From 62f9cdd82c52dc05cb89a742d21aba013ce526d4 Mon Sep 17 00:00:00 2001 From: Hongjun Ni Date: Tue, 4 Jul 2017 20:11:57 +0800 Subject: Add PPPoE Plugin Supports 64K PPPoE sessions This plugin adds three graph nodes: 1) pppoe-input for PPPoE decapsulation 2) pppoe-encap for PPPoE encapsulation 3) pppoe-tap-dispatch for control plane process Below is the configuration to make PPPoE CP and DP work: vim /etc/vpp/startup.conf tuntap { enable ethernet name newtap } create pppoe tap tap-if-index 1 //Configure it after a subscriber's PPPoE discovery and PPP link establishment succeeds: create pppoe session client-ip 100.1.2.1 session-id 1 client-mac 00:11:01:00:00:01 show pppoe fib show pppoe session Change-Id: I73e724b6bf7c3e4181a9914c5752da1fa72d7e60 Signed-off-by: Hongjun Ni --- src/plugins/pppoe/pppoe.h | 295 ++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 295 insertions(+) create mode 100644 src/plugins/pppoe/pppoe.h (limited to 'src/plugins/pppoe/pppoe.h') diff --git a/src/plugins/pppoe/pppoe.h b/src/plugins/pppoe/pppoe.h new file mode 100644 index 00000000..37d628eb --- /dev/null +++ b/src/plugins/pppoe/pppoe.h @@ -0,0 +1,295 @@ +/* + *------------------------------------------------------------------ + * Copyright (c) 2017 Intel 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 _PPPOE_H +#define _PPPOE_H + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + + +typedef struct +{ + u8 ver_type; + u8 code; + u16 session_id; + u16 length; + u16 ppp_proto; +} pppoe_header_t; + +#define PPPOE_VER_TYPE 0x11 +#define PPPOE_PADS 0x65 + +typedef struct +{ + /* Rewrite string */ + u8 *rewrite; + + /* pppoe session_id in HOST byte order */ + u16 session_id; + + /* session client addresses */ + ip46_address_t client_ip; + + /* the index of tx interface for pppoe encaped packet */ + u32 encap_if_index; + + /** FIB indices - inner IP packet lookup here */ + u32 decap_fib_index; + + u8 local_mac[6]; + u8 client_mac[6]; + + /* vnet intfc index */ + u32 sw_if_index; + u32 hw_if_index; + +} pppoe_session_t; + +#define foreach_pppoe_input_next \ +_(DROP, "error-drop") \ +_(IP4_INPUT, "ip4-input") \ +_(IP6_INPUT, "ip6-input" ) \ +_(CP_INPUT, "pppoe-tap-dispatch" ) \ + +typedef enum +{ +#define _(s,n) PPPOE_INPUT_NEXT_##s, + foreach_pppoe_input_next +#undef _ + PPPOE_INPUT_N_NEXT, +} pppoe_input_next_t; + +typedef enum +{ +#define pppoe_error(n,s) PPPOE_ERROR_##n, +#include +#undef pppoe_error + PPPOE_N_ERROR, +} pppoe_input_error_t; + + +#define MTU 1500 +#define MTU_BUFFERS ((MTU + VLIB_BUFFER_DATA_SIZE - 1) / VLIB_BUFFER_DATA_SIZE) +#define NUM_BUFFERS_TO_ALLOC 32 + +/* + * The size of pppoe session table + */ +#define PPPOE_NUM_BUCKETS (128 * 1024) +#define PPPOE_MEMORY_SIZE (16<<20) + +/* *INDENT-OFF* */ +/* + * The PPPoE key is the mac address and session ID + */ +typedef struct +{ + union + { + struct + { + u16 session_id; + u8 mac[6]; + } fields; + struct + { + u32 w0; + u32 w1; + } words; + u64 raw; + }; +} pppoe_entry_key_t; +/* *INDENT-ON* */ + +/* *INDENT-OFF* */ +/* + * The PPPoE entry results + */ +typedef struct +{ + union + { + struct + { + u32 sw_if_index; + + u32 session_index; + + } fields; + u64 raw; + }; +} pppoe_entry_result_t; +/* *INDENT-ON* */ + +typedef struct +{ + /* For DP: vector of encap session instances, */ + pppoe_session_t *sessions; + + /* For CP: vector of CP path */ + BVT (clib_bihash) session_table; + + /* Free vlib hw_if_indices */ + u32 *free_pppoe_session_hw_if_indices; + + /* Mapping from sw_if_index to session index */ + u32 *session_index_by_sw_if_index; + + /* used for pppoe cp path */ + u32 tap_if_index; + + /* API message ID base */ + u16 msg_id_base; + + /* convenience */ + vlib_main_t *vlib_main; + vnet_main_t *vnet_main; + +} pppoe_main_t; + +extern pppoe_main_t pppoe_main; + +extern vlib_node_registration_t pppoe_input_node; +extern vlib_node_registration_t pppoe_encap_node; +extern vlib_node_registration_t pppoe_tap_dispatch_node; + +u8 *format_pppoe_encap_trace (u8 * s, va_list * args); + +typedef struct +{ + u8 is_add; + u8 is_ip6; + u16 session_id; + ip46_address_t client_ip; + u32 encap_if_index; + u32 decap_fib_index; + u8 local_mac[6]; + u8 client_mac[6]; +} vnet_pppoe_add_del_session_args_t; + +int vnet_pppoe_add_del_session + (vnet_pppoe_add_del_session_args_t * a, u32 * sw_if_indexp); + +typedef struct +{ + u8 is_add; + u32 client_if_index; + u32 tap_if_index; +} vnet_pppoe_add_del_tap_args_t; + +always_inline u64 +pppoe_make_key (u8 * mac_address, u16 session_id) +{ + u64 temp; + + /* + * The mac address in memory is A:B:C:D:E:F + * The session_id in register is H:L + */ +#if CLIB_ARCH_IS_LITTLE_ENDIAN + /* + * Create the in-register key as F:E:D:C:B:A:H:L + * In memory the key is L:H:A:B:C:D:E:F + */ + temp = *((u64 *) (mac_address)) << 16; + temp = (temp & ~0xffff) | (u64) (session_id); +#else + /* + * Create the in-register key as H:L:A:B:C:D:E:F + * In memory the key is H:L:A:B:C:D:E:F + */ + temp = *((u64 *) (mac_address)) >> 16; + temp = temp | (((u64) session_id) << 48); +#endif + + return temp; +} + +static_always_inline void +pppoe_lookup_1 (BVT (clib_bihash) * session_table, + pppoe_entry_key_t * cached_key, + pppoe_entry_result_t * cached_result, + u8 * mac0, + u16 session_id0, + pppoe_entry_key_t * key0, + u32 * bucket0, pppoe_entry_result_t * result0) +{ + /* set up key */ + key0->raw = pppoe_make_key (mac0, session_id0); + *bucket0 = ~0; + + if (key0->raw == cached_key->raw) + { + /* Hit in the one-entry cache */ + result0->raw = cached_result->raw; + } + else + { + /* Do a regular session table lookup */ + BVT (clib_bihash_kv) kv; + + kv.key = key0->raw; + kv.value = ~0ULL; + BV (clib_bihash_search_inline) (session_table, &kv); + result0->raw = kv.value; + + /* Update one-entry cache */ + cached_key->raw = key0->raw; + cached_result->raw = result0->raw; + } +} + +static_always_inline void +pppoe_update_1 (BVT (clib_bihash) * session_table, + u8 * mac0, + u16 session_id0, + pppoe_entry_key_t * key0, + u32 * bucket0, pppoe_entry_result_t * result0) +{ + /* set up key */ + key0->raw = pppoe_make_key (mac0, session_id0); + *bucket0 = ~0; + + /* Update the entry */ + BVT (clib_bihash_kv) kv; + kv.key = key0->raw; + kv.value = result0->raw; + BV (clib_bihash_add_del) (session_table, &kv, 1 /* is_add */ ); + +} +#endif /* _PPPOE_H */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ -- cgit 1.2.3-korg From 43161a873375ddf156cf6fbe8764bfc206b38fa0 Mon Sep 17 00:00:00 2001 From: Neale Ranns Date: Sat, 12 Aug 2017 02:12:00 -0700 Subject: PPPoE usses a midchain adjacency stack on an interface-tx DPO 1) introduce an interface-tx DPO. This is a simple wrapper around a sw_if_index. enhance DPO stacking functions to allow per-instance next-nodes and hence allow children to stack onto the interface per-instance tx node and not on 'interface-output'. 2) update PPPoE code to use ta midchain stack on a interface-tx DPO of the encap-interface. This remove the need for pppoe_encap node (which is replaced by the adj-midchain-tx) and interface-output node is no longer used (see above). Since PPPoE encap node is no longer needed, the PPPoE seesion does not need to be retrieved in the data-path, hence the cahce misses are removed. Change-Id: Id8b40f53daa14889a9c51d802e14fed7fba4399a Signed-off-by: Neale Ranns --- src/plugins/pppoe.am | 23 +-- src/plugins/pppoe/pppoe.c | 144 ++++++++++--- src/plugins/pppoe/pppoe.h | 6 - src/plugins/pppoe/pppoe_encap.c | 384 ---------------------------------- src/vnet.am | 3 +- src/vnet/adj/adj_midchain.c | 126 ++++++++---- src/vnet/dpo/dpo.c | 107 ++++++---- src/vnet/dpo/dpo.h | 19 +- src/vnet/dpo/interface_dpo.c | 446 ---------------------------------------- src/vnet/dpo/interface_dpo.h | 67 ------ src/vnet/dpo/interface_rx_dpo.c | 445 +++++++++++++++++++++++++++++++++++++++ src/vnet/dpo/interface_rx_dpo.h | 69 +++++++ src/vnet/dpo/interface_tx_dpo.c | 92 +++++++++ src/vnet/dpo/interface_tx_dpo.h | 33 +++ src/vnet/fib/fib_path.c | 14 +- src/vnet/fib/fib_test.c | 14 +- test/test_pppoe.py | 3 +- 17 files changed, 952 insertions(+), 1043 deletions(-) delete mode 100644 src/plugins/pppoe/pppoe_encap.c delete mode 100644 src/vnet/dpo/interface_dpo.c delete mode 100644 src/vnet/dpo/interface_dpo.h create mode 100644 src/vnet/dpo/interface_rx_dpo.c create mode 100644 src/vnet/dpo/interface_rx_dpo.h create mode 100644 src/vnet/dpo/interface_tx_dpo.c