate ip6 payload length */ - payload_length = vlib_buffer_length_in_chain (vm, b); - payload_length -= sizeof (*ip6h); - - ip6h->payload_length = clib_host_to_net_u16 (payload_length); - - ip6h->hop_limit = 0xff; - ip6h->protocol = proto; - clib_memcpy (ip6h->src_address.as_u8, src->as_u8, - sizeof (ip6h->src_address)); - clib_memcpy (ip6h->dst_address.as_u8, dst->as_u8, - sizeof (ip6h->src_address)); - - return ip6h; -} - -void * pkt_push_ip (vlib_main_t * vm, vlib_buffer_t * b, ip_address_t * src, ip_address_t * dst, u32 proto) { @@ -210,12 +155,12 @@ pkt_push_ip (vlib_main_t * vm, vlib_buffer_t * b, ip_address_t * src, switch (ip_addr_version (src)) { case IP4: - return pkt_push_ipv4 (vm, b, &ip_addr_v4 (src), &ip_addr_v4 (dst), - proto); + return vlib_buffer_push_ip4 (vm, b, &ip_addr_v4 (src), + &ip_addr_v4 (dst), proto); break; case IP6: - return pkt_push_ipv6 (vm, b, &ip_addr_v6 (src), &ip_addr_v6 (dst), - proto); + return vlib_buffer_push_ip6 (vm, b, &ip_addr_v6 (src), + &ip_addr_v6 (dst), proto); break; } diff --git a/src/vnet/lisp-cp/packets.h b/src/vnet/lisp-cp/packets.h index 212a1d78163..f6da3bf4b87 100644 --- a/src/vnet/lisp-cp/packets.h +++ b/src/vnet/lisp-cp/packets.h @@ -26,51 +26,6 @@ void *pkt_push_udp_and_ip (vlib_main_t * vm, vlib_buffer_t * b, u16 sp, void *pkt_push_ecm_hdr (vlib_buffer_t * b); -always_inline u8 * -vlib_buffer_get_tail (vlib_buffer_t * b) -{ - return b->data + b->current_data + b->current_length; -} - -always_inline void * -vlib_buffer_put_uninit (vlib_buffer_t * b, u8 size) -{ - /* XXX should make sure there's enough space! */ - void *p = vlib_buffer_get_tail (b); - b->current_length += size; - return p; -} - -always_inline void * -vlib_buffer_push_uninit (vlib_buffer_t * b, u8 size) -{ - /* XXX should make sure there's enough space! */ - ASSERT (b->current_data >= size); - b->current_data -= size; - b->current_length += size; - - return vlib_buffer_get_current (b); -} - -always_inline void * -vlib_buffer_make_headroom (vlib_buffer_t * b, u8 size) -{ - /* XXX should make sure there's enough space! */ - b->current_data += size; - return vlib_buffer_get_current (b); -} - -always_inline void * -vlib_buffer_pull (vlib_buffer_t * b, u8 size) -{ - if (b->current_length < size) - return 0; - - void *data = vlib_buffer_get_current (b); - vlib_buffer_advance (b, size); - return data; -} - /* *INDENT-ON* */ /* diff --git a/src/vnet/lisp-gpe/interface.c b/src/vnet/lisp-gpe/interface.c index 13359277700..292c7e6a74e 100644 --- a/src/vnet/lisp-gpe/interface.c +++ b/src/vnet/lisp-gpe/interface.c @@ -23,7 +23,7 @@ #include <vppinfra/hash.h> #include <vnet/vnet.h> #include <vnet/ip/ip.h> -#include <vnet/ip/udp.h> +#include <vnet/udp/udp.h> #include <vnet/ethernet/ethernet.h> #include <vnet/lisp-gpe/lisp_gpe.h> #include <vnet/lisp-gpe/lisp_gpe_fwd_entry.h> diff --git a/src/vnet/lisp-gpe/lisp_gpe.h b/src/vnet/lisp-gpe/lisp_gpe.h index c898a7daf6b..b5a50ec6884 100644 --- a/src/vnet/lisp-gpe/lisp_gpe.h +++ b/src/vnet/lisp-gpe/lisp_gpe.h @@ -27,10 +27,12 @@ #include <vnet/l2/l2_input.h> #include <vnet/ethernet/ethernet.h> #include <vnet/ip/ip4_packet.h> -#include <vnet/ip/udp.h> +#include <vnet/udp/udp.h> #include <vnet/lisp-cp/lisp_types.h> #include <vnet/lisp-gpe/lisp_gpe_packet.h> #include <vnet/adj/adj_types.h> +#include <vppinfra/bihash_24_8.h> +#include <vppinfra/bihash_template.h> /** IP4-UDP-LISP encap header */ /* *INDENT-OFF* */ diff --git a/src/vnet/lisp-gpe/lisp_gpe_adjacency.c b/src/vnet/lisp-gpe/lisp_gpe_adjacency.c index 65006b81f1d..dbcf7134103 100644 --- a/src/vnet/lisp-gpe/lisp_gpe_adjacency.c +++ b/src/vnet/lisp-gpe/lisp_gpe_adjacency.c @@ -25,6 +25,8 @@ #include <vnet/lisp-gpe/lisp_gpe_tunnel.h> #include <vnet/fib/fib_entry.h> #include <vnet/adj/adj_midchain.h> +#include <vppinfra/bihash_24_8.h> +#include <vppinfra/bihash_template.h> /** * Memory pool of all adjacencies diff --git a/src/vnet/session/application.c b/src/vnet/session/application.c new file mode 100644 index 00000000000..a561e7d17f0 --- /dev/null +++ b/src/vnet/session/application.c @@ -0,0 +1,343 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/session/application.h> +#include <vnet/session/session.h> + +/* + * Pool from which we allocate all applications + */ +static application_t *app_pool; + +/* + * Hash table of apps by api client index + */ +static uword *app_by_api_client_index; + +int +application_api_queue_is_full (application_t * app) +{ + unix_shared_memory_queue_t *q; + + /* builtin servers are always OK */ + if (app->api_client_index == ~0) + return 0; + + q = vl_api_client_index_to_input_queue (app->api_client_index); + if (!q) + return 1; + + if (q->cursize == q->maxsize) + return 1; + return 0; +} + +static void +application_table_add (application_t * app) +{ + hash_set (app_by_api_client_index, app->api_client_index, app->index); +} + +static void +application_table_del (application_t * app) +{ + hash_unset (app_by_api_client_index, app->api_client_index); +} + +application_t * +application_lookup (u32 api_client_index) +{ + uword *p; + p = hash_get (app_by_api_client_index, api_client_index); + if (p) + return application_get (p[0]); + + return 0; +} + +void +application_del (application_t * app) +{ + session_manager_main_t *smm = vnet_get_session_manager_main (); + api_main_t *am = &api_main; + void *oldheap; + session_manager_t *sm; + + if (app->mode == APP_SERVER) + { + sm = session_manager_get (app->session_manager_index); + session_manager_del (smm, sm); + } + + /* Free the event fifo in the /vpe-api shared-memory segment */ + oldheap = svm_push_data_heap (am->vlib_rp); + if (app->event_queue) + unix_shared_memory_queue_free (app->event_queue); + svm_pop_heap (oldheap); + + application_table_del (app); + + pool_put (app_pool, app); +} + +application_t * +application_new (application_type_t type, session_type_t sst, + u32 api_client_index, u32 flags, session_cb_vft_t * cb_fns) +{ + session_manager_main_t *smm = vnet_get_session_manager_main (); + api_main_t *am = &api_main; + application_t *app; + void *oldheap; + session_manager_t *sm; + + pool_get (app_pool, app); + memset (app, 0, sizeof (*app)); + + /* Allocate event fifo in the /vpe-api shared-memory segment */ + oldheap = svm_push_data_heap (am->vlib_rp); + + /* Allocate server event queue */ + app->event_queue = + unix_shared_memory_queue_init (128 /* nels $$$$ config */ , + sizeof (session_fifo_event_t), + 0 /* consumer pid */ , + 0 + /* (do not) signal when queue non-empty */ + ); + + svm_pop_heap (oldheap); + + /* If a server, allocate session manager */ + if (type == APP_SERVER) + { + pool_get (smm->session_managers, sm); + memset (sm, 0, sizeof (*sm)); + + app->session_manager_index = sm - smm->session_managers; + } + else if (type == APP_CLIENT) + { + /* Allocate connect session manager if needed */ + if (smm->connect_manager_index[sst] == INVALID_INDEX) + connects_session_manager_init (smm, sst); + app->session_manager_index = smm->connect_manager_index[sst]; + } + + app->mode = type; + app->index = application_get_index (app); + app->session_type = sst; + app->api_client_index = api_client_index; + app->flags = flags; + app->cb_fns = *cb_fns; + + /* Add app to lookup by api_client_index table */ + application_table_add (app); + + return app; +} + +application_t * +application_get (u32 index) +{ + return pool_elt_at_index (app_pool, index); +} + +u32 +application_get_index (application_t * app) +{ + return app - app_pool; +} + +int +application_server_init (application_t * server, u32 segment_size, + u32 add_segment_size, u32 rx_fifo_size, + u32 tx_fifo_size, u8 ** segment_name) +{ + session_manager_main_t *smm = vnet_get_session_manager_main (); + session_manager_t *sm; + int rv; + + sm = session_manager_get (server->session_manager_index); + + /* Add first segment */ + if ((rv = session_manager_add_first_segment (smm, sm, segment_size, + segment_name))) + { + return rv; + } + + /* Setup session manager */ + sm->add_segment_size = add_segment_size; + sm->rx_fifo_size = rx_fifo_size; + sm->tx_fifo_size = tx_fifo_size; + sm->add_segment = sm->add_segment_size != 0; + return 0; +} + +u8 * +format_application_server (u8 * s, va_list * args) +{ + application_t *srv = va_arg (*args, application_t *); + int verbose = va_arg (*args, int); + vl_api_registration_t *regp; + stream_session_t *listener; + u8 *server_name, *str, *seg_name; + u32 segment_size; + + if (srv == 0) + { + if (verbose) + s = format (s, "%-40s%-20s%-15s%-15s%-10s", "Connection", "Server", + "Segment", "API Client", "Cookie"); + else + s = format (s, "%-40s%-20s", "Connection", "Server"); + + return s; + } + + regp = vl_api_client_index_to_registration (srv->api_client_index); + if (!regp) + server_name = format (0, "%s%c", regp->name, 0); + else + server_name = regp->name; + + listener = stream_session_listener_get (srv->session_type, + srv->session_index); + str = format (0, "%U", format_stream_session, listener, verbose); + + session_manager_get_segment_info (listener->server_segment_index, &seg_name, + &segment_size); + if (verbose) + { + s = format (s, "%-40s%-20s%-20s%-10d%-10d", str, server_name, + seg_name, srv->api_client_index, srv->accept_cookie); + } + else + s = format (s, "%-40s%-20s", str, server_name); + return s; +} + +u8 * +format_application_client (u8 * s, va_list * args) +{ + application_t *client = va_arg (*args, application_t *); + int verbose = va_arg (*args, int); + stream_session_t *session; + u8 *str, *seg_name; + u32 segment_size; + + if (client == 0) + { + if (verbose) + s = + format (s, "%-40s%-20s%-10s", "Connection", "Segment", + "API Client"); + else + s = format (s, "%-40s", "Connection"); + + return s; + } + + session = stream_session_get (client->session_index, client->thread_index); + str = format (0, "%U", format_stream_session, session, verbose); + + session_manager_get_segment_info (session->server_segment_index, &seg_name, + &segment_size); + if (verbose) + { + s = format (s, "%-40s%-20s%-10d%", str, seg_name, + client->api_client_index); + } + else + s = format (s, "%-40s", str); + return s; +} + +static clib_error_t * +show_app_command_fn (vlib_main_t * vm, unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + application_t *app; + int do_server = 0; + int do_client = 0; + int verbose = 0; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "server")) + do_server = 1; + else if (unformat (input, "client")) + do_client = 1; + else if (unformat (input, "verbose")) + verbose = 1; + else + break; + } + + if (do_server) + { + if (pool_elts (app_pool)) + { + vlib_cli_output (vm, "%U", format_application_server, + 0 /* header */ , + verbose); + /* *INDENT-OFF* */ + pool_foreach (app, app_pool, + ({ + if (app->mode == APP_SERVER) + vlib_cli_output (vm, "%U", format_application_server, app, + verbose); + })); + /* *INDENT-ON* */ + } + else + vlib_cli_output (vm, "No active server bindings"); + } + + if (do_client) + { + if (pool_elts (app_pool)) + { + vlib_cli_output (vm, "%U", format_application_client, + 0 /* header */ , + verbose); + /* *INDENT-OFF* */ + pool_foreach (app, app_pool, + ({ + if (app->mode == APP_CLIENT) + vlib_cli_output (vm, "%U", format_application_client, app, + verbose); + })); + /* *INDENT-ON* */ + } + else + vlib_cli_output (vm, "No active server bindings"); + } + + return 0; +} + +VLIB_CLI_COMMAND (show_app_command, static) = +{ +.path = "show app",.short_help = + "show app [server|client] [verbose]",.function = show_app_command_fn,}; + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/application.h b/src/vnet/session/application.h new file mode 100644 index 00000000000..027d6967fd7 --- /dev/null +++ b/src/vnet/session/application.h @@ -0,0 +1,120 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef SRC_VNET_SESSION_APPLICATION_H_ +#define SRC_VNET_SESSION_APPLICATION_H_ + +#include <vnet/vnet.h> +#include <vnet/session/session.h> + +typedef enum +{ + APP_SERVER, + APP_CLIENT +} application_type_t; + +typedef struct _stream_session_cb_vft +{ + /** Notify server of new segment */ + int (*add_segment_callback) (u32 api_client_index, const u8 * seg_name, + u32 seg_size); + + /** Notify server of newly accepted session */ + int (*session_accept_callback) (stream_session_t * new_session); + + /* Connection request callback */ + int (*session_connected_callback) (u32 api_client_index, + stream_session_t * s, u8 code); + + /** Notify app that session is closing */ + void (*session_disconnect_callback) (stream_session_t * s); + + /** Notify app that session was reset */ + void (*session_reset_callback) (stream_session_t * s); + + /* Direct RX callback, for built-in servers */ + int (*builtin_server_rx_callback) (stream_session_t * session); + + /* Redirect connection to local server */ + int (*redirect_connect_callback) (u32 api_client_index, void *mp); +} session_cb_vft_t; + +typedef struct _application +{ + /** Index in server pool */ + u32 index; + + /** Flags */ + u32 flags; + + /** Binary API connection index, ~0 if internal */ + u32 api_client_index; + + /* */ + u32 api_context; + + /** Application listens for events on this svm queue */ + unix_shared_memory_queue_t *event_queue; + + /** Stream session type */ + u8 session_type; + + /* Stream server mode: accept or connect */ + u8 mode; + + u32 session_manager_index; + + /* + * Bind/Listen specific + */ + + /** Accept cookie, for multiple session flavors ($$$ maybe) */ + u32 accept_cookie; + + /** Index of the listen session or connect session */ + u32 session_index; + + /** Session thread index for client connect sessions */ + u32 thread_index; + + /* + * Callbacks: shoulder-taps for the server/client + */ + session_cb_vft_t cb_fns; +} application_t; + +application_t *application_new (application_type_t type, session_type_t sst, + u32 api_client_index, u32 flags, + session_cb_vft_t * cb_fns); +void application_del (application_t * app); +application_t *application_get (u32 index); +application_t *application_lookup (u32 api_client_index); +u32 application_get_index (application_t * app); + +int +application_server_init (application_t * server, u32 segment_size, + u32 add_segment_size, u32 rx_fifo_size, + u32 tx_fifo_size, u8 ** segment_name); +int application_api_queue_is_full (application_t * app); + +#endif /* SRC_VNET_SESSION_APPLICATION_H_ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/application_interface.c b/src/vnet/session/application_interface.c new file mode 100644 index 00000000000..0ea77fd8c4e --- /dev/null +++ b/src/vnet/session/application_interface.c @@ -0,0 +1,459 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#include <vnet/session/application_interface.h> + +#include <vnet/session/session.h> +#include <vlibmemory/api.h> +#include <vnet/dpo/load_balance.h> +#include <vnet/fib/ip4_fib.h> + +/** @file + VPP's application/session API bind/unbind/connect/disconnect calls +*/ + +static u8 +ip_is_zero (ip46_address_t * ip46_address, u8 is_ip4) +{ + if (is_ip4) + return (ip46_address->ip4.as_u32 == 0); + else + return (ip46_address->as_u64[0] == 0 && ip46_address->as_u64[1] == 0); +} + +static u8 +ip_is_local (ip46_address_t * ip46_address, u8 is_ip4) +{ + fib_node_index_t fei; + fib_entry_flag_t flags; + fib_prefix_t prefix; + + /* Check if requester is local */ + if (is_ip4) + { + prefix.fp_len = 32; + prefix.fp_proto = FIB_PROTOCOL_IP4; + } + else + { + prefix.fp_len = 128; + prefix.fp_proto = FIB_PROTOCOL_IP6; + } + + clib_memcpy (&prefix.fp_addr, ip46_address, sizeof (ip46_address)); + fei = fib_table_lookup (0, &prefix); + flags = fib_entry_get_flags (fei); + + return (flags & FIB_ENTRY_FLAG_LOCAL); +} + +int +api_parse_session_handle (u64 handle, u32 * session_index, u32 * thread_index) +{ + session_manager_main_t *smm = vnet_get_session_manager_main (); + stream_session_t *pool; + + *thread_index = handle & 0xFFFFFFFF; + *session_index = handle >> 32; + + if (*thread_index >= vec_len (smm->sessions)) + return VNET_API_ERROR_INVALID_VALUE; + + pool = smm->sessions[*thread_index]; + + if (pool_is_free_index (pool, *session_index)) + return VNET_API_ERROR_INVALID_VALUE_2; + + return 0; +} + +int +vnet_bind_i (u32 api_client_index, ip46_address_t * ip46, u16 port_host_order, + session_type_t sst, u64 * options, session_cb_vft_t * cb_fns, + application_t ** app, u32 * len_seg_name, char *seg_name) +{ + u8 *segment_name = 0; + application_t *server = 0; + stream_session_t *listener; + u8 is_ip4; + + listener = + stream_session_lookup_listener (ip46, + clib_host_to_net_u16 (port_host_order), + sst); + + if (listener) + return VNET_API_ERROR_ADDRESS_IN_USE; + + if (application_lookup (api_client_index)) + { + clib_warning ("Only one bind supported for now"); + return VNET_API_ERROR_ADDRESS_IN_USE; + } + + is_ip4 = SESSION_TYPE_IP4_UDP == sst || SESSION_TYPE_IP4_TCP == sst; + if (!ip_is_zero (ip46, is_ip4) && !ip_is_local (ip46, is_ip4)) + return VNET_API_ERROR_INVALID_VALUE; + + /* Allocate and initialize stream server */ + server = application_new (APP_SERVER, sst, api_client_index, + options[SESSION_OPTIONS_FLAGS], cb_fns); + + application_server_init (server, options[SESSION_OPTIONS_SEGMENT_SIZE], + options[SESSION_OPTIONS_ADD_SEGMENT_SIZE], + options[SESSION_OPTIONS_RX_FIFO_SIZE], + options[SESSION_OPTIONS_TX_FIFO_SIZE], + &segment_name); + + /* Setup listen path down to transport */ + stream_session_start_listen (server->index, ip46, port_host_order); + + /* + * Return values + */ + + ASSERT (vec_len (segment_name) <= 128); + *len_seg_name = vec_len (segment_name); + memcpy (seg_name, segment_name, *len_seg_name); + *app = server; + + return 0; +} + +int +vnet_unbind_i (u32 api_client_index) +{ + application_t *server; + + /* + * Find the stream_server_t corresponding to the api client + */ + server = application_lookup (api_client_index); + if (!server) + return VNET_API_ERROR_INVALID_VALUE_2; + + /* Clear the listener */ + stream_session_stop_listen (server->index); + application_del (server); + + return 0; +} + +int +vnet_connect_i (u32 api_client_index, u32 api_context, session_type_t sst, + ip46_address_t * ip46, u16 port, u64 * options, void *mp, + session_cb_vft_t * cb_fns) +{ + stream_session_t *listener; + application_t *server, *app; + + /* + * Figure out if connecting to a local server + */ + listener = stream_session_lookup_listener (ip46, + clib_host_to_net_u16 (port), + sst); + if (listener) + { + server = application_get (listener->app_index); + + /* + * Server is willing to have a direct fifo connection created + * instead of going through the state machine, etc. + */ + if (server->flags & SESSION_OPTIONS_FLAGS_USE_FIFO) + return server->cb_fns. + redirect_connect_callback (server->api_client_index, mp); + } + + /* Create client app */ + app = application_new (APP_CLIENT, sst, api_client_index, + options[SESSION_OPTIONS_FLAGS], cb_fns); + + app->api_context = api_context; + + /* + * Not connecting to a local server. Create regular session + */ + stream_session_open (sst, ip46, port, app->index); + + return 0; +} + +/** + * unformat a vnet URI + * + * fifo://name + * tcp://ip46-addr:port + * udp://ip46-addr:port + * + * u8 ip46_address[16]; + * u16 port_in_host_byte_order; + * stream_session_type_t sst; + * u8 *fifo_name; + * + * if (unformat (input, "%U", unformat_vnet_uri, &ip46_address, + * &sst, &port, &fifo_name)) + * etc... + * + */ +uword +unformat_vnet_uri (unformat_input_t * input, va_list * args) +{ + ip46_address_t *address = va_arg (*args, ip46_address_t *); + session_type_t *sst = va_arg (*args, session_type_t *); + u16 *port = va_arg (*args, u16 *); + + if (unformat (input, "tcp://%U/%d", unformat_ip4_address, &address->ip4, + port)) + { + *sst = SESSION_TYPE_IP4_TCP; + return 1; + } + if (unformat (input, "udp://%U/%d", unformat_ip4_address, &address->ip4, + port)) + { + *sst = SESSION_TYPE_IP4_UDP; + return 1; + } + if (unformat (input, "udp://%U/%d", unformat_ip6_address, &address->ip6, + port)) + { + *sst = SESSION_TYPE_IP6_UDP; + return 1; + } + if (unformat (input, "tcp://%U/%d", unformat_ip6_address, &address->ip6, + port)) + { + *sst = SESSION_TYPE_IP6_TCP; + return 1; + } + + return 0; +} + +int +parse_uri (char *uri, session_type_t * sst, ip46_address_t * addr, + u16 * port_number_host_byte_order) +{ + unformat_input_t _input, *input = &_input; + + /* Make sure */ + uri = (char *) format (0, "%s%c", uri, 0); + + /* Parse uri */ + unformat_init_string (input, uri, strlen (uri)); + if (!unformat (input, "%U", unformat_vnet_uri, addr, sst, + port_number_host_byte_order)) + { + unformat_free (input); + return VNET_API_ERROR_INVALID_VALUE; + } + unformat_free (input); + + return 0; +} + +int +vnet_bind_uri (vnet_bind_args_t * a) +{ + application_t *server = 0; + u16 port_host_order; + session_type_t sst = SESSION_N_TYPES; + ip46_address_t ip46; + int rv; + + memset (&ip46, 0, sizeof (ip46)); + rv = parse_uri (a->uri, &sst, &ip46, &port_host_order); + if (rv) + return rv; + + if ((rv = vnet_bind_i (a->api_client_index, &ip46, port_host_order, sst, + a->options, a->session_cb_vft, &server, + &a->segment_name_length, a->segment_name))) + return rv; + + a->server_event_queue_address = (u64) server->event_queue; + return 0; +} + +session_type_t +session_type_from_proto_and_ip (session_api_proto_t proto, u8 is_ip4) +{ + if (proto == SESSION_PROTO_TCP) + { + if (is_ip4) + return SESSION_TYPE_IP4_TCP; + else + return SESSION_TYPE_IP6_TCP; + } + else + { + if (is_ip4) + return SESSION_TYPE_IP4_UDP; + else + return SESSION_TYPE_IP6_UDP; + } + + return SESSION_N_TYPES; +} + +int +vnet_unbind_uri (char *uri, u32 api_client_index) +{ + u16 port_number_host_byte_order; + session_type_t sst = SESSION_N_TYPES; + ip46_address_t ip46_address; + stream_session_t *listener; + int rv; + + rv = parse_uri (uri, &sst, &ip46_address, &port_number_host_byte_order); + if (rv) + return rv; + + listener = + stream_session_lookup_listener (&ip46_address, + clib_host_to_net_u16 + (port_number_host_byte_order), sst); + + if (!listener) + return VNET_API_ERROR_ADDRESS_NOT_IN_USE; + + /* External client? */ + if (api_client_index != ~0) + { + ASSERT (vl_api_client_index_to_registration (api_client_index)); + } + + return vnet_unbind_i (api_client_index); +} + +int +vnet_connect_uri (vnet_connect_args_t * a) +{ + ip46_address_t ip46_address; + u16 port; + session_type_t sst; + application_t *app; + int rv; + + app = application_lookup (a->api_client_index); + if (app) + { + clib_warning ("Already have a connect from this app"); + return VNET_API_ERROR_INVALID_VALUE_2; + } + + /* Parse uri */ + rv = parse_uri (a->uri, &sst, &ip46_address, &port); + if (rv) + return rv; + + return vnet_connect_i (a->api_client_index, a->api_context, sst, + &ip46_address, port, a->options, a->mp, + a->session_cb_vft); +} + +int +vnet_disconnect_session (u32 client_index, u32 session_index, + u32 thread_index) +{ + stream_session_t *session; + + session = stream_session_get (session_index, thread_index); + stream_session_disconnect (session); + + return 0; +} + + +int +vnet_bind (vnet_bind_args_t * a) +{ + application_t *server = 0; + session_type_t sst = SESSION_N_TYPES; + int rv; + + sst = session_type_from_proto_and_ip (a->proto, a->tep.is_ip4); + if ((rv = vnet_bind_i (a->api_client_index, &a->tep.ip, a->tep.port, sst, + a->options, a->session_cb_vft, &server, + &a->segment_name_length, a->segment_name))) + return rv; + + a->server_event_queue_address = (u64) server->event_queue; + a->handle = (u64) a->tep.vrf << 32 | (u64) server->session_index; + return 0; +} + +int +vnet_unbind (vnet_unbind_args_t * a) +{ + application_t *server; + + if (a->api_client_index != ~0) + { + ASSERT (vl_api_client_index_to_registration (a->api_client_index)); + } + + /* Make sure this is the right one */ + server = application_lookup (a->api_client_index); + ASSERT (server->session_index == (0xFFFFFFFF & a->handle)); + + /* TODO use handle to disambiguate namespaces/vrfs */ + return vnet_unbind_i (a->api_client_index); +} + +int +vnet_connect (vnet_connect_args_t * a) +{ + session_type_t sst; + application_t *app; + + app = application_lookup (a->api_client_index); + if (app) + { + clib_warning ("Already have a connect from this app"); + return VNET_API_ERROR_INVALID_VALUE_2; + } + + sst = session_type_from_proto_and_ip (a->proto, a->tep.is_ip4); + return vnet_connect_i (a->api_client_index, a->api_context, sst, &a->tep.ip, + a->tep.port, a->options, a->mp, a->session_cb_vft); +} + +int +vnet_disconnect (vnet_disconnect_args_t * a) +{ + stream_session_t *session; + u32 session_index, thread_index; + + if (api_parse_session_handle (a->handle, &session_index, &thread_index)) + { + clib_warning ("Invalid handle"); + return -1; + } + + session = stream_session_get (session_index, thread_index); + stream_session_disconnect (session); + + return 0; +} + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/application_interface.h b/src/vnet/session/application_interface.h new file mode 100644 index 00000000000..8d87c067841 --- /dev/null +++ b/src/vnet/session/application_interface.h @@ -0,0 +1,136 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#ifndef __included_uri_h__ +#define __included_uri_h__ + +#include <vlib/vlib.h> +#include <vnet/vnet.h> +#include <svm/svm_fifo_segment.h> +#include <vnet/session/session.h> +#include <vnet/session/application.h> +#include <vnet/session/transport.h> + +typedef enum _session_api_proto +{ + SESSION_PROTO_TCP, + SESSION_PROTO_UDP +} session_api_proto_t; + +typedef struct _vnet_bind_args_t +{ + union + { + char *uri; + struct + { + transport_endpoint_t tep; + session_api_proto_t proto; + }; + }; + + u32 api_client_index; + u64 *options; + session_cb_vft_t *session_cb_vft; + + /* + * Results + */ + char *segment_name; + u32 segment_name_length; + u64 server_event_queue_address; + u64 handle; +} vnet_bind_args_t; + +typedef struct _vnet_unbind_args_t +{ + union + { + char *uri; + u64 handle; + }; + u32 api_client_index; +} vnet_unbind_args_t; + +typedef struct _vnet_connect_args +{ + union + { + char *uri; + struct + { + transport_endpoint_t tep; + session_api_proto_t proto; + }; + }; + u32 api_client_index; + u32 api_context; + u64 *options; + session_cb_vft_t *session_cb_vft; + + /* Used for redirects */ + void *mp; +} vnet_connect_args_t; + +typedef struct _vnet_disconnect_args_t +{ + u64 handle; + u32 api_client_index; +} vnet_disconnect_args_t; + +/* Bind / connect options */ +typedef enum +{ + SESSION_OPTIONS_FLAGS, + SESSION_OPTIONS_SEGMENT_SIZE, + SESSION_OPTIONS_ADD_SEGMENT_SIZE, + SESSION_OPTIONS_RX_FIFO_SIZE, + SESSION_OPTIONS_TX_FIFO_SIZE, + SESSION_OPTIONS_ACCEPT_COOKIE, + SESSION_OPTIONS_N_OPTIONS +} session_options_index_t; + +/** Server can handle delegated connect requests from local clients */ +#define SESSION_OPTIONS_FLAGS_USE_FIFO (1<<0) + +/** Server wants vpp to add segments when out of memory for fifos */ +#define SESSION_OPTIONS_FLAGS_ADD_SEGMENT (1<<1) + +#define VNET_CONNECT_REDIRECTED 123 + +int vnet_bind_uri (vnet_bind_args_t *); +int vnet_unbind_uri (char *uri, u32 api_client_index); +int vnet_connect_uri (vnet_connect_args_t * a); +int +vnet_disconnect_session (u32 client_index, u32 session_index, + u32 thread_index); + +int vnet_bind (vnet_bind_args_t * a); +int vnet_connect (vnet_connect_args_t * a); +int vnet_unbind (vnet_unbind_args_t * a); +int vnet_disconnect (vnet_disconnect_args_t * a); + +int +api_parse_session_handle (u64 handle, u32 * session_index, + u32 * thread_index); + +#endif /* __included_uri_h__ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/hashes.c b/src/vnet/session/hashes.c new file mode 100644 index 00000000000..1808dd73f90 --- /dev/null +++ b/src/vnet/session/hashes.c @@ -0,0 +1,28 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +/** Generate typed init functions for multiple hash table styles... */ + +#include <vppinfra/bihash_16_8.h> +#include <vppinfra/bihash_template.h> + +#include <vppinfra/bihash_template.c> + +#undef __included_bihash_template_h__ + +#include <vppinfra/bihash_48_8.h> +#include <vppinfra/bihash_template.h> + +#include <vppinfra/bihash_template.c> diff --git a/src/vnet/session/node.c b/src/vnet/session/node.c new file mode 100644 index 00000000000..e467f4e99e2 --- /dev/null +++ b/src/vnet/session/node.c @@ -0,0 +1,435 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vlib/vlib.h> +#include <vnet/vnet.h> +#include <vnet/pg/pg.h> +#include <vnet/ip/ip.h> + +#include <vnet/tcp/tcp.h> + +#include <vppinfra/hash.h> +#include <vppinfra/error.h> +#include <vppinfra/elog.h> +#include <vlibmemory/unix_shared_memory_queue.h> + +#include <vnet/udp/udp_packet.h> +#include <vnet/lisp-cp/packets.h> +#include <math.h> + +vlib_node_registration_t session_queue_node; + +typedef struct +{ + u32 session_index; + u32 server_thread_index; +} session_queue_trace_t; + +/* packet trace format function */ +static u8 * +format_session_queue_trace (u8 * s, va_list * args) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); + session_queue_trace_t *t = va_arg (*args, session_queue_trace_t *); + + s = format (s, "SESSION_QUEUE: session index %d, server thread index %d", + t->session_index, t->server_thread_index); + return s; +} + +vlib_node_registration_t session_queue_node; + +#define foreach_session_queue_error \ +_(TX, "Packets transmitted") \ +_(TIMER, "Timer events") + +typedef enum +{ +#define _(sym,str) SESSION_QUEUE_ERROR_##sym, + foreach_session_queue_error +#undef _ + SESSION_QUEUE_N_ERROR, +} session_queue_error_t; + +static char *session_queue_error_strings[] = { +#define _(sym,string) string, + foreach_session_queue_error +#undef _ +}; + +static u32 session_type_to_next[] = { + SESSION_QUEUE_NEXT_TCP_IP4_OUTPUT, + SESSION_QUEUE_NEXT_IP4_LOOKUP, + SESSION_QUEUE_NEXT_TCP_IP6_OUTPUT, + SESSION_QUEUE_NEXT_IP6_LOOKUP, +}; + +always_inline int +session_fifo_rx_i (vlib_main_t * vm, vlib_node_runtime_t * node, + session_manager_main_t * smm, session_fifo_event_t * e0, + stream_session_t * s0, u32 thread_index, int *n_tx_packets, + u8 peek_data) +{ + u32 n_trace = vlib_get_trace_count (vm, node); + u32 left_to_snd0, max_len_to_snd0, len_to_deq0, n_bufs, snd_space0; + u32 n_frame_bytes, n_frames_per_evt; + transport_connection_t *tc0; + transport_proto_vft_t *transport_vft; + u32 next_index, next0, *to_next, n_left_to_next, bi0; + vlib_buffer_t *b0; + u32 rx_offset; + u16 snd_mss0; + u8 *data0; + int i; + + next_index = next0 = session_type_to_next[s0->session_type]; + + transport_vft = session_get_transport_vft (s0->session_type); + tc0 = transport_vft->get_connection (s0->connection_index, thread_index); + + /* Make sure we have space to send and there's something to dequeue */ + snd_space0 = transport_vft->send_space (tc0); + snd_mss0 = transport_vft->send_mss (tc0); + + if (snd_space0 == 0 || svm_fifo_max_dequeue (s0->server_tx_fifo) == 0 + || snd_mss0 == 0) + return 0; + + ASSERT (e0->enqueue_length > 0); + + /* Ensure we're not writing more than transport window allows */ + max_len_to_snd0 = clib_min (e0->enqueue_length, snd_space0); + + if (peek_data) + { + /* Offset in rx fifo from where to peek data */ + rx_offset = transport_vft->rx_fifo_offset (tc0); + } + + /* TODO check if transport is willing to send len_to_snd0 + * bytes (Nagle) */ + + n_frame_bytes = snd_mss0 * VLIB_FRAME_SIZE; + n_frames_per_evt = ceil ((double) max_len_to_snd0 / n_frame_bytes); + + n_bufs = vec_len (smm->tx_buffers[thread_index]); + left_to_snd0 = max_len_to_snd0; + for (i = 0; i < n_frames_per_evt; i++) + { + /* Make sure we have at least one full frame of buffers ready */ + if (PREDICT_FALSE (n_bufs < VLIB_FRAME_SIZE)) + { + vec_validate (smm->tx_buffers[thread_index], + n_bufs + VLIB_FRAME_SIZE - 1); + n_bufs += + vlib_buffer_alloc (vm, &smm->tx_buffers[thread_index][n_bufs], + VLIB_FRAME_SIZE); + + /* buffer shortage + * XXX 0.9 because when debugging we might not get a full frame */ + if (PREDICT_FALSE (n_bufs < 0.9 * VLIB_FRAME_SIZE)) + { + /* Keep track of how much we've dequeued and exit */ + e0->enqueue_length -= max_len_to_snd0 - left_to_snd0; + return -1; + } + + _vec_len (smm->tx_buffers[thread_index]) = n_bufs; + } + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + while (left_to_snd0 && n_left_to_next) + { + /* Get free buffer */ + n_bufs--; + bi0 = smm->tx_buffers[thread_index][n_bufs]; + _vec_len (smm->tx_buffers[thread_index]) = n_bufs; + + b0 = vlib_get_buffer (vm, bi0); + b0->error = 0; + b0->flags = VLIB_BUFFER_TOTAL_LENGTH_VALID + | VNET_BUFFER_LOCALLY_ORIGINATED; + b0->current_data = 0; + + /* RX on the local interface. tx in default fib */ + vnet_buffer (b0)->sw_if_index[VLIB_RX] = 0; + vnet_buffer (b0)->sw_if_index[VLIB_TX] = (u32) ~ 0; + + /* usual speculation, or the enqueue_x1 macro will barf */ + to_next[0] = bi0; + to_next += 1; + n_left_to_next -= 1; + + VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b0); + if (PREDICT_FALSE (n_trace > 0)) + { + session_queue_trace_t *t0; + vlib_trace_buffer (vm, node, next_index, b0, + 1 /* follow_chain */ ); + vlib_set_trace_count (vm, node, --n_trace); + t0 = vlib_add_trace (vm, node, b0, sizeof (*t0)); + t0->session_index = s0->session_index; + t0->server_thread_index = s0->thread_index; + } + + if (1) + { + ELOG_TYPE_DECLARE (e) = + { + .format = "evt-dequeue: id %d length %d",.format_args = + "i4i4",}; + struct + { + u32 data[2]; + } *ed; + ed = ELOG_DATA (&vm->elog_main, e); + ed->data[0] = e0->event_id; + ed->data[1] = e0->enqueue_length; + } + + len_to_deq0 = (left_to_snd0 < snd_mss0) ? left_to_snd0 : snd_mss0; + + /* Make room for headers */ + data0 = vlib_buffer_make_headroom (b0, MAX_HDRS_LEN); + + /* Dequeue the data + * TODO 1) peek instead of dequeue + * 2) buffer chains */ + if (peek_data) + { + int n_bytes_read; + n_bytes_read = svm_fifo_peek (s0->server_tx_fifo, s0->pid, + rx_offset, len_to_deq0, data0); + if (n_bytes_read < 0) + goto dequeue_fail; + + /* Keep track of progress locally, transport is also supposed to + * increment it independently when pushing header */ + rx_offset += n_bytes_read; + } + else + { + if (svm_fifo_dequeue_nowait (s0->server_tx_fifo, s0->pid, + len_to_deq0, data0) < 0) + goto dequeue_fail; + } + + b0->current_length = len_to_deq0; + + /* Ask transport to push header */ + transport_vft->push_header (tc0, b0); + + left_to_snd0 -= len_to_deq0; + *n_tx_packets = *n_tx_packets + 1; + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, + to_next, n_left_to_next, + bi0, next0); + } + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + /* If we couldn't dequeue all bytes store progress */ + if (max_len_to_snd0 < e0->enqueue_length) + { + e0->enqueue_length -= max_len_to_snd0; + vec_add1 (smm->evts_partially_read[thread_index], *e0); + } + return 0; + +dequeue_fail: + /* Can't read from fifo. Store event rx progress, save as partially read, + * return buff to free list and return */ + e0->enqueue_length -= max_len_to_snd0 - left_to_snd0; + vec_add1 (smm->evts_partially_read[thread_index], *e0); + + to_next -= 1; + n_left_to_next += 1; + _vec_len (smm->tx_buffers[thread_index]) += 1; + + clib_warning ("dequeue fail"); + return 0; +} + +int +session_fifo_rx_peek (vlib_main_t * vm, vlib_node_runtime_t * node, + session_manager_main_t * smm, session_fifo_event_t * e0, + stream_session_t * s0, u32 thread_index, int *n_tx_pkts) +{ + return session_fifo_rx_i (vm, node, smm, e0, s0, thread_index, n_tx_pkts, + 1); +} + +int +session_fifo_rx_dequeue (vlib_main_t * vm, vlib_node_runtime_t * node, + session_manager_main_t * smm, + session_fifo_event_t * e0, stream_session_t * s0, + u32 thread_index, int *n_tx_pkts) +{ + return session_fifo_rx_i (vm, node, smm, e0, s0, thread_index, n_tx_pkts, + 0); +} + +static uword +session_queue_node_fn (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * frame) +{ + session_manager_main_t *smm = vnet_get_session_manager_main (); + session_fifo_event_t *my_fifo_events, *e; + u32 n_to_dequeue; + unix_shared_memory_queue_t *q; + int n_tx_packets = 0; + u32 my_thread_index = vm->cpu_index; + int i, rv; + + /* + * Update TCP time + */ + tcp_update_time (vlib_time_now (vm), my_thread_index); + + /* + * Get vpp queue events + */ + q = smm->vpp_event_queues[my_thread_index]; + if (PREDICT_FALSE (q == 0)) + return 0; + + /* min number of events we can dequeue without blocking */ + n_to_dequeue = q->cursize; + if (n_to_dequeue == 0) + return 0; + + my_fifo_events = smm->fifo_events[my_thread_index]; + + /* If we didn't manage to process previous events try going + * over them again without dequeuing new ones. + * XXX: Block senders to sessions that can't keep up */ + if (vec_len (my_fifo_events) >= 100) + goto skip_dequeue; + + /* See you in the next life, don't be late */ + if (pthread_mutex_trylock (&q->mutex)) + return 0; + + for (i = 0; i < n_to_dequeue; i++) + { + vec_add2 (my_fifo_events, e, 1); + unix_shared_memory_queue_sub_raw (q, (u8 *) e); + } + + /* The other side of the connection is not polling */ + if (q->cursize < (q->maxsize / 8)) + (void) pthread_cond_broadcast (&q->condvar); + pthread_mutex_unlock (&q->mutex); + + smm->fifo_events[my_thread_index] = my_fifo_events; + +skip_dequeue: + + for (i = 0; i < n_to_dequeue; i++) + { + svm_fifo_t *f0; /* $$$ prefetch 1 ahead maybe */ + stream_session_t *s0; + u32 server_session_index0, server_thread_index0; + session_fifo_event_t *e0; + + e0 = &my_fifo_events[i]; + f0 = e0->fifo; + server_session_index0 = f0->server_session_index; + server_thread_index0 = f0->server_thread_index; + + /* $$$ add multiple event queues, per vpp worker thread */ + ASSERT (server_thread_index0 == my_thread_index); + + s0 = pool_elt_at_index (smm->sessions[my_thread_index], + server_session_index0); + + ASSERT (s0->thread_index == my_thread_index); + + switch (e0->event_type) + { + case FIFO_EVENT_SERVER_TX: + /* Spray packets in per session type frames, since they go to + * different nodes */ + rv = (smm->session_rx_fns[s0->session_type]) (vm, node, smm, e0, s0, + my_thread_index, + &n_tx_packets); + if (rv < 0) + goto done; + + break; + + default: + clib_warning ("unhandled event type %d", e0->event_type); + } + } + +done: + + /* Couldn't process all events. Probably out of buffers */ + if (PREDICT_FALSE (i < n_to_dequeue)) + { + session_fifo_event_t *partially_read = + smm->evts_partially_read[my_thread_index]; + vec_add (partially_read, &my_fifo_events[i], n_to_dequeue - i); + vec_free (my_fifo_events); + smm->fifo_events[my_thread_index] = partially_read; + smm->evts_partially_read[my_thread_index] = 0; + } + else + { + vec_free (smm->fifo_events[my_thread_index]); + smm->fifo_events[my_thread_index] = + smm->evts_partially_read[my_thread_index]; + smm->evts_partially_read[my_thread_index] = 0; + } + + vlib_node_increment_counter (vm, session_queue_node.index, + SESSION_QUEUE_ERROR_TX, n_tx_packets); + + return n_tx_packets; +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (session_queue_node) = +{ + .function = session_queue_node_fn, + .name = "session-queue", + .format_trace = format_session_queue_trace, + .type = VLIB_NODE_TYPE_INPUT, + .n_errors = ARRAY_LEN (session_queue_error_strings), + .error_strings = session_queue_error_strings, + .n_next_nodes = SESSION_QUEUE_N_NEXT, + /* .state = VLIB_NODE_STATE_DISABLED, enable on-demand? */ + /* edit / add dispositions here */ + .next_nodes = + { + [SESSION_QUEUE_NEXT_DROP] = "error-drop", + [SESSION_QUEUE_NEXT_IP4_LOOKUP] = "ip4-lookup", + [SESSION_QUEUE_NEXT_IP6_LOOKUP] = "ip6-lookup", + [SESSION_QUEUE_NEXT_TCP_IP4_OUTPUT] = "tcp4-output", + [SESSION_QUEUE_NEXT_TCP_IP6_OUTPUT] = "tcp6-output", + }, +}; +/* *INDENT-ON* */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/session.api