diff options
Diffstat (limited to 'src')
| -rw-r--r-- | src/vlib/buffer_funcs.h | 62 | ||||
| -rw-r--r-- | src/vnet.am | 2 | ||||
| -rw-r--r-- | src/vnet/buffer.h | 12 | ||||
| -rw-r--r-- | src/vnet/ip/icmp46_packet.h | 1 | ||||
| -rw-r--r-- | src/vnet/ip/icmp6.h | 2 | ||||
| -rw-r--r-- | src/vnet/ip/ip.api | 28 | ||||
| -rw-r--r-- | src/vnet/ip/ip4_error.h | 9 | ||||
| -rwxr-xr-x | src/vnet/ip/ip4_forward.c | 13 | ||||
| -rw-r--r-- | src/vnet/ip/ip4_input.c | 2 | ||||
| -rw-r--r-- | src/vnet/ip/ip4_input.h | 1 | ||||
| -rw-r--r-- | src/vnet/ip/ip4_reassembly.c | 1277 | ||||
| -rw-r--r-- | src/vnet/ip/ip4_reassembly.h | 49 | ||||
| -rw-r--r-- | src/vnet/ip/ip6_error.h | 9 | ||||
| -rw-r--r-- | src/vnet/ip/ip6_forward.c | 2 | ||||
| -rw-r--r-- | src/vnet/ip/ip6_input.h | 1 | ||||
| -rw-r--r-- | src/vnet/ip/ip6_packet.h | 3 | ||||
| -rw-r--r-- | src/vnet/ip/ip6_reassembly.c | 1347 | ||||
| -rw-r--r-- | src/vnet/ip/ip6_reassembly.h | 49 | ||||
| -rw-r--r-- | src/vnet/ip/ip_api.c | 61 | ||||
| -rw-r--r-- | src/vnet/ip/lookup.c | 2 | ||||
| -rw-r--r-- | src/vnet/ip/lookup.h | 1 | ||||
| -rw-r--r-- | src/vnet/ip/protocols.def | 5 |
22 files changed, 2930 insertions, 8 deletions
diff --git a/src/vlib/buffer_funcs.h b/src/vlib/buffer_funcs.h index 4831eb5ebac..cc56db7762c 100644 --- a/src/vlib/buffer_funcs.h +++ b/src/vlib/buffer_funcs.h @@ -1004,6 +1004,68 @@ vlib_validate_buffer_set_in_use (vlib_buffer_t * b, u32 expected) #endif } +/** minimum data size of first buffer in a buffer chain */ +#define VLIB_BUFFER_CHAIN_MIN_FIRST_DATA_SIZE (256) + +/** + * @brief compress buffer chain in a way where the first buffer is at least + * VLIB_BUFFER_CHAIN_MIN_FIRST_DATA_SIZE long + * + * @param[in] vm - vlib_main + * @param[in,out] first - first buffer in chain + * @param[in,out] discard_vector - vector of buffer indexes which were removed + * from the chain + */ +always_inline void +vlib_buffer_chain_compress (vlib_main_t * vm, + vlib_buffer_t * first, u32 ** discard_vector) +{ + if (first->current_length >= VLIB_BUFFER_CHAIN_MIN_FIRST_DATA_SIZE || + !(first->flags & VLIB_BUFFER_NEXT_PRESENT)) + { + /* this is already big enough or not a chain */ + return; + } + /* probe free list to find allocated buffer size to avoid overfill */ + u32 index; + vlib_buffer_free_list_t *free_list = + vlib_buffer_get_buffer_free_list (vm, first, &index); + + u32 want_first_size = clib_min (VLIB_BUFFER_CHAIN_MIN_FIRST_DATA_SIZE, + free_list->n_data_bytes - + first->current_data); + do + { + vlib_buffer_t *second = vlib_get_buffer (vm, first->next_buffer); + u32 need = want_first_size - first->current_length; + u32 amount_to_copy = clib_min (need, second->current_length); + clib_memcpy (((u8 *) vlib_buffer_get_current (first)) + + first->current_length, + vlib_buffer_get_current (second), amount_to_copy); + first->current_length += amount_to_copy; + vlib_buffer_advance (second, amount_to_copy); + if (first->flags & VLIB_BUFFER_TOTAL_LENGTH_VALID) + { + first->total_length_not_including_first_buffer -= amount_to_copy; + } + if (!second->current_length) + { + vec_add1 (*discard_vector, first->next_buffer); + if (second->flags & VLIB_BUFFER_NEXT_PRESENT) + { + first->next_buffer = second->next_buffer; + } + else + { + first->flags &= ~VLIB_BUFFER_NEXT_PRESENT; + } + second->flags &= ~VLIB_BUFFER_NEXT_PRESENT; + } + } + while ((first->current_length < want_first_size) && + (first->flags & VLIB_BUFFER_NEXT_PRESENT)); +} + #endif /* included_vlib_buffer_funcs_h */ /* diff --git a/src/vnet.am b/src/vnet.am index 7eb4159e75c..8f884748d0f 100644 --- a/src/vnet.am +++ b/src/vnet.am @@ -343,6 +343,7 @@ libvnet_la_SOURCES += \ vnet/ip/ip4_pg.c \ vnet/ip/ip4_source_and_port_range_check.c \ vnet/ip/ip4_source_check.c \ + vnet/ip/ip4_reassembly.c \ vnet/ip/ip6_format.c \ vnet/ip/ip6_forward.c \ vnet/ip/ip6_punt_drop.c \ @@ -350,6 +351,7 @@ libvnet_la_SOURCES += \ vnet/ip/ip6_input.c \ vnet/ip/ip6_neighbor.c \ vnet/ip/ip6_pg.c \ + vnet/ip/ip6_reassembly.c \ vnet/ip/ip_api.c \ vnet/ip/ip_checksum.c \ vnet/ip/ip_frag.c \ diff --git a/src/vnet/buffer.h b/src/vnet/buffer.h index c9f1d6ef6a1..5d7273aa04f 100644 --- a/src/vnet/buffer.h +++ b/src/vnet/buffer.h @@ -177,6 +177,18 @@ typedef struct u8 code; u32 data; } icmp; + + /* reassembly */ + struct + { + u16 fragment_first; + u16 fragment_last; + u16 range_first; + u16 range_last; + u32 next_range_bi; + u16 ip6_frag_hdr_offset; + u16 estimated_mtu; + } reass; }; } ip; diff --git a/src/vnet/ip/icmp46_packet.h b/src/vnet/ip/icmp46_packet.h index a86cbd57bdb..e761af8bf16 100644 --- a/src/vnet/ip/icmp46_packet.h +++ b/src/vnet/ip/icmp46_packet.h @@ -149,6 +149,7 @@ _ (parameter_problem, 0, erroneous_header_field) \ _ (parameter_problem, 1, unrecognized_next_header) \ _ (parameter_problem, 2, unrecognized_option) \ + _ (parameter_problem, 3, first_fragment_has_incomplete_header_chain) \ _ (router_renumbering, 0, command) \ _ (router_renumbering, 1, result) \ _ (node_information_request, 0, data_contains_ip6_address) \ diff --git a/src/vnet/ip/icmp6.h b/src/vnet/ip/icmp6.h index 9a3487b1c5c..1cac8240466 100644 --- a/src/vnet/ip/icmp6.h +++ b/src/vnet/ip/icmp6.h @@ -45,7 +45,7 @@ _ (DEST_UNREACH_SENT, "destination unreachable response sent") \ _ (PACKET_TOO_BIG_SENT, "packet too big response sent") \ _ (TTL_EXPIRE_SENT, "hop limit exceeded response sent") \ - _ (PARAM_PROBLEM_SENT, "parameter Pproblem response sent") \ + _ (PARAM_PROBLEM_SENT, "parameter problem response sent") \ _ (DROP, "error message dropped") diff --git a/src/vnet/ip/ip.api b/src/vnet/ip/ip.api index 0da0a61fb75..6ed5a9d6bdb 100644 --- a/src/vnet/ip/ip.api +++ b/src/vnet/ip/ip.api @@ -812,6 +812,34 @@ autoreply define ioam_disable u16 id; }; +autoreply define ip_reassembly_set +{ + u32 client_index; + u32 context; + u32 timeout_ms; + u32 max_reassemblies; + u32 expire_walk_interval_ms; + u8 is_ip6; +}; + +define ip_reassembly_get +{ + u32 client_index; + u32 context; + u8 is_ip6; +}; + +define ip_reassembly_get_reply +{ + u32 client_index; + u32 context; + i32 retval; + u32 timeout_ms; + u32 max_reassemblies; + u32 expire_walk_interval_ms; + u8 is_ip6; +}; + /* * Local Variables: * eval: (c-set-style "gnu") diff --git a/src/vnet/ip/ip4_error.h b/src/vnet/ip/ip4_error.h index 4e545aeda3c..00b1c6daad5 100644 --- a/src/vnet/ip/ip4_error.h +++ b/src/vnet/ip/ip4_error.h @@ -71,14 +71,19 @@ _ (UNICAST_SOURCE_CHECK_FAILS, "ip4 unicast source check fails") \ \ /* Spoofed packets in ip4-rewrite-local */ \ - _(SPOOFED_LOCAL_PACKETS, "ip4 spoofed local-address packet drops") \ + _ (SPOOFED_LOCAL_PACKETS, "ip4 spoofed local-address packet drops") \ \ /* Errors singalled by ip4-inacl */ \ _ (INACL_TABLE_MISS, "input ACL table-miss drops") \ _ (INACL_SESSION_DENY, "input ACL session deny drops") \ \ /* Erros from mfib-forward */ \ - _ (RPF_FAILURE, "Multicast RPF check failed") + _ (RPF_FAILURE, "Multicast RPF check failed") \ + \ + /* Errors signalled by ip4-reassembly */ \ + _ (REASS_DUPLICATE_FRAGMENT, "duplicate/overlapping fragments") \ + _ (REASS_LIMIT_REACHED, "drops due to concurrent reassemblies limit") \ + _ (REASS_TIMEOUT, "fragments dropped due to reassembly timeout") typedef enum { diff --git a/src/vnet/ip/ip4_forward.c b/src/vnet/ip/ip4_forward.c index 3ddf6dfa783..2a21135bb3e 100755 --- a/src/vnet/ip/ip4_forward.c +++ b/src/vnet/ip/ip4_forward.c @@ -1582,8 +1582,12 @@ ip4_local_inline (vlib_main_t * vm, /* 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; - proto1 = ip4_is_fragment (ip1) ? 0xfe : ip1->protocol; + proto0 = + ip4_is_fragment (ip0) ? IP_PROTOCOL_VPP_FRAGMENTATION : + ip0->protocol; + proto1 = + ip4_is_fragment (ip1) ? IP_PROTOCOL_VPP_FRAGMENTATION : + ip1->protocol; if (head_of_feature_arc == 0) goto skip_checks; @@ -1748,7 +1752,9 @@ ip4_local_inline (vlib_main_t * vm, /* 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; + proto0 = + ip4_is_fragment (ip0) ? IP_PROTOCOL_VPP_FRAGMENTATION : + ip0->protocol; if (head_of_feature_arc == 0 || p0->flags & VNET_BUFFER_F_IS_NATED) goto skip_check; @@ -1839,6 +1845,7 @@ VLIB_REGISTER_NODE (ip4_local_node) = [IP_LOCAL_NEXT_PUNT] = "ip4-punt", [IP_LOCAL_NEXT_UDP_LOOKUP] = "ip4-udp-lookup", [IP_LOCAL_NEXT_ICMP] = "ip4-icmp-input", + [IP_LOCAL_NEXT_REASSEMBLY] = "ip4-reassembly", }, }; /* *INDENT-ON* */ diff --git a/src/vnet/ip/ip4_input.c b/src/vnet/ip/ip4_input.c index 121f40f473d..61c86ef112d 100644 --- a/src/vnet/ip/ip4_input.c +++ b/src/vnet/ip/ip4_input.c @@ -289,6 +289,7 @@ VLIB_REGISTER_NODE (ip4_input_node) = { [IP4_INPUT_NEXT_LOOKUP] = "ip4-lookup", [IP4_INPUT_NEXT_LOOKUP_MULTICAST] = "ip4-mfib-forward-lookup", [IP4_INPUT_NEXT_ICMP_ERROR] = "ip4-icmp-error", + [IP4_INPUT_NEXT_REASSEMBLY] = "ip4-reassembly", }, .format_buffer = format_ip4_header, @@ -311,6 +312,7 @@ VLIB_REGISTER_NODE (ip4_input_no_checksum_node,static) = { [IP4_INPUT_NEXT_LOOKUP] = "ip4-lookup", [IP4_INPUT_NEXT_LOOKUP_MULTICAST] = "ip4-mfib-forward-lookup", [IP4_INPUT_NEXT_ICMP_ERROR] = "ip4-icmp-error", + [IP4_INPUT_NEXT_REASSEMBLY] = "ip4-reassembly", }, .format_buffer = format_ip4_header, diff --git a/src/vnet/ip/ip4_input.h b/src/vnet/ip/ip4_input.h index 75306a34db3..2692848f8ad 100644 --- a/src/vnet/ip/ip4_input.h +++ b/src/vnet/ip/ip4_input.h @@ -52,6 +52,7 @@ typedef enum IP4_INPUT_NEXT_LOOKUP, IP4_INPUT_NEXT_LOOKUP_MULTICAST, IP4_INPUT_NEXT_ICMP_ERROR, + IP4_INPUT_NEXT_REASSEMBLY, IP4_INPUT_N_NEXT, } ip4_input_next_t; diff --git a/src/vnet/ip/ip4_reassembly.c b/src/vnet/ip/ip4_reassembly.c new file mode 100644 index 00000000000..6b8d66532fa --- /dev/null +++ b/src/vnet/ip/ip4_reassembly.c @@ -0,0 +1,1277 @@ +/* + * 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 IPv4 Reassembly. + * + * This file contains the source code for IPv4 reassembly. + */ + +#include <vppinfra/vec.h> +#include <vnet/vnet.h> +#include <vnet/ip/ip.h> +#include <vppinfra/bihash_24_8.h> +#include <vnet/ip/ip4_reassembly.h> + +#define MSEC_PER_SEC 1000 +#define IP4_REASS_TIMEOUT_DEFAULT_MS 100 +#define IP4_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS 10000 // 10 seconds default +#define IP4_REASS_MAX_REASSEMBLIES_DEAFULT 1024 +#define IP4_REASS_HT_LOAD_FACTOR (0.75) + +#define IP4_REASS_DEBUG_BUFFERS 0 +#if IP4_REASS_DEBUG_BUFFERS +#define IP4_REASS_DEBUG_BUFFER(bi, what) \ + do \ + { \ + u32 _bi = bi; \ + printf (#what "buffer %u", _bi); \ + vlib_buffer_t *_b = vlib_get_buffer (vm, _bi); \ + while (_b->flags & VLIB_BUFFER_NEXT_PRESENT) \ + { \ + _bi = _b->next_buffer; \ + printf ("[%u]", _bi); \ + _b = vlib_get_buffer (vm, _bi); \ + } \ + printf ("\n"); \ + fflush (stdout); \ + } \ + while (0) +#else +#define IP4_REASS_DEBUG_BUFFER(...) +#endif + +static vlib_node_registration_t ip4_reass_node; + +typedef struct +{ + union + { + struct + { + // align