create mode 100644 src/vnet/dpo/interface_tx_dpo.h (limited to 'src/plugins/pppoe/pppoe.h') diff --git a/src/plugins/pppoe.am b/src/plugins/pppoe.am index 28bd20a0..06ed60b4 100644 --- a/src/plugins/pppoe.am +++ b/src/plugins/pppoe.am @@ -14,27 +14,26 @@ vppapitestplugins_LTLIBRARIES += pppoe_test_plugin.la vppplugins_LTLIBRARIES += pppoe_plugin.la -pppoe_plugin_la_SOURCES = \ - pppoe/pppoe_decap.c \ - pppoe/pppoe_encap.c \ - pppoe/pppoe_tap.c \ - pppoe/pppoe_tap_node.c \ - pppoe/pppoe.c \ +pppoe_plugin_la_SOURCES = \ + pppoe/pppoe_decap.c \ + pppoe/pppoe_tap.c \ + pppoe/pppoe_tap_node.c \ + pppoe/pppoe.c \ pppoe/pppoe_api.c -BUILT_SOURCES += \ - pppoe/pppoe.api.h \ +BUILT_SOURCES += \ + pppoe/pppoe.api.h \ pppoe/pppoe.api.json API_FILES += pppoe/pppoe.api nobase_apiinclude_HEADERS += \ - pppoe/pppoe_all_api_h.h \ - pppoe/pppoe_msg_enum.h \ + pppoe/pppoe_all_api_h.h \ + pppoe/pppoe_msg_enum.h \ pppoe/pppoe.api.h -pppoe_test_plugin_la_SOURCES = \ - pppoe/pppoe_test.c \ +pppoe_test_plugin_la_SOURCES = \ + pppoe/pppoe_test.c \ pppoe/pppoe_plugin.api.h # vi:syntax=automake diff --git a/src/plugins/pppoe/pppoe.c b/src/plugins/pppoe/pppoe.c index cb587e29..e09ac7d9 100644 --- a/src/plugins/pppoe/pppoe.c +++ b/src/plugins/pppoe/pppoe.c @@ -24,11 +24,13 @@ #include #include #include -#include +#include #include #include #include #include +#include +#include #include #include @@ -85,7 +87,6 @@ pppoe_interface_admin_up_down (vnet_main_t * vnm, u32 hw_if_index, u32 flags) VNET_DEVICE_CLASS (pppoe_device_class,static) = { .name = "PPPPOE", .format_device_name = format_pppoe_name, - .format_tx_trace = format_pppoe_encap_trace, .tx_function = dummy_interface_tx, .admin_up_down_function = pppoe_interface_admin_up_down, }; @@ -99,27 +100,19 @@ format_pppoe_header_with_length (u8 * s, va_list * args) return s; } -/* *INDENT-OFF* */ -VNET_HW_INTERFACE_CLASS (pppoe_hw_class) = -{ - .name = "PPPPOE", - .format_header = format_pppoe_header_with_length, - .build_rewrite = default_build_rewrite, - .flags = VNET_HW_INTERFACE_CLASS_FLAG_P2P, -}; -/* *INDENT-ON* */ - -#define foreach_copy_field \ -_(session_id) \ -_(encap_if_index) \ -_(decap_fib_index) \ -_(client_ip) - -static void -eth_pppoe_rewrite (pppoe_session_t * t, bool is_ip6) +static u8 * +pppoe_build_rewrite (vnet_main_t * vnm, + u32 sw_if_index, + vnet_link_t link_type, const void *dst_address) { - u8 *rw = 0; int len = sizeof (pppoe_header_t) + sizeof (ethernet_header_t); + pppoe_main_t *pem = &pppoe_main; + pppoe_session_t *t; + u32 session_id; + u8 *rw = 0; + + session_id = pem->session_index_by_sw_if_index[sw_if_index]; + t = pool_elt_at_index (pem->sessions, session_id); vec_validate_aligned (rw, len - 1, CLIB_CACHE_LINE_BYTES); @@ -134,21 +127,112 @@ eth_pppoe_rewrite (pppoe_session_t * t, bool is_ip6) pppoe->session_id = clib_host_to_net_u16 (t->session_id); pppoe->length = 0; /* To be filled in at run-time */ - if (!is_ip6) + switch (link_type) { + case VNET_LINK_IP4: pppoe->ppp_proto = clib_host_to_net_u16 (PPP_PROTOCOL_ip4); + break; + case VNET_LINK_IP6: + pppoe->ppp_proto = clib_host_to_net_u16 (PPP_PROTOCOL_ip6); + break; + default: + break; } - else + + return rw; +} + +/** + * @brief Fixup the adj rewrite post encap. Insert the packet's length + */ +static void +pppoe_fixup (vlib_main_t * vm, ip_adjacency_t * adj, vlib_buffer_t * b0) +{ + pppoe_header_t *pppoe0; + + pppoe0 = vlib_buffer_get_current (b0); + + pppoe0->length = clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0) + - sizeof (pppoe_header_t) + - sizeof (ethernet_header_t)); +} + +static void +pppoe_update_adj (vnet_main_t * vnm, u32 sw_if_index, adj_index_t ai) +{ + pppoe_main_t *pem = &pppoe_main; + dpo_id_t dpo = DPO_INVALID; + ip_adjacency_t *adj; + pppoe_session_t *t; + u32 session_id; + + ASSERT (ADJ_INDEX_INVALID != ai); + + adj = adj_get (ai); + + switch (adj->lookup_next_index) { - pppoe->ppp_proto = clib_host_to_net_u16 (PPP_PROTOCOL_ip6); + case IP_LOOKUP_NEXT_ARP: + case IP_LOOKUP_NEXT_GLEAN: + adj_nbr_midchain_update_rewrite (ai, pppoe_fixup, + ADJ_FLAG_NONE, + pppoe_build_rewrite (vnm, + sw_if_index, + adj->ia_link, + NULL)); + break; + case IP_LOOKUP_NEXT_MCAST: + /* + * Construct a partial rewrite from the known ethernet mcast dest MAC + * There's no MAC fixup, so the last 2 parameters are 0 + */ + adj_mcast_midchain_update_rewrite (ai, pppoe_fixup, + ADJ_FLAG_NONE, + pppoe_build_rewrite (vnm, + sw_if_index, + adj->ia_link, + NULL), 0, 0); + break; + + case IP_LOOKUP_NEXT_DROP: + case IP_LOOKUP_NEXT_PUNT: + case IP_LOOKUP_NEXT_LOCAL: + case IP_LOOKUP_NEXT_REWRITE: + case IP_LOOKUP_NEXT_MIDCHAIN: + case IP_LOOKUP_NEXT_MCAST_MIDCHAIN: + case IP_LOOKUP_NEXT_ICMP_ERROR: + case IP_LOOKUP_N_NEXT: + ASSERT (0); + break; } - t->rewrite = rw; - _vec_len (t->rewrite) = len; + session_id = pem->session_index_by_sw_if_index[sw_if_index]; + t = pool_elt_at_index (pem->sessions, session_id); + interface_tx_dpo_add_or_lock (vnet_link_to_dpo_proto (adj->ia_link), + t->encap_if_index, &dpo); + + adj_nbr_midchain_stack (ai, &dpo); - return; + dpo_reset (&dpo); } +/* *INDENT-OFF* */ +VNET_HW_INTERFACE_CLASS (pppoe_hw_class) = +{ + .name = "PPPPOE", + .format_header = format_pppoe_header_with_length, + .build_rewrite = pppoe_build_rewrite, + .update_adjacency = pppoe_update_adj, + .flags = VNET_HW_INTERFACE_CLASS_FLAG_P2P, +}; +/* *INDENT-ON* */ + +#define foreach_copy_field \ +_(session_id) \ +_(encap_if_index) \ +_(decap_fib_index) \ +_(client_ip) + static bool pppoe_decap_next_is_valid (pppoe_main_t * pem, u32 is_ip6, u32 decap_fib_index) @@ -231,8 +315,6 @@ int vnet_pppoe_add_del_session clib_memcpy (t->client_mac, a->client_mac, 6); - eth_pppoe_rewrite (t, is_ip6); - /* update pppoe fib with session_index */ result.fields.session_index = t - pem->sessions; pppoe_update_1 (&pem->session_table, @@ -285,9 +367,6 @@ int vnet_pppoe_add_del_session vnet_sw_interface_set_flags (vnm, sw_if_index, VNET_SW_INTERFACE_FLAG_ADMIN_UP); - /* Set pppoe session output node */ - hi->output_node_index = pppoe_encap_node.index; - /* add reverse route for client ip */ fib_table_entry_path_add (a->decap_fib_index, &pfx, FIB_SOURCE_PLUGIN_HI, FIB_ENTRY_FLAG_NONE, @@ -328,7 +407,6 @@ int vnet_pppoe_add_del_session sw_if_index, ~0, 1, FIB_ROUTE_PATH_FLAG_NONE); - vec_free (t->rewrite); pool_put (pem->sessions, t); } diff --git a/src/plugins/pppoe/pppoe.h b/src/plugins/pppoe/pppoe.h index 37d628eb..b06c068f 100644 --- a/src/plugins/pppoe/pppoe.h +++ b/src/plugins/pppoe/pppoe.h @@ -48,9 +48,6 @@ typedef struct typedef struct { - /* Rewrite string */ - u8 *rewrite; - /* pppoe session_id in HOST byte order */ u16 session_id; @@ -177,11 +174,8 @@ typedef struct extern pppoe_main_t pppoe_main; extern vlib_node_registration_t pppoe_input_node; -extern vlib_node_registration_t pppoe_encap_node; extern vlib_node_registration_t pppoe_tap_dispatch_node; -u8 *format_pppoe_encap_trace (u8 * s, va_list * args); - typedef struct { u8 is_add; diff --git a/src/plugins/pppoe/pppoe_encap.c b/src/plugins/pppoe/pppoe_encap.c deleted file mode 100644 index 69bec61d..00000000 --- a/src/plugins/pppoe/pppoe_encap.c +++ /dev/null @@ -1,384 +0,0 @@ -/* - * Copyright (c) 2017 Intel 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 -#include -#include -#include -#include -#include - -/* Statistics (not all errors) */ -#define foreach_pppoe_encap_error \ -_(ENCAPSULATED, "good packets encapsulated") - -static char * pppoe_encap_error_strings[] = { -#define _(sym,string) string, - foreach_pppoe_encap_error -#undef _ -}; - -typedef enum { -#define _(sym,str) PPPOE_ENCAP_ERROR_##sym, - foreach_pppoe_encap_error -#undef _ - PPPOE_ENCAP_N_ERROR, -} pppoe_encap_error_t; - -#define foreach_pppoe_encap_next \ -_(DROP, "error-drop") \ -_(INTERFACE, "interface-output" ) \ - -typedef enum -{ -#define _(s,n) PPPOE_ENCAP_NEXT_##s, - foreach_pppoe_encap_next -#undef _ - PPPOE_ENCAP_N_NEXT, -} pppoe_encap_next_t; - -typedef struct { - u32 session_index; - u32 session_id; -} pppoe_encap_trace_t; - -u8 * format_pppoe_encap_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 *); - pppoe_encap_trace_t * t - = va_arg (*args, pppoe_encap_trace_t *); - - s = format (s, "PPPOE encap to pppoe_session%d session_id %d", - t->session_index, t->session_id); - return s; -} - - -#define foreach_fixed_header2_offset \ - _(0) _(1) - - -static uword -pppoe_encap (vlib_main_t * vm, - vlib_node_runtime_t * node, - vlib_frame_t * from_frame) -{ - u32 n_left_from, next_index, * from, * to_next; - pppoe_main_t * pem = &pppoe_main; - vnet_main_t * vnm = pem->vnet_main; - vnet_interface_main_t * im = &vnm->interface_main; - u32 pkts_encapsulated = 0; - u32 thread_index = vlib_get_thread_index(); - u32 stats_sw_if_index, stats_n_packets, stats_n_bytes; - u32 sw_if_index0 = 0, sw_if_index1 = 0; - u32 next0 = 0, next1 = 0; - pppoe_session_t * t0 = NULL, * t1 = NULL; - - from = vlib_frame_vector_args (from_frame); - n_left_from = from_frame->n_vectors; - - next_index = node->cached_next_index; - stats_sw_if_index = node->runtime_data[0]; - stats_n_packets = stats_n_bytes = 0; - - 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 >= 4 && n_left_to_next >= 2) - { - u32 bi0, bi1; - vlib_buffer_t * b0, * b1; - u32 len0, len1; - ethernet_header_t * eth0, * eth1; - pppoe_header_t * pppoe0, * pppoe1; - u64 * copy_src0, * copy_dst0; - u64 * copy_src1, * copy_dst1; - u16 * copy_src_last0, * copy_dst_last0; - u16 * copy_src_last1, * copy_dst_last1; - u16 new_l0, new_l1; - u32 session_id0, session_id1; - - /* Prefetch next iteration. */ - { - vlib_buffer_t * p2, * p3; - - p2 = vlib_get_buffer (vm, from[2]); - p3 = vlib_get_buffer (vm, from[3]); - - vlib_prefetch_buffer_header (p2, LOAD); - vlib_prefetch_buffer_header (p3, LOAD); - - CLIB_PREFETCH (p2->data, 2*CLIB_CACHE_LINE_BYTES, LOAD); - CLIB_PREFETCH (p3->data, 2*CLIB_CACHE_LINE_BYTES, LOAD); - } - - bi0 = from[0]; - bi1 = from[1]; - to_next[0] = bi0; - to_next[1] = bi1; - from += 2; - to_next += 2; - n_left_to_next -= 2; - n_left_from -= 2; - - b0 = vlib_get_buffer (vm, bi0); - b1 = vlib_get_buffer (vm, bi1); - - /* Get next node index and if-index from session */ - sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX]; - session_id0 = pem->session_index_by_sw_if_index[sw_if_index0]; - t0 = pool_elt_at_index(pem->sessions, session_id0); - next0 = PPPOE_ENCAP_NEXT_INTERFACE; - vnet_buffer(b0)->sw_if_index[VLIB_TX] = t0->encap_if_index; - - /* Get next node index and if-index from session */ - sw_if_index1 = vnet_buffer(b1)->sw_if_index[VLIB_TX]; - session_id1 = pem->session_index_by_sw_if_index[sw_if_index1]; - t1 = pool_elt_at_index(pem->sessions, session_id1); - next1 = PPPOE_ENCAP_NEXT_INTERFACE; - vnet_buffer(b1)->sw_if_index[VLIB_TX] = t1->encap_if_index; - - /* Apply the rewrite string. $$$$ vnet_rewrite? */ - vlib_buffer_advance (b0, -(word)_vec_len(t0->rewrite)); - vlib_buffer_advance (b1, -(word)_vec_len(t1->rewrite)); - - eth0 = (ethernet_header_t *)(vlib_buffer_get_current(b0)); - eth1 = (ethernet_header_t *)(vlib_buffer_get_current(b1)); - - /* Copy the fixed header */ - copy_dst0 = (u64 *) eth0; - copy_src0 = (u64 *) t0->rewrite; - copy_dst1 = (u64 *) eth1; - copy_src1 = (u64 *) t1->rewrite; - /* Copy first 8-bytes at a time */ -#define _(offs) copy_dst0[offs] = copy_src0[offs]; - foreach_fixed_header2_offset; -#undef _ - /* Last 6 octets. Hopefully gcc will be our friend */ - copy_dst_last0 = (u16 *)(©_dst0[2]); - copy_src_last0 = (u16 *)(©_src0[2]); - copy_dst_last0[0] = copy_src_last0[0]; - copy_dst_last0[1] = copy_src_last0[1]; - copy_dst_last0[2] = copy_src_last0[2]; - -#define _(offs) copy_dst1[offs] = copy_src1[offs]; - foreach_fixed_header2_offset; -#undef _ - /* Last 6 octets. Hopefully gcc will be our friend */ - copy_dst_last1 = (u16 *)(©_dst1[2]); - copy_src_last1 = (u16 *)(©_src1[2]); - copy_dst_last1[0] = copy_src_last1[0]; - copy_dst_last1[1] = copy_src_last1[1]; - copy_dst_last1[2] = copy_src_last1[2]; - - /* Fix PPPoE length */ - new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0) - - sizeof (*pppoe0) - sizeof(*eth0)); - pppoe0 = (pppoe_header_t *)(eth0 + 1); - pppoe0->length = new_l0; - - new_l1 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b1) - - sizeof (*pppoe1) - sizeof(*eth1)); - pppoe1 = (pppoe_header_t *)(eth1 + 1); - pppoe1->length = new_l1; - - pkts_encapsulated += 2; - len0 = vlib_buffer_length_in_chain (vm, b0); - len1 = vlib_buffer_length_in_chain (vm, b1); - stats_n_packets += 2; - stats_n_bytes += len0 + len1; - - /* Batch stats increment on the same pppoe session so counter is not - incremented per packet. Note stats are still incremented for deleted - and admin-down session where packets are dropped. It is not worthwhile - to check for this rare case and affect normal path performance. */ - if (PREDICT_FALSE ((sw_if_index0 != stats_sw_if_index) || - (sw_if_index1 != stats_sw_if_index))) - { - stats_n_packets -= 2; - stats_n_bytes -= len0 + len1; - if (sw_if_index0 == sw_if_index1) - { - if (stats_n_packets) - vlib_increment_combined_counter - (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX, - thread_index, stats_sw_if_index, - stats_n_packets, stats_n_bytes); - stats_sw_if_index = sw_if_index0; - stats_n_packets = 2; - stats_n_bytes = len0 + len1; - } - else - { - vlib_increment_combined_counter - (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX, - thread_index, sw_if_index0, 1, len0); - vlib_increment_combined_counter - (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX, - thread_index, sw_if_index1, 1, len1); - } - } - - if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED)) - { - pppoe_encap_trace_t *tr = - vlib_add_trace (vm, node, b0, sizeof (*tr)); - tr->session_index = t0 - pem->sessions; - tr->session_id = t0->session_id; - } - - if (PREDICT_FALSE(b1->flags & VLIB_BUFFER_IS_TRACED)) - { - pppoe_encap_trace_t *tr = - vlib_add_trace (vm, node, b1, sizeof (*tr)); - tr->session_index = t1 - pem->sessions; - tr->session_id = t1->session_id; - } - - vlib_validate_buffer_enqueue_x2 (vm, node, next_index, - to_next, n_left_to_next, - bi0, bi1, next0, next1); - } - - while (n_left_from > 0 && n_left_to_next > 0) - { - u32 bi0; - vlib_buffer_t * b0; - ethernet_header_t * eth0; - pppoe_header_t * pppoe0; - u64 * copy_src0, * copy_dst0; - u16 * copy_src_last0, * copy_dst_last0; - u16 new_l0; - u32 len0; - u32 session_id0; - - 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); - - /* Get next node index and if-index from session */ - sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX]; - session_id0 = pem->session_index_by_sw_if_index[sw_if_index0]; - t0 = pool_elt_at_index(pem->sessions, session_id0); - next0 = PPPOE_ENCAP_NEXT_INTERFACE; - vnet_buffer(b0)->sw_if_index[VLIB_TX] = t0->encap_if_index; - - /* Apply the rewrite string. $$$$ vnet_rewrite? */ - vlib_buffer_advance (b0, -(word)_vec_len(t0->rewrite)); - - eth0 = (ethernet_header_t *)(vlib_buffer_get_current(b0)); - /* Copy the fixed header */ - copy_dst0 = (u64 *) eth0; - copy_src0 = (u64 *) t0->rewrite; - - /* Copy first 8-bytes at a time */ -#define _(offs) copy_dst0[offs] = copy_src0[offs]; - foreach_fixed_header2_offset; -#undef _ - /* Last 6 octets. Hopefully gcc will be our friend */ - copy_dst_last0 = (u16 *)(©_dst0[2]); - copy_src_last0 = (u16 *)(©_src0[2]); - copy_dst_last0[0] = copy_src_last0[0]; - copy_dst_last0[1] = copy_src_last0[1]; - copy_dst_last0[2] = copy_src_last0[2]; - - /* Fix PPPoE length */ - new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0) - - sizeof (*pppoe0) - sizeof(*eth0)); - pppoe0 = (pppoe_header_t *)(eth0 + 1); - pppoe0->length = new_l0; - - pkts_encapsulated ++; - len0 = vlib_buffer_length_in_chain (vm, b0); - stats_n_packets += 1; - stats_n_bytes += len0; - - /* Batch stats increment on the same pppoe session so counter is not - incremented per packet. Note stats are still incremented for deleted - and admin-down session where packets are dropped. It is not worthwhile - to check for this rare case and affect normal path performance. */ - if (PREDICT_FALSE (sw_if_index0 != stats_sw_if_index)) - { - stats_n_packets -= 1; - stats_n_bytes -= len0; - if (stats_n_packets) - vlib_increment_combined_counter - (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX, - thread_index, stats_sw_if_index, - stats_n_packets, stats_n_bytes); - stats_n_packets = 1; - stats_n_bytes = len0; - stats_sw_if_index = sw_if_index0; - } - - if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED)) - { - pppoe_encap_trace_t *tr = - vlib_add_trace (vm, node, b0, sizeof (*tr)); - tr->session_index = t0 - pem->sessions; - tr->session_id = t0->session_id; - } - 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); - } - - /* Do we still need this now that session tx stats