b/src/vnet/session/session.api new file mode 100644 index 00000000000..a7b28c1dba0 --- /dev/null +++ b/src/vnet/session/session.api @@ -0,0 +1,429 @@ +/* + * Copyright (c) 2015-2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + + /** \brief Bind to a given URI + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param accept_cookie - sender accept cookie, to identify this bind flavor + @param uri - a URI, e.g. "tcp://0.0.0.0/0/80" [ipv4] + "tcp://::/0/80" [ipv6] etc. + @param options - socket options, fifo sizes, etc. +*/ +define bind_uri { + u32 client_index; + u32 context; + u32 accept_cookie; + u32 initial_segment_size; + u8 uri[128]; + u64 options[16]; +}; + +/** \brief Unbind a given URI + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param uri - a URI, e.g. "tcp://0.0.0.0/0/80" [ipv4] + "tcp://::/0/80" [ipv6], etc. + @param options - socket options, fifo sizes, etc. +*/ +define unbind_uri { + u32 client_index; + u32 context; + u8 uri[128]; +}; + +/** \brief Connect to a given URI + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param accept_cookie - sender accept cookie, to identify this bind flavor + @param uri - a URI, e.g. "tcp4://0.0.0.0/0/80" + "tcp6://::/0/80" [ipv6], etc. + @param options - socket options, fifo sizes, etc. +*/ +define connect_uri { + u32 client_index; + u32 context; + u8 uri[128]; + u64 client_queue_address; + u64 options[16]; +}; + +/** \brief Bind reply + @param context - sender context, to match reply w/ request + @param retval - return code for the request + @param event_queue_address - vpp event queue address or 0 if this + connection shouldn't send events + @param segment_name_length - length of segment name + @param segment_name - name of segment client needs to attach to +*/ +define bind_uri_reply { + u32 context; + i32 retval; + u64 server_event_queue_address; + u8 segment_name_length; + u32 segment_size; + u8 segment_name[128]; +}; + +/** \brief unbind reply + @param context - sender context, to match reply w/ request + @param retval - return code for the request +*/ +define unbind_uri_reply { + u32 context; + i32 retval; +}; + +/** \brief vpp->client, connect reply + @param context - sender context, to match reply w/ request + @param retval - return code for the request + @param server_rx_fifo - rx (vpp -> vpp-client) fifo address + @param server_tx_fifo - tx (vpp-client -> vpp) fifo address + @param session_index - session index; + @param session_thread_index - session thread index + @param session_type - session thread type + @param vpp_event_queue_address - vpp's event queue address + @param client_event_queue_address - client's event queue address + @param segment_name_length - non-zero if the client needs to attach to + the fifo segment + @param segment_name - set if the client needs to attach to the segment +*/ +define connect_uri_reply { + u32 context; + i32 retval; + u64 server_rx_fifo; + u64 server_tx_fifo; + u32 session_index; + u32 session_thread_index; + u8 session_type; + u64 client_event_queue_address; + u64 vpp_event_queue_address; + u32 segment_size; + u8 segment_name_length; + u8 segment_name[128]; +}; + +/** \brief vpp->client, please map an additional shared memory segment + @param context - sender context, to match reply w/ request + @param segment_name - +*/ +define map_another_segment { + u32 client_index; + u32 context; + u32 segment_size; + u8 segment_name[128]; +}; + +/** \brief client->vpp + @param context - sender context, to match reply w/ request + @param retval - return code for the request +*/ +define map_another_segment_reply { + u32 context; + i32 retval; +}; + +/** \brief vpp->client, accept this session + @param context - sender context, to match reply w/ request + @param accept_cookie - tells client which bind flavor just occurred + @param rx_fifo_address - rx (vpp -> vpp-client) fifo address + @param tx_fifo_address - tx (vpp-client -> vpp) fifo address + @param session_index - index of new session + @param session_thread_index - thread index of new session + @param vpp_event_queue_address - vpp's event queue address + @param session_type - type of session + +*/ +define accept_session { + u32 client_index; + u32 context; + u32 accept_cookie; + u64 server_rx_fifo; + u64 server_tx_fifo; + u32 session_index; + u32 session_thread_index; + u64 vpp_event_queue_address; + u8 session_type; +}; + +/** \brief client->vpp, reply to an accept message + @param context - sender context, to match reply w/ request + @param retval - return code for the request + @param session_index - session index from accept_session / connect_reply + @param session_thread_index - thread index from accept_session / + connect_reply +*/ +define accept_session_reply { + u32 context; + i32 retval; + u8 session_type; + u8 session_thread_index; + u32 session_index; +}; + +/** \brief bidirectional disconnect API + @param client_index - opaque cookie to identify the sender + client to vpp direction only + @param context - sender context, to match reply w/ request + @param session_index - cookie #1 from accept_session / connect_reply + @param session_thread_index - cookie #2 +*/ +define disconnect_session { + u32 client_index; + u32 context; + u32 session_index; + u32 session_thread_index; +}; + +/** \brief bidirectional disconnect reply API + @param client_index - opaque cookie to identify the sender + client to vpp direction only + @param context - sender context, to match reply w/ request + @param retval - return code for the request + @param session_index - session index from accept_session / connect_reply + @param session_thread_index - thread index from accept_session / + connect_reply +*/ +define disconnect_session_reply { + u32 client_index; + u32 context; + i32 retval; + u32 session_index; + u32 session_thread_index; +}; + +/** \brief vpp->client reset session API + @param client_index - opaque cookie to identify the sender + client to vpp direction only + @param context - sender context, to match reply w/ request + @param session_index - session index from accept_session / connect_reply + @param session_thread_index - thread index from accept_session / + connect_reply +*/ +define reset_session { + u32 client_index; + u32 context; + u32 session_index; + u32 session_thread_index; +}; + +/** \brief client->vpp reset session reply + @param client_index - opaque cookie to identify the sender + client to vpp direction only + @param context - sender context, to match reply w/ request + @param retval - return code for the request + @param session_index - session index from accept_session / connect_reply + @param session_thread_index - thread index from accept_session / + connect_reply +*/ +define reset_session_reply { + u32 client_index; + u32 context; + i32 retval; + u32 session_index; + u32 session_thread_index; +}; + +/** \brief Bind to an ip:port pair for a given transport protocol + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param vrf - bind namespace + @param is_ip4 - flag that is 1 if ip address family is IPv4 + @param ip - ip address + @param port - port + @param proto - protocol 0 - TCP 1 - UDP + @param options - socket options, fifo sizes, etc. +*/ +define bind_sock { + u32 client_index; + u32 context; + u32 vrf; + u8 is_ip4; + u8 ip[16]; + u16 port; + u8 proto; + u64 options[16]; +}; + +/** \brief Unbind + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param handle - bind handle obtained from bind reply +*/ +define unbind_sock { + u32 client_index; + u32 context; + u64 handle; +}; + +/** \brief Connect to a remote peer + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param vrf - connection namespace + @param is_ip4 - flag that is 1 if ip address family is IPv4 + @param ip - ip address + @param port - port + @param proto - protocol 0 - TCP 1 - UDP + @param client_queue_address - client's API queue address. Non-zero when + used to perform redirects + @param options - socket options, fifo sizes, etc. +*/ +define connect_sock { + u32 client_index; + u32 context; + u32 vrf; + u8 is_ip4; + u8 ip[16]; + u16 port; + u8 proto; + u64 client_queue_address; + u64 options[16]; +}; + +/** \brief Bind reply + @param context - sender context, to match reply w/ request + @param handle - bind handle + @param retval - return code for the request + @param event_queue_address - vpp event queue address or 0 if this + connection shouldn't send events + @param segment_name_length - length of segment name + @param segment_name - name of segment client needs to attach to +*/ +define bind_sock_reply { + u32 context; + u64 handle; + i32 retval; + u64 server_event_queue_address; + u32 segment_size; + u8 segment_name_length; + u8 segment_name[128]; +}; + +/** \brief unbind reply + @param context - sender context, to match reply w/ request + @param retval - return code for the request +*/ +define unbind_sock_reply { + u32 context; + i32 retval; +}; + +/** \brief vpp/server->client, connect reply + @param context - sender context, to match reply w/ request + @param retval - return code for the request + @param handle - connection handle + @param server_rx_fifo - rx (vpp -> vpp-client) fifo address + @param server_tx_fifo - tx (vpp-client -> vpp) fifo address + @param vpp_event_queue_address - vpp's event queue address + @param client_event_queue_address - client's event queue address + @param segment_name_length - non-zero if the client needs to attach to + the fifo segment + @param segment_name - set if the client needs to attach to the segment +*/ +define connect_sock_reply { + u32 context; + i32 retval; + u64 handle; + u64 server_rx_fifo; + u64 server_tx_fifo; + u64 client_event_queue_address; + u64 vpp_event_queue_address; + u32 segment_size; + u8 segment_name_length; + u8 segment_name[128]; +}; + +/** \brief bidirectional disconnect API + @param client_index - opaque cookie to identify the sender + client to vpp direction only + @param context - sender context, to match reply w/ request + @param handle - session handle obtained through accept/connect +*/ +define disconnect_sock { + u32 client_index; + u32 context; + u64 handle; +}; + +/** \brief bidirectional disconnect reply API + @param client_index - opaque cookie to identify the sender + client to vpp direction only + @param client_context - sender context, to match reply w/ request + @param handle - session handle obtained through accept/connect +*/ +define disconnect_sock_reply { + u32 client_index; + u32 context; + i32 retval; + u64 handle; +}; + +/** \brief vpp->client, accept this session + @param context - sender context, to match reply w/ request + @param accept_cookie - tells client which bind flavor just occurred + @param handle - session handle obtained through accept/connect + @param rx_fifo_address - rx (vpp -> vpp-client) fifo address + @param tx_fifo_address - tx (vpp-client -> vpp) fifo address + @param vpp_event_queue_address - vpp's event queue address +*/ +define accept_sock { + u32 client_index; + u32 context; + u32 accept_cookie; + u64 handle; + u64 server_rx_fifo; + u64 server_tx_fifo; + u64 vpp_event_queue_address; +}; + +/** \brief client->vpp, reply to an accept message + @param context - sender context, to match reply w/ request + @param retval - return code for the request + @param handle - session handle obtained through accept/connect +*/ +define accept_sock_reply { + u32 context; + i32 retval; + u64 handle; +}; + +/** \brief vpp->client reset session API + @param client_index - opaque cookie to identify the sender + client to vpp direction only + @param context - sender context, to match reply w/ request + @param handle - session handle obtained through accept/connect +*/ +define reset_sock { + u32 client_index; + u32 context; + u64 handle; +}; + +/** \brief client->vpp reset session reply + @param client_index - opaque cookie to identify the sender + client to vpp direction only + @param context - sender context, to match reply w/ request + @param handle - session handle obtained through accept/connect +*/ +define reset_sock_reply { + u32 client_index; + u32 context; + i32 retval; + u64 handle; +}; +/* + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */
\ No newline at end of file diff --git a/src/vnet/session/session.c b/src/vnet/session/session.c new file mode 100644 index 00000000000..539da613367 --- /dev/null +++ b/src/vnet/session/session.c @@ -0,0 +1,1286 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +/** + * @file + * @brief Session and session manager + */ + +#include <vnet/session/session.h> +#include <vlibmemory/api.h> +#include <vnet/dpo/load_balance.h> +#include <vnet/fib/ip4_fib.h> +#include <vnet/session/application.h> + +/** + * Per-type vector of transport protocol virtual function tables + */ +static transport_proto_vft_t *tp_vfts; + +session_manager_main_t session_manager_main; + +/* + * Session lookup key; (src-ip, dst-ip, src-port, dst-port, session-type) + * Value: (owner thread index << 32 | session_index); + */ +static void +stream_session_table_add_for_tc (u8 sst, transport_connection_t * tc, + u64 value) +{ + session_manager_main_t *smm = &session_manager_main; + session_kv4_t kv4; + session_kv6_t kv6; + + switch (sst) + { + case SESSION_TYPE_IP4_UDP: + case SESSION_TYPE_IP4_TCP: + make_v4_ss_kv_from_tc (&kv4, tc); + kv4.value = value; + clib_bihash_add_del_16_8 (&smm->v4_session_hash, &kv4, 1 /* is_add */ ); + break; + case SESSION_TYPE_IP6_UDP: + case SESSION_TYPE_IP6_TCP: + make_v6_ss_kv_from_tc (&kv6, tc); + kv6.value = value; + clib_bihash_add_del_48_8 (&smm->v6_session_hash, &kv6, 1 /* is_add */ ); + break; + default: + clib_warning ("Session type not supported"); + ASSERT (0); + } +} + +void +stream_session_table_add (session_manager_main_t * smm, stream_session_t * s, + u64 value) +{ + transport_connection_t *tc; + + tc = tp_vfts[s->session_type].get_connection (s->connection_index, + s->thread_index); + stream_session_table_add_for_tc (s->session_type, tc, value); +} + +static void +stream_session_half_open_table_add (u8 sst, transport_connection_t * tc, + u64 value) +{ + session_manager_main_t *smm = &session_manager_main; + session_kv4_t kv4; + session_kv6_t kv6; + + switch (sst) + { + case SESSION_TYPE_IP4_UDP: + case SESSION_TYPE_IP4_TCP: + make_v4_ss_kv_from_tc (&kv4, tc); + kv4.value = value; + clib_bihash_add_del_16_8 (&smm->v4_half_open_hash, &kv4, + 1 /* is_add */ ); + break; + case SESSION_TYPE_IP6_UDP: + case SESSION_TYPE_IP6_TCP: + make_v6_ss_kv_from_tc (&kv6, tc); + kv6.value = value; + clib_bihash_add_del_48_8 (&smm->v6_half_open_hash, &kv6, + 1 /* is_add */ ); + break; + default: + clib_warning ("Session type not supported"); + ASSERT (0); + } +} + +static int +stream_session_table_del_for_tc (session_manager_main_t * smm, u8 sst, + transport_connection_t * tc) +{ + session_kv4_t kv4; + session_kv6_t kv6; + + switch (sst) + { + case SESSION_TYPE_IP4_UDP: + case SESSION_TYPE_IP4_TCP: + make_v4_ss_kv_from_tc (&kv4, tc); + return clib_bihash_add_del_16_8 (&smm->v4_session_hash, &kv4, + 0 /* is_add */ ); + break; + case SESSION_TYPE_IP6_UDP: + case SESSION_TYPE_IP6_TCP: + make_v6_ss_kv_from_tc (&kv6, tc); + return clib_bihash_add_del_48_8 (&smm->v6_session_hash, &kv6, + 0 /* is_add */ ); + break; + default: + clib_warning ("Session type not supported"); + ASSERT (0); + } + + return 0; +} + +static int +stream_session_table_del (session_manager_main_t * smm, stream_session_t * s) +{ + transport_connection_t *ts; + + ts = tp_vfts[s->session_type].get_connection (s->connection_index, + s->thread_index); + return stream_session_table_del_for_tc (smm, s->session_type, ts); +} + +static void +stream_session_half_open_table_del (session_manager_main_t * smm, u8 sst, + transport_connection_t * tc) +{ + session_kv4_t kv4; + session_kv6_t kv6; + + switch (sst) + { + case SESSION_TYPE_IP4_UDP: + case SESSION_TYPE_IP4_TCP: + make_v4_ss_kv_from_tc (&kv4, tc); + clib_bihash_add_del_16_8 (&smm->v4_half_open_hash, &kv4, + 0 /* is_add */ ); + break; + case SESSION_TYPE_IP6_UDP: + case SESSION_TYPE_IP6_TCP: + make_v6_ss_kv_from_tc (&kv6, tc); + clib_bihash_add_del_48_8 (&smm->v6_half_open_hash, &kv6, + 0 /* is_add */ ); + break; + default: + clib_warning ("Session type not supported"); + ASSERT (0); + } +} + +stream_session_t * +stream_session_lookup_listener4 (ip4_address_t * lcl, u16 lcl_port, u8 proto) +{ + session_manager_main_t *smm = &session_manager_main; + session_kv4_t kv4; + int rv; + + make_v4_listener_kv (&kv4, lcl, lcl_port, proto); + rv = clib_bihash_search_inline_16_8 (&smm->v4_session_hash, &kv4); + if (rv == 0) + return pool_elt_at_index (smm->listen_sessions[proto], (u32) kv4.value); + + /* Zero out the lcl ip */ + kv4.key[0] = 0; + rv = clib_bihash_search_inline_16_8 (&smm->v4_session_hash, &kv4); + if (rv == 0) + return pool_elt_at_index (smm->listen_sessions[proto], kv4.value); + + return 0; +} + +/** Looks up a session based on the 5-tuple passed as argument. + * + * First it tries to find an established session, if this fails, it tries + * finding a listener session if this fails, it tries a lookup with a + * wildcarded local source (listener bound to all interfaces) + */ +stream_session_t * +stream_session_lookup4 (ip4_address_t * lcl, ip4_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto, + u32 my_thread_index) +{ + session_manager_main_t *smm = &session_manager_main; + session_kv4_t kv4; + int rv; + + /* Lookup session amongst established ones */ + make_v4_ss_kv (&kv4, lcl, rmt, lcl_port, rmt_port, proto); + rv = clib_bihash_search_inline_16_8 (&smm->v4_session_hash, &kv4); + if (rv == 0) + return stream_session_get_tsi (kv4.value, my_thread_index); + + /* If nothing is found, check if any listener is available */ + return stream_session_lookup_listener4 (lcl, lcl_port, proto); +} + +stream_session_t * +stream_session_lookup_listener6 (ip6_address_t * lcl, u16 lcl_port, u8 proto) +{ + session_manager_main_t *smm = &session_manager_main; + session_kv6_t kv6; + int rv; + + make_v6_listener_kv (&kv6, lcl, lcl_port, proto); + rv = clib_bihash_search_inline_48_8 (&smm->v6_session_hash, &kv6); + if (rv == 0) + return pool_elt_at_index (smm->listen_sessions[proto], kv6.value); + + /* Zero out the lcl ip */ + kv6.key[0] = kv6.key[1] = 0; + rv = clib_bihash_search_inline_48_8 (&smm->v6_session_hash, &kv6); + if (rv == 0) + return pool_elt_at_index (smm->listen_sessions[proto], kv6.value); + + return 0; +} + +/* Looks up a session based on the 5-tuple passed as argument. + * First it tries to find an established session, if this fails, it tries + * finding a listener session if this fails, it tries a lookup with a + * wildcarded local source (listener bound to all interfaces) */ +stream_session_t * +stream_session_lookup6 (ip6_address_t * lcl, ip6_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto, + u32 my_thread_index) +{ + session_manager_main_t *smm = vnet_get_session_manager_main (); + session_kv6_t kv6; + int rv; + + make_v6_ss_kv (&kv6, lcl, rmt, lcl_port, rmt_port, proto); + rv = clib_bihash_search_inline_48_8 (&smm->v6_session_hash, &kv6); + if (rv == 0) + return stream_session_get_tsi (kv6.value, my_thread_index); + + /* If nothing is found, check if any listener is available */ + return stream_session_lookup_listener6 (lcl, lcl_port, proto); +} + +stream_session_t * +stream_session_lookup_listener (ip46_address_t * lcl, u16 lcl_port, u8 proto) +{ + switch (proto) + { + case SESSION_TYPE_IP4_UDP: + case SESSION_TYPE_IP4_TCP: + return stream_session_lookup_listener4 (&lcl->ip4, lcl_port, proto); + break; + case SESSION_TYPE_IP6_UDP: + case SESSION_TYPE_IP6_TCP: + return stream_session_lookup_listener6 (&lcl->ip6, lcl_port, proto); + break; + } + return 0; +} + +static u64 +stream_session_half_open_lookup (session_manager_main_t * smm, + ip46_address_t * lcl, ip46_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto) +{ + session_kv4_t kv4; + session_kv6_t kv6; + int rv; + + switch (proto) + { + case SESSION_TYPE_IP4_UDP: + case SESSION_TYPE_IP4_TCP: + make_v4_ss_kv (&kv4, &lcl->ip4, &rmt->ip4, lcl_port, rmt_port, proto); + rv = clib_bihash_search_inline_16_8 (&smm->v4_half_open_hash, &kv4); + + if (rv == 0) + return kv4.value; + + return (u64) ~ 0; + break; + case SESSION_TYPE_IP6_UDP: + case SESSION_TYPE_IP6_TCP: + make_v6_ss_kv (&kv6, &lcl->ip6, &rmt->ip6, lcl_port, rmt_port, proto); + rv = clib_bihash_search_inline_48_8 (&smm->v6_half_open_hash, &kv6); + + if (rv == 0) + return kv6.value; + + return (u64) ~ 0; + break; + } + return 0; +} + +transport_connection_t * +stream_session_lookup_transport4 (session_manager_main_t * smm, + ip4_address_t * lcl, ip4_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto, + u32 my_thread_index) +{ + session_kv4_t kv4; + stream_session_t *s; + int rv; + + /* Lookup session amongst established ones */ + make_v4_ss_kv (&kv4, lcl, rmt, lcl_port, rmt_port, proto); + rv = clib_bihash_search_inline_16_8 (&smm->v4_session_hash, &kv4); + if (rv == 0) + { + s = stream_session_get_tsi (kv4.value, my_thread_index); + + return tp_vfts[s->session_type].get_connection (s->connection_index, + my_thread_index); + } + + /* If nothing is found, check if any listener is available */ + s = stream_session_lookup_listener4 (lcl, lcl_port, proto); + if (s) + return tp_vfts[s->session_type].get_listener (s->connection_index); + + /* Finally, try half-open connections */ + rv = clib_bihash_search_inline_16_8 (&smm->v4_half_open_hash, &kv4); + if (rv == 0) + return tp_vfts[proto].get_half_open (kv4.value & 0xFFFFFFFF); + + return 0; +} + +transport_connection_t * +stream_session_lookup_transport6 (session_manager_main_t * smm, + ip6_address_t * lcl, ip6_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto, + u32 my_thread_index) +{ + stream_session_t *s; + session_kv6_t kv6; + int rv; + + make_v6_ss_kv (&kv6, lcl, rmt, lcl_port, rmt_port, proto); + rv = clib_bihash_search_inline_48_8 (&smm->v6_session_hash, &kv6); + if (rv == 0) + { + s = stream_session_get_tsi (kv6.value, my_thread_index); + + return tp_vfts[s->session_type].get_connection (s->connection_index, + my_thread_index); + } + + /* If nothing is found, check if any listener is available */ + s = stream_session_lookup_listener6 (lcl, lcl_port, proto); + if (s) + return tp_vfts[s->session_type].get_listener (s->connection_index); + + /* Finally, try half-open connections */ + rv = clib_bihash_search_inline_48_8 (&smm->v6_half_open_hash, &kv6); + if (rv == 0) + return tp_vfts[s->session_type].get_half_open (kv6.value & 0xFFFFFFFF); + + return 0; +} + +/** + * Allocate vpp event queue (once) per worker thread + */ +void +vpp_session_event_queue_allocate (session_manager_main_t * smm, + u32 thread_index) +{ + api_main_t *am = &api_main; + void *oldheap; + + if (smm->vpp_event_queues[thread_index] == 0) + { + /* Allocate event fifo in the /vpe-api shared-memory segment */ + oldheap = svm_push_data_heap (am->vlib_rp); + + smm->vpp_event_queues[thread_index] = + unix_shared_memory_queue_init (2048 /* nels $$$$ config */ , + sizeof (session_fifo_event_t), + 0 /* consumer pid */ , + 0 + /* (do not) send signal when queue non-empty */ + ); + + svm_pop_heap (oldheap); + } +} + +void +session_manager_get_segment_info (u32 index, u8 ** name, u32 * size) +{ + svm_fifo_segment_private_t *s; + s = svm_fifo_get_segment (index); + *name = s->h->segment_name; + *size = s->ssvm.ssvm_size; +} + +always_inline int +session_manager_add_segment_i (session_manager_main_t * smm, + session_manager_t * sm, + u32 segment_size, u8 * segment_name) +{ + svm_fifo_segment_create_args_t _ca, *ca = &_ca; + int rv; + + memset (ca, 0, sizeof (*ca)); + + ca->segment_name = (char *) segment_name; + ca->segment_size = segment_size; + + rv = svm_fifo_segment_create (ca); + if (rv) + { + clib_warning ("svm_fifo_segment_create ('%s', %d) failed", + ca->segment_name, ca->segment_size); + vec_free (segment_name); + return -1; + } + + vec_add1 (sm->segment_indices, ca->new_segment_index); + + return 0; +} + +static int +session_manager_add_segment (session_manager_main_t * smm, + session_manager_t * sm) +{ + u8 *segment_name; + svm_fifo_segment_create_args_t _ca, *ca = &_ca; + u32 add_segment_size; + u32 default_segment_size = 128 << 10; + + memset (ca, 0, sizeof (*ca)); + segment_name = format (0, "%d-%d%c", getpid (), + smm->unique_segment_name_counter++, 0); + add_segment_size = + sm->add_segment_size ? sm->add_segment_size : default_segment_size; + + return session_manager_add_segment_i (smm, sm, add_segment_size, + segment_name); +} + +int +session_manager_add_first_segment (session_manager_main_t * smm, + session_manager_t * sm, u32 segment_size, + u8 ** segment_name) +{ + svm_fifo_segment_create_args_t _ca, *ca = &_ca; + memset (ca, 0, sizeof (*ca)); + *segment_name = format (0, "%d-%d%c", getpid (), + smm->unique_segment_name_counter++, 0); + return session_manager_add_segment_i (smm, sm, segment_size, *segment_name); +} + +void +session_manager_del (session_manager_main_t * smm, session_manager_t * sm) +{ + u32 *deleted_sessions = 0; + u32 *deleted_thread_indices = 0; + int i, j; + + /* Across all fifo segments used by the server */ + for (j = 0; j < vec_len (sm->segment_indices); j++) + { + svm_fifo_segment_private_t *fifo_segment; + svm_fifo_t **fifos; + /* Vector of fifos allocated in the segment */ + fifo_segment = svm_fifo_get_segment (sm->segment_indices[j]); + fifos = (svm_fifo_t **) fifo_segment->h->fifos; + + /* + * Remove any residual sessions from the session lookup table + * Don't bother deleting the individual fifos, we're going to + * throw away the fifo segment in a minute. + */ + for (i = 0; i < vec_len (fifos); i++) + { + svm_fifo_t *fifo; + u32 session_index, thread_index; + stream_session_t *session; + + fifo = fifos[i]; + session_index = fifo->server_session_index; + thread_index = fifo->server_thread_index; + + session = pool_elt_at_index (smm->sessions[thread_index], + session_index); + + /* Add to the deleted_sessions vector (once!) */ + if (!session->is_deleted) + { + session->is_deleted = 1; + vec_add1 (deleted_sessions, + session - smm->sessions[thread_index]); + vec_add1 (deleted_thread_indices, thread_index); + } + } + + for (i = 0; i < vec_len (deleted_sessions); i++) + { + stream_session_t *session; + + session = + pool_elt_at_index (smm->sessions[deleted_thread_indices[i]], + deleted_sessions[i]); + + /* Instead of directly removing the session call disconnect */ + stream_session_disconnect (session); + + /* + stream_session_table_del (smm, session); + pool_put(smm->sessions[deleted_thread_indices[i]], session); + */ + } + + vec_reset_length (deleted_sessions); + vec_reset_length (deleted_thread_indices); + + /* Instead of removing the segment, test when removing the session if + * the segment can be removed + */ + /* svm_fifo_segment_delete (fifo_segment); */ + } + + vec_free (deleted_sessions); + vec_free (deleted_thread_indices); +} + +int +session_manager_allocate_session_fifos (session_manager_main_t * smm, + session_manager_t * sm, + svm_fifo_t ** server_rx_fifo, + svm_fifo_t ** server_tx_fifo, + u32 * fifo_segment_index, + u8 * added_a_segment) +{ + svm_fifo_segment_private_t *fifo_segment; + u32 fifo_size, default_fifo_size = 8192 /* TODO config */ ; + int i; + + *added_a_segment = 0; + + /* Allocate svm fifos */ + ASSERT (vec_len (sm->segment_indices)); + +again: + for (i = 0; i < vec_len (sm->segment_indices); i++) + { + *fifo_segment_index = sm->segment_indices[i]; + fifo_segment = svm_fifo_get_segment (*fifo_segment_index); + + fifo_size = sm->rx_fifo_size; + fifo_size = (fifo_size == 0) ? default_fifo_size : fifo_size; + *server_rx_fifo = svm_fifo_segment_alloc_fifo (fifo_segment, fifo_size); + + fifo_size = sm->tx_fifo_size; + fifo_size = (fifo_size == 0) ? default_fifo_size : fifo_size; + *server_tx_fifo = svm_fifo_segment_alloc_fifo (fifo_segment, fifo_size); + + if (*server_rx_fifo == 0) + { + /* This would be very odd, but handle it... */ + if (*server_tx_fifo != 0) + { + svm_fifo_segment_free_fifo (fifo_segment, *server_tx_fifo); + *server_tx_fifo = 0; + } + continue; + } + if (*server_tx_fifo == 0) + { + if (*server_rx_fifo != 0) + { + svm_fifo_segment_free_fifo (fifo_segment, *server_rx_fifo); + *server_rx_fifo = 0; + } + continue; + } + break; + } + + /* See if we're supposed to create another segment */ + if (*server_rx_fifo == 0) + { + if (sm->add_segment) + { + if (*added_a_segment) + { + clib_warning ("added a segment, still cant allocate a fifo"); + return SESSION_ERROR_NEW_SEG_NO_SPACE; + } + + if (session_manager_add_segment (smm, sm)) + return VNET_API_ERROR_URI_FIFO_CREATE_FAILED; + + *added_a_segment = 1; + goto again; + } + else + return SESSION_ERROR_NO_SPACE; + } + return 0; +} + +int +stream_session_create_i (session_manager_main_t * smm, application_t * app, + transport_connection_t * tc, + stream_session_t ** ret_s) +{ + int rv; + svm_fifo_t *server_rx_fifo = 0, *server_tx_fifo = 0; + u32 fifo_segment_index; + u32 pool_index, seg_size; + stream_session_t *s; + u64 value; + u32 thread_index = tc->thread_index; + session_manager_t *sm; + u8 segment_added; + u8 *seg_name; + + sm = session_manager_get (app->session_manager_index); + + /* Check the API queue */ + if (app->mode == APP_SERVER && application_api_queue_is_full (app)) + return SESSION_ERROR_API_QUEUE_FULL; + + if ((rv = session_manager_allocate_session_fifos (smm, sm, &server_rx_fifo, + &server_tx_fifo, + &fifo_segment_index, + &segment_added))) + return rv; + + if (segment_added && app->mode == APP_SERVER) + { + /* Send an API message to the external server, to map new segment */ + ASSERT (app->cb_fns.add_segment_callback); + + session_manager_get_segment_info (fifo_segment_index, &seg_name, + &seg_size); + if (app->cb_fns.add_segment_callback (app->api_client_index, seg_name, + seg_size)) + return VNET_API_ERROR_URI_FIFO_CREATE_FAILED; + } + + /* Create the session */ + pool_get (smm->sessions[thread_index], s); + memset (s, 0, sizeof (*s)); + + /* Initialize backpointers */ + pool_index = s - smm->sessions[thread_index]; + server_rx_fifo->server_session_index = pool_index; + server_rx_fifo->server_thread_index = thread_index; + + server_tx_fifo->server_session_index = pool_index; + server_tx_fifo->server_thread_index = thread_index; + + s->server_rx_fifo = server_rx_fifo; + s->server_tx_fifo = server_tx_fifo; + + /* Initialize state machine, such as it is... */ + s->session_type = app->session_type; + s->session_state = SESSION_STATE_CONNECTING; + s->app_index = application_get_index (app); + s->server_segment_index = fifo_segment_index; + s->thread_index = thread_index; + s->session_index = pool_index; + + /* Attach transport to session */ + s->connection_index = tc->c_index; + + /* Attach session to transport */ + tc->s_index = s->session_index; + + /* Add to the main lookup table */ + value = (((u64) thread_index) << 32) | (u64) s->session_index; + stream_session_table_add_for_tc (app->session_type, tc, value); + + *ret_s = s; + + return 0; +} + +/* + * Enqueue data for delivery to session peer. Does not notify peer of enqueue + * event but on request can queue notification events for later delivery by + * calling stream_server_flush_enqueue_events(). + * + * @param tc Transport connection which is to be enqueued data + * @param data Data to be enqueued + * @param len Length of data to be enqueued + * @param queue_event Flag to indicate if peer is to be notified or if event + * is to be queued. The former is useful when more data is + * enqueued and only one event is to be generated. + * @return Number of bytes enqueued or a negative value if enqueueing failed. + */ +int +stream_session_enqueue_data (transport_connection_t * tc, u8 * data, u16 len, + u8 queue_event) +{ + stream_session_t *s; + int enqueued; + + s = stream_session_get (tc->s_index, tc->thread_index); + + /* Make sure there's enough space left. We might've filled the pipes */ + if (PREDICT_FALSE (len > svm_fifo_max_enqueue (s->server_rx_fifo))) + return -1; + + enqueued = svm_fifo_enqueue_nowait (s->server_rx_fifo, s->pid, len, data); + + if (queue_event) + { + /* Queue RX event on this fifo. Eventually these will need to be flushed + * by calling stream_server_flush_enqueue_events () */ + session_manager_main_t *smm = vnet_get_session_manager_main (); + u32 thread_index = s->thread_index; + u32 my_enqueue_epoch = smm->current_enqueue_epoch[thread_index]; + + if (s->enqueue_epoch != my_enqueue_epoch) + { + s->enqueue_epoch = my_enqueue_epoch; + vec_add1 (smm->session_indices_to_enqueue_by_thread[thread_index], + s - smm->sessions[thread_index]); + } + } + + return enqueued; +} + +/** Check if we have space in rx fifo to push more bytes */ +u8 +stream_session_no_space (transport_connection_t * tc, u32 thread_index, + u16 data_len) +{ + stream_session_t *s = stream_session_get (tc->c_index, thread_index); + + if (PREDICT_FALSE (s->session_state != SESSION_STATE_READY)) + return 1; + + if (data_len > svm_fifo_max_enqueue (s->server_rx_fifo)) + return 1; + + return 0; +} + +u32 +stream_session_peek_bytes (transport_connection_t * tc, u8 * buffer, + u32 offset, u32 max_bytes) +{ + stream_session_t *s = stream_session_get (tc->s_index, tc->thread_index); + return svm_fifo_peek (s->server_tx_fifo, s->pid, offset, max_bytes, buffer); +} + +u32 +stream_session_dequeue_drop (transport_connection_t * tc, u32 max_bytes) +{ + stream_session_t *s = stream_session_get (tc->s_index, tc->thread_index); + return svm_fifo_dequeue_drop (s->server_tx_fifo, s->pid, max_bytes); +} + +/** + * Notify session peer that new data has been enqueued. + * + * @param s Stream session for which the event is to be generated. + * @param block Flag to indicate if call should block if event queue is full. + * + * @return 0 on succes or negative number if failed to send notification. + */ +static int +stream_session_enqueue_notify (stream_session_t * s, u8 block) +{ + application_t *app; + session_fifo_event_t evt; + unix_shared_memory_queue_t *q; + static u32 serial_number; + + if (PREDICT_FALSE (s->session_state == SESSION_STATE_CLOSED)) + return 0; + + /* Get session's server */ + app = application_get (s->app_index); + + /* Fabricate event */ + evt.fifo = s->server_rx_fifo; + evt.event_type = FIFO_EVENT_SERVER_RX; + evt.event_id = serial_number++; + evt.enqueue_length = svm_fifo_max_dequeue (s->server_rx_fifo); + + /* Add event to server's event queue */ + q = app->event_queue; + + /* Based on request block (or not) for lack of space */ + if (block || PREDICT_TRUE (q->cursize < q->maxsize)) + unix_shared_memory_queue_add (app->event_queue, (u8 *) & evt, + 0 /* do wait for mutex */ ); + else + return -1; + + if (1) + { + ELOG_TYPE_DECLARE (e) = + { + .format = "evt-enqueue: id %d length %d",.format_args = "i4i4",}; + struct + { + u32 data[2]; + } *ed; + ed = ELOG_DATA (&vlib_global_main.elog_main, e); + ed->data[0] = evt.event_id; + ed->data[1] = evt.enqueue_length; + } + + return 0; +} + +/** + * Flushes queue of sessions that are to be notified of new data + * enqueued events. + * + * @param thread_index Thread index for which the flush is to be performed. + * @return 0 on success or a positive number indicating the number of + * failures due to API queue being full. + */ +int +session_manager_flush_enqueue_events (u32 thread_index) +{ + session_manager_main_t *smm = &session_manager_main; + u32 *session_indices_to_enqueue; + int i, errors = 0; + + session_indices_to_enqueue = + smm->session_indices_to_enqueue_by_thread[thread_index]; + + for (i = 0; i < vec_len (session_indices_to_enqueue); i++) + { + stream_session_t *s0; + + /* Get session */ + s0 = stream_session_get (session_indices_to_enqueue[i], thread_index); + if (stream_session_enqueue_notify (s0, 0 /* don't block */ )) + { + errors++; + } + } + + vec_reset_length (session_indices_to_enqueue); + + smm->session_indices_to_enqueue_by_thread[thread_index] = + session_indices_to_enqueue; + + /* Increment enqueue epoch for next round */ + smm->current_enqueue_epoch[thread_index]++; + + return errors; +} + +/* + * Start listening on server's ip/port pair for requested transport. + * + * Creates a 'dummy' stream session with state LISTENING to be used in session + * lookups, prior to establishing connection. Requests transport to build + * it's own specific listening connection. + */ +int +stream_session_start_listen (u32 server_index, ip46_address_t * ip, u16 port) +{ + session_manager_main_t *smm = &session_manager_main; + stream_session_t *s; + transport_connection_t *tc; + application_t *srv; + u32 tci; + + srv = application_get (server_index); + + pool_get (smm->listen_sessions[srv->session_type], s); + memset (s, 0, sizeof (*s)); + + s->session_type = srv->session_type; + s->session_state = SESSION_STATE_LISTENING; + s->session_index = s - smm->listen_sessions[srv->session_type]; + s->app_index = srv->index; + + /* Transport bind/listen */ + tci = tp_vfts[srv->session_type].bind (smm->vlib_main, s->session_index, ip, + port); + + /* Attach transport to session */ + s->connection_index = tci; + tc = tp_vfts[srv->session_type].get_listener (tci); + + srv->session_index = s->session_index; + + /* Add to the main lookup table */ + stream_session_table_add_for_tc (s->session_type, tc, s->session_index); + + return 0; +} + +void +stream_session_stop_listen (u32 server_index) +{ + session_manager_main_t *smm = &session_manager_main; + stream_session_t *listener; + transport_connection_t *tc; + application_t *srv; + + srv = application_get (server_index); + listener = pool_elt_at_index (smm->listen_sessions[srv->session_type], + srv->session_index); + + tc = tp_vfts[srv->session_type].get_listener (listener->connection_index); + stream_session_table_del_for_tc (smm, listener->session_type, tc); + + tp_vfts[srv->session_type].unbind (smm->vlib_main, + listener->connection_index); + pool_put (smm->listen_sessions[srv->session_type], listener); +} + +int +connect_server_add_segment_cb (application_t * ss, char *segment_name, + u32 segment_size) +{ + /* Does exactly nothing, but die */ + ASSERT (0); + return 0; +} + +void +connects_session_manager_init (session_manager_main_t * smm, u8 session_type) +{ + session_manager_t *sm; + u32 connect_fifo_size = 8 << 10; /* Config? */ + u32 default_segment_size = 1 << 20; + + pool_get (smm->session_managers, sm); + memset (sm, 0, sizeof (*sm)); + + sm->add_segment_size = default_segment_size; + sm->rx_fifo_size = connect_fifo_size; + sm->tx_fifo_size = connect_fifo_size; + sm->add_segment = 1; + + session_manager_add_segment (smm, sm); + smm->connect_manager_index[session_type] = sm - smm->session_managers; +} + +void +stream_session_connect_notify (transport_connection_t * tc, u8 sst, + u8 is_fail) +{ + session_manager_main_t *smm = &session_manager_main; + application_t *app; + stream_session_t *new_s = 0; + u64 value; + + value = stream_session_half_open_lookup (smm, &tc->lcl_ip, &tc->rmt_ip, + tc->lcl_port, tc->rmt_port, + tc->proto); + if (value == HALF_OPEN_LOOKUP_INVALID_VALUE) + { + clib_warning ("This can't be good!"); + return; + } + + app = application_get (value >> 32); + + if (!is_fail) + { + /* Create new session (server segments are allocated if needed) */ + if (stream_session_create_i (smm, app, tc, &new_s)) + return; + + app->session_index = stream_session_get_index (new_s); + app->thread_index = new_s->thread_index; + + /* Allocate vpp event queue for this thread if needed */ + vpp_session_event_queue_allocate (smm, tc->thread_index); + } + + /* Notify client */ + app->cb_fns.session_connected_callback (app->api_client_index, new_s, + is_fail); + + /* Cleanup session lookup */ + stream_session_half_open_table_del (smm, sst, tc); +} + +void +stream_session_accept_notify (transport_connection_t * tc) +{ + application_t *server; + stream_session_t *s; + + s = stream_session_get (tc->s_index, tc->thread_index); + server = application_get (s->app_index); + server->cb_fns.session_accept_callback (s); +} + +/** + * Notification from transport that connection is being closed. + * + * A disconnect is sent to application but state is not removed. Once + * disconnect is acknowledged by application, session disconnect is called. + * Ultimately this leads to close being called on transport (passive close). + */ +void +stream_session_disconnect_notify (transport_connection_t * tc) +{ + application_t *server; + stream_session_t *s; + + s = stream_session_get (tc->s_index, tc->thread_index); + server = application_get (s->app_index); + server->cb_fns.session_disconnect_callback (s); +} + +/** + * Cleans up session and associated app if needed. + */ +void +stream_session_delete (stream_session_t * s) +{ + session_manager_main_t *smm = vnet_get_session_manager_main (); + svm_fifo_segment_private_t *fifo_segment; + application_t *app; + int rv; + + /* delete from the main lookup table */ + rv = stream_session_table_del (smm, s); + + if (rv) + clib_warning ("hash delete error, rv %d", rv); + + /* Cleanup fifo segments */ + fifo_segment = svm_fifo_get_segment (s->server_segment_index); + svm_fifo_segment_free_fifo (fifo_segment, s->server_rx_fifo); + svm_fifo_segment_free_fifo (fifo_segment, s->server_tx_fifo); + + /* Cleanup app if client */ + app = application_get (s->app_index); + if (app->mode == APP_CLIENT) + { + application_del (app); + } + else if (app->mode == APP_SERVER) + { + session_manager_t *sm; + svm_fifo_segment_private_t *fifo_segment; + svm_fifo_t **fifos; + u32 fifo_index; + + sm = session_manager_get (app->session_manager_index); + + /* Delete fifo */ + fifo_segment = svm_fifo_get_segment (s->server_segment_index); + fifos = (svm_fifo_t **) fifo_segment->h->fifos; + + fifo_index = svm_fifo_segment_index (fifo_segment); + + /* Remove segment only if it holds no fifos and not the first */ + if (sm->segment_indices[0] != fifo_index && vec_len (fifos) == 0) + svm_fifo_segment_delete (fifo_segment); + } + + pool_put (smm->sessions[s->thread_index], s); +} + +/** + * Notification from transport that connection is being deleted + * + * This should be called only on previously fully established sessions. For + * instance failed connects should call stream_session_connect_notify and + * indicate that the connect has failed. + */ +void +stream_session_delete_notify (transport_connection_t * tc) +{ + stream_session_t *s; + + s = stream_session_get_if_valid (tc->s_index, tc->thread_index); + if (!s) + { + clib_warning ("Surprised!"); + return; + } + stream_session_delete (s); +} + +/** + * Notify application that connection has been reset. + */ +void +stream_session_reset_notify (transport_connection_t * tc) +{ + stream_session_t *s; + application_t *app; + s = stream_session_get (tc->s_index, tc->thread_index); + + app = application_get (s->app_index); + app->cb_fns.session_reset_callback (s); +} + +/** + * Accept a stream session. Optionally ping the server by callback. + */ +int +stream_session_accept (transport_connection_t * tc, u32 listener_index, + u8 sst, u8 notify) +{ + session_manager_main_t *smm = &session_manager_main; + application_t *server; + stream_session_t *s, *listener; + + int rv; + + /* Find the server */ + listener = pool_elt_at_index (smm->listen_sessions[sst], listener_index); + server = application_get (listener->app_index); + + if ((rv = stream_session_create_i (smm, server, tc, &s))) + return rv; + + /* Allocate vpp event queue for this thread if needed */ + vpp_session_event_queue_allocate (smm, tc->thread_index); + + /* Shoulder-tap the server */ + if (notify) + { + server->cb_fns.session_accept_callback (s); + } + + return 0; +} + +void +stream_session_open (u8 sst, ip46_address_t * addr, u16 port_host_byte_order, + u32 app_index) +{ + transport_connection_t *tc; + u32 tci; + u64 value; + + /* Ask transport to open connection */ + tci = tp_vfts[sst].open (addr, port_host_byte_order); + + /* Get transport connection */ + tc = tp_vfts[sst].get_half_open (tci); + + /* Store api_client_index and transport connection index */ + value = (((u64) app_index) << 32) | (u64) tc->c_index; + + /* Add to the half-open lookup table */ + stream_session_half_open_table_add (sst, tc, value); +} + +/** + * Disconnect session and propagate to transport. This should eventually + * result in a delete notification that allows us to cleanup session state. + * Called for both active/passive disconnects. + */ +void +stream_session_disconnect (stream_session_t * s) +{ + tp_vfts[s->session_type].close (s->connection_index, s->thread_index); + s->session_state = SESSION_STATE_CLOSED; +} + +/** + * Cleanup transport and session state. + */ +void +stream_session_cleanup (stream_session_t * s) +{ + tp_vfts[s->session_type].cleanup (s->connection_index, s->thread_index); + stream_session_delete (s); +} + +void +session_register_transport (u8 type, const transport_proto_vft_t * vft) +{ + session_manager_main_t *smm = vnet_get_session_manager_main (); + + vec_validate (tp_vfts, type); + tp_vfts[type] = *vft; + + /* If an offset function is provided, then peek instead of dequeue */ + smm->session_rx_fns[type] = + (vft->rx_fifo_offset) ? session_fifo_rx_peek : session_fifo_rx_dequeue; +} + +transport_proto_vft_t * +session_get_transport_vft (u8 type) +{ + if (type >= vec_len (tp_vfts)) + return 0; + return &tp_vfts[type]; +} + +static clib_error_t * +session_manager_main_init (vlib_main_t * vm) +{ + u32 num_threads; + vlib_thread_main_t *vtm = vlib_get_thread_main (); + session_manager_main_t *smm = &session_manager_main; + int i; + + smm->vlib_main = vm; + smm->vnet_main = vnet_get_main (); + + num_threads = 1 /* main thread */ + vtm->n_threads; + + if (num_threads < 1) + return clib_error_return (0, "n_thread_stacks not set"); + + /* $$$ config parameters */ + svm_fifo_segment_init (0x200000000ULL /* first segment base VA */ , + 20 /* timeout in seconds */ ); + + /* configure per-thread ** vectors */ + vec_validate (smm->sessions, num_threads - 1); + vec_validate (smm->session_indices_to_enqueue_by_thread, num_threads - 1); + vec_validate (smm->tx_buffers, num_threads - 1); + vec_validate (smm->fifo_events, num_threads - 1); + vec_validate (smm->evts_partially_read, num_threads - 1); + vec_validate (smm->current_enqueue_epoch, num_threads - 1); + vec_validate (smm->vpp_event_queues, num_threads - 1); + + /* $$$$ preallocate hack config parameter */ + for (i = 0; i < 200000; i++) + { + stream_session_t *ss; + pool_get (smm->sessions[0], ss); + memset (ss, 0, sizeof (*ss)); + } + + for (i = 0; i < 200000; i++) + pool_put_index (smm->sessions[0], i); + + clib_bihash_init_16_8 (&smm->v4_session_hash, "v4 session table", + 200000 /* $$$$ config parameter nbuckets */ , + (64 << 20) /*$$$ config parameter table size */ ); + clib_bihash_init_48_8 (&smm->v6_session_hash, "v6 session table", + 200000 /* $$$$ config parameter nbuckets */ , + (64 << 20) /*$$$ config parameter table size */ ); + + clib_bihash_init_16_8 (&smm->v4_half_open_hash, "v4 half-open table", + 200000 /* $$$$ config parameter nbuckets */ , + (64 << 20) /*$$$ config parameter table size */ ); + clib_bihash_init_48_8 (&smm->v6_half_open_hash, "v6 half-open table", + 200000 /* $$$$ config parameter nbuckets */ , + (64 << 20) /*$$$ config parameter table size */ ); + + for (i = 0; i < SESSION_N_TYPES; i++) + smm->connect_manager_index[i] = INVALID_INDEX; + + return 0; +} + +VLIB_INIT_FUNCTION (session_manager_main_init); + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/session.h b/src/vnet/session/session.h new file mode 100644 index 00000000000..cf14cca9f30 --- /dev/null +++ b/src/vnet/session/session.h @@ -0,0 +1,380 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#ifndef __included_session_h__ +#define __included_session_h__ + +#include <vnet/session/transport.h> +#include <vlibmemory/unix_shared_memory_queue.h> +#include <vlibmemory/api.h> +#include <vppinfra/sparse_vec.h> +#include <svm/svm_fifo_segment.h> + +#define HALF_OPEN_LOOKUP_INVALID_VALUE ((u64)~0) +#define INVALID_INDEX ((u32)~0) + +/* TODO decide how much since we have pre-data as well */ +#define MAX_HDRS_LEN 100 /* Max number of bytes for headers */ + +typedef enum +{ + FIFO_EVENT_SERVER_RX, + FIFO_EVENT_SERVER_TX, + FIFO_EVENT_TIMEOUT, + FIFO_EVENT_SERVER_EXIT, +} fifo_event_type_t; + +#define foreach_session_input_error \ +_(NO_SESSION, "No session drops") \ +_(NO_LISTENER, "No listener for dst port drops") \ +_(ENQUEUED, "Packets pushed into rx fifo") \ +_(NOT_READY, "Session not ready packets") \ +_(FIFO_FULL, "Packets dropped for lack of rx fifo space") \ +_(EVENT_FIFO_FULL, "Events not sent for lack of event fifo space") \ +_(API_QUEUE_FULL, "Sessions not created for lack of API queue space") \ +_(NEW_SEG_NO_SPACE, "Created segment, couldn't allocate a fifo pair") \ +_(NO_SPACE, "Couldn't allocate a fifo pair") + +typedef enum +{ +#define _(sym,str) SESSION_ERROR_##sym, + foreach_session_input_error +#undef _ + SESSION_N_ERROR, +} session_error_t; + +/* Event queue input node static next indices */ +typedef enum +{ + SESSION_QUEUE_NEXT_DROP, + SESSION_QUEUE_NEXT_TCP_IP4_OUTPUT, + SESSION_QUEUE_NEXT_IP4_LOOKUP, + SESSION_QUEUE_NEXT_TCP_IP6_OUTPUT, + SESSION_QUEUE_NEXT_IP6_LOOKUP, + SESSION_QUEUE_N_NEXT, +} session_queue_next_t; + +#define foreach_session_type \ + _(IP4_TCP, ip4_tcp) \ + _(IP4_UDP, ip4_udp) \ + _(IP6_TCP, ip6_tcp) \ + _(IP6_UDP, ip6_udp) + +typedef enum +{ +#define _(A, a) SESSION_TYPE_##A, + foreach_session_type +#undef _ + SESSION_N_TYPES, +} session_type_t; + +/* + * Application session state + */ +typedef enum +{ + SESSION_STATE_LISTENING, + SESSION_STATE_CONNECTING, + SESSION_STATE_READY, + SESSION_STATE_CLOSED, + SESSION_STATE_N_STATES, +} stream_session_state_t; + +typedef CLIB_PACKED (struct + { + svm_fifo_t * fifo; + u8 event_type; + /* $$$$ for event logging */ + u16 event_id; + u32 enqueue_length; + }) session_fifo_event_t; + +typedef struct _stream_session_t +{ + /** Type */ + u8 session_type; + + /** State */ + u8 session_state; + + /** Session index in per_thread pool */ + u32 session_index; + + /** Transport specific */ + u32 connection_index; + + u8 thread_index; + + /** Application specific */ + u32 pid; + + /** fifo pointers. Once allocated, these do not move */ + svm_fifo_t *server_rx_fifo; + svm_fifo_t *server_tx_fifo; + + /** To avoid n**2 "one event per frame" check */ + u8 enqueue_epoch; + + /** used during unbind processing */ + u8 is_deleted; + + /** stream server pool index */ + u32 app_index; + + /** svm segment index */ + u32 server_segment_index; +} stream_session_t; + +typedef struct _session_manager +{ + /** segments mapped by this server */ + u32 *segment_indices; + + /** Session fifo sizes. They are provided for binds and take default + * values for connects */ + u32 rx_fifo_size; + u32 tx_fifo_size; + + /** Configured additional segment size */ + u32 add_segment_size; + + /** Flag that indicates if additional segments should be created */ + u8 add_segment; +} session_manager_t; + +/* Forward definition */ +typedef struct _session_manager_main session_manager_main_t; + +typedef int + (session_fifo_rx_fn) (vlib_main_t * vm, vlib_node_runtime_t * node, + session_manager_main_t * smm, + session_fifo_event_t * e0, stream_session_t * s0, + u32 thread_index, int *n_tx_pkts); + +extern session_fifo_rx_fn session_fifo_rx_peek; +extern session_fifo_rx_fn session_fifo_rx_dequeue; + +struct _session_manager_main +{ + /** Lookup tables for established sessions and listeners */ + clib_bihash_16_8_t v4_session_hash; + clib_bihash_48_8_t v6_session_hash; + + /** Lookup tables for half-open sessions */ + clib_bihash_16_8_t v4_half_open_hash; + clib_bihash_48_8_t v6_half_open_hash; + + /** Per worker thread session pools */ + stream_session_t **sessions; + + /** Pool of listen sessions. Same type as stream sessions to ease lookups */ + stream_session_t *listen_sessions[SESSION_N_TYPES]; + + /** Sparse vector to map dst port to stream server */ + u16 *stream_server_by_dst_port[SESSION_N_TYPES]; + + /** per-worker enqueue epoch counters */ + u8 *current_enqueue_epoch; + + /** Per-worker thread vector of sessions to enqueue */ + u32 **session_indices_to_enqueue_by_thread; + + /** per-worker tx buffer free lists */ + u32 **tx_buffers; + + /** Per worker-thread vector of partially read events */ + session_fifo_event_t **evts_partially_read; + + /** per-worker active event vectors */ + session_fifo_event_t **fifo_events; + + /** vpp fifo event queue */ + unix_shared_memory_queue_t **vpp_event_queues; + + /** Unique segment name counter */ + u32 unique_segment_name_counter; + + /* Connection manager used by incoming connects */ + u32 connect_manager_index[SESSION_N_TYPES]; + + session_manager_t *session_managers; + + /** Per transport rx function that can either dequeue or peek */ + session_fifo_rx_fn *session_rx_fns[SESSION_N_TYPES]; + + /* Convenience */ + vlib_main_t *vlib_main; + vnet_main_t *vnet_main; +}; + +extern session_manager_main_t session_manager_main; + +/* + * Session manager function + */ +always_inline session_manager_main_t * +vnet_get_session_manager_main () +{ + return &session_manager_main; +} + +always_inline session_manager_t * +session_manager_get (u32 index) +{ + return pool_elt_at_index (session_manager_main.session_managers, index); +} + +always_inline unix_shared_memory_queue_t * +session_manager_get_vpp_event_queue (u32 thread_index) +{ + return session_manager_main.vpp_event_queues[thread_index]; +} + +always_inline session_manager_t * +connects_session_manager_get (session_manager_main_t * smm, + session_type_t session_type) +{ + return pool_elt_at_index (smm->session_managers, + smm->connect_manager_index[session_type]); +} + +void session_manager_get_segment_info (u32 index, u8 ** name, u32 * size); +int session_manager_flush_enqueue_events (u32 thread_index); +int +session_manager_add_first_segment (session_manager_main_t * smm, + session_manager_t * sm, u32 segment_size, + u8 ** segment_name); +void +session_manager_del (session_manager_main_t * smm, session_manager_t * sm); +void +connects_session_manager_init (session_manager_main_t * smm, u8 session_type); + +/* + * Stream session functions + */ + +stream_session_t *stream_session_lookup_listener4 (ip4_address_t * lcl, + u16 lcl_port, u8 proto); +stream_session_t *stream_session_lookup4 (ip4_address_t * lcl, + ip4_address_t * rmt, u16 lcl_port, + u16 rmt_port, u8 proto, + u32 thread_index); +stream_session_t *stream_session_lookup_listener6 (ip6_address_t * lcl, + u16 lcl_port, u8 proto); +stream_session_t *stream_session_lookup6 (ip6_address_t * lcl, + ip6_address_t * rmt, u16 lcl_port, + u16 rmt_port, u8, u32 thread_index); +transport_connection_t + * stream_session_lookup_transport4 (session_manager_main_t * smm, + ip4_address_t * lcl, + ip4_address_t * rmt, u16 lcl_port, + u16 rmt_port, u8 proto, + u32 thread_index); +transport_connection_t + * stream_session_lookup_transport6 (session_manager_main_t * smm, + ip6_address_t * lcl, + ip6_address_t * rmt, u16 lcl_port, + u16 rmt_port, u8 proto, + u32 thread_index); +stream_session_t *stream_session_lookup_listener (ip46_address_t * lcl, + u16 lcl_port, u8 proto); + +always_inline stream_session_t * +stream_session_get_tsi (u64 ti_and_si, u32 thread_index) +{ + ASSERT ((u32) (ti_and_si >> 32) == thread_index); + return pool_elt_at_index (session_manager_main.sessions[thread_index], + ti_and_si & 0xFFFFFFFFULL); +} + +always_inline stream_session_t * +stream_session_get (u64 si, u32 thread_index) +{ + return pool_elt_at_index (session_manager_main.sessions[thread_index], si); +} + +always_inline stream_session_t * +stream_session_get_if_valid (u64 si, u32 thread_index) +{ + if (thread_index >= vec_len (session_manager_main.sessions)) + return 0; + + if (pool_is_free_index (session_manager_main.sessions[thread_index], si)) + return 0; + + return pool_elt_at_index (session_manager_main.sessions[thread_index], si); +} + +always_inline stream_session_t * +stream_session_listener_get (u8 sst, u64 si) +{ + return pool_elt_at_index (session_manager_main.listen_sessions[sst], si); +} + +always_inline u32 +stream_session_get_index (stream_session_t * s) +{ + if (s->session_state == SESSION_STATE_LISTENING) + return s - session_manager_main.listen_sessions[s->session_type]; + + return s - session_manager_main.sessions[s->thread_index]; +} + +always_inline u32 +stream_session_max_enqueue (transport_connection_t * tc) +{ + stream_session_t *s = stream_session_get (tc->s_index, tc->thread_index); + return svm_fifo_max_enqueue (s->server_rx_fifo); +} + +int +stream_session_enqueue_data (transport_connection_t * tc, u8 * data, u16 len, + u8 queue_event); +u32 +stream_session_peek_bytes (transport_connection_t * tc, u8 * buffer, + u32 offset, u32 max_bytes); +u32 stream_session_dequeue_drop (transport_connection_t * tc, u32 max_bytes); + +void +stream_session_connect_notify (transport_connection_t * tc, u8 sst, + u8 is_fail); +void stream_session_accept_notify (transport_connection_t * tc); +void stream_session_disconnect_notify (transport_connection_t * tc); +void stream_session_delete_notify (transport_connection_t * tc); +void stream_session_reset_notify (transport_connection_t * tc); +int +stream_session_accept (transport_connection_t * tc, u32 listener_index, + u8 sst, u8 notify); +void stream_session_open (u8 sst, ip46_address_t * addr, + u16 port_host_byte_order, u32 api_client_index); +void stream_session_disconnect (stream_session_t * s); +void stream_session_cleanup (stream_session_t * s); +int +stream_session_start_listen (u32 server_index, ip46_address_t * ip, u16 port); +void stream_session_stop_listen (u32 server_index); + +u8 *format_stream_session (u8 * s, va_list * args); + +void session_register_transport (u8 type, const transport_proto_vft_t * vft); +transport_proto_vft_t *session_get_transport_vft (u8 type); + +#endif /* __included_session_h__ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/session_api.c b/src/vnet/session/session_api.c new file mode 100644 index 00000000000..9d068684636 --- /dev/null +++ b/src/vnet/session/session_api.c @@ -0,0 +1,821 @@ +/* + * Copyright (c) 2015-2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/vnet.h> +#include <vlibmemory/api.h> +#include <vnet/session/application.h> + +#include <vnet/vnet_msg_enum.h> +#include "application_interface.h" + +#define vl_typedefs /* define message structures */ +#include <vnet/vnet_all_api_h.h> +#undef vl_typedefs + +#define vl_endianfun /* define message structures */ +#include <vnet/vnet_all_api_h.h> +#undef vl_endianfun + +/* instantiate all the print functions we know about */ +#define vl_print(handle, ...) vlib_cli_output (handle, __VA_ARGS__) +#define vl_printfun +#include <vnet/vnet_all_api_h.h> +#undef vl_printfun + +#include <vlibapi/api_helper_macros.h> + +#define foreach_session_api_msg \ +_(MAP_ANOTHER_SEGMENT_REPLY, map_another_segment_reply) \ +_(BIND_URI, bind_uri) \ +_(UNBIND_URI, unbind_uri) \ +_(CONNECT_URI, connect_uri) \ +_(DISCONNECT_SESSION, disconnect_session) \ +_(DISCONNECT_SESSION_REPLY, disconnect_session_reply) \ +_(ACCEPT_SESSION_REPLY, accept_session_reply) \ +_(RESET_SESSION_REPLY, reset_session_reply) \ +_(BIND_SOCK, bind_sock) \ +_(UNBIND_SOCK, unbind_sock) \ +_(CONNECT_SOCK, connect_sock) \ +_(DISCONNECT_SOCK, disconnect_sock) \ +_(DISCONNECT_SOCK_REPLY, disconnect_sock_reply) \ +_(ACCEPT_SOCK_REPLY, accept_sock_reply) \ +_(RESET_SOCK_REPLY, reset_sock_reply) \ + +static int +send_add_segment_callback (u32 api_client_index, const u8 * segment_name, + u32 segment_size) +{ + vl_api_map_another_segment_t *mp; + unix_shared_memory_queue_t *q; + + q = vl_api_client_index_to_input_queue (api_client_index); + + if (!q) + return -1; + + mp = vl_msg_api_alloc (sizeof (*mp)); + memset (mp, 0, sizeof (*mp)); + mp->_vl_msg_id = clib_host_to_net_u16 (VL_API_MAP_ANOTHER_SEGMENT); + mp->segment_size = segment_size; + strncpy ((char *) mp->segment_name, (char *) segment_name, + sizeof (mp->segment_name) - 1); + + vl_msg_api_send_shmem (q, (u8 *) & mp); + + return 0; +} + +static int +send_session_accept_uri_callback (stream_session_t * s) +{ + vl_api_accept_session_t *mp; + unix_shared_memory_queue_t *q, *vpp_queue; + application_t *server = application_get (s->app_index); + + q = vl_api_client_index_to_input_queue (server->api_client_index); + vpp_queue = session_manager_get_vpp_event_queue (s->thread_index); + + if (!q) + return -1; + + mp = vl_msg_api_alloc (sizeof (*mp)); + mp->_vl_msg_id = clib_host_to_net_u16 (VL_API_ACCEPT_SESSION); + + /* Note: session_type is the first octet in all types of sessions */ + + mp->accept_cookie = server->accept_cookie; + mp->server_rx_fifo = (u64) s->server_rx_fifo; + mp->server_tx_fifo = (u64) s->server_tx_fifo; + mp->session_thread_index = s->thread_index; + mp->session_index = s->session_index; + mp->session_type = s->session_type; + mp->vpp_event_queue_address = (u64) vpp_queue; + vl_msg_api_send_shmem (q, (u8 *) & mp); + + return 0; +} + +static void +send_session_disconnect_uri_callback (stream_session_t * s) +{ + vl_api_disconnect_session_t *mp; + unix_shared_memory_queue_t *q; + application_t *app = application_get (s->app_index); + + q = vl_api_client_index_to_input_queue (app->api_client_index); + + if (!q) + return; + + mp = vl_msg_api_alloc (sizeof (*mp)); + memset (mp, 0, sizeof (*mp)); + mp->_vl_msg_id = clib_host_to_net_u16 (VL_API_DISCONNECT_SESSION); + + mp->session_thread_index = s->thread_index; + mp->session_index = s->session_index; + vl_msg_api_send_shmem (q, (u8 *) & mp); +} + +static int +send_session_connected_uri_callback (u32 api_client_index, + stream_session_t * s, u8 is_fail) +{ + vl_api_connect_uri_reply_t *mp; + unix_shared_memory_queue_t *q; + application_t *app = application_lookup (api_client_index); + u8 *seg_name; + unix_shared_memory_queue_t *vpp_queue; + + q = vl_api_client_index_to_input_queue (app->api_client_index); + + if (!q) + return -1; + + mp = vl_msg_api_alloc (sizeof (*mp)); + mp->_vl_msg_id = clib_host_to_net_u16 (VL_API_CONNECT_URI_REPLY); + mp->context = app->api_context; + mp->retval = is_fail; + if (!is_fail) + { + vpp_queue = session_manager_get_vpp_event_queue (s->thread_index); + mp->server_rx_fifo = (u64) s->server_rx_fifo; + mp->server_tx_fifo = (u64) s->server_tx_fifo; + mp->session_thread_index = s->thread_index; + mp->session_index = s->session_index; + mp->session_type = s->session_type; + mp->vpp_event_queue_address = (u64) vpp_queue; + mp->client_event_queue_address = (u64) app->event_queue; + + session_manager_get_segment_info (s->server_segment_index, &seg_name, + &mp->segment_size); + mp->segment_name_length = vec_len (seg_name); + if (mp->segment_name_length) + clib_memcpy (mp->segment_name, seg_name, mp->segment_name_length); + } + + vl_msg_api_send_shmem (q, (u8 *) & mp); + + /* Remove client if connect failed */ + if (is_fail) + application_del (app); + + return 0; +} + +/** + * Redirect a connect_uri message to the indicated server. + * Only sent if the server has bound the related port with + * URI_OPTIONS_FLAGS_USE_FIFO + */ +static int +redirect_connect_uri_callback (u32 server_api_client_index, void *mp_arg) +{ + vl_api_connect_uri_t *mp = mp_arg; + unix_shared_memory_queue_t *server_q, *client_q; + vlib_main_t *vm = vlib_get_main (); + f64 timeout = vlib_time_now (vm) + 0.5; + int rv = 0; + + server_q = vl_api_client_index_to_input_queue (server_api_client_index); + + if (!server_q) + { + rv = VNET_API_ERROR_INVALID_VALUE; + goto out; + } + + client_q = vl_api_client_index_to_input_queue (mp->client_index); + if (!client_q) + { + rv = VNET_API_ERROR_INVALID_VALUE_2; + goto out; + } + + /* Tell the server the client's API queue address, so it can reply */ + mp->client_queue_address = (u64) client_q; + + /* + * Bounce message handlers MUST NOT block the data-plane. + * Spin waiting for the queue lock, but + */ + + while (vlib_time_now (vm) < timeout) + { + rv = + unix_shared_memory_queue_add (server_q, (u8 *) & mp, 1 /*nowait */ ); + switch (rv) + { + /* correctly enqueued */ + case 0: + return VNET_CONNECT_REDIRECTED; + + /* continue spinning, wait for pthread_mutex_trylock to work */ + case -1: + continue; + + /* queue stuffed, drop the msg */ + case -2: + rv = VNET_API_ERROR_QUEUE_FULL; + goto out; + } + } +out: + /* Dispose of the message */ + vl_msg_api_free (mp); + return rv; +} + +static u64 +make_session_handle (stream_session_t * s) +{ + return (u64) s->session_index << 32 | (u64) s->thread_index; +} + +static int +send_session_accept_callback (stream_session_t * s) +{ + vl_api_accept_sock_t *mp; + unix_shared_memory_queue_t *q, *vpp_queue; + application_t *server = application_get (s->app_index); + + q = vl_api_client_index_to_input_queue (server->api_client_index); + vpp_queue = session_manager_get_vpp_event_queue (s->thread_index); + + if (!q) + return -1; + + mp = vl_msg_api_alloc (sizeof (*mp)); + mp->_vl_msg_id = clib_host_to_net_u16 (VL_API_ACCEPT_SOCK); + + /* Note: session_type is the first octet in all types of sessions */ + + mp->accept_cookie = server->accept_cookie; + mp->server_rx_fifo = (u64) s->server_rx_fifo; + mp->server_tx_fifo = (u64) s->server_tx_fifo; + mp->handle = make_session_handle (s); + mp->vpp_event_queue_address = (u64) vpp_queue; + vl_msg_api_send_shmem (q, (u8 *) & mp); + + return 0; +} + +static int +send_session_connected_callback (u32 api_client_index, stream_session_t * s, + u8 is_fail) +{ + vl_api_connect_sock_reply_t *mp; + unix_shared_memory_queue_t *q; + application_t *app = application_lookup (api_client_index); + u8 *seg_name; + unix_shared_memory_queue_t *vpp_queue; + + q = vl_api_client_index_to_input_queue (app->api_client_index); + + if (!q) + return -1; + + mp = vl_msg_api_alloc (sizeof (*mp)); + mp->_vl_msg_id = clib_host_to_net_u16 (VL_API_CONNECT_SOCK_REPLY); + mp->context = app->api_context; + mp->retval = is_fail; + if (!is_fail) + { + vpp_queue = session_manager_get_vpp_event_queue (s->thread_index); + mp->server_rx_fifo = (u64) s->server_rx_fifo; + mp->server_tx_fifo = (u64) s->server_tx_fifo; + mp->handle = make_session_handle (s); + mp->vpp_event_queue_address = (u64) vpp_queue; + mp->client_event_queue_address = (u64) app->event_queue; + + session_manager_get_segment_info (s->server_segment_index, &seg_name, + &mp->segment_size); + mp->segment_name_length = vec_len (seg_name); + if (mp->segment_name_length) + clib_memcpy (mp->segment_name, seg_name, mp->segment_name_length); + } + + vl_msg_api_send_shmem (q, (u8 *) & mp); + + /* Remove client if connect failed */ + if (is_fail) + application_del (app); + + return 0; +} + +static void +send_session_disconnect_callback (stream_session_t * s) +{ + vl_api_disconnect_sock_t *mp; + unix_shared_memory_queue_t *q; + application_t *app = application_get (s->app_index); + + q = vl_api_client_index_to_input_queue (app->api_client_index); + + if (!q) + return; + + mp = vl_msg_api_alloc (sizeof (*mp)); + memset (mp, 0, sizeof (*mp)); + mp->_vl_msg_id = clib_host_to_net_u16 (VL_API_DISCONNECT_SOCK); + + mp->handle = make_session_handle (s); + vl_msg_api_send_shmem (q, (u8 *) & mp); +} + +/** + * Redirect a connect_uri message to the indicated server. + * Only sent if the server has bound the related port with + * URI_OPTIONS_FLAGS_USE_FIFO + */ +static int +redirect_connect_callback (u32 server_api_client_index, void *mp_arg) +{ + vl_api_connect_sock_t *mp = mp_arg; + unix_shared_memory_queue_t *server_q, *client_q; + vlib_main_t *vm = vlib_get_main (); + f64 timeout = vlib_time_now (vm) + 0.5; + int rv = 0; + + server_q = vl_api_client_index_to_input_queue (server_api_client_index); + + if (!server_q) + { + rv = VNET_API_ERROR_INVALID_VALUE; + goto out; + } + + client_q = vl_api_client_index_to_input_queue (mp->client_index); + if (!client_q) + { + rv = VNET_API_ERROR_INVALID_VALUE_2; + goto out; + } + + /* Tell the server the client's API queue address, so it can reply */ + mp->client_queue_address = (u64) client_q; + + /* + * Bounce message handlers MUST NOT block the data-plane. + * Spin waiting for the queue lock, but + */ + + while (vlib_time_now (vm) < timeout) + { + rv = + unix_shared_memory_queue_add (server_q, (u8 *) & mp, 1 /*nowait */ ); + switch (rv) + { + /* correctly enqueued */ + case 0: + return VNET_CONNECT_REDIRECTED; + + /* continue spinning, wait for pthread_mutex_trylock to work */ + case -1: + continue; + + /* queue stuffed, drop the msg */ + case -2: + rv = VNET_API_ERROR_QUEUE_FULL; + goto out; + } + } +out: + /* Dispose of the message */ + vl_msg_api_free (mp); + return rv; +} + +static session_cb_vft_t uri_session_cb_vft = { + .session_accept_callback = send_session_accept_uri_callback, + .session_disconnect_callback = send_session_disconnect_uri_callback, + .session_connected_callback = send_session_connected_uri_callback, + .add_segment_callback = send_add_segment_callback, + .redirect_connect_callback = redirect_connect_uri_callback +}; + +static session_cb_vft_t session_cb_vft = { + .session_accept_callback = send_session_accept_callback, + .session_disconnect_callback = send_session_disconnect_callback, + .session_connected_callback = send_session_connected_callback, + .add_segment_callback = send_add_segment_callback, + .redirect_connect_callback = redirect_connect_callback +}; + +static int +api_session_not_valid (u32 session_index, u32 thread_index) +{ + session_manager_main_t *smm = vnet_get_session_manager_main (); + stream_session_t *pool; + + if (thread_index >= vec_len (smm->sessions)) + return VNET_API_ERROR_INVALID_VALUE; + + pool = smm->sessions[thread_index]; + + if (pool_is_free_index (pool, session_index)) + return VNET_API_ERROR_INVALID_VALUE_2; + + return 0; +} + +static void +vl_api_bind_uri_t_handler (vl_api_bind_uri_t * mp) +{ + vl_api_bind_uri_reply_t *rmp; + vnet_bind_args_t _a, *a = &_a; + char segment_name[128]; + u32 segment_name_length; + int rv; + + _Static_assert (sizeof (u64) * SESSION_OPTIONS_N_OPTIONS <= + sizeof (mp->options), + "Out of options, fix api message definition"); + + segment_name_length = ARRAY_LEN (segment_name); + + memset (a, 0, sizeof (*a)); + + a->uri = (char *) mp->uri; + a->api_client_index = mp->client_index; + a->options = mp->options; + a->segment_name = segment_name; + a->segment_name_length = segment_name_length; + a->session_cb_vft = &uri_session_cb_vft; + + a->options[SESSION_OPTIONS_SEGMENT_SIZE] = mp->initial_segment_size; + a->options[SESSION_OPTIONS_ACCEPT_COOKIE] = mp->accept_cookie; + rv = vnet_bind_uri (a); + + /* *INDENT-OFF* */ + REPLY_MACRO2 (VL_API_BIND_URI_REPLY, ({ + rmp->retval = rv; + if (!rv) + { + rmp->segment_name_length = 0; + /* $$$$ policy? */ + rmp->segment_size = mp->initial_segment_size; + if (segment_name_length) + { + memcpy (rmp->segment_name, segment_name, segment_name_length); + rmp->segment_name_length = segment_name_length; + } + rmp->server_event_queue_address = a->server_event_queue_address; + } + })); + /* *INDENT-ON* */ + +} + +static void +vl_api_unbind_uri_t_handler (vl_api_unbind_uri_t * mp) +{ + vl_api_unbind_uri_reply_t *rmp; + int rv; + + rv = vnet_unbind_uri ((char *) mp->uri, mp->client_index); + + REPLY_MACRO (VL_API_UNBIND_URI_REPLY); +} + +static void +vl_api_connect_uri_t_handler (vl_api_connect_uri_t * mp) +{ + vnet_connect_args_t _a, *a = &_a; + + a->uri = (char *) mp->uri; + a->api_client_index = mp->client_index; + a->api_context = mp->context; + a->options = mp->options; + a->session_cb_vft = &uri_session_cb_vft; + a->mp = mp; + vnet_connect_uri (a); +} + +static void +vl_api_disconnect_session_t_handler (vl_api_disconnect_session_t * mp) +{ + vl_api_disconnect_session_reply_t *rmp; + int rv; + + rv = api_session_not_valid (mp->session_index, mp->session_thread_index); + if (!rv) + rv = vnet_disconnect_session (mp->client_index, mp->session_index, + mp->session_thread_index); + + REPLY_MACRO (VL_API_DISCONNECT_SESSION_REPLY); +} + +static void +vl_api_disconnect_session_reply_t_handler (vl_api_disconnect_session_reply_t * + mp) +{ + if (api_session_not_valid (mp->session_index, mp->session_thread_index)) + { + clib_warning ("Invalid session!"); + return; + } + + /* Client objected to disconnecting the session, log and continue */ + if (mp->retval) + { + clib_warning ("client retval %d", mp->retval); + return; + } + + /* Disconnect has been confirmed. Confirm close to transport */ + vnet_disconnect_session (mp->client_index, mp->session_index, + mp->session_thread_index); +} + +static void +vl_api_reset_session_reply_t_handler (vl_api_reset_session_reply_t * mp) +{ + stream_session_t *s; + + if (api_session_not_valid (mp->session_index, mp->session_thread_index)) + { + clib_warning ("Invalid session!"); + return; + } + + /* Client objected to resetting the session, log and continue */ + if (mp->retval) + { + clib_warning ("client retval %d", mp->retval); + return; + } + + s = stream_session_get (mp->session_index, mp->session_thread_index); + + /* This comes as a response to a reset, transport only waiting for + * confirmation to remove connection state, no need to disconnect */ + stream_session_cleanup (s); +} + +static void +vl_api_accept_session_reply_t_handler (vl_api_accept_session_reply_t * mp) +{ + stream_session_t *s; + int rv; + + if (api_session_not_valid (mp->session_index, mp->session_thread_index)) + return; + + s = stream_session_get (mp->session_index, mp->session_thread_index); + rv = mp->retval; + + if (rv) + { + /* Server isn't interested, kill the session */ + stream_session_disconnect (s); + return; + } + + s->session_state = SESSION_STATE_READY; +} + +static void +vl_api_map_another_segment_reply_t_handler (vl_api_map_another_segment_reply_t + * mp) +{ + clib_warning ("not implemented"); +} + +static void +vl_api_bind_sock_t_handler (vl_api_bind_sock_t * mp) +{ + vl_api_bind_sock_reply_t *rmp; + vnet_bind_args_t _a, *a = &_a; + char segment_name[128]; + u32 segment_name_length; + int rv; + + STATIC_ASSERT (sizeof (u64) * SESSION_OPTIONS_N_OPTIONS <= + sizeof (mp->options), + "Out of options, fix api message definition"); + + segment_name_length = ARRAY_LEN (segment_name); + + memset (a, 0, sizeof (*a)); + + clib_memcpy (&a->tep.ip, mp->ip, + (mp->is_ip4 ? sizeof (ip4_address_t) : + sizeof (ip6_address_t))); + a->tep.is_ip4 = mp->is_ip4; + a->tep.port = mp->port; + a->tep.vrf = mp->vrf; + + a->api_client_index = mp->client_index; + a->options = mp->options; + a->segment_name = segment_name; + a->segment_name_length = segment_name_length; + a->session_cb_vft = &session_cb_vft; + + rv = vnet_bind_uri (a); + + /* *INDENT-OFF* */ + REPLY_MACRO2 (VL_API_BIND_SOCK_REPLY, ({ + rmp->retval = rv; + if (!rv) + { + rmp->segment_name_length = 0; + rmp->segment_size = mp->options[SESSION_OPTIONS_SEGMENT_SIZE]; + if (segment_name_length) + { + memcpy(rmp->segment_name, segment_name, segment_name_length); + rmp->segment_name_length = segment_name_length; + } + rmp->server_event_queue_address = a->server_event_queue_address; + } + })); + /* *INDENT-ON* */ +} + +static void +vl_api_unbind_sock_t_handler (vl_api_unbind_sock_t * mp) +{ + vl_api_unbind_sock_reply_t *rmp; + vnet_unbind_args_t _a, *a = &_a; + int rv; + + a->api_client_index = mp->client_index; + a->handle = mp->handle; + + rv = vnet_unbind (a); + + REPLY_MACRO (VL_API_UNBIND_SOCK_REPLY); +} + +static void +vl_api_connect_sock_t_handler (vl_api_connect_sock_t * mp) +{ + vnet_connect_args_t _a, *a = &_a; + + clib_memcpy (&a->tep.ip, mp->ip, + (mp->is_ip4 ? sizeof (ip4_address_t) : + sizeof (ip6_address_t))); + a->tep.is_ip4 = mp->is_ip4; + a->tep.port = mp->port; + a->tep.vrf = mp->vrf; + a->options = mp->options; + a->session_cb_vft = &session_cb_vft; + a->api_context = mp->context; + a->mp = mp; + + vnet_connect (a); +} + +static void +vl_api_disconnect_sock_t_handler (vl_api_disconnect_sock_t * mp) +{ + vnet_disconnect_args_t _a, *a = &_a; + vl_api_disconnect_sock_reply_t *rmp; + int rv; + + a->api_client_index = mp->client_index; + a->handle = mp->handle; + rv = vnet_disconnect (a); + + REPLY_MACRO (VL_API_DISCONNECT_SOCK_REPLY); +} + +static void +vl_api_disconnect_sock_reply_t_handler (vl_api_disconnect_sock_reply_t * mp) +{ + vnet_disconnect_args_t _a, *a = &_a; + + /* Client objected to disconnecting the session, log and continue */ + if (mp->retval) + { + clib_warning ("client retval %d", mp->retval); + return; + } + + a->api_client_index = mp->client_index; + a->handle = mp->handle; + + vnet_disconnect (a); +} + +static void +vl_api_reset_sock_reply_t_handler (vl_api_reset_sock_reply_t * mp) +{ + stream_session_t *s; + u32 session_index, thread_index; + + /* Client objected to resetting the session, log and continue */ + if (mp->retval) + { + clib_warning ("client retval %d", mp->retval); + return; + } + + if (api_parse_session_handle (mp->handle, &session_index, &thread_index)) + { + clib_warning ("Invalid handle"); + return; + } + + s = stream_session_get (session_index, thread_index); + + /* This comes as a response to a reset, transport only waiting for + * confirmation to remove connection state, no need to disconnect */ + stream_session_cleanup (s); +} + +static void +vl_api_accept_sock_reply_t_handler (vl_api_accept_sock_reply_t * mp) +{ + stream_session_t *s; + u32 session_index, thread_index; + + if (api_parse_session_handle (mp->handle, &session_index, &thread_index)) + { + clib_warning ("Invalid handle"); + return; + } + s = stream_session_get (session_index, thread_index); + + if (mp->retval) + { + /* Server isn't interested, kill the session */ + stream_session_disconnect (s); + return; + } + + s->session_state = SESSION_STATE_READY; +} + +#define vl_msg_name_crc_list +#include <vnet/vnet_all_api_h.h> +#undef vl_msg_name_crc_list + +static void +setup_message_id_table (api_main_t * am) +{ +#define _(id,n,crc) vl_msg_api_add_msg_name_crc (am, #n "_" #crc, id); + foreach_vl_msg_name_crc_session; +#undef _ +} + +/* + * session_api_hookup + * Add uri's API message handlers to the table. + * vlib has alread mapped shared memory and + * added the client registration handlers. + * See .../open-repo/vlib/memclnt_vlib.c:memclnt_process() + */ +static clib_error_t * +session_api_hookup (vlib_main_t * vm) +{ + api_main_t *am = &api_main; + +#define _(N,n) \ + vl_msg_api_set_handlers(VL_API_##N, #n, \ + vl_api_##n##_t_handler, \ + vl_noop_handler, \ + vl_api_##n##_t_endian, \ + vl_api_##n##_t_print, \ + sizeof(vl_api_##n##_t), 1); + foreach_session_api_msg; +#undef _ + + /* + * Messages which bounce off the data-plane to + * an API client. Simply tells the message handling infra not + * to free the message. + * + * Bounced message handlers MUST NOT block the data plane + */ + am->message_bounce[VL_API_CONNECT_URI] = 1; + am->message_bounce[VL_API_CONNECT_SOCK] = 1; + + /* + * Set up the (msg_name, crc, message-id) table + */ + setup_message_id_table (am); + + return 0; +} + +VLIB_API_INIT_FUNCTION (session_api_hookup); +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/session_cli.c b/src/vnet/session/session_cli.c new file mode 100644 index 