by making this 4 octets even though its a 2 octets field + u32 xx_id; + ip4_address_t src; + ip4_address_t dst; + // align by making this 4 octets even though its a 2 octets field + u32 frag_id; + // align by making this 4 octets even though its a 1 octet field + u32 proto; + u32 unused; + }; + u64 as_u64[3]; + }; +} ip4_reass_key_t; + +always_inline u32 +ip4_reass_buffer_get_data_offset_no_check (vlib_buffer_t * b) +{ + vnet_buffer_opaque_t *vnb = vnet_buffer (b); + return vnb->ip.reass.range_first - vnb->ip.reass.fragment_first; +} + +always_inline u32 +ip4_reass_buffer_get_data_offset (vlib_buffer_t * b) +{ + vnet_buffer_opaque_t *vnb = vnet_buffer (b); + ASSERT (vnb->ip.reass.range_first >= vnb->ip.reass.fragment_first); + return ip4_reass_buffer_get_data_offset_no_check (b); +} + +always_inline u16 +ip4_reass_buffer_get_data_len_no_check (vlib_buffer_t * b) +{ + vnet_buffer_opaque_t *vnb = vnet_buffer (b); + return clib_min (vnb->ip.reass.range_last, vnb->ip.reass.fragment_last) - + (vnb->ip.reass.fragment_first + ip4_reass_buffer_get_data_offset (b)) + 1; +} + +always_inline u16 +ip4_reass_buffer_get_data_len (vlib_buffer_t * b) +{ + vnet_buffer_opaque_t *vnb = vnet_buffer (b); + ASSERT (vnb->ip.reass.range_last > vnb->ip.reass.fragment_first); + return ip4_reass_buffer_get_data_len_no_check (b); +} + +typedef struct +{ + // hash table key + ip4_reass_key_t key; + f64 first_heard; + // time when last packet was received + f64 last_heard; + // internal id of this reassembly + u32 id; + // buffer index of first buffer in this reassembly context + u32 first_bi; + // last octet of packet, ~0 until fragment without more_fragments arrives + u32 last_packet_octet; + // length of data collected so far + u32 data_len; + // trace operation counter + u32 trace_op_counter; +} ip4_reass_t; + +typedef struct +{ + // IPv4 config + u32 timeout_ms; + f64 timeout; + u32 expire_walk_interval_ms; + u32 max_reass_n; + + // IPv4 runtime + ip4_reass_t *pool; + clib_bihash_24_8_t hash; + u32 reass_n; + u32 id_counter; + u32 buffers_n; + + // convenience + vlib_main_t *vlib_main; + vnet_main_t *vnet_main; + + // node index of ip4-drop node + u32 ip4_drop_idx; + u32 ip4_reass_expire_node_idx; + +} ip4_reass_main_t; + +ip4_reass_main_t ip4_reass_main; + +typedef enum +{ + IP4_REASSEMBLY_NEXT_INPUT, + IP4_REASSEMBLY_NEXT_DROP, + IP4_REASSEMBLY_N_NEXT, +} ip4_reass_next_t; + +typedef enum +{ + RANGE_NEW, + RANGE_SHRINK, + RANGE_DISCARD, + RANGE_OVERLAP, + FINALIZE, +} ip4_reass_trace_operation_e; + +typedef struct +{ + u16 range_first; + u16 range_last; + u32 range_bi; + i32 data_offset; + u32 data_len; + u32 first_bi; +} ip4_reass_range_trace_t; + +typedef struct +{ + ip4_reass_trace_operation_e action; + u32 pool_index; + u32 reass_id; + ip4_reass_range_trace_t trace_range; + u32 size_diff; + u32 op_id; + u32 fragment_first; + u32 fragment_last; + u32 total_data_len; +} ip4_reass_trace_t; + +void +ip4_reass_trace_details (vlib_main_t * vm, u32 bi, + ip4_reass_range_trace_t * trace) +{ + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + vnet_buffer_opaque_t *vnb = vnet_buffer (b); + trace->range_first = vnb->ip.reass.range_first; + trace->range_last = vnb->ip.reass.range_last; + trace->data_offset = ip4_reass_buffer_get_data_offset_no_check (b); + trace->data_len = ip4_reass_buffer_get_data_len_no_check (b); + trace->range_bi = bi; +} + +u8 * +format_ip4_reass_range_trace (u8 * s, va_list * args) +{ + ip4_reass_range_trace_t *trace = va_arg (*args, ip4_reass_range_trace_t *); + s = format (s, "range: [%u, %u], off %d, len %u, bi %u", trace->range_first, + trace->range_last, trace->data_offset, trace->data_len, + trace->range_bi); + return s; +} + +u8 * +format_ip4_reass_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 *); + ip4_reass_trace_t *t = va_arg (*args, ip4_reass_trace_t *); + s = format (s, "reass id: %u, op id: %u ", t->reass_id, t->op_id); + u32 indent = format_get_indent (s); + s = format (s, "first bi: %u, data len: %u, ip/fragment[%u, %u]", + t->trace_range.first_bi, t->total_data_len, t->fragment_first, + t->fragment_last); + switch (t->action) + { + case RANGE_SHRINK: + s = format (s, "\n%Ushrink %U by %u", format_white_space, indent, + format_ip4_reass_range_trace, &t->trace_range, + t->size_diff); + break; + case RANGE_DISCARD: + s = format (s, "\n%Udiscard %U", format_white_space, indent, + format_ip4_reass_range_trace, &t->trace_range); + break; + case RANGE_NEW: + s = format (s, "\n%Unew %U", format_white_space, indent, + format_ip4_reass_range_trace, &t->trace_range); + break; + case RANGE_OVERLAP: + s = format (s, "\n%Uoverlapping/ignored %U", format_white_space, indent, + format_ip4_reass_range_trace, &t->trace_range); + break; + case FINALIZE: + s = format (s, "\n%Ufinalize reassembly", format_white_space, indent); + break; + } + return s; +} + +void +ip4_reass_add_trace (vlib_main_t * vm, vlib_node_runtime_t * node, + ip4_reass_main_t * rm, ip4_reass_t * reass, u32 bi, + ip4_reass_trace_operation_e action, u32 size_diff) +{ + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + vnet_buffer_opaque_t *vnb = vnet_buffer (b); + ip4_reass_trace_t *t = vlib_add_trace (vm, node, b, sizeof (t[0])); + t->pool_index = reass - rm->pool; + t->reass_id = reass->id; + t->action = action; + ip4_reass_trace_details (vm, bi, &t->trace_range); + t->size_diff = size_diff; + t->op_id = reass->trace_op_counter; + ++reass->trace_op_counter; + t->fragment_first = vnb->ip.reass.fragment_first; + t->fragment_last = vnb->ip.reass.fragment_last; + t->trace_range.first_bi = reass->first_bi; + t->total_data_len = reass->data_len; +#if 0 + static u8 *s = NULL; + s = format (s, "%U", format_ip4_reass_trace, NULL, NULL, t); + printf ("%.*s\n", vec_len (s), s); + fflush (stdout); + vec_reset_length (s); +#endif +} + +void +ip4_reass_free (ip4_reass_main_t * rm, ip4_reass_t * reass) +{ + clib_bihash_kv_24_8_t kv; + kv.key[0] = reass->key.as_u64[0]; + kv.key[1] = reass->key.as_u64[1]; + kv.key[2] = reass->key.as_u64[2]; + clib_bihash_add_del_24_8 (&rm->hash, &kv, 0); + pool_put (rm->pool, reass); + --rm->reass_n; +} + +static void +ip4_reass_on_timeout (vlib_main_t * vm, ip4_reass_main_t * rm, + ip4_reass_t * reass, u32 ** vec_drop_timeout) +{ + u32 range_bi = reass->first_bi; + vlib_buffer_t *range_b; + vnet_buffer_opaque_t *range_vnb; + while (~0 != range_bi) + { + range_b = vlib_get_buffer (vm, range_bi); + range_vnb = vnet_buffer (range_b); + u32 bi = range_bi; + while (~0 != bi) + { + vec_add1 (*vec_drop_timeout, bi); + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + if (b->flags & VLIB_BUFFER_NEXT_PRESENT) + { + bi = b->next_buffer; + b->flags &= ~VLIB_BUFFER_NEXT_PRESENT; + } + else + { + bi = ~0; + } + } + range_bi = range_vnb->ip.reass.next_range_bi; + } +} + +ip4_reass_t * +ip4_reass_find_or_create (vlib_main_t * vm, ip4_reass_main_t * rm, + ip4_reass_key_t * k, u32 ** vec_drop_timeout) +{ + ip4_reass_t *reass = NULL; + f64 now = vlib_time_now (rm->vlib_main); + clib_bihash_kv_24_8_t kv, value; + kv.key[0] = k->as_u64[0]; + kv.key[1] = k->as_u64[1]; + kv.key[2] = k->as_u64[2]; + + if (!clib_bihash_search_24_8 (&rm->hash, &kv, &value)) + { + reass = pool_elt_at_index (rm->pool, value.value); + if (now > reass->last_heard + rm->timeout) + { + ip4_reass_on_timeout (vm, rm, reass, vec_drop_timeout); + ip4_reass_free (rm, reass); + reass = NULL; + } + } + + if (reass) + { + reass->last_heard = now; + return reass; + } + + if (rm->reass_n >= rm->max_reass_n) + { + reass = NULL; + return reass; + } + else + { + pool_get (rm->pool, reass); + memset (reass, 0, sizeof (*reass)); + reass->id = rm->id_counter; + ++rm->id_counter; + reass->first_bi = ~0; + reass->last_packet_octet = ~0; + reass->data_len = 0; + ++rm->reass_n; + } + + reass->key.as_u64[0] = kv.key[0] = k->as_u64[0]; + reass->key.as_u64[1] = kv.key[1] = k->as_u64[1]; + reass->key.as_u64[2] = kv.key[2] = k->as_u64[2]; + kv.value = reass - rm->pool; + reass->last_heard = now; + + if (clib_bihash_add_del_24_8 (&rm->hash, &kv, 1)) + { + ip4_reass_free (rm, reass); + reass = NULL; + } + + return reass; +} + +void +ip4_reass_finalize (vlib_main_t * vm, vlib_node_runtime_t * node, + ip4_reass_main_t * rm, ip4_reass_t * reass, u32 * bi0, + u32 * next0, vlib_error_t * error0, u32 next_input, + u32 ** vec_drop_compress, u32 ** vec_drop_overlap) +{ + ASSERT (~0 != reass->first_bi); + vlib_buffer_t *first_b = vlib_get_buffer (vm, reass->first_bi); + vlib_buffer_t *last_b = NULL; + u32 sub_chain_bi = reass->first_bi; + u32 total_length = 0; + u32 buf_cnt = 0; + u32 dropped_cnt = 0; + do + { + u32 tmp_bi = sub_chain_bi; + vlib_buffer_t *tmp = vlib_get_buffer (vm, tmp_bi); + ip4_header_t *ip = vlib_buffer_get_current (tmp); + u32 data_len = ip4_reass_buffer_get_data_len (tmp); + u32 trim_front = + ip4_header_bytes (ip) + ip4_reass_buffer_get_data_offset (tmp); + u32 trim_end = + vlib_buffer_length_in_chain (vm, tmp) - trim_front - data_len; + if (tmp_bi == reass->first_bi) + { + /* first buffer - keep ip4 header */ + ASSERT (0 == ip4_reass_buffer_get_data_offset (tmp)); + trim_front = 0; + trim_end = vlib_buffer_length_in_chain (vm, tmp) - data_len - + ip4_header_bytes (ip); + } + u32 keep_data = + vlib_buffer_length_in_chain (vm, tmp) - trim_front - trim_end; + while (1) + { + ++buf_cnt; + if (trim_front) + { + if (trim_front > tmp->current_length) + { + /* drop whole buffer */ + vec_add1 (*vec_drop_compress, tmp_bi); + ++dropped_cnt; + trim_front -= tmp->current_length; + ASSERT (tmp->flags & VLIB_BUFFER_NEXT_PRESENT); + tmp->flags &= ~VLIB_BUFFER_NEXT_PRESENT; + tmp_bi = tmp->next_buffer; + tmp = vlib_get_buffer (vm, tmp_bi); + continue; + } + else + { + vlib_buffer_advance (tmp, trim_front); + trim_front = 0; + } + } + if (keep_data) + { + if (last_b) + { + last_b->flags |= VLIB_BUFFER_NEXT_PRESENT; + last_b->next_buffer = tmp_bi; + } + last_b = tmp; + if (keep_data <= tmp->current_length) + { + tmp->current_length = keep_data; + keep_data = 0; + } + else + { + keep_data -= tmp->current_length; + ASSERT (tmp->flags & VLIB_BUFFER_NEXT_PRESENT); + } + total_length += tmp->current_length; + } + else + { + vec_add1 (*vec_drop_overlap, tmp_bi); + ASSERT (reass->first_bi != tmp_bi); + ++dropped_cnt; + } + if (tmp->flags & VLIB_BUFFER_NEXT_PRESENT) + { + tmp_bi = tmp->next_buffer; + tmp = vlib_get_buffer (vm, tmp->next_buffer); + } + else + { + break; + } + } + sub_chain_bi = + vnet_buffer (vlib_get_buffer (vm, sub_chain_bi))->ip. + reass.next_range_bi; + } + while (~0 != sub_chain_bi); + last_b->flags &= ~VLIB_BUFFER_NEXT_PRESENT; + ASSERT (rm->buffers_n >= (buf_cnt - dropped_cnt)); + rm->buffers_n -= buf_cnt - dropped_cnt; + ASSERT (total_length >= first_b->current_length); + total_length -= first_b->current_length; + first_b->flags |= VLIB_BUFFER_TOTAL_LENGTH_VALID; + first_b->total_length_not_including_first_buffer = total_length; + ip4_header_t *ip = vlib_buffer_get_current (first_b); + ip->flags_and_fragment_offset = 0; + ip->length = clib_host_to_net_u16 (first_b->current_length + total_length); + ip->checksum = ip4_header_checksum (ip); + vlib_buffer_chain_compress (vm, first_b, vec_drop_compress); + if (PREDICT_FALSE (first_b->flags & VLIB_BUFFER_IS_TRACED)) + { + ip4_reass_add_trace (vm, node, rm, reass, reass->first_bi, FINALIZE, 0); +#if 0 + // following code does a hexdump of packet fragments to stdout ... + do + { + u32 bi = reass->first_bi; + u8 *s = NULL; + while (~0 != bi) + { + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + s = format (s, "%u: %U\n", bi, format_hexdump, + vlib_buffer_get_current (b), b->current_length); + if (b->flags & VLIB_BUFFER_NEXT_PRESENT) + { + bi = b->next_buffer; + } + else + { + break; + } + } + printf ("%.