is kept? */ - vlib_node_increment_counter (vm, node->node_index, - PPPOE_ENCAP_ERROR_ENCAPSULATED, - pkts_encapsulated); - - /* Increment any remaining batch stats */ - if (stats_n_packets) - { - vlib_increment_combined_counter - (im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX, - thread_index, stats_sw_if_index, stats_n_packets, stats_n_bytes); - node->runtime_data[0] = stats_sw_if_index; - } - - return from_frame->n_vectors; -} - -VLIB_REGISTER_NODE (pppoe_encap_node) = { - .function = pppoe_encap, - .name = "pppoe-encap", - .vector_size = sizeof (u32), - .format_trace = format_pppoe_encap_trace, - .type = VLIB_NODE_TYPE_INTERNAL, - .n_errors = ARRAY_LEN(pppoe_encap_error_strings), - .error_strings = pppoe_encap_error_strings, - .n_next_nodes = PPPOE_ENCAP_N_NEXT, - .next_nodes = { -#define _(s,n) [PPPOE_ENCAP_NEXT_##s] = n, - foreach_pppoe_encap_next -#undef _ - }, -}; - -VLIB_NODE_FUNCTION_MULTIARCH (pppoe_encap_node, pppoe_encap) - diff --git a/src/vnet.am b/src/vnet.am index 9821069a..182ec1f4 100644 --- a/src/vnet.am +++ b/src/vnet.am @@ -1011,7 +1011,8 @@ libvnet_la_SOURCES += \ vnet/dpo/lookup_dpo.c \ vnet/dpo/classify_dpo.c \ vnet/dpo/replicate_dpo.c \ - vnet/dpo/interface_dpo.c \ + vnet/dpo/interface_rx_dpo.c \ + vnet/dpo/interface_tx_dpo.c \ vnet/dpo/mpls_disposition.c \ vnet/dpo/mpls_label_dpo.c diff --git a/src/vnet/adj/adj_midchain.c b/src/vnet/adj/adj_midchain.c index a93a1c3e..e9a510b0 100644 --- a/src/vnet/adj/adj_midchain.c +++ b/src/vnet/adj/adj_midchain.c @@ -65,8 +65,7 @@ adj_midchain_tx_inline (vlib_main_t * vm, /* set up to enqueue to our disposition with index = next_index */ vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); - - while (n_left_from >= 4 && n_left_to_next > 2) + while (n_left_from >= 8 && n_left_to_next > 4) { u32 bi0, adj_index0, next0; const ip_adjacency_t * adj0; @@ -76,49 +75,75 @@ adj_midchain_tx_inline (vlib_main_t * vm, const ip_adjacency_t * adj1; const dpo_id_t *dpo1; vlib_buffer_t * b1; + u32 bi2, adj_index2, next2; + const ip_adjacency_t * adj2; + const dpo_id_t *dpo2; + vlib_buffer_t * b2; + u32 bi3, adj_index3, next3; + const ip_adjacency_t * adj3; + const dpo_id_t *dpo3; + vlib_buffer_t * b3; /* Prefetch next iteration. */ { - vlib_buffer_t * p2, * p3; - - p2 = vlib_get_buffer (vm, from[2]); - p3 = vlib_get_buffer (vm, from[3]); - - vlib_prefetch_buffer_header (p2, LOAD); - vlib_prefetch_buffer_header (p3, LOAD); - - CLIB_PREFETCH (p2->data, CLIB_CACHE_LINE_BYTES, STORE); - CLIB_PREFETCH (p3->data, CLIB_CACHE_LINE_BYTES, STORE); + vlib_buffer_t * p4, * p5; + vlib_buffer_t * 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); } bi0 = from[0]; to_next[0] = bi0; bi1 = from[1]; to_next[1] = bi1; + bi2 = from[2]; + to_next[2] = bi2; + bi3 = from[3]; + to_next[3] = bi3; - from += 2; - to_next += 2; - n_left_from -= 2; - n_left_to_next -= 2; + 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); /* Follow the DPO on which the midchain is stacked */ adj_index0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX]; adj_index1 = vnet_buffer(b1)->ip.adj_index[VLIB_TX]; + adj_index2 = vnet_buffer(b2)->ip.adj_index[VLIB_TX]; + adj_index3 = vnet_buffer(b3)->ip.adj_index[VLIB_TX]; adj0 = adj_get(adj_index0); adj1 = adj_get(adj_index1); + adj2 = adj_get(adj_index2); + adj3 = adj_get(adj_index3); dpo0 = &adj0->sub_type.midchain.next_dpo; dpo1 = &adj1->sub_type.midchain.next_dpo; + dpo2 = &adj2->sub_type.midchain.next_dpo; + dpo3 = &adj3->sub_type.midchain.next_dpo; next0 = dpo0->dpoi_next_node; next1 = dpo1->dpoi_next_node; + next2 = dpo2->dpoi_next_node; + next3 = dpo3->dpoi_next_node; - vnet_buffer(b1)->ip.adj_index[VLIB_TX] = dpo1->dpoi_index; - vnet_buffer(b0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index; + vnet_buffer(b1)->ip.adj_index[VLIB_TX] = dpo1->dpoi_index; + vnet_buffer(b0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index; + vnet_buffer(b2)->ip.adj_index[VLIB_TX] = dpo2->dpoi_index; + vnet_buffer(b3)->ip.adj_index[VLIB_TX] = dpo3->dpoi_index; if (interface_count) { @@ -134,6 +159,18 @@ adj_midchain_tx_inline (vlib_main_t * vm, adj1->rewrite_header.sw_if_index, 1, vlib_buffer_length_in_chain (vm, b1)); + vlib_increment_combined_counter (im->combined_sw_if_counters + + VNET_INTERFACE_COUNTER_TX, + thread_index, + adj2->rewrite_header.sw_if_index, + 1, + vlib_buffer_length_in_chain (vm, b2)); + vlib_increment_combined_counter (im->combined_sw_if_counters + + VNET_INTERFACE_COUNTER_TX, + thread_index, + adj3->rewrite_header.sw_if_index, + 1, + vlib_buffer_length_in_chain (vm, b3)); } if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED)) @@ -148,11 +185,23 @@ adj_midchain_tx_inline (vlib_main_t * vm, b1, sizeof (*tr)); tr->ai = adj_index1; } + if (PREDICT_FALSE(b2->flags & VLIB_BUFFER_IS_TRACED)) + { + adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node, + b2, sizeof (*tr)); + tr->ai = adj_index2; + } + if (PREDICT_FALSE(b3->flags & VLIB_BUFFER_IS_TRACED)) + { + adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node, + b3, sizeof (*tr)); + tr->ai = adj_index3; + } - vlib_validate_buffer_enqueue_x2 (vm, node, next_index, + vlib_validate_buffer_enqueue_x4 (vm, node, next_index, to_next, n_left_to_next, - bi0, bi1, - next0, next1); + bi0, bi1, bi2, bi3, + next0, next1, next2, next3); } while (n_left_from > 0 && n_left_to_next > 0) { @@ -175,7 +224,7 @@ adj_midchain_tx_inline (vlib_main_t * vm, adj0 = adj_get(adj_index0); dpo0 = &adj0->sub_type.midchain.next_dpo; next0 = dpo0->dpoi_next_node; - vnet_buffer(b0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index; + vnet_buffer(b0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index; if (interface_count) { @@ -392,6 +441,17 @@ adj_nbr_midchain_get_tx_node (ip_adjacency_t *adj) adj_midchain_tx_node.index); } +static u32 +adj_nbr_midchain_get_feature_node (ip_adjacency_t *adj) +{ + if (adj->ia_flags & ADJ_FLAG_MIDCHAIN_NO_COUNT) + { + return (adj_midchain_tx_no_count_feature_node[adj->ia_link]); + } + + return (adj_midchain_tx_feature_node[adj->ia_link]); +} + /** * adj_midchain_setup * @@ -414,10 +474,7 @@ adj_midchain_setup (adj_index_t adj_index, adj->ia_flags |= flags; arc_index = adj_midchain_get_feature_arc_index_for_link_type (adj); - feature_index = (flags & ADJ_FLAG_MIDCHAIN_NO_COUNT) ? - adj_midchain_tx_no_count_feature_node[adj->ia_link] : - adj_midchain_tx_feature_node[adj->ia_link]; - + feature_index = adj_nbr_midchain_get_feature_node(adj); tx_node = adj_nbr_midchain_get_tx_node(adj); vnet_feature_enable_disable_with_index (arc_index, feature_index, @@ -432,8 +489,8 @@ adj_midchain_setup (adj_index_t adj_index, * need to get to the stacked child's node. */ dpo_stack_from_node(tx_node, - &adj->sub_type.midchain.next_dpo, - drop_dpo_get(vnet_link_to_dpo_proto(adj->ia_link))); + &adj->sub_type.midchain.next_dpo, + drop_dpo_get(vnet_link_to_dpo_proto(adj->ia_link))); } /** @@ -495,10 +552,9 @@ adj_nbr_midchain_unstack (adj_index_t adj_index) * stack on the drop */ dpo_stack(DPO_ADJACENCY_MIDCHAIN, - vnet_link_to_dpo_proto(adj->ia_link), - &adj->sub_type.midchain.next_dpo, - drop_dpo_get(vnet_link_to_dpo_proto(adj->ia_link))); - + vnet_link_to_dpo_proto(adj->ia_link), + &adj->sub_type.midchain.next_dpo, + drop_dpo_get(vnet_link_to_dpo_proto(adj->ia_link))); CLIB_MEMORY_BARRIER(); } @@ -537,9 +593,9 @@ format_adj_midchain (u8* s, va_list *ap) format_vnet_rewrite, &adj->rewrite_header, sizeof (adj->rewrite_data), indent); s = format (s, "\n%Ustacked-on:\n%U%U", - format_white_space, indent, - format_white_space, indent+2, - format_dpo_id, &adj->sub_type.midchain.next_dpo, indent+2); + format_white_space, indent, + format_white_space, indent+2, + format_dpo_id, &adj->sub_type.midchain.next_dpo, indent+2); return (s); } diff --git a/src/vnet/dpo/dpo.c b/src/vnet/dpo/dpo.c index aa770838..bd18b66b 100644 --- a/src/vnet/dpo/dpo.c +++ b/src/vnet/dpo/dpo.c @@ -37,7 +37,8 @@ #include #include #include -#include +#include +#include #include /** @@ -275,6 +276,29 @@ dpo_is_adj (const dpo_id_t *dpo) (dpo->dpoi_type == DPO_ADJACENCY_GLEAN)); } +static u32 * +dpo_default_get_next_node (const dpo_id_t *dpo) +{ + u32 *node_indices = NULL; + const char *node_name; + u32 ii = 0; + + node_name = dpo_nodes[dpo->dpoi_type][dpo->dpoi_proto][ii]; + while (NULL != node_name) + { + vlib_node_t *node; + + node = vlib_get_node_by_name(vlib_get_main(), (u8*) node_name); + ASSERT(NULL != node); + vec_add1(node_indices, node->index); + + ++ii; + node_name = dpo_nodes[dpo->dpoi_type][dpo->dpoi_proto][ii]; + } + + return (node_indices); +} + void dpo_register (dpo_type_t type, const dpo_vft_t *vft, @@ -282,6 +306,10 @@ dpo_register (dpo_type_t type, { vec_validate(dpo_vfts, type); dpo_vfts[type] = *vft; + if (NULL == dpo_vfts[type].dv_get_next_node) + { + dpo_vfts[type].dv_get_next_node = dpo_default_get_next_node; + } vec_validate(dpo_nodes, type); dpo_nodes[type] = nodes; @@ -340,24 +368,25 @@ dpo_get_next_node (dpo_type_t child_type, */ if (~0 == dpo_edges[child_type][child_proto][parent_type][parent_proto]) { - vlib_node_t *parent_node, *child_node; + vlib_node_t *child_node; + u32 *parent_indices; vlib_main_t *vm; - u32 edge ,pp, cc; + u32 edge, *pi, cc; vm = vlib_get_main(); - vlib_worker_thread_barrier_sync(vm); - + ASSERT(NULL != dpo_vfts[parent_type].dv_get_next_node); ASSERT(NULL != dpo_nodes[child_type]); ASSERT(NULL != dpo_nodes[child_type][child_proto]); - ASSERT(NULL != dpo_nodes[parent_type]); - ASSERT(NULL != dpo_nodes[parent_type][parent_proto]); cc = 0; + parent_indices = dpo_vfts[parent_type].dv_get_next_node(parent_dpo); + + vlib_worker_thread_barrier_sync(vm); /* - * create a graph arc from each of the parent's registered node types, - * to each of the childs. + * create a graph arc from each of the child's registered node types, + * to each of the parent's. */ while (NULL != dpo_nodes[child_type][child_proto][cc]) { @@ -365,17 +394,9 @@ dpo_get_next_node (dpo_type_t child_type, vlib_get_node_by_name(vm, (u8*) dpo_nodes[child_type][child_proto][cc]); - pp = 0; - - while (NULL != dpo_nodes[parent_type][parent_proto][pp]) + vec_foreach(pi, parent_indices) { - parent_node = - vlib_get_node_by_name(vm, - (u8*) dpo_nodes[parent_type][parent_proto][pp]); - - edge = vlib_node_add_next(vm, - child_node->index, - parent_node->index); + edge = vlib_node_add_next(vm, child_node->index, *pi); if (~0 == dpo_edges[child_type][child_proto][parent_type][parent_proto]) { @@ -385,12 +406,12 @@ dpo_get_next_node (dpo_type_t child_type, { ASSERT(dpo_edges[child_type][child_proto][parent_type][parent_proto] == edge); } - pp++; } cc++; } vlib_worker_thread_barrier_release(vm); + vec_free(parent_indices); } return (dpo_edges[child_type][child_proto][parent_type][parent_proto]); @@ -451,38 +472,39 @@ dpo_stack_from_node (u32 child_node_index, dpo_id_t *dpo, const dpo_id_t *parent) { - dpo_proto_t parent_proto; - vlib_node_t *parent_node; dpo_type_t parent_type; + u32 *parent_indices; vlib_main_t *vm; - u32 edge; + u32 edge, *pi; + edge = 0; parent_type = parent->dpoi_type; - parent_proto = parent->dpoi_proto; - vm = vlib_get_main(); - ASSERT(NULL != dpo_nodes[parent_type]); - ASSERT(NULL != dpo_nodes[parent_type][parent_proto]); + ASSERT(NULL != dpo_vfts[parent_type].dv_get_next_node); + parent_indices = dpo_vfts[parent_type].dv_get_next_node(parent); + ASSERT(parent_indices); - parent_node = - vlib_get_node_by_name(vm, (u8*) dpo_nodes[parent_type][parent_proto][0]); - - edge = vlib_node_get_next(vm, - child_node_index, - parent_node->index); - - if (~0 == edge) + /* + * This loop is purposefully written with the worker thread lock in the + * inner loop because; + * 1) the likelihood that the edge does not exist is smaller + * 2) the likelihood there is more than one node is even smaller + * so we are optimising for not need to take the lock + */ + vec_foreach(pi, parent_indices) { - vlib_worker_thread_barrier_sync(vm); + edge = vlib_node_get_next(vm, child_node_index, *pi); - edge = vlib_node_add_next(vm, - child_node_index, - parent_node->index); + if (~0 == edge) + { + vlib_worker_thread_barrier_sync(vm); - vlib_worker_thread_barrier_release(vm); - } + edge = vlib_node_add_next(vm, child_node_index, *pi); + vlib_worker_thread_barrier_release(vm); + } + } dpo_stack_i(edge, dpo, parent); } @@ -498,7 +520,8 @@ dpo_module_init (vlib_main_t * vm) lookup_dpo_module_init(); ip_null_dpo_module_init(); replicate_module_init(); - interface_dpo_module_init(); + interface_rx_dpo_module_init(); + interface_tx_dpo_module_init(); mpls_disp_dpo_module_init(); return (NULL); diff --git a/src/vnet/dpo/dpo.h b/src/vnet/dpo/dpo.h index 42fc51d4..33562968 100644 --- a/src/vnet/dpo/dpo.h +++ b/src/vnet/dpo/dpo.h @@ -112,7 +112,8 @@ typedef enum dpo_type_t_ { DPO_MPLS_LABEL, DPO_MPLS_DISPOSITION, DPO_MFIB_ENTRY, - DPO_INTERFACE, + DPO_INTERFACE_RX, + DPO_INTERFACE_TX, DPO_LAST, } __attribute__((packed)) dpo_type_t; @@ -138,7 +139,8 @@ typedef enum dpo_type_t_ { [DPO_MPLS_LABEL] = "dpo-mpls-label", \ [DPO_MPLS_DISPOSITION] = "dpo-mpls-diposition", \ [DPO_MFIB_ENTRY] = "dpo-mfib_entry", \ - [DPO_INTERFACE] = "dpo-interface" \ + [DPO_INTERFACE_RX] = "dpo-interface-rx", \ + [DPO_INTERFACE_TX] = "dpo-interface-tx" \ } /** @@ -331,6 +333,12 @@ typedef void (*dpo_unlock_fn_t)(dpo_id_t *dpo); */ typedef void (*dpo_mem_show_t)(void); +/** + * @brief Given a DPO instance return a vector of node indices that + * the type/instance will use. + */ +typedef u32* (*dpo_get_next_node_t)(const dpo_id_t *dpo); + /** * @brief A virtual function table regisitered for a DPO type */ @@ -352,6 +360,13 @@ typedef struct dpo_vft_t_ * A show memory usage function */ dpo_mem_show_t dv_mem_show; + /** + * A function to get the next VLIB node given an instance + * of the DPO. If this is null, then the node's name MUST be + * retreiveable from the nodes names array passed in the register + * function + */ + dpo_get_next_node_t dv_get_next_node; } dpo_vft_t; diff --git a/src/vnet/dpo/interface_dpo.c b/src/vnet/dpo/interface_dpo.c deleted file mode 100644 index 780bfa2a..00000000 --- a/src/vnet/dpo/interface_dpo.c +++ /dev/null @@ -1,446 +0,0 @@ -/* - * 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 -#include - -/* - * The 'DB' of interface DPOs. - * There is only one per-interface per-protocol, so this is a per-interface - * vector - */ -static index_t *interface_dpo_db[DPO_PROTO_NUM]; - -static interface_dpo_t * -interface_dpo_alloc (void) -{ - interface_dpo_t *ido; - - pool_get(interface_dpo_pool, ido); - - return (ido); -} - -static inline interface_dpo_t * -interface_dpo_get_from_dpo (const dpo_id_t *dpo) -{ - ASSERT(DPO_INTERFACE == dpo->dpoi_type); - - return (interface_dpo_get(dpo->dpoi_index)); -} - -static inline index_t -interface_dpo_get_index (interface_dpo_t *ido) -{ - return (ido - interface_dpo_pool); -} - -static void -interface_dpo_lock (dpo_id_t *dpo) -{ - interface_dpo_t *ido; - - ido = interface_dpo_get_from_dpo(dpo); - ido->ido_locks++; -} - -static void -interface_dpo_unlock (dpo_id_t *dpo) -{ - interface_dpo_t *ido; - - ido = interface_dpo_get_from_dpo(dpo); - ido->ido_locks--; - - if (0 == ido->ido_locks) - { - interface_dpo_db[ido->ido_proto][ido->ido_sw_if_index] = - INDEX_INVALID; - pool_put(interface_dpo_pool, ido); - } -} - -/* - * interface_dpo_add_or_lock - * - * Add/create and lock a new or lock an existing for the interface DPO - * on the interface and protocol given - */ -void -interface_dpo_add_or_lock (dpo_proto_t proto, - u32 sw_if_index, - dpo_id_t *dpo) -{ - interface_dpo_t *ido; - - vec_validate_init_empty(interface_dpo_db[proto], - sw_if_index, - INDEX_INVALID); - - if (INDEX_INVALID == interface_dpo_db[proto][sw_if_index]) - { - ido = interface_dpo_alloc(); - - ido->ido_sw_if_index = sw_if_index; - ido->ido_proto = proto; - - interface_dpo_db[proto][sw_if_index] = - interface_dpo_get_index(ido); - } - else - { - ido = interface_dpo_get(interface_dpo_db[proto][sw_if_index]); - } - - dpo_set(dpo, DPO_INTERFACE, proto, interface_dpo_get_index(ido)); -} - - -static clib_error_t * -interface_dpo_interface_state_change (vnet_main_t * vnm, - u32 sw_if_index, - u32 flags) -{ - /* - */ - return (NULL); -} - -VNET_SW_INTERFACE_ADMIN_UP_DOWN_FUNCTION( - interface_dpo_interface_state_change); - -/** - * @brief Registered callback for HW interface state changes - */ -static clib_error_t * -interface_dpo_hw_interface_state_change (vnet_main_t * vnm, - u32 hw_if_index, - u32 flags) -{ - return (NULL); -} - -VNET_HW_INTERFACE_LINK_UP_DOWN_FUNCTION( - interface_dpo_hw_interface_state_change); - -static clib_error_t * -interface_dpo_interface_delete (vnet_main_t * vnm, - u32 sw_if_index, - u32 is_add) -{ - return (NULL); -} - -VNET_SW_INTERFACE_ADD_DEL_FUNCTION( - interface_dpo_interface_delete); - -u8* -format_interface_dpo (u8* s, va_list *ap) -{ - index_t index = va_arg(*ap, index_t); - CLIB_UNUSED(u32 indent) = va_arg(*ap, u32); - vnet_main_t * vnm = vnet_get_main(); - interface_dpo_t *ido = interface_dpo_get(index); - - return (format(s, "%U-dpo: %U", - format_vnet_sw_interface_name, - vnm, - vnet_get_sw_interface(vnm, ido->ido_sw_if_index), - format_dpo_proto, ido->ido_proto)); -} - -static void -interface_dpo_mem_show (void) -{ - fib_show_memory_usage("Interface", - pool_elts(interface_dpo_pool), - pool_len(interface_dpo_pool), - sizeof(interface_dpo_t)); -} - - -const static dpo_vft_t interface_dpo_vft = { - .dv_lock = interface_dpo_lock, - .dv_unlock = interface_dpo_unlock, - .dv_format = format_interface_dpo, - .dv_mem_show = interface_dpo_mem_show, -}; - -/** - * @brief The per-protocol VLIB graph nodes that are assigned to a glean - * object. - * - * this means that these graph nodes are ones from which a glean is the - * parent object in the DPO-graph. - */ -const static char* const interface_dpo_ip4_nodes[] = -{ - "interface-dpo-ip4", - NULL, -}; -const static char* const interface_dpo_ip6_nodes[] = -{ - "interface-dpo-ip4", - NULL, -}; -const static char* const interface_dpo_l2_nodes[] = -{ - "interface-dpo-l2", - NULL, -}; - -const static char* const * const interface_dpo_nodes[DPO_PROTO_NUM] = -{ - [DPO_PROTO_IP4] = interface_dpo_ip4_nodes, - [DPO_PROTO_IP6] = interface_dpo_ip6_nodes, - [DPO_PROTO_ETHERNET] = interface_dpo_l2_nodes, - [DPO_PROTO_MPLS] = NULL, -}; - -void -interface_dpo_module_init (void) -{ - dpo_register(DPO_INTERFACE, - &interface_dpo_vft, - interface_dpo_nodes); -} - -/** - * @brief Interface DPO trace data - */ -typedef struct interface_dpo_trace_t_ -{ - u32 sw_if_index; -} interface_dpo_trace_t; - -typedef enum interface_dpo_next_t_ -{ - INTERFACE_DPO_DROP = 0, - INTERFACE_DPO_INPUT = 1, -} interface_dpo_next_t; - -always_inline uword -interface_dpo_inline (vlib_main_t * vm, - vlib_node_runtime_t * node, - vlib_frame_t * from_frame) -{ - u32 n_left_from, next_index, * from, * to_next; - u32 thread_index = vlib_get_thread_index (); - vnet_interface_main_t *im; - - im = &vnet_get_main ()->interface_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 >= 4 && n_left_to_next > 2) - { - const interface_dpo_t *ido0, *ido1; - u32 bi0, idoi0, bi1, idoi1; - vlib_buffer_t *b0, *b1; - - bi0 = from[0]; - to_next[0] = bi0; - bi1 = from[1]; - to_next[1] = bi1; - from += 2; - to_next += 2; - n_left_from -= 2; - n_left_to_next -= 2; - - b0 = vlib_get_buffer (vm, bi0); - b1 = vlib_get_buffer (vm, bi1); - - idoi0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX]; - idoi1 = vnet_buffer(b1)->ip.adj_index[VLIB_TX]; - ido0 = interface_dpo_get(idoi0); - ido1 = interface_dpo_get(idoi1); - - vnet_buffer(b0)->sw_if_index[VLIB_RX] = ido0->ido_sw_if_index; - vnet_buffer(b1)->sw_if_index[VLIB_RX] = ido1->ido_sw_if_index; - - vlib_increment_combined_counter (im->combined_sw_if_counters - + VNET_INTERFACE_COUNTER_RX, - thread_index, - ido0->ido_sw_if_index, - 1, - vlib_buffer_length_in_chain (vm, b0)); - vlib_increment_combined_counter (im->combined_sw_if_counters - + VNET_INTERFACE_COUNTER_RX, - thread_index, - ido1->ido_sw_if_index, - 1, - vlib_buffer_length_in_chain (vm, b1)); - - if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED)) - { - interface_dpo_trace_t *tr0; - - tr0 = vlib_add_trace (vm, node, b0, sizeof (*tr0)); - tr0->sw_if_index = ido0->ido_sw_if_index; - } - if (PREDICT_FALSE(b1->flags & VLIB_BUFFER_IS_TRACED)) - { - interface_dpo_trace_t *tr1; - - tr1 = vlib_add_trace (vm, node, b1, sizeof (*tr1)); - tr1->sw_if_index = ido1->ido_sw_if_index; - } - - vlib_validate_buffer_enqueue_x2(vm, node, next_index, to_next, - n_left_to_next, bi0, bi1, - INTERFACE_DPO_INPUT, - INTERFACE_DPO_INPUT); - } - - while (n_left_from > 0 && n_left_to_next > 0) - { - const interface_dpo_t * ido0; - vlib_buffer_t * b0; - u32 bi0, idoi0; - - 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); - - idoi0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX]; - ido0 = interface_dpo_get(idoi0); - - /* Swap the RX interface of the packet to the one the - * interface DPR represents */ - vnet_buffer(b0)->sw_if_index[VLIB_RX] = ido0->ido_sw_if_index; - - /* Bump the interface's RX coutners */ - vlib_increment_combined_counter (im->combined_sw_if_counters - + VNET_INTERFACE_COUNTER_RX, - thread_index, - ido0->ido_sw_if_index, - 1, - vlib_buffer_length_in_chain (vm, b0)); - - if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED)) - { - interface_dpo_trace_t *tr; - - tr = vlib_add_trace (vm, node, b0, sizeof (*tr)); - tr->sw_if_index = ido0->ido_sw_if_index; - } - - vlib_validate_buffer_enqueue_x1(vm, node, next_index, to_next, - n_left_to_next, bi0, - INTERFACE_DPO_INPUT); - } - vlib_put_next_frame (vm, node, next_index, n_left_to_next); - } - return from_frame->n_vectors; -} - -static u8 * -format_interface_dpo_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 *); - interface_dpo_trace_t * t = va_arg (*args, interface_dpo_trace_t *); - uword indent = format_get_indent (s); - s = format (s, "%U sw_if_index:%d", - format_white_space, indent, - t->sw_if_index); - return s; -} - -static uword -interface_dpo_ip4 (vlib_main_t * vm, - vlib_node_runtime_t * node, - vlib_frame_t * from_frame) -{ - return (interface_dpo_inline(vm, node, from_frame)); -} - -static uword -interface_dpo_ip6 (vlib_main_t * vm, - vlib_node_runtime_t * node, - vlib_frame_t * from_frame) -{ - return (interface_dpo_inline(vm, node, from_frame)); -} - -static uword -interface_dpo_l2 (vlib_main_t * vm, - vlib_node_runtime_t * node, - vlib_frame_t * from_frame) -{ - return (interface_dpo_inline(vm, node, from_frame)); -} - -VLIB_REGISTER_NODE (interface_dpo_ip4_node) = { - .function = interface_dpo_ip4, - .name = "interface-dpo-ip4", - .vector_size = sizeof (u32), - .format_trace = format_interface_dpo_trace, - - .n_next_nodes = 2, - .next_nodes = { - [INTERFACE_DPO_DROP] = "ip4-drop", - [INTERFACE_DPO_INPUT] = "ip4-input", - }, -}; - -VLIB_NODE_FUNCTION_MULTIARCH (interface_dpo_ip4_node, - interface_dpo_ip4) - -VLIB_REGISTER_NODE (interface_dpo_ip6_node) = { - .function = interface_dpo_ip6, - .name = "interface-dpo-ip6", - .vector_size = sizeof (u32), - .format_trace = format_interface_dpo_trace, - - .n_next_nodes = 2, - .next_nodes = { - [INTERFACE_DPO_DROP] = "ip6-drop", - [INTERFACE_DPO_INPUT] = "ip6-input", - }, -}; - -VLIB_NODE_FUNCTION_MULTIARCH (interface_dpo_ip6_node, - interface_dpo_ip6) - -VLIB_REGISTER_NODE (interface_dpo_l2_node) = { - .function = interface_dpo_l2, - .name = "interface-dpo-l2", - .vector_size = sizeof (u32), - .format_trace = format_interface_dpo_trace, - - .n_next_nodes = 2, - .next_nodes = { - [INTERFACE_DPO_DROP] = "error-drop", - [INTERFACE_DPO_INPUT] = "l2-input", - }, -}; - -VLIB_NODE_FUNCTION_MULTIARCH (interface_dpo_l2_node, - interface_dpo_l2) - diff --git a/src/vnet/dpo/interface_dpo.h b/src/vnet/dpo/interface_dpo.h deleted file mode 100644 index 1538dfbb..00000000 --- a/src/vnet/dpo/interface_dpo.h +++ /dev/null @@ -1,67 +0,0 @@ -/* - * 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. - */ -/** - * @brief - * The data-path object representing interfaceing the packet, i.e. it's for-us - */ - -#ifndef __INTERFACE_DPO_H__ -#define __INTERFACE_DPO_H__ - -#include - -typedef struct interface_dpo_t_ -{ - /** - * The Software interface index that the packets will be given - * as the ingress/rx interface - */ - u32 ido_sw_if_index; - - /** - * next VLIB node. A '-input' node. - */ - u32 ido_next_node; - - /** - * DPO protocol that the packets will have as they 'ingress' - * on this interface - */ - dpo_proto_t ido_proto; - - /** - * number of locks. - */ - u16 ido_locks; -} interface_dpo_t; - -extern void interface_dpo_add_or_lock (dpo_proto_t proto, - u32 sw_if_index, - dpo_id_t *dpo); - -extern void interface_dpo_module_init(void); - -/** - * @brief pool of all interface DPOs - */ -interface_dpo_t *interface_dpo_pool; - -static inline interface_dpo_t * -interface_dpo_get (index_t index) -{ - return (pool_elt_at_index(interface_dpo_pool, index)); -} - -#endif diff --git a/src/vnet/dpo/interface_rx_dpo.c b/src/vnet/dpo/interface_rx_dpo.c new file mode 100644 index 00000000..a624f514 --- /dev/null +++ b/src/vnet/dpo/interface_rx_dpo.c @@ -0,0 +1,445 @@ +/* + * 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 +#include + +/* + * The 'DB' of interface DPOs. + * There is only one per-interface per-protocol, so this is a per-interface + * vector + */ +static index_t *interface_rx_dpo_db[DPO_PROTO_NUM]; + +static interface_rx_dpo_t * +interface_rx_dpo_alloc (void) +{ + interface_rx_dpo_t *ido; + + pool_get(interface_rx_dpo_pool, ido); + + return (ido); +} + +static inline interface_rx_dpo_t * +interface_rx_dpo_get_from_dpo (const dpo_id_t *dpo) +{ + ASSERT(DPO_INTERFACE_RX == dpo->dpoi_type); + + return (interface_rx_dpo_get(dpo->dpoi_index)); +} + +static inline index_t +interface_rx_dpo_get_index (interface_rx_dpo_t *ido) +{ + return (ido - interface_rx_dpo_pool); +} + +static void +interface_rx_dpo_lock (dpo_id_t *dpo) +{ + interface_rx_dpo_t *ido; + + ido = interface_rx_dpo_get_from_dpo(dpo); + ido->ido_locks++; +} + +static void +interface_rx_dpo_unlock (dpo_id_t *dpo) +{ + interface_rx_dpo_t *ido; + + ido = interface_rx_dpo_get_from_dpo(dpo); + ido->ido_locks--; + + if (0 == ido->ido_locks) + { + interface_rx_dpo_db[ido->ido_proto][ido->ido_sw_if_index] = + INDEX_INVALID; + pool_put(interface_rx_dpo_pool, ido); + } +} + +/* + * interface_rx_dpo_add_or_lock + * + * Add/create and lock a new or lock an existing for the interface DPO + * on the interface and protocol given + */ +void +interface_rx_dpo_add_or_lock (dpo_proto_t proto, + u32 sw_if_index, + dpo_id_t *dpo) +{ + interface_rx_dpo_t *ido; + + vec_validate_init_empty(interface_rx_dpo_db[proto], + sw_if_index, + INDEX_INVALID); + + if (INDEX_INVALID == interface_rx_dpo_db[proto][sw_if_index]) + { + ido = interface_rx_dpo_alloc(); + + ido->ido_sw_if_index = sw_if_index; + ido->ido_proto = proto; + + interface_rx_dpo_db[proto][sw_if_index] = + interface_rx_dpo_get_index(ido); + } + else + { + ido = interface_rx_dpo_get(interface_rx_dpo_db[proto][sw_if_index]); + } + + dpo_set(dpo, DPO_INTERFACE_RX, proto, interface_rx_dpo_get_index(ido)); +} + + +static clib_error_t * +interface_rx_dpo_interface_state_change (vnet_main_t * vnm, + u32 sw_if_index, + u32 flags) +{ + /* + */ + return (NULL); +} + +VNET_SW_INTERFACE_ADMIN_UP_DOWN_FUNCTION( + interface_rx_dpo_interface_state_change); + +/** + * @brief Registered callback for HW interface state changes + */ +static clib_error_t * +interface_rx_dpo_hw_interface_state_change (vnet_main_t * vnm, + u32 hw_if_index, + u32 flags) +{ + return (NULL); +} + +VNET_HW_INTERFACE_LINK_UP_DOWN_FUNCTION( + interface_rx_dpo_hw_interface_state_change); + +static clib_error_t * +interface_rx_dpo_interface_delete (vnet_main_t * vnm, + u32 sw_if_index, + u32 is_add) +{ + return (NULL); +} + +VNET_SW_INTERFACE_ADD_DEL_FUNCTION( + interface_rx_dpo_interface_delete); + +u8* +format_interface_rx_dpo (u8* s, va_list *ap) +{ + index_t index = va_arg(*ap, index_t); + CLIB_UNUSED(u32 indent) = va_arg(*ap, u32); + vnet_main_t * vnm = vnet_get_main(); + interface_rx_dpo_t *ido = interface_rx_dpo_get(index); + + return (format(s, "%U-dpo: %U", + format_vnet_sw_interface_name, + vnm, + vnet_get_sw_interface(vnm, ido->ido_sw_if_index), + format_dpo_proto, ido->ido_proto)); +} + +static void +interface_rx_dpo_mem_show (void) +{ + fib_show_memory_usage("Interface", + pool_elts(interface_rx_dpo_pool), + pool_len(interface_rx_dpo_pool), + sizeof(interface_rx_dpo_t)); +} + + +const static dpo_vft_t interface_rx_dpo_vft = { + .dv_lock = interface_rx_dpo_lock, + .dv_unlock = interface_rx_dpo_unlock, + .dv_format = format_interface_rx_dpo, + .dv_mem_show = interface_rx_dpo_mem_show, +}; + +/** + * @brief The per-protocol VLIB graph nodes that are assigned to a glean + * object. + * + * this means that these graph nodes are ones from which a glean is the + * parent object in the DPO-graph. + */ +const static char* const interface_rx_dpo_ip4_nodes[] = +{ + "interface-rx-dpo-ip4", + NULL, +}; +const static char* const interface_rx_dpo_ip6_nodes[] = +{ + "interface-rx-dpo-ip6", + NULL, +}; +const static char* const interface_rx_dpo_l2_nodes[] = +{ + "interface-rx-dpo-l2", + NULL, +}; + +const static char* const * const interface_rx_dpo_nodes[DPO_PROTO_NUM] = +{ + [DPO_PROTO_IP4] = interface_rx_dpo_ip4_nodes, + [DPO_PROTO_IP6] = interface_rx_dpo_ip6_nodes, + [DPO_PROTO_ETHERNET] = interface_rx_dpo_l2_nodes, + [DPO_PROTO_MPLS] = NULL, +}; + +void +interface_rx_dpo_module_init (void) +{ + dpo_register(DPO_INTERFACE_RX, + &interface_rx_dpo_vft, + interface_rx_dpo_nodes); +} + +/** + * @brief Interface DPO trace data + */ +typedef struct interface_rx_dpo_trace_t_ +{ + u32 sw_if_index; +} interface_rx_dpo_trace_t; + +typedef enum interface_rx_dpo_next_t_ +{ + INTERFACE_RX_DPO_DROP = 0, + INTERFACE_RX_DPO_INPUT = 1, +} interface_rx_dpo_next_t; + +always_inline uword +interface_rx_dpo_inline (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + u32 n_left_from, next_index, * from, * to_next; + u32 thread_index = vlib_get_thread_index (); + vnet_interface_main_t *im; + + im = &vnet_get_main ()->interface_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 >= 4 && n_left_to_next > 2) + { + const interface_rx_dpo_t *ido0, *ido1; + u32 bi0, idoi0, bi1, idoi1; + vlib_buffer_t *b0, *b1; + + bi0 = from[0]; + to_next[0] = bi0; + bi1 = from[1]; + to_next[1] = bi1; + from += 2; + to_next += 2; + n_left_from -= 2; + n_left_to_next -= 2; + + b0 = vlib_get_buffer (vm, bi0); + b1 = vlib_get_buffer (vm, bi1); + + idoi0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX]; + idoi1 = vnet_buffer(b1)->ip.adj_index[VLIB_TX]; + ido0 = interface_rx_dpo_get(idoi0); + ido1 = interface_rx_dpo_get(idoi1); + + vnet_buffer(b0)->sw_if_index[VLIB_RX] = ido0->ido_sw_if_index; + vnet_buffer(b1)->sw_if_index[VLIB_RX] = ido1->ido_sw_if_index; + + vlib_increment_combined_counter (im->combined_sw_if_counters + + VNET_INTERFACE_COUNTER_RX, + thread_index, + ido0->ido_sw_if_index, + 1, + vlib_buffer_length_in_chain (vm, b0)); + vlib_increment_combined_counter (im->combined_sw_if_counters + + VNET_INTERFACE_COUNTER_RX, + thread_index, + ido1->ido_sw_if_index, + 1, + vlib_buffer_length_in_chain (vm, b1)); + + if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED)) + { + interface_rx_dpo_trace_t *tr0; + + tr0 = vlib_add_trace (vm, node, b0, sizeof (*tr0)); + tr0->sw_if_index = ido0->ido_sw_if_index; + } + if (PREDICT_FALSE(b1->flags & VLIB_BUFFER_IS_TRACED)) + { + interface_rx_dpo_trace_t *tr1; + + tr1 = vlib_add_trace (vm, node, b1, sizeof (*tr1)); + tr1->sw_if_index = ido1->ido_sw_if_index; + } + + vlib_validate_buffer_enqueue_x2(vm, node, next_index, to_next, + n_left_to_next, bi0, bi1, + INTERFACE_RX_DPO_INPUT, + INTERFACE_RX_DPO_INPUT); + } + + while (n_left_from > 0 && n_left_to_next > 0) + { + const interface_rx_dpo_t * ido0; + vlib_buffer_t * b0; + u32 bi0, idoi0; + + 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); + + idoi0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX]; + ido0 = interface_rx_dpo_get(idoi0); + + /* Swap the RX interface of the packet to the one the + * interface DPR represents */ + vnet_buffer(b0)->sw_if_index[VLIB_RX] = ido0->ido_sw_if_index; + + /* Bump the interface's RX coutners */ + vlib_increment_combined_counter (im->combined_sw_if_counters + + VNET_INTERFACE_COUNTER_RX, + thread_index, + ido0->ido_sw_if_index, + 1, + vlib_buffer_length_in_chain (vm, b0)); + + if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED)) + { + interface_rx_dpo_trace_t *tr; + + tr = vlib_add_trace (vm, node, b0, sizeof (*tr)); + tr->sw_if_index = ido0->ido_sw_if_index; + } + + vlib_validate_buffer_enqueue_x1(vm, node, next_index, to_next, + n_left_to_next, bi0, + INTERFACE_RX_DPO_INPUT); + } + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + return from_frame->n_vectors; +} + +static u8 * +format_interface_rx_dpo_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 *); + interface_rx_dpo_trace_t * t = va_arg (*args, interface_rx_dpo_trace_t *); + uword indent = format_get_indent (s); + s = format (s, "%U sw_if_index:%d", + format_white_space, indent, + t->sw_if_index); + return s; +} + +static uword +interface_rx_dpo_ip4 (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return (interface_rx_dpo_inline(vm, node, from_frame)); +} + +static uword +interface_rx_dpo_ip6 (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return (interface_rx_dpo_inline(vm, node, from_frame)); +} + +static uword +interface_rx_dpo_l2 (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return (interface_rx_dpo_inline(vm, node, from_frame)); +} + +VLIB_REGISTER_NODE (interface_rx_dpo_ip4_node) = { + .function = interface_rx_dpo_ip4, + .name = "interface-rx-dpo-ip4", + .vector_size = sizeof (u32), + .format_trace = format_interface_rx_dpo_trace, + + .n_next_nodes = 2, + .next_nodes = { + [INTERFACE_RX_DPO_DROP] = "ip4-drop", + [INTERFACE_RX_DPO_INPUT] = "ip4-input", + }, +}; + +VLIB_NODE_FUNCTION_MULTIARCH (interface_rx_dpo_ip4_node, + interface_rx_dpo_ip4) + +VLIB_REGISTER_NODE (interface_rx_dpo_ip6_node) = { + .function = interface_rx_dpo_ip6, + .name = "interface-rx-dpo-ip6", + .vector_size = sizeof (u32), + .format_trace = format_interface_rx_dpo_trace, + + .n_next_nodes = 2, + .next_nodes = { + [INTERFACE_RX_DPO_DROP] = "ip6-drop", + [INTERFACE_RX_DPO_INPUT] = "ip6-input", + }, +}; + +VLIB_NODE_FUNCTION_MULTIARCH (interface_rx_dpo_ip6_node, + interface_rx_dpo_ip6) + +VLIB_REGISTER_NODE (interface_rx_dpo_l2_node) = { + .function = interface_rx_dpo_l2, + .name = "interface-rx-dpo-l2", + .vector_size = sizeof (u32), + .format_trace = format_interface_rx_dpo_trace, + + .n_next_nodes = 2, + .next_nodes = { + [INTERFACE_RX_DPO_DROP] = "error-drop", + [INTERFACE_RX_DPO_INPUT] = "l2-input", + }, +}; + +VLIB_NODE_FUNCTION_MULTIARCH (interface_rx_dpo_l2_node, + interface_rx_dpo_l2) diff --git a/src/vnet/dpo/interface_rx_dpo.h b/src/vnet/dpo/interface_rx_dpo.h new file mode 100644 index 00000000..edecce08 --- /dev/null +++ b/src/vnet/dpo/interface_rx_dpo.h @@ -0,0 +1,69 @@ +/* + * 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 __INTERFACE_RX_DPO_H__ +#define __INTERFACE_RX_DPO_H__ + +#include + +/** + * @brief + * The data-path object representing a change of receive interface. + * If a packet encounters an object of this type in the data-path, it's + * RX interface is changed. + */ +typedef struct interface_rx_dpo_t_ +{ + /** + * The Software interface index that the packets will be given + * as the ingress/rx interface + */ + u32 ido_sw_if_index; + + /** + * next VLIB node. A '-input' node. + */ + u32 ido_next_node; + + /** + * DPO protocol that the packets will have as they 'ingress' + * on this interface + */ + dpo_proto_t ido_proto; + + /** + * number of locks. + */ + u16 ido_locks; +} interface_rx_dpo_t; + +extern void interface_rx_dpo_add_or_lock (dpo_proto_t proto, + u32 sw_if_index, + dpo_id_t *dpo); + +extern void interface_rx_dpo_module_init(void); + +/** + * @brief pool of all interface DPOs + */ +interface_rx_dpo_t *interface_rx_dpo_pool; + +static inline interface_rx_dpo_t * +interface_rx_dpo_get (index_t index) +{ + return (pool_elt_at_index(interface_rx_dpo_pool, index)); +} + +#endif diff --git a/src/vnet/dpo/interface_tx_dpo.c b/src/vnet/dpo/interface_tx_dpo.c new file mode 100644 index 00000000..f7c8bfda --- /dev/null +++ b/src/vnet/dpo/interface_tx_dpo.c @@ -0,0 +1,92 @@ +/* + * 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 +#include + +/* + * We do not lock nor unlock these DPOs since there is nothing to lock + * all we do is construct DPO object wrappers around a sw_if_index + */ +static void +interface_tx_dpo_lock (dpo_id_t *dpo) +{ +} + +static void +interface_tx_dpo_unlock (dpo_id_t *dpo) +{ +} + +/* + * interface_tx_dpo_add_or_lock + * + * construct DPO object wrappers around a sw_if_index + */ +void +interface_tx_dpo_add_or_lock (dpo_proto_t proto, + u32 sw_if_index, + dpo_id_t *dpo) +{ + dpo_set(dpo, DPO_INTERFACE_TX, proto, sw_if_index); +} + +u8* +format_interface_tx_dpo (u8* s, va_list *ap) +{ + index_t index = va_arg(*ap, index_t); + CLIB_UNUSED(u32 indent) = va_arg(*ap, u32); + vnet_main_t * vnm = vnet_get_main(); + + return (format(s, "%U-dpo:", + format_vnet_sw_interface_name, + vnm, + vnet_get_sw_interface(vnm, index))); +} + +static void +interface_tx_dpo_mem_show (void) +{ +} + +u32* +interface_tx_dpo_get_next_node (const dpo_id_t *dpo) +{ + u32 *node_indices = NULL; + + /* + * return the interface's TX node for the wrapped sw_if_index + */ + vec_add1(node_indices, + vnet_tx_node_index_for_sw_interface(vnet_get_main(), + dpo->dpoi_index)); + + return (node_indices); +} + +const static dpo_vft_t interface_tx_dpo_vft = { + .dv_lock = interface_tx_dpo_lock, + .dv_unlock = interface_tx_dpo_unlock, + .dv_format = format_interface_tx_dpo, + .dv_mem_show = interface_tx_dpo_mem_show, + .dv_get_next_node = interface_tx_dpo_get_next_node, +}; + +void +interface_tx_dpo_module_init (void) +{ + dpo_register(DPO_INTERFACE_TX, &interface_tx_dpo_vft, NULL); +} + diff --git a/src/vnet/dpo/interface_tx_dpo.h b/src/vnet/dpo/interface_tx_dpo.h new file mode 100644 index 00000000..0c560ada --- /dev/null +++ b/src/vnet/dpo/interface_tx_dpo.h @@ -0,0 +1,33 @@ +/* + * 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. + */ +/** + * @brief + * The data-path object representing transmitting the packet on a n interface. + * This is a convenient DPO wrapper around a simple interface transmit and thus + * allows us to represent direct interface transmit in the DPO model. + */ + +#ifndef __INTERFACE_TX_DPO_H__ +#define __INTERFACE_TX_DPO_H__ + +#include + +extern void interface_tx_dpo_add_or_lock (dpo_proto_t proto, + u32 sw_if_index, + dpo_id_t *dpo); + +extern void interface_tx_dpo_module_init(void); + +#endif diff --git a/src/vnet/fib/fib_path.c b/src/vnet/fib/fib_path.c index 58050ccb..cf5a463d 100644 --- a/src/vnet/fib/fib_path.c +++ b/src/vnet/fib/fib_path.c @@ -21,7 +21,7 @@ #include #include #include -#include +#include #include #include @@ -1707,9 +1707,9 @@ fib_path_resolve (fib_node_index_t path_index) /* * Resolve via a receive DPO. */ - interface_dpo_add_or_lock(path->fp_nh_proto, - path->intf_rx.fp_interface, - &path->fp_dpo); + interface_rx_dpo_add_or_lock(path->fp_nh_proto, + path->intf_rx.fp_interface, + &path->fp_dpo); break; } case FIB_PATH_TYPE_EXCLUSIVE: @@ -2041,9 +2041,9 @@ fib_path_contribute_forwarding (fib_node_index_t path_index, /* * Create the adj needed for sending IP multicast traffic */ - interface_dpo_add_or_lock(fib_forw_chain_type_to_dpo_proto(fct), - path->attached.fp_interface, - dpo); + interface_rx_dpo_add_or_lock(fib_forw_chain_type_to_dpo_proto(fct), + path->attached.fp_interface, + dpo); break; case FIB_PATH_TYPE_RECEIVE: case FIB_PATH_TYPE_SPECIAL: diff --git a/src/vnet/fib/fib_test.c b/src/vnet/fib/fib_test.c index 7ca4cb3a..6867cca8 100644 --- a/src/vnet/fib/fib_test.c +++ b/src/vnet/fib/fib_test.c @@ -25,7 +25,7 @@ #include #include #include -#include +#include #include #include @@ -407,7 +407,7 @@ fib_test_validate_rep_v (const replicate_t *rep, } break; case FT_REP_INTF: - FIB_TEST_LB((DPO_INTERFACE == dpo->dpoi_type), + FIB_TEST_LB((DPO_INTERFACE_RX == dpo->dpoi_type), "bucket %d stacks on %U", bucket, format_dpo_type, dpo->dpoi_type); @@ -589,7 +589,7 @@ fib_test_validate_lb_v (const load_balance_t *lb, exp->adj.adj); break; case FT_LB_INTF: - FIB_TEST_I((DPO_INTERFACE == dpo->dpoi_type), + FIB_TEST_I((DPO_INTERFACE_RX == dpo->dpoi_type), "bucket %d stacks on %U", bucket, format_dpo_type, dpo->dpoi_type); @@ -8523,7 +8523,7 @@ lfib_test (void) */ dpo_id_t idpo = DPO_INVALID; - interface_dpo_add_or_lock(DPO_PROTO_IP4, + interface_rx_dpo_add_or_lock(DPO_PROTO_IP4, tm->hw[0]->sw_if_index, &idpo); @@ -8667,9 +8667,9 @@ lfib_test (void) FIB_TEST(lb_count == pool_elts(load_balance_pool), "Load-balance resources freed %d of %d", lb_count, pool_elts(load_balance_pool)); - FIB_TEST(0 == pool_elts(interface_dpo_pool), - "interface_dpo resources freed %d of %d", - 0, pool_elts(interface_dpo_pool)); + FIB_TEST(0 == pool_elts(interface_rx_dpo_pool), + "interface_rx_dpo resources freed %d of %d", + 0, pool_elts(interface_rx_dpo_pool)); return (0); } diff --git a/test/test_pppoe.py b/test/test_pppoe.py index 0baf4546..1d0aeffd 100644 --- a/test/test_pppoe.py +++ b/test/test_pppoe.py @@ -281,7 +281,7 @@ class TestPPPoE(VppTestCase): # self.vapi.cli("clear trace") tx2 = self.create_stream_ip4(self.pg1, self.pg0, - self.pg0.remote_ip4, self.dst_ip) + self.pg0.remote_ip4, self.dst_ip, 65) self.pg1.add_stream(tx2) self.pg_enable_capture(self.pg_interfaces) @@ -293,6 +293,7 @@ class TestPPPoE(VppTestCase): self.logger.info(self.vapi.cli("show pppoe fib")) self.logger.info(self.vapi.cli("show pppoe session")) self.logger.info(self.vapi.cli("show ip fib")) + self.logger.info(self.vapi.cli("show adj")) # # test case cleanup -- cgit 1.2.3-korg