00000000000..b2943a1ccc5 --- /dev/null +++ b/src/vnet/session/session_cli.c @@ -0,0 +1,189 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#include <vnet/session/application.h> +#include <vnet/session/session.h> + +/** + * Format stream session as per the following format + * + * verbose: + * "Connection", "Rx fifo", "Tx fifo", "Session Index" + * non-verbose: + * "Connection" + */ +u8 * +format_stream_session (u8 * s, va_list * args) +{ + stream_session_t *ss = va_arg (*args, stream_session_t *); + int verbose = va_arg (*args, int); + transport_proto_vft_t *tp_vft; + u8 *str = 0; + + tp_vft = session_get_transport_vft (ss->session_type); + + if (verbose) + str = format (0, "%-20llp%-20llp%-15lld", ss->server_rx_fifo, + ss->server_tx_fifo, stream_session_get_index (ss)); + + if (ss->session_state == SESSION_STATE_READY) + { + s = format (s, "%-40U%v", tp_vft->format_connection, + ss->connection_index, ss->thread_index, str); + } + else if (ss->session_state == SESSION_STATE_LISTENING) + { + s = format (s, "%-40U%v", tp_vft->format_listener, ss->connection_index, + str); + } + else if (ss->session_state == SESSION_STATE_READY) + { + s = + format (s, "%-40U%v", tp_vft->format_half_open, ss->connection_index, + str); + } + else if (ss->session_state == SESSION_STATE_CLOSED) + { + s = format (s, "[CL] %-40U%v", tp_vft->format_connection, + ss->connection_index, ss->thread_index, str); + } + else + { + clib_warning ("Session in unknown state!"); + } + + vec_free (str); + + return s; +} + +static clib_error_t * +show_session_command_fn (vlib_main_t * vm, unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + session_manager_main_t *smm = &session_manager_main; + int verbose = 0, i; + stream_session_t *pool; + stream_session_t *s; + u8 *str = 0; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "verbose")) + verbose = 1; + else + break; + } + + for (i = 0; i < vec_len (smm->sessions); i++) + { + u32 once_per_pool; + pool = smm->sessions[i]; + + once_per_pool = 1; + + if (pool_elts (pool)) + { + + vlib_cli_output (vm, "Thread %d: %d active sessions", + i, pool_elts (pool)); + if (verbose) + { + if (once_per_pool) + { + str = format (str, "%-40s%-20s%-20s%-15s", + "Connection", "Rx fifo", "Tx fifo", + "Session Index"); + vlib_cli_output (vm, "%v", str); + vec_reset_length (str); + once_per_pool = 0; + } + + /* *INDENT-OFF* */ + pool_foreach (s, pool, + ({ + vlib_cli_output (vm, "%U", format_stream_session, s, verbose); + })); + /* *INDENT-ON* */ + } + } + else + vlib_cli_output (vm, "Thread %d: no active sessions", i); + } + vec_free (str); + + return 0; +} + +VLIB_CLI_COMMAND (show_uri_command, static) = +{ +.path = "show session",.short_help = "show session [verbose]",.function = + show_session_command_fn,}; + + +static clib_error_t * +clear_session_command_fn (vlib_main_t * vm, unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + session_manager_main_t *smm = &session_manager_main; + u32 thread_index = 0; + u32 session_index = ~0; + stream_session_t *pool, *session; + application_t *server; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "thread %d", &thread_index)) + ; + else if (unformat (input, "session %d", &session_index)) + ; + else + return clib_error_return (0, "unknown input `%U'", + format_unformat_error, input); + } + + if (session_index == ~0) + return clib_error_return (0, "session <nn> required, but not set."); + + if (thread_index > vec_len (smm->sessions)) + return clib_error_return (0, "thread %d out of range [0-%d]", + thread_index, vec_len (smm->sessions)); + + pool = smm->sessions[thread_index]; + + if (pool_is_free_index (pool, session_index)) + return clib_error_return (0, "session %d not active", session_index); + + session = pool_elt_at_index (pool, session_index); + server = application_get (session->app_index); + + /* Disconnect both app and transport */ + server->cb_fns.session_disconnect_callback (session); + + return 0; +} + +VLIB_CLI_COMMAND (clear_uri_session_command, static) = +{ +.path = "clear session",.short_help = + "clear session thread <thread> session <index>",.function = + clear_session_command_fn,}; + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/transport.c b/src/vnet/session/transport.c new file mode 100644 index 00000000000..abd94ba4f1d --- /dev/null +++ b/src/vnet/session/transport.c @@ -0,0 +1,64 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/session/transport.h> + +u32 +transport_endpoint_lookup (transport_endpoint_table_t *ht, ip46_address_t *ip, + u16 port) +{ + clib_bihash_kv_24_8_t kv; + int rv; + + kv.key[0] = ip->as_u64[0]; + kv.key[1] = ip->as_u64[1]; + kv.key[2] = port; + + rv = clib_bihash_search_inline_24_8 (ht, &kv); + if (rv == 0) + return kv.value; + + return TRANSPORT_ENDPOINT_INVALID_INDEX; +} + +void +transport_endpoint_table_add (transport_endpoint_table_t *ht, + transport_endpoint_t *te, u32 value) +{ + clib_bihash_kv_24_8_t kv; + + kv.key[0] = te->ip.as_u64[0]; + kv.key[1] = te->ip.as_u64[1]; + kv.key[2] = te->port; + kv.value = value; + + clib_bihash_add_del_24_8 (ht, &kv, 1); +} + +void +transport_endpoint_table_del (transport_endpoint_table_t *ht, + transport_endpoint_t *te) +{ + clib_bihash_kv_24_8_t kv; + + kv.key[0] = te->ip.as_u64[0]; + kv.key[1] = te->ip.as_u64[1]; + kv.key[2] = te->port; + + clib_bihash_add_del_24_8 (ht, &kv, 0); +} + + + diff --git a/src/vnet/session/transport.h b/src/vnet/session/transport.h new file mode 100644 index 00000000000..2d4415ba6a3 --- /dev/null +++ b/src/vnet/session/transport.h @@ -0,0 +1,250 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef VNET_VNET_URI_TRANSPORT_H_ +#define VNET_VNET_URI_TRANSPORT_H_ + +#include <vnet/vnet.h> +#include <vnet/ip/ip.h> +#include <vppinfra/bihash_16_8.h> +#include <vppinfra/bihash_48_8.h> + +/* + * Protocol independent transport properties associated to a session + */ +typedef struct _transport_connection +{ + ip46_address_t rmt_ip; /**< Remote IP */ + ip46_address_t lcl_ip; /**< Local IP */ + u16 lcl_port; /**< Local port */ + u16 rmt_port; /**< Remote port */ + u8 proto; /**< Transport protocol id */ + + u32 s_index; /**< Parent session index */ + u32 c_index; /**< Connection index in transport pool */ + u8 is_ip4; /**< Flag if IP4 connection */ + u32 thread_index; /**< Worker-thread index */ + + /** Macros for 'derived classes' where base is named "connection" */ +#define c_lcl_ip connection.lcl_ip +#define c_rmt_ip connection.rmt_ip +#define c_lcl_ip4 connection.lcl_ip.ip4 +#define c_rmt_ip4 connection.rmt_ip.ip4 +#define c_lcl_ip6 connection.lcl_ip.ip6 +#define c_rmt_ip6 connection.rmt_ip.ip6 +#define c_lcl_port connection.lcl_port +#define c_rmt_port connection.rmt_port +#define c_proto connection.proto +#define c_state connection.state +#define c_s_index connection.s_index +#define c_c_index connection.c_index +#define c_is_ip4 connection.is_ip4 +#define c_thread_index connection.thread_index +} transport_connection_t; + +/* + * Transport protocol virtual function table + */ +typedef struct _transport_proto_vft +{ + /* + * Setup + */ + u32 (*bind) (vlib_main_t *, u32, ip46_address_t *, u16); + u32 (*unbind) (vlib_main_t *, u32); + int (*open) (ip46_address_t * addr, u16 port_host_byte_order); + void (*close) (u32 conn_index, u32 thread_index); + void (*cleanup) (u32 conn_index, u32 thread_index); + + /* + * Transmission + */ + u32 (*push_header) (transport_connection_t * tconn, vlib_buffer_t * b); + u16 (*send_mss) (transport_connection_t * tc); + u32 (*send_space) (transport_connection_t * tc); + u32 (*rx_fifo_offset) (transport_connection_t * tc); + + /* + * Connection retrieval + */ + transport_connection_t *(*get_connection) (u32 conn_idx, u32 thread_idx); + transport_connection_t *(*get_listener) (u32 conn_index); + transport_connection_t *(*get_half_open) (u32 conn_index); + + /* + * Format + */ + u8 *(*format_connection) (u8 * s, va_list * args); + u8 *(*format_listener) (u8 * s, va_list * args); + u8 *(*format_half_open) (u8 * s, va_list * args); + +} transport_proto_vft_t; + +/* 16 octets */ +typedef CLIB_PACKED (struct + { + union + { + struct + { + ip4_address_t src; ip4_address_t dst; + u16 src_port; + u16 dst_port; + /* align by making this 4 octets even though its a 1-bit field + * NOTE: avoid key overlap with other transports that use 5 tuples for + * session identification. + */ + u32 proto; + }; + u64 as_u64[2]; + }; + }) v4_connection_key_t; + +typedef CLIB_PACKED (struct + { + union + { + struct + { + /* 48 octets */ + ip6_address_t src; ip6_address_t dst; + u16 src_port; + u16 dst_port; u32 proto; u8 unused_for_now[8]; + }; u64 as_u64[6]; + }; + }) v6_connection_key_t; + +typedef clib_bihash_kv_16_8_t session_kv4_t; +typedef clib_bihash_kv_48_8_t session_kv6_t; + +always_inline void +make_v4_ss_kv (session_kv4_t * kv, ip4_address_t * lcl, ip4_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto) +{ + v4_connection_key_t key; + memset (&key, 0, sizeof (v4_connection_key_t)); + + key.src.as_u32 = lcl->as_u32; + key.dst.as_u32 = rmt->as_u32; + key.src_port = lcl_port; + key.dst_port = rmt_port; + key.proto = proto; + + kv->key[0] = key.as_u64[0]; + kv->key[1] = key.as_u64[1]; + kv->value = ~0ULL; +} + +always_inline void +make_v4_listener_kv (session_kv4_t * kv, ip4_address_t * lcl, u16 lcl_port, + u8 proto) +{ + v4_connection_key_t key; + memset (&key, 0, sizeof (v4_connection_key_t)); + + key.src.as_u32 = lcl->as_u32; + key.dst.as_u32 = 0; + key.src_port = lcl_port; + key.dst_port = 0; + key.proto = proto; + + kv->key[0] = key.as_u64[0]; + kv->key[1] = key.as_u64[1]; + kv->value = ~0ULL; +} + +always_inline void +make_v4_ss_kv_from_tc (session_kv4_t * kv, transport_connection_t * t) +{ + return make_v4_ss_kv (kv, &t->lcl_ip.ip4, &t->rmt_ip.ip4, t->lcl_port, + t->rmt_port, t->proto); +} + +always_inline void +make_v6_ss_kv (session_kv6_t * kv, ip6_address_t * lcl, ip6_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto) +{ + v6_connection_key_t key; + memset (&key, 0, sizeof (v6_connection_key_t)); + + key.src.as_u64[0] = lcl->as_u64[0]; + key.src.as_u64[1] = lcl->as_u64[1]; + key.dst.as_u64[0] = rmt->as_u64[0]; + key.dst.as_u64[1] = rmt->as_u64[1]; + key.src_port = lcl_port; + key.dst_port = rmt_port; + key.proto = proto; + + kv->key[0] = key.as_u64[0]; + kv->key[1] = key.as_u64[1]; + kv->value = ~0ULL; +} + +always_inline void +make_v6_listener_kv (session_kv6_t * kv, ip6_address_t * lcl, u16 lcl_port, + u8 proto) +{ + v6_connection_key_t key; + memset (&key, 0, sizeof (v6_connection_key_t)); + + key.src.as_u64[0] = lcl->as_u64[0]; + key.src.as_u64[1] = lcl->as_u64[1]; + key.dst.as_u64[0] = 0; + key.dst.as_u64[1] = 0; + key.src_port = lcl_port; + key.dst_port = 0; + key.proto = proto; + + kv->key[0] = key.as_u64[0]; + kv->key[1] = key.as_u64[1]; + kv->value = ~0ULL; +} + +always_inline void +make_v6_ss_kv_from_tc (session_kv6_t * kv, transport_connection_t * t) +{ + make_v6_ss_kv (kv, &t->lcl_ip.ip6, &t->rmt_ip.ip6, t->lcl_port, + t->rmt_port, t->proto); +} + +typedef struct _transport_endpoint +{ + ip46_address_t ip; + u16 port; + u8 is_ip4; + u32 vrf; +} transport_endpoint_t; + +typedef clib_bihash_24_8_t transport_endpoint_table_t; + +#define TRANSPORT_ENDPOINT_INVALID_INDEX ((u32)~0) + +u32 +transport_endpoint_lookup (transport_endpoint_table_t * ht, + ip46_address_t * ip, u16 port); +void transport_endpoint_table_add (transport_endpoint_table_t * ht, + transport_endpoint_t * te, u32 value); +void transport_endpoint_table_del (transport_endpoint_table_t * ht, + transport_endpoint_t * te); + +#endif /* VNET_VNET_URI_TRANSPORT_H_ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp.c b/src/vnet/tcp/tcp.c new file mode 100644 index 00000000000..0f9b7097b42 --- /dev/null +++ b/src/vnet/tcp/tcp.c @@ -0,0 +1,708 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/tcp/tcp.h> +#include <vnet/session/session.h> +#include <vnet/fib/fib.h> +#include <math.h> + +tcp_main_t tcp_main; + +static u32 +tcp_connection_bind (vlib_main_t * vm, u32 session_index, ip46_address_t * ip, + u16 port_host_byte_order, u8 is_ip4) +{ + tcp_main_t *tm = &tcp_main; + tcp_connection_t *listener; + + pool_get (tm->listener_pool, listener); + memset (listener, 0, sizeof (*listener)); + + listener->c_c_index = listener - tm->listener_pool; + listener->c_lcl_port = clib_host_to_net_u16 (port_host_byte_order); + + if (is_ip4) + listener->c_lcl_ip4.as_u32 = ip->ip4.as_u32; + else + clib_memcpy (&listener->c_lcl_ip6, &ip->ip6, sizeof (ip6_address_t)); + + listener->c_s_index = session_index; + listener->c_proto = SESSION_TYPE_IP4_TCP; + listener->state = TCP_STATE_LISTEN; + listener->c_is_ip4 = 1; + + return listener->c_c_index; +} + +u32 +tcp_session_bind_ip4 (vlib_main_t * vm, u32 session_index, + ip46_address_t * ip, u16 port_host_byte_order) +{ + return tcp_connection_bind (vm, session_index, ip, port_host_byte_order, 1); +} + +u32 +tcp_session_bind_ip6 (vlib_main_t * vm, u32 session_index, + ip46_address_t * ip, u16 port_host_byte_order) +{ + return tcp_connection_bind (vm, session_index, ip, port_host_byte_order, 0); + +} + +static void +tcp_session_unbind (u32 listener_index) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + pool_put_index (tm->listener_pool, listener_index); +} + +u32 +tcp_session_unbind_ip4 (vlib_main_t * vm, u32 listener_index) +{ + tcp_session_unbind (listener_index); + return 0; +} + +u32 +tcp_session_unbind_ip6 (vlib_main_t * vm, u32 listener_index) +{ + tcp_session_unbind (listener_index); + return 0; +} + +transport_connection_t * +tcp_session_get_listener (u32 listener_index) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + tcp_connection_t *tc; + tc = pool_elt_at_index (tm->listener_pool, listener_index); + return &tc->connection; +} + +/** + * Cleans up connection state. + * + * No notifications. + */ +void +tcp_connection_cleanup (tcp_connection_t * tc) +{ + tcp_main_t *tm = &tcp_main; + u32 tepi; + transport_endpoint_t *tep; + + /* Cleanup local endpoint if this was an active connect */ + tepi = transport_endpoint_lookup (&tm->local_endpoints_table, &tc->c_lcl_ip, + tc->c_lcl_port); + + /*XXX lock */ + if (tepi != TRANSPORT_ENDPOINT_INVALID_INDEX) + { + tep = pool_elt_at_index (tm->local_endpoints, tepi); + transport_endpoint_table_del (&tm->local_endpoints_table, tep); + pool_put (tm->local_endpoints, tep); + } + + /* Make sure all timers are cleared */ + tcp_connection_timers_reset (tc); + + /* Check if half-open */ + if (tc->state == TCP_STATE_SYN_SENT) + pool_put (tm->half_open_connections, tc); + else + pool_put (tm->connections[tc->c_thread_index], tc); +} + +/** + * Connection removal. + * + * This should be called only once connection enters CLOSED state. Note + * that it notifies the session of the removal event, so if the goal is to + * just remove the connection, call tcp_connection_cleanup instead. + */ +void +tcp_connection_del (tcp_connection_t * tc) +{ + stream_session_delete_notify (&tc->connection); + tcp_connection_cleanup (tc); +} + +/** + * Begin connection closing procedure. + * + * If at the end the connection is not in CLOSED state, it is not removed. + * Instead, we rely on on TCP to advance through state machine to either + * 1) LAST_ACK (passive close) whereby when the last ACK is received + * tcp_connection_del is called. This notifies session of the delete and + * calls cleanup. + * 2) TIME_WAIT (active close) whereby after 2MSL the 2MSL timer triggers + * and cleanup is called. + */ +void +tcp_connection_close (tcp_connection_t * tc) +{ + /* Send FIN if needed */ + if (tc->state == TCP_STATE_ESTABLISHED || tc->state == TCP_STATE_SYN_RCVD + || tc->state == TCP_STATE_CLOSE_WAIT) + tcp_send_fin (tc); + + /* Switch state */ + if (tc->state == TCP_STATE_ESTABLISHED || tc->state == TCP_STATE_SYN_RCVD) + tc->state = TCP_STATE_FIN_WAIT_1; + else if (tc->state == TCP_STATE_SYN_SENT) + tc->state = TCP_STATE_CLOSED; + else if (tc->state == TCP_STATE_CLOSE_WAIT) + tc->state = TCP_STATE_LAST_ACK; + + /* Half-close connections are not supported XXX */ + + if (tc->state == TCP_STATE_CLOSED) + tcp_connection_del (tc); +} + +void +tcp_session_close (u32 conn_index, u32 thread_index) +{ + tcp_connection_t *tc; + tc = tcp_connection_get (conn_index, thread_index); + tcp_connection_close (tc); +} + +void +tcp_session_cleanup (u32 conn_index, u32 thread_index) +{ + tcp_connection_t *tc; + tc = tcp_connection_get (conn_index, thread_index); + tcp_connection_cleanup (tc); +} + +void * +ip_interface_get_first_ip (u32 sw_if_index, u8 is_ip4) +{ + ip_lookup_main_t *lm4 = &ip4_main.lookup_main; + ip_lookup_main_t *lm6 = &ip6_main.lookup_main; + ip_interface_address_t *ia = 0; + + if (is_ip4) + { + /* *INDENT-OFF* */ + foreach_ip_interface_address (lm4, ia, sw_if_index, 1 /* unnumbered */ , + ({ + return ip_interface_address_get_address (lm4, ia); + })); + /* *INDENT-ON* */ + } + else + { + /* *INDENT-OFF* */ + foreach_ip_interface_address (lm6, ia, sw_if_index, 1 /* unnumbered */ , + ({ + return ip_interface_address_get_address (lm6, ia); + })); + /* *INDENT-ON* */ + } + + return 0; +} + +/** + * Allocate local port and add if successful add entry to local endpoint + * table to mark the pair as used. + */ +u16 +tcp_allocate_local_port (tcp_main_t * tm, ip46_address_t * ip) +{ + u8 unique = 0; + transport_endpoint_t *tep; + u32 time_now, tei; + u16 min = 1024, max = 65535, tries; /* XXX configurable ? */ + + tries = max - min; + time_now = tcp_time_now (); + + /* Start at random point or max */ + pool_get (tm->local_endpoints, tep); + clib_memcpy (&tep->ip, ip, sizeof (*ip)); + tep->port = random_u32 (&time_now) << 16; + tep->port = tep->port < min ? max : tep->port; + + /* Search for first free slot */ + while (tries) + { + tei = transport_endpoint_lookup (&tm->local_endpoints_table, &tep->ip, + tep->port); + if (tei == TRANSPORT_ENDPOINT_INVALID_INDEX) + { + unique = 1; + break; + } + + tep->port--; + + if (tep->port < min) + tep->port = max; + + tries--; + } + + if (unique) + { + transport_endpoint_table_add (&tm->local_endpoints_table, tep, + tep - tm->local_endpoints); + + return tep->port; + } + + /* Failed */ + pool_put (tm->local_endpoints, tep); + return -1; +} + +/** + * Initialize all connection timers as invalid + */ +void +tcp_connection_timers_init (tcp_connection_t * tc) +{ + int i; + + /* Set all to invalid */ + for (i = 0; i < TCP_N_TIMERS; i++) + { + tc->timers[i] = TCP_TIMER_HANDLE_INVALID; + } + + tc->rto = TCP_RTO_INIT; +} + +/** + * Stop all connection timers + */ +void +tcp_connection_timers_reset (tcp_connection_t * tc) +{ + int i; + for (i = 0; i < TCP_N_TIMERS; i++) + { + tcp_timer_reset (tc, i); + } +} + +/** Initialize tcp connection variables + * + * Should be called after having received a msg from the peer, i.e., a SYN or + * a SYNACK, such that connection options have already been exchanged. */ +void +tcp_connection_init_vars (tcp_connection_t * tc) +{ + tcp_connection_timers_init (tc); + tcp_set_snd_mss (tc); + tc->sack_sb.head = TCP_INVALID_SACK_HOLE_INDEX; + tcp_cc_init (tc); +} + +int +tcp_connection_open (ip46_address_t * rmt_addr, u16 rmt_port, u8 is_ip4) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + tcp_connection_t *tc; + fib_prefix_t prefix; + u32 fei, sw_if_index; + ip46_address_t lcl_addr; + u16 lcl_port; + + /* + * Find the local address and allocate port + */ + memset (&lcl_addr, 0, sizeof (lcl_addr)); + + /* Find a FIB path to the destination */ + clib_memcpy (&prefix.fp_addr, rmt_addr, sizeof (*rmt_addr)); + prefix.fp_proto = is_ip4 ? FIB_PROTOCOL_IP4 : FIB_PROTOCOL_IP6; + prefix.fp_len = is_ip4 ? 32 : 128; + + fei = fib_table_lookup (0, &prefix); + + /* Couldn't find route to destination. Bail out. */ + if (fei == FIB_NODE_INDEX_INVALID) + return -1; + + sw_if_index = fib_entry_get_resolving_interface (fei); + + if (sw_if_index == (u32) ~ 0) + return -1; + + if (is_ip4) + { + ip4_address_t *ip4; + ip4 = ip_interface_get_first_ip (sw_if_index, 1); + lcl_addr.ip4.as_u32 = ip4->as_u32; + } + else + { + ip6_address_t *ip6; + ip6 = ip_interface_get_first_ip (sw_if_index, 0); + clib_memcpy (&lcl_addr.ip6, ip6, sizeof (*ip6)); + } + + /* Allocate source port */ + lcl_port = tcp_allocate_local_port (tm, &lcl_addr); + if (lcl_port < 1) + return -1; + + /* + * Create connection and send SYN + */ + + pool_get (tm->half_open_connections, tc); + memset (tc, 0, sizeof (*tc)); + + clib_memcpy (&tc->c_rmt_ip, rmt_addr, sizeof (ip46_address_t)); + clib_memcpy (&tc->c_lcl_ip, &lcl_addr, sizeof (ip46_address_t)); + tc->c_rmt_port = clib_host_to_net_u16 (rmt_port); + tc->c_lcl_port = clib_host_to_net_u16 (lcl_port); + tc->c_c_index = tc - tm->half_open_connections; + tc->c_is_ip4 = is_ip4; + + /* The other connection vars will be initialized after SYN ACK */ + tcp_connection_timers_init (tc); + + tcp_send_syn (tc); + + tc->state = TCP_STATE_SYN_SENT; + + return tc->c_c_index; +} + +int +tcp_session_open_ip4 (ip46_address_t * addr, u16 port) +{ + return tcp_connection_open (addr, port, 1); +} + +int +tcp_session_open_ip6 (ip46_address_t * addr, u16 port) +{ + return tcp_connection_open (addr, port, 0); +} + +u8 * +format_tcp_session_ip4 (u8 * s, va_list * args) +{ + u32 tci = va_arg (*args, u32); + u32 thread_index = va_arg (*args, u32); + tcp_connection_t *tc; + + tc = tcp_connection_get (tci, thread_index); + + s = format (s, "[%s] %U:%d->%U:%d", "tcp", format_ip4_address, + &tc->c_lcl_ip4, clib_net_to_host_u16 (tc->c_lcl_port), + format_ip4_address, &tc->c_rmt_ip4, + clib_net_to_host_u16 (tc->c_rmt_port)); + + return s; +} + +u8 * +format_tcp_session_ip6 (u8 * s, va_list * args) +{ + u32 tci = va_arg (*args, u32); + u32 thread_index = va_arg (*args, u32); + tcp_connection_t *tc = tcp_connection_get (tci, thread_index); + s = format (s, "[%s] %U:%d->%U:%d", "tcp", format_ip6_address, + &tc->c_lcl_ip6, clib_net_to_host_u16 (tc->c_lcl_port), + format_ip6_address, &tc->c_rmt_ip6, + clib_net_to_host_u16 (tc->c_rmt_port)); + return s; +} + +u8 * +format_tcp_listener_session_ip4 (u8 * s, va_list * args) +{ + u32 tci = va_arg (*args, u32); + tcp_connection_t *tc = tcp_listener_get (tci); + s = format (s, "[%s] %U:%d->%U:%d", "tcp", format_ip4_address, + &tc->c_lcl_ip4, clib_net_to_host_u16 (tc->c_lcl_port), + format_ip4_address, &tc->c_rmt_ip4, + clib_net_to_host_u16 (tc->c_rmt_port)); + return s; +} + +u8 * +format_tcp_listener_session_ip6 (u8 * s, va_list * args) +{ + u32 tci = va_arg (*args, u32); + tcp_connection_t *tc = tcp_listener_get (tci); + s = format (s, "[%s] %U:%d->%U:%d", "tcp", format_ip6_address, + &tc->c_lcl_ip6, clib_net_to_host_u16 (tc->c_lcl_port), + format_ip6_address, &tc->c_rmt_ip6, + clib_net_to_host_u16 (tc->c_rmt_port)); + return s; +} + +u8 * +format_tcp_half_open_session_ip4 (u8 * s, va_list * args) +{ + u32 tci = va_arg (*args, u32); + tcp_connection_t *tc = tcp_half_open_connection_get (tci); + s = format (s, "[%s] %U:%d->%U:%d", "tcp", format_ip4_address, + &tc->c_lcl_ip4, clib_net_to_host_u16 (tc->c_lcl_port), + format_ip4_address, &tc->c_rmt_ip4, + clib_net_to_host_u16 (tc->c_rmt_port)); + return s; +} + +u8 * +format_tcp_half_open_session_ip6 (u8 * s, va_list * args) +{ + u32 tci = va_arg (*args, u32); + tcp_connection_t *tc = tcp_half_open_connection_get (tci); + s = format (s, "[%s] %U:%d->%U:%d", "tcp", format_ip6_address, + &tc->c_lcl_ip6, clib_net_to_host_u16 (tc->c_lcl_port), + format_ip6_address, &tc->c_rmt_ip6, + clib_net_to_host_u16 (tc->c_rmt_port)); + return s; +} + +transport_connection_t * +tcp_session_get_transport (u32 conn_index, u32 thread_index) +{ + tcp_connection_t *tc = tcp_connection_get (conn_index, thread_index); + return &tc->connection; +} + +transport_connection_t * +tcp_half_open_session_get_transport (u32 conn_index) +{ + tcp_connection_t *tc = tcp_half_open_connection_get (conn_index); + return &tc->connection; +} + +u16 +tcp_session_send_mss (transport_connection_t * trans_conn) +{ + tcp_connection_t *tc = (tcp_connection_t *) trans_conn; + return tc->snd_mss; +} + +u32 +tcp_session_send_space (transport_connection_t * trans_conn) +{ + tcp_connection_t *tc = (tcp_connection_t *) trans_conn; + return tcp_available_snd_space (tc); +} + +u32 +tcp_session_rx_fifo_offset (transport_connection_t * trans_conn) +{ + tcp_connection_t *tc = (tcp_connection_t *) trans_conn; + return (tc->snd_una_max - tc->snd_una); +} + +/* *INDENT-OFF* */ +const static transport_proto_vft_t tcp4_proto = { + .bind = tcp_session_bind_ip4, + .unbind = tcp_session_unbind_ip4, + .push_header = tcp_push_header, + .get_connection = tcp_session_get_transport, + .get_listener = tcp_session_get_listener, + .get_half_open = tcp_half_open_session_get_transport, + .open = tcp_session_open_ip4, + .close = tcp_session_close, + .cleanup = tcp_session_cleanup, + .send_mss = tcp_session_send_mss, + .send_space = tcp_session_send_space, + .rx_fifo_offset = tcp_session_rx_fifo_offset, + .format_connection = format_tcp_session_ip4, + .format_listener = format_tcp_listener_session_ip4, + .format_half_open = format_tcp_half_open_session_ip4 +}; + +const static transport_proto_vft_t tcp6_proto = { + .bind = tcp_session_bind_ip6, + .unbind = tcp_session_unbind_ip6, + .push_header = tcp_push_header, + .get_connection = tcp_session_get_transport, + .get_listener = tcp_session_get_listener, + .get_half_open = tcp_half_open_session_get_transport, + .open = tcp_session_open_ip6, + .close = tcp_session_close, + .cleanup = tcp_session_cleanup, + .send_mss = tcp_session_send_mss, + .send_space = tcp_session_send_space, + .rx_fifo_offset = tcp_session_rx_fifo_offset, + .format_connection = format_tcp_session_ip6, + .format_listener = format_tcp_listener_session_ip6, + .format_half_open = format_tcp_half_open_session_ip6 +}; +/* *INDENT-ON* */ + +void +tcp_timer_keep_handler (u32 conn_index) +{ + u32 cpu_index = os_get_cpu_number (); + tcp_connection_t *tc; + + tc = tcp_connection_get (conn_index, cpu_index); + tc->timers[TCP_TIMER_KEEP] = TCP_TIMER_HANDLE_INVALID; + + tcp_connection_close (tc); +} + +void +tcp_timer_establish_handler (u32 conn_index) +{ + tcp_connection_t *tc; + u8 sst; + + tc = tcp_half_open_connection_get (conn_index); + tc->timers[TCP_TIMER_ESTABLISH] = TCP_TIMER_HANDLE_INVALID; + + ASSERT (tc->state == TCP_STATE_SYN_SENT); + + sst = tc->c_is_ip4 ? SESSION_TYPE_IP4_TCP : SESSION_TYPE_IP6_TCP; + stream_session_connect_notify (&tc->connection, sst, 1 /* fail */ ); + + tcp_connection_cleanup (tc); +} + +void +tcp_timer_2msl_handler (u32 conn_index) +{ + u32 cpu_index = os_get_cpu_number (); + tcp_connection_t *tc; + + tc = tcp_connection_get (conn_index, cpu_index); + tc->timers[TCP_TIMER_2MSL] = TCP_TIMER_HANDLE_INVALID; + + tcp_connection_del (tc); +} + +/* *INDENT-OFF* */ +static timer_expiration_handler *timer_expiration_handlers[TCP_N_TIMERS] = +{ + tcp_timer_retransmit_handler, + tcp_timer_delack_handler, + 0, + tcp_timer_keep_handler, + tcp_timer_2msl_handler, + tcp_timer_retransmit_syn_handler, + tcp_timer_establish_handler +}; +/* *INDENT-ON* */ + +static void +tcp_expired_timers_dispatch (u32 * expired_timers) +{ + int i; + u32 connection_index, timer_id; + + for (i = 0; i < vec_len (expired_timers); i++) + { + /* Get session index and timer id */ + connection_index = expired_timers[i] & 0x0FFFFFFF; + timer_id = expired_timers[i] >> 28; + + /* Handle expiration */ + (*timer_expiration_handlers[timer_id]) (connection_index); + } +} + +void +tcp_initialize_timer_wheels (tcp_main_t * tm) +{ + tw_timer_wheel_16t_2w_512sl_t *tw; + vec_foreach (tw, tm->timer_wheels) + { + tw_timer_wheel_init_16t_2w_512sl (tw, tcp_expired_timers_dispatch, + 100e-3 /* timer period 100ms */ , ~0); + tw->last_run_time = vlib_time_now (tm->vlib_main); + } +} + +clib_error_t * +tcp_init (vlib_main_t * vm) +{ + ip_main_t *im = &ip_main; + ip_protocol_info_t *pi; + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_thread_main_t *vtm = vlib_get_thread_main (); + clib_error_t *error = 0; + u32 num_threads; + + tm->vlib_main = vm; + tm->vnet_main = vnet_get_main (); + + if ((error = vlib_call_init_function (vm, ip_main_init))) + return error; + if ((error = vlib_call_init_function (vm, ip4_lookup_init))) + return error; + if ((error = vlib_call_init_function (vm, ip6_lookup_init))) + return error; + + /* + * Registrations + */ + + /* Register with IP */ + pi = ip_get_protocol_info (im, IP_PROTOCOL_TCP); + if (pi == 0) + return clib_error_return (0, "TCP protocol info AWOL"); + pi->format_header = format_tcp_header; + pi->unformat_pg_edit = unformat_pg_tcp_header; + + ip4_register_protocol (IP_PROTOCOL_TCP, tcp4_input_node.index); + + /* Register as transport with URI */ + session_register_transport (SESSION_TYPE_IP4_TCP, &tcp4_proto); + session_register_transport (SESSION_TYPE_IP6_TCP, &tcp6_proto); + + /* + * Initialize data structures + */ + + num_threads = 1 /* main thread */ + vtm->n_threads; + vec_validate (tm->connections, num_threads - 1); + + /* Initialize per worker thread tx buffers (used for control messages) */ + vec_validate (tm->tx_buffers, num_threads - 1); + + /* Initialize timer wheels */ + vec_validate (tm->timer_wheels, num_threads - 1); + tcp_initialize_timer_wheels (tm); + + vec_validate (tm->delack_connections, num_threads - 1); + + /* Initialize clocks per tick for TCP timestamp. Used to compute + * monotonically increasing timestamps. */ + tm->tstamp_ticks_per_clock = vm->clib_time.seconds_per_clock + / TCP_TSTAMP_RESOLUTION; + + clib_bihash_init_24_8 (&tm->local_endpoints_table, "local endpoint table", + 200000 /* $$$$ config parameter nbuckets */ , + (64 << 20) /*$$$ config parameter table size */ ); + + return error; +} + +VLIB_INIT_FUNCTION (tcp_init); + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp.h b/src/vnet/tcp/tcp.h new file mode 100644 index 00000000000..22f00a63273 --- /dev/null +++ b/src/vnet/tcp/tcp.h @@ -0,0 +1,624 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef _vnet_tcp_h_ +#define _vnet_tcp_h_ + +#include <vnet/vnet.h> +#include <vnet/ip/ip.h> +#include <vnet/tcp/tcp_packet.h> +#include <vnet/tcp/tcp_timer.h> +#include <vnet/session/transport.h> +#include <vnet/session/session.h> + +#define TCP_TICK 10e-3 /**< TCP tick period (s) */ +#define THZ 1/TCP_TICK /**< TCP tick frequency */ +#define TCP_TSTAMP_RESOLUTION TCP_TICK /**< Time stamp resolution */ +#define TCP_PAWS_IDLE 24 * 24 * 60 * 60 * THZ /**< 24 days */ +#define TCP_MAX_OPTION_SPACE 40 + +#define TCP_DUPACK_THRESHOLD 3 +#define TCP_DEFAULT_RX_FIFO_SIZE 64 << 10 + +/** TCP FSM state definitions as per RFC793. */ +#define foreach_tcp_fsm_state \ + _(CLOSED, "CLOSED") \ + _(LISTEN, "LISTEN") \ + _(SYN_SENT, "SYN_SENT") \ + _(SYN_RCVD, "SYN_RCVD") \ + _(ESTABLISHED, "ESTABLISHED") \ + _(CLOSE_WAIT, "CLOSE_WAIT") \ + _(FIN_WAIT_1, "FIN_WAIT_1") \ + _(LAST_ACK, "LAST_ACK") \ + _(CLOSING, "CLOSING") \ + _(FIN_WAIT_2, "FIN_WAIT_2") \ + _(TIME_WAIT, "TIME_WAIT") + +typedef enum _tcp_state +{ +#define _(sym, str) TCP_STATE_##sym, + foreach_tcp_fsm_state +#undef _ + TCP_N_STATES +} tcp_state_t; + +format_function_t format_tcp_state; + +/** TCP timers */ +#define foreach_tcp_timer \ + _(RETRANSMIT, "RETRANSMIT") \ + _(DELACK, "DELAYED ACK") \ + _(PERSIST, "PERSIST") \ + _(KEEP, "KEEP") \ + _(2MSL, "2MSL") \ + _(RETRANSMIT_SYN, "RETRANSMIT_SYN") \ + _(ESTABLISH, "ESTABLISH") + +typedef enum _tcp_timers +{ +#define _(sym, str) TCP_TIMER_##sym, + foreach_tcp_timer +#undef _ + TCP_N_TIMERS +} tcp_timers_e; + +typedef void (timer_expiration_handler) (u32 index); + +extern timer_expiration_handler tcp_timer_delack_handler; +extern timer_expiration_handler tcp_timer_retransmit_handler; +extern timer_expiration_handler tcp_timer_retransmit_syn_handler; + +#define TCP_TIMER_HANDLE_INVALID ((u32) ~0) + +/* Timer delays as multiples of 100ms */ +#define TCP_TO_TIMER_TICK TCP_TICK*10 /* Period for converting from TCP + * ticks to timer units */ +#define TCP_DELACK_TIME 1 /* 0.1s */ +#define TCP_ESTABLISH_TIME 750 /* 75s */ +#define TCP_2MSL_TIME 300 /* 30s */ + +#define TCP_RTO_MAX 60 * THZ /* Min max RTO (60s) as per RFC6298 */ +#define TCP_RTT_MAX 30 * THZ /* 30s (probably too much) */ +#define TCP_RTO_SYN_RETRIES 3 /* SYN retries without doubling RTO */ +#define TCP_RTO_INIT 1 * THZ /* Initial retransmit timer */ + +void tcp_update_time (f64 now, u32 thread_index); + +/** TCP connection flags */ +#define foreach_tcp_connection_flag \ + _(DELACK, "Delay ACK") \ + _(SNDACK, "Send ACK") \ + _(BURSTACK, "Burst ACK set") \ + _(SENT_RCV_WND0, "Sent 0 receive window") \ + _(RECOVERY, "Recovery on") \ + _(FAST_RECOVERY, "Fast Recovery on") + +typedef enum _tcp_connection_flag_bits +{ +#define _(sym, str) TCP_CONN_##sym##_BIT, + foreach_tcp_connection_flag +#undef _ + TCP_CONN_N_FLAG_BITS +} tcp_connection_flag_bits_e; + +typedef enum _tcp_connection_flag +{ +#define _(sym, str) TCP_CONN_##sym = 1 << TCP_CONN_##sym##_BIT, + foreach_tcp_connection_flag +#undef _ + TCP_CONN_N_FLAGS +} tcp_connection_flags_e; + +/** TCP buffer flags */ +#define foreach_tcp_buf_flag \ + _ (ACK) /**< Sending ACK. */ \ + _ (DUPACK) /**< Sending DUPACK. */ \ + +enum +{ +#define _(f) TCP_BUF_BIT_##f, + foreach_tcp_buf_flag +#undef _ + TCP_N_BUF_BITS, +}; + +enum +{ +#define _(f) TCP_BUF_FLAG_##f = 1 << TCP_BUF_BIT_##f, + foreach_tcp_buf_flag +#undef _ +}; + +#define TCP_MAX_SACK_BLOCKS 5 /**< Max number of SACK blocks stored */ +#define TCP_INVALID_SACK_HOLE_INDEX ((u32)~0) + +typedef struct _sack_scoreboard_hole +{ + u32 next; /**< Index for next entry in linked list */ + u32 prev; /**< Index for previous entry in linked list */ + u32 start; /**< Start sequence number */ + u32 end; /**< End sequence number */ +} sack_scoreboard_hole_t; + +typedef struct _sack_scoreboard +{ + sack_scoreboard_hole_t *holes; /**< Pool of holes */ + u32 head; /**< Index to first entry */ + u32 sacked_bytes; /**< Number of bytes sacked in sb */ +} sack_scoreboard_t; + +typedef enum _tcp_cc_algorithm_type +{ + TCP_CC_NEWRENO, +} tcp_cc_algorithm_type_e; + +typedef struct _tcp_cc_algorithm tcp_cc_algorithm_t; + +typedef enum _tcp_cc_ack_t +{ + TCP_CC_ACK, + TCP_CC_DUPACK, + TCP_CC_PARTIALACK +} tcp_cc_ack_t; + +typedef struct _tcp_connection +{ + transport_connection_t connection; /**< Common transport data. First! */ + + u8 state; /**< TCP state as per tcp_state_t */ + u16 flags; /**< Connection flags (see tcp_conn_flags_e) */ + u32 timers[TCP_N_TIMERS]; /**< Timer handles into timer wheel */ + + /* TODO RFC4898 */ + + /** Send sequence variables RFC793 */ + u32 snd_una; /**< oldest unacknowledged sequence number */ + u32 snd_una_max; /**< newest unacknowledged sequence number + 1*/ + u32 snd_wnd; /**< send window */ + u32 snd_wl1; /**< seq number used for last snd.wnd update */ + u32 snd_wl2; /**< ack number used for last snd.wnd update */ + u32 snd_nxt; /**< next seq number to be sent */ + + /** Receive sequence variables RFC793 */ + u32 rcv_nxt; /**< next sequence number expected */ + u32 rcv_wnd; /**< receive window we expect */ + + u32 rcv_las; /**< rcv_nxt at last ack sent/rcv_wnd update */ + u32 iss; /**< initial sent sequence */ + u32 irs; /**< initial remote sequence */ + + /* Options */ + tcp_options_t opt; /**< TCP connection options parsed */ + u8 rcv_wscale; /**< Window scale to advertise to peer */ + u8 snd_wscale; /**< Window scale to use when sending */ + u32 tsval_recent; /**< Last timestamp received */ + u32 tsval_recent_age; /**< When last updated tstamp_recent*/ + + sack_block_t *snd_sacks; /**< Vector of SACKs to send. XXX Fixed size? */ + sack_scoreboard_t sack_sb; /**< SACK "scoreboard" that tracks holes */ + + u8 rcv_dupacks; /**< Number of DUPACKs received */ + u8 snt_dupacks; /**< Number of DUPACKs sent in a burst */ + + /* Congestion control */ + u32 cwnd; /**< Congestion window */ + u32 ssthresh; /**< Slow-start threshold */ + u32 prev_ssthresh; /**< ssthresh before congestion */ + u32 bytes_acked; /**< Bytes acknowledged by current segment */ + u32 rtx_bytes; /**< Retransmitted bytes */ + u32 tsecr_last_ack; /**< Timestamp echoed to us in last health ACK */ + tcp_cc_algorithm_t *cc_algo; /**< Congestion control algorithm */ + + /* RTT and RTO */ + u32 rto; /**< Retransmission timeout */ + u32 rto_boff; /**< Index for RTO backoff */ + u32 srtt; /**< Smoothed RTT */ + u32 rttvar; /**< Smoothed mean RTT difference. Approximates variance */ + u32 rtt_ts; /**< Timestamp for tracked ACK */ + u32 rtt_seq; /**< Sequence number for tracked ACK */ + + u16 snd_mss; /**< Send MSS */ +} tcp_connection_t; + +struct _tcp_cc_algorithm +{ + void (*rcv_ack) (tcp_connection_t * tc); + void (*rcv_cong_ack) (tcp_connection_t * tc, tcp_cc_ack_t ack); + void (*congestion) (tcp_connection_t * tc); + void (*recovered) (tcp_connection_t * tc); + void (*init) (tcp_connection_t * tc); +}; + +#define tcp_fastrecovery_on(tc) (tc)->flags |= TCP_CONN_FAST_RECOVERY +#define tcp_fastrecovery_off(tc) (tc)->flags &= ~TCP_CONN_FAST_RECOVERY +#define tcp_in_fastrecovery(tc) ((tc)->flags & TCP_CONN_FAST_RECOVERY) +#define tcp_in_recovery(tc) ((tc)->flags & (TCP_CONN_FAST_RECOVERY | TCP_CONN_RECOVERY)) +#define tcp_recovery_off(tc) ((tc)->flags &= ~(TCP_CONN_FAST_RECOVERY | TCP_CONN_RECOVERY)) +#define tcp_in_slowstart(tc) (tc->cwnd < tc->ssthresh) + +typedef enum +{ + TCP_IP4, + TCP_IP6, + TCP_N_AF, +} tcp_af_t; + +typedef enum _tcp_error +{ +#define tcp_error(n,s) TCP_ERROR_##n, +#include <vnet/tcp/tcp_error.def> +#undef tcp_error + TCP_N_ERROR, +} tcp_error_t; + +typedef struct _tcp_lookup_dispatch +{ + u8 next, error; +} tcp_lookup_dispatch_t; + +typedef struct _tcp_main +{ + /* Per-worker thread tcp connection pools */ + tcp_connection_t **connections; + + /* Pool of listeners. */ + tcp_connection_t *listener_pool; + + /** Dispatch table by state and flags */ + tcp_lookup_dispatch_t dispatch_table[TCP_N_STATES][64]; + + u8 log2_tstamp_clocks_per_tick; + f64 tstamp_ticks_per_clock; + + /** per-worker tx buffer free lists */ + u32 **tx_buffers; + + /* Per worker-thread timer wheel for connections timers */ + tw_timer_wheel_16t_2w_512sl_t *timer_wheels; + + /* Convenience per worker-thread vector of connections to DELACK */ + u32 **delack_connections; + + /* Pool of half-open connections on which we've sent a SYN */ + tcp_connection_t *half_open_connections; + + /* Pool of local TCP endpoints */ + transport_endpoint_t *local_endpoints; + + /* Local endpoints lookup table */ + transport_endpoint_table_t local_endpoints_table; + + /* Congestion control algorithms registered */ + tcp_cc_algorithm_t *cc_algos; + + /* convenience */ + vlib_main_t *vlib_main; + vnet_main_t *vnet_main; + ip4_main_t *ip4_main; + ip6_main_t *ip6_main; +} tcp_main_t; + +extern tcp_main_t tcp_main; +extern vlib_node_registration_t tcp4_input_node; +extern vlib_node_registration_t tcp6_input_node; +extern vlib_node_registration_t tcp4_output_node; +extern vlib_node_registration_t tcp6_output_node; + +always_inline tcp_main_t * +vnet_get_tcp_main () +{ + return &tcp_main; +} + +always_inline tcp_connection_t * +tcp_connection_get (u32 conn_index, u32 thread_index) +{ + return pool_elt_at_index (tcp_main.connections[thread_index], conn_index); +} + +always_inline tcp_connection_t * +tcp_connection_get_if_valid (u32 conn_index, u32 thread_index) +{ + if (tcp_main.connections[thread_index] == 0) + return 0; + if (pool_is_free_index (tcp_main.connections[thread_index], conn_index)) + return 0; + return pool_elt_at_index (tcp_main.connections[thread_index], conn_index); +} + +void tcp_connection_close (tcp_connection_t * tc); +void tcp_connection_cleanup (tcp_connection_t * tc); +void tcp_connection_del (tcp_connection_t * tc); + +always_inline tcp_connection_t * +tcp_listener_get (u32 tli) +{ + return pool_elt_at_index (tcp_main.listener_pool, tli); +} + +always_inline tcp_connection_t * +tcp_half_open_connection_get (u32 conn_index) +{ + return pool_elt_at_index (tcp_main.half_open_connections, conn_index); +} + +void tcp_make_ack (tcp_connection_t * ts, vlib_buffer_t * b); +void tcp_make_finack (tcp_connection_t * tc, vlib_buffer_t * b); +void tcp_make_synack (tcp_connection_t * ts, vlib_buffer_t * b); +void tcp_send_reset (vlib_buffer_t * pkt, u8 is_ip4); +void tcp_send_syn (tcp_connection_t * tc); +void tcp_send_fin (tcp_connection_t * tc); +void tcp_set_snd_mss (tcp_connection_t * tc); + +always_inline u32 +tcp_end_seq (tcp_header_t * th, u32 len) +{ + return th->seq_number + tcp_is_syn (th) + tcp_is_fin (th) + len; +} + +/* Modulo arithmetic for TCP sequence numbers */ +#define seq_lt(_s1, _s2) ((i32)((_s1)-(_s2)) < 0) +#define seq_leq(_s1, _s2) ((i32)((_s1)-(_s2)) <= 0) +#define seq_gt(_s1, _s2) ((i32)((_s1)-(_s2)) > 0) +#define seq_geq(_s1, _s2) ((i32)((_s1)-(_s2)) >= 0) + +/* Modulo arithmetic for timestamps */ +#define timestamp_lt(_t1, _t2) ((i32)((_t1)-(_t2)) < 0) +#define timestamp_leq(_t1, _t2) ((i32)((_t1)-(_t2)) <= 0) + +always_inline u32 +tcp_flight_size (const tcp_connection_t * tc) +{ + return tc->snd_una_max - tc->snd_una - tc->sack_sb.sacked_bytes + + tc->rtx_bytes; +} + +/** + * Initial cwnd as per RFC5681 + */ +always_inline u32 +tcp_initial_cwnd (const tcp_connection_t * tc) +{ + if (tc->snd_mss > 2190) + return 2 * tc->snd_mss; + else if (tc->snd_mss > 1095) + return 3 * tc->snd_mss; + else + return 4 * tc->snd_mss; +} + +always_inline u32 +tcp_loss_wnd (const tcp_connection_t * tc) +{ + return tc->snd_mss; +} + +always_inline u32 +tcp_available_wnd (const tcp_connection_t * tc) +{ + return clib_min (tc->cwnd, tc->snd_wnd); +} + +always_inline u32 +tcp_available_snd_space (const tcp_connection_t * tc) +{ + u32 available_wnd = tcp_available_wnd (tc); + u32 flight_size = tcp_flight_size (tc); + + if (available_wnd <= flight_size) + return 0; + + return available_wnd - flight_size; +} + +void tcp_retransmit_first_unacked (tcp_connection_t * tc); + +void tcp_fast_retransmit (tcp_connection_t * tc); + +always_inline u32 +tcp_time_now (void) +{ + return clib_cpu_time_now () * tcp_main.tstamp_ticks_per_clock; +} + +u32 tcp_push_header (transport_connection_t * tconn, vlib_buffer_t * b); + +u32 +tcp_prepare_retransmit_segment (tcp_connection_t * tc, vlib_buffer_t * b, + u32 max_bytes); + +void tcp_connection_timers_init (tcp_connection_t * tc); +void tcp_connection_timers_reset (tcp_connection_t * tc); + +void tcp_connection_init_vars (tcp_connection_t * tc); + +always_inline void +tcp_connection_force_ack (tcp_connection_t * tc, vlib_buffer_t * b) +{ + /* Reset flags, make sure ack is sent */ + tc->flags = TCP_CONN_SNDACK; + vnet_buffer (b)->tcp.flags &= ~TCP_BUF_FLAG_DUPACK; +} + +always_inline void +tcp_timer_set (tcp_connection_t * tc, u8 timer_id, u32 interval) +{ + tc->timers[timer_id] + = tw_timer_start_16t_2w_512sl (&tcp_main.timer_wheels[tc->c_thread_index], + tc->c_c_index, timer_id, interval); +} + +always_inline void +tcp_retransmit_timer_set (tcp_main_t * tm, tcp_connection_t * tc) +{ + /* XXX Switch to faster TW */ + tcp_timer_set (tc, TCP_TIMER_RETRANSMIT, + clib_max (tc->rto * TCP_TO_TIMER_TICK, 1)); +} + +always_inline void +tcp_timer_reset (tcp_connection_t * tc, u8 timer_id) +{ + if (tc->timers[timer_id] == TCP_TIMER_HANDLE_INVALID) + return; + + tw_timer_stop_16t_2w_512sl (&tcp_main.timer_wheels[tc->c_thread_index], + tc->timers[timer_id]); + tc->timers[timer_id] = TCP_TIMER_HANDLE_INVALID; +} + +always_inline void +tcp_timer_update (tcp_connection_t * tc, u8 timer_id, u32 interval) +{ + if (tc->timers[timer_id] != TCP_TIMER_HANDLE_INVALID) + tw_timer_stop_16t_2w_512sl (&tcp_main.timer_wheels[tc->c_thread_index], + tc->timers[timer_id]); + tc->timers[timer_id] = + tw_timer_start_16t_2w_512sl (&tcp_main.timer_wheels[tc->c_thread_index], + tc->c_c_index, timer_id, interval); +} + +always_inline u8 +tcp_timer_is_active (tcp_connection_t * tc, tcp_timers_e timer) +{ + return tc->timers[timer] != TCP_TIMER_HANDLE_INVALID; +} + +void +scoreboard_remove_hole (sack_scoreboard_t * sb, + sack_scoreboard_hole_t * hole); + +always_inline sack_scoreboard_hole_t * +scoreboard_next_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole) +{ + if (hole->next != TCP_INVALID_SACK_HOLE_INDEX) + return pool_elt_at_index (sb->holes, hole->next); + return 0; +} + +always_inline sack_scoreboard_hole_t * +scoreboard_first_hole (sack_scoreboard_t * sb) +{ + if (sb->head != TCP_INVALID_SACK_HOLE_INDEX) + return pool_elt_at_index (sb->holes, sb->head); + return 0; +} + +always_inline void +scoreboard_clear (sack_scoreboard_t * sb) +{ + sack_scoreboard_hole_t *hole = scoreboard_first_hole (sb); + while ((hole = scoreboard_first_hole (sb))) + { + scoreboard_remove_hole (sb, hole); + } +} + +always_inline u32 +scoreboard_hole_bytes (sack_scoreboard_hole_t * hole) +{ + return hole->end - hole->start; +} + +always_inline void +tcp_cc_algo_register (tcp_cc_algorithm_type_e type, + const tcp_cc_algorithm_t * vft) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + vec_validate (tm->cc_algos, type); + + tm->cc_algos[type] = *vft; +} + +always_inline tcp_cc_algorithm_t * +tcp_cc_algo_get (tcp_cc_algorithm_type_e type) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + return &tm->cc_algos[type]; +} + +void tcp_cc_init (tcp_connection_t * tc); + +/** + * Push TCP header to buffer + * + * @param vm - vlib_main + * @param b - buffer to write the header to + * @param sp_net - source port net order + * @param dp_net - destination port net order + * @param seq - sequence number net order + * @param ack - ack number net order + * @param tcp_hdr_opts_len - header and options length in bytes + * @param flags - header flags + * @param wnd - window size + * + * @return - pointer to start of TCP header + */ +always_inline void * +vlib_buffer_push_tcp_net_order (vlib_buffer_t * b, u16 sp, u16 dp, u32 seq, + u32 ack, u8 tcp_hdr_opts_len, u8 flags, + u16 wnd) +{ + tcp_header_t *th; + + th = vlib_buffer_push_uninit (b, tcp_hdr_opts_len); + + th->src_port = sp; + th->dst_port = dp; + th->seq_number = seq; + th->ack_number = ack; + th->data_offset_and_reserved = (tcp_hdr_opts_len >> 2) << 4; + th->flags = flags; + th->window = wnd; + th->checksum = 0; + th->urgent_pointer = 0; + return th; +} + +/** + * Push TCP header to buffer + * + * @param vm - vlib_main + * @param b - buffer to write the header to + * @param sp_net - source port net order + * @param dp_net - destination port net order + * @param seq - sequence number host order + * @param ack - ack number host order + * @param tcp_hdr_opts_len - header and options length in bytes + * @param flags - header flags + * @param wnd - window size + * + * @return - pointer to start of TCP header + */ +always_inline void * +vlib_buffer_push_tcp (vlib_buffer_t * b, u16 sp_net, u16 dp_net, u32 seq, + u32 ack, u8 tcp_hdr_opts_len, u8 flags, u16 wnd) +{ + return vlib_buffer_push_tcp_net_order (b, sp_net, dp_net, + clib_host_to_net_u32 (seq), + clib_host_to_net_u32 (ack), + tcp_hdr_opts_len, flags, + clib_host_to_net_u16 (wnd)); +} + +#endif /* _vnet_tcp_h_ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp_error.def b/src/vnet/tcp/tcp_error.def new file mode 100644 index 00000000000..cff5ec13a70 --- /dev/null +++ b/src/vnet/tcp/tcp_error.def @@ -0,0 +1,35 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +tcp_error (NONE, "no error") +tcp_error (NO_LISTENER, "no listener for dst port") +tcp_error (LOOKUP_DROPS, "lookup drops") +tcp_error (DISPATCH, "Dispatch error") +tcp_error (ENQUEUED, "Packets pushed into rx fifo") +tcp_error (PURE_ACK, "Pure acks") +tcp_error (SYNS_RCVD, "SYNs received") +tcp_error (SYN_ACKS_RCVD, "SYN-ACKs received") +tcp_error (NOT_READY, "Session not ready for packets") +tcp_error (FIFO_FULL, "Packets dropped for lack of rx fifo space") +tcp_error (EVENT_FIFO_FULL, "Events not sent for lack of event fifo space") +tcp_error (API_QUEUE_FULL, "Sessions not created for lack of API queue space") +tcp_error (CREATE_SESSION_FAIL, "Sessions couldn't be allocated") +tcp_error (SEGMENT_INVALID, "Invalid segment") +tcp_error (ACK_INVALID, "Invalid ACK") +tcp_error (ACK_DUP, "Duplicate ACK") +tcp_error (ACK_OLD, "Old ACK") +tcp_error (PKTS_SENT, "Packets sent") +tcp_error (FILTERED_DUPACKS, "Filtered duplicate ACKs") +tcp_error (RST_SENT, "Resets sent")
\ No newline at end of file diff --git a/src/vnet/tcp/tcp_format.c b/src/vnet/tcp/tcp_format.c new file mode 100644 index 00000000000..7136741d212 --- /dev/null +++ b/src/vnet/tcp/tcp_format.c @@ -0,0 +1,136 @@ +/* + * Copyright (c) 2015 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +/* + * tcp/tcp_format.c: tcp formatting + * + * Copyright (c) 2008 Eliot Dresselhaus + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice shall be + * included in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +#include <vnet/ip/ip.h> +#include <vnet/tcp/tcp.h> + +static u8 * +format_tcp_flags (u8 * s, va_list * args) +{ + int flags = va_arg (*args, int); + +#define _(f) if (flags & TCP_FLAG_##f) s = format (s, "%s, ", #f); + foreach_tcp_flag +#undef _ + return s; +} + +/* Format TCP header. */ +u8 * +format_tcp_header (u8 * s, va_list * args) +{ + tcp_header_t *tcp = va_arg (*args, tcp_header_t *); + u32 max_header_bytes = va_arg (*args, u32); + u32 header_bytes; + uword indent; + + /* Nothing to do. */ + if (max_header_bytes < sizeof (tcp[0])) + return format (s, "TCP header truncated"); + + indent = format_get_indent (s); + indent += 2; + header_bytes = tcp_header_bytes (tcp); + + s = format (s, "TCP: %d -> %d", clib_net_to_host_u16 (tcp->src), + clib_net_to_host_u16 (tcp->dst)); + + s = format (s, "\n%Useq. 0x%08x ack 0x%08x", format_white_space, indent, + clib_net_to_host_u32 (tcp->seq_number), + clib_net_to_host_u32 (tcp->ack_number)); + + s = format (s, "\n%Uflags %U, tcp header: %d bytes", format_white_space, + indent, format_tcp_flags, tcp->flags, header_bytes); + + s = format (s, "\n%Uwindow %d, checksum 0x%04x", format_white_space, indent, + clib_net_to_host_u16 (tcp->window), + clib_net_to_host_u16 (tcp->checksum)); + + +#if 0 + /* Format TCP options. */ + { + u8 *o; + u8 *option_start = (void *) (tcp + 1); + u8 *option_end = (void *) tcp + header_bytes; + + for (o = option_start; o < option_end;) + { + u32 length = o[1]; + switch (o[0]) + { + case TCP_OPTION_END: + length = 1; + o = option_end; + break; + + case TCP_OPTION_NOOP: + length = 1; + break; + + } + } + } +#endif + + /* Recurse into next protocol layer. */ + if (max_header_bytes != 0 && header_bytes < max_header_bytes) + { + ip_main_t *im = &ip_main; + tcp_udp_port_info_t *pi; + + pi = ip_get_tcp_udp_port_info (im, tcp->dst); + + if (pi && pi->format_header) + s = format (s, "\n%U%U", format_white_space, indent - 2, + pi->format_header, + /* next protocol header */ (void *) tcp + header_bytes, + max_header_bytes - header_bytes); + } + + return s; +} + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp_input.c b/src/vnet/tcp/tcp_input.c new file mode 100644 index 00000000000..daa0683b48e --- /dev/null +++ b/src/vnet/tcp/tcp_input.c @@ -0,0 +1,2316 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vppinfra/sparse_vec.h> +#include <vnet/tcp/tcp_packet.h> +#include <vnet/tcp/tcp.h> +#include <vnet/session/session.h> +#include <math.h> + +static char *tcp_error_strings[] = { +#define tcp_error(n,s) s, +#include <vnet/tcp/tcp_error.def> +#undef tcp_error +}; + +/* All TCP nodes have the same outgoing arcs */ +#define foreach_tcp_state_next \ + _ (DROP, "error-drop") \ + _ (TCP4_OUTPUT, "tcp4-output") \ + _ (TCP6_OUTPUT, "tcp6-output") + +typedef enum _tcp_established_next +{ +#define _(s,n) TCP_ESTABLISHED_NEXT_##s, + foreach_tcp_state_next +#undef _ + TCP_ESTABLISHED_N_NEXT, +} tcp_established_next_t; + +typedef enum _tcp_rcv_process_next +{ +#define _(s,n) TCP_RCV_PROCESS_NEXT_##s, + foreach_tcp_state_next +#undef _ + TCP_RCV_PROCESS_N_NEXT, +} tcp_rcv_process_next_t; + +typedef enum _tcp_syn_sent_next +{ +#define _(s,n) TCP_SYN_SENT_NEXT_##s, + foreach_tcp_state_next +#undef _ + TCP_SYN_SENT_N_NEXT, +} tcp_syn_sent_next_t; + +typedef enum _tcp_listen_next +{ +#define _(s,n) TCP_LISTEN_NEXT_##s, + foreach_tcp_state_next +#undef _ + TCP_LISTEN_N_NEXT, +} tcp_listen_next_t; + +/* Generic, state independent indices */ +typedef enum _tcp_state_next +{ +#define _(s,n) TCP_NEXT_##s, + foreach_tcp_state_next +#undef _ + TCP_STATE_N_NEXT, +} tcp_state_next_t; + +#define tcp_next_output(is_ip4) (is_ip4 ? TCP_NEXT_TCP4_OUTPUT \ + : TCP_NEXT_TCP6_OUTPUT) + +vlib_node_registration_t tcp4_established_node; +vlib_node_registration_t tcp6_established_node; + +/** + * Validate segment sequence number. As per RFC793: + * + * Segment Receive Test + * Length Window + * ------- ------- ------------------------------------------- + * 0 0 SEG.SEQ = RCV.NXT + * 0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND + * >0 0 not acceptable + * >0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND + * or RCV.NXT =< SEG.SEQ+SEG.LEN-1 < RCV.NXT+RCV.WND + * + * This ultimately consists in checking if segment falls within the window. + * The one important difference compared to RFC793 is that we use rcv_las, + * or the rcv_nxt at last ack sent instead of rcv_nxt since that's the + * peer's reference when computing our receive window. + * + * This accepts only segments within the window. + */ +always_inline u8 +tcp_segment_in_rcv_wnd (tcp_connection_t * tc, u32 seq, u32 end_seq) +{ + return seq_leq (end_seq, tc->rcv_las + tc->rcv_wnd) + && seq_geq (seq, tc->rcv_nxt); +} + +void +tcp_options_parse (tcp_header_t * th, tcp_options_t * to) +{ + const u8 *data; + u8 opt_len, opts_len, kind; + int j; + sack_block_t b; + + opts_len = (tcp_doff (th) << 2) - sizeof (tcp_header_t); + data = (const u8 *) (th + 1); + + /* Zero out all flags but those set in SYN */ + to->flags &= (TCP_OPTS_FLAG_SACK_PERMITTED | TCP_OPTS_FLAG_WSCALE); + + for (; opts_len > 0; opts_len -= opt_len, data += opt_len) + { + kind = data[0]; + + /* Get options length */ + if (kind == TCP_OPTION_EOL) + break; + else if (kind == TCP_OPTION_NOOP) + opt_len = 1; + else + { + /* broken options */ + if (opts_len < 2) + break; + opt_len = data[1]; + + /* weird option length */ + if (opt_len < 2 || opt_len > opts_len) + break; + } + + /* Parse options */ + switch (kind) + { + case TCP_OPTION_MSS: + if ((opt_len == TCP_OPTION_LEN_MSS) && tcp_syn (th)) + { + to->flags |= TCP_OPTS_FLAG_MSS; + to->mss = clib_net_to_host_u16 (*(u16 *) (data + 2)); + } + break; + case TCP_OPTION_WINDOW_SCALE: + if ((opt_len == TCP_OPTION_LEN_WINDOW_SCALE) && tcp_syn (th)) + { + to->flags |= TCP_OPTS_FLAG_WSCALE; + to->wscale = data[2]; + if (to->wscale > TCP_MAX_WND_SCALE) + { + clib_warning ("Illegal window scaling value: %d", + to->wscale); + to->wscale = TCP_MAX_WND_SCALE; + } + } + break; + case TCP_OPTION_TIMESTAMP: + if (opt_len == TCP_OPTION_LEN_TIMESTAMP) + { + to->flags |= TCP_OPTS_FLAG_TSTAMP; + to->tsval = clib_net_to_host_u32 (*(u32 *) (data + 2)); + to->tsecr = clib_net_to_host_u32 (*(u32 *) (data + 6)); + } + break; + case TCP_OPTION_SACK_PERMITTED: + if (opt_len == TCP_OPTION_LEN_SACK_PERMITTED && tcp_syn (th)) + to->flags |= TCP_OPTS_FLAG_SACK_PERMITTED; + break; + case TCP_OPTION_SACK_BLOCK: + /* If SACK permitted was not advertised or a SYN, break */ + if ((to->flags & TCP_OPTS_FLAG_SACK_PERMITTED) == 0 || tcp_syn (th)) + break; + + /* If too short or not correctly formatted, break */ + if (opt_len < 10 || ((opt_len - 2) % TCP_OPTION_LEN_SACK_BLOCK)) + break; + + to->flags |= TCP_OPTS_FLAG_SACK; + to->n_sack_blocks = (opt_len - 2) / TCP_OPTION_LEN_SACK_BLOCK; + vec_reset_length (to->sacks); + for (j = 0; j < to->n_sack_blocks; j++) + { + b.start = clib_net_to_host_u32 (*(u32 *) (data + 2 + 4 * j)); + b.end = clib_net_to_host_u32 (*(u32 *) (data + 6 + 4 * j)); + vec_add1 (to->sacks, b); + } + break; + default: + /* Nothing to see here */ + continue; + } + } +} + +always_inline int +tcp_segment_check_paws (tcp_connection_t * tc) +{ + /* XXX normally test for timestamp should be lt instead of leq, but for + * local testing this is not enough */ + return tcp_opts_tstamp (&tc->opt) && tc->tsval_recent + && timestamp_lt (tc->opt.tsval, tc->tsval_recent); +} + +/** + * Validate incoming segment as per RFC793 p. 69 and RFC1323 p. 19 + * + * It first verifies if segment has a wrapped sequence number (PAWS) and then + * does the processing associated to the first four steps (ignoring security + * and precedence): sequence number, rst bit and syn bit checks. + * + * @return 0 if segments passes validation. + */ +static int +tcp_segment_validate (vlib_main_t * vm, tcp_connection_t * tc0, + vlib_buffer_t * b0, tcp_header_t * th0, u32 * next0) +{ + u8 paws_failed; + + if (PREDICT_FALSE (!tcp_ack (th0) && !tcp_rst (th0) && !tcp_syn (th0))) + return -1; + + tcp_options_parse (th0, &tc0->opt); + + /* RFC1323: Check against wrapped sequence numbers (PAWS). If we have + * timestamp to echo and it's less than tsval_recent, drop segment + * but still send an ACK in order to retain TCP's mechanism for detecting + * and recovering from half-open connections */ + paws_failed = tcp_segment_check_paws (tc0); + if (paws_failed) + { + clib_warning ("paws failed"); + + /* If it just so happens that a segment updates tsval_recent for a + * segment over 24 days old, invalidate tsval_recent. */ + if (timestamp_lt (tc0->tsval_recent_age + TCP_PAWS_IDLE, + tcp_time_now ())) + { + /* Age isn't reset until we get a valid tsval (bsd inspired) */ + tc0->tsval_recent = 0; + } + else + { + /* Drop after ack if not rst */ + if (!tcp_rst (th0)) + { + tcp_make_ack (tc0, b0); + *next0 = tcp_next_output (tc0->c_is_ip4); + return -1; + } + } + } + + /* 1st: check sequence number */ + if (!tcp_segment_in_rcv_wnd (tc0, vnet_buffer (b0)->tcp.seq_number, + vnet_buffer (b0)->tcp.seq_end)) + { + if (!tcp_rst (th0)) + { + /* Send dup ack */ + tcp_make_ack (tc0, b0); + *next0 = tcp_next_output (tc0->c_is_ip4); + } + return -1; + } + + /* 2nd: check the RST bit */ + if (tcp_rst (th0)) + { + /* Notify session that connection has been reset. Switch + * state to closed and await for session to do the cleanup. */ + stream_session_reset_notify (&tc0->connection); + tc0->state = TCP_STATE_CLOSED; + return -1; + } + + /* 3rd: check security and precedence (skip) */ + + /* 4th: check the SYN bit */ + if (tcp_syn (th0)) + { + tcp_send_reset (b0, tc0->c_is_ip4); + return -1; + } + + /* If PAWS passed and segment in window, save timestamp */ + if (!paws_failed) + { + tc0->tsval_recent = tc0->opt.tsval; + tc0->tsval_recent_age = tcp_time_now (); + } + + return 0; +} + +always_inline int +tcp_rcv_ack_is_acceptable (tcp_connection_t * tc0, vlib_buffer_t * tb0) +{ + /* SND.UNA =< SEG.ACK =< SND.NXT */ + return (seq_leq (tc0->snd_una, vnet_buffer (tb0)->tcp.ack_number) + && seq_leq (vnet_buffer (tb0)->tcp.ack_number, tc0->snd_nxt)); +} + +/** + * Compute smoothed RTT as per VJ's '88 SIGCOMM and RFC6298 + * + * Note that although the original article, srtt and rttvar are scaled + * to minimize round-off errors, here we don't. Instead, we rely on + * better precision time measurements. + * + * TODO support us rtt resolution + */ +static void +tcp_estimate_rtt (tcp_connection_t * tc, u32 mrtt) +{ + int err; + + if (tc->srtt != 0) + { + err = mrtt - tc->srtt; + tc->srtt += err >> 3; + + /* XXX Drop in RTT results in RTTVAR increase and bigger RTO. + * The increase should be bound */ + tc->rttvar += (clib_abs (err) - tc->rttvar) >> 2; + } + else + { + /* First measurement. */ + tc->srtt = mrtt; + tc->rttvar = mrtt << 1; + } +} + +/** Update RTT estimate and RTO timer + * + * Measure RTT: We have two sources of RTT measurements: TSOPT and ACK + * timing. Middle boxes are known to fiddle with TCP options so we + * should give higher priority to ACK timing. + * + * return 1 if valid rtt 0 otherwise + */ +static int +tcp_update_rtt (tcp_connection_t * tc, u32 ack) +{ + u32 mrtt = 0; + + /* Karn's rule, part 1. Don't use retransmitted segments to estimate + * RTT because they're ambiguous. */ + if (tc->rtt_seq && seq_gt (ack, tc->rtt_seq) && !tc->rto_boff) + { + mrtt = tcp_time_now () - tc->rtt_ts; + tc->rtt_seq = 0; + } + + /* As per RFC7323 TSecr can be used for RTTM only if the segment advances + * snd_una, i.e., the left side of the send window: + * seq_lt (tc->snd_una, ack). Note: last condition could be dropped, we don't + * try to update rtt for dupacks */ + else if (tcp_opts_tstamp (&tc->opt) && tc->opt.tsecr && tc->bytes_acked) + { + mrtt = tcp_time_now () - tc->opt.tsecr; + } + + /* Ignore dubious measurements */ + if (mrtt == 0 || mrtt > TCP_RTT_MAX) + return 0; + + tcp_estimate_rtt (tc, mrtt); + + tc->rto = clib_min (tc->srtt + (tc->rttvar << 2), TCP_RTO_MAX); + + return 1; +} + +/** + * Dequeue bytes that have been acked and while at it update RTT estimates. + */ +static void +tcp_dequeue_acked (tcp_connection_t * tc, u32 ack) +{ + /* Dequeue the newly ACKed bytes */ + stream_session_dequeue_drop (&tc->connection, tc->bytes_acked); + + /* Update rtt and rto */ + if (tcp_update_rtt (tc, ack)) + { + /* Good ACK received and valid RTT, make sure retransmit backoff is 0 */ + tc->rto_boff = 0; + } +} + +/** Check if dupack as per RFC5681 Sec. 2 */ +always_inline u8 +tcp_ack_is_dupack (tcp_connection_t * tc, vlib_buffer_t * b, u32 new_snd_wnd) +{ + return ((vnet_buffer (b)->tcp.ack_number == tc->snd_una) + && seq_gt (tc->snd_una_max, tc->snd_una) + && (vnet_buffer (b)->tcp.seq_end == vnet_buffer (b)->tcp.seq_number) + && (new_snd_wnd == tc->snd_wnd)); +} + +void +scoreboard_remove_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole) +{ + sack_scoreboard_hole_t *next, *prev; + + if (hole->next != TCP_INVALID_SACK_HOLE_INDEX) + { + next = pool_elt_at_index (sb->holes, hole->next); + next->prev = hole->prev; + } + + if (hole->prev != TCP_INVALID_SACK_HOLE_INDEX) + { + prev = pool_elt_at_index (sb->holes, hole->prev); + prev->next = hole->next; + } + else + { + sb->head = hole->next; + } + + pool_put (sb->holes, hole); +} + +sack_scoreboard_hole_t * +scoreboard_insert_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * prev, + u32 start, u32 end) +{ + sack_scoreboard_hole_t *hole, *next; + u32 hole_index; + + pool_get (sb->holes, hole); + memset (hole, 0, sizeof (*hole)); + + hole->start = start; + hole->end = end; + hole_index = hole - sb->holes; + + if (prev) + { + hole->prev = prev - sb->holes; + hole->next = prev->next; + + if ((next = scoreboard_next_hole (sb, hole))) + next->prev = hole_index; + + prev->next = hole_index; + } + else + { + sb->head = hole_index; + hole->prev = TCP_INVALID_SACK_HOLE_INDEX; + hole->next = TCP_INVALID_SACK_HOLE_INDEX; + } + + return hole; +} + +static void +tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) +{ + sack_scoreboard_t *sb = &tc->sack_sb; + sack_block_t *blk, tmp; + sack_scoreboard_hole_t *hole, *next_hole; + u32 blk_index = 0; + int i, j; + + if (!tcp_opts_sack (tc) && sb->head == TCP_INVALID_SACK_HOLE_INDEX) + return; + + /* Remove invalid blocks */ + vec_foreach (blk, tc->opt.sacks) + { + if (seq_lt (blk->start, blk->end) + && seq_gt (blk->start, tc->snd_una) + && seq_gt (blk->start, ack) && seq_lt (blk->end, tc->snd_nxt)) + continue; + + vec_del1 (tc->opt.sacks, blk - tc->opt.sacks); + } + + /* Add block for cumulative ack */ + if (seq_gt (ack, tc->snd_una)) + { + tmp.start = tc->snd_una; + tmp.end = ack; + vec_add1 (tc->opt.sacks, tmp); + } + + if (vec_len (tc->opt.sacks) == 0) + return; + + /* Make sure blocks are ordered */ + for (i = 0; i < vec_len (tc->opt.sacks); i++) + for (j = i; j < vec_len (tc->opt.sacks); j++) + if (seq_lt (tc->opt.sacks[j].start, tc->opt.sacks[i].start)) + { + tmp = tc->opt.sacks[i]; + tc->opt.sacks[i] = tc->opt.sacks[j]; + tc->opt.sacks[j] = tmp; + } + + /* If no holes, insert the first that covers all outstanding bytes */ + if (sb->head == TCP_INVALID_SACK_HOLE_INDEX) + { + scoreboard_insert_hole (sb, 0, tc->snd_una, tc->snd_una_max); + } + + /* Walk the holes with the SACK blocks */ + hole = pool_elt_at_index (sb->holes, sb->head); + while (hole && blk_index < vec_len (tc->opt.sacks)) + { + blk = &tc->opt.sacks[blk_index]; + + if (seq_leq (blk->start, hole->start)) + { + /* Block covers hole. Remove hole */ + if (seq_geq (blk->end, hole->end)) + { + next_hole = scoreboard_next_hole (sb, hole); + + /* Byte accounting */ + if (seq_lt (hole->end, ack)) + { + /* Bytes lost because snd wnd left edge advances */ + if (seq_lt (next_hole->start, ack)) + sb->sacked_bytes -= next_hole->start - hole->end; + else + sb->sacked_bytes -= ack - hole->end; + } + else + { + sb->sacked_bytes += scoreboard_hole_bytes (hole); + } + + scoreboard_remove_hole (sb, hole); + hole = next_hole; + } + /* Partial overlap */ + else + { + sb->sacked_bytes += blk->end - hole->start; + hole->start = blk->end; + blk_index++; + } + } + else + { + /* Hole must be split */ + if (seq_leq (blk->end, hole->end)) + { + sb->sacked_bytes += blk->end - blk->start; + scoreboard_insert_hole (sb, hole, blk->end, hole->end); + hole->end = blk->start - 1; + blk_index++; + } + else + { + sb->sacked_bytes += hole->end - blk->start + 1; + hole->end = blk->start - 1; + hole = scoreboard_next_hole (sb, hole); + } + } + } +} + +/** Update snd_wnd + * + * If (SND.WL1 < SEG.SEQ or (SND.WL1 = SEG.SEQ and SND.WL2 =< SEG.ACK)), set + * SND.WND <- SEG.WND, set SND.WL1 <- SEG.SEQ, and set SND.WL2 <- SEG.ACK */ +static void +tcp_update_snd_wnd (tcp_connection_t * tc, u32 seq, u32 ack, u32 snd_wnd) +{ + if (tc->snd_wl1 < seq || (tc->snd_wl1 == seq && tc->snd_wl2 <= ack)) + { + tc->snd_wnd = snd_wnd; + tc->snd_wl1 = seq; + tc->snd_wl2 = ack; + } +} + +static void +tcp_cc_congestion (tcp_connection_t * tc) +{ + tc->cc_algo->congestion (tc); +} + +static void +tcp_cc_recover (tcp_connection_t * tc) +{ + if (tcp_in_fastrecovery (tc)) + { + tc->cc_algo->recovered (tc); + tcp_recovery_off (tc); + } + else if (tcp_in_recovery (tc)) + { + tcp_recovery_off (tc); + tc->cwnd = tcp_loss_wnd (tc); + } +} + +static void +tcp_cc_rcv_ack (tcp_connection_t * tc) +{ + u8 partial_ack; + + if (tcp_in_recovery (tc)) + { + partial_ack = seq_lt (tc->snd_una, tc->snd_una_max); + if (!partial_ack) + { + /* Clear retransmitted bytes. */ + tc->rtx_bytes = 0; + tcp_cc_recover (tc); + } + else + { + /* Clear retransmitted bytes. XXX should we clear all? */ + tc->rtx_bytes = 0; + tc->cc_algo->rcv_cong_ack (tc, TCP_CC_PARTIALACK); + + /* Retransmit first unacked segment */ + tcp_retransmit_first_unacked (tc); + } + } + else + { + tc->cc_algo->rcv_ack (tc); + } + + tc->rcv_dupacks = 0; + tc->tsecr_last_ack = tc->opt.tsecr; +} + +static void +tcp_cc_rcv_dupack (tcp_connection_t * tc, u32 ack) +{ + ASSERT (tc->snd_una == ack); + + tc->rcv_dupacks++; + if (tc->rcv_dupacks == TCP_DUPACK_THRESHOLD) + { + /* RFC6582 NewReno heuristic to avoid multiple fast retransmits */ + if (tc->opt.tsecr != tc->tsecr_last_ack) + { + tc->rcv_dupacks = 0; + return; + } + + tcp_fastrecovery_on (tc); + + /* Handle congestion and dupack */ + tcp_cc_congestion (tc); + tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK); + + tcp_fast_retransmit (tc); + + /* Post retransmit update cwnd to ssthresh and account for the + * three segments that have left the network and should've been + * buffered at the receiver */ + tc->cwnd = tc->ssthresh + TCP_DUPACK_THRESHOLD * tc->snd_mss; + } + else if (tc->rcv_dupacks > TCP_DUPACK_THRESHOLD) + { + ASSERT (tcp_in_fastrecovery (tc)); + + tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK); + } +} + +void +tcp_cc_init (tcp_connection_t * tc) +{ + tc->cc_algo = tcp_cc_algo_get (TCP_CC_NEWRENO); + tc->cc_algo->init (tc); +} + +static int +tcp_rcv_ack (tcp_connection_t * tc, vlib_buffer_t * b, + tcp_header_t * th, u32 * next, u32 * error) +{ + u32 new_snd_wnd; + + /* If the ACK acks something not yet sent (SEG.ACK > SND.NXT) then send an + * ACK, drop the segment, and return */ + if (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_nxt)) + { + tcp_make_ack (tc, b); + *next = tcp_next_output (tc->c_is_ip4); + *error = TCP_ERROR_ACK_INVALID; + return -1; + } + + /* If old ACK, discard */ + if (seq_lt (vnet_buffer (b)->tcp.ack_number, tc->snd_una)) + { + *error = TCP_ERROR_ACK_OLD; + return -1; + } + + if (tcp_opts_sack_permitted (&tc->opt)) + tcp_rcv_sacks (tc, vnet_buffer (b)->tcp.ack_number); + + new_snd_wnd = clib_net_to_host_u32 (th->window) << tc->snd_wscale; + + if (tcp_ack_is_dupack (tc, b, new_snd_wnd)) + { + tcp_cc_rcv_dupack (tc, vnet_buffer (b)->tcp.ack_number); + *error = TCP_ERROR_ACK_DUP; + return -1; + } + + /* Valid ACK */ + tc->bytes_acked = vnet_buffer (b)->tcp.ack_number - tc->snd_una; + tc->snd_una = vnet_buffer (b)->tcp.ack_number; + + /* Dequeue ACKed packet and update RTT */ + tcp_dequeue_acked (tc, vnet_buffer (b)->tcp.ack_number); + + tcp_update_snd_wnd (tc, vnet_buffer (b)->tcp.seq_number, + vnet_buffer (b)->tcp.ack_number, new_snd_wnd); + + /* Updates congestion control (slow start/congestion avoidance) */ + tcp_cc_rcv_ack (tc); + + /* If everything has been acked, stop retransmit timer + * otherwise update */ + if (tc->snd_una == tc->snd_una_max) + tcp_timer_reset (tc, TCP_TIMER_RETRANSMIT); + else + tcp_timer_update (tc, TCP_TIMER_RETRANSMIT, tc->rto); + + return 0; +} + +/** + * Build SACK list as per RFC2018. + * + * Makes sure the first block contains the segment that generated the current + * ACK and the following ones are the ones most recently reported in SACK + * blocks. + * + * @param tc TCP connection for which the SACK list is updated + * @param start Start sequence number of the newest SACK block + * @param end End sequence of the newest SACK block + */ +static void +tcp_update_sack_list (tcp_connection_t * tc, u32 start, u32 end) +{ + sack_block_t *new_list = 0, block; + u32 n_elts; + int i; + u8 new_head = 0; + + /* If the first segment is ooo add it to the list. Last write might've moved + * rcv_nxt over the first segment. */ + if (seq_lt (tc->rcv_nxt, start)) + { + block.start = start; + block.end = end; + vec_add1 (new_list, block); + new_head = 1; + } + + /* Find the blocks still worth keeping. */ + for (i = 0; i < vec_len (tc->snd_sacks); i++) + { + /* Discard if: + * 1) rcv_nxt advanced beyond current block OR + * 2) Segment overlapped by the first segment, i.e., it has been merged + * into it.*/ + if (seq_leq (tc->snd_sacks[i].start, tc->rcv_nxt) + || seq_leq (tc->snd_sacks[i].start, end)) + continue; + + /* Save subsequent segments to new SACK list. */ + n_elts = clib_min (vec_len (tc->snd_sacks) - i, + TCP_MAX_SACK_BLOCKS - new_head); + vec_insert_elts (new_list, &tc->snd_sacks[i], n_elts, new_head); + break; + } + + /* Replace old vector with new one */ + vec_free (tc->snd_sacks); + tc->snd_sacks = new_list; +} + +/** Enqueue data for delivery to application */ +always_inline u32 +tcp_session_enqueue_data (tcp_connection_t * tc, vlib_buffer_t * b, + u16 data_len) +{ + int written; + + /* Pure ACK. Update rcv_nxt and be done. */ + if (PREDICT_FALSE (data_len == 0)) + { + tc->rcv_nxt = vnet_buffer (b)->tcp.seq_end; + return TCP_ERROR_PURE_ACK; + } + + written = stream_session_enqueue_data (&tc->connection, + vlib_buffer_get_current (b), + data_len, 1 /* queue event */ ); + + /* Update rcv_nxt */ + if (PREDICT_TRUE (written == data_len)) + { + tc->rcv_nxt = vnet_buffer (b)->tcp.seq_end; + } + /* If more data written than expected, account for out-of-order bytes. */ + else if (written > data_len) + { + tc->rcv_nxt = vnet_buffer (b)->tcp.seq_end + written - data_len; + + /* Send ACK confirming the update */ + tc->flags |= TCP_CONN_SNDACK; + + /* Update SACK list if need be */ + if (tcp_opts_sack_permitted (&tc->opt)) + { + /* Remove SACK blocks that have been delivered */ + tcp_update_sack_list (tc, tc->rcv_nxt, tc->rcv_nxt); + } + } + else + { + ASSERT (0); + return TCP_ERROR_FIFO_FULL; + } + + return TCP_ERROR_ENQUEUED; +} + +/** Enqueue out-of-order data */ +always_inline u32 +tcp_session_enqueue_ooo (tcp_connection_t * tc, vlib_buffer_t * b, + u16 data_len) +{ + stream_session_t *s0; + u32 offset, seq; + + s0 = stream_session_get (tc->c_s_index, tc->c_thread_index); + seq = vnet_buffer (b)->tcp.seq_number; + offset = seq - tc->rcv_nxt; + + if (svm_fifo_enqueue_with_offset (s0->server_rx_fifo, s0->pid, offset, + data_len, vlib_buffer_get_current (b))) + return TCP_ERROR_FIFO_FULL; + + /* Update SACK list if in use */ + if (tcp_opts_sack_permitted (&tc->opt)) + { + ooo_segment_t *newest; + u32 start, end; + + /* Get the newest segment from the fifo */ + newest = svm_fifo_newest_ooo_segment (s0->server_rx_fifo); + start = tc->rcv_nxt + ooo_segment_offset (s0->server_rx_fifo, newest); + end = tc->rcv_nxt + ooo_segment_end_offset (s0->server_rx_fifo, newest); + + tcp_update_sack_list (tc, start, end); + } + + return TCP_ERROR_ENQUEUED; +} + +/** + * Check if ACK could be delayed. DELACK timer is set only after frame is + * processed so this can return true for a full bursts of packets. + */ +always_inline int +tcp_can_delack (tcp_connection_t * tc) +{ + /* If there's no DELACK timer set and the last window sent wasn't 0 we + * can safely delay. */ + if (!tcp_timer_is_active (tc, TCP_TIMER_DELACK) + && (tc->flags & TCP_CONN_SENT_RCV_WND0) == 0 + && (tc->flags & TCP_CONN_SNDACK) == 0) + return 1; + + return 0; +} + +static int +tcp_segment_rcv (tcp_main_t * tm, tcp_connection_t * tc, vlib_buffer_t * b, + u16 n_data_bytes, u32 * next0) +{ + u32 error = 0; + + /* Handle out-of-order data */ + if (PREDICT_FALSE (vnet_buffer (b)->tcp.seq_number != tc->rcv_nxt)) + { + error = tcp_session_enqueue_ooo (tc, b, n_data_bytes); + + /* Don't send more than 3 dupacks per burst + * XXX decide if this is good */ + if (tc->snt_dupacks < 3) + { + /* RFC2581: Send DUPACK for fast retransmit */ + tcp_make_ack (tc, b); + *next0 = tcp_next_output (tc->c_is_ip4); + + /* Mark as DUPACK. We may filter these in output if + * the burst fills the holes. */ + vnet_buffer (b)->tcp.flags = TCP_BUF_FLAG_DUPACK; + + tc->snt_dupacks++; + } + + goto done; + } + + /* In order data, enqueue. Fifo figures out by itself if any out-of-order + * segments can be enqueued after fifo tail offset changes. */ + error = tcp_session_enqueue_data (tc, b, n_data_bytes); + + /* Check if ACK can be delayed */ + if (tcp_can_delack (tc)) + { + /* Nothing to do for pure ACKs */ + if (n_data_bytes == 0) + goto done; + + /* If connection has not been previously marked for delay ack + * add it to the list and flag it */ + if (!tc->flags & TCP_CONN_DELACK) + { + vec_add1 (tm->delack_connections[tc->c_thread_index], + tc->c_c_index); + tc->flags |= TCP_CONN_DELACK; + } + } + else + { + /* Check if a packet has already been enqueued to output for burst. + * If yes, then drop this one, otherwise, let it pass through to + * output */ + if ((tc->flags & TCP_CONN_BURSTACK) == 0) + { + *next0 = tcp_next_output (tc->c_is_ip4); + tcp_make_ack (tc, b); + error = TCP_ERROR_ENQUEUED; + + /* TODO: maybe add counter to ensure N acks will be sent/burst */ + tc->flags |= TCP_CONN_BURSTACK; + } + } + +done: + return error; +} + +void +delack_timers_init (tcp_main_t * tm, u32 thread_index) +{ + tcp_connection_t *tc; + u32 i, *conns; + tw_timer_wheel_16t_2w_512sl_t *tw; + + tw = &tm->timer_wheels[thread_index]; + conns = tm->delack_connections[thread_index]; + for (i = 0; i < vec_len (conns); i++) + { + tc = pool_elt_at_index (tm->connections[thread_index], conns[i]); + ASSERT (0 != tc); + + tc->timers[TCP_TIMER_DELACK] + = tw_timer_start_16t_2w_512sl (tw, conns[i], + TCP_TIMER_DELACK, TCP_DELACK_TIME); + } + vec_reset_length (tm->delack_connections[thread_index]); +} + +always_inline uword +tcp46_established_inline (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame, int is_ip4) +{ + u32 n_left_from, next_index, *from, *to_next; + u32 my_thread_index = vm->cpu_index, errors = 0; + tcp_main_t *tm = vnet_get_tcp_main (); + + from = vlib_frame_vector_args (from_frame); + n_left_from = from_frame->n_vectors; + + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + tcp_header_t *th0 = 0; + tcp_connection_t *tc0; + ip4_header_t *ip40; + ip6_header_t *ip60; + u32 n_advance_bytes0, n_data_bytes0; + u32 next0 = TCP_ESTABLISHED_NEXT_DROP, error0 = TCP_ERROR_ENQUEUED; + + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index, + my_thread_index); + + /* Checksum computed by ipx_local no need to compute again */ + + if (is_ip4) + { + ip40 = vlib_buffer_get_current (b0); + th0 = ip4_next_header (ip40); + n_advance_bytes0 = (ip4_header_bytes (ip40) + + tcp_header_bytes (th0)); + n_data_bytes0 = clib_net_to_host_u16 (ip40->length) + - n_advance_bytes0; + } + else + { + ip60 = vlib_buffer_get_current (b0); + th0 = ip6_next_header (ip60); + n_advance_bytes0 = tcp_header_bytes (th0); + n_data_bytes0 = clib_net_to_host_u16 (ip60->payload_length) + - n_advance_bytes0; + n_advance_bytes0 += sizeof (ip60[0]); + } + + /* SYNs, FINs and data consume sequence numbers */ + vnet_buffer (b0)->tcp.seq_end = vnet_buffer (b0)->tcp.seq_number + + tcp_is_syn (th0) + tcp_is_fin (th0) + n_data_bytes0; + + /* TODO header prediction fast path */ + + /* 1-4: check SEQ, RST, SYN */ + if (PREDICT_FALSE (tcp_segment_validate (vm, tc0, b0, th0, &next0))) + { + error0 = TCP_ERROR_SEGMENT_INVALID; + goto drop; + } + + /* 5: check the ACK field */ + if (tcp_rcv_ack (tc0, b0, th0, &next0, &error0)) + { + goto drop; + } + + /* 6: check the URG bit TODO */ + + /* 7: process the segment text */ + vlib_buffer_advance (b0, n_advance_bytes0); + error0 = tcp_segment_rcv (tm, tc0, b0, n_data_bytes0, &next0); + + /* 8: check the FIN bit */ + if (tcp_fin (th0)) + { + /* Send ACK and enter CLOSE-WAIT */ + tcp_make_ack (tc0, b0); + tcp_connection_force_ack (tc0, b0); + next0 = tcp_next_output (tc0->c_is_ip4); + tc0->state = TCP_STATE_CLOSE_WAIT; + stream_session_disconnect_notify (&tc0->connection); + } + + drop: + b0->error = node->errors[error0]; + if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) + { + + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + errors = session_manager_flush_enqueue_events (my_thread_index); + if (errors) + { + if (is_ip4) + vlib_node_increment_counter (vm, tcp4_established_node.index, + TCP_ERROR_EVENT_FIFO_FULL, errors); + else + vlib_node_increment_counter (vm, tcp6_established_node.index, + TCP_ERROR_EVENT_FIFO_FULL, errors); + } + + delack_timers_init (tm, my_thread_index); + + return from_frame->n_vectors; +} + +static uword +tcp4_established (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_established_inline (vm, node, from_frame, 1 /* is_ip4 */ ); +} + +static uword +tcp6_established (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_established_inline (vm, node, from_frame, 0 /* is_ip4 */ ); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp4_established_node) = +{ + .function = tcp4_established, + .name = "tcp4-established", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR,.error_strings = tcp_error_strings, + .n_next_nodes = TCP_ESTABLISHED_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp4_established_node, tcp4_established); + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp6_established_node) = +{ + .function = tcp6_established, + .name = "tcp6-established", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_ESTABLISHED_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + }, +}; +/* *INDENT-ON* */ + + +VLIB_NODE_FUNCTION_MULTIARCH (tcp6_established_node, tcp6_established); + +vlib_node_registration_t tcp4_syn_sent_node; +vlib_node_registration_t tcp6_syn_sent_node; + +always_inline uword +tcp46_syn_sent_inline (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame, int is_ip4) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + u32 n_left_from, next_index, *from, *to_next; + u32 my_thread_index = vm->cpu_index, errors = 0; + u8 sst = is_ip4 ? SESSION_TYPE_IP4_TCP : SESSION_TYPE_IP6_TCP; + + from = vlib_frame_vector_args (from_frame); + n_left_from = from_frame->n_vectors; + + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0, ack0, seq0; + vlib_buffer_t *b0; + tcp_header_t *tcp0 = 0; + tcp_connection_t *tc0; + ip4_header_t *ip40; + ip6_header_t *ip60; + u32 n_advance_bytes0, n_data_bytes0; + tcp_connection_t *new_tc0; + u32 next0 = TCP_SYN_SENT_NEXT_DROP, error0 = TCP_ERROR_ENQUEUED; + + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + tc0 = + tcp_half_open_connection_get (vnet_buffer (b0)-> + tcp.connection_index); + + ack0 = vnet_buffer (b0)->tcp.ack_number; + seq0 = vnet_buffer (b0)->tcp.seq_number; + + /* Checksum computed by ipx_local no need to compute again */ + + if (is_ip4) + { + ip40 = vlib_buffer_get_current (b0); + tcp0 = ip4_next_header (ip40); + n_advance_bytes0 = (ip4_header_bytes (ip40) + + tcp_header_bytes (tcp0)); + n_data_bytes0 = clib_net_to_host_u16 (ip40->length) + - n_advance_bytes0; + } + else + { + ip60 = vlib_buffer_get_current (b0); + tcp0 = ip6_next_header (ip60); + n_advance_bytes0 = tcp_header_bytes (tcp0); + n_data_bytes0 = clib_net_to_host_u16 (ip60->payload_length) + - n_advance_bytes0; + n_advance_bytes0 += sizeof (ip60[0]); + } + + if (PREDICT_FALSE + (!tcp_ack (tcp0) && !tcp_rst (tcp0) && !tcp_syn (tcp0))) + goto drop; + + /* SYNs, FINs and data consume sequence numbers */ + vnet_buffer (b0)->tcp.seq_end = seq0 + tcp_is_syn (tcp0) + + tcp_is_fin (tcp0) + n_data_bytes0; + + /* + * 1. check the ACK bit + */ + + /* + * If the ACK bit is set + * If SEG.ACK =< ISS, or SEG.ACK > SND.NXT, send a reset (unless + * the RST bit is set, if so drop the segment and return) + * <SEQ=SEG.ACK><CTL=RST> + * and discard the segment. Return. + * If SND.UNA =< SEG.ACK =< SND.NXT then the ACK is acceptable. + */ + if (tcp_ack (tcp0)) + { + if (ack0 <= tc0->iss || ack0 > tc0->snd_nxt) + { + if (!tcp_rst (tcp0)) + tcp_send_reset (b0, is_ip4); + + goto drop; + } + + /* Make sure ACK is valid */ + if (tc0->snd_una > ack0) + goto drop; + } + + /* + * 2. check the RST bit + */ + + if (tcp_rst (tcp0)) + { + /* If ACK is acceptable, signal client that peer is not + * willing to accept connection and drop connection*/ + if (tcp_ack (tcp0)) + { + stream_session_connect_notify (&tc0->connection, sst, + 1 /* fail */ ); + tcp_connection_cleanup (tc0); + } + goto drop; + } + + /* + * 3. check the security and precedence (skipped) + */ + + /* + * 4. check the SYN bit + */ + + /* No SYN flag. Drop. */ + if (!tcp_syn (tcp0)) + goto drop; + + /* Stop connection establishment and retransmit timers */ + tcp_timer_reset (tc0, TCP_TIMER_ESTABLISH); + tcp_timer_reset (tc0, TCP_TIMER_RETRANSMIT_SYN); + + /* Valid SYN or SYN-ACK. Move connection from half-open pool to + * current thread pool. */ + pool_get (tm->connections[my_thread_index], new_tc0); + clib_memcpy (new_tc0, tc0, sizeof (*new_tc0)); + + new_tc0->c_thread_index = my_thread_index; + + /* Cleanup half-open connection XXX lock */ + pool_put (tm->half_open_connections, tc0); + + new_tc0->rcv_nxt = vnet_buffer (b0)->tcp.seq_end; + new_tc0->irs = seq0; + + /* Parse options */ + tcp_options_parse (tcp0, &new_tc0->opt); + tcp_connection_init_vars (new_tc0); + + if (tcp_opts_tstamp (&new_tc0->opt)) + { + new_tc0->tsval_recent = new_tc0->opt.tsval; + new_tc0->tsval_recent_age = tcp_time_now (); + } + + if (tcp_opts_wscale (&new_tc0->opt)) + new_tc0->snd_wscale = new_tc0->opt.wscale; + + new_tc0->snd_wnd = clib_net_to_host_u32 (tcp0->window) + << new_tc0->snd_wscale; + new_tc0->snd_wl1 = seq0; + new_tc0->snd_wl2 = ack0; + + /* SYN-ACK: See if we can switch to ESTABLISHED state */ + if (tcp_ack (tcp0)) + { + /* Our SYN is ACKed: we have iss < ack = snd_una */ + + /* TODO Dequeue acknowledged segments if we support Fast Open */ + new_tc0->snd_una = ack0; + new_tc0->state = TCP_STATE_ESTABLISHED; + + /* Notify app that we have connection */ + stream_session_connect_notify (&new_tc0->connection, sst, 0); + + /* Make sure after data segment processing ACK is sent */ + new_tc0->flags |= TCP_CONN_SNDACK; + } + /* SYN: Simultaneous open. Change state to SYN-RCVD and send SYN-ACK */ + else + { + new_tc0->state = TCP_STATE_SYN_RCVD; + + /* Notify app that we have connection XXX */ + stream_session_connect_notify (&new_tc0->connection, sst, 0); + + tcp_make_synack (new_tc0, b0); + next0 = tcp_next_output (is_ip4); + + goto drop; + } + + /* Read data, if any */ + if (n_data_bytes0) + { + error0 = + tcp_segment_rcv (tm, new_tc0, b0, n_data_bytes0, &next0); + if (error0 == TCP_ERROR_PURE_ACK) + error0 = TCP_ERROR_SYN_ACKS_RCVD; + } + else + { + tcp_make_ack (new_tc0, b0); + next0 = tcp_next_output (new_tc0->c_is_ip4); + } + + drop: + + b0->error = error0 ? node->errors[error0] : 0; + if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) + { + + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + errors = session_manager_flush_enqueue_events (my_thread_index); + if (errors) + { + if (is_ip4) + vlib_node_increment_counter (vm, tcp4_established_node.index, + TCP_ERROR_EVENT_FIFO_FULL, errors); + else + vlib_node_increment_counter (vm, tcp6_established_node.index, + TCP_ERROR_EVENT_FIFO_FULL, errors); + } + + return from_frame->n_vectors; +} + +static uword +tcp4_syn_sent (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_syn_sent_inline (vm, node, from_frame, 1 /* is_ip4 */ ); +} + +static uword +tcp6_syn_sent_rcv (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_syn_sent_inline (vm, node, from_frame, 0 /* is_ip4 */ ); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp4_syn_sent_node) = +{ + .function = tcp4_syn_sent, + .name = "tcp4-syn-sent", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_SYN_SENT_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp4_syn_sent_node, tcp4_syn_sent); + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp6_syn_sent_node) = +{ + .function = tcp6_syn_sent_rcv, + .name = "tcp6-syn-sent", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_SYN_SENT_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + } +,}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp6_syn_sent_node, tcp6_syn_sent_rcv); +/** + * Handles reception for all states except LISTEN, SYN-SEND and ESTABLISHED + * as per RFC793 p. 64 + */ +always_inline uword +tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame, int is_ip4) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + u32 n_left_from, next_index, *from, *to_next; + u32 my_thread_index = vm->cpu_index, errors = 0; + + from = vlib_frame_vector_args (from_frame); + n_left_from = from_frame->n_vectors; + + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + tcp_header_t *tcp0 = 0; + tcp_connection_t *tc0; + ip4_header_t *ip40; + ip6_header_t *ip60; + u32 n_advance_bytes0, n_data_bytes0; + u32 next0 = TCP_RCV_PROCESS_NEXT_DROP, error0 = TCP_ERROR_ENQUEUED; + + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index, + my_thread_index); + + /* Checksum computed by ipx_local no need to compute again */ + + if (is_ip4) + { + ip40 = vlib_buffer_get_current (b0); + tcp0 = ip4_next_header (ip40); + n_advance_bytes0 = (ip4_header_bytes (ip40) + + tcp_header_bytes (tcp0)); + n_data_bytes0 = clib_net_to_host_u16 (ip40->length) + - n_advance_bytes0; + } + else + { + ip60 = vlib_buffer_get_current (b0); + tcp0 = ip6_next_header (ip60); + n_advance_bytes0 = tcp_header_bytes (tcp0); + n_data_bytes0 = clib_net_to_host_u16 (ip60->payload_length) + - n_advance_bytes0; + n_advance_bytes0 += sizeof (ip60[0]); + } + + /* SYNs, FINs and data consume sequence numbers */ + vnet_buffer (b0)->tcp.seq_end = vnet_buffer (b0)->tcp.seq_number + + tcp_is_syn (tcp0) + tcp_is_fin (tcp0) + n_data_bytes0; + + /* + * Special treatment for CLOSED + */ + switch (tc0->state) + { + case TCP_STATE_CLOSED: + goto drop; + break; + } + + /* + * For all other states (except LISTEN) + */ + + /* 1-4: check SEQ, RST, SYN */ + if (PREDICT_FALSE + (tcp_segment_validate (vm, tc0, b0, tcp0, &next0))) + { + error0 = TCP_ERROR_SEGMENT_INVALID; + goto drop; + } + + /* 5: check the ACK field */ + switch (tc0->state) + { + case TCP_STATE_SYN_RCVD: + /* + * If the segment acknowledgment is not acceptable, form a + * reset segment, + * <SEQ=SEG.ACK><CTL=RST> + * and send it. + */ + if (!tcp_rcv_ack_is_acceptable (tc0, b0)) + { + tcp_send_reset (b0, is_ip4); + goto drop; + } + /* Switch state to ESTABLISHED */ + tc0->state = TCP_STATE_ESTABLISHED; + + /* Initialize session variables */ + tc0->snd_una = vnet_buffer (b0)->tcp.ack_number; + tc0->snd_wnd = clib_net_to_host_u32 (tcp0->window) + << tc0->opt.wscale; + tc0->snd_wl1 = vnet_buffer (b0)->tcp.seq_number; + tc0->snd_wl2 = vnet_buffer (b0)->tcp.ack_number; + + /* Shoulder tap the server */ + stream_session_accept_notify (&tc0->connection); + + tcp_timer_reset (tc0, TCP_TIMER_RETRANSMIT_SYN); + break; + case TCP_STATE_ESTABLISHED: + /* We can get packets in established state here because they + * were enqueued before state change */ + if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) + goto drop; + + break; + case TCP_STATE_FIN_WAIT_1: + /* In addition to the processing for the ESTABLISHED state, if + * our FIN is now acknowledged then enter FIN-WAIT-2 and + * continue processing in that state. */ + if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) + goto drop; + tc0->state = TCP_STATE_FIN_WAIT_2; + /* Stop all timers, 2MSL will be set lower */ + tcp_connection_timers_reset (tc0); + break; + case TCP_STATE_FIN_WAIT_2: + /* In addition to the processing for the ESTABLISHED state, if + * the retransmission queue is empty, the user's CLOSE can be + * acknowledged ("ok") but do not delete the TCB. */ + if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) + goto drop; + /* check if rtx queue is empty and ack CLOSE TODO */ + break; + case TCP_STATE_CLOSE_WAIT: + /* Do the same processing as for the ESTABLISHED state. */ + if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) + goto drop; + break; + case TCP_STATE_CLOSING: + /* In addition to the processing for the ESTABLISHED state, if + * the ACK acknowledges our FIN then enter the TIME-WAIT state, + * otherwise ignore the segment. */ + if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) + goto drop; + + /* XXX test that send queue empty */ + tc0->state = TCP_STATE_TIME_WAIT; + goto drop; + + break; + case TCP_STATE_LAST_ACK: + /* The only thing that can arrive in this state is an + * acknowledgment of our FIN. If our FIN is now acknowledged, + * delete the TCB, enter the CLOSED state, and return. */ + + if (!tcp_rcv_ack_is_acceptable (tc0, b0)) + goto drop; + + tcp_connection_del (tc0); + goto drop; + + break; + case TCP_STATE_TIME_WAIT: + /* The only thing that can arrive in this state is a + * retransmission of the remote FIN. Acknowledge it, and restart + * the 2 MSL timeout. */ + + /* TODO */ + goto drop; + break; + default: + ASSERT (0); + } + + /* 6: check the URG bit TODO */ + + /* 7: process the segment text */ + switch (tc0->state) + { + case TCP_STATE_ESTABLISHED: + case TCP_STATE_FIN_WAIT_1: + case TCP_STATE_FIN_WAIT_2: + error0 = tcp_segment_rcv (tm, tc0, b0, n_data_bytes0, &next0); + break; + case TCP_STATE_CLOSE_WAIT: + case TCP_STATE_CLOSING: + case TCP_STATE_LAST_ACK: + case TCP_STATE_TIME_WAIT: + /* This should not occur, since a FIN has been received from the + * remote side. Ignore the segment text. */ + break; + } + + /* 8: check the FIN bit */ + if (!tcp_fin (tcp0)) + goto drop; + + switch (tc0->state) + { + case TCP_STATE_ESTABLISHED: + case TCP_STATE_SYN_RCVD: + /* Send FIN-ACK notify app and enter CLOSE-WAIT */ + tcp_connection_timers_reset (tc0); + tcp_make_finack (tc0, b0); + next0 = tcp_next_output (tc0->c_is_ip4); + stream_session_disconnect_notify (&tc0->connection); + tc0->state = TCP_STATE_CLOSE_WAIT; + break; + case TCP_STATE_CLOSE_WAIT: + case TCP_STATE_CLOSING: + case TCP_STATE_LAST_ACK: + /* move along .. */ + break; + case TCP_STATE_FIN_WAIT_1: + tc0->state = TCP_STATE_TIME_WAIT; + tcp_connection_timers_reset (tc0); + tcp_timer_set (tc0, TCP_TIMER_2MSL, TCP_2MSL_TIME); + break; + case TCP_STATE_FIN_WAIT_2: + /* Got FIN, send ACK! */ + tc0->state = TCP_STATE_TIME_WAIT; + tcp_timer_set (tc0, TCP_TIMER_2MSL, TCP_2MSL_TIME); + tcp_make_ack (tc0, b0); + next0 = tcp_next_output (is_ip4); + break; + case TCP_STATE_TIME_WAIT: + /* Remain in the TIME-WAIT state. Restart the 2 MSL time-wait + * timeout. + */ + tcp_timer_update (tc0, TCP_TIMER_2MSL, TCP_2MSL_TIME); + break; + } + + b0->error = error0 ? node->errors[error0] : 0; + + drop: + if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) + { + + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + errors = session_manager_flush_enqueue_events (my_thread_index); + if (errors) + { + if (is_ip4) + vlib_node_increment_counter (vm, tcp4_established_node.index, + TCP_ERROR_EVENT_FIFO_FULL, errors); + else + vlib_node_increment_counter (vm, tcp6_established_node.index, + TCP_ERROR_EVENT_FIFO_FULL, errors); + } + + return from_frame->n_vectors; +} + +static uword +tcp4_rcv_process (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_rcv_process_inline (vm, node, from_frame, 1 /* is_ip4 */ ); +} + +static uword +tcp6_rcv_process (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_rcv_process_inline (vm, node, from_frame, 0 /* is_ip4 */ ); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp4_rcv_process_node) = +{ + .function = tcp4_rcv_process, + .name = "tcp4-rcv-process", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_RCV_PROCESS_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp4_rcv_process_node, tcp4_rcv_process); + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp6_rcv_process_node) = +{ + .function = tcp6_rcv_process, + .name = "tcp6-rcv-process", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_RCV_PROCESS_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp6_rcv_process_node, tcp6_rcv_process); + +vlib_node_registration_t tcp4_listen_node; +vlib_node_registration_t tcp6_listen_node; + +/** + * LISTEN state processing as per RFC 793 p. 65 + */ +always_inline uword +tcp46_listen_inline (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame, int is_ip4) +{ + u32 n_left_from, next_index, *from, *to_next; + u32 my_thread_index = vm->cpu_index; + tcp_main_t *tm = vnet_get_tcp_main (); + u8 sst = is_ip4 ? SESSION_TYPE_IP4_TCP : SESSION_TYPE_IP6_TCP; + + from = vlib_frame_vector_args (from_frame); + n_left_from = from_frame->n_vectors; + + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + tcp_header_t *th0 = 0; + tcp_connection_t *lc0; + ip4_header_t *ip40; + ip6_header_t *ip60; + tcp_connection_t *child0; + u32 error0 = TCP_ERROR_SYNS_RCVD, next0 = TCP_LISTEN_NEXT_DROP; + + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + lc0 = tcp_listener_get (vnet_buffer (b0)->tcp.connection_index); + + if (is_ip4) + { + ip40 = vlib_buffer_get_current (b0); + th0 = ip4_next_header (ip40); + } + else + { + ip60 = vlib_buffer_get_current (b0); + th0 = ip6_next_header (ip60); + } + + /* Create child session. For syn-flood protection use filter */ + + /* 1. first check for an RST */ + if (tcp_rst (th0)) + goto drop; + + /* 2. second check for an ACK */ + if (tcp_ack (th0)) + { + tcp_send_reset (b0, is_ip4); + goto drop; + } + + /* 3. check for a SYN (did that already) */ + + /* Create child session and send SYN-ACK */ + pool_get (tm->connections[my_thread_index], child0); + memset (child0, 0, sizeof (*child0)); + + child0->c_c_index = child0 - tm->connections[my_thread_index]; + child0->c_lcl_port = lc0->c_lcl_port; + child0->c_rmt_port = th0->src_port; + child0->c_is_ip4 = is_ip4; + child0->c_thread_index = my_thread_index; + + if (is_ip4) + { + child0->c_lcl_ip4.as_u32 = ip40->dst_address.as_u32; + child0->c_rmt_ip4.as_u32 = ip40->src_address.as_u32; + } + else + { + clib_memcpy (&child0->c_lcl_ip6, &ip60->dst_address, + sizeof (ip6_address_t)); + clib_memcpy (&child0->c_rmt_ip6, &ip60->src_address, + sizeof (ip6_address_t)); + } + + if (stream_session_accept (&child0->connection, lc0->c_s_index, sst, + 0 /* notify */ )) + { + error0 = TCP_ERROR_CREATE_SESSION_FAIL; + goto drop; + } + + tcp_options_parse (th0, &child0->opt); + tcp_connection_init_vars (child0); + + child0->irs = vnet_buffer (b0)->tcp.seq_number; + child0->rcv_nxt = vnet_buffer (b0)->tcp.seq_number + 1; + child0->state = TCP_STATE_SYN_RCVD; + + /* RFC1323: TSval timestamps sent on {SYN} and {SYN,ACK} + * segments are used to initialize PAWS. */ + if (tcp_opts_tstamp (&child0->opt)) + { + child0->tsval_recent = child0->opt.tsval; + child0->tsval_recent_age = tcp_time_now (); + } + + /* Reuse buffer to make syn-ack and send */ + tcp_make_synack (child0, b0); + next0 = tcp_next_output (is_ip4); + + drop: + if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) + { + + } + + b0->error = error0 ? node->errors[error0] : 0; + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + return from_frame->n_vectors; +} + +static uword +tcp4_listen (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_listen_inline (vm, node, from_frame, 1 /* is_ip4 */ ); +} + +static uword +tcp6_listen (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_listen_inline (vm, node, from_frame, 0 /* is_ip4 */ ); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp4_listen_node) = +{ + .function = tcp4_listen, + .name = "tcp4-listen", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_LISTEN_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_LISTEN_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp4_listen_node, tcp4_listen); + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp6_listen_node) = +{ + .function = tcp6_listen, + .name = "tcp6-listen", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_LISTEN_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_LISTEN_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp6_listen_node, tcp6_listen); + +vlib_node_registration_t tcp4_input_node; +vlib_node_registration_t tcp6_input_node; + +typedef enum _tcp_input_next +{ + TCP_INPUT_NEXT_DROP, + TCP_INPUT_NEXT_LISTEN, + TCP_INPUT_NEXT_RCV_PROCESS, + TCP_INPUT_NEXT_SYN_SENT, + TCP_INPUT_NEXT_ESTABLISHED, + TCP_INPUT_NEXT_RESET, + TCP_INPUT_N_NEXT +} tcp_input_next_t; + +#define foreach_tcp4_input_next \ + _ (DROP, "error-drop") \ + _ (LISTEN, "tcp4-listen") \ + _ (RCV_PROCESS, "tcp4-rcv-process") \ + _ (SYN_SENT, "tcp4-syn-sent") \ + _ (ESTABLISHED, "tcp4-established") \ + _ (RESET, "tcp4-reset") + +#define foreach_tcp6_input_next \ + _ (DROP, "error-drop") \ + _ (LISTEN, "tcp6-listen") \ + _ (RCV_PROCESS, "tcp6-rcv-process") \ + _ (SYN_SENT, "tcp6-syn-sent") \ + _ (ESTABLISHED, "tcp6-established") \ + _ (RESET, "tcp6-reset") + +typedef struct +{ + u16 src_port; + u16 dst_port; + u8 state; +} tcp_rx_trace_t; + +const char *tcp_fsm_states[] = { +#define _(sym, str) str, + foreach_tcp_fsm_state +#undef _ +}; + +u8 * +format_tcp_state (u8 * s, va_list * args) +{ + tcp_state_t *state = va_arg (*args, tcp_state_t *); + + if (state[0] < TCP_N_STATES) + s = format (s, "%s", tcp_fsm_states[state[0]]); + else + s = format (s, "UNKNOWN"); + + return s; +} + +u8 * +format_tcp_rx_trace (u8 * s, va_list * args) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); + tcp_rx_trace_t *t = va_arg (*args, tcp_rx_trace_t *); + + s = format (s, "TCP: src-port %d dst-port %U%s\n", + clib_net_to_host_u16 (t->src_port), + clib_net_to_host_u16 (t->dst_port), format_tcp_state, t->state); + + return s; +} + +#define filter_flags (TCP_FLAG_SYN|TCP_FLAG_ACK|TCP_FLAG_RST|TCP_FLAG_FIN) + +always_inline uword +tcp46_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame, int is_ip4) +{ + u32 n_left_from, next_index, *from, *to_next; + u32 my_thread_index = vm->cpu_index; + tcp_main_t *tm = vnet_get_tcp_main (); + session_manager_main_t *ssm = vnet_get_session_manager_main (); + + from = vlib_frame_vector_args (from_frame); + n_left_from = from_frame->n_vectors; + + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + tcp_header_t *tcp0 = 0; + tcp_connection_t *tc0; + ip4_header_t *ip40; + ip6_header_t *ip60; + u32 error0 = TCP_ERROR_NO_LISTENER, next0 = TCP_INPUT_NEXT_DROP; + u8 flags0; + + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + + if (is_ip4) + { + ip40 = vlib_buffer_get_current (b0); + tcp0 = ip4_next_header (ip40); + + /* lookup session */ + tc0 = + (tcp_connection_t *) stream_session_lookup_transport4 (ssm, + &ip40->dst_address, + &ip40->src_address, + tcp0->dst_port, + tcp0->src_port, + SESSION_TYPE_IP4_TCP, + my_thread_index); + } + else + { + ip60 = vlib_buffer_get_current (b0); + tcp0 = ip6_next_header (ip60); + tc0 = + (tcp_connection_t *) stream_session_lookup_transport6 (ssm, + &ip60->src_address, + &ip60->dst_address, + tcp0->src_port, + tcp0->dst_port, + SESSION_TYPE_IP6_TCP, + my_thread_index); + } + + /* Session exists */ + if (PREDICT_TRUE (0 != tc0)) + { + /* Save connection index */ + vnet_buffer (b0)->tcp.connection_index = tc0->c_c_index; + vnet_buffer (b0)->tcp.seq_number = + clib_net_to_host_u32 (tcp0->seq_number); + vnet_buffer (b0)->tcp.ack_number = + clib_net_to_host_u32 (tcp0->ack_number); + + flags0 = tcp0->flags & filter_flags; + next0 = tm->dispatch_table[tc0->state][flags0].next; + error0 = tm->dispatch_table[tc0->state][flags0].error; + + if (PREDICT_FALSE (error0 == TCP_ERROR_DISPATCH)) + { + /* Overload tcp flags to store state */ + vnet_buffer (b0)->tcp.flags = tc0->state; + } + } + else + { + /* Send reset */ + next0 = TCP_INPUT_NEXT_RESET; + error0 = TCP_ERROR_NO_LISTENER; + vnet_buffer (b0)->tcp.flags = 0; + } + + b0->error = error0 ? node->errors[error0] : 0; + + if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) + { + + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + return from_frame->n_vectors; +} + +static uword +tcp4_input (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_input_inline (vm, node, from_frame, 1 /* is_ip4 */ ); +} + +static uword +tcp6_input (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_input_inline (vm, node, from_frame, 0 /* is_ip4 */ ); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp4_input_node) = +{ + .function = tcp4_input, + .name = "tcp4-input", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_INPUT_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_INPUT_NEXT_##s] = n, + foreach_tcp4_input_next +#undef _ + }, + .format_buffer = format_tcp_header, + .format_trace = format_tcp_rx_trace, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp4_input_node, tcp4_input); + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp6_input_node) = +{ + .function = tcp6_input, + .name = "tcp6-input", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_INPUT_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_INPUT_NEXT_##s] = n, + foreach_tcp6_input_next +#undef _ + }, + .format_buffer = format_tcp_header, + .format_trace = format_tcp_rx_trace, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp6_input_node, tcp6_input); +void +tcp_update_time (f64 now, u32 thread_index) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + tw_timer_expire_timers_16t_2w_512sl (&tm->timer_wheels[thread_index], now); +} + +static void +tcp_dispatch_table_init (tcp_main_t * tm) +{ + int i, j; + for (i = 0; i < ARRAY_LEN (tm->dispatch_table); i++) + for (j = 0; j < ARRAY_LEN (tm->dispatch_table[i]); j++) + { + tm->dispatch_table[i][j].next = TCP_INPUT_NEXT_DROP; + tm->dispatch_table[i][j].error = TCP_ERROR_DISPATCH; + } + +#define _(t,f,n,e) \ +do { \ + tm->dispatch_table[TCP_STATE_##t][f].next = (n); \ + tm->dispatch_table[TCP_STATE_##t][f].error = (e); \ +} while (0) + + /* SYNs for new connections -> tcp-listen. */ + _(LISTEN, TCP_FLAG_SYN, TCP_INPUT_NEXT_LISTEN, TCP_ERROR_NONE); + /* ACK for for a SYN-ACK -> tcp-rcv-process. */ + _(SYN_RCVD, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); + /* SYN-ACK for a SYN */ + _(SYN_SENT, TCP_FLAG_SYN | TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, + TCP_ERROR_NONE); + _(SYN_SENT, TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE); + _(SYN_SENT, TCP_FLAG_RST, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE); + _(SYN_SENT, TCP_FLAG_RST | TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, + TCP_ERROR_NONE); + /* ACK for for established connection -> tcp-established. */ + _(ESTABLISHED, TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE); + /* FIN for for established connection -> tcp-established. */ + _(ESTABLISHED, TCP_FLAG_FIN, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE); + _(ESTABLISHED, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED, + TCP_ERROR_NONE); + /* ACK or FIN-ACK to our FIN */ + _(FIN_WAIT_1, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); + _(FIN_WAIT_1, TCP_FLAG_ACK | TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, + TCP_ERROR_NONE); + /* FIN in reply to our FIN from the other side */ + _(FIN_WAIT_1, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); + /* FIN confirming that the peer (app) has closed */ + _(FIN_WAIT_2, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); + _(FIN_WAIT_2, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, + TCP_ERROR_NONE); + _(LAST_ACK, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); +#undef _ +} + +clib_error_t * +tcp_input_init (vlib_main_t * vm) +{ + clib_error_t *error = 0; + tcp_main_t *tm = vnet_get_tcp_main (); + + if ((error = vlib_call_init_function (vm, tcp_init))) + return error; + + /* Initialize dispatch table. */ + tcp_dispatch_table_init (tm); + + return error; +} + +VLIB_INIT_FUNCTION (tcp_input_init); + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp_newreno.c b/src/vnet/tcp/tcp_newreno.c new file mode 100644 index 00000000000..856dffe4af9 --- /dev/null +++ b/src/vnet/tcp/tcp_newreno.c @@ -0,0 +1,93 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/tcp/tcp.h> + +void +newreno_congestion (tcp_connection_t * tc) +{ + tc->prev_ssthresh = tc->ssthresh; + tc->ssthresh = clib_max (tcp_flight_size (tc) / 2, 2 * tc->snd_mss); +} + +void +newreno_recovered (tcp_connection_t * tc) +{ + tc->cwnd = tc->ssthresh; +} + +void +newreno_rcv_ack (tcp_connection_t * tc) +{ + if (tcp_in_slowstart (tc)) + { + tc->cwnd += clib_min (tc->snd_mss, tc->bytes_acked); + } + else + { + /* Round up to 1 if needed */ + tc->cwnd += clib_max (tc->snd_mss * tc->snd_mss / tc->cwnd, 1); + } +} + +void +newreno_rcv_cong_ack (tcp_connection_t * tc, tcp_cc_ack_t ack_type) +{ + if (ack_type == TCP_CC_DUPACK) + { + tc->cwnd += tc->snd_mss; + } + else if (ack_type == TCP_CC_PARTIALACK) + { + tc->cwnd -= tc->bytes_acked; + if (tc->bytes_acked > tc->snd_mss) + tc->bytes_acked += tc->snd_mss; + } +} + +void +newreno_conn_init (tcp_connection_t * tc) +{ + tc->ssthresh = tc->snd_wnd; + tc->cwnd = tcp_initial_cwnd (tc); +} + +const static tcp_cc_algorithm_t tcp_newreno = { + .congestion = newreno_congestion, + .recovered = newreno_recovered, + .rcv_ack = newreno_rcv_ack, + .rcv_cong_ack = newreno_rcv_cong_ack, + .init = newreno_conn_init +}; + +clib_error_t * +newreno_init (vlib_main_t * vm) +{ + clib_error_t *error = 0; + + tcp_cc_algo_register (TCP_CC_NEWRENO, &tcp_newreno); + + return error; +} + +VLIB_INIT_FUNCTION (newreno_init); + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp_output.c b/src/vnet/tcp/tcp_output.c new file mode 100644 index 00000000000..dbcf1f74975 --- /dev/null +++ b/src/vnet/tcp/tcp_output.c @@ -0,0 +1,1412 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/tcp/tcp.h> +#include <vnet/lisp-cp/packets.h> + +vlib_node_registration_t tcp4_output_node; +vlib_node_registration_t tcp6_output_node; + +typedef enum _tcp_output_nect +{ + TCP_OUTPUT_NEXT_DROP, + TCP_OUTPUT_NEXT_IP_LOOKUP, + TCP_OUTPUT_N_NEXT +} tcp_output_next_t; + +#define foreach_tcp4_output_next \ + _ (DROP, "error-drop") \ + _ (IP_LOOKUP, "ip4-lookup") + +#define foreach_tcp6_output_next \ + _ (DROP, "error-drop") \ + _ (IP_LOOKUP, "ip6-lookup") + +static char *tcp_error_strings[] = { +#define tcp_error(n,s) s, +#include <vnet/tcp/tcp_error.def> +#undef tcp_error +}; + +typedef struct +{ + u16 src_port; + u16 dst_port; + u8 state; +} tcp_tx_trace_t; + +u16 dummy_mtu = 400; + +u8 * +format_tcp_tx_trace (u8 * s, va_list * args) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); + + s = format (s, "TBD\n"); + + return s; +} + +void +tcp_set_snd_mss (tcp_connection_t * tc) +{ + u16 snd_mss; + + /* TODO find our iface MTU */ + snd_mss = dummy_mtu; + + /* TODO cache mss and consider PMTU discovery */ + snd_mss = tc->opt.mss < snd_mss ? tc->opt.mss : snd_mss; + + tc->snd_mss = snd_mss; + + if (tc->snd_mss == 0) + { + clib_warning ("snd mss is 0"); + tc->snd_mss = dummy_mtu; + } +} + +static u8 +tcp_window_compute_scale (u32 available_space) +{ + u8 wnd_scale = 0; + while (wnd_scale < TCP_MAX_WND_SCALE + && (available_space >> wnd_scale) > TCP_WND_MAX) + wnd_scale++; + return wnd_scale; +} + +/** + * Compute initial window and scale factor. As per RFC1323, window field in + * SYN and SYN-ACK segments is never scaled. + */ +u32 +tcp_initial_window_to_advertise (tcp_connection_t * tc) +{ + u32 available_space; + + /* Initial wnd for SYN. Fifos are not allocated yet. + * Use some predefined value */ + if (tc->state != TCP_STATE_SYN_RCVD) + { + return TCP_DEFAULT_RX_FIFO_SIZE; + } + + available_space = stream_session_max_enqueue (&tc->connection); + tc->rcv_wscale = tcp_window_compute_scale (available_space); + tc->rcv_wnd = clib_min (available_space, TCP_WND_MAX << tc->rcv_wscale); + + return clib_min (tc->rcv_wnd, TCP_WND_MAX); +} + +/** + * Compute and return window to advertise, scaled as per RFC1323 + */ +u32 +tcp_window_to_advertise (tcp_connection_t * tc, tcp_state_t state) +{ + u32 available_space, wnd, scaled_space; + + if (state != TCP_STATE_ESTABLISHED) + return tcp_initial_window_to_advertise (tc); + + available_space = stream_session_max_enqueue (&tc->connection); + scaled_space = available_space >> tc->rcv_wscale; + + /* Need to update scale */ + if (PREDICT_FALSE ((scaled_space == 0 && available_space != 0)) + || (scaled_space >= TCP_WND_MAX)) + tc->rcv_wscale = tcp_window_compute_scale (available_space); + + wnd = clib_min (available_space, TCP_WND_MAX << tc->rcv_wscale); + tc->rcv_wnd = wnd; + + return wnd >> tc->rcv_wscale; +} + +/** + * Write TCP options to segment. + */ +u32 +tcp_options_write (u8 * data, tcp_options_t * opts) +{ + u32 opts_len = 0; + u32 buf, seq_len = 4; + + if (tcp_opts_mss (opts)) + { + *data++ = TCP_OPTION_MSS; + *data++ = TCP_OPTION_LEN_MSS; + buf = clib_host_to_net_u16 (opts->mss); + clib_memcpy (data, &buf, sizeof (opts->mss)); + data += sizeof (opts->mss); + opts_len += TCP_OPTION_LEN_MSS; + } + + if (tcp_opts_wscale (opts)) + { + *data++ = TCP_OPTION_WINDOW_SCALE; + *data++ = TCP_OPTION_LEN_WINDOW_SCALE; + *data++ = opts->wscale; + opts_len += TCP_OPTION_LEN_WINDOW_SCALE; + } + + if (tcp_opts_sack_permitted (opts)) + { + *data++ = TCP_OPTION_SACK_PERMITTED; + *data++ = TCP_OPTION_LEN_SACK_PERMITTED; + opts_len += TCP_OPTION_LEN_SACK_PERMITTED; + } + + if (tcp_opts_tstamp (opts)) + { + *data++ = TCP_OPTION_TIMESTAMP; + *data++ = TCP_OPTION_LEN_TIMESTAMP; + buf = clib_host_to_net_u32 (opts->tsval); + clib_memcpy (data, &buf, sizeof (opts->tsval)); + data += sizeof (opts->tsval); + buf = clib_host_to_net_u32 (opts->tsecr); + clib_memcpy (data, &buf, sizeof (opts->tsecr)); + data += sizeof (opts->tsecr); + opts_len += TCP_OPTION_LEN_TIMESTAMP; + } + + if (tcp_opts_sack (opts)) + { + int i; + u32 n_sack_blocks = clib_min (vec_len (opts->sacks), + TCP_OPTS_MAX_SACK_BLOCKS); + + if (n_sack_blocks != 0) + { + *data++ = TCP_OPTION_SACK_BLOCK; + *data++ = 2 + n_sack_blocks * TCP_OPTION_LEN_SACK_BLOCK; + for (i = 0; i < n_sack_blocks; i++) + { + buf = clib_host_to_net_u32 (opts->sacks[i].start); + clib_memcpy (data, &buf, seq_len); + data += seq_len; + buf = clib_host_to_net_u32 (opts->sacks[i].end); + clib_memcpy (data, &buf, seq_len); + data += seq_len; + } + opts_len += 2 + n_sack_blocks * TCP_OPTION_LEN_SACK_BLOCK; + } + } + + /* Terminate TCP options */ + if (opts_len % 4) + { + *data++ = TCP_OPTION_EOL; + opts_len += TCP_OPTION_LEN_EOL; + } + + /* Pad with zeroes to a u32 boundary */ + while (opts_len % 4) + { + *data++ = TCP_OPTION_NOOP; + opts_len += TCP_OPTION_LEN_NOOP; + } + return opts_len; +} + +always_inline int +tcp_make_syn_options (tcp_options_t * opts, u32 initial_wnd) +{ + u8 len = 0; + + opts->flags |= TCP_OPTS_FLAG_MSS; + opts->mss = dummy_mtu; /*XXX discover that */ + len += TCP_OPTION_LEN_MSS; + + opts->flags |= TCP_OPTS_FLAG_WSCALE; + opts->wscale = tcp_window_compute_scale (initial_wnd); + len += TCP_OPTION_LEN_WINDOW_SCALE; + + opts->flags |= TCP_OPTS_FLAG_TSTAMP; + opts->tsval = tcp_time_now (); + opts->tsecr = 0; + len += TCP_OPTION_LEN_TIMESTAMP; + + opts->flags |= TCP_OPTS_FLAG_SACK_PERMITTED; + len += TCP_OPTION_LEN_SACK_PERMITTED; + + /* Align to needed boundary */ + len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN; + return len; +} + +always_inline int +tcp_make_synack_options (tcp_connection_t * tc, tcp_options_t * opts) +{ + u8 len = 0; + + opts->flags |= TCP_OPTS_FLAG_MSS; + opts->mss = dummy_mtu; /*XXX discover that */ + len += TCP_OPTION_LEN_MSS; + + if (tcp_opts_wscale (&tc->opt)) + { + opts->flags |= TCP_OPTS_FLAG_WSCALE; + opts->wscale = tc->rcv_wscale; + len += TCP_OPTION_LEN_WINDOW_SCALE; + } + + if (tcp_opts_tstamp (&tc->opt)) + { + opts->flags |= TCP_OPTS_FLAG_TSTAMP; + opts->tsval = tcp_time_now (); + opts->tsecr = tc->tsval_recent; + len += TCP_OPTION_LEN_TIMESTAMP; + } + + if (tcp_opts_sack_permitted (&tc->opt)) + { + opts->flags |= TCP_OPTS_FLAG_SACK_PERMITTED; + len += TCP_OPTION_LEN_SACK_PERMITTED; + } + + /* Align to needed boundary */ + len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN; + return len; +} + +always_inline int +tcp_make_established_options (tcp_connection_t * tc, tcp_options_t * opts) +{ + u8 len = 0; + + opts->flags = 0; + + if (tcp_opts_tstamp (&tc->opt)) + { + opts->flags |= TCP_OPTS_FLAG_TSTAMP; + opts->tsval = tcp_time_now (); + opts->tsecr = tc->tsval_recent; + len += TCP_OPTION_LEN_TIMESTAMP; + } + if (tcp_opts_sack_permitted (&tc->opt)) + { + if (vec_len (tc->snd_sacks)) + { + opts->flags |= TCP_OPTS_FLAG_SACK; + opts->sacks = tc->snd_sacks; + opts->n_sack_blocks = vec_len (tc->snd_sacks); + len += 2 + TCP_OPTION_LEN_SACK_BLOCK * opts->n_sack_blocks; + } + } + + /* Align to needed boundary */ + len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN; + return len; +} + +always_inline int +tcp_make_options (tcp_connection_t * tc, tcp_options_t * opts, + tcp_state_t state) +{ + switch (state) + { + case TCP_STATE_ESTABLISHED: + case TCP_STATE_FIN_WAIT_1: + return tcp_make_established_options (tc, opts); + case TCP_STATE_SYN_RCVD: + return tcp_make_synack_options (tc, opts); + case TCP_STATE_SYN_SENT: + return tcp_make_syn_options (opts, + tcp_initial_window_to_advertise (tc)); + default: + clib_warning ("Not handled!"); + return 0; + } +} + +#define tcp_get_free_buffer_index(tm, bidx) \ +do { \ + u32 *my_tx_buffers, n_free_buffers; \ + u32 cpu_index = tm->vlib_main->cpu_index; \ + my_tx_buffers = tm->tx_buffers[cpu_index]; \ + if (PREDICT_FALSE(vec_len (my_tx_buffers) == 0)) \ + { \ + n_free_buffers = 32; /* TODO config or macro */ \ + vec_validate (my_tx_buffers, n_free_buffers - 1); \ + _vec_len(my_tx_buffers) = vlib_buffer_alloc_from_free_list ( \ + tm->vlib_main, my_tx_buffers, n_free_buffers, \ + VLIB_BUFFER_DEFAULT_FREE_LIST_INDEX); \ + tm->tx_buffers[cpu_index] = my_tx_buffers; \ + } \ + /* buffer shortage */ \ + if (PREDICT_FALSE (vec_len (my_tx_buffers) == 0)) \ + return; \ + *bidx = my_tx_buffers[_vec_len (my_tx_buffers)-1]; \ + _vec_len (my_tx_buffers) -= 1; \ +} while (0) + +always_inline void +tcp_reuse_buffer (vlib_main_t * vm, vlib_buffer_t * b) +{ + vlib_buffer_t *it = b; + do + { + it->current_data = 0; + it->current_length = 0; + it->total_length_not_including_first_buffer = 0; + } + while ((it->flags & VLIB_BUFFER_NEXT_PRESENT) + && (it = vlib_get_buffer (vm, it->next_buffer))); + + /* Leave enough space for headers */ + vlib_buffer_make_headroom (b, MAX_HDRS_LEN); +} + +/** + * Prepare ACK + */ +void +tcp_make_ack_i (tcp_connection_t * tc, vlib_buffer_t * b, tcp_state_t state, + u8 flags) +{ + tcp_options_t _snd_opts, *snd_opts = &_snd_opts; + u8 tcp_opts_len, tcp_hdr_opts_len; + tcp_header_t *th; + u16 wnd; + + wnd = tcp_window_to_advertise (tc, state); + + /* Make and write options */ + tcp_opts_len = tcp_make_established_options (tc, snd_opts); + tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t); + + th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->snd_nxt, + tc->rcv_nxt, tcp_hdr_opts_len, flags, wnd); + + tcp_options_write ((u8 *) (th + 1), snd_opts); + + /* Mark as ACK */ + vnet_buffer (b)->tcp.connection_index = tc->c_c_index; +} + +/** + * Convert buffer to ACK + */ +void +tcp_make_ack (tcp_connection_t * tc, vlib_buffer_t * b) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + + tcp_reuse_buffer (vm, b); + tcp_make_ack_i (tc, b, TCP_STATE_ESTABLISHED, TCP_FLAG_ACK); + vnet_buffer (b)->tcp.flags = TCP_BUF_FLAG_ACK; +} + +/** + * Convert buffer to FIN-ACK + */ +void +tcp_make_finack (tcp_connection_t * tc, vlib_buffer_t * b) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + + tcp_reuse_buffer (vm, b); + tcp_make_ack_i (tc, b, TCP_STATE_ESTABLISHED, TCP_FLAG_ACK | TCP_FLAG_FIN); + + /* Reset flags, make sure ack is sent */ + tc->flags = TCP_CONN_SNDACK; + vnet_buffer (b)->tcp.flags &= ~TCP_BUF_FLAG_DUPACK; + + tc->snd_nxt += 1; +} + +/** + * Convert buffer to SYN-ACK + */ +void +tcp_make_synack (tcp_connection_t * tc, vlib_buffer_t * b) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + tcp_options_t _snd_opts, *snd_opts = &_snd_opts; + u8 tcp_opts_len, tcp_hdr_opts_len; + tcp_header_t *th; + u16 initial_wnd; + u32 time_now; + + memset (snd_opts, 0, sizeof (*snd_opts)); + + tcp_reuse_buffer (vm, b); + + /* Set random initial sequence */ + time_now = tcp_time_now (); + + tc->iss = random_u32 (&time_now); + tc->snd_una = tc->iss; + tc->snd_nxt = tc->iss + 1; + tc->snd_una_max = tc->snd_nxt; + + initial_wnd = tcp_initial_window_to_advertise (tc); + + /* Make and write options */ + tcp_opts_len = tcp_make_synack_options (tc, snd_opts); + tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t); + + th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->iss, + tc->rcv_nxt, tcp_hdr_opts_len, + TCP_FLAG_SYN | TCP_FLAG_ACK, initial_wnd); + + tcp_options_write ((u8 *) (th + 1), snd_opts); + + vnet_buffer (b)->tcp.connection_index = tc->c_c_index; + vnet_buffer (b)->tcp.flags = TCP_BUF_FLAG_ACK; + + /* Init retransmit timer */ + tcp_retransmit_timer_set (tm, tc); +} + +always_inline void +tcp_enqueue_to_ip_lookup (vlib_main_t * vm, vlib_buffer_t * b, u32 bi, + u8 is_ip4) +{ + u32 *to_next, next_index; + vlib_frame_t *f; + + b->flags |= VNET_BUFFER_LOCALLY_ORIGINATED; + b->error = 0; + + /* Default FIB for now */ + vnet_buffer (b)->sw_if_index[VLIB_TX] = 