*s\n", vec_len (s), s); + fflush (stdout); + vec_free (s); + } + while (0); +#endif + } + *bi0 = reass->first_bi; + *next0 = next_input; + *error0 = IP4_ERROR_NONE; + ip4_reass_free (rm, reass); + reass = NULL; +} + +static u32 +ip4_reass_get_buffer_chain_length (vlib_main_t * vm, vlib_buffer_t * b) +{ + u32 len = 0; + while (b) + { + ++len; + if (PREDICT_FALSE (b->flags & VLIB_BUFFER_NEXT_PRESENT)) + { + b = vlib_get_buffer (vm, b->next_buffer); + } + else + { + break; + } + } + return len; +} + +static void +ip4_reass_insert_range_in_chain (vlib_main_t * vm, + ip4_reass_main_t * rm, + ip4_reass_t * reass, + u32 prev_range_bi, u32 new_next_bi) +{ + + vlib_buffer_t *new_next_b = vlib_get_buffer (vm, new_next_bi); + vnet_buffer_opaque_t *new_next_vnb = vnet_buffer (new_next_b); + if (~0 != prev_range_bi) + { + vlib_buffer_t *prev_b = vlib_get_buffer (vm, prev_range_bi); + vnet_buffer_opaque_t *prev_vnb = vnet_buffer (prev_b); + new_next_vnb->ip.reass.next_range_bi = prev_vnb->ip.reass.next_range_bi; + prev_vnb->ip.reass.next_range_bi = new_next_bi; + } + else + { + if (~0 != reass->first_bi) + { + new_next_vnb->ip.reass.next_range_bi = reass->first_bi; + } + reass->first_bi = new_next_bi; + } + reass->data_len += ip4_reass_buffer_get_data_len (new_next_b); + rm->buffers_n += ip4_reass_get_buffer_chain_length (vm, new_next_b); +} + +static void +ip4_reass_remove_range_from_chain (vlib_main_t * vm, + vlib_node_runtime_t * node, + ip4_reass_main_t * rm, + u32 ** vec_drop_overlap, + ip4_reass_t * reass, u32 prev_range_bi, + u32 discard_bi) +{ + vlib_buffer_t *discard_b = vlib_get_buffer (vm, discard_bi); + vnet_buffer_opaque_t *discard_vnb = vnet_buffer (discard_b); + if (~0 != prev_range_bi) + { + vlib_buffer_t *prev_b = vlib_get_buffer (vm, prev_range_bi); + vnet_buffer_opaque_t *prev_vnb = vnet_buffer (prev_b); + ASSERT (prev_vnb->ip.reass.next_range_bi == discard_bi); + prev_vnb->ip.reass.next_range_bi = discard_vnb->ip.reass.next_range_bi; + } + else + { + reass->first_bi = discard_vnb->ip.reass.next_range_bi; + } + reass->data_len -= ip4_reass_buffer_get_data_len (discard_b); + while (1) + { + vec_add1 (*vec_drop_overlap, discard_bi); + if (PREDICT_FALSE (discard_b->flags & VLIB_BUFFER_IS_TRACED)) + { + ip4_reass_add_trace (vm, node, rm, reass, discard_bi, RANGE_DISCARD, + 0); + } + if (discard_b->flags & VLIB_BUFFER_NEXT_PRESENT) + { + discard_b->flags &= ~VLIB_BUFFER_NEXT_PRESENT; + discard_bi = discard_b->next_buffer; + discard_b = vlib_get_buffer (vm, discard_bi); + } + else + { + break; + } + } +} + +void +ip4_reass_update (vlib_main_t * vm, vlib_node_runtime_t * node, + ip4_reass_main_t * rm, ip4_reass_t * reass, u32 * bi0, + u32 * next0, vlib_error_t * error0, + u32 ** vec_drop_overlap, u32 ** vec_drop_compress, + u32 next_input, u32 next_drop) +{ + int consumed = 0; + vlib_buffer_t *fb = vlib_get_buffer (vm, *bi0); + ip4_header_t *fip = vlib_buffer_get_current (fb); + ASSERT (fb->current_length >= sizeof (*fip)); + vnet_buffer_opaque_t *fvnb = vnet_buffer (fb); + u32 fragment_first = fvnb->ip.reass.fragment_first = + ip4_get_fragment_offset_bytes (fip); + u32 fragment_length = + clib_net_to_host_u16 (fip->length) - ip4_header_bytes (fip); + u32 fragment_last = fvnb->ip.reass.fragment_last = + fragment_first + fragment_length - 1; + int more_fragments = ip4_get_fragment_more (fip); + u32 candidate_range_bi = reass->first_bi; + u32 prev_range_bi = ~0; + fvnb->ip.reass.range_first = fragment_first; + fvnb->ip.reass.range_last = fragment_last; + fvnb->ip.reass.next_range_bi = ~0; + if (!more_fragments) + { + reass->last_packet_octet = fragment_last; + } + if (~0 == reass->first_bi) + { + // starting a new reassembly + ip4_reass_insert_range_in_chain (vm, rm, reass, prev_range_bi, *bi0); + if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED)) + { + ip4_reass_add_trace (vm, node, rm, reass, *bi0, RANGE_NEW, 0); + } + *bi0 = ~0; + fvnb->ip.reass.estimated_mtu = clib_net_to_host_u16 (fip->length); + return; + } + fvnb->ip.reass.estimated_mtu = clib_min (clib_net_to_host_u16 (fip->length), + fvnb->ip.reass.estimated_mtu); + while (~0 != candidate_range_bi) + { + vlib_buffer_t *candidate_b = vlib_get_buffer (vm, candidate_range_bi); + vnet_buffer_opaque_t *candidate_vnb = vnet_buffer (candidate_b); + if (fragment_first > candidate_vnb->ip.reass.range_last) + { + // this fragments starts after candidate range + prev_range_bi = candidate_range_bi; + candidate_range_bi = candidate_vnb->ip.reass.next_range_bi; + if (candidate_vnb->ip.reass.range_last < fragment_last && + ~0 == candidate_range_bi) + { + // special case - this fragment falls beyond all known ranges + ip4_reass_insert_range_in_chain (vm, rm, reass, prev_range_bi, + *bi0); + consumed = 1; + break; + } + continue; + } + if (fragment_last < candidate_vnb->ip.reass.range_first) + { + // this fragment ends before candidate range without any overlap + ip4_reass_insert_range_in_chain (vm, rm, reass, prev_range_bi, + *bi0); + consumed = 1; + } + else + { + if (fragment_first >= candidate_vnb->ip.reass.range_first && + fragment_last <= candidate_vnb->ip.reass.range_last) + { + // this fragment is a (sub)part of existing range, ignore it + if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED)) + { + ip4_reass_add_trace (vm, node, rm, reass, *bi0, + RANGE_OVERLAP, 0); + } + break; + } + int discard_candidate = 0; + if (fragment_first < candidate_vnb->ip.reass.range_first) + { + u32 overlap = + fragment_last - candidate_vnb->ip.reass.range_first + 1; + if (overlap < ip4_reass_buffer_get_data_len (candidate_b)) + { + candidate_vnb->ip.reass.range_first += overlap; + ASSERT (reass->data_len >= overlap); + reass->data_len -= overlap; + if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED)) + { + ip4_reass_add_trace (vm, node, rm, reass, + candidate_range_bi, RANGE_SHRINK, + overlap); + } + ip4_reass_insert_range_in_chain (vm, rm, reass, + prev_range_bi, *bi0); + consumed = 1; + } + else + { + discard_candidate = 1; + } + } + else if (fragment_last > candidate_vnb->ip.reass.range_last) + { + u32 overlap = + candidate_vnb->ip.reass.range_last - fragment_first + 1; + if (overlap < ip4_reass_buffer_get_data_len (candidate_b)) + { + fvnb->ip.reass.range_first += overlap; + if (~0 != candidate_vnb->ip.reass.next_range_bi) + { + prev_range_bi = candidate_range_bi; + candidate_range_bi = + candidate_vnb->ip.reass.next_range_bi; + continue; + } + else + { + // special case - last range discarded + ip4_reass_insert_range_in_chain (vm, rm, reass, + candidate_range_bi, + *bi0); + consumed = 1; + } + } + else + { + discard_candidate = 1; + } + } + else + { + discard_candidate = 1; + } + if (discard_candidate) + { + u32 next_range_bi = candidate_vnb->ip.reass.next_range_bi; + // discard candidate range, probe next range + ip4_reass_remove_range_from_chain (vm, node, rm, + vec_drop_overlap, reass, + prev_range_bi, + candidate_range_bi); + if (~0 != next_range_bi) + { + candidate_range_bi = next_range_bi; + continue; + } + else + { + // special case - last range discarded + ip4_reass_insert_range_in_chain (vm, rm, reass, + prev_range_bi, *bi0); + consumed = 1; + } + } + } + break; + } + if (consumed) + { + if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED)) + { + ip4_reass_add_trace (vm, node, rm, reass, *bi0, RANGE_NEW, 0); + } + } + if (~0 != reass->last_packet_octet && + reass->data_len == reass->last_packet_octet + 1) + { + ip4_reass_finalize (vm, node, rm, reass, bi0, next0, error0, next_input, + vec_drop_compress, vec_drop_overlap); + } + else + { + if (consumed) + { + *bi0 = ~0; + } + else + { + *next0 = next_drop; + *error0 = IP4_ERROR_REASS_DUPLICATE_FRAGMENT; + } + } +} + +always_inline uword +ip4_reassembly (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * frame) +{ + u32 *from = vlib_frame_vector_args (frame); + u32 n_left_from, n_left_to_next, *to_next, next_index; + ip4_reass_main_t *rm = &ip4_reass_main; + + n_left_from = frame->n_vectors; + next_index = node->cached_next_index; + static u32 *vec_drop_timeout = NULL; // indexes of buffers which timed out + static u32 *vec_drop_overlap = NULL; // indexes of buffers which were discarded due to overlap + static u32 *vec_drop_compress = NULL; // indexes of buffers dicarded due to buffer compression + while (n_left_from > 0 || vec_len (vec_drop_timeout) > 0 || + vec_len (vec_drop_overlap) > 0) + { + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (vec_len (vec_drop_timeout) > 0 && n_left_to_next > 0) + { + u32 bi = vec_pop (vec_drop_timeout); + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + b->error = node->errors[IP4_ERROR_REASS_TIMEOUT]; + to_next[0] = bi; + to_next += 1; + n_left_to_next -= 1; + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi, + IP4_REASSEMBLY_NEXT_DROP); + IP4_REASS_DEBUG_BUFFER (bi, enqueue_drop_timeout); + ASSERT (rm->buffers_n > 0); + --rm->buffers_n; + } + + while (vec_len (vec_drop_overlap) > 0 && n_left_to_next > 0) + { + u32 bi = vec_pop (vec_drop_overlap); + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + b->error = node->errors[IP4_ERROR_REASS_DUPLICATE_FRAGMENT]; + to_next[0] = bi; + to_next += 1; + n_left_to_next -= 1; + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi, + IP4_REASSEMBLY_NEXT_DROP); + IP4_REASS_DEBUG_BUFFER (bi, enqueue_drop_duplicate_fragment); + ASSERT (rm->buffers_n > 0); + --rm->buffers_n; + } + + while (vec_len (vec_drop_compress) > 0 && n_left_to_next > 0) + { + u32 bi = vec_pop (vec_drop_compress); + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + b->error = node->errors[IP4_ERROR_NONE]; + to_next[0] = bi; + to_next += 1; + n_left_to_next -= 1; + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi, + IP4_REASSEMBLY_NEXT_DROP); + IP4_REASS_DEBUG_BUFFER (bi, enqueue_drop_compress); + ASSERT (rm->buffers_n > 0); + --rm->buffers_n; + } + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + u32 next0; //, error0; + + bi0 = from[0]; + b0 = vlib_get_buffer (vm, bi0); + + ip4_header_t *ip0 = vlib_buffer_get_current (b0); + ip4_reass_key_t k; + k.src.as_u32 = ip0->src_address.as_u32; + k.dst.as_u32 = ip0->dst_address.as_u32; + k.xx_id = vnet_buffer (b0)->sw_if_index[VLIB_RX]; + k.frag_id = ip0->fragment_id; + k.proto = ip0->protocol; + k.unused = 0; + ip4_reass_t *reass = + ip4_reass_find_or_create (vm, rm, &k, &vec_drop_timeout); + + u32 error0 = IP4_ERROR_NONE; + if (reass) + { + ip4_reass_update (vm, node, rm, reass, &bi0, &next0, &error0, + &vec_drop_overlap, &vec_drop_compress, + IP4_REASSEMBLY_NEXT_INPUT, + IP4_REASSEMBLY_NEXT_DROP); + } + else + { + next0 = IP4_REASSEMBLY_NEXT_DROP; + error0 = IP4_ERROR_REASS_LIMIT_REACHED; + } + + b0->error = node->errors[error0]; + + if (bi0 != ~0) + { + to_next[0] = bi0; + to_next += 1; + n_left_to_next -= 1; + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + IP4_REASS_DEBUG_BUFFER (bi0, enqueue_next); + } + + from += 1; + n_left_from -= 1; + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + return frame->n_vectors; +} + +static char *ip4_reassembly_error_strings[] = { +#define _(sym, string) string, + foreach_ip4_error +#undef _ +}; + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (ip4_reass_node, static) = { + .function = ip4_reassembly, + .name = "ip4-reassembly", + .vector_size = sizeof (u32), + .format_trace = format_ip4_reass_trace, + .n_errors = ARRAY_LEN (ip4_reassembly_error_strings), + .error_strings = ip4_reassembly_error_strings, + .n_next_nodes = IP4_REASSEMBLY_N_NEXT, + .next_nodes = + { + [IP4_REASSEMBLY_NEXT_INPUT] = "ip4-input", + [IP4_REASSEMBLY_NEXT_DROP] = "ip4-drop", + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (ip4_reass_node, ip4_reassembly) + static u32 ip4_reass_get_nbuckets () +{ + ip4_reass_main_t *rm = &ip4_reass_main; + u32 nbuckets; + u8 i; + + nbuckets = (u32) (rm->max_reass_n / IP4_REASS_HT_LOAD_FACTOR); + + for (i = 0; i < 31; i++) + if ((1 << i) >= nbuckets) + break; + nbuckets = 1 << i; + + return nbuckets; +} + +typedef enum +{ + IP4_EVENT_CONFIG_CHANGED = 1, +} ip4_reass_event_t; + +typedef struct +{ + int failure; + clib_bihash_24_8_t *new_hash; +} ip4_rehash_cb_ctx; + +void +ip4_rehash_cb (clib_bihash_kv_24_8_t * kv, void *_ctx) +{ + ip4_rehash_cb_ctx *ctx = _ctx; + if (clib_bihash_add_del_24_8 (ctx->new_hash, kv, 1)) + { + ctx->failure = 1; + } +} + +vnet_api_error_t +ip4_reass_set (u32 timeout_ms, u32 max_reassemblies, + u32 expire_walk_interval_ms) +{ + u32 old_nbuckets = ip4_reass_get_nbuckets (); + ip4_reass_main.timeout_ms = timeout_ms; + ip4_reass_main.timeout = (f64) timeout_ms / (f64) MSEC_PER_SEC; + ip4_reass_main.max_reass_n = max_reassemblies; + ip4_reass_main.expire_walk_interval_ms = expire_walk_interval_ms; + vlib_process_signal_event (ip4_reass_main.vlib_main, + ip4_reass_main.ip4_reass_expire_node_idx, + IP4_EVENT_CONFIG_CHANGED, 0); + u32 new_nbuckets = ip4_reass_get_nbuckets (); + if (ip4_reass_main.max_reass_n > 0 && new_nbuckets > 1 && + new_nbuckets != old_nbuckets) + { + clib_bihash_24_8_t new_hash; + memset (&new_hash, 0, sizeof (new_hash)); + ip4_rehash_cb_ctx ctx; + ctx.failure = 0; + ctx.new_hash = &new_hash; + clib_bihash_init_24_8 (&new_hash, "ip4-reass", new_nbuckets, + new_nbuckets * 1024); + clib_bihash_foreach_key_value_pair_24_8 (&ip4_reass_main.hash, + ip4_rehash_cb, &ctx); + if (ctx.failure) + { + clib_bihash_free_24_8 (&new_hash); + return -1; + } + else + { + clib_bihash_free_24_8 (&ip4_reass_main.hash); + clib_memcpy (&ip4_reass_main.hash, &new_hash, + sizeof (ip4_reass_main.hash)); + } + } + return 0; +} + +vnet_api_error_t +ip4_reass_get (u32 * timeout_ms, u32 * max_reassemblies, + u32 * expire_walk_interval_ms) +{ + *timeout_ms = ip4_reass_main.timeout_ms; + *max_reassemblies = ip4_reass_main.max_reass_n; + *expire_walk_interval_ms = ip4_reass_main.expire_walk_interval_ms; + return 0; +} + +clib_error_t * +ip4_reass_init_function (vlib_main_t * vm) +{ + ip4_reass_main_t *rm = &ip4_reass_main; + clib_error_t *error = 0; + u32 nbuckets; + + rm->vlib_main = vm; + rm->vnet_main = vnet_get_main (); + + rm->reass_n = 0; + pool_alloc (rm->pool, rm->max_reass_n); + ip4_reass_set (IP4_REASS_TIMEOUT_DEFAULT_MS, + IP4_REASS_MAX_REASSEMBLIES_DEAFULT, + IP4_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS); + + nbuckets = ip4_reass_get_nbuckets (); + clib_bihash_init_24_8 (&rm->hash, "ip4-reass", nbuckets, nbuckets * 1024); + + vlib_node_t *node = vlib_get_node_by_name (vm, (u8 *) "ip4-drop"); + ASSERT (node); + rm->ip4_drop_idx = node->index; + node = vlib_get_node_by_name (vm, (u8 *) "ip4-reassembly-expire-walk"); + ASSERT (node); + rm->ip4_reass_expire_node_idx = node->index; + return error; +} + +VLIB_INIT_FUNCTION (ip4_reass_init_function); + +static uword +ip4_reass_walk_expired (vlib_main_t * vm, + vlib_node_runtime_t * node, vlib_frame_t * f) +{ + ip4_reass_main_t *rm = &ip4_reass_main; + uword event_type, *event_data = 0; + + while (true) + { + vlib_process_wait_for_event_or_clock (vm, + (f64) rm->expire_walk_interval_ms + / (f64) MSEC_PER_SEC); + event_type = vlib_process_get_events (vm, &event_data); + + switch (event_type) + { + case ~0: /* no events => timeout */ + /* nothing to do here */ + break; + case IP4_EVENT_CONFIG_CHANGED: + break; + default: + clib_warning ("BUG: event type 0x%wx", event_type); + break; + } + f64 now = vlib_time_now (vm); + + ip4_reass_t *reass; + u32 *vec_drop_timeout = NULL; + int *pool_indexes_to_free = NULL; + + int index; + /* *INDENT-OFF* */ + pool_foreach_index (index, rm->pool, ({ + reass = pool_elt_at_index (rm->pool, index); + if (now > reass->last_heard + rm->timeout) + { + vec_add1 (pool_indexes_to_free, index); + } + })); + /* *INDENT-ON* */ + int *i; + /* *INDENT-OFF* */ + vec_foreach (i, pool_indexes_to_free) + { + ip4_reass_t *reass = pool_elt_at_index (rm->pool, i[0]); + ip4_reass_on_timeout (vm, rm, reass, &vec_drop_timeout); + ip4_reass_free (rm, reass); + } + /* *INDENT-ON* */ + + while (vec_len (vec_drop_timeout) > 0) + { + vlib_frame_t *f = vlib_get_frame_to_node (vm, rm->ip4_drop_idx); + u32 *to_next = vlib_frame_vector_args (f); + u32 n_left_to_next = VLIB_FRAME_SIZE - f->n_vectors; + u32 n_trace = 0; + while (vec_len (vec_drop_timeout) > 0 && n_left_to_next > 0) + { + u32 bi = vec_pop (vec_drop_timeout); + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + if (PREDICT_FALSE (b->flags & VLIB_BUFFER_IS_TRACED)) + { + if (pool_is_free_index (vm->trace_main.trace_buffer_pool, + b->trace_index)) + { + /* the trace is gone, don't trace this buffer anymore */ + b->flags &= ~VLIB_BUFFER_IS_TRACED; + } + else + { + ++n_trace; + } + } + b->error = node->errors[IP4_ERROR_REASS_TIMEOUT]; + to_next[0] = bi; + ++f->n_vectors; + to_next += 1; + n_left_to_next -= 1; + IP4_REASS_DEBUG_BUFFER (bi, enqueue_drop_timeout_walk); + ASSERT (rm->buffers_n > 0); + --rm->buffers_n; + } + if (PREDICT_FALSE (n_trace > 0)) + { + f->flags |= VLIB_FRAME_TRACE; + } + vlib_put_frame_to_node (vm, rm->ip4_drop_idx, f); + } + + vec_free (pool_indexes_to_free); + vec_free (vec_drop_timeout); + if (event_data) + { + _vec_len (event_data) = 0; + } + } + + return 0; +} + +static vlib_node_registration_t ip4_reass_expire_node; + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (ip4_reass_expire_node, static) = { + .function = ip4_reass_walk_expired, + .type = VLIB_NODE_TYPE_PROCESS, + .name = "ip4-reassembly-expire-walk", + .format_trace = format_ip4_reass_trace, + .n_errors = ARRAY_LEN (ip4_reassembly_error_strings), + .error_strings = ip4_reassembly_error_strings, + +}; +/* *INDENT-ON* */ + +static u8 * +format_ip4_reass_key (u8 * s, va_list * args) +{ + ip4_reass_key_t *key = va_arg (*args, ip4_reass_key_t *); + s = format (s, "xx_id: %u, src: %U, dst: %U, frag_id: %u, proto: %u", + key->xx_id, format_ip4_address, &key->src, format_ip4_address, + &key->dst, clib_net_to_host_u16 (key->frag_id), key->proto); + return s; +} + +static u8 * +format_ip4_reass (u8 * s, va_list * args) +{ + vlib_main_t *vm = va_arg (*args, vlib_main_t *); + ip4_reass_t *reass = va_arg (*args, ip4_reass_t *); + + s = format (s, "ID: %u, key: %U\n first_bi: %u, data_len: %u, " + "last_packet_octet: %u, trace_op_counter: %u\n", + reass->id, format_ip4_reass_key, &reass->key, reass->first_bi, + reass->data_len, reass->last_packet_octet, + reass->trace_op_counter); + u32 bi = reass->first_bi; + u32 counter = 0; + while (~0 != bi) + { + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + vnet_buffer_opaque_t *vnb = vnet_buffer (b); + s = format (s, " #%03u: range: [%u, %u], bi: %u, off: %d, len: %u, " + "fragment[%u, %u]\n", + counter, vnb->ip.reass.range_first, + vnb->ip.reass.range_last, bi, + ip4_reass_buffer_get_data_offset_no_check (b), + ip4_reass_buffer_get_data_len_no_check (b), + vnb->ip.reass.fragment_first, vnb->ip.reass.fragment_last); + if (b->flags & VLIB_BUFFER_NEXT_PRESENT) + { + bi = b->next_buffer; + } + else + { + bi = ~0; + } + } + return s; +} + +static clib_error_t * +show_ip4_reass (vlib_main_t * vm, unformat_input_t * input, + CLIB_UNUSED (vlib_cli_command_t * lmd)) +{ + ip4_reass_main_t *rm = &ip4_reass_main; + + vlib_cli_output (vm, "---------------------"); + vlib_cli_output (vm, "IP4 reassembly status"); + vlib_cli_output (vm, "---------------------"); + if (unformat (input, "details")) + { + ip4_reass_t *reass; + /* *INDENT-OFF* */ + pool_foreach (reass, rm->pool, { + vlib_cli_output (vm, "%U", format_ip4_reass, vm, reass); + }); + /* *INDENT-ON* */ + } + vlib_cli_output (vm, "---------------------"); + vlib_cli_output (vm, "Current IP4 reassemblies count: %lu\n", rm->reass_n); + vlib_cli_output (vm, + "Maximum configured concurrent IP4 reassemblies: %lu\n", + (long unsigned) rm->max_reass_n); + vlib_cli_output (vm, "Buffers in use: %lu\n", + (long unsigned) rm->buffers_n); + return 0; +} + +/* *INDENT-OFF* */ +VLIB_CLI_COMMAND (show_ip4_reassembly_cmd, static) = { + .path = "show ip4-reassembly", + .short_help = "show ip4-reassembly [details]", + .function = show_ip4_reass, +}; +/* *INDENT-ON* */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/ip/ip4_reassembly.h b/src/vnet/ip/ip4_reassembly.h new file mode 100644 index 00000000000..a48c07ef061 --- /dev/null +++ b/src/vnet/ip/ip4_reassembly.h @@ -0,0 +1,49 @@ +/* + * 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 IPv4 Reassembly. + * + * This file contains the source code for IPv4 reassembly. + */ + +#ifndef __included_ip4_reassembly_h__ +#define __included_ip4_reassembly_h__ + +#include <vnet/api_errno.h> +#include <vnet/vnet.h> + +/** + * @brief set ip4 reassembly configuration + */ +vnet_api_error_t ip4_reass_set (u32 timeout_ms, u32 max_reassemblies, + u32 expire_walk_interval_ms); + +/** + * @brief get ip4 reassembly configuration + */ +vnet_api_error_t ip4_reass_get (u32 * timeout_ms, u32 * max_reassemblies, + u32 * expire_walk_interval_ms); + +#endif /* __included_ip4_reassembly_h */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/ip/ip6_error.h b/src/vnet/ip/ip6_error.h index 7bb4b77b5a3..19f20625a10 100644 --- a/src/vnet/ip/ip6_error.h +++ b/src/vnet/ip/ip6_error.h @@ -71,7 +71,14 @@ \ /* Erros singalled by ip6-inacl */ \ _ (INACL_TABLE_MISS, "input ACL table-miss drops") \ - _ (INACL_SESSION_DENY, "input ACL session deny drops") + _ (INACL_SESSION_DENY, "input ACL session deny drops") \ + \ + /* Errors signalled by ip6-reassembly */ \ + _ (REASS_MISSING_UPPER, "missing-upper layer drops") \ + _ (REASS_DUPLICATE_FRAGMENT, "duplicate fragments") \ + _ (REASS_OVERLAPPING_FRAGMENT, "overlapping fragments") \ + _ (REASS_LIMIT_REACHED, "drops due to concurrent reassemblies limit") \ + _ (REASS_TIMEOUT, "fragments dropped due to reassembly timeout") typedef enum { diff --git a/src/vnet/ip/ip6_forward.c b/src/vnet/ip/ip6_forward.c index c1c9ec0b586..5b06ca7609d 100644 --- a/src/vnet/ip/ip6_forward.c +++ b/src/vnet/ip/ip6_forward.c @@ -1578,6 +1578,7 @@ ip6_local_inline (vlib_main_t * vm, vlib_node_runtime_t * node, next0 = lm->local_next_by_ip_protocol[ip0->protocol]; next0 = error0 != IP6_ERROR_UNKNOWN_PROTOCOL ? IP_LOCAL_NEXT_DROP : next0; + p0->error = error_node->errors[error0]; if (head_of_feature_arc) @@ -1619,6 +1620,7 @@ VLIB_REGISTER_NODE (ip6_local_node, static) = [IP_LOCAL_NEXT_PUNT] = "ip6-punt", [IP_LOCAL_NEXT_UDP_LOOKUP] = "ip6-udp-lookup", [IP_LOCAL_NEXT_ICMP] = "ip6-icmp-input", + [IP_LOCAL_NEXT_REASSEMBLY] = "ip6-reassembly", }, }; /* *INDENT-ON* */ diff --git a/src/vnet/ip/ip6_input.h b/src/vnet/ip/ip6_input.h index 4c0d78459b1..fe993caa889 100644 --- a/src/vnet/ip/ip6_input.h +++ b/src/vnet/ip/ip6_input.h @@ -41,6 +41,7 @@ #define included_ip6_input_h #include <vnet/ip/ip.h> +#include <vnet/ip/icmp6.h> extern char *ip6_error_strings[]; diff --git a/src/vnet/ip/ip6_packet.h b/src/vnet/ip/ip6_packet.h index c0c745e27fe..6b724370e76 100644 --- a/src/vnet/ip/ip6_packet.h +++ b/src/vnet/ip/ip6_packet.h @@ -519,6 +519,9 @@ typedef CLIB_PACKED (struct { #define ip6_frag_hdr_offset(hdr) \ (clib_net_to_host_u16((hdr)->fragment_offset_and_more) >> 3) +#define ip6_frag_hdr_offset_bytes(hdr) \ + (8 * ip6_frag_hdr_offset(hdr)) + #define ip6_frag_hdr_more(hdr) \ (clib_net_to_host_u16((hdr)->fragment_offset_and_more) & 0x1) diff --git a/src/vnet/ip/ip6_reassembly.c b/src/vnet/ip/ip6_reassembly.c new file mode 100644 index 00000000000..d448032b913 --- /dev/null +++ b/src/vnet/ip/ip6_reassembly.c @@ -0,0 +1,1347 @@ +/* + * 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 IPv6 Reassembly. + * + * This file contains the source code for IPv6 reassembly. + */ + +#include <vppinfra/vec.h> +#include <vnet/vnet.h> +#include <vnet/ip/ip.h> +#include <vppinfra/bihash_48_8.h> +#include <vnet/ip/ip6_reassembly.h> + +#define MSEC_PER_SEC 1000 +#define IP6_REASS_TIMEOUT_DEFAULT_MS 100 +#define IP6_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS 10000 // 10 seconds default +#define IP6_REASS_MAX_REASSEMBLIES_DEAFULT 1024 +#define IP6_REASS_HT_LOAD_FACTOR (0.75) + +static vlib_node_registration_t ip6_reass_node; + +typedef struct +{ + union + { + struct + { + ip6_address_t src; + ip6_address_t dst; + // align by making this 4 octets even though its a 2 octets field + u32 xx_id; + // align by making this 4 octets even though its a 2 octets field + u32 frag_id; + // align by making this 4 octets even though its a 1 octet field + u32 proto; + u32 unused; + }; + u64 as_u64[6]; + }; +} ip6_reass_key_t; + +always_inline u32 +ip6_reass_buffer_get_data_offset_no_check (vlib_buffer_t * b) +{ + vnet_buffer_opaque_t *vnb = vnet_buffer (b); + return vnb->ip.reass.range_first - vnb->ip.reass.fragment_first; +} + +always_inline u32 +ip6_reass_buffer_get_data_offset (vlib_buffer_t * b) +{ + vnet_buffer_opaque_t *vnb = vnet_buffer (b); + ASSERT (vnb->ip.reass.range_first >= vnb->ip.reass.fragment_first); + return ip6_reass_buffer_get_data_offset_no_check (b); +} + +always_inline u16 +ip6_reass_buffer_get_data_len_no_check (vlib_buffer_t * b) +{ + vnet_buffer_opaque_t *vnb = vnet_buffer (b); + return clib_min (vnb->ip.reass.range_last, vnb->ip.reass.fragment_last) - + (vnb->ip.reass.fragment_first + ip6_reass_buffer_get_data_offset (b)) + 1; +} + +always_inline u16 +ip6_reass_buffer_get_data_len (vlib_buffer_t * b) +{ + vnet_buffer_opaque_t *vnb = vnet_buffer (b); + ASSERT (vnb->ip.reass.range_last > vnb->ip.reass.fragment_first); + return ip6_reass_buffer_get_data_len_no_check (b); +} + +typedef struct +{ + // hash table key + ip6_reass_key_t key; + // time when first packet was received + f64 first_heard; + // time when last packet was received + f64 last_heard; + // internal id of this reassembly + u32 id; + // buffer index of first buffer in this reassembly context + u32 first_bi; + // last octet of packet, ~0 until fragment without more_fragments arrives + u32 last_packet_octet; + // length of data collected so far + u32 data_len; + // trace operation counter + u32 trace_op_counter; +} ip6_reass_t; + +typedef struct +{ + // IPv6 config + u32 timeout_ms; + f64 timeout; + u32 expire_walk_interval_ms; + u32 max_reass_n; + + // IPv6 runtime + ip6_reass_t *pool; + clib_bihash_48_8_t hash; + u32 reass_n; + u32 id_counter; + u32 buffers_n; + + // convenience + vlib_main_t *vlib_main; + vnet_main_t *vnet_main; + + // node index of ip6-drop node + u32 ip6_drop_idx; + u32 ip6_icmp_error_idx; + u32 ip6_reass_expire_node_idx; + +} ip6_reass_main_t; + +ip6_reass_main_t ip6_reass_main; + +typedef enum +{ + IP6_REASSEMBLY_NEXT_INPUT, + IP6_REASSEMBLY_NEXT_DROP, + IP6_REASSEMBLY_NEXT_ICMP_ERROR, + IP6_REASSEMBLY_N_NEXT, +} ip6_reass_next_t; + +typedef enum +{ + RANGE_NEW, + RANGE_OVERLAP, + ICMP_ERROR_RT_EXCEEDED, + ICMP_ERROR_FL_TOO_BIG, + ICMP_ERROR_FL_NOT_MULT_8, + FINALIZE, +} ip6_reass_trace_operation_e; + +typedef struct +{ + u16 range_first; + u16 range_last; + u32 range_bi; + i32 data_offset; + u32 data_len; + u32 first_bi; +} ip6_reass_range_trace_t; + +typedef struct +{ + ip6_reass_trace_operation_e action; + u32 pool_index; + u32 reass_id; + ip6_reass_range_trace_t trace_range; + u32 size_diff; + u32 op_id; + u32 fragment_first; + u32 fragment_last; + u32 total_data_len; +} ip6_reass_trace_t; + +static void +ip6_reass_trace_details (vlib_main_t * vm, u32 bi, + ip6_reass_range_trace_t * trace) +{ + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + vnet_buffer_opaque_t *vnb = vnet_buffer (b); + trace->range_first = vnb->ip.reass.range_first; + trace->range_last = vnb->ip.reass.range_last; + trace->data_offset = ip6_reass_buffer_get_data_offset_no_check (b); + trace->data_len = ip6_reass_buffer_get_data_len_no_check (b); + trace->range_bi = bi; +} + +u8 * +format_ip6_reass_range_trace (u8 * s, va_list * args) +{ + ip6_reass_range_trace_t *trace = va_arg (*args, ip6_reass_range_trace_t *); + s = format (s, "range: [%u, %u], off %d, len %u, bi %u", trace->range_first, + trace->range_last, trace->data_offset, trace->data_len, + trace->range_bi); + return s; +} + +u8 * +format_ip6_reass_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 *); + ip6_reass_trace_t *t = va_arg (*args, ip6_reass_trace_t *); + s = format (s, "reass id: %u, op id: %u ", t->reass_id, t->op_id); + u32 indent = format_get_indent (s); + s = format (s, "first bi: %u, data len: %u, ip/fragment[%u, %u]", + t->trace_range.first_bi, t->total_data_len, t->fragment_first, + t->fragment_last); + switch (t->action) + { + case RANGE_NEW: + s = format (s, "\n%Unew %U", format_white_space, indent, + format_ip6_reass_range_trace, &t->trace_range); + break; + case RANGE_OVERLAP: + s = format (s, "\n%Uoverlap %U", format_white_space, indent, + format_ip6_reass_range_trace, &t->trace_range); + break; + case ICMP_ERROR_FL_TOO_BIG: + s = format (s, "\n%Uicmp-error - frag_len > 65535 %U", + format_white_space, indent, format_ip6_reass_range_trace, + &t->trace_range); + break; + case ICMP_ERROR_FL_NOT_MULT_8: + s = format (s, "\n%Uicmp-error - frag_len mod 8 != 0 %U", + format_white_space, indent, format_ip6_reass_range_trace, + &t->trace_range); + break; + case ICMP_ERROR_RT_EXCEEDED: + s = format (s, "\n%Uicmp-error - reassembly time exceeded", + format_white_space, indent); + break; + case FINALIZE: + s = format (s, "\n%Ufinalize reassembly", format_white_space, indent); + break; + } + return s; +} + +static void +ip6_reass_add_trace (vlib_main_t * vm, vlib_node_runtime_t * node, + ip6_reass_main_t * rm, ip6_reass_t * reass, + u32 bi, ip6_reass_trace_operation_e action, + u32 size_diff) +{ + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + vnet_buffer_opaque_t *vnb = vnet_buffer (b); + ip6_reass_trace_t *t = vlib_add_trace (vm, node, b, sizeof (t[0])); + t->pool_index = reass - rm->pool; + t->reass_id = reass->id; + t->action = action; + ip6_reass_trace_details (vm, bi, &t->trace_range); + t->size_diff = size_diff; + t->op_id = reass->trace_op_counter; + ++reass->trace_op_counter; + t->fragment_first = vnb->ip.reass.fragment_first; + t->fragment_last = vnb->ip.reass.fragment_last; + t->trace_range.first_bi = reass->first_bi; + t->total_data_len = reass->data_len; +#if 0 + static u8 *s = NULL; + s = format (s, "%U", format_ip6_reass_trace, NULL, NULL, t); + printf ("%.*s\n", vec_len (s), s); + fflush (stdout); + vec_reset_length (s); +#endif +} + +static void +ip6_reass_free (ip6_reass_main_t * rm, ip6_reass_t * reass) +{ + clib_bihash_kv_48_8_t kv; + kv.key[0] = reass->key.as_u64[0]; + kv.key[1] = reass->key.as_u64[1]; + kv.key[2] = reass->key.as_u64[2]; + kv.key[3] = reass->key.as_u64[3]; + kv.key[4] = reass->key.as_u64[4]; + kv.key[5] = reass->key.as_u64[5]; + clib_bihash_add_del_48_8 (&rm->hash, &kv, 0); + pool_put (rm->pool, reass); + --rm->reass_n; +} + +static void +ip6_reass_drop_all (vlib_main_t * vm, ip6_reass_main_t * rm, + ip6_reass_t * reass, u32 ** vec_drop_bi) +{ + u32 range_bi = reass->first_bi; + vlib_buffer_t *range_b; + vnet_buffer_opaque_t *range_vnb; + while (~0 != range_bi) + { + range_b = vlib_get_buffer (vm, range_bi); + range_vnb = vnet_buffer (range_b); + u32 bi = range_bi; + while (~0 != bi) + { + vec_add1 (*vec_drop_bi, bi); + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + if (b->flags & VLIB_BUFFER_NEXT_PRESENT) + { + bi = b->next_buffer; + b->flags &= ~VLIB_BUFFER_NEXT_PRESENT; + } + else + { + bi = ~0; + } + } + range_bi = range_vnb->ip.reass.next_range_bi; + } +} + +static void +ip6_reass_on_timeout (vlib_main_t * vm, vlib_node_runtime_t * node, + ip6_reass_main_t * rm, ip6_reass_t * reass, + u32 * icmp_bi, u32 ** vec_timeout) +{ + if (~0 == reass->first_bi) + { + return; + } + vlib_buffer_t *b = vlib_get_buffer (vm, reass->first_bi); + if (0 == vnet_buffer (b)->ip.reass.fragment_first) + { + *icmp_bi = reass->first_bi; + if (PREDICT_FALSE (b->flags & VLIB_BUFFER_IS_TRACED)) + { + ip6_reass_add_trace (vm, node, rm, reass, reass->first_bi, + ICMP_ERROR_RT_EXCEEDED, 0); + } + // fragment with offset zero received - send icmp message back + if (b->flags & VLIB_BUFFER_NEXT_PRESENT) + { + // separate first buffer from chain and steer it towards icmp node + b->flags &= ~VLIB_BUFFER_NEXT_PRESENT; + reass->first_bi = b->next_buffer; + } + else + { + reass->first_bi = vnet_buffer (b)->ip.reass.next_range_bi; + } + icmp6_error_set_vnet_buffer (b, ICMP6_time_exceeded, + ICMP6_time_exceeded_fragment_reassembly_time_exceeded, + 0); + } + ip6_reass_drop_all (vm, rm, reass, vec_timeout); +} + +static ip6_reass_t * +ip6_reass_find_or_create (vlib_main_t * vm, + vlib_node_runtime_t * node, + ip6_reass_main_t * rm, + ip6_reass_key_t * k, u32 * icmp_bi, + u32 ** vec_timeout) +{ + ip6_reass_t *reass = NULL; + f64 now = vlib_time_now (rm->vlib_main); + clib_bihash_kv_48_8_t kv, value; + kv.key[0] = k->as_u64[0]; + kv.key[1] = k->as_u64[1]; + kv.key[2] = k->as_u64[2]; + kv.key[3] = k->as_u64[3]; + kv.key[4] = k->as_u64[4]; + kv.key[5] = k->as_u64[5]; + + if (!clib_bihash_search_48_8 (&rm->hash, &kv, &value)) + { + reass = pool_elt_at_index (rm->pool, value.value); + if (now > reass->last_heard + rm->timeout) + { + ip6_reass_on_timeout (vm, node, rm, reass, icmp_bi, vec_timeout); + ip6_reass_free (rm, reass); + reass = NULL; + } + } + + if (reass) + { + reass->last_heard = now; + return reass; + } + + if (rm->reass_n >= rm->max_reass_n) + { + reass = NULL; + return reass; + } + else + { + pool_get (rm->pool, reass); + memset (reass, 0, sizeof (*reass)); + reass->id = rm->id_counter; + ++rm->id_counter; + reass->first_bi = ~0; + reass->last_packet_octet = ~0; + reass->data_len = 0; + ++rm->reass_n; + } + + reass->key.as_u64[0] = kv.key[0] = k->as_u64[0]; + reass->key.as_u64[1] = kv.key[1] = k->as_u64[1]; + reass->key.as_u64[2] = kv.key[2] = k->as_u64[2]; + reass->key.as_u64[3] = kv.key[3] = k->as_u64[3]; + reass->key.as_u64[4] = kv.key[4] = k->as_u64[4]; + reass->key.as_u64[5] = kv.key[5] = k->as_u64[5]; + kv.value = reass - rm->pool; + reass->last_heard = now; + + if (clib_bihash_add_del_48_8 (&rm->hash, &kv, 1)) + { + ip6_reass_free (rm, reass); + reass = NULL; + } + + return reass; +} + +void +ip6_reass_finalize (vlib_main_t * vm, vlib_node_runtime_t * node, + ip6_reass_main_t * rm, ip6_reass_t * reass, u32 * bi0, + u32 * next0, vlib_error_t * error0, u32 next_input, + u32 ** vec_drop_compress) +{ + ASSERT (~0 != reass->first_bi); + *bi0 = reass->first_bi; + *next0 = next_input; + *error0 = IP6_ERROR_NONE; + ip6_frag_hdr_t *frag_hdr; + vlib_buffer_t *last_b = NULL; + u32 sub_chain_bi = reass->first_bi; + u32 total_length = 0; + u32 buf_cnt = 0; + u32 dropped_cnt = 0; + do + { + u32 tmp_bi = sub_chain_bi; + vlib_buffer_t *tmp = vlib_get_buffer (vm, tmp_bi); + u32 data_len = ip6_reass_buffer_get_data_len (tmp); + u32 trim_front = vnet_buffer (tmp)->ip.reass.ip6_frag_hdr_offset + + sizeof (*frag_hdr) + ip6_reass_buffer_get_data_offset (tmp); + u32 trim_end = + vlib_buffer_length_in_chain (vm, tmp) - trim_front - data_len; + if (tmp_bi == reass->first_bi) + { + /* first buffer - keep ip6 header */ + ASSERT (0 == ip6_reass_buffer_get_data_offset (tmp)); + trim_front = 0; + trim_end = vlib_buffer_length_in_chain (vm, tmp) - data_len - + (vnet_buffer (tmp)->ip.reass.ip6_frag_hdr_offset + + sizeof (*frag_hdr)); + } + u32 keep_data = + vlib_buffer_length_in_chain (vm, tmp) - trim_front - trim_end; + while (1) + { + ++buf_cnt; + if (trim_front) + { + if (trim_front > tmp->current_length) + { + /* drop whole buffer */ + vec_add1 (*vec_drop_compress, tmp_bi); + ++dropped_cnt; + trim_front -= tmp->current_length; + ASSERT (tmp->flags & VLIB_BUFFER_NEXT_PRESENT); + tmp->flags &= ~VLIB_BUFFER_NEXT_PRESENT; + tmp_bi = tmp->next_buffer; + tmp = vlib_get_buffer (vm, tmp_bi); + continue; + } + else + { + vlib_buffer_advance (tmp, trim_front); + trim_front = 0; + } + } + if (keep_data) + { + if (last_b) + { + last_b->flags |= VLIB_BUFFER_NEXT_PRESENT; + last_b->next_buffer = tmp_bi; + } + last_b = tmp; + if (keep_data <= tmp->current_length) + { + tmp->current_length = keep_data; + keep_data = 0; + } + else + { + keep_data -= tmp->current_length; + ASSERT (tmp->flags & VLIB_BUFFER_NEXT_PRESENT); + } + total_length += tmp->current_length; + } + else + { + vec_add1 (*vec_drop_compress, tmp_bi); + ASSERT (reass->first_bi != tmp_bi); + ++dropped_cnt; + } + if (tmp->flags & VLIB_BUFFER_NEXT_PRESENT) + { + tmp_bi = tmp->next_buffer; + tmp = vlib_get_buffer (vm, tmp->next_buffer); + } + else + { + break; + } + } + sub_chain_bi = + vnet_buffer (vlib_get_buffer (vm, sub_chain_bi))->ip. + reass.next_range_bi; + } + while (~0 != sub_chain_bi); + last_b->flags &= ~VLIB_BUFFER_NEXT_PRESENT; + vlib_buffer_t *first_b = vlib_get_buffer (vm, reass->first_bi); + ASSERT (total_length >= first_b->current_length); + total_length -= first_b->current_length; + first_b->flags |= VLIB_BUFFER_TOTAL_LENGTH_VALID; + first_b->total_length_not_including_first_buffer = total_length; + // drop fragment header + vnet_buffer_opaque_t *first_b_vnb = vnet_buffer (first_b); + ip6_header_t *ip = vlib_buffer_get_current (first_b); + u16 ip6_frag_hdr_offset = first_b_vnb->ip.reass.ip6_frag_hdr_offset; + ip6_ext_header_t *prev_hdr; + ip6_ext_header_find_t (ip, prev_hdr, frag_hdr, + IP_PROTOCOL_IPV6_FRAGMENTATION); + if (prev_hdr) + { + prev_hdr->next_hdr = frag_hdr->next_hdr; + } + else + { + ip->protocol = frag_hdr->next_hdr; + } + ASSERT ((u8 *) frag_hdr - (u8 *) ip == ip6_frag_hdr_offset); + memmove (frag_hdr, (u8 *) frag_hdr + sizeof (*frag_hdr), + first_b->current_length - ip6_frag_hdr_offset - + sizeof (ip6_frag_hdr_t)); + first_b->current_length -= sizeof (*frag_hdr); + ip->payload_length = + clib_host_to_net_u16 (total_length + first_b->current_length - + sizeof (*ip)); + ip6_reass_free (rm, reass); + vlib_buffer_chain_compress (vm, first_b, vec_drop_compress); + if (PREDICT_FALSE (first_b->flags & VLIB_BUFFER_IS_TRACED)) + { + ip6_reass_add_trace (vm, node, rm, reass, reass->first_bi, FINALIZE, 0); +#if 0 + // following code does a hexdump of packet fragments to stdout ... + do + { + u32 bi = reass->first_bi; + u8 *s = NULL; + while (~0 != bi) + { + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + s = format (s, "%u: %U\n", bi, format_hexdump, + vlib_buffer_get_current (b), b->current_length); + if (b->flags & VLIB_BUFFER_NEXT_PRESENT) + { + bi = b->next_buffer; + } + else + { + break; + } + } + printf ("%.*s\n", vec_len (s), s); + fflush (stdout); + vec_free (s); + } + while (0); +#endif + } + reass = NULL; +} + +static u32 +ip6_reass_get_buffer_chain_length (vlib_main_t * vm, vlib_buffer_t * b) +{ + u32 len = 0; + while (b) + { + ++len; + if (PREDICT_FALSE (b->flags & VLIB_BUFFER_NEXT_PRESENT)) + { + b = vlib_get_buffer (vm, b->next_buffer); + } + else + { + break; + } + } + return len; +} + +static void +ip6_reass_insert_range_in_chain (vlib_main_t * vm, + ip6_reass_main_t * rm, + ip6_reass_t * reass, + u32 prev_range_bi, u32 new_next_bi) +{ + + vlib_buffer_t *new_next_b = vlib_get_buffer (vm, new_next_bi); + vnet_buffer_opaque_t *new_next_vnb = vnet_buffer (new_next_b); + if (~0 != prev_range_bi) + { + vlib_buffer_t *prev_b = vlib_get_buffer (vm, prev_range_bi); + vnet_buffer_opaque_t *prev_vnb = vnet_buffer (prev_b); + new_next_vnb->ip.reass.next_range_bi = prev_vnb->ip.reass.next_range_bi; + prev_vnb->ip.reass.next_range_bi = new_next_bi; + } + else + { + if (~0 != reass->first_bi) + { + new_next_vnb->ip.reass.next_range_bi = reass->first_bi; + } + reass->first_bi = new_next_bi; + } + reass->data_len += ip6_reass_buffer_get_data_len (new_next_b); + rm->buffers_n += ip6_reass_get_buffer_chain_length (vm, new_next_b); +} + +void +ip6_reass_update (vlib_main_t * vm, vlib_node_runtime_t * node, + ip6_reass_main_t * rm, ip6_reass_t * reass, u32 * bi0, + u32 * next0, vlib_error_t * error0, + ip6_frag_hdr_t * frag_hdr, u32 ** vec_drop_overlap, + u32 ** vec_drop_compress, u32 next_input, u32 next_drop, + u32 next_icmp_error) +{ + int consumed = 0; + vlib_buffer_t *fb = vlib_get_buffer (vm, *bi0); + vnet_buffer_opaque_t *fvnb = vnet_buffer (fb); + fvnb->ip.reass.ip6_frag_hdr_offset = + (u8 *) frag_hdr - (u8 *) vlib_buffer_get_current (fb); + ip6_header_t *fip = vlib_buffer_get_current (fb); + ASSERT (fb->current_length > sizeof (*fip)); + ASSERT (fvnb->ip.reass.ip6_frag_hdr_offset > 0 && + fvnb->ip.reass.ip6_frag_hdr_offset < fb->current_length); + u32 fragment_first = fvnb->ip.reass.fragment_first = + ip6_frag_hdr_offset_bytes (frag_hdr); + u32 fragment_length = + vlib_buffer_length_in_chain (vm, fb) - + (fvnb->ip.reass.ip6_frag_hdr_offset + sizeof (*frag_hdr)); + u32 fragment_last = fvnb->ip.reass.fragment_last = + fragment_first + fragment_length - 1; + int more_fragments = ip6_frag_hdr_more (frag_hdr); + u32 candidate_range_bi = reass->first_bi; + u32 prev_range_bi = ~0; + fvnb->ip.reass.range_first = fragment_first; + fvnb->ip.reass.range_last = fragment_last; + fvnb->ip.reass.next_range_bi = ~0; + if (more_fragments && 0 != fragment_length % 8) + { + if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED)) + { + ip6_reass_add_trace (vm, node, rm, reass, *bi0, + ICMP_ERROR_FL_NOT_MULT_8, 0); + } + *next0 = next_icmp_error; + icmp6_error_set_vnet_buffer (fb, ICMP6_parameter_problem, + ICMP6_parameter_problem_erroneous_header_field, + (u8 *) & fip->payload_length - (u8 *) fip); + return; + } + if (fragment_first + fragment_length > 65535) + { + if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED)) + { + ip6_reass_add_trace (vm, node, rm, reass, *bi0, + ICMP_ERROR_FL_TOO_BIG, 0); + } + *next0 = next_icmp_error; + ip6_header_t *ip0 = vlib_buffer_get_current (fb); + icmp6_error_set_vnet_buffer (fb, ICMP6_parameter_problem, + ICMP6_parameter_problem_erroneous_header_field, + (u8 *) & frag_hdr->fragment_offset_and_more + - (u8 *) ip0); + return; + } + if (!more_fragments) + { + reass->last_packet_octet = fragment_last; + } + if (~0 == reass->first_bi) + { + // starting a new reassembly + ip6_reass_insert_range_in_chain (vm, rm, reass, prev_range_bi, *bi0); + if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED)) + { + ip6_reass_add_trace (vm, node, rm, reass, *bi0, RANGE_NEW, 0); + } + *bi0 = ~0; + return; + } + fvnb->ip.reass.estimated_mtu = + clib_min (clib_net_to_host_u16 (fip->payload_length), + fvnb->ip.reass.estimated_mtu); + while (~0 != candidate_range_bi) + { + vlib_buffer_t *candidate_b = vlib_get_buffer (vm, candidate_range_bi); + vnet_buffer_opaque_t *candidate_vnb = vnet_buffer (candidate_b); + if (fragment_first > candidate_vnb->ip.reass.range_last) + { + // this fragments starts after candidate range + prev_range_bi = candidate_range_bi; + candidate_range_bi = candidate_vnb->ip.reass.next_range_bi; + if (candidate_vnb->ip.reass.range_last < fragment_last && + ~0 == candidate_range_bi) + { + // special case - this fragment falls beyond all known ranges + ip6_reass_insert_range_in_chain (vm, rm, reass, prev_range_bi, + *bi0); + consumed = 1; + break; + } + continue; + } + if (fragment_last < candidate_vnb->ip.reass.range_first) + { + // this fragment ends before candidate range without any overlap + ip6_reass_insert_range_in_chain (vm, rm, reass, prev_range_bi, + *bi0); + consumed = 1; + } + else if (fragment_first == candidate_vnb->ip.reass.range_first && + fragment_last == candidate_vnb->ip.reass.range_last) + { + // duplicate fragment - ignore + } + else + { + // overlapping fragment - not allowed by RFC 8200 + ip6_reass_drop_all (vm, rm, reass, vec_drop_overlap); + ip6_reass_free (rm, reass); + if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED)) + { + ip6_reass_add_trace (vm, node, rm, reass, *bi0, RANGE_OVERLAP, + 0); + } + *next0 = next_drop; + *error0 = IP6_ERROR_REASS_OVERLAPPING_FRAGMENT; + } + break; + } + if (consumed) + { + if (PREDICT_FALSE (fb->flags & VLIB_BUFFER_IS_TRACED)) + { + ip6_reass_add_trace (vm, node, rm, reass, *bi0, RANGE_NEW, 0); + } + } + if (~0 != reass->last_packet_octet && + reass->data_len == reass->last_packet_octet + 1) + { + ip6_reass_finalize (vm, node, rm, reass, bi0, next0, error0, next_input, + vec_drop_compress); + } + else + { + if (consumed) + { + *bi0 = ~0; + } + else + { + *next0 = next_drop; + *error0 = IP6_ERROR_REASS_DUPLICATE_FRAGMENT; + } + } +} + +always_inline uword +ip6_reassembly (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * frame) +{ + u32 *from = vlib_frame_vector_args (frame); + u32 n_left_from, n_left_to_next, *to_next, next_index; + ip6_reass_main_t *rm = &ip6_reass_main; + + n_left_from = frame->n_vectors; + next_index = node->cached_next_index; + static u32 *vec_timeout = NULL; // indexes of buffers which timed out + static u32 *vec_drop_overlap = NULL; // indexes of buffers dropped due to overlap + static u32 *vec_drop_compress = NULL; // indexes of buffers dropped due to buffer compression + while (n_left_from > 0 || vec_len (vec_timeout) > 0 || + vec_len (vec_drop_overlap) > 0) + { + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (vec_len (vec_timeout) > 0 && n_left_to_next > 0) + { + u32 bi = vec_pop (vec_timeout); + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + b->error = node->errors[IP6_ERROR_REASS_TIMEOUT]; + to_next[0] = bi; + to_next += 1; + n_left_to_next -= 1; + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi, + IP6_REASSEMBLY_NEXT_DROP); + ASSERT (rm->buffers_n > 0); + --rm->buffers_n; + } + + while (vec_len (vec_drop_overlap) > 0 && n_left_to_next > 0) + { + u32 bi = vec_pop (vec_drop_overlap); + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + b->error = node->errors[IP6_ERROR_REASS_OVERLAPPING_FRAGMENT]; + to_next[0] = bi; + to_next += 1; + n_left_to_next -= 1; + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi, + IP6_REASSEMBLY_NEXT_DROP); + ASSERT (rm->buffers_n > 0); + --rm->buffers_n; + } + + while (vec_len (vec_drop_compress) > 0 && n_left_to_next > 0) + { + u32 bi = vec_pop (vec_drop_compress); + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + b->error = node->errors[IP6_ERROR_NONE]; + to_next[0] = bi; + to_next += 1; + n_left_to_next -= 1; + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi, + IP6_REASSEMBLY_NEXT_DROP); + ASSERT (rm->buffers_n > 0); + --rm->buffers_n; + } + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + u32 next0; //, error0; + + bi0 = from[0]; + b0 = vlib_get_buffer (vm, bi0); + + ip6_header_t *ip0 = vlib_buffer_get_current (b0); + ip6_frag_hdr_t *frag_hdr; + ip6_ext_header_t *prev_hdr; + ip6_ext_header_find_t (ip0, prev_hdr, frag_hdr, + IP_PROTOCOL_IPV6_FRAGMENTATION); + if (0 == ip6_frag_hdr_offset (frag_hdr)) + { + // first fragment - verify upper-layer is present + ip6_ext_header_t *tmp = (ip6_ext_header_t *) frag_hdr; + while (ip6_ext_hdr (tmp->next_hdr)) + { + tmp = ip6_ext_next_header (tmp); + } + if (IP_PROTOCOL_IP6_NONXT == tmp->next_hdr) + { + icmp6_error_set_vnet_buffer (b0, ICMP6_parameter_problem, + ICMP6_parameter_problem_first_fragment_has_incomplete_header_chain, + 0); + b0->error = node->errors[IP6_ERROR_REASS_MISSING_UPPER]; + + to_next[0] = bi0; + to_next += 1; + n_left_to_next -= 1; + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, + to_next, n_left_to_next, + bi0, + IP6_REASSEMBLY_NEXT_ICMP_ERROR); + goto next; + } + } + vnet_buffer (b0)->ip.reass.ip6_frag_hdr_offset = + (u8 *) frag_hdr - (u8 *) ip0; + + ip6_reass_key_t k; + k.src.as_u64[0] = ip0->src_address.as_u64[0]; + k.src.as_u64[1] = ip0->src_address.as_u64[1]; + k.dst.as_u64[0] = ip0->dst_address.as_u64[0]; + k.dst.as_u64[1] = ip0->dst_address.as_u64[1]; + k.xx_id = vnet_buffer (b0)->sw_if_index[VLIB_RX]; + k.frag_id = frag_hdr->identification; + k.proto = ip0->protocol; + k.unused = 0; + u32 icmp_bi = ~0; + ip6_reass_t *reass = + ip6_reass_find_or_create (vm, node, rm, &k, &icmp_bi, + &vec_timeout); + + u32 error0 = IP6_ERROR_NONE; + if (reass) + { + ip6_reass_update (vm, node, rm, reass, &bi0, &next0, &error0, + frag_hdr, &vec_drop_overlap, + &vec_drop_compress, IP6_REASSEMBLY_NEXT_INPUT, + IP6_REASSEMBLY_NEXT_DROP, + IP6_REASSEMBLY_NEXT_ICMP_ERROR); + } + else + { + next0 = IP6_REASSEMBLY_NEXT_DROP; + error0 = IP6_ERROR_REASS_LIMIT_REACHED; + } + + b0->error = node->errors[error0]; + + if (~0 != bi0) + { + to_next[0] = bi0; + to_next += 1; + n_left_to_next -= 1; + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + } + + if (~0 != icmp_bi) + { + next0 = IP6_REASSEMBLY_NEXT_ICMP_ERROR; + to_next[0] = icmp_bi; + to_next += 1; + n_left_to_next -= 1; + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, icmp_bi, + next0); + } + next: + from += 1; + n_left_from -= 1; + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + return frame->n_vectors; +} + +static char *ip6_reassembly_error_strings[] = { +#define _(sym, string) string, + foreach_ip6_error +#undef _ +}; + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (ip6_reass_node, static) = { + .function = ip6_reassembly, + .name = "ip6-reassembly", + .vector_size = sizeof (u32), + .format_trace = format_ip6_reass_trace, + .n_errors = ARRAY_LEN (ip6_reassembly_error_strings), + .error_strings = ip6_reassembly_error_strings, + .n_next_nodes = IP6_REASSEMBLY_N_NEXT, + .next_nodes = + { + [IP6_REASSEMBLY_NEXT_INPUT] = "ip6-input", + [IP6_REASSEMBLY_NEXT_DROP] = "ip6-drop", + [IP6_REASSEMBLY_NEXT_ICMP_ERROR] = "ip6-icmp-error", + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (ip6_reass_node, ip6_reassembly) + static u32 ip6_reass_get_nbuckets () +{ + ip6_reass_main_t *rm = &ip6_reass_main; + u32 nbuckets; + u8 i; + + nbuckets = (u32) (rm->max_reass_n / IP6_REASS_HT_LOAD_FACTOR); + + for (i = 0; i < 31; i++) + if ((1 << i) >= nbuckets) + break; + nbuckets = 1 << i; + + return nbuckets; +} + +typedef enum +{ + IP6_EVENT_CONFIG_CHANGED = 1, +} ip6_reass_event_t; + +typedef struct +{ + int failure; + clib_bihash_48_8_t *new_hash; +} ip6_rehash_cb_ctx; + +void +ip6_rehash_cb (clib_bihash_kv_48_8_t * kv, void *_ctx) +{ + ip6_rehash_cb_ctx *ctx = _ctx; + if (clib_bihash_add_del_48_8 (ctx->new_hash, kv, 1)) + { + ctx->failure = 1; + } +} + +vnet_api_error_t +ip6_reass_set (u32 timeout_ms, u32 max_reassemblies, + u32 expire_walk_interval_ms) +{ + u32 old_nbuckets = ip6_reass_get_nbuckets (); + ip6_reass_main.timeout_ms = timeout_ms; + ip6_reass_main.timeout = (f64) timeout_ms / (f64) MSEC_PER_SEC; + ip6_reass_main.max_reass_n = max_reassemblies; + ip6_reass_main.expire_walk_interval_ms = expire_walk_interval_ms; + vlib_process_signal_event (ip6_reass_main.vlib_main, + ip6_reass_main.ip6_reass_expire_node_idx, + IP6_EVENT_CONFIG_CHANGED, 0); + u32 new_nbuckets = ip6_reass_get_nbuckets (); + if (ip6_reass_main.max_reass_n > 0 && new_nbuckets > 1 && + new_nbuckets != old_nbuckets) + { + clib_bihash_48_8_t new_hash; + memset (&new_hash, 0, sizeof (new_hash)); + ip6_rehash_cb_ctx ctx; + ctx.failure = 0; + ctx.new_hash = &new_hash; + clib_bihash_init_48_8 (&new_hash, "ip6-reass", new_nbuckets, + new_nbuckets * 1024); + clib_bihash_foreach_key_value_pair_48_8 (&ip6_reass_main.hash, + ip6_rehash_cb, &ctx); + if (ctx.failure) + { + clib_bihash_free_48_8 (&new_hash); + return -1; + } + else + { + clib_bihash_free_48_8 (&ip6_reass_main.hash); + clib_memcpy (&ip6_reass_main.hash, &new_hash, + sizeof (ip6_reass_main.hash)); + } + } + return 0; +} + +vnet_api_error_t +ip6_reass_get (u32 * timeout_ms, u32 * max_reassemblies, + u32 * expire_walk_interval_ms) +{ + *timeout_ms = ip6_reass_main.timeout_ms; + *max_reassemblies = ip6_reass_main.max_reass_n; + *expire_walk_interval_ms = ip6_reass_main.expire_walk_interval_ms; + return 0; +} + +clib_error_t * +ip6_reass_init_function (vlib_main_t * vm) +{ + ip6_reass_main_t *rm = &ip6_reass_main; + clib_error_t *error = 0; + u32 nbuckets; + + rm->vlib_main = vm; + rm->vnet_main = vnet_get_main (); + + rm->reass_n = 0; + pool_alloc (rm->pool, rm->max_reass_n); + ip6_reass_set (IP6_REASS_TIMEOUT_DEFAULT_MS, + IP6_REASS_MAX_REASSEMBLIES_DEAFULT, + IP6_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS); + + nbuckets = ip6_reass_get_nbuckets (); + clib_bihash_init_48_8 (&rm->hash, "ip6-reass", nbuckets, nbuckets * 1024); + + vlib_node_t *node = vlib_get_node_by_name (vm, (u8 *) "ip6-drop"); + ASSERT (node); + rm->ip6_drop_idx = node->index; + node = vlib_get_node_by_name (vm, (u8 *) "ip6-reassembly-expire-walk"); + ASSERT (node); + rm->ip6_reass_expire_node_idx = node->index; + node = vlib_get_node_by_name (vm, (u8 *) "ip6-icmp-error"); + ASSERT (node); + rm->ip6_icmp_error_idx = node->index; + + if ((error = vlib_call_init_function (vm, ip_main_init))) + return error; + ip6_register_protocol (IP_PROTOCOL_IPV6_FRAGMENTATION, + ip6_reass_node.index); + return error; +} + +VLIB_INIT_FUNCTION (ip6_reass_init_function); + +static uword +ip6_reass_walk_expired (vlib_main_t * vm, + vlib_node_runtime_t * node, vlib_frame_t * f) +{ + ip6_reass_main_t *rm = &ip6_reass_main; + uword event_type, *event_data = 0; + + while (true) + { + vlib_process_wait_for_event_or_clock (vm, + (f64) rm->expire_walk_interval_ms + / (f64) MSEC_PER_SEC); + event_type = vlib_process_get_events (vm, &event_data); + + switch (event_type) + { + case ~0: /* no events => timeout */ + /* nothing to do here */ + break; + case IP6_EVENT_CONFIG_CHANGED: + break; + default: + clib_warning ("BUG: event type 0x%wx", event_type); + break; + } + f64 now = vlib_time_now (vm); + + ip6_reass_t *reass; + u32 *vec_timeout = NULL; + int *pool_indexes_to_free = NULL; + + int index; + /* *INDENT-OFF* */ + pool_foreach_index (index, rm->pool, ({ + reass = pool_elt_at_index (rm->pool, index); + if (now > reass->last_heard + rm->timeout) + { + vec_add1 (pool_indexes_to_free, index); + } + })); + /* *INDENT-ON* */ + int *i; + u32 *vec_icmp_bi = NULL; + /* *INDENT-OFF* */ + vec_foreach (i, pool_indexes_to_free) + { + ip6_reass_t *reass = pool_elt_at_index (rm->pool, i[0]); + u32 icmp_bi = ~0; + ip6_reass_on_timeout (vm, node, rm, reass, &icmp_bi, &vec_timeout); + if (~0 != icmp_bi) + { + vec_add1 (vec_icmp_bi, icmp_bi); + } + ip6_reass_free (rm, reass); + } + /* *INDENT-ON* */ + + while (vec_len (vec_timeout) > 0) + { + vlib_frame_t *f = vlib_get_frame_to_node (vm, rm->ip6_drop_idx); + u32 *to_next = vlib_frame_vector_args (f); + u32 n_left_to_next = VLIB_FRAME_SIZE - f->n_vectors; + u32 n_trace = 0; + while (vec_len (vec_timeout) > 0 && n_left_to_next > 0) + { + u32 bi = vec_pop (vec_timeout); + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + if (PREDICT_FALSE (b->flags & VLIB_BUFFER_IS_TRACED)) + { + if (pool_is_free_index (vm->trace_main.trace_buffer_pool, + b->trace_index)) + { + /* the trace is gone, don't trace this buffer anymore */ + b->flags &= ~VLIB_BUFFER_IS_TRACED; + } + else + { + ++n_trace; + } + } + b->error = node->errors[IP6_ERROR_REASS_TIMEOUT]; + to_next[0] = bi; + ++f->n_vectors; + to_next += 1; + n_left_to_next -= 1; + ASSERT (rm->buffers_n > 0); + --rm->buffers_n; + } + if (PREDICT_FALSE (n_trace > 0)) + { + f->flags |= VLIB_FRAME_TRACE; + } + vlib_put_frame_to_node (vm, rm->ip6_drop_idx, f); + } + + while (vec_len (vec_icmp_bi) > 0) + { + vlib_frame_t *f = + vlib_get_frame_to_node (vm, rm->ip6_icmp_error_idx); + u32 *to_next = vlib_frame_vector_args (f); + u32 n_left_to_next = VLIB_FRAME_SIZE - f->n_vectors; + u32 n_trace = 0; + while (vec_len (vec_icmp_bi) > 0 && n_left_to_next > 0) + { + u32 bi = vec_pop (vec_icmp_bi); + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + if (PREDICT_FALSE (b->flags & VLIB_BUFFER_IS_TRACED)) + { + if (pool_is_free_index (vm->trace_main.trace_buffer_pool, + b->trace_index)) + { + /* the trace is gone, don't trace this buffer anymore */ + b->flags &= ~VLIB_BUFFER_IS_TRACED; + } + else + { + ++n_trace; + } + } + b->error = node->errors[IP6_ERROR_REASS_TIMEOUT]; + to_next[0] = bi; + ++f->n_vectors; + to_next += 1; + n_left_to_next -= 1; + ASSERT (rm->buffers_n > 0); + --rm->buffers_n; + } + if (PREDICT_FALSE (n_trace > 0)) + { + f->flags |= VLIB_FRAME_TRACE; + } + vlib_put_frame_to_node (vm, rm->ip6_icmp_error_idx, f); + } + + vec_free (pool_indexes_to_free); + vec_free (vec_timeout); + vec_free (vec_icmp_bi); + if (event_data) + { + _vec_len (event_data) = 0; + } + } + + return 0; +} + +static vlib_node_registration_t ip6_reass_expire_node; + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (ip6_reass_expire_node, static) = { + .function = ip6_reass_walk_expired, + .format_trace = format_ip6_reass_trace, + .type = VLIB_NODE_TYPE_PROCESS, + .name = "ip6-reassembly-expire-walk", + + .n_errors = ARRAY_LEN (ip6_reassembly_error_strings), + .error_strings = ip6_reassembly_error_strings, + +}; +/* *INDENT-ON* */ + +static u8 * +format_ip6_reass_key (u8 * s, va_list * args) +{ + ip6_reass_key_t *key = va_arg (*args, ip6_reass_key_t *); + s = format (s, "xx_id: %u, src: %U, dst: %U, frag_id: %u, proto: %u", + key->xx_id, format_ip6_address, &key->src, format_ip6_address, + &key->dst, clib_net_to_host_u16 (key->frag_id), key->proto); + return s; +} + +static u8 * +format_ip6_reass (u8 * s, va_list * args) +{ + vlib_main_t *vm = va_arg (*args, vlib_main_t *); + ip6_reass_t *reass = va_arg (*args, ip6_reass_t *); + + s = format (s, "ID: %u, key: %U\n first_bi: %u, data_len: %u, " + "last_packet_octet: %u, trace_op_counter: %u\n", + reass->id, format_ip6_reass_key, &reass->key, reass->first_bi, + reass->data_len, reass->last_packet_octet, + reass->trace_op_counter); + u32 bi = reass->first_bi; + u32 counter = 0; + while (~0 != bi) + { + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + vnet_buffer_opaque_t *vnb = vnet_buffer (b); + s = format (s, " #%03u: range: [%u, %u], bi: %u, off: %d, len: %u, " + "fragment[%u, %u]\n", + counter, vnb->ip.reass.range_first, + vnb->ip.reass.range_last, bi, + ip6_reass_buffer_get_data_offset_no_check (b), + ip6_reass_buffer_get_data_len_no_check (b), + vnb->ip.reass.fragment_first, vnb->ip.reass.fragment_last); + if (b->flags & VLIB_BUFFER_NEXT_PRESENT) + { + bi = b->next_buffer; + } + else + { + bi = ~0; + } + } + return s; +} + +static clib_error_t * +show_ip6_reass (vlib_main_t * vm, unformat_input_t * input, + CLIB_UNUSED (vlib_cli_command_t * lmd)) +{ + ip6_reass_main_t *rm = &ip6_reass_main; + + vlib_cli_output (vm, "---------------------"); + vlib_cli_output (vm, "IP6 reassembly status"); + vlib_cli_output (vm, "---------------------"); + if (unformat (input, "details")) + { + ip6_reass_t *reass; + /* *INDENT-OFF* */ + pool_foreach (reass, rm->pool, { + vlib_cli_output (vm, "%U", format_ip6_reass, vm, reass); + }); + /* *INDENT-ON* */ + } + vlib_cli_output (vm, "---------------------"); + vlib_cli_output (vm, "Current IP6 reassemblies count: %lu\n", rm->reass_n); + vlib_cli_output (vm, + "Maximum configured concurrent IP6 reassemblies: %lu\n", + (long unsigned) rm->max_reass_n); + vlib_cli_output (vm, "Buffers in use: %lu\n", + (long unsigned) rm->buffers_n); + return 0; +} + +/* *INDENT-OFF* */ +VLIB_CLI_COMMAND (show_ip6_reassembly_cmd, static) = { + .path = "show ip6-reassembly", + .short_help = "show ip6-reassembly [details]", + .function = show_ip6_reass, +}; +/* *INDENT-ON* */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/ip/ip6_reassembly.h b/src/vnet/ip/ip6_reassembly.h new file mode 100644 index 00000000000..0de4e04bb2b --- /dev/null +++ b/src/vnet/ip/ip6_reassembly.h @@ -0,0 +1,49 @@ +/* + * 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 IPv6 Reassembly. + * + * This file contains the source code for IPv6 reassembly. + */ + +#ifndef __included_ip6_reassembly_h__ +#define __included_ip6_reassembly_h__ + +#include <vnet/api_errno.h> +#include <vnet/vnet.h> + +/** + * @brief set ip6 reassembly configuration + */ +vnet_api_error_t ip6_reass_set (u32 timeout_ms, u32 max_reassemblies, + u32 expire_walk_interval_ms); + +/** + * @brief get ip6 reassembly configuration + */ +vnet_api_error_t ip6_reass_get (u32 * timeout_ms, u32 * max_reassemblies, + u32 * expire_walk_interval_ms); + +#endif /* __included_ip6_reassembly_h */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/ip/ip_api.c b/src/vnet/ip/ip_api.c index c0889ebfa6b..60fa2faf926 100644 --- a/src/vnet/ip/ip_api.c +++ b/src/vnet/ip/ip_api.c @@ -42,6 +42,8 @@ #include <vnet/fib/ip4_fib.h> #include <vnet/fib/ip6_fib.h> #include <vnet/ip/ip6_hop_by_hop.h> +#include <vnet/ip/ip4_reassembly.h> +#include <vnet/ip/ip6_reassembly.h> #include <vnet/vnet_msg_enum.h> @@ -97,7 +99,9 @@ _(IOAM_DISABLE, ioam_disable) \ _(IP_SOURCE_AND_PORT_RANGE_CHECK_ADD_DEL, \ ip_source_and_port_range_check_add_del) \ _(IP_SOURCE_AND_PORT_RANGE_CHECK_INTERFACE_ADD_DEL, \ - ip_source_and_port_range_check_interface_add_del) + ip_source_and_port_range_check_interface_add_del) \ +_(IP_REASSEMBLY_SET, ip_reassembly_set) \ +_(IP_REASSEMBLY_GET, ip_reassembly_get) extern void stats_dslock_with_hint (int hint, int tag); extern void stats_dsunlock (void); @@ -2786,6 +2790,61 @@ vl_api_set_arp_neighbor_limit_t_handler (vl_api_set_arp_neighbor_limit_t * mp) REPLY_MACRO (VL_API_SET_ARP_NEIGHBOR_LIMIT_REPLY); } +void +vl_api_ip_reassembly_set_t_handler (vl_api_ip_reassembly_set_t * mp) +{ + vl_api_ip_reassembly_set_reply_t *rmp; + int rv = 0; + if (mp->is_ip6) + { + rv = ip6_reass_set (clib_net_to_host_u32 (mp->timeout_ms), + clib_net_to_host_u32 (mp->max_reassemblies), + clib_net_to_host_u32 (mp->expire_walk_interval_ms)); + } + else + { + rv = ip4_reass_set (clib_net_to_host_u32 (mp->timeout_ms), + clib_net_to_host_u32 (mp->max_reassemblies), + clib_net_to_host_u32 (mp->expire_walk_interval_ms)); + } + + REPLY_MACRO (VL_API_IP_REASSEMBLY_SET_REPLY); +} + +void +vl_api_ip_reassembly_get_t_handler (vl_api_ip_reassembly_get_t * mp) +{ + unix_shared_memory_queue_t *q; + + q = vl_api_client_index_to_input_queue (mp->client_index); + + if (q == 0) + return; + + vl_api_ip_reassembly_get_reply_t *rmp = vl_msg_api_alloc (sizeof (*rmp)); + memset (rmp, 0, sizeof (*rmp)); + rmp->_vl_msg_id = ntohs (VL_API_IP_REASSEMBLY_GET_REPLY); + rmp->context = mp->context; + rmp->retval = 0; + if (mp->is_ip6) + { + rmp->is_ip6 = 1; + ip6_reass_get (&rmp->timeout_ms, &rmp->max_reassemblies, + &rmp->expire_walk_interval_ms); + } + else + { + rmp->is_ip6 = 0; + ip4_reass_get (&rmp->timeout_ms, &rmp->max_reassemblies, + &rmp->expire_walk_interval_ms); + } + rmp->timeout_ms = clib_host_to_net_u32 (rmp->timeout_ms); + rmp->max_reassemblies = clib_host_to_net_u32 (rmp->max_reassemblies); + rmp->expire_walk_interval_ms = + clib_host_to_net_u32 (rmp->expire_walk_interval_ms); + vl_msg_api_send_shmem (q, (u8 *) & rmp); +} + #define vl_msg_name_crc_list #include <vnet/ip/ip.api.h> #undef vl_msg_name_crc_list diff --git a/src/vnet/ip/lookup.c b/src/vnet/ip/lookup.c index 67e46bd8cfa..f2880bf8cd9 100644 --- a/src/vnet/ip/lookup.c +++ b/src/vnet/ip/lookup.c @@ -230,6 +230,8 @@ ip_lookup_init (ip_lookup_main_t * lm, u32 is_ip6) } lm->local_next_by_ip_protocol[IP_PROTOCOL_UDP] = IP_LOCAL_NEXT_UDP_LOOKUP; + lm->local_next_by_ip_protocol[IP_PROTOCOL_VPP_FRAGMENTATION] = + IP_LOCAL_NEXT_REASSEMBLY; lm->local_next_by_ip_protocol[is_ip6 ? IP_PROTOCOL_ICMP6 : IP_PROTOCOL_ICMP] = IP_LOCAL_NEXT_ICMP; lm->builtin_protocol_by_ip_protocol[IP_PROTOCOL_UDP] = diff --git a/src/vnet/ip/lookup.h b/src/vnet/ip/lookup.h index a89546f5d92..9fe7a1b0085 100644 --- a/src/vnet/ip/lookup.h +++ b/src/vnet/ip/lookup.h @@ -111,6 +111,7 @@ typedef enum IP_LOCAL_NEXT_PUNT, IP_LOCAL_NEXT_UDP_LOOKUP, IP_LOCAL_NEXT_ICMP, + IP_LOCAL_NEXT_REASSEMBLY, IP_LOCAL_N_NEXT, } ip_local_next_t; diff --git a/src/vnet/ip/protocols.def b/src/vnet/ip/protocols.def index 77fab31da05..4ec0bbc9284 100644 --- a/src/vnet/ip/protocols.def +++ b/src/vnet/ip/protocols.def @@ -158,5 +158,10 @@ ip_protocol (134, RSVP_E2E_IGNORE) ip_protocol (135, MOBILITY) ip_protocol (136, UDP_LITE) ip_protocol (137, MPLS_IN_IP) +/* + * VPPs way of dealing with fragments is to mark them as 0xfe, to be picked + * by the corresponding nodes based on this protocol + */ +ip_protocol (0xfe, VPP_FRAGMENTATION) ip_protocol (255, RESERVED) |