0; + + /* Send to IP lookup */ + next_index = is_ip4 ? ip4_lookup_node.index : ip6_lookup_node.index; + f = vlib_get_frame_to_node (vm, next_index); + + /* Enqueue the packet */ + to_next = vlib_frame_vector_args (f); + to_next[0] = bi; + f->n_vectors = 1; + vlib_put_frame_to_node (vm, next_index, f); +} + +int +tcp_make_reset_in_place (vlib_main_t * vm, vlib_buffer_t * b0, + tcp_state_t state, u32 my_thread_index, u8 is_ip4) +{ + u8 tcp_hdr_len = sizeof (tcp_header_t); + ip4_header_t *ih4; + ip6_header_t *ih6; + tcp_header_t *th0; + ip4_address_t src_ip40; + ip6_address_t src_ip60; + u16 src_port0; + u32 tmp; + + /* Find IP and TCP headers */ + if (is_ip4) + { + ih4 = vlib_buffer_get_current (b0); + th0 = ip4_next_header (ih4); + } + else + { + ih6 = vlib_buffer_get_current (b0); + th0 = ip6_next_header (ih6); + } + + /* Swap src and dst ip */ + if (is_ip4) + { + ASSERT ((ih4->ip_version_and_header_length & 0xF0) == 0x40); + src_ip40.as_u32 = ih4->src_address.as_u32; + ih4->src_address.as_u32 = ih4->dst_address.as_u32; + ih4->dst_address.as_u32 = src_ip40.as_u32; + + /* Chop the end of the pkt */ + b0->current_length += ip4_header_bytes (ih4) + tcp_hdr_len; + } + else + { + ASSERT ((ih6->ip_version_traffic_class_and_flow_label & 0xF0) == 0x60); + clib_memcpy (&src_ip60, &ih6->src_address, sizeof (ip6_address_t)); + clib_memcpy (&ih6->src_address, &ih6->dst_address, + sizeof (ip6_address_t)); + clib_memcpy (&ih6->dst_address, &src_ip60, sizeof (ip6_address_t)); + + /* Chop the end of the pkt */ + b0->current_length += sizeof (ip6_header_t) + tcp_hdr_len; + } + + /* Try to determine what/why we're actually resetting and swap + * src and dst ports */ + if (state == TCP_STATE_CLOSED) + { + if (!tcp_syn (th0)) + return -1; + + tmp = clib_net_to_host_u32 (th0->seq_number); + + /* Got a SYN for no listener. */ + th0->flags = TCP_FLAG_RST | TCP_FLAG_ACK; + th0->ack_number = clib_host_to_net_u32 (tmp + 1); + th0->seq_number = 0; + + } + else if (state >= TCP_STATE_SYN_SENT) + { + th0->flags = TCP_FLAG_RST | TCP_FLAG_ACK; + th0->seq_number = th0->ack_number; + th0->ack_number = 0; + } + + src_port0 = th0->src_port; + th0->src_port = th0->dst_port; + th0->dst_port = src_port0; + th0->window = 0; + th0->data_offset_and_reserved = (tcp_hdr_len >> 2) << 4; + th0->urgent_pointer = 0; + + /* Compute checksum */ + if (is_ip4) + { + th0->checksum = ip4_tcp_udp_compute_checksum (vm, b0, ih4); + } + else + { + int bogus = ~0; + th0->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b0, ih6, &bogus); + ASSERT (!bogus); + } + + return 0; +} + +/** + * Send reset without reusing existing buffer + */ +void +tcp_send_reset (vlib_buffer_t * pkt, u8 is_ip4) +{ + vlib_buffer_t *b; + u32 bi; + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + u8 tcp_hdr_len, flags = 0; + tcp_header_t *th, *pkt_th; + u32 seq, ack; + ip4_header_t *ih4, *pkt_ih4; + ip6_header_t *ih6, *pkt_ih6; + + tcp_get_free_buffer_index (tm, &bi); + b = vlib_get_buffer (vm, bi); + + /* Leave enough space for headers */ + vlib_buffer_make_headroom (b, MAX_HDRS_LEN); + + /* Make and write options */ + tcp_hdr_len = sizeof (tcp_header_t); + + if (is_ip4) + { + pkt_ih4 = vlib_buffer_get_current (pkt); + pkt_th = ip4_next_header (pkt_ih4); + } + else + { + pkt_ih6 = vlib_buffer_get_current (pkt); + pkt_th = ip6_next_header (pkt_ih6); + } + + if (tcp_ack (pkt_th)) + { + flags = TCP_FLAG_RST; + seq = pkt_th->ack_number; + ack = 0; + } + else + { + flags = TCP_FLAG_RST | TCP_FLAG_ACK; + seq = 0; + ack = clib_host_to_net_u32 (vnet_buffer (pkt)->tcp.seq_end); + } + + th = vlib_buffer_push_tcp_net_order (b, pkt_th->dst_port, pkt_th->src_port, + seq, ack, tcp_hdr_len, flags, 0); + + /* Swap src and dst ip */ + if (is_ip4) + { + ASSERT ((pkt_ih4->ip_version_and_header_length & 0xF0) == 0x40); + ih4 = vlib_buffer_push_ip4 (vm, b, &pkt_ih4->dst_address, + &pkt_ih4->src_address, IP_PROTOCOL_TCP); + th->checksum = ip4_tcp_udp_compute_checksum (vm, b, ih4); + } + else + { + int bogus = ~0; + pkt_ih6 = (ip6_header_t *) (pkt_th - 1); + ASSERT ((pkt_ih6->ip_version_traffic_class_and_flow_label & 0xF0) == + 0x60); + ih6 = + vlib_buffer_push_ip6 (vm, b, &pkt_ih6->dst_address, + &pkt_ih6->src_address, IP_PROTOCOL_TCP); + th->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b, ih6, &bogus); + ASSERT (!bogus); + } + + tcp_enqueue_to_ip_lookup (vm, b, bi, is_ip4); +} + +void +tcp_push_ip_hdr (tcp_main_t * tm, tcp_connection_t * tc, vlib_buffer_t * b) +{ + tcp_header_t *th = vlib_buffer_get_current (b); + + if (tc->c_is_ip4) + { + ip4_header_t *ih; + ih = vlib_buffer_push_ip4 (tm->vlib_main, b, &tc->c_lcl_ip4, + &tc->c_rmt_ip4, IP_PROTOCOL_TCP); + th->checksum = ip4_tcp_udp_compute_checksum (tm->vlib_main, b, ih); + } + else + { + ip6_header_t *ih; + int bogus = ~0; + + ih = vlib_buffer_push_ip6 (tm->vlib_main, b, &tc->c_lcl_ip6, + &tc->c_rmt_ip6, IP_PROTOCOL_TCP); + th->checksum = ip6_tcp_udp_icmp_compute_checksum (tm->vlib_main, b, ih, + &bogus); + ASSERT (!bogus); + } +} + +/** + * Send SYN + * + * Builds a SYN packet for a half-open connection and sends it to ipx_lookup. + * The packet is not forwarded through tcpx_output to avoid doing lookups + * in the half_open pool. + */ +void +tcp_send_syn (tcp_connection_t * tc) +{ + vlib_buffer_t *b; + u32 bi; + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + u8 tcp_hdr_opts_len, tcp_opts_len; + tcp_header_t *th; + u32 time_now; + u16 initial_wnd; + tcp_options_t snd_opts; + + tcp_get_free_buffer_index (tm, &bi); + b = vlib_get_buffer (vm, bi); + + /* Leave enough space for headers */ + vlib_buffer_make_headroom (b, MAX_HDRS_LEN); + + /* Set random initial sequence */ + time_now = tcp_time_now (); + + tc->iss = random_u32 (&time_now); + tc->snd_una = tc->iss; + tc->snd_una_max = tc->snd_nxt = tc->iss + 1; + + initial_wnd = tcp_initial_window_to_advertise (tc); + + /* Make and write options */ + memset (&snd_opts, 0, sizeof (snd_opts)); + tcp_opts_len = tcp_make_syn_options (&snd_opts, initial_wnd); + tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t); + + th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->iss, + tc->rcv_nxt, tcp_hdr_opts_len, TCP_FLAG_SYN, + initial_wnd); + + tcp_options_write ((u8 *) (th + 1), &snd_opts); + + /* Measure RTT with this */ + tc->rtt_ts = tcp_time_now (); + tc->rtt_seq = tc->snd_nxt; + + /* Start retransmit trimer */ + tcp_timer_set (tc, TCP_TIMER_RETRANSMIT_SYN, tc->rto * TCP_TO_TIMER_TICK); + tc->rto_boff = 0; + + /* Set the connection establishment timer */ + tcp_timer_set (tc, TCP_TIMER_ESTABLISH, TCP_ESTABLISH_TIME); + + tcp_push_ip_hdr (tm, tc, b); + tcp_enqueue_to_ip_lookup (vm, b, bi, tc->c_is_ip4); +} + +always_inline void +tcp_enqueue_to_output (vlib_main_t * vm, vlib_buffer_t * b, u32 bi, u8 is_ip4) +{ + u32 *to_next, next_index; + vlib_frame_t *f; + + b->flags |= VNET_BUFFER_LOCALLY_ORIGINATED; + b->error = 0; + + /* Decide where to send the packet */ + next_index = is_ip4 ? tcp4_output_node.index : tcp6_output_node.index; + f = vlib_get_frame_to_node (vm, next_index); + + /* Enqueue the packet */ + to_next = vlib_frame_vector_args (f); + to_next[0] = bi; + f->n_vectors = 1; + vlib_put_frame_to_node (vm, next_index, f); +} + +/** + * Send FIN + */ +void +tcp_send_fin (tcp_connection_t * tc) +{ + vlib_buffer_t *b; + u32 bi; + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + + tcp_get_free_buffer_index (tm, &bi); + b = vlib_get_buffer (vm, bi); + + /* Leave enough space for headers */ + vlib_buffer_make_headroom (b, MAX_HDRS_LEN); + + tcp_make_finack (tc, b); + + tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4); +} + +always_inline u8 +tcp_make_state_flags (tcp_state_t next_state) +{ + switch (next_state) + { + case TCP_STATE_ESTABLISHED: + return TCP_FLAG_ACK; + case TCP_STATE_SYN_RCVD: + return TCP_FLAG_SYN | TCP_FLAG_ACK; + case TCP_STATE_SYN_SENT: + return TCP_FLAG_SYN; + case TCP_STATE_LAST_ACK: + case TCP_STATE_FIN_WAIT_1: + return TCP_FLAG_FIN; + default: + clib_warning ("Shouldn't be here!"); + } + return 0; +} + +/** + * Push TCP header and update connection variables + */ +static void +tcp_push_hdr_i (tcp_connection_t * tc, vlib_buffer_t * b, + tcp_state_t next_state) +{ + u32 advertise_wnd, data_len; + u8 tcp_opts_len, tcp_hdr_opts_len, opts_write_len, flags; + tcp_options_t _snd_opts, *snd_opts = &_snd_opts; + tcp_header_t *th; + + data_len = b->current_length; + vnet_buffer (b)->tcp.flags = 0; + + /* Make and write options */ + memset (snd_opts, 0, sizeof (*snd_opts)); + tcp_opts_len = tcp_make_options (tc, snd_opts, next_state); + tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t); + + /* Get rcv window to advertise */ + advertise_wnd = tcp_window_to_advertise (tc, next_state); + flags = tcp_make_state_flags (next_state); + + /* Push header and options */ + th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->snd_nxt, + tc->rcv_nxt, tcp_hdr_opts_len, flags, + advertise_wnd); + + opts_write_len = tcp_options_write ((u8 *) (th + 1), snd_opts); + + ASSERT (opts_write_len == tcp_opts_len); + + /* Tag the buffer with the connection index */ + vnet_buffer (b)->tcp.connection_index = tc->c_c_index; + + tc->snd_nxt += data_len; +} + +/* Send delayed ACK when timer expires */ +void +tcp_timer_delack_handler (u32 index) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + u32 thread_index = os_get_cpu_number (); + tcp_connection_t *tc; + vlib_buffer_t *b; + u32 bi; + + tc = tcp_connection_get (index, thread_index); + + /* Get buffer */ + tcp_get_free_buffer_index (tm, &bi); + b = vlib_get_buffer (vm, bi); + + /* Fill in the ACK */ + tcp_make_ack (tc, b); + + tc->timers[TCP_TIMER_DELACK] = TCP_TIMER_HANDLE_INVALID; + tc->flags &= ~TCP_CONN_DELACK; + + tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4); +} + +/** Build a retransmit segment + * + * @return the number of bytes in the segment or 0 if there's nothing to + * retransmit + * */ +u32 +tcp_prepare_retransmit_segment (tcp_connection_t * tc, vlib_buffer_t * b, + u32 max_bytes) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + u32 n_bytes, offset = 0; + sack_scoreboard_hole_t *hole; + u32 hole_size; + + tcp_reuse_buffer (vm, b); + + ASSERT (tc->state == TCP_STATE_ESTABLISHED); + ASSERT (max_bytes != 0); + + if (tcp_opts_sack_permitted (&tc->opt)) + { + /* XXX get first hole not retransmitted yet */ + hole = scoreboard_first_hole (&tc->sack_sb); + if (!hole) + return 0; + + offset = hole->start - tc->snd_una; + hole_size = hole->end - hole->start; + + ASSERT (hole_size); + + if (hole_size < max_bytes) + max_bytes = hole_size; + } + else + { + if (seq_geq (tc->snd_nxt, tc->snd_una_max)) + return 0; + } + + n_bytes = stream_session_peek_bytes (&tc->connection, + vlib_buffer_get_current (b), offset, + max_bytes); + ASSERT (n_bytes != 0); + + tc->snd_nxt += n_bytes; + tcp_push_hdr_i (tc, b, tc->state); + + return n_bytes; +} + +static void +tcp_timer_retransmit_handler_i (u32 index, u8 is_syn) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + u32 thread_index = os_get_cpu_number (); + tcp_connection_t *tc; + vlib_buffer_t *b; + u32 bi, max_bytes, snd_space; + + if (is_syn) + { + tc = tcp_half_open_connection_get (index); + } + else + { + tc = tcp_connection_get (index, thread_index); + } + + /* Make sure timer handle is set to invalid */ + tc->timers[TCP_TIMER_RETRANSMIT] = TCP_TIMER_HANDLE_INVALID; + + /* Increment RTO backoff (also equal to number of retries) */ + tc->rto_boff += 1; + + /* Go back to first un-acked byte */ + tc->snd_nxt = tc->snd_una; + + /* Get buffer */ + tcp_get_free_buffer_index (tm, &bi); + b = vlib_get_buffer (vm, bi); + + if (tc->state == TCP_STATE_ESTABLISHED) + { + tcp_fastrecovery_off (tc); + + /* Exponential backoff */ + tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX); + + /* Figure out what and how many bytes we can send */ + snd_space = tcp_available_snd_space (tc); + max_bytes = clib_min (tc->snd_mss, snd_space); + tcp_prepare_retransmit_segment (tc, b, max_bytes); + + tc->rtx_bytes += max_bytes; + + /* No fancy recovery for now! */ + scoreboard_clear (&tc->sack_sb); + } + else + { + /* Retransmit for SYN/SYNACK */ + ASSERT (tc->state == TCP_STATE_SYN_RCVD + || tc->state == TCP_STATE_SYN_SENT); + + /* Try without increasing RTO a number of times. If this fails, + * start growing RTO exponentially */ + if (tc->rto_boff > TCP_RTO_SYN_RETRIES) + tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX); + + vlib_buffer_make_headroom (b, MAX_HDRS_LEN); + tcp_push_hdr_i (tc, b, tc->state); + } + + if (!is_syn) + { + tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4); + + /* Re-enable retransmit timer */ + tcp_retransmit_timer_set (tm, tc); + } + else + { + ASSERT (tc->state == TCP_STATE_SYN_SENT); + + /* This goes straight to ipx_lookup */ + tcp_push_ip_hdr (tm, tc, b); + tcp_enqueue_to_ip_lookup (vm, b, bi, tc->c_is_ip4); + + /* Re-enable retransmit timer */ + tcp_timer_set (tc, TCP_TIMER_RETRANSMIT_SYN, + tc->rto * TCP_TO_TIMER_TICK); + } +} + +void +tcp_timer_retransmit_handler (u32 index) +{ + tcp_timer_retransmit_handler_i (index, 0); +} + +void +tcp_timer_retransmit_syn_handler (u32 index) +{ + tcp_timer_retransmit_handler_i (index, 1); +} + +/** + * Retansmit first unacked segment */ +void +tcp_retransmit_first_unacked (tcp_connection_t * tc) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + u32 snd_nxt = tc->snd_nxt; + vlib_buffer_t *b; + u32 bi; + + tc->snd_nxt = tc->snd_una; + + /* Get buffer */ + tcp_get_free_buffer_index (tm, &bi); + b = vlib_get_buffer (tm->vlib_main, bi); + + tcp_prepare_retransmit_segment (tc, b, tc->snd_mss); + tcp_enqueue_to_output (tm->vlib_main, b, bi, tc->c_is_ip4); + + tc->snd_nxt = snd_nxt; + tc->rtx_bytes += tc->snd_mss; +} + +void +tcp_fast_retransmit (tcp_connection_t * tc) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + u32 snd_space, max_bytes, n_bytes, bi; + vlib_buffer_t *b; + + ASSERT (tcp_in_fastrecovery (tc)); + + clib_warning ("fast retransmit!"); + + /* Start resending from first un-acked segment */ + tc->snd_nxt = tc->snd_una; + + snd_space = tcp_available_snd_space (tc); + + while (snd_space) + { + tcp_get_free_buffer_index (tm, &bi); + b = vlib_get_buffer (tm->vlib_main, bi); + + max_bytes = clib_min (tc->snd_mss, snd_space); + n_bytes = tcp_prepare_retransmit_segment (tc, b, max_bytes); + + /* Nothing left to retransmit */ + if (n_bytes == 0) + return; + + tcp_enqueue_to_output (tm->vlib_main, b, bi, tc->c_is_ip4); + + snd_space -= n_bytes; + } + + /* If window allows, send new data */ + tc->snd_nxt = tc->snd_una_max; +} + +always_inline u32 +tcp_session_has_ooo_data (tcp_connection_t * tc) +{ + stream_session_t *s = + stream_session_get (tc->c_s_index, tc->c_thread_index); + return svm_fifo_has_ooo_data (s->server_rx_fifo); +} + +always_inline uword +tcp46_output_inline (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame, int is_ip4) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + u32 n_left_from, next_index, *from, *to_next; + u32 my_thread_index = vm->cpu_index; + + from = vlib_frame_vector_args (from_frame); + n_left_from = from_frame->n_vectors; + + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + tcp_connection_t *tc0; + tcp_header_t *th0; + u32 error0 = TCP_ERROR_PKTS_SENT, next0 = TCP_OUTPUT_NEXT_IP_LOOKUP; + + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index, + my_thread_index); + th0 = vlib_buffer_get_current (b0); + + if (is_ip4) + { + ip4_header_t *ih0; + ih0 = vlib_buffer_push_ip4 (vm, b0, &tc0->c_lcl_ip4, + &tc0->c_rmt_ip4, IP_PROTOCOL_TCP); + th0->checksum = ip4_tcp_udp_compute_checksum (vm, b0, ih0); + } + else + { + ip6_header_t *ih0; + int bogus = ~0; + + ih0 = vlib_buffer_push_ip6 (vm, b0, &tc0->c_lcl_ip6, + &tc0->c_rmt_ip6, IP_PROTOCOL_TCP); + th0->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b0, ih0, + &bogus); + ASSERT (!bogus); + } + + /* Filter out DUPACKs if there are no OOO segments left */ + if (PREDICT_FALSE + (vnet_buffer (b0)->tcp.flags & TCP_BUF_FLAG_DUPACK)) + { + tc0->snt_dupacks--; + ASSERT (tc0->snt_dupacks >= 0); + if (!tcp_session_has_ooo_data (tc0)) + { + error0 = TCP_ERROR_FILTERED_DUPACKS; + next0 = TCP_OUTPUT_NEXT_DROP; + goto done; + } + } + + /* Retransmitted SYNs do reach this but it should be harmless */ + tc0->rcv_las = tc0->rcv_nxt; + + /* Stop DELACK timer and fix flags */ + tc0->flags &= + ~(TCP_CONN_SNDACK | TCP_CONN_DELACK | TCP_CONN_BURSTACK); + if (tcp_timer_is_active (tc0, TCP_TIMER_DELACK)) + { + tcp_timer_reset (tc0, TCP_TIMER_DELACK); + } + + /* If not retransmitting + * 1) update snd_una_max (SYN, SYNACK, new data, FIN) + * 2) If we're not tracking an ACK, start tracking */ + if (seq_lt (tc0->snd_una_max, tc0->snd_nxt)) + { + tc0->snd_una_max = tc0->snd_nxt; + if (tc0->rtt_ts == 0) + { + tc0->rtt_ts = tcp_time_now (); + tc0->rtt_seq = tc0->snd_nxt; + } + } + + /* Set the retransmit timer if not set already and not + * doing a pure ACK */ + if (!tcp_timer_is_active (tc0, TCP_TIMER_RETRANSMIT) + && tc0->snd_nxt != tc0->snd_una) + { + tcp_retransmit_timer_set (tm, tc0); + tc0->rto_boff = 0; + } + + /* set fib index to default and lookup node */ + /* XXX network virtualization (vrf/vni) */ + vnet_buffer (b0)->sw_if_index[VLIB_RX] = 0; + vnet_buffer (b0)->sw_if_index[VLIB_TX] = (u32) ~ 0; + + b0->flags |= VNET_BUFFER_LOCALLY_ORIGINATED; + + done: + b0->error = error0 != 0 ? node->errors[error0] : 0; + if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) + { + + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + return from_frame->n_vectors; +} + +static uword +tcp4_output (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_output_inline (vm, node, from_frame, 1 /* is_ip4 */ ); +} + +static uword +tcp6_output (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_output_inline (vm, node, from_frame, 0 /* is_ip4 */ ); +} + +VLIB_REGISTER_NODE (tcp4_output_node) = +{ + .function = tcp4_output,.name = "tcp4-output", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32),.n_errors = TCP_N_ERROR,.error_strings = + tcp_error_strings,.n_next_nodes = TCP_OUTPUT_N_NEXT,.next_nodes = + { +#define _(s,n) [TCP_OUTPUT_NEXT_##s] = n, + foreach_tcp4_output_next +#undef _ + } +,.format_buffer = format_tcp_header,.format_trace = format_tcp_tx_trace,}; + +VLIB_NODE_FUNCTION_MULTIARCH (tcp4_output_node, tcp4_output) +VLIB_REGISTER_NODE (tcp6_output_node) = +{ + .function = tcp6_output,.name = "tcp6-output", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32),.n_errors = TCP_N_ERROR,.error_strings = + tcp_error_strings,.n_next_nodes = TCP_OUTPUT_N_NEXT,.next_nodes = + { +#define _(s,n) [TCP_OUTPUT_NEXT_##s] = n, + foreach_tcp6_output_next +#undef _ + } +,.format_buffer = format_tcp_header,.format_trace = format_tcp_tx_trace,}; + +VLIB_NODE_FUNCTION_MULTIARCH (tcp6_output_node, tcp6_output) u32 +tcp_push_header (transport_connection_t * tconn, vlib_buffer_t * b) +{ + tcp_connection_t *tc; + + tc = (tcp_connection_t *) tconn; + tcp_push_hdr_i (tc, b, TCP_STATE_ESTABLISHED); + return 0; +} + +typedef enum _tcp_reset_next +{ + TCP_RESET_NEXT_DROP, + TCP_RESET_NEXT_IP_LOOKUP, + TCP_RESET_N_NEXT +} tcp_reset_next_t; + +#define foreach_tcp4_reset_next \ + _(DROP, "error-drop") \ + _(IP_LOOKUP, "ip4-lookup") + +#define foreach_tcp6_reset_next \ + _(DROP, "error-drop") \ + _(IP_LOOKUP, "ip6-lookup") + +static uword +tcp46_send_reset_inline (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame, u8 is_ip4) +{ + u32 n_left_from, next_index, *from, *to_next; + u32 my_thread_index = vm->cpu_index; + + from = vlib_frame_vector_args (from_frame); + n_left_from = from_frame->n_vectors; + + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + u32 error0 = TCP_ERROR_RST_SENT, next0 = TCP_RESET_NEXT_IP_LOOKUP; + + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + + if (tcp_make_reset_in_place (vm, b0, vnet_buffer (b0)->tcp.flags, + my_thread_index, is_ip4)) + { + error0 = TCP_ERROR_LOOKUP_DROPS; + next0 = TCP_RESET_NEXT_DROP; + goto done; + } + + /* Prepare to send to IP lookup */ + vnet_buffer (b0)->sw_if_index[VLIB_TX] = 0; + next0 = TCP_RESET_NEXT_IP_LOOKUP; + + done: + b0->error = error0 != 0 ? node->errors[error0] : 0; + b0->flags |= VNET_BUFFER_LOCALLY_ORIGINATED; + if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) + { + + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + } + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + return from_frame->n_vectors; +} + +static uword +tcp4_send_reset (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_send_reset_inline (vm, node, from_frame, 1); +} + +static uword +tcp6_send_reset (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_send_reset_inline (vm, node, from_frame, 0); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp4_reset_node) = { + .function = tcp4_send_reset, + .name = "tcp4-reset", + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_RESET_N_NEXT, + .next_nodes = { +#define _(s,n) [TCP_RESET_NEXT_##s] = n, + foreach_tcp4_reset_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp6_reset_node) = { + .function = tcp6_send_reset, + .name = "tcp6-reset", + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_RESET_N_NEXT, + .next_nodes = { +#define _(s,n) [TCP_RESET_NEXT_##s] = n, + foreach_tcp6_reset_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp_packet.h b/src/vnet/tcp/tcp_packet.h new file mode 100644 index 00000000000..866c5fd6046 --- /dev/null +++ b/src/vnet/tcp/tcp_packet.h @@ -0,0 +1,184 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef included_tcp_packet_h +#define included_tcp_packet_h + +#include <vnet/vnet.h> + +/* TCP flags bit 0 first. */ +#define foreach_tcp_flag \ + _ (FIN) /**< No more data from sender. */ \ + _ (SYN) /**< Synchronize sequence numbers. */ \ + _ (RST) /**< Reset the connection. */ \ + _ (PSH) /**< Push function. */ \ + _ (ACK) /**< Ack field significant. */ \ + _ (URG) /**< Urgent pointer field significant. */ \ + _ (ECE) /**< ECN-echo. Receiver got CE packet */ \ + _ (CWR) /**< Sender reduced congestion window */ + +enum +{ +#define _(f) TCP_FLAG_BIT_##f, + foreach_tcp_flag +#undef _ + TCP_N_FLAG_BITS, +}; + +enum +{ +#define _(f) TCP_FLAG_##f = 1 << TCP_FLAG_BIT_##f, + foreach_tcp_flag +#undef _ +}; + +typedef struct _tcp_header +{ + union + { + struct + { + u16 src_port; /**< Source port. */ + u16 dst_port; /**< Destination port. */ + }; + struct + { + u16 src, dst; + }; + }; + + u32 seq_number; /**< Sequence number of the first data octet in this + * segment, except when SYN is present. If SYN + * is present the seq number is is the ISN and the + * first data octet is ISN+1 */ + u32 ack_number; /**< Acknowledgement number if ACK is set. It contains + * the value of the next sequence number the sender + * of the segment is expecting to receive. */ + u8 data_offset_and_reserved; + u8 flags; /**< Flags: see the macro above */ + u16 window; /**< Number of bytes sender is willing to receive. */ + + u16 checksum; /**< Checksum of TCP pseudo header and data. */ + u16 urgent_pointer; /**< Seq number of the byte after the urgent data. */ +} __attribute__ ((packed)) tcp_header_t; + +/* Flag tests that return 0 or !0 */ +#define tcp_doff(_th) ((_th)->data_offset_and_reserved >> 4) +#define tcp_fin(_th) ((_th)->flags & TCP_FLAG_FIN) +#define tcp_syn(_th) ((_th)->flags & TCP_FLAG_SYN) +#define tcp_rst(_th) ((_th)->flags & TCP_FLAG_RST) +#define tcp_psh(_th) ((_th)->flags & TCP_FLAG_PSH) +#define tcp_ack(_th) ((_th)->flags & TCP_FLAG_ACK) +#define tcp_urg(_th) ((_th)->flags & TCP_FLAG_URG) +#define tcp_ece(_th) ((_th)->flags & TCP_FLAG_ECE) +#define tcp_cwr(_th) ((_th)->flags & TCP_FLAG_CWR) + +/* Flag tests that return 0 or 1 */ +#define tcp_is_syn(_th) !!((_th)->flags & TCP_FLAG_SYN) +#define tcp_is_fin(_th) !!((_th)->flags & TCP_FLAG_FIN) + +always_inline int +tcp_header_bytes (tcp_header_t * t) +{ + return tcp_doff (t) * sizeof (u32); +} + +/* + * TCP options. + */ + +typedef enum tcp_option_type +{ + TCP_OPTION_EOL = 0, /**< End of options. */ + TCP_OPTION_NOOP = 1, /**< No operation. */ + TCP_OPTION_MSS = 2, /**< Limit MSS. */ + TCP_OPTION_WINDOW_SCALE = 3, /**< Window scale. */ + TCP_OPTION_SACK_PERMITTED = 4, /**< Selective Ack permitted. */ + TCP_OPTION_SACK_BLOCK = 5, /**< Selective Ack block. */ + TCP_OPTION_TIMESTAMP = 8, /**< Timestamps. */ + TCP_OPTION_UTO = 28, /**< User timeout. */ + TCP_OPTION_AO = 29, /**< Authentication Option. */ +} tcp_option_type_t; + +#define foreach_tcp_options_flag \ + _ (MSS) /**< MSS advertised in SYN */ \ + _ (TSTAMP) /**< Timestamp capability advertised in SYN */ \ + _ (WSCALE) /**< Wnd scale capability advertised in SYN */ \ + _ (SACK_PERMITTED) /**< SACK capability advertised in SYN */ \ + _ (SACK) /**< SACK present */ + +enum +{ +#define _(f) TCP_OPTS_FLAG_BIT_##f, + foreach_tcp_options_flag +#undef _ + TCP_OPTIONS_N_FLAG_BITS, +}; + +enum +{ +#define _(f) TCP_OPTS_FLAG_##f = 1 << TCP_OPTS_FLAG_BIT_##f, + foreach_tcp_options_flag +#undef _ +}; + +typedef struct _sack_block +{ + u32 start; /**< Start sequence number */ + u32 end; /**< End sequence number */ +} sack_block_t; + +typedef struct +{ + u8 flags; /** Option flags, see above */ + + /* Received options */ + u16 mss; /**< Maximum segment size advertised by peer */ + u8 wscale; /**< Window scale advertised by peer */ + u32 tsval; /**< Peer's timestamp value */ + u32 tsecr; /**< Echoed/reflected time stamp */ + sack_block_t *sacks; /**< SACK blocks received */ + u8 n_sack_blocks; /**< Number of SACKs blocks */ +} tcp_options_t; + +/* Flag tests that return 0 or !0 */ +#define tcp_opts_mss(_to) ((_to)->flags & TCP_OPTS_FLAG_MSS) +#define tcp_opts_tstamp(_to) ((_to)->flags & TCP_OPTS_FLAG_TSTAMP) +#define tcp_opts_wscale(_to) ((_to)->flags & TCP_OPTS_FLAG_WSCALE) +#define tcp_opts_sack(_to) ((_to)->flags & TCP_OPTS_FLAG_SACK) +#define tcp_opts_sack_permitted(_to) ((_to)->flags & TCP_OPTS_FLAG_SACK_PERMITTED) + +/* TCP option lengths */ +#define TCP_OPTION_LEN_EOL 1 +#define TCP_OPTION_LEN_NOOP 1 +#define TCP_OPTION_LEN_MSS 4 +#define TCP_OPTION_LEN_WINDOW_SCALE 3 +#define TCP_OPTION_LEN_SACK_PERMITTED 2 +#define TCP_OPTION_LEN_TIMESTAMP 10 +#define TCP_OPTION_LEN_SACK_BLOCK 8 + +#define TCP_WND_MAX 65535U +#define TCP_MAX_WND_SCALE 14 /* See RFC 1323 */ +#define TCP_OPTS_ALIGN 4 +#define TCP_OPTS_MAX_SACK_BLOCKS 3 +#endif /* included_tcp_packet_h */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp_pg.c b/src/vnet/tcp/tcp_pg.c new file mode 100644 index 00000000000..dc3240496df --- /dev/null +++ b/src/vnet/tcp/tcp_pg.c @@ -0,0 +1,236 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +/* + * ip/tcp_pg: TCP packet-generator interface + * + * Copyright (c) 2008 Eliot Dresselhaus + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice shall be + * included in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +#include <vnet/ip/ip.h> +#include <vnet/pg/pg.h> + +/* TCP flags bit 0 first. */ +#define foreach_tcp_flag \ + _ (FIN) \ + _ (SYN) \ + _ (RST) \ + _ (PSH) \ + _ (ACK) \ + _ (URG) \ + _ (ECE) \ + _ (CWR) + +static void +tcp_pg_edit_function (pg_main_t * pg, + pg_stream_t * s, + pg_edit_group_t * g, + u32 * packets, + u32 n_packets) +{ + vlib_main_t * vm = vlib_get_main(); + u32 ip_offset, tcp_offset; + + tcp_offset = g->start_byte_offset; + ip_offset = (g-1)->start_byte_offset; + + while (n_packets >= 1) + { + vlib_buffer_t * p0; + ip4_header_t * ip0; + tcp_header_t * tcp0; + ip_csum_t sum0; + u32 tcp_len0; + + p0 = vlib_get_buffer (vm, packets[0]); + n_packets -= 1; + packets += 1; + + ASSERT (p0->current_data == 0); + ip0 = (void *) (p0->data + ip_offset); + tcp0 = (void *) (p0->data + tcp_offset); + tcp_len0 = clib_net_to_host_u16 (ip0->length) - sizeof (ip0[0]); + + /* Initialize checksum with header. */ + if (BITS (sum0) == 32) + { + sum0 = clib_mem_unaligned (&ip0->src_address, u32); + sum0 = ip_csum_with_carry (sum0, clib_mem_unaligned (&ip0->dst_address, u32)); + } + else + sum0 = clib_mem_unaligned (&ip0->src_address, u64); + + sum0 = ip_csum_with_carry + (sum0, clib_host_to_net_u32 (tcp_len0 + (ip0->protocol << 16))); + + /* Invalidate possibly old checksum. */ + tcp0->checksum = 0; + + sum0 = ip_incremental_checksum_buffer (vm, p0, tcp_offset, tcp_len0, sum0); + + tcp0->checksum = ~ ip_csum_fold (sum0); + } +} + +typedef struct { + pg_edit_t src, dst; + pg_edit_t seq_number, ack_number; + pg_edit_t data_offset_and_reserved; +#define _(f) pg_edit_t f##_flag; + foreach_tcp_flag +#undef _ + pg_edit_t window; + pg_edit_t checksum; + pg_edit_t urgent_pointer; +} pg_tcp_header_t; + +static inline void +pg_tcp_header_init (pg_tcp_header_t * p) +{ + /* Initialize fields that are not bit fields in the IP header. */ +#define _(f) pg_edit_init (&p->f, tcp_header_t, f); + _ (src); + _ (dst); + _ (seq_number); + _ (ack_number); + _ (window); + _ (checksum); + _ (urgent_pointer); +#undef _ + + /* Initialize bit fields. */ +#define _(f) \ + pg_edit_init_bitfield (&p->f##_flag, tcp_header_t, \ + flags, \ + TCP_FLAG_BIT_##f, 1); + + foreach_tcp_flag +#undef _ + + pg_edit_init_bitfield (&p->data_offset_and_reserved, tcp_header_t, + data_offset_and_reserved, + 4, 4); +} + +uword +unformat_pg_tcp_header (unformat_input_t * input, va_list * args) +{ + pg_stream_t * s = va_arg (*args, pg_stream_t *); + pg_tcp_header_t * p; + u32 group_index; + + p = pg_create_edit_group (s, sizeof (p[0]), sizeof (tcp_header_t), + &group_index); + pg_tcp_header_init (p); + + /* Defaults. */ + pg_edit_set_fixed (&p->seq_number, 0); + pg_edit_set_fixed (&p->ack_number, 0); + + pg_edit_set_fixed (&p->data_offset_and_reserved, + sizeof (tcp_header_t) / sizeof (u32)); + + pg_edit_set_fixed (&p->window, 4096); + pg_edit_set_fixed (&p->urgent_pointer, 0); + +#define _(f) pg_edit_set_fixed (&p->f##_flag, 0); + foreach_tcp_flag +#undef _ + + p->checksum.type = PG_EDIT_UNSPECIFIED; + + if (! unformat (input, "TCP: %U -> %U", + unformat_pg_edit, + unformat_tcp_udp_port, &p->src, + unformat_pg_edit, + unformat_tcp_udp_port, &p->dst)) + goto error; + + /* Parse options. */ + while (1) + { + if (unformat (input, "window %U", + unformat_pg_edit, + unformat_pg_number, &p->window)) + ; + + else if (unformat (input, "checksum %U", + unformat_pg_edit, + unformat_pg_number, &p->checksum)) + ; + + /* Flags. */ +#define _(f) else if (unformat (input, #f)) pg_edit_set_fixed (&p->f##_flag, 1); + foreach_tcp_flag +#undef _ + + /* Can't parse input: try next protocol level. */ + else + break; + } + + { + ip_main_t * im = &ip_main; + u16 dst_port; + tcp_udp_port_info_t * pi; + + pi = 0; + if (p->dst.type == PG_EDIT_FIXED) + { + dst_port = pg_edit_get_value (&p->dst, PG_EDIT_LO); + pi = ip_get_tcp_udp_port_info (im, dst_port); + } + + if (pi && pi->unformat_pg_edit + && unformat_user (input, pi->unformat_pg_edit, s)) + ; + + else if (! unformat_user (input, unformat_pg_payload, s)) + goto error; + + if (p->checksum.type == PG_EDIT_UNSPECIFIED) + { + pg_edit_group_t * g = pg_stream_get_group (s, group_index); + g->edit_function = tcp_pg_edit_function; + g->edit_function_opaque = 0; + } + + return 1; + } + + error: + /* Free up any edits we may have added. */ + pg_free_edit_group (s); + return 0; +} + diff --git a/src/vnet/tcp/tcp_syn_filter4.c b/src/vnet/tcp/tcp_syn_filter4.c new file mode 100644 index 00000000000..c7605a30c98 --- /dev/null +++ b/src/vnet/tcp/tcp_syn_filter4.c @@ -0,0 +1,542 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vlib/vlib.h> +#include <vnet/vnet.h> +#include <vnet/pg/pg.h> +#include <vppinfra/error.h> +#include <vnet/feature/feature.h> +#include <vnet/ip/ip.h> +#include <vppinfra/xxhash.h> + +typedef struct +{ + f64 next_reset; + f64 reset_interval; + u8 *syn_counts; +} syn_filter4_runtime_t; + +typedef struct +{ + u32 next_index; + int not_a_syn; + u8 filter_value; +} syn_filter4_trace_t; + +/* packet trace format function */ +static u8 * +format_syn_filter4_trace (u8 * s, va_list * args) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); + syn_filter4_trace_t *t = va_arg (*args, syn_filter4_trace_t *); + + s = format (s, "SYN_FILTER4: next index %d, %s", + t->next_index, t->not_a_syn ? "not a syn" : "syn"); + if (t->not_a_syn == 0) + s = format (s, ", filter value %d\n", t->filter_value); + else + s = format (s, "\n"); + return s; +} + +static vlib_node_registration_t syn_filter4_node; + +#define foreach_syn_filter_error \ +_(THROTTLED, "TCP SYN packet throttle drops") \ +_(OK, "TCP SYN packets passed") + +typedef enum +{ +#define _(sym,str) SYN_FILTER_ERROR_##sym, + foreach_syn_filter_error +#undef _ + SYN_FILTER_N_ERROR, +} syn_filter_error_t; + +static char *syn_filter4_error_strings[] = { +#define _(sym,string) string, + foreach_syn_filter_error +#undef _ +}; + +typedef enum +{ + SYN_FILTER_NEXT_DROP, + SYN_FILTER_N_NEXT, +} syn_filter_next_t; + +extern vnet_feature_arc_registration_t vnet_feat_arc_ip4_local; + +static uword +syn_filter4_node_fn (vlib_main_t * vm, + vlib_node_runtime_t * node, vlib_frame_t * frame) +{ + u32 n_left_from, *from, *to_next; + syn_filter_next_t next_index; + u32 ok_syn_packets = 0; + vnet_feature_main_t *fm = &feature_main; + u8 arc_index = vnet_feat_arc_ip4_local.feature_arc_index; + vnet_feature_config_main_t *cm = &fm->feature_config_mains[arc_index]; + syn_filter4_runtime_t *rt = (syn_filter4_runtime_t *) node->runtime_data; + f64 now = vlib_time_now (vm); + /* Shut up spurious gcc warnings. */ + u8 *c0 = 0, *c1 = 0, *c2 = 0, *c3 = 0; + + from = vlib_frame_vector_args (frame); + n_left_from = frame->n_vectors; + next_index = node->cached_next_index; + + if (now > rt->next_reset) + { + memset (rt->syn_counts, 0, vec_len (rt->syn_counts)); + rt->next_reset = now + rt->reset_interval; + } + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from >= 8 && n_left_to_next >= 4) + { + u32 bi0, bi1, bi2, bi3; + vlib_buffer_t *b0, *b1, *b2, *b3; + u32 next0, next1, next2, next3; + ip4_header_t *ip0, *ip1, *ip2, *ip3; + tcp_header_t *tcp0, *tcp1, *tcp2, *tcp3; + u32 not_a_syn0 = 1, not_a_syn1 = 1, not_a_syn2 = 1, not_a_syn3 = 1; + u64 hash0, hash1, hash2, hash3; + + /* Prefetch next iteration. */ + { + vlib_buffer_t *p4, *p5, *p6, *p7; + + p4 = vlib_get_buffer (vm, from[4]); + p5 = vlib_get_buffer (vm, from[5]); + p6 = vlib_get_buffer (vm, from[6]); + p7 = vlib_get_buffer (vm, from[7]); + + vlib_prefetch_buffer_header (p4, LOAD); + vlib_prefetch_buffer_header (p5, LOAD); + vlib_prefetch_buffer_header (p6, LOAD); + vlib_prefetch_buffer_header (p7, LOAD); + + CLIB_PREFETCH (p4->data, CLIB_CACHE_LINE_BYTES, STORE); + CLIB_PREFETCH (p5->data, CLIB_CACHE_LINE_BYTES, STORE); + CLIB_PREFETCH (p6->data, CLIB_CACHE_LINE_BYTES, STORE); + CLIB_PREFETCH (p7->data, CLIB_CACHE_LINE_BYTES, STORE); + } + + /* speculatively enqueue b0 and b1 to the current next frame */ + to_next[0] = bi0 = from[0]; + to_next[1] = bi1 = from[1]; + to_next[2] = bi2 = from[2]; + to_next[3] = bi3 = from[3]; + from += 4; + to_next += 4; + n_left_from -= 4; + n_left_to_next -= 4; + + b0 = vlib_get_buffer (vm, bi0); + b1 = vlib_get_buffer (vm, bi1); + b2 = vlib_get_buffer (vm, bi2); + b3 = vlib_get_buffer (vm, bi3); + + vnet_get_config_data + (&cm->config_main, &b0->current_config_index, + &next0, 0 /* sizeof (c0[0]) */ ); + vnet_get_config_data + (&cm->config_main, &b1->current_config_index, + &next1, 0 /* sizeof (c0[0]) */ ); + vnet_get_config_data + (&cm->config_main, &b2->current_config_index, + &next2, 0 /* sizeof (c0[0]) */ ); + vnet_get_config_data + (&cm->config_main, &b3->current_config_index, + &next3, 0 /* sizeof (c0[0]) */ ); + + /* Not TCP? */ + ip0 = vlib_buffer_get_current (b0); + if (ip0->protocol != IP_PROTOCOL_TCP) + goto trace00; + + tcp0 = ip4_next_header (ip0); + /* + * Not a SYN? + * $$$$ hack: the TCP bitfield flags seem not to compile + * correct code. + */ + if (PREDICT_TRUE (!(tcp0->flags & 0x2))) + goto trace00; + + not_a_syn0 = 0; + hash0 = clib_xxhash ((u64) ip0->src_address.as_u32); + c0 = &rt->syn_counts[hash0 & (_vec_len (rt->syn_counts) - 1)]; + if (PREDICT_FALSE (*c0 >= 0x80)) + { + next0 = SYN_FILTER_NEXT_DROP; + b0->error = node->errors[SYN_FILTER_ERROR_THROTTLED]; + goto trace00; + } + *c0 += 1; + ok_syn_packets++; + + trace00: + if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE) + && (b0->flags & VLIB_BUFFER_IS_TRACED))) + { + syn_filter4_trace_t *t = + vlib_add_trace (vm, node, b0, sizeof (*t)); + t->not_a_syn = not_a_syn0; + t->next_index = next0; + t->filter_value = not_a_syn0 ? 0 : *c0; + } + + /* Not TCP? */ + ip1 = vlib_buffer_get_current (b1); + if (ip1->protocol != IP_PROTOCOL_TCP) + goto trace01; + + tcp1 = ip4_next_header (ip1); + /* + * Not a SYN? + * $$$$ hack: the TCP bitfield flags seem not to compile + * correct code. + */ + if (PREDICT_TRUE (!(tcp1->flags & 0x2))) + goto trace01; + + not_a_syn1 = 0; + hash1 = clib_xxhash ((u64) ip1->src_address.as_u32); + c1 = &rt->syn_counts[hash1 & (_vec_len (rt->syn_counts) - 1)]; + if (PREDICT_FALSE (*c1 >= 0x80)) + { + next1 = SYN_FILTER_NEXT_DROP; + b1->error = node->errors[SYN_FILTER_ERROR_THROTTLED]; + goto trace01; + } + *c1 += 1; + ok_syn_packets++; + + trace01: + if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE) + && (b1->flags & VLIB_BUFFER_IS_TRACED))) + { + syn_filter4_trace_t *t = + vlib_add_trace (vm, node, b1, sizeof (*t)); + t->not_a_syn = not_a_syn1; + t->next_index = next1; + t->filter_value = not_a_syn1 ? 0 : *c1; + } + + /* Not TCP? */ + ip2 = vlib_buffer_get_current (b2); + if (ip2->protocol != IP_PROTOCOL_TCP) + goto trace02; + + tcp2 = ip4_next_header (ip2); + /* + * Not a SYN? + * $$$$ hack: the TCP bitfield flags seem not to compile + * correct code. + */ + if (PREDICT_TRUE (!(tcp2->flags & 0x2))) + goto trace02; + + not_a_syn2 = 0; + hash2 = clib_xxhash ((u64) ip2->src_address.as_u32); + c2 = &rt->syn_counts[hash2 & (_vec_len (rt->syn_counts) - 1)]; + if (PREDICT_FALSE (*c2 >= 0x80)) + { + next2 = SYN_FILTER_NEXT_DROP; + b2->error = node->errors[SYN_FILTER_ERROR_THROTTLED]; + goto trace02; + } + *c2 += 1; + ok_syn_packets++; + + trace02: + if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE) + && (b2->flags & VLIB_BUFFER_IS_TRACED))) + { + syn_filter4_trace_t *t = + vlib_add_trace (vm, node, b2, sizeof (*t)); + t->not_a_syn = not_a_syn2; + t->next_index = next2; + t->filter_value = not_a_syn2 ? 0 : *c2; + } + + /* Not TCP? */ + ip3 = vlib_buffer_get_current (b3); + if (ip3->protocol != IP_PROTOCOL_TCP) + goto trace03; + + tcp3 = ip4_next_header (ip3); + /* + * Not a SYN? + * $$$$ hack: the TCP bitfield flags seem not to compile + * correct code. + */ + if (PREDICT_TRUE (!(tcp3->flags & 0x2))) + goto trace03; + + not_a_syn3 = 0; + hash3 = clib_xxhash ((u64) ip3->src_address.as_u32); + c3 = &rt->syn_counts[hash3 & (_vec_len (rt->syn_counts) - 1)]; + if (PREDICT_FALSE (*c3 >= 0x80)) + { + next3 = SYN_FILTER_NEXT_DROP; + b3->error = node->errors[SYN_FILTER_ERROR_THROTTLED]; + goto trace03; + } + *c3 += 1; + ok_syn_packets++; + + trace03: + if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE) + && (b3->flags & VLIB_BUFFER_IS_TRACED))) + { + syn_filter4_trace_t *t = + vlib_add_trace (vm, node, b3, sizeof (*t)); + t->not_a_syn = not_a_syn3; + t->next_index = next3; + t->filter_value = not_a_syn3 ? 0 : *c3; + } + vlib_validate_buffer_enqueue_x4 (vm, node, next_index, + to_next, n_left_to_next, + bi0, bi1, bi2, bi3, + next0, next1, next2, next3); + } + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + u32 next0; + ip4_header_t *ip0; + tcp_header_t *tcp0; + u32 not_a_syn0 = 1; + u32 hash0; + u8 *c0; + + /* speculatively enqueue b0 to the current next frame */ + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + + vnet_get_config_data + (&cm->config_main, &b0->current_config_index, + &next0, 0 /* sizeof (c0[0]) */ ); + + /* Not TCP? */ + ip0 = vlib_buffer_get_current (b0); + if (ip0->protocol != IP_PROTOCOL_TCP) + goto trace0; + + tcp0 = ip4_next_header (ip0); + /* + * Not a SYN? + * $$$$ hack: the TCP bitfield flags seem not to compile + * correct code. + */ + if (PREDICT_TRUE (!(tcp0->flags & 0x2))) + goto trace0; + + not_a_syn0 = 0; + hash0 = clib_xxhash ((u64) ip0->src_address.as_u32); + c0 = &rt->syn_counts[hash0 & (_vec_len (rt->syn_counts) - 1)]; + if (PREDICT_FALSE (*c0 >= 0x80)) + { + next0 = SYN_FILTER_NEXT_DROP; + b0->error = node->errors[SYN_FILTER_ERROR_THROTTLED]; + goto trace0; + } + *c0 += 1; + ok_syn_packets++; + + trace0: + + if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE) + && (b0->flags & VLIB_BUFFER_IS_TRACED))) + { + syn_filter4_trace_t *t = + vlib_add_trace (vm, node, b0, sizeof (*t)); + t->not_a_syn = not_a_syn0; + t->next_index = next0; + t->filter_value = not_a_syn0 ? 0 : *c0; + } + + /* verify speculative enqueue, maybe switch current next frame */ + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, + to_next, n_left_to_next, + bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + vlib_node_increment_counter (vm, syn_filter4_node.index, + SYN_FILTER_ERROR_OK, ok_syn_packets); + return frame->n_vectors; +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (syn_filter4_node, static) = +{ + .function = syn_filter4_node_fn, + .name = "syn-filter-4", + .vector_size = sizeof (u32), + .format_trace = format_syn_filter4_trace, + .type = VLIB_NODE_TYPE_INTERNAL, + + .runtime_data_bytes = sizeof (syn_filter4_runtime_t), + .n_errors = ARRAY_LEN(syn_filter4_error_strings), + .error_strings = syn_filter4_error_strings, + + .n_next_nodes = SYN_FILTER_N_NEXT, + + /* edit / add dispositions here */ + .next_nodes = { + [SYN_FILTER_NEXT_DROP] = "error-drop", + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (syn_filter4_node, syn_filter4_node_fn); + +/* *INDENT-OFF* */ +VNET_FEATURE_INIT (syn_filter_4, static) = +{ + .arc_name = "ip4-local", + .node_name = "syn-filter-4", + .runs_before = VNET_FEATURES("ip4-local-end-of-arc"), +}; +/* *INDENT-ON* */ + +int +syn_filter_enable_disable (u32 sw_if_index, int enable_disable) +{ + vnet_main_t *vnm = vnet_get_main (); + vnet_sw_interface_t *sw; + int rv = 0; + + /* Utterly wrong? */ + if (pool_is_free_index (vnm->interface_main.sw_interfaces, sw_if_index)) + return VNET_API_ERROR_INVALID_SW_IF_INDEX; + + /* Not a physical port? */ + sw = vnet_get_sw_interface (vnm, sw_if_index); + if (sw->type != VNET_SW_INTERFACE_TYPE_HARDWARE) + return VNET_API_ERROR_INVALID_SW_IF_INDEX; + + if (enable_disable) + { + vlib_main_t *vm = vlib_get_main (); + syn_filter4_runtime_t *rt; + + rt = vlib_node_get_runtime_data (vm, syn_filter4_node.index); + vec_validate (rt->syn_counts, 1023); + /* + * Given perfect disperson / optimal hashing results: + * Allow 128k (successful) syns/sec. 1024, buckets each of which + * absorb 128 syns before filtering. Reset table once a second. + * Reality bites, lets try resetting once every 100ms. + */ + rt->reset_interval = 0.1; /* reset interval in seconds */ + } + + rv = vnet_feature_enable_disable ("ip4-local", "syn-filter-4", + sw_if_index, enable_disable, 0, 0); + + return rv; +} + +static clib_error_t * +syn_filter_enable_disable_command_fn (vlib_main_t * vm, + unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + vnet_main_t *vnm = vnet_get_main (); + u32 sw_if_index = ~0; + int enable_disable = 1; + int rv; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "disable")) + enable_disable = 0; + else if (unformat (input, "%U", unformat_vnet_sw_interface, + vnm, &sw_if_index)) + ; + else + break; + } + + if (sw_if_index == ~0) + return clib_error_return (0, "Please specify an interface..."); + + rv = syn_filter_enable_disable (sw_if_index, enable_disable); + + switch (rv) + { + case 0: + break; + + case VNET_API_ERROR_INVALID_SW_IF_INDEX: + return clib_error_return + (0, "Invalid interface, only works on physical ports"); + break; + + case VNET_API_ERROR_UNIMPLEMENTED: + return clib_error_return (0, + "Device driver doesn't support redirection"); + break; + + case VNET_API_ERROR_INVALID_VALUE: + return clib_error_return (0, "feature arc not found"); + + case VNET_API_ERROR_INVALID_VALUE_2: + return clib_error_return (0, "feature node not found"); + + default: + return clib_error_return (0, "syn_filter_enable_disable returned %d", + rv); + } + return 0; +} + +/* *INDENT-OFF* */ +VLIB_CLI_COMMAND (sr_content_command, static) = +{ + .path = "ip syn filter", + .short_help = "ip syn filter <interface-name> [disable]", + .function = syn_filter_enable_disable_command_fn, +}; +/* *INDENT-ON* */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp_timer.h b/src/vnet/tcp/tcp_timer.h new file mode 100644 index 00000000000..fa25268c1bf --- /dev/null +++ b/src/vnet/tcp/tcp_timer.h @@ -0,0 +1,29 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#ifndef __included_tcp_timer_h__ +#define __included_tcp_timer_h__ + +#include <vppinfra/tw_timer_16t_2w_512sl.h> +#include <vppinfra/tw_timer_16t_1w_2048sl.h> + +#endif /* __included_tcp_timer_h__ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/udp/builtin_server.c b/src/vnet/udp/builtin_server.c new file mode 100644 index 00000000000..afa66ba4453 --- /dev/null +++ b/src/vnet/udp/builtin_server.c @@ -0,0 +1,239 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +/** @file + udp builtin server +*/ + +#include <vnet/udp/udp.h> +#include <vnet/session/session.h> +#include <vnet/session/application_interface.h> + +/** per-worker built-in server copy buffers */ +u8 **copy_buffers; + +static int +builtin_session_create_callback (stream_session_t * s) +{ + /* Simple version: declare session ready-to-go... */ + s->session_state = SESSION_STATE_READY; + return 0; +} + +static void +builtin_session_disconnect_callback (stream_session_t * s) +{ + stream_session_disconnect (s); +} + +static int +builtin_server_rx_callback (stream_session_t * s) +{ + svm_fifo_t *rx_fifo, *tx_fifo; + u32 this_transfer; + int actual_transfer; + u8 *my_copy_buffer; + session_fifo_event_t evt; + unix_shared_memory_queue_t *q; + + my_copy_buffer = copy_buffers[s->thread_index]; + rx_fifo = s->server_rx_fifo; + tx_fifo = s->server_tx_fifo; + + this_transfer = svm_fifo_max_enqueue (tx_fifo) + < svm_fifo_max_dequeue (rx_fifo) ? + svm_fifo_max_enqueue (tx_fifo) : svm_fifo_max_dequeue (rx_fifo); + + vec_validate (my_copy_buffer, this_transfer - 1); + _vec_len (my_copy_buffer) = this_transfer; + + actual_transfer = svm_fifo_dequeue_nowait (rx_fifo, 0, this_transfer, + my_copy_buffer); + ASSERT (actual_transfer == this_transfer); + actual_transfer = svm_fifo_enqueue_nowait (tx_fifo, 0, this_transfer, + my_copy_buffer); + + copy_buffers[s->thread_index] = my_copy_buffer; + + /* Fabricate TX event, send to ourselves */ + evt.fifo = tx_fifo; + evt.event_type = FIFO_EVENT_SERVER_TX; + /* $$$$ for event logging */ + evt.enqueue_length = actual_transfer; + evt.event_id = 0; + q = session_manager_get_vpp_event_queue (s->thread_index); + unix_shared_memory_queue_add (q, (u8 *) & evt, 0 /* do wait for mutex */ ); + + return 0; +} + +/* *INDENT-OFF* */ +static session_cb_vft_t builtin_server = { + .session_accept_callback = builtin_session_create_callback, + .session_disconnect_callback = builtin_session_disconnect_callback, + .builtin_server_rx_callback = builtin_server_rx_callback +}; +/* *INDENT-ON* */ + +static int +bind_builtin_uri_server (u8 * uri) +{ + vnet_bind_args_t _a, *a = &_a; + char segment_name[128]; + u32 segment_name_length; + int rv; + u64 options[16]; + + segment_name_length = ARRAY_LEN (segment_name); + + memset (a, 0, sizeof (*a)); + memset (options, 0, sizeof (options)); + + a->uri = (char *) uri; + a->api_client_index = ~0; /* built-in server */ + a->segment_name = segment_name; + a->segment_name_length = segment_name_length; + a->session_cb_vft = &builtin_server; + + options[SESSION_OPTIONS_ACCEPT_COOKIE] = 0x12345678; + options[SESSION_OPTIONS_SEGMENT_SIZE] = (2 << 30); /*$$$$ config / arg */ + a->options = options; + + rv = vnet_bind_uri (a); + + return rv; +} + +static int +unbind_builtin_uri_server (u8 * uri) +{ + int rv; + + rv = vnet_unbind_uri ((char *) uri, ~0 /* client_index */ ); + + return rv; +} + +static clib_error_t * +builtin_server_init (vlib_main_t * vm) +{ + vlib_thread_main_t *vtm = vlib_get_thread_main (); + u32 num_threads; + + num_threads = 1 /* main thread */ + vtm->n_threads; + + vec_validate (copy_buffers, num_threads - 1); + return 0; +} + +VLIB_INIT_FUNCTION (builtin_server_init); + +static clib_error_t * +builtin_uri_bind_command_fn (vlib_main_t * vm, + unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + u8 *uri = 0; + int rv; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "uri %s", &uri)) + ; + else + break; + } + + if (uri == 0) + return clib_error_return (0, "uri to bind not specified..."); + + rv = bind_builtin_uri_server (uri); + + vec_free (uri); + + switch (rv) + { + case 0: + break; + + default: + return clib_error_return (0, "bind_uri_server returned %d", rv); + break; + } + + return 0; +} + +/* *INDENT-OFF* */ +VLIB_CLI_COMMAND (builtin_uri_bind_command, static) = +{ + .path = "builtin uri bind", + .short_help = "builtin uri bind", + .function = builtin_uri_bind_command_fn, +}; +/* *INDENT-ON* */ + +static clib_error_t * +builtin_uri_unbind_command_fn (vlib_main_t * vm, + unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + u8 *uri = 0; + int rv; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "uri %s", &uri)) + ; + else + break; + } + + if (uri == 0) + return clib_error_return (0, "uri to unbind not specified..."); + + rv = unbind_builtin_uri_server (uri); + + vec_free (uri); + + switch (rv) + { + case 0: + break; + + default: + return clib_error_return (0, "unbind_uri_server returned %d", rv); + break; + } + + return 0; +} + +/* *INDENT-OFF* */ +VLIB_CLI_COMMAND (builtin_uri_unbind_command, static) = +{ + .path = "builtin uri unbind", + .short_help = "builtin uri unbind", + .function = builtin_uri_unbind_command_fn, +}; +/* *INDENT-ON* */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/udp/udp.c b/src/vnet/udp/udp.c new file mode 100644 index 00000000000..9e740466637 --- /dev/null +++ b/src/vnet/udp/udp.c @@ -0,0 +1,342 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +/** @file + udp state machine, etc. +*/ + +#include <vnet/udp/udp.h> +#include <vnet/session/session.h> +#include <vnet/dpo/load_balance.h> +#include <vnet/fib/ip4_fib.h> + +udp_uri_main_t udp_uri_main; + +u32 +udp_session_bind_ip4 (vlib_main_t * vm, u32 session_index, + ip46_address_t * ip, u16 port_number_host_byte_order) +{ + udp_uri_main_t *um = vnet_get_udp_main (); + udp_connection_t *listener; + + pool_get (um->udp_listeners, listener); + memset (listener, 0, sizeof (udp_connection_t)); + listener->c_lcl_port = clib_host_to_net_u16 (port_number_host_byte_order); + listener->c_lcl_ip4.as_u32 = ip->ip4.as_u32; + listener->c_proto = SESSION_TYPE_IP4_UDP; + udp_register_dst_port (um->vlib_main, port_number_host_byte_order, + udp4_uri_input_node.index, 1 /* is_ipv4 */ ); + return 0; +} + +u32 +udp_session_bind_ip6 (vlib_main_t * vm, u32 session_index, + ip46_address_t * ip, u16 port_number_host_byte_order) +{ + udp_uri_main_t *um = vnet_get_udp_main (); + udp_connection_t *listener; + + pool_get (um->udp_listeners, listener); + listener->c_lcl_port = clib_host_to_net_u16 (port_number_host_byte_order); + clib_memcpy (&listener->c_lcl_ip6, &ip->ip6, sizeof (ip6_address_t)); + listener->c_proto = SESSION_TYPE_IP6_UDP; + udp_register_dst_port (um->vlib_main, port_number_host_byte_order, + udp4_uri_input_node.index, 0 /* is_ipv4 */ ); + return 0; +} + +u32 +udp_session_unbind_ip4 (vlib_main_t * vm, u32 listener_index) +{ + udp_connection_t *listener; + listener = udp_listener_get (listener_index); + + /* deregister the udp_local mapping */ + udp_unregister_dst_port (vm, listener->c_lcl_port, 1 /* is_ipv4 */ ); + return 0; +} + +u32 +udp_session_unbind_ip6 (vlib_main_t * vm, u32 listener_index) +{ + udp_connection_t *listener; + + listener = udp_listener_get (listener_index); + + /* deregister the udp_local mapping */ + udp_unregister_dst_port (vm, listener->c_lcl_port, 0 /* is_ipv4 */ ); + return 0; +} + +transport_connection_t * +udp_session_get_listener (u32 listener_index) +{ + udp_connection_t *us; + + us = udp_listener_get (listener_index); + return &us->connection; +} + +u32 +udp_push_header (transport_connection_t * tconn, vlib_buffer_t * b) +{ + udp_connection_t *us; + u8 *data; + udp_header_t *udp; + + us = (udp_connection_t *) tconn; + + if (tconn->is_ip4) + { + ip4_header_t *ip; + + data = vlib_buffer_get_current (b); + udp = (udp_header_t *) (data - sizeof (udp_header_t)); + ip = (ip4_header_t *) ((u8 *) udp - sizeof (ip4_header_t)); + + /* Build packet header, swap rx key src + dst fields */ + ip->src_address.as_u32 = us->c_lcl_ip4.as_u32; + ip->dst_address.as_u32 = us->c_rmt_ip4.as_u32; + ip->ip_version_and_header_length = 0x45; + ip->ttl = 254; + ip->protocol = IP_PROTOCOL_UDP; + ip->length = clib_host_to_net_u16 (b->current_length + sizeof (*udp)); + ip->checksum = ip4_header_checksum (ip); + + udp->src_port = us->c_lcl_port; + udp->dst_port = us->c_rmt_port; + udp->length = clib_host_to_net_u16 (b->current_length); + udp->checksum = 0; + + b->current_length = sizeof (*ip) + sizeof (*udp); + return SESSION_QUEUE_NEXT_IP4_LOOKUP; + } + else + { + vlib_main_t *vm = vlib_get_main (); + ip6_header_t *ip; + u16 payload_length; + int bogus = ~0; + + data = vlib_buffer_get_current (b); + udp = (udp_header_t *) (data - sizeof (udp_header_t)); + ip = (ip6_header_t *) ((u8 *) udp - sizeof (ip6_header_t)); + + /* Build packet header, swap rx key src + dst fields */ + clib_memcpy (&ip->src_address, &us->c_lcl_ip6, sizeof (ip6_address_t)); + clib_memcpy (&ip->dst_address, &us->c_rmt_ip6, sizeof (ip6_address_t)); + + ip->ip_version_traffic_class_and_flow_label = + clib_host_to_net_u32 (0x6 << 28); + + ip->hop_limit = 0xff; + ip->protocol = IP_PROTOCOL_UDP; + + payload_length = vlib_buffer_length_in_chain (vm, b); + payload_length -= sizeof (*ip); + + ip->payload_length = clib_host_to_net_u16 (payload_length); + + udp->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b, ip, &bogus); + ASSERT (!bogus); + + udp->src_port = us->c_lcl_port; + udp->dst_port = us->c_rmt_port; + udp->length = clib_host_to_net_u16 (b->current_length); + udp->checksum = 0; + + b->current_length = sizeof (*ip) + sizeof (*udp); + + return SESSION_QUEUE_NEXT_IP6_LOOKUP; + } +} + +transport_connection_t * +udp_session_get (u32 connection_index, u32 my_thread_index) +{ + udp_uri_main_t *um = vnet_get_udp_main (); + + udp_connection_t *us; + us = + pool_elt_at_index (um->udp_sessions[my_thread_index], connection_index); + return &us->connection; +} + +void +udp_session_close (u32 connection_index, u32 my_thread_index) +{ + udp_uri_main_t *um = vnet_get_udp_main (); + pool_put_index (um->udp_sessions[my_thread_index], connection_index); +} + +u8 * +format_udp_session_ip4 (u8 * s, va_list * args) +{ + u32 uci = va_arg (*args, u32); + u32 thread_index = va_arg (*args, u32); + udp_connection_t *u4; + + u4 = udp_connection_get (uci, thread_index); + + s = format (s, "[%s] %U:%d->%U:%d", "udp", format_ip4_address, + &u4->c_lcl_ip4, clib_net_to_host_u16 (u4->c_lcl_port), + format_ip4_address, &u4->c_rmt_ip4, + clib_net_to_host_u16 (u4->c_rmt_port)); + return s; +} + +u8 * +format_udp_session_ip6 (u8 * s, va_list * args) +{ + u32 uci = va_arg (*args, u32); + u32 thread_index = va_arg (*args, u32); + udp_connection_t *tc = udp_connection_get (uci, thread_index); + s = format (s, "[%s] %U:%d->%U:%d", "udp", format_ip6_address, + &tc->c_lcl_ip6, clib_net_to_host_u16 (tc->c_lcl_port), + format_ip6_address, &tc->c_rmt_ip6, + clib_net_to_host_u16 (tc->c_rmt_port)); + return s; +} + +u8 * +format_udp_listener_session_ip4 (u8 * s, va_list * args) +{ + u32 tci = va_arg (*args, u32); + udp_connection_t *tc = udp_listener_get (tci); + s = format (s, "[%s] %U:%d->%U:%d", "udp", format_ip4_address, + &tc->c_lcl_ip4, clib_net_to_host_u16 (tc->c_lcl_port), + format_ip4_address, &tc->c_rmt_ip4, + clib_net_to_host_u16 (tc->c_rmt_port)); + return s; +} + +u8 * +format_udp_listener_session_ip6 (u8 * s, va_list * args) +{ + u32 tci = va_arg (*args, u32); + udp_connection_t *tc = udp_listener_get (tci); + s = format (s, "[%s] %U:%d->%U:%d", "udp", format_ip6_address, + &tc->c_lcl_ip6, clib_net_to_host_u16 (tc->c_lcl_port), + format_ip6_address, &tc->c_rmt_ip6, + clib_net_to_host_u16 (tc->c_rmt_port)); + return s; +} + +u16 +udp_send_mss_uri (transport_connection_t * t) +{ + /* TODO figure out MTU of output interface */ + return 400; +} + +u32 +udp_send_space_uri (transport_connection_t * t) +{ + /* No constraint on TX window */ + return ~0; +} + +int +udp_open_connection (ip46_address_t * addr, u16 port) +{ + clib_warning ("Not implemented"); + return 0; +} + +/* *INDENT-OFF* */ +const static transport_proto_vft_t udp4_proto = { + .bind = udp_session_bind_ip4, + .open = udp_open_connection, + .unbind = udp_session_unbind_ip4, + .push_header = udp_push_header, + .get_connection = udp_session_get, + .get_listener = udp_session_get_listener, + .close = udp_session_close, + .send_mss = udp_send_mss_uri, + .send_space = udp_send_space_uri, + .format_connection = format_udp_session_ip4, + .format_listener = format_udp_listener_session_ip4 +}; + +const static transport_proto_vft_t udp6_proto = { + .bind = udp_session_bind_ip6, + .open = udp_open_connection, + .unbind = udp_session_unbind_ip6, + .push_header = udp_push_header, + .get_connection = udp_session_get, + .get_listener = udp_session_get_listener, + .close = udp_session_close, + .send_mss = udp_send_mss_uri, + .send_space = udp_send_space_uri, + .format_connection = format_udp_session_ip6, + .format_listener = format_udp_listener_session_ip6 +}; +/* *INDENT-ON* */ + +static clib_error_t * +udp_init (vlib_main_t * vm) +{ + udp_uri_main_t *um = vnet_get_udp_main (); + ip_main_t *im = &ip_main; + vlib_thread_main_t *tm = vlib_get_thread_main (); + u32 num_threads; + clib_error_t *error = 0; + ip_protocol_info_t *pi; + + um->vlib_main = vm; + um->vnet_main = vnet_get_main (); + + if ((error = vlib_call_init_function (vm, ip_main_init))) + return error; + if ((error = vlib_call_init_function (vm, ip4_lookup_init))) + return error; + if ((error = vlib_call_init_function (vm, ip6_lookup_init))) + return error; + + /* + * Registrations + */ + + /* IP registration */ + pi = ip_get_protocol_info (im, IP_PROTOCOL_UDP); + if (pi == 0) + return clib_error_return (0, "UDP protocol info AWOL"); + pi->format_header = format_udp_header; + pi->unformat_pg_edit = unformat_pg_udp_header; + + + /* Register as transport with URI */ + session_register_transport (SESSION_TYPE_IP4_UDP, &udp4_proto); + session_register_transport (SESSION_TYPE_IP6_UDP, &udp6_proto); + + /* + * Initialize data structures + */ + + num_threads = 1 /* main thread */ + tm->n_threads; + vec_validate (um->udp_sessions, num_threads - 1); + + return error; +} + +VLIB_INIT_FUNCTION (udp_init); + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/ip/udp.h b/src/vnet/udp/udp.h index bad58b5d440..7ab26ce96f0 100644 --- a/src/vnet/ip/udp.h +++ b/src/vnet/udp/udp.h @@ -1,7 +1,5 @@ /* - * ip/udp.h: udp protocol - * - * Copyright (c) 2013 Cisco and/or its affiliates. + * Copyright (c) 2017 Cisco and/or its affiliates. * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at: @@ -14,22 +12,67 @@ * See the License for the specific language governing permissions and * limitations under the License. */ - -#ifndef included_udp_h -#define included_udp_h +#ifndef __included_udp_h__ +#define __included_udp_h__ #include <vnet/vnet.h> -#include <vnet/ip/udp_packet.h> +#include <vnet/udp/udp_packet.h> #include <vnet/ip/ip.h> #include <vnet/ip/ip4.h> #include <vnet/ip/ip4_packet.h> #include <vnet/pg/pg.h> #include <vnet/ip/format.h> +#include <vnet/ip/ip.h> +#include <vnet/session/transport.h> + +typedef struct +{ + transport_connection_t connection; /** must be first */ + + /** ersatz MTU to limit fifo pushes to test data size */ + u32 mtu; +} udp_connection_t; + +typedef struct _udp_uri_main +{ + /* Per-worker thread udp connection pools */ + udp_connection_t **udp_sessions; + udp_connection_t *udp_listeners; + + /* convenience */ + vlib_main_t *vlib_main; + vnet_main_t *vnet_main; + ip4_main_t *ip4_main; + ip6_main_t *ip6_main; +} udp_uri_main_t; + +extern udp_uri_main_t udp_uri_main; +extern vlib_node_registration_t udp4_uri_input_node; + +always_inline udp_uri_main_t * +vnet_get_udp_main () +{ + return &udp_uri_main; +} + +always_inline udp_connection_t * +udp_connection_get (u32 conn_index, u32 thread_index) +{ + return pool_elt_at_index (udp_uri_main.udp_sessions[thread_index], + conn_index); +} + +always_inline udp_connection_t * +udp_listener_get (u32 conn_index) +{ + return pool_elt_at_index (udp_uri_main.udp_listeners, conn_index); +} + typedef enum { #define udp_error(n,s) UDP_ERROR_##n, -#include <vnet/ip/udp_error.def> +#include <vnet/udp/udp_error.def> #undef udp_error UDP_N_ERROR, } udp_error_t; @@ -123,6 +166,10 @@ void udp_register_dst_port (vlib_main_t * vm, udp_dst_port_t dst_port, u32 node_index, u8 is_ip4); +void +udp_unregister_dst_port (vlib_main_t * vm, + udp_dst_port_t dst_port, u8 is_ip4); + void udp_punt_unknown (vlib_main_t * vm, u8 is_ip4, u8 is_add); always_inline void @@ -304,8 +351,6 @@ ip_udp_encap_two (vlib_main_t * vm, vlib_buffer_t * b0, vlib_buffer_t * b1, } } -#endif /* included_udp_h */ - /* * fd.io coding-style-patch-verification: ON * @@ -313,3 +358,5 @@ ip_udp_encap_two (vlib_main_t * vm, vlib_buffer_t * b0, vlib_buffer_t * b1, * eval: (c-set-style "gnu") * End: */ + +#endif /* __included_udp_h__ */ diff --git a/src/vnet/ip/udp_error.def b/src/vnet/udp/udp_error.def index bfdae0acc77..bfdae0acc77 100644 --- a/src/vnet/ip/udp_error.def +++ b/src/vnet/udp/udp_error.def diff --git a/src/vnet/ip/udp_format.c b/src/vnet/udp/udp_format.c index abdf561e8c9..abdf561e8c9 100644 --- a/src/vnet/ip/udp_format.c +++ b/src/vnet/udp/udp_format.c diff --git a/src/vnet/udp/udp_input.c b/src/vnet/udp/udp_input.c new file mode 100644 index 00000000000..4d509335c32 --- /dev/null +++ b/src/vnet/udp/udp_input.c @@ -0,0 +1,314 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vlib/vlib.h> +#include <vnet/vnet.h> +#include <vnet/pg/pg.h> +#include <vnet/ip/ip.h> + +#include <vnet/udp/udp.h> +#include <vppinfra/hash.h> +#include <vppinfra/error.h> +#include <vppinfra/elog.h> + +#include <vnet/udp/udp_packet.h> + +#include <vlibmemory/api.h> +#include "../session/application_interface.h" + +vlib_node_registration_t udp4_uri_input_node; + +typedef struct +{ + u32 session; + u32 disposition; + u32 thread_index; +} udp4_uri_input_trace_t; + +/* packet trace format function */ +static u8 * +format_udp4_uri_input_trace (u8 * s, va_list * args) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); + udp4_uri_input_trace_t *t = va_arg (*args, udp4_uri_input_trace_t *); + + s = format (s, "UDP4_URI_INPUT: session %d, disposition %d, thread %d", + t->session, t->disposition, t->thread_index); + return s; +} + +typedef enum +{ + UDP4_URI_INPUT_NEXT_DROP, + UDP4_URI_INPUT_N_NEXT, +} udp4_uri_input_next_t; + +static char *udp4_uri_input_error_strings[] = { +#define _(sym,string) string, + foreach_session_input_error +#undef _ +}; + +static uword +udp4_uri_input_node_fn (vlib_main_t * vm, + vlib_node_runtime_t * node, vlib_frame_t * frame) +{ + u32 n_left_from, *from, *to_next; + udp4_uri_input_next_t next_index; + udp_uri_main_t *um = vnet_get_udp_main (); + session_manager_main_t *smm = vnet_get_session_manager_main (); + u32 my_thread_index = vm->cpu_index; + u8 my_enqueue_epoch; + u32 *session_indices_to_enqueue; + static u32 serial_number; + int i; + + my_enqueue_epoch = ++smm->current_enqueue_epoch[my_thread_index]; + + from = vlib_frame_vector_args (frame); + n_left_from = frame->n_vectors; + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + u32 next0 = UDP4_URI_INPUT_NEXT_DROP; + u32 error0 = SESSION_ERROR_ENQUEUED; + udp_header_t *udp0; + ip4_header_t *ip0; + stream_session_t *s0; + svm_fifo_t *f0; + u16 udp_len0; + u8 *data0; + + /* speculatively enqueue b0 to the current next frame */ + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + + /* udp_local hands us a pointer to the udp data */ + + data0 = vlib_buffer_get_current (b0); + udp0 = (udp_header_t *) (data0 - sizeof (*udp0)); + + /* $$$$ fixme: udp_local doesn't do ip options correctly anyhow */ + ip0 = (ip4_header_t *) (((u8 *) udp0) - sizeof (*ip0)); + s0 = 0; + + /* lookup session */ + s0 = stream_session_lookup4 (&ip0->dst_address, &ip0->src_address, + udp0->dst_port, udp0->src_port, + SESSION_TYPE_IP4_UDP, my_thread_index); + + /* no listener */ + if (PREDICT_FALSE (s0 == 0)) + { + error0 = SESSION_ERROR_NO_LISTENER; + goto trace0; + } + + f0 = s0->server_rx_fifo; + + /* established hit */ + if (PREDICT_TRUE (s0->session_state == SESSION_STATE_READY)) + { + udp_len0 = clib_net_to_host_u16 (udp0->length); + + if (PREDICT_FALSE (udp_len0 > svm_fifo_max_enqueue (f0))) + { + error0 = SESSION_ERROR_FIFO_FULL; + goto trace0; + } + + svm_fifo_enqueue_nowait (f0, 0 /* pid */ , + udp_len0 - sizeof (*udp0), + (u8 *) (udp0 + 1)); + + b0->error = node->errors[SESSION_ERROR_ENQUEUED]; + + /* We need to send an RX event on this fifo */ + if (s0->enqueue_epoch != my_enqueue_epoch) + { + s0->enqueue_epoch = my_enqueue_epoch; + + vec_add1 (smm->session_indices_to_enqueue_by_thread + [my_thread_index], + s0 - smm->sessions[my_thread_index]); + } + } + /* listener hit */ + else if (s0->session_state == SESSION_STATE_LISTENING) + { + udp_connection_t *us; + int rv; + + error0 = SESSION_ERROR_NOT_READY; + + /* + * create udp transport session + */ + pool_get (um->udp_sessions[my_thread_index], us); + + us->mtu = 1024; /* $$$$ policy */ + + us->c_lcl_ip4.as_u32 = ip0->dst_address.as_u32; + us->c_rmt_ip4.as_u32 = ip0->src_address.as_u32; + us->c_lcl_port = udp0->dst_port; + us->c_rmt_port = udp0->src_port; + us->c_proto = SESSION_TYPE_IP4_UDP; + us->c_c_index = us - um->udp_sessions[my_thread_index]; + + /* + * create stream session and attach the udp session to it + */ + rv = stream_session_accept (&us->connection, s0->session_index, + SESSION_TYPE_IP4_UDP, + 1 /*notify */ ); + if (rv) + error0 = rv; + + } + else + { + + error0 = SESSION_ERROR_NOT_READY; + goto trace0; + } + + trace0: + b0->error = node->errors[error0]; + + if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE) + && (b0->flags & VLIB_BUFFER_IS_TRACED))) + { + udp4_uri_input_trace_t *t = + vlib_add_trace (vm, node, b0, sizeof (*t)); + + t->session = ~0; + if (s0) + t->session = s0 - smm->sessions[my_thread_index]; + t->disposition = error0; + t->thread_index = my_thread_index; + } + + /* verify speculative enqueue, maybe switch current next frame */ + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, + to_next, n_left_to_next, + bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + /* Send enqueue events */ + + session_indices_to_enqueue = + smm->session_indices_to_enqueue_by_thread[my_thread_index]; + + for (i = 0; i < vec_len (session_indices_to_enqueue); i++) + { + session_fifo_event_t evt; + unix_shared_memory_queue_t *q; + stream_session_t *s0; + application_t *server0; + + /* Get session */ + s0 = pool_elt_at_index (smm->sessions[my_thread_index], + session_indices_to_enqueue[i]); + + /* Get session's server */ + server0 = application_get (s0->app_index); + + /* Built-in server? Deliver the goods... */ + if (server0->cb_fns.builtin_server_rx_callback) + { + server0->cb_fns.builtin_server_rx_callback (s0); + continue; + } + + /* Fabricate event */ + evt.fifo = s0->server_rx_fifo; + evt.event_type = FIFO_EVENT_SERVER_RX; + evt.event_id = serial_number++; + evt.enqueue_length = svm_fifo_max_dequeue (s0->server_rx_fifo); + + /* Add event to server's event queue */ + q = server0->event_queue; + + /* Don't block for lack of space */ + if (PREDICT_TRUE (q->cursize < q->maxsize)) + unix_shared_memory_queue_add (server0->event_queue, (u8 *) & evt, + 0 /* do wait for mutex */ ); + else + { + vlib_node_increment_counter (vm, udp4_uri_input_node.index, + SESSION_ERROR_FIFO_FULL, 1); + } + if (1) + { + ELOG_TYPE_DECLARE (e) = + { + .format = "evt-enqueue: id %d length %d",.format_args = "i4i4",}; + struct + { + u32 data[2]; + } *ed; + ed = ELOG_DATA (&vlib_global_main.elog_main, e); + ed->data[0] = evt.event_id; + ed->data[1] = evt.enqueue_length; + } + } + + vec_reset_length (session_indices_to_enqueue); + + smm->session_indices_to_enqueue_by_thread[my_thread_index] = + session_indices_to_enqueue; + + return frame->n_vectors; +} + +VLIB_REGISTER_NODE (udp4_uri_input_node) = +{ + .function = udp4_uri_input_node_fn,.name = "udp4-uri-input",.vector_size = + sizeof (u32),.format_trace = format_udp4_uri_input_trace,.type = + VLIB_NODE_TYPE_INTERNAL,.n_errors = + ARRAY_LEN (udp4_uri_input_error_strings),.error_strings = + udp4_uri_input_error_strings,.n_next_nodes = UDP4_URI_INPUT_N_NEXT, + /* edit / add dispositions here */ + .next_nodes = + { + [UDP4_URI_INPUT_NEXT_DROP] = "error-drop",} +,}; + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/ip/udp_local.c b/src/vnet/udp/udp_local.c index 13ab6e4fb32..6b239f73481 100644 --- a/src/vnet/ip/udp_local.c +++ b/src/vnet/udp/udp_local.c @@ -17,8 +17,8 @@ #include <vlib/vlib.h> #include <vnet/pg/pg.h> -#include <vnet/ip/udp.h> -#include <vnet/ip/udp_packet.h> +#include <vnet/udp/udp.h> +#include <vnet/udp/udp_packet.h> #include <vppinfra/sparse_vec.h> udp_main_t udp_main; @@ -528,6 +528,27 @@ udp_register_dst_port (vlib_main_t * vm, } void +udp_unregister_dst_port (vlib_main_t * vm, udp_dst_port_t dst_port, u8 is_ip4) +{ + udp_main_t *um = &udp_main; + udp_dst_port_info_t *pi; + udp_input_runtime_t *rt; + u16 *n; + + pi = udp_get_dst_port_info (um, dst_port, is_ip4); + /* Not registered? Fagedaboudit */ + if (!pi) + return; + + /* Kill the mapping. Don't bother killing the pi, it may be back. */ + rt = vlib_node_get_runtime_data + (vm, is_ip4 ? udp4_input_node.index : udp6_input_node.index); + n = sparse_vec_validate (rt->next_by_dst_port, + clib_host_to_net_u16 (dst_port)); + n[0] = SPARSE_VEC_INVALID_INDEX; +} + +void udp_punt_unknown (vlib_main_t * vm, u8 is_ip4, u8 is_add) { udp_input_runtime_t *rt; diff --git a/src/vnet/ip/udp_packet.h b/src/vnet/udp/udp_packet.h index beea3059246..beea3059246 100644 --- a/src/vnet/ip/udp_packet.h +++ b/src/vnet/udp/udp_packet.h diff --git a/src/vnet/ip/udp_pg.c b/src/vnet/udp/udp_pg.c index c9d8d38ca4a..c9d8d38ca4a 100644 --- a/src/vnet/ip/udp_pg.c +++ b/src/vnet/udp/udp_pg.c diff --git a/src/vnet/vnet_all_api_h.h b/src/vnet/vnet_all_api_h.h index 142acedc73f..c4075db65c3 100644 --- a/src/vnet/vnet_all_api_h.h +++ b/src/vnet/vnet_all_api_h.h @@ -51,6 +51,7 @@ #include <vnet/lisp-cp/lisp.api.h> #include <vnet/lisp-gpe/lisp_gpe.api.h> #include <vnet/lisp-cp/one.api.h> +#include <vnet/session/session.api.h> #include <vnet/mpls/mpls.api.h> #include <vnet/sr/sr.api.h> #include <vnet/classify/classify.api.h> diff --git a/src/vnet/vxlan-gpe/vxlan_gpe.h b/src/vnet/vxlan-gpe/vxlan_gpe.h index 1b4bc44e7bb..e768d2309ff 100644 --- a/src/vnet/vxlan-gpe/vxlan_gpe.h +++ b/src/vnet/vxlan-gpe/vxlan_gpe.h @@ -29,7 +29,7 @@ #include <vnet/vxlan-gpe/vxlan_gpe_packet.h> #include <vnet/ip/ip4_packet.h> #include <vnet/ip/ip6_packet.h> -#include <vnet/ip/udp.h> +#include <vnet/udp/udp.h> /** * @brief VXLAN GPE header struct diff --git a/src/vnet/vxlan/vxlan.h b/src/vnet/vxlan/vxlan.h index adfa3a8e3c8..dca1cd1285d 100644 --- a/src/vnet/vxlan/vxlan.h +++ b/src/vnet/vxlan/vxlan.h @@ -26,7 +26,7 @@ #include <vnet/vxlan/vxlan_packet.h> #include <vnet/ip/ip4_packet.h> #include <vnet/ip/ip6_packet.h> -#include <vnet/ip/udp.h> +#include <vnet/udp/udp.h> #include <vnet/dpo/dpo.h> #include <vnet/adj/adj_types.h> diff --git a/src/vpp/api/vpe.api b/src/vpp/api/vpe.api index 24f482936d0..2d6e4f3716c 100644 --- a/src/vpp/api/vpe.api +++ b/src/vpp/api/vpe.api @@ -38,6 +38,7 @@ * IPSEC-GRE APIs: see .../src/vnet/ipsec-gre/{ipsec_gre.api, ipsec_gre_api.c} * LISP APIs: see .../src/vnet/lisp/{lisp.api, lisp_api.c} * LISP-GPE APIs: see .../src/vnet/lisp-gpe/{lisp_gpe.api, lisp_gpe_api.c} + * SESSION APIs: .../vnet/session/{session.api session_api.c} * MPLS APIs: see .../src/vnet/mpls/{mpls.api, mpls_api.c} * SR APIs: see .../src/vnet/sr/{sr.api, sr_api.c} * DPDK APIs: see ... /src/vnet/devices/dpdk/{dpdk.api, dpdk_api.c} diff --git a/src/vppinfra.am b/src/vppinfra.am index 8d375958702..4b9f0c29447 100644 --- a/src/vppinfra.am +++ b/src/vppinfra.am @@ -157,7 +157,9 @@ nobase_include_HEADERS = \ vppinfra/asm_mips.h \ vppinfra/asm_x86.h \ vppinfra/bihash_8_8.h \ + vppinfra/bihash_16_8.h \ vppinfra/bihash_24_8.h \ + vppinfra/bihash_48_8.h \ vppinfra/bihash_template.h \ vppinfra/bihash_template.c \ vppinfra/bitmap.h \ @@ -206,6 +208,7 @@ nobase_include_HEADERS = \ vppinfra/timer.h \ vppinfra/tw_timer_2t_1w_2048sl.h \ vppinfra/tw_timer_16t_2w_512sl.h \ + vppinfra/tw_timer_16t_1w_2048sl.h \ vppinfra/tw_timer_template.h \ vppinfra/tw_timer_template.c \ vppinfra/types.h \ @@ -261,6 +264,8 @@ CLIB_CORE = \ vppinfra/tw_timer_2t_1w_2048sl.c \ vppinfra/tw_timer_16t_2w_512sl.h \ vppinfra/tw_timer_16t_2w_512sl.c \ + vppinfra/tw_timer_16t_1w_2048sl.h \ + vppinfra/tw_timer_16t_1w_2048sl.c \ vppinfra/unformat.c \ vppinfra/vec.c \ vppinfra/vector.c \ diff --git a/src/vppinfra/bihash_16_8.h b/src/vppinfra/bihash_16_8.h new file mode 100644 index 00000000000..ce80f70e50a --- /dev/null +++ b/src/vppinfra/bihash_16_8.h @@ -0,0 +1,103 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#undef BIHASH_TYPE + +#define BIHASH_TYPE _16_8 +#define BIHASH_KVP_PER_PAGE 4 + +#ifndef __included_bihash_16_8_h__ +#define __included_bihash_16_8_h__ + +#include <vppinfra/heap.h> +#include <vppinfra/format.h> +#include <vppinfra/pool.h> +#include <vppinfra/xxhash.h> + +typedef struct +{ + u64 key[2]; + u64 value; +} clib_bihash_kv_16_8_t; + +static inline int +clib_bihash_is_free_16_8 (clib_bihash_kv_16_8_t * v) +{ + /* Free values are memset to 0xff, check a bit... */ + if (v->key[0] == ~0ULL && v->value == ~0ULL) + return 1; + return 0; +} + +#if __SSE4_2__ +#ifndef __defined_crc_u32__ +#define __defined_crc_u32__ +static inline u32 +crc_u32 (u32 data, u32 value) +{ + __asm__ volatile ("crc32l %[data], %[value];":[value] "+r" (value):[data] + "rm" (data)); + return value; +} +#endif /* __defined_crc_u32__ */ + +static inline u64 +clib_bihash_hash_16_8 (clib_bihash_kv_16_8_t * v) +{ + u32 *dp = (u32 *) & v->key[0]; + u32 value = 0; + + value = crc_u32 (dp[0], value); + value = crc_u32 (dp[1], value); + value = crc_u32 (dp[2], value); + value = crc_u32 (dp[3], value); + + return value; +} +#else +static inline u64 +clib_bihash_hash_16_8 (clib_bihash_kv_16_8_t * v) +{ + u64 tmp = v->key[0] ^ v->key[1]; + return clib_xxhash (tmp); +} +#endif + +static inline u8 * +format_bihash_kvp_16_8 (u8 * s, va_list * args) +{ + clib_bihash_kv_16_8_t *v = va_arg (*args, clib_bihash_kv_16_8_t *); + + s = format (s, "key %llu %llu value %llu", v->key[0], v->key[1], v->value); + return s; +} + +static inline int +clib_bihash_key_compare_16_8 (u64 * a, u64 * b) +{ + return ((a[0] ^ b[0]) | (a[1] ^ b[1])) == 0; +} + +#undef __included_bihash_template_h__ +#include <vppinfra/bihash_template.h> + +#endif /* __included_bihash_16_8_h__ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vppinfra/bihash_48_8.h b/src/vppinfra/bihash_48_8.h new file mode 100644 index 00000000000..1a6e76913ed --- /dev/null +++ b/src/vppinfra/bihash_48_8.h @@ -0,0 +1,116 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#undef BIHASH_TYPE + +#define BIHASH_TYPE _48_8 +#define BIHASH_KVP_PER_PAGE 4 + +#ifndef __included_bihash_48_8_h__ +#define __included_bihash_48_8_h__ + +#include <vppinfra/heap.h> +#include <vppinfra/format.h> +#include <vppinfra/pool.h> +#include <vppinfra/xxhash.h> + +typedef struct +{ + u64 key[6]; + u64 value; +} clib_bihash_kv_48_8_t; + +static inline int +clib_bihash_is_free_48_8 (const clib_bihash_kv_48_8_t * v) +{ + /* Free values are memset to 0xff, check a bit... */ + if (v->key[0] == ~0ULL && v->value == ~0ULL) + return 1; + return 0; +} + +#if __SSE4_2__ +#ifndef __defined_crc_u32__ +#define __defined_crc_u32__ +static inline u32 +crc_u32 (u32 data, u32 value) +{ + __asm__ volatile ("crc32l %[data], %[value];":[value] "+r" (value):[data] + "rm" (data)); + return value; +} +#endif /* __defined_crc_u32__ */ + +static inline u64 +clib_bihash_hash_48_8 (const clib_bihash_kv_48_8_t * v) +{ + const u32 *dp = (const u32 *) &v->key[0]; + u32 value = 0; + + value = crc_u32 (dp[0], value); + value = crc_u32 (dp[1], value); + value = crc_u32 (dp[2], value); + value = crc_u32 (dp[3], value); + value = crc_u32 (dp[4], value); + value = crc_u32 (dp[5], value); + value = crc_u32 (dp[6], value); + value = crc_u32 (dp[7], value); + value = crc_u32 (dp[8], value); + value = crc_u32 (dp[9], value); + value = crc_u32 (dp[10], value); + value = crc_u32 (dp[11], value); + + return value; +} +#else +static inline u64 +clib_bihash_hash_48_8 (const clib_bihash_kv_48_8_t * v) +{ + u64 tmp = v->key[0] ^ v->key[1] ^ v->key[2] ^ v->key[3] ^ v->key[4] + ^ v->key[5]; + return clib_xxhash (tmp); +} +#endif + +static inline u8 * +format_bihash_kvp_48_8 (u8 * s, va_list * args) +{ + clib_bihash_kv_48_8_t *v = va_arg (*args, clib_bihash_kv_48_8_t *); + + s = format (s, "key %llu %llu %llu %llu %llu %llu value %llu", v->key[0], + v->key[1], v->key[2], v->key[3], v->key[4], v->key[5], + v->value); + return s; +} + +static inline int +clib_bihash_key_compare_48_8 (const u64 * a, const u64 * b) +{ + return ((a[0] ^ b[0]) | (a[1] ^ b[1]) | (a[2] ^ b[2]) | (a[3] ^ b[3]) + | (a[4] ^ b[4]) | (a[5] ^ b[5])) == 0; +} + +#undef __included_bihash_template_h__ +#include <vppinfra/bihash_template.h> + +#endif /* __included_bihash_48_8_h__ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vppinfra/tw_timer_16t_1w_2048sl.c b/src/vppinfra/tw_timer_16t_1w_2048sl.c new file mode 100644 index 00000000000..3f342045841 --- /dev/null +++ b/src/vppinfra/tw_timer_16t_1w_2048sl.c @@ -0,0 +1,26 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vppinfra/error.h> +#include "tw_timer_16t_1w_2048sl.h" +#include "tw_timer_template.c" + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vppinfra/tw_timer_16t_1w_2048sl.h b/src/vppinfra/tw_timer_16t_1w_2048sl.h new file mode 100644 index 00000000000..685ac31ed07 --- /dev/null +++ b/src/vppinfra/tw_timer_16t_1w_2048sl.h @@ -0,0 +1,46 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef __included_tw_timer_16t_2w_512sl_h__ +#define __included_tw_timer_16t_2w_512sl_h__ + +/* ... So that a client app can create multiple wheel geometries */ +#undef TW_TIMER_WHEELS +#undef TW_SLOTS_PER_RING +#undef TW_RING_SHIFT +#undef TW_RING_MASK +#undef TW_TIMERS_PER_OBJECT +#undef LOG2_TW_TIMERS_PER_OBJECT +#undef TW_SUFFIX + +#define TW_TIMER_WHEELS 1 +#define TW_SLOTS_PER_RING 2048 +#define TW_RING_SHIFT 11 +#define TW_RING_MASK (TW_SLOTS_PER_RING -1) +#define TW_TIMERS_PER_OBJECT 16 +#define LOG2_TW_TIMERS_PER_OBJECT 4 +#define TW_SUFFIX _16t_1w_2048sl + +#include <vppinfra/tw_timer_template.h> + +#endif /* __included_tw_timer_16t_2w_512sl_h__ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ |