diff options
Diffstat (limited to 'src/vnet/ip/reass')
-rw-r--r-- | src/vnet/ip/reass/ip4_full_reass.c | 687 | ||||
-rw-r--r-- | src/vnet/ip/reass/ip4_full_reass.h | 3 | ||||
-rw-r--r-- | src/vnet/ip/reass/ip4_sv_reass.c | 486 | ||||
-rw-r--r-- | src/vnet/ip/reass/ip4_sv_reass.h | 1 | ||||
-rw-r--r-- | src/vnet/ip/reass/ip6_full_reass.c | 741 | ||||
-rw-r--r-- | src/vnet/ip/reass/ip6_full_reass.h | 2 | ||||
-rw-r--r-- | src/vnet/ip/reass/ip6_sv_reass.c | 309 | ||||
-rw-r--r-- | src/vnet/ip/reass/ip6_sv_reass.h | 1 | ||||
-rw-r--r-- | src/vnet/ip/reass/reassembly.rst | 221 |
9 files changed, 1725 insertions, 726 deletions
diff --git a/src/vnet/ip/reass/ip4_full_reass.c b/src/vnet/ip/reass/ip4_full_reass.c index d2069c0876c..bab7d479dcf 100644 --- a/src/vnet/ip/reass/ip4_full_reass.c +++ b/src/vnet/ip/reass/ip4_full_reass.c @@ -23,16 +23,21 @@ #include <vppinfra/vec.h> #include <vnet/vnet.h> #include <vnet/ip/ip.h> +#include <vnet/ip/ip.api_enum.h> #include <vppinfra/fifo.h> #include <vppinfra/bihash_16_8.h> #include <vnet/ip/reass/ip4_full_reass.h> #include <stddef.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_TIMEOUT_DEFAULT_MS 200 + +/* As there are only 1024 reass context per thread, either the DDOS attacks + * or fractions of real timeouts, would consume these contexts quickly and + * running out context space and unable to perform reassembly */ +#define IP4_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS 50 // 50 ms default #define IP4_REASS_MAX_REASSEMBLIES_DEFAULT 1024 -#define IP4_REASS_MAX_REASSEMBLY_LENGTH_DEFAULT 3 +#define IP4_REASS_MAX_REASSEMBLY_LENGTH_DEFAULT 3 #define IP4_REASS_HT_LOAD_FACTOR (0.75) #define IP4_REASS_DEBUG_BUFFERS 0 @@ -68,21 +73,19 @@ typedef enum typedef struct { - union + struct { - struct - { - u32 xx_id; - ip4_address_t src; - ip4_address_t dst; - u16 frag_id; - u8 proto; - u8 unused; - }; - u64 as_u64[2]; + u16 frag_id; + u8 proto; + u8 unused; + u32 fib_index; + ip4_address_t src; + ip4_address_t dst; }; } ip4_full_reass_key_t; +STATIC_ASSERT_SIZEOF (ip4_full_reass_key_t, 16); + typedef union { struct @@ -155,6 +158,8 @@ typedef struct ip4_full_reass_t *pool; u32 reass_n; u32 id_counter; + // for pacing the main thread timeouts + u32 last_id; clib_spinlock_t lock; } ip4_full_reass_per_thread_t; @@ -177,17 +182,19 @@ typedef struct // convenience vlib_main_t *vlib_main; - // node index of ip4-drop node - u32 ip4_drop_idx; u32 ip4_full_reass_expire_node_idx; /** Worker handoff */ u32 fq_index; + u32 fq_local_index; u32 fq_feature_index; u32 fq_custom_index; // reference count for enabling/disabling feature - per interface u32 *feature_use_refcount_per_intf; + + // whether local fragmented packets are reassembled or not + int is_local_reass_enabled; } ip4_full_reass_main_t; extern ip4_full_reass_main_t ip4_full_reass_main; @@ -219,6 +226,7 @@ typedef enum RANGE_OVERLAP, FINALIZE, HANDOFF, + PASSTHROUGH, } ip4_full_reass_trace_operation_e; typedef struct @@ -329,6 +337,9 @@ format_ip4_full_reass_trace (u8 * s, va_list * args) format (s, "handoff from thread #%u to thread #%u", t->thread_id, t->thread_id_to); break; + case PASSTHROUGH: + s = format (s, "passthrough - not a fragment"); + break; } return s; } @@ -404,13 +415,16 @@ ip4_full_reass_free (ip4_full_reass_main_t * rm, ip4_full_reass_per_thread_t * rt, ip4_full_reass_t * reass) { - clib_bihash_kv_16_8_t kv; - kv.key[0] = reass->key.as_u64[0]; - kv.key[1] = reass->key.as_u64[1]; + clib_bihash_kv_16_8_t kv = {}; + clib_memcpy_fast (&kv, &reass->key, sizeof (kv.key)); clib_bihash_add_del_16_8 (&rm->hash, &kv, 0); return ip4_full_reass_free_ctx (rt, reass); } +/* n_left_to_next, and to_next are taken as input params, as this function + * could be called from a graphnode, where its managing local copy of these + * variables, and ignoring those and still trying to enqueue the buffers + * with local variables would cause either buffer leak or corruption */ always_inline void ip4_full_reass_drop_all (vlib_main_t *vm, vlib_node_runtime_t *node, ip4_full_reass_t *reass) @@ -419,58 +433,103 @@ ip4_full_reass_drop_all (vlib_main_t *vm, vlib_node_runtime_t *node, vlib_buffer_t *range_b; vnet_buffer_opaque_t *range_vnb; u32 *to_free = NULL; + 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) + + if (~0 != range_bi) { - vec_add1 (to_free, 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; - } + vec_add1 (to_free, range_bi); } + range_bi = range_vnb->ip.reass.next_range_bi; } + /* send to next_error_index */ - if (~0 != reass->error_next_index) + if (~0 != reass->error_next_index && + reass->error_next_index < node->n_next_nodes) { - u32 n_left_to_next, *to_next, next_index; + u32 n_free = vec_len (to_free); + + /* record number of packets sent to custom app */ + vlib_node_increment_counter (vm, node->node_index, + IP4_ERROR_REASS_TO_CUSTOM_APP, n_free); + + if (node->flags & VLIB_NODE_FLAG_TRACE) + for (u32 i = 0; i < n_free; i++) + { + vlib_buffer_t *b = vlib_get_buffer (vm, to_free[i]); + if (PREDICT_FALSE (b->flags & VLIB_BUFFER_IS_TRACED)) + ip4_full_reass_add_trace (vm, node, reass, to_free[i], + RANGE_DISCARD, 0, ~0); + } - next_index = reass->error_next_index; - u32 bi = ~0; + vlib_buffer_enqueue_to_single_next (vm, node, to_free, + reass->error_next_index, n_free); + } + else + { + vlib_buffer_free (vm, to_free, vec_len (to_free)); + } + vec_free (to_free); +} - while (vec_len (to_free) > 0) - { - vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); +always_inline void +sanitize_reass_buffers_add_missing (vlib_main_t *vm, ip4_full_reass_t *reass, + u32 *bi0) +{ + u32 range_bi = reass->first_bi; + vlib_buffer_t *range_b; + vnet_buffer_opaque_t *range_vnb; - while (vec_len (to_free) > 0 && n_left_to_next > 0) + while (~0 != range_bi) + { + range_b = vlib_get_buffer (vm, range_bi); + range_vnb = vnet_buffer (range_b); + u32 bi = range_bi; + if (~0 != bi) + { + if (bi == *bi0) + *bi0 = ~0; + if (range_b->flags & VLIB_BUFFER_NEXT_PRESENT) { - bi = vec_pop (to_free); - - if (~0 != bi) + u32 _bi = bi; + vlib_buffer_t *_b = vlib_get_buffer (vm, _bi); + while (_b->flags & VLIB_BUFFER_NEXT_PRESENT) { - to_next[0] = bi; - to_next += 1; - n_left_to_next -= 1; + if (_b->next_buffer != range_vnb->ip.reass.next_range_bi) + { + _bi = _b->next_buffer; + _b = vlib_get_buffer (vm, _bi); + } + else + { + _b->flags &= ~VLIB_BUFFER_NEXT_PRESENT; + break; + } } } - vlib_put_next_frame (vm, node, next_index, n_left_to_next); + range_bi = range_vnb->ip.reass.next_range_bi; } } - else + if (*bi0 != ~0) { - vlib_buffer_free (vm, to_free, vec_len (to_free)); + vlib_buffer_t *fb = vlib_get_buffer (vm, *bi0); + vnet_buffer_opaque_t *fvnb = vnet_buffer (fb); + if (~0 != reass->first_bi) + { + fvnb->ip.reass.next_range_bi = reass->first_bi; + reass->first_bi = *bi0; + } + else + { + reass->first_bi = *bi0; + fvnb->ip.reass.next_range_bi = ~0; + } + *bi0 = ~0; } - vec_free (to_free); } always_inline void @@ -484,10 +543,10 @@ ip4_full_reass_init (ip4_full_reass_t * reass) } always_inline ip4_full_reass_t * -ip4_full_reass_find_or_create (vlib_main_t * vm, vlib_node_runtime_t * node, - ip4_full_reass_main_t * rm, - ip4_full_reass_per_thread_t * rt, - ip4_full_reass_kv_t * kv, u8 * do_handoff) +ip4_full_reass_find_or_create (vlib_main_t *vm, vlib_node_runtime_t *node, + ip4_full_reass_main_t *rm, + ip4_full_reass_per_thread_t *rt, + ip4_full_reass_kv_t *kv, u8 *do_handoff) { ip4_full_reass_t *reass; f64 now; @@ -510,6 +569,8 @@ again: if (now > reass->last_heard + rm->timeout) { + vlib_node_increment_counter (vm, node->node_index, + IP4_ERROR_REASS_TIMEOUT, 1); ip4_full_reass_drop_all (vm, node, reass); ip4_full_reass_free (rm, rt, reass); reass = NULL; @@ -538,8 +599,7 @@ again: ++rt->reass_n; } - reass->key.as_u64[0] = kv->kv.key[0]; - reass->key.as_u64[1] = kv->kv.key[1]; + clib_memcpy_fast (&reass->key, &kv->kv.key, sizeof (reass->key)); kv->v.reass_index = (reass - rt->pool); kv->v.memory_owner_thread_index = vm->thread_index; reass->last_heard = now; @@ -568,7 +628,6 @@ ip4_full_reass_finalize (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_buffer_t *last_b = NULL; u32 sub_chain_bi = reass->first_bi; u32 total_length = 0; - u32 buf_cnt = 0; do { u32 tmp_bi = sub_chain_bi; @@ -605,7 +664,6 @@ ip4_full_reass_finalize (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_buffer_length_in_chain (vm, tmp) - trim_front - trim_end; while (1) { - ++buf_cnt; if (trim_front) { if (trim_front > tmp->current_length) @@ -755,6 +813,16 @@ ip4_full_reass_finalize (vlib_main_t * vm, vlib_node_runtime_t * node, *next0 = reass->next_index; } vnet_buffer (first_b)->ip.reass.estimated_mtu = reass->min_fragment_length; + + /* Keep track of number of successfully reassembled packets and number of + * fragments reassembled */ + vlib_node_increment_counter (vm, node->node_index, IP4_ERROR_REASS_SUCCESS, + 1); + + vlib_node_increment_counter (vm, node->node_index, + IP4_ERROR_REASS_FRAGMENTS_REASSEMBLED, + reass->fragments_n); + *error0 = IP4_ERROR_NONE; ip4_full_reass_free (rm, rt, reass); reass = NULL; @@ -1090,199 +1158,216 @@ ip4_full_reass_update (vlib_main_t * vm, vlib_node_runtime_t * node, } always_inline uword -ip4_full_reass_inline (vlib_main_t * vm, vlib_node_runtime_t * node, - vlib_frame_t * frame, ip4_full_reass_node_type_t type) +ip4_full_reass_inline (vlib_main_t *vm, vlib_node_runtime_t *node, + vlib_frame_t *frame, ip4_full_reass_node_type_t type, + bool is_local) { u32 *from = vlib_frame_vector_args (frame); - u32 n_left_from, n_left_to_next, *to_next, next_index; + u32 n_left, n_next = 0, to_next[VLIB_FRAME_SIZE]; ip4_full_reass_main_t *rm = &ip4_full_reass_main; ip4_full_reass_per_thread_t *rt = &rm->per_thread_data[vm->thread_index]; + u16 nexts[VLIB_FRAME_SIZE]; + clib_spinlock_lock (&rt->lock); - n_left_from = frame->n_vectors; - next_index = node->cached_next_index; - while (n_left_from > 0) + n_left = frame->n_vectors; + while (n_left > 0) { - vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); - - while (n_left_from > 0 && n_left_to_next > 0) - { - u32 bi0; - vlib_buffer_t *b0; - u32 next0; - u32 error0 = IP4_ERROR_NONE; + u32 bi0; + vlib_buffer_t *b0; + u32 next0; + u32 error0 = IP4_ERROR_NONE; - bi0 = from[0]; - b0 = vlib_get_buffer (vm, bi0); + bi0 = from[0]; + b0 = vlib_get_buffer (vm, bi0); - ip4_header_t *ip0 = vlib_buffer_get_current (b0); - if (!ip4_get_fragment_more (ip0) && !ip4_get_fragment_offset (ip0)) + ip4_header_t *ip0 = vlib_buffer_get_current (b0); + if (!ip4_get_fragment_more (ip0) && !ip4_get_fragment_offset (ip0)) + { + // this is a whole packet - no fragmentation + if (CUSTOM != type) { - // this is a whole packet - no fragmentation - if (CUSTOM != type) - { - next0 = IP4_FULL_REASS_NEXT_INPUT; - } - else - { - next0 = vnet_buffer (b0)->ip.reass.next_index; - } - goto packet_enqueue; + next0 = IP4_FULL_REASS_NEXT_INPUT; } - const u32 fragment_first = ip4_get_fragment_offset_bytes (ip0); - const u32 fragment_length = - clib_net_to_host_u16 (ip0->length) - ip4_header_bytes (ip0); - const u32 fragment_last = fragment_first + fragment_length - 1; - if (fragment_first > fragment_last || fragment_first + fragment_length > UINT16_MAX - 20 || (fragment_length < 8 && ip4_get_fragment_more (ip0))) // 8 is minimum frag length per RFC 791 + else { - next0 = IP4_FULL_REASS_NEXT_DROP; - error0 = IP4_ERROR_REASS_MALFORMED_PACKET; - goto packet_enqueue; + next0 = vnet_buffer (b0)->ip.reass.next_index; } - ip4_full_reass_kv_t kv; - u8 do_handoff = 0; - - kv.k.as_u64[0] = - (u64) vec_elt (ip4_main.fib_index_by_sw_if_index, - vnet_buffer (b0)->sw_if_index[VLIB_RX]) | - (u64) ip0->src_address.as_u32 << 32; - kv.k.as_u64[1] = - (u64) ip0->dst_address. - as_u32 | (u64) ip0->fragment_id << 32 | (u64) ip0->protocol << 48; - - ip4_full_reass_t *reass = - ip4_full_reass_find_or_create (vm, node, rm, rt, &kv, - &do_handoff); - - if (reass) + ip4_full_reass_add_trace (vm, node, NULL, bi0, PASSTHROUGH, 0, ~0); + goto packet_enqueue; + } + + if (is_local && !rm->is_local_reass_enabled) + { + next0 = IP4_FULL_REASS_NEXT_DROP; + goto packet_enqueue; + } + + const u32 fragment_first = ip4_get_fragment_offset_bytes (ip0); + const u32 fragment_length = + clib_net_to_host_u16 (ip0->length) - ip4_header_bytes (ip0); + const u32 fragment_last = fragment_first + fragment_length - 1; + + /* Keep track of received fragments */ + vlib_node_increment_counter (vm, node->node_index, + IP4_ERROR_REASS_FRAGMENTS_RCVD, 1); + + if (fragment_first > fragment_last || + fragment_first + fragment_length > UINT16_MAX - 20 || + (fragment_length < 8 && // 8 is minimum frag length per RFC 791 + ip4_get_fragment_more (ip0))) + { + next0 = IP4_FULL_REASS_NEXT_DROP; + error0 = IP4_ERROR_REASS_MALFORMED_PACKET; + goto packet_enqueue; + } + + u32 fib_index = (vnet_buffer (b0)->sw_if_index[VLIB_TX] == (u32) ~0) ? + vec_elt (ip4_main.fib_index_by_sw_if_index, + vnet_buffer (b0)->sw_if_index[VLIB_RX]) : + vnet_buffer (b0)->sw_if_index[VLIB_TX]; + + ip4_full_reass_kv_t kv = { .k.fib_index = fib_index, + .k.src.as_u32 = ip0->src_address.as_u32, + .k.dst.as_u32 = ip0->dst_address.as_u32, + .k.frag_id = ip0->fragment_id, + .k.proto = ip0->protocol + + }; + u8 do_handoff = 0; + + ip4_full_reass_t *reass = + ip4_full_reass_find_or_create (vm, node, rm, rt, &kv, &do_handoff); + + if (reass) + { + const u32 fragment_first = ip4_get_fragment_offset_bytes (ip0); + if (0 == fragment_first) { - const u32 fragment_first = ip4_get_fragment_offset_bytes (ip0); - if (0 == fragment_first) - { - reass->sendout_thread_index = vm->thread_index; - } + reass->sendout_thread_index = vm->thread_index; } + } - if (PREDICT_FALSE (do_handoff)) + if (PREDICT_FALSE (do_handoff)) + { + next0 = IP4_FULL_REASS_NEXT_HANDOFF; + vnet_buffer (b0)->ip.reass.owner_thread_index = + kv.v.memory_owner_thread_index; + } + else if (reass) + { + u32 handoff_thread_idx; + u32 counter = ~0; + switch (ip4_full_reass_update (vm, node, rm, rt, reass, &bi0, &next0, + &error0, CUSTOM == type, + &handoff_thread_idx)) { + case IP4_REASS_RC_OK: + /* nothing to do here */ + break; + case IP4_REASS_RC_HANDOFF: next0 = IP4_FULL_REASS_NEXT_HANDOFF; + b0 = vlib_get_buffer (vm, bi0); vnet_buffer (b0)->ip.reass.owner_thread_index = - kv.v.memory_owner_thread_index; - } - else if (reass) - { - u32 handoff_thread_idx; - switch (ip4_full_reass_update - (vm, node, rm, rt, reass, &bi0, &next0, - &error0, CUSTOM == type, &handoff_thread_idx)) - { - case IP4_REASS_RC_OK: - /* nothing to do here */ - break; - case IP4_REASS_RC_HANDOFF: - next0 = IP4_FULL_REASS_NEXT_HANDOFF; - b0 = vlib_get_buffer (vm, bi0); - vnet_buffer (b0)->ip.reass.owner_thread_index = - handoff_thread_idx; - break; - case IP4_REASS_RC_TOO_MANY_FRAGMENTS: - vlib_node_increment_counter (vm, node->node_index, - IP4_ERROR_REASS_FRAGMENT_CHAIN_TOO_LONG, - 1); - ip4_full_reass_drop_all (vm, node, reass); - ip4_full_reass_free (rm, rt, reass); - goto next_packet; - break; - case IP4_REASS_RC_NO_BUF: - vlib_node_increment_counter (vm, node->node_index, - IP4_ERROR_REASS_NO_BUF, 1); - ip4_full_reass_drop_all (vm, node, reass); - ip4_full_reass_free (rm, rt, reass); - goto next_packet; - break; - case IP4_REASS_RC_INTERNAL_ERROR: - /* drop everything and start with a clean slate */ - vlib_node_increment_counter (vm, node->node_index, - IP4_ERROR_REASS_INTERNAL_ERROR, - 1); - ip4_full_reass_drop_all (vm, node, reass); - ip4_full_reass_free (rm, rt, reass); - goto next_packet; - break; - } + handoff_thread_idx; + break; + case IP4_REASS_RC_TOO_MANY_FRAGMENTS: + counter = IP4_ERROR_REASS_FRAGMENT_CHAIN_TOO_LONG; + break; + case IP4_REASS_RC_NO_BUF: + counter = IP4_ERROR_REASS_NO_BUF; + break; + case IP4_REASS_RC_INTERNAL_ERROR: + counter = IP4_ERROR_REASS_INTERNAL_ERROR; + /* Sanitization is needed in internal error cases only, as + * the incoming packet is already dropped in other cases, + * also adding bi0 back to the reassembly list, fixes the + * leaking of buffers during internal errors. + * + * Also it doesnt make sense to send these buffers custom + * app, these fragments are with internal errors */ + sanitize_reass_buffers_add_missing (vm, reass, &bi0); + reass->error_next_index = ~0; + break; } - else + + if (~0 != counter) { - next0 = IP4_FULL_REASS_NEXT_DROP; - error0 = IP4_ERROR_REASS_LIMIT_REACHED; + vlib_node_increment_counter (vm, node->node_index, counter, 1); + ip4_full_reass_drop_all (vm, node, reass); + ip4_full_reass_free (rm, rt, reass); + goto next_packet; } + } + else + { + next0 = IP4_FULL_REASS_NEXT_DROP; + error0 = IP4_ERROR_REASS_LIMIT_REACHED; + } + packet_enqueue: - packet_enqueue: - - if (bi0 != ~0) + if (bi0 != ~0) + { + /* bi0 might have been updated by reass_finalize, reload */ + b0 = vlib_get_buffer (vm, bi0); + if (IP4_ERROR_NONE != error0) { - to_next[0] = bi0; - to_next += 1; - n_left_to_next -= 1; + b0->error = node->errors[error0]; + } - /* bi0 might have been updated by reass_finalize, reload */ - b0 = vlib_get_buffer (vm, bi0); - if (IP4_ERROR_NONE != error0) + if (next0 == IP4_FULL_REASS_NEXT_HANDOFF) + { + if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) { - b0->error = node->errors[error0]; + ip4_full_reass_add_trace ( + vm, node, NULL, bi0, HANDOFF, 0, + vnet_buffer (b0)->ip.reass.owner_thread_index); } + } + else if (FEATURE == type && IP4_ERROR_NONE == error0) + { + vnet_feature_next (&next0, b0); + } - if (next0 == IP4_FULL_REASS_NEXT_HANDOFF) - { - if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) - { - ip4_full_reass_add_trace ( - vm, node, NULL, bi0, HANDOFF, 0, - vnet_buffer (b0)->ip.reass.owner_thread_index); - } - } - else if (FEATURE == type && IP4_ERROR_NONE == error0) - { - vnet_feature_next (&next0, b0); - } - vlib_validate_buffer_enqueue_x1 (vm, node, next_index, - to_next, n_left_to_next, - bi0, next0); - IP4_REASS_DEBUG_BUFFER (bi0, enqueue_next); + /* Increment the counter to-custom-app also as this fragment is + * also going to application */ + if (CUSTOM == type) + { + vlib_node_increment_counter (vm, node->node_index, + IP4_ERROR_REASS_TO_CUSTOM_APP, 1); } - next_packet: - from += 1; - n_left_from -= 1; + to_next[n_next] = bi0; + nexts[n_next] = next0; + n_next++; + IP4_REASS_DEBUG_BUFFER (bi0, enqueue_next); } - vlib_put_next_frame (vm, node, next_index, n_left_to_next); + next_packet: + from += 1; + n_left -= 1; } clib_spinlock_unlock (&rt->lock); + + vlib_buffer_enqueue_to_next (vm, node, to_next, nexts, n_next); return frame->n_vectors; } -static char *ip4_full_reass_error_strings[] = { -#define _(sym, string) string, - foreach_ip4_error -#undef _ -}; - VLIB_NODE_FN (ip4_full_reass_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip4_full_reass_inline (vm, node, frame, NORMAL); + return ip4_full_reass_inline (vm, node, frame, NORMAL, false /* is_local */); } VLIB_REGISTER_NODE (ip4_full_reass_node) = { .name = "ip4-full-reassembly", .vector_size = sizeof (u32), .format_trace = format_ip4_full_reass_trace, - .n_errors = ARRAY_LEN (ip4_full_reass_error_strings), - .error_strings = ip4_full_reass_error_strings, + .n_errors = IP4_N_ERROR, + .error_counters = ip4_error_counters, .n_next_nodes = IP4_FULL_REASS_N_NEXT, .next_nodes = { @@ -1293,19 +1378,42 @@ VLIB_REGISTER_NODE (ip4_full_reass_node) = { }, }; +VLIB_NODE_FN (ip4_local_full_reass_node) +(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame) +{ + return ip4_full_reass_inline (vm, node, frame, NORMAL, true /* is_local */); +} + +VLIB_REGISTER_NODE (ip4_local_full_reass_node) = { + .name = "ip4-local-full-reassembly", + .vector_size = sizeof (u32), + .format_trace = format_ip4_full_reass_trace, + .n_errors = IP4_N_ERROR, + .error_counters = ip4_error_counters, + .n_next_nodes = IP4_FULL_REASS_N_NEXT, + .next_nodes = + { + [IP4_FULL_REASS_NEXT_INPUT] = "ip4-input", + [IP4_FULL_REASS_NEXT_DROP] = "ip4-drop", + [IP4_FULL_REASS_NEXT_HANDOFF] = "ip4-local-full-reassembly-handoff", + + }, +}; + VLIB_NODE_FN (ip4_full_reass_node_feature) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip4_full_reass_inline (vm, node, frame, FEATURE); + return ip4_full_reass_inline (vm, node, frame, FEATURE, + false /* is_local */); } VLIB_REGISTER_NODE (ip4_full_reass_node_feature) = { .name = "ip4-full-reassembly-feature", .vector_size = sizeof (u32), .format_trace = format_ip4_full_reass_trace, - .n_errors = ARRAY_LEN (ip4_full_reass_error_strings), - .error_strings = ip4_full_reass_error_strings, + .n_errors = IP4_N_ERROR, + .error_counters = ip4_error_counters, .n_next_nodes = IP4_FULL_REASS_N_NEXT, .next_nodes = { @@ -1316,26 +1424,26 @@ VLIB_REGISTER_NODE (ip4_full_reass_node_feature) = { }; VNET_FEATURE_INIT (ip4_full_reass_feature, static) = { - .arc_name = "ip4-unicast", - .node_name = "ip4-full-reassembly-feature", - .runs_before = VNET_FEATURES ("ip4-lookup", - "ipsec4-input-feature"), - .runs_after = 0, + .arc_name = "ip4-unicast", + .node_name = "ip4-full-reassembly-feature", + .runs_before = VNET_FEATURES ("ip4-lookup", "ipsec4-input-feature", + "ip4-sv-reassembly-feature"), + .runs_after = 0, }; VLIB_NODE_FN (ip4_full_reass_node_custom) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip4_full_reass_inline (vm, node, frame, CUSTOM); + return ip4_full_reass_inline (vm, node, frame, CUSTOM, false /* is_local */); } VLIB_REGISTER_NODE (ip4_full_reass_node_custom) = { .name = "ip4-full-reassembly-custom", .vector_size = sizeof (u32), .format_trace = format_ip4_full_reass_trace, - .n_errors = ARRAY_LEN (ip4_full_reass_error_strings), - .error_strings = ip4_full_reass_error_strings, + .n_errors = IP4_N_ERROR, + .error_counters = ip4_error_counters, .n_next_nodes = IP4_FULL_REASS_N_NEXT, .next_nodes = { @@ -1345,15 +1453,6 @@ VLIB_REGISTER_NODE (ip4_full_reass_node_custom) = { }, }; -VNET_FEATURE_INIT (ip4_full_reass_custom, static) = { - .arc_name = "ip4-unicast", - .node_name = "ip4-full-reassembly-feature", - .runs_before = VNET_FEATURES ("ip4-lookup", - "ipsec4-input-feature"), - .runs_after = 0, -}; - - #ifndef CLIB_MARCH_VARIANT uword ip4_full_reass_custom_register_next_node (uword node_index) @@ -1369,7 +1468,9 @@ ip4_full_reass_get_nbuckets () u32 nbuckets; u8 i; - nbuckets = (u32) (rm->max_reass_n / IP4_REASS_HT_LOAD_FACTOR); + /* need more mem with more workers */ + nbuckets = (u32) (rm->max_reass_n * (vlib_num_workers () + 1) / + IP4_REASS_HT_LOAD_FACTOR); for (i = 0; i < 31; i++) if ((1 << i) >= nbuckets) @@ -1495,17 +1596,17 @@ ip4_full_reass_init_function (vlib_main_t * vm) nbuckets = ip4_full_reass_get_nbuckets (); clib_bihash_init_16_8 (&rm->hash, "ip4-dr", nbuckets, nbuckets * 1024); - node = vlib_get_node_by_name (vm, (u8 *) "ip4-drop"); - ASSERT (node); - rm->ip4_drop_idx = node->index; - rm->fq_index = vlib_frame_queue_main_init (ip4_full_reass_node.index, 0); + rm->fq_local_index = + vlib_frame_queue_main_init (ip4_local_full_reass_node.index, 0); rm->fq_feature_index = vlib_frame_queue_main_init (ip4_full_reass_node_feature.index, 0); rm->fq_custom_index = vlib_frame_queue_main_init (ip4_full_reass_node_custom.index, 0); rm->feature_use_refcount_per_intf = NULL; + rm->is_local_reass_enabled = 1; + return error; } @@ -1547,6 +1648,7 @@ ip4_full_reass_walk_expired (vlib_main_t *vm, vlib_node_runtime_t *node, uword thread_index = 0; int index; const uword nthreads = vlib_num_workers () + 1; + for (thread_index = 0; thread_index < nthreads; ++thread_index) { ip4_full_reass_per_thread_t *rt = @@ -1554,13 +1656,39 @@ ip4_full_reass_walk_expired (vlib_main_t *vm, vlib_node_runtime_t *node, clib_spinlock_lock (&rt->lock); vec_reset_length (pool_indexes_to_free); - pool_foreach_index (index, rt->pool) { - reass = pool_elt_at_index (rt->pool, index); - if (now > reass->last_heard + rm->timeout) - { - vec_add1 (pool_indexes_to_free, index); - } - } + + /* Pace the number of timeouts handled per thread,to avoid barrier + * sync issues in real world scenarios */ + + u32 beg = rt->last_id; + /* to ensure we walk at least once per sec per context */ + u32 end = + beg + (IP4_REASS_MAX_REASSEMBLIES_DEFAULT * + IP4_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS / MSEC_PER_SEC + + 1); + if (end > vec_len (rt->pool)) + { + end = vec_len (rt->pool); + rt->last_id = 0; + } + else + { + rt->last_id = end; + } + + pool_foreach_stepping_index (index, beg, end, rt->pool) + { + reass = pool_elt_at_index (rt->pool, index); + if (now > reass->last_heard + rm->timeout) + { + vec_add1 (pool_indexes_to_free, index); + } + } + + if (vec_len (pool_indexes_to_free)) + vlib_node_increment_counter (vm, node->node_index, + IP4_ERROR_REASS_TIMEOUT, + vec_len (pool_indexes_to_free)); int *i; vec_foreach (i, pool_indexes_to_free) { @@ -1575,7 +1703,7 @@ ip4_full_reass_walk_expired (vlib_main_t *vm, vlib_node_runtime_t *node, vec_free (pool_indexes_to_free); if (event_data) { - _vec_len (event_data) = 0; + vec_set_len (event_data, 0); } } @@ -1583,13 +1711,12 @@ ip4_full_reass_walk_expired (vlib_main_t *vm, vlib_node_runtime_t *node, } VLIB_REGISTER_NODE (ip4_full_reass_expire_node) = { - .function = ip4_full_reass_walk_expired, - .type = VLIB_NODE_TYPE_PROCESS, - .name = "ip4-full-reassembly-expire-walk", - .format_trace = format_ip4_full_reass_trace, - .n_errors = ARRAY_LEN (ip4_full_reass_error_strings), - .error_strings = ip4_full_reass_error_strings, - + .function = ip4_full_reass_walk_expired, + .type = VLIB_NODE_TYPE_PROCESS, + .name = "ip4-full-reassembly-expire-walk", + .format_trace = format_ip4_full_reass_trace, + .n_errors = IP4_N_ERROR, + .error_counters = ip4_error_counters, }; static u8 * @@ -1597,9 +1724,8 @@ format_ip4_full_reass_key (u8 * s, va_list * args) { ip4_full_reass_key_t *key = va_arg (*args, ip4_full_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, + format (s, "fib_index: %u, src: %U, dst: %U, frag_id: %u, proto: %u", + key->fib_index, format_ip4_address, &key->src, format_ip4_address, &key->dst, clib_net_to_host_u16 (key->frag_id), key->proto); return s; } @@ -1750,10 +1876,10 @@ format_ip4_full_reass_handoff_trace (u8 * s, va_list * args) } always_inline uword -ip4_full_reass_handoff_node_inline (vlib_main_t * vm, - vlib_node_runtime_t * node, - vlib_frame_t * frame, - ip4_full_reass_node_type_t type) +ip4_full_reass_handoff_node_inline (vlib_main_t *vm, vlib_node_runtime_t *node, + vlib_frame_t *frame, + ip4_full_reass_node_type_t type, + bool is_local) { ip4_full_reass_main_t *rm = &ip4_full_reass_main; @@ -1772,7 +1898,14 @@ ip4_full_reass_handoff_node_inline (vlib_main_t * vm, switch (type) { case NORMAL: - fq_index = rm->fq_index; + if (is_local) + { + fq_index = rm->fq_local_index; + } + else + { + fq_index = rm->fq_index; + } break; case FEATURE: fq_index = rm->fq_feature_index; @@ -1782,7 +1915,6 @@ ip4_full_reass_handoff_node_inline (vlib_main_t * vm, break; default: clib_warning ("Unexpected `type' (%d)!", type); - ASSERT (0); } while (n_left_from > 0) @@ -1816,7 +1948,8 @@ VLIB_NODE_FN (ip4_full_reass_handoff_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip4_full_reass_handoff_node_inline (vm, node, frame, NORMAL); + return ip4_full_reass_handoff_node_inline (vm, node, frame, NORMAL, + false /* is_local */); } @@ -1834,16 +1967,36 @@ VLIB_REGISTER_NODE (ip4_full_reass_handoff_node) = { }, }; +VLIB_NODE_FN (ip4_local_full_reass_handoff_node) +(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame) +{ + return ip4_full_reass_handoff_node_inline (vm, node, frame, NORMAL, + true /* is_local */); +} + +VLIB_REGISTER_NODE (ip4_local_full_reass_handoff_node) = { + .name = "ip4-local-full-reassembly-handoff", + .vector_size = sizeof (u32), + .n_errors = ARRAY_LEN(ip4_full_reass_handoff_error_strings), + .error_strings = ip4_full_reass_handoff_error_strings, + .format_trace = format_ip4_full_reass_handoff_trace, + + .n_next_nodes = 1, + + .next_nodes = { + [0] = "error-drop", + }, +}; VLIB_NODE_FN (ip4_full_reass_feature_handoff_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip4_full_reass_handoff_node_inline (vm, node, frame, FEATURE); + return ip4_full_reass_handoff_node_inline (vm, node, frame, FEATURE, + false /* is_local */); } - VLIB_REGISTER_NODE (ip4_full_reass_feature_handoff_node) = { .name = "ip4-full-reass-feature-hoff", .vector_size = sizeof (u32), @@ -1863,10 +2016,10 @@ VLIB_NODE_FN (ip4_full_reass_custom_handoff_node) (vlib_main_t * vm, node, vlib_frame_t * frame) { - return ip4_full_reass_handoff_node_inline (vm, node, frame, CUSTOM); + return ip4_full_reass_handoff_node_inline (vm, node, frame, CUSTOM, + false /* is_local */); } - VLIB_REGISTER_NODE (ip4_full_reass_custom_handoff_node) = { .name = "ip4-full-reass-custom-hoff", .vector_size = sizeof (u32), @@ -1906,8 +2059,28 @@ ip4_full_reass_enable_disable_with_refcnt (u32 sw_if_index, int is_enable) "ip4-full-reassembly-feature", sw_if_index, 0, 0, 0); } - return -1; + return 0; } + +void +ip4_local_full_reass_enable_disable (int enable) +{ + if (enable) + { + ip4_full_reass_main.is_local_reass_enabled = 1; + } + else + { + ip4_full_reass_main.is_local_reass_enabled = 0; + } +} + +int +ip4_local_full_reass_enabled () +{ + return ip4_full_reass_main.is_local_reass_enabled; +} + #endif /* diff --git a/src/vnet/ip/reass/ip4_full_reass.h b/src/vnet/ip/reass/ip4_full_reass.h index 000c80c5906..5df8107ca48 100644 --- a/src/vnet/ip/reass/ip4_full_reass.h +++ b/src/vnet/ip/reass/ip4_full_reass.h @@ -47,6 +47,9 @@ int ip4_full_reass_enable_disable_with_refcnt (u32 sw_if_index, int is_enable); uword ip4_full_reass_custom_register_next_node (uword node_index); + +void ip4_local_full_reass_enable_disable (int enable); +int ip4_local_full_reass_enabled (); #endif /* __included_ip4_full_reass_h__ */ /* diff --git a/src/vnet/ip/reass/ip4_sv_reass.c b/src/vnet/ip/reass/ip4_sv_reass.c index cd5e19b65d3..7c3c2fff217 100644 --- a/src/vnet/ip/reass/ip4_sv_reass.c +++ b/src/vnet/ip/reass/ip4_sv_reass.c @@ -48,7 +48,7 @@ typedef struct { struct { - u32 xx_id; + u32 fib_index; ip4_address_t src; ip4_address_t dst; u16 frag_id; @@ -150,6 +150,7 @@ typedef struct /** Worker handoff */ u32 fq_index; u32 fq_feature_index; + u32 fq_custom_context_index; // reference count for enabling/disabling feature - per interface u32 *feature_use_refcount_per_intf; @@ -189,6 +190,7 @@ typedef struct u8 ip_proto; u16 l4_src_port; u16 l4_dst_port; + int l4_layer_truncated; } ip4_sv_reass_trace_t; extern vlib_node_registration_t ip4_sv_reass_node; @@ -225,6 +227,10 @@ format_ip4_sv_reass_trace (u8 * s, va_list * args) s = format (s, "[not-fragmented]"); break; } + if (t->l4_layer_truncated) + { + s = format (s, " [l4-layer-truncated]"); + } return s; } @@ -232,7 +238,8 @@ static void ip4_sv_reass_add_trace (vlib_main_t *vm, vlib_node_runtime_t *node, ip4_sv_reass_t *reass, u32 bi, ip4_sv_reass_trace_operation_e action, u32 ip_proto, - u16 l4_src_port, u16 l4_dst_port) + u16 l4_src_port, u16 l4_dst_port, + int l4_layer_truncated) { vlib_buffer_t *b = vlib_get_buffer (vm, bi); if (pool_is_free_index @@ -253,6 +260,7 @@ ip4_sv_reass_add_trace (vlib_main_t *vm, vlib_node_runtime_t *node, t->ip_proto = ip_proto; t->l4_src_port = l4_src_port; t->l4_dst_port = l4_dst_port; + t->l4_layer_truncated = l4_layer_truncated; #if 0 static u8 *s = NULL; s = format (s, "%U", format_ip4_sv_reass_trace, NULL, NULL, t); @@ -314,6 +322,8 @@ ip4_sv_reass_find_or_create (vlib_main_t * vm, ip4_sv_reass_main_t * rm, ip4_sv_reass_t *reass = NULL; f64 now = vlib_time_now (vm); +again: + if (!clib_bihash_search_16_8 (&rm->hash, &kv->kv, &kv->kv)) { if (vm->thread_index != kv->v.thread_index) @@ -368,10 +378,14 @@ ip4_sv_reass_find_or_create (vlib_main_t * vm, ip4_sv_reass_main_t * rm, kv->v.thread_index = vm->thread_index; reass->last_heard = now; - if (clib_bihash_add_del_16_8 (&rm->hash, &kv->kv, 1)) + int rv = clib_bihash_add_del_16_8 (&rm->hash, &kv->kv, 2); + if (rv) { ip4_sv_reass_free (vm, rm, rt, reass); reass = NULL; + // if other worker created a context already work with the other copy + if (-2 == rv) + goto again; } return reass; @@ -407,9 +421,10 @@ ip4_sv_reass_update (vlib_main_t *vm, vlib_node_runtime_t *node, vlib_buffer_t *b0 = vlib_get_buffer (vm, bi0); if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) { - ip4_sv_reass_add_trace (vm, node, reass, bi0, REASS_FINISH, - reass->ip_proto, reass->l4_src_port, - reass->l4_dst_port); + ip4_sv_reass_add_trace ( + vm, node, reass, bi0, REASS_FINISH, reass->ip_proto, + reass->l4_src_port, reass->l4_dst_port, + vnet_buffer (b0)->ip.reass.l4_layer_truncated); } } vec_add1 (reass->cached_buffers, bi0); @@ -417,8 +432,9 @@ ip4_sv_reass_update (vlib_main_t *vm, vlib_node_runtime_t *node, { if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) { - ip4_sv_reass_add_trace (vm, node, reass, bi0, REASS_FRAGMENT_CACHE, - ~0, ~0, ~0); + ip4_sv_reass_add_trace ( + vm, node, reass, bi0, REASS_FRAGMENT_CACHE, ~0, ~0, ~0, + vnet_buffer (b0)->ip.reass.l4_layer_truncated); } if (vec_len (reass->cached_buffers) > rm->max_reass_len) { @@ -428,15 +444,33 @@ ip4_sv_reass_update (vlib_main_t *vm, vlib_node_runtime_t *node, return rc; } +always_inline int +l4_layer_truncated (ip4_header_t *ip) +{ + static const int l4_layer_length[256] = { + [IP_PROTOCOL_TCP] = sizeof (tcp_header_t), + [IP_PROTOCOL_UDP] = sizeof (udp_header_t), + [IP_PROTOCOL_ICMP] = sizeof (icmp46_header_t), + }; + + return ((u8 *) ip + ip4_header_bytes (ip) + l4_layer_length[ip->protocol] > + (u8 *) ip + clib_net_to_host_u16 (ip->length)); +} + always_inline uword -ip4_sv_reass_inline (vlib_main_t * vm, vlib_node_runtime_t * node, - vlib_frame_t * frame, bool is_feature, - bool is_output_feature, bool is_custom) +ip4_sv_reass_inline (vlib_main_t *vm, vlib_node_runtime_t *node, + vlib_frame_t *frame, bool is_feature, + bool is_output_feature, bool is_custom, + bool with_custom_context) { u32 *from = vlib_frame_vector_args (frame); - u32 n_left_from, n_left_to_next, *to_next, next_index; + u32 n_left_from, n_left_to_next, *to_next, *to_next_aux, next_index; ip4_sv_reass_main_t *rm = &ip4_sv_reass_main; ip4_sv_reass_per_thread_t *rt = &rm->per_thread_data[vm->thread_index]; + u32 *context; + if (with_custom_context) + context = vlib_frame_aux_args (frame); + clib_spinlock_lock (&rt->lock); n_left_from = frame->n_vectors; @@ -482,6 +516,7 @@ ip4_sv_reass_inline (vlib_main_t * vm, vlib_node_runtime_t * node, (is_output_feature ? 1 : 0) * vnet_buffer (b1)-> ip.save_rewrite_length); + if (PREDICT_FALSE (ip4_get_fragment_more (ip0) || ip4_get_fragment_offset (ip0)) || (ip4_get_fragment_more (ip1) || ip4_get_fragment_offset (ip1))) @@ -506,29 +541,40 @@ ip4_sv_reass_inline (vlib_main_t * vm, vlib_node_runtime_t * node, } vnet_buffer (b0)->ip.reass.is_non_first_fragment = 0; vnet_buffer (b0)->ip.reass.ip_proto = ip0->protocol; - if (IP_PROTOCOL_TCP == ip0->protocol) + if (l4_layer_truncated (ip0)) { - vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags = - ((tcp_header_t *) (ip0 + 1))->flags; - vnet_buffer (b0)->ip.reass.tcp_ack_number = - ((tcp_header_t *) (ip0 + 1))->ack_number; - vnet_buffer (b0)->ip.reass.tcp_seq_number = - ((tcp_header_t *) (ip0 + 1))->seq_number; + vnet_buffer (b0)->ip.reass.l4_layer_truncated = 1; + vnet_buffer (b0)->ip.reass.l4_src_port = 0; + vnet_buffer (b0)->ip.reass.l4_dst_port = 0; } - else if (IP_PROTOCOL_ICMP == ip0->protocol) + else { - vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags = - ((icmp46_header_t *) (ip0 + 1))->type; + vnet_buffer (b0)->ip.reass.l4_layer_truncated = 0; + if (IP_PROTOCOL_TCP == ip0->protocol) + { + vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags = + ((tcp_header_t *) (ip0 + 1))->flags; + vnet_buffer (b0)->ip.reass.tcp_ack_number = + ((tcp_header_t *) (ip0 + 1))->ack_number; + vnet_buffer (b0)->ip.reass.tcp_seq_number = + ((tcp_header_t *) (ip0 + 1))->seq_number; + } + else if (IP_PROTOCOL_ICMP == ip0->protocol) + { + vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags = + ((icmp46_header_t *) (ip0 + 1))->type; + } + vnet_buffer (b0)->ip.reass.l4_src_port = ip4_get_port (ip0, 1); + vnet_buffer (b0)->ip.reass.l4_dst_port = ip4_get_port (ip0, 0); } - vnet_buffer (b0)->ip.reass.l4_src_port = ip4_get_port (ip0, 1); - vnet_buffer (b0)->ip.reass.l4_dst_port = ip4_get_port (ip0, 0); if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) { - ip4_sv_reass_add_trace (vm, node, NULL, from[(b - 2) - bufs], - REASS_PASSTHROUGH, - vnet_buffer (b0)->ip.reass.ip_proto, - vnet_buffer (b0)->ip.reass.l4_src_port, - vnet_buffer (b0)->ip.reass.l4_dst_port); + ip4_sv_reass_add_trace ( + vm, node, NULL, from[(b - 2) - bufs], REASS_PASSTHROUGH, + vnet_buffer (b0)->ip.reass.ip_proto, + vnet_buffer (b0)->ip.reass.l4_src_port, + vnet_buffer (b0)->ip.reass.l4_dst_port, + vnet_buffer (b0)->ip.reass.l4_layer_truncated); } if (is_feature) { @@ -541,35 +587,48 @@ ip4_sv_reass_inline (vlib_main_t * vm, vlib_node_runtime_t * node, } vnet_buffer (b1)->ip.reass.is_non_first_fragment = 0; vnet_buffer (b1)->ip.reass.ip_proto = ip1->protocol; - if (IP_PROTOCOL_TCP == ip1->protocol) + if (l4_layer_truncated (ip1)) { - vnet_buffer (b1)->ip.reass.icmp_type_or_tcp_flags = - ((tcp_header_t *) (ip1 + 1))->flags; - vnet_buffer (b1)->ip.reass.tcp_ack_number = - ((tcp_header_t *) (ip1 + 1))->ack_number; - vnet_buffer (b1)->ip.reass.tcp_seq_number = - ((tcp_header_t *) (ip1 + 1))->seq_number; + vnet_buffer (b1)->ip.reass.l4_layer_truncated = 1; + vnet_buffer (b1)->ip.reass.l4_src_port = 0; + vnet_buffer (b1)->ip.reass.l4_dst_port = 0; } - else if (IP_PROTOCOL_ICMP == ip1->protocol) + else { - vnet_buffer (b1)->ip.reass.icmp_type_or_tcp_flags = - ((icmp46_header_t *) (ip1 + 1))->type; + vnet_buffer (b1)->ip.reass.l4_layer_truncated = 0; + if (IP_PROTOCOL_TCP == ip1->protocol) + { + vnet_buffer (b1)->ip.reass.icmp_type_or_tcp_flags = + ((tcp_header_t *) (ip1 + 1))->flags; + vnet_buffer (b1)->ip.reass.tcp_ack_number = + ((tcp_header_t *) (ip1 + 1))->ack_number; + vnet_buffer (b1)->ip.reass.tcp_seq_number = + ((tcp_header_t *) (ip1 + 1))->seq_number; + } + else if (IP_PROTOCOL_ICMP == ip1->protocol) + { + vnet_buffer (b1)->ip.reass.icmp_type_or_tcp_flags = + ((icmp46_header_t *) (ip1 + 1))->type; + } + vnet_buffer (b1)->ip.reass.l4_src_port = ip4_get_port (ip1, 1); + vnet_buffer (b1)->ip.reass.l4_dst_port = ip4_get_port (ip1, 0); } - vnet_buffer (b1)->ip.reass.l4_src_port = ip4_get_port (ip1, 1); - vnet_buffer (b1)->ip.reass.l4_dst_port = ip4_get_port (ip1, 0); if (PREDICT_FALSE (b1->flags & VLIB_BUFFER_IS_TRACED)) { - ip4_sv_reass_add_trace (vm, node, NULL, from[(b - 1) - bufs], - REASS_PASSTHROUGH, - vnet_buffer (b1)->ip.reass.ip_proto, - vnet_buffer (b1)->ip.reass.l4_src_port, - vnet_buffer (b1)->ip.reass.l4_dst_port); + ip4_sv_reass_add_trace ( + vm, node, NULL, from[(b - 1) - bufs], REASS_PASSTHROUGH, + vnet_buffer (b1)->ip.reass.ip_proto, + vnet_buffer (b1)->ip.reass.l4_src_port, + vnet_buffer (b1)->ip.reass.l4_dst_port, + vnet_buffer (b1)->ip.reass.l4_layer_truncated); } n_left_from -= 2; next[0] = next0; next[1] = next1; next += 2; + if (with_custom_context) + context += 2; } while (n_left_from > 0) @@ -608,34 +667,45 @@ ip4_sv_reass_inline (vlib_main_t * vm, vlib_node_runtime_t * node, } vnet_buffer (b0)->ip.reass.is_non_first_fragment = 0; vnet_buffer (b0)->ip.reass.ip_proto = ip0->protocol; - if (IP_PROTOCOL_TCP == ip0->protocol) + if (l4_layer_truncated (ip0)) { - vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags = - ((tcp_header_t *) (ip0 + 1))->flags; - vnet_buffer (b0)->ip.reass.tcp_ack_number = - ((tcp_header_t *) (ip0 + 1))->ack_number; - vnet_buffer (b0)->ip.reass.tcp_seq_number = - ((tcp_header_t *) (ip0 + 1))->seq_number; + vnet_buffer (b0)->ip.reass.l4_layer_truncated = 1; } - else if (IP_PROTOCOL_ICMP == ip0->protocol) + else { - vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags = - ((icmp46_header_t *) (ip0 + 1))->type; + vnet_buffer (b0)->ip.reass.l4_layer_truncated = 0; + if (IP_PROTOCOL_TCP == ip0->protocol) + { + vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags = + ((tcp_header_t *) (ip0 + 1))->flags; + vnet_buffer (b0)->ip.reass.tcp_ack_number = + ((tcp_header_t *) (ip0 + 1))->ack_number; + vnet_buffer (b0)->ip.reass.tcp_seq_number = + ((tcp_header_t *) (ip0 + 1))->seq_number; + } + else if (IP_PROTOCOL_ICMP == ip0->protocol) + { + vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags = + ((icmp46_header_t *) (ip0 + 1))->type; + } + vnet_buffer (b0)->ip.reass.l4_src_port = ip4_get_port (ip0, 1); + vnet_buffer (b0)->ip.reass.l4_dst_port = ip4_get_port (ip0, 0); } - vnet_buffer (b0)->ip.reass.l4_src_port = ip4_get_port (ip0, 1); - vnet_buffer (b0)->ip.reass.l4_dst_port = ip4_get_port (ip0, 0); if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) { - ip4_sv_reass_add_trace (vm, node, NULL, from[(b - 1) - bufs], - REASS_PASSTHROUGH, - vnet_buffer (b0)->ip.reass.ip_proto, - vnet_buffer (b0)->ip.reass.l4_src_port, - vnet_buffer (b0)->ip.reass.l4_dst_port); + ip4_sv_reass_add_trace ( + vm, node, NULL, from[(b - 1) - bufs], REASS_PASSTHROUGH, + vnet_buffer (b0)->ip.reass.ip_proto, + vnet_buffer (b0)->ip.reass.l4_src_port, + vnet_buffer (b0)->ip.reass.l4_dst_port, + vnet_buffer (b0)->ip.reass.l4_layer_truncated); } n_left_from -= 1; next[0] = next0; next += 1; + if (with_custom_context) + context += 1; } vlib_buffer_enqueue_to_next (vm, node, from, (u16 *) nexts, @@ -649,7 +719,11 @@ slow_path: while (n_left_from > 0) { - vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + if (with_custom_context) + vlib_get_next_frame_with_aux_safe (vm, node, next_index, to_next, + to_next_aux, n_left_to_next); + else + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); while (n_left_from > 0 && n_left_to_next > 0) { @@ -657,6 +731,7 @@ slow_path: vlib_buffer_t *b0; u32 next0; u32 error0 = IP4_ERROR_NONE; + u8 forward_context = 0; bi0 = from[0]; b0 = vlib_get_buffer (vm, bi0); @@ -679,29 +754,42 @@ slow_path: } vnet_buffer (b0)->ip.reass.is_non_first_fragment = 0; vnet_buffer (b0)->ip.reass.ip_proto = ip0->protocol; - if (IP_PROTOCOL_TCP == ip0->protocol) + if (l4_layer_truncated (ip0)) { - vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags = - ((tcp_header_t *) (ip0 + 1))->flags; - vnet_buffer (b0)->ip.reass.tcp_ack_number = - ((tcp_header_t *) (ip0 + 1))->ack_number; - vnet_buffer (b0)->ip.reass.tcp_seq_number = - ((tcp_header_t *) (ip0 + 1))->seq_number; + vnet_buffer (b0)->ip.reass.l4_layer_truncated = 1; + vnet_buffer (b0)->ip.reass.l4_src_port = 0; + vnet_buffer (b0)->ip.reass.l4_dst_port = 0; } - else if (IP_PROTOCOL_ICMP == ip0->protocol) + else { - vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags = - ((icmp46_header_t *) (ip0 + 1))->type; + vnet_buffer (b0)->ip.reass.l4_layer_truncated = 0; + if (IP_PROTOCOL_TCP == ip0->protocol) + { + vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags = + ((tcp_header_t *) (ip0 + 1))->flags; + vnet_buffer (b0)->ip.reass.tcp_ack_number = + ((tcp_header_t *) (ip0 + 1))->ack_number; + vnet_buffer (b0)->ip.reass.tcp_seq_number = + ((tcp_header_t *) (ip0 + 1))->seq_number; + } + else if (IP_PROTOCOL_ICMP == ip0->protocol) + { + vnet_buffer (b0)->ip.reass.icmp_type_or_tcp_flags = + ((icmp46_header_t *) (ip0 + 1))->type; + } + vnet_buffer (b0)->ip.reass.l4_src_port = + ip4_get_port (ip0, 1); + vnet_buffer (b0)->ip.reass.l4_dst_port = + ip4_get_port (ip0, 0); } - vnet_buffer (b0)->ip.reass.l4_src_port = ip4_get_port (ip0, 1); - vnet_buffer (b0)->ip.reass.l4_dst_port = ip4_get_port (ip0, 0); if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) { ip4_sv_reass_add_trace ( vm, node, NULL, bi0, REASS_PASSTHROUGH, vnet_buffer (b0)->ip.reass.ip_proto, vnet_buffer (b0)->ip.reass.l4_src_port, - vnet_buffer (b0)->ip.reass.l4_dst_port); + vnet_buffer (b0)->ip.reass.l4_dst_port, + vnet_buffer (b0)->ip.reass.l4_layer_truncated); } goto packet_enqueue; } @@ -719,13 +807,17 @@ slow_path: ip4_sv_reass_kv_t kv; u8 do_handoff = 0; - kv.k.as_u64[0] = - (u64) vec_elt (ip4_main.fib_index_by_sw_if_index, - vnet_buffer (b0)->sw_if_index[VLIB_RX]) | - (u64) ip0->src_address.as_u32 << 32; - kv.k.as_u64[1] = - (u64) ip0->dst_address. - as_u32 | (u64) ip0->fragment_id << 32 | (u64) ip0->protocol << 48; + if (with_custom_context) + kv.k.as_u64[0] = (u64) *context | (u64) ip0->src_address.as_u32 + << 32; + else + kv.k.as_u64[0] = + (u64) vec_elt (ip4_main.fib_index_by_sw_if_index, + vnet_buffer (b0)->sw_if_index[VLIB_RX]) | + (u64) ip0->src_address.as_u32 << 32; + kv.k.as_u64[1] = (u64) ip0->dst_address.as_u32 | + (u64) ip0->fragment_id << 32 | + (u64) ip0->protocol << 48; ip4_sv_reass_t *reass = ip4_sv_reass_find_or_create (vm, rm, rt, &kv, &do_handoff); @@ -735,6 +827,8 @@ slow_path: next0 = IP4_SV_REASSEMBLY_NEXT_HANDOFF; vnet_buffer (b0)->ip.reass.owner_thread_index = kv.v.thread_index; + if (with_custom_context) + forward_context = 1; goto packet_enqueue; } @@ -771,31 +865,32 @@ slow_path: { ip4_sv_reass_add_trace ( vm, node, reass, bi0, REASS_FRAGMENT_FORWARD, - reass->ip_proto, reass->l4_src_port, reass->l4_dst_port); + reass->ip_proto, reass->l4_src_port, reass->l4_dst_port, + vnet_buffer (b0)->ip.reass.l4_layer_truncated); } goto packet_enqueue; } ip4_sv_reass_rc_t rc = ip4_sv_reass_update (vm, node, rm, ip0, reass, bi0); + u32 counter = ~0; switch (rc) { case IP4_SV_REASS_RC_OK: /* nothing to do here */ break; case IP4_SV_REASS_RC_TOO_MANY_FRAGMENTS: - vlib_node_increment_counter (vm, node->node_index, - IP4_ERROR_REASS_FRAGMENT_CHAIN_TOO_LONG, - 1); - ip4_sv_reass_free (vm, rm, rt, reass); - goto next_packet; + counter = IP4_ERROR_REASS_FRAGMENT_CHAIN_TOO_LONG; break; case IP4_SV_REASS_RC_UNSUPP_IP_PROTO: - vlib_node_increment_counter (vm, node->node_index, - IP4_ERROR_REASS_UNSUPP_IP_PROT, 1); + counter = IP4_ERROR_REASS_UNSUPP_IP_PROT; + break; + } + if (~0 != counter) + { + vlib_node_increment_counter (vm, node->node_index, counter, 1); ip4_sv_reass_free (vm, rm, rt, reass); goto next_packet; - break; } if (reass->is_complete) { @@ -843,13 +938,15 @@ slow_path: { ip4_sv_reass_add_trace ( vm, node, reass, bi0, REASS_FRAGMENT_FORWARD, - reass->ip_proto, reass->l4_src_port, reass->l4_dst_port); + reass->ip_proto, reass->l4_src_port, reass->l4_dst_port, + vnet_buffer (b0)->ip.reass.l4_layer_truncated); } vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, n_left_to_next, bi0, next0); } - _vec_len (reass->cached_buffers) = 0; // buffers are owned by frame now + vec_set_len (reass->cached_buffers, + 0); // buffers are owned by frame now } goto next_packet; @@ -862,13 +959,26 @@ slow_path: b0 = vlib_get_buffer (vm, bi0); vnet_feature_next (&next0, b0); } - vlib_validate_buffer_enqueue_x1 (vm, node, next_index, - to_next, n_left_to_next, - bi0, next0); + if (with_custom_context && forward_context) + { + if (to_next_aux) + { + to_next_aux[0] = *context; + to_next_aux += 1; + } + vlib_validate_buffer_enqueue_with_aux_x1 ( + vm, node, next_index, to_next, to_next_aux, n_left_to_next, + bi0, *context, next0); + } + else + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); next_packet: from += 1; n_left_from -= 1; + if (with_custom_context) + context += 1; } vlib_put_next_frame (vm, node, next_index, n_left_to_next); @@ -879,28 +989,21 @@ done: return frame->n_vectors; } -static char *ip4_sv_reass_error_strings[] = { -#define _(sym, string) string, - foreach_ip4_error -#undef _ -}; - VLIB_NODE_FN (ip4_sv_reass_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip4_sv_reass_inline (vm, node, frame, false /* is_feature */ , - false /* is_output_feature */ , - false /* is_custom */ ); + return ip4_sv_reass_inline ( + vm, node, frame, false /* is_feature */, false /* is_output_feature */, + false /* is_custom */, false /* with_custom_context */); } -/* *INDENT-OFF* */ VLIB_REGISTER_NODE (ip4_sv_reass_node) = { .name = "ip4-sv-reassembly", .vector_size = sizeof (u32), .format_trace = format_ip4_sv_reass_trace, - .n_errors = ARRAY_LEN (ip4_sv_reass_error_strings), - .error_strings = ip4_sv_reass_error_strings, + .n_errors = IP4_N_ERROR, + .error_counters = ip4_error_counters, .n_next_nodes = IP4_SV_REASSEMBLY_N_NEXT, .next_nodes = { @@ -910,24 +1013,22 @@ VLIB_REGISTER_NODE (ip4_sv_reass_node) = { }, }; -/* *INDENT-ON* */ VLIB_NODE_FN (ip4_sv_reass_node_feature) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip4_sv_reass_inline (vm, node, frame, true /* is_feature */ , - false /* is_output_feature */ , - false /* is_custom */ ); + return ip4_sv_reass_inline ( + vm, node, frame, true /* is_feature */, false /* is_output_feature */, + false /* is_custom */, false /* with_custom_context */); } -/* *INDENT-OFF* */ VLIB_REGISTER_NODE (ip4_sv_reass_node_feature) = { .name = "ip4-sv-reassembly-feature", .vector_size = sizeof (u32), .format_trace = format_ip4_sv_reass_trace, - .n_errors = ARRAY_LEN (ip4_sv_reass_error_strings), - .error_strings = ip4_sv_reass_error_strings, + .n_errors = IP4_N_ERROR, + .error_counters = ip4_error_counters, .n_next_nodes = IP4_SV_REASSEMBLY_N_NEXT, .next_nodes = { @@ -936,34 +1037,30 @@ VLIB_REGISTER_NODE (ip4_sv_reass_node_feature) = { [IP4_SV_REASSEMBLY_NEXT_HANDOFF] = "ip4-sv-reass-feature-hoff", }, }; -/* *INDENT-ON* */ -/* *INDENT-OFF* */ VNET_FEATURE_INIT (ip4_sv_reass_feature) = { .arc_name = "ip4-unicast", .node_name = "ip4-sv-reassembly-feature", .runs_before = VNET_FEATURES ("ip4-lookup"), .runs_after = 0, }; -/* *INDENT-ON* */ VLIB_NODE_FN (ip4_sv_reass_node_output_feature) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip4_sv_reass_inline (vm, node, frame, true /* is_feature */ , - true /* is_output_feature */ , - false /* is_custom */ ); + return ip4_sv_reass_inline ( + vm, node, frame, true /* is_feature */, true /* is_output_feature */, + false /* is_custom */, false /* with_custom_context */); } -/* *INDENT-OFF* */ VLIB_REGISTER_NODE (ip4_sv_reass_node_output_feature) = { .name = "ip4-sv-reassembly-output-feature", .vector_size = sizeof (u32), .format_trace = format_ip4_sv_reass_trace, - .n_errors = ARRAY_LEN (ip4_sv_reass_error_strings), - .error_strings = ip4_sv_reass_error_strings, + .n_errors = IP4_N_ERROR, + .error_counters = ip4_error_counters, .n_next_nodes = IP4_SV_REASSEMBLY_N_NEXT, .next_nodes = { @@ -972,24 +1069,20 @@ VLIB_REGISTER_NODE (ip4_sv_reass_node_output_feature) = { [IP4_SV_REASSEMBLY_NEXT_HANDOFF] = "ip4-sv-reass-feature-hoff", }, }; -/* *INDENT-ON* */ -/* *INDENT-OFF* */ VNET_FEATURE_INIT (ip4_sv_reass_output_feature) = { .arc_name = "ip4-output", .node_name = "ip4-sv-reassembly-output-feature", .runs_before = 0, .runs_after = 0, }; -/* *INDENT-ON* */ -/* *INDENT-OFF* */ VLIB_REGISTER_NODE (ip4_sv_reass_custom_node) = { .name = "ip4-sv-reassembly-custom-next", .vector_size = sizeof (u32), .format_trace = format_ip4_sv_reass_trace, - .n_errors = ARRAY_LEN (ip4_sv_reass_error_strings), - .error_strings = ip4_sv_reass_error_strings, + .n_errors = IP4_N_ERROR, + .error_counters = ip4_error_counters, .n_next_nodes = IP4_SV_REASSEMBLY_N_NEXT, .next_nodes = { @@ -999,15 +1092,39 @@ VLIB_REGISTER_NODE (ip4_sv_reass_custom_node) = { }, }; -/* *INDENT-ON* */ VLIB_NODE_FN (ip4_sv_reass_custom_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip4_sv_reass_inline (vm, node, frame, false /* is_feature */ , - false /* is_output_feature */ , - true /* is_custom */ ); + return ip4_sv_reass_inline ( + vm, node, frame, false /* is_feature */, false /* is_output_feature */, + true /* is_custom */, false /* with_custom_context */); +} + +VLIB_REGISTER_NODE (ip4_sv_reass_custom_context_node) = { + .name = "ip4-sv-reassembly-custom-context", + .vector_size = sizeof (u32), + .aux_size = sizeof(u32), + .format_trace = format_ip4_sv_reass_trace, + .n_errors = IP4_N_ERROR, + .error_counters = ip4_error_counters, + .n_next_nodes = IP4_SV_REASSEMBLY_N_NEXT, + .next_nodes = + { + [IP4_SV_REASSEMBLY_NEXT_INPUT] = "ip4-input", + [IP4_SV_REASSEMBLY_NEXT_DROP] = "ip4-drop", + [IP4_SV_REASSEMBLY_NEXT_HANDOFF] = "ip4-sv-reassembly-custom-context-handoff", + + }, +}; + +VLIB_NODE_FN (ip4_sv_reass_custom_context_node) +(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame) +{ + return ip4_sv_reass_inline ( + vm, node, frame, false /* is_feature */, false /* is_output_feature */, + true /* is_custom */, true /* with_custom_context */); } #ifndef CLIB_MARCH_VARIANT @@ -1152,6 +1269,8 @@ ip4_sv_reass_init_function (vlib_main_t * vm) rm->fq_index = vlib_frame_queue_main_init (ip4_sv_reass_node.index, 0); rm->fq_feature_index = vlib_frame_queue_main_init (ip4_sv_reass_node_feature.index, 0); + rm->fq_custom_context_index = + vlib_frame_queue_main_init (ip4_sv_reass_custom_context_node.index, 0); rm->feature_use_refcount_per_intf = NULL; rm->output_feature_use_refcount_per_intf = NULL; @@ -1204,7 +1323,6 @@ ip4_sv_reass_walk_expired (vlib_main_t *vm, clib_spinlock_lock (&rt->lock); vec_reset_length (pool_indexes_to_free); - /* *INDENT-OFF* */ pool_foreach_index (index, rt->pool) { reass = pool_elt_at_index (rt->pool, index); if (now > reass->last_heard + rm->timeout) @@ -1212,15 +1330,12 @@ ip4_sv_reass_walk_expired (vlib_main_t *vm, vec_add1 (pool_indexes_to_free, index); } } - /* *INDENT-ON* */ int *i; - /* *INDENT-OFF* */ vec_foreach (i, pool_indexes_to_free) { ip4_sv_reass_t *reass = pool_elt_at_index (rt->pool, i[0]); ip4_sv_reass_free (vm, rm, rt, reass); } - /* *INDENT-ON* */ clib_spinlock_unlock (&rt->lock); } @@ -1228,33 +1343,29 @@ ip4_sv_reass_walk_expired (vlib_main_t *vm, vec_free (pool_indexes_to_free); if (event_data) { - _vec_len (event_data) = 0; + vec_set_len (event_data, 0); } } return 0; } -/* *INDENT-OFF* */ VLIB_REGISTER_NODE (ip4_sv_reass_expire_node) = { - .function = ip4_sv_reass_walk_expired, - .type = VLIB_NODE_TYPE_PROCESS, - .name = "ip4-sv-reassembly-expire-walk", - .format_trace = format_ip4_sv_reass_trace, - .n_errors = ARRAY_LEN (ip4_sv_reass_error_strings), - .error_strings = ip4_sv_reass_error_strings, - + .function = ip4_sv_reass_walk_expired, + .type = VLIB_NODE_TYPE_PROCESS, + .name = "ip4-sv-reassembly-expire-walk", + .format_trace = format_ip4_sv_reass_trace, + .n_errors = IP4_N_ERROR, + .error_counters = ip4_error_counters, }; -/* *INDENT-ON* */ static u8 * format_ip4_sv_reass_key (u8 * s, va_list * args) { ip4_sv_reass_key_t *key = va_arg (*args, ip4_sv_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, + format (s, "fib_index: %u, src: %U, dst: %U, frag_id: %u, proto: %u", + key->fib_index, format_ip4_address, &key->src, format_ip4_address, &key->dst, clib_net_to_host_u16 (key->frag_id), key->proto); return s; } @@ -1313,11 +1424,9 @@ show_ip4_reass (vlib_main_t * vm, clib_spinlock_lock (&rt->lock); if (details) { - /* *INDENT-OFF* */ pool_foreach (reass, rt->pool) { vlib_cli_output (vm, "%U", format_ip4_sv_reass, vm, reass); } - /* *INDENT-ON* */ } sum_reass_n += rt->reass_n; clib_spinlock_unlock (&rt->lock); @@ -1341,13 +1450,11 @@ show_ip4_reass (vlib_main_t * vm, return 0; } -/* *INDENT-OFF* */ VLIB_CLI_COMMAND (show_ip4_sv_reass_cmd, static) = { .path = "show ip4-sv-reassembly", .short_help = "show ip4-sv-reassembly [details]", .function = show_ip4_reass, }; -/* *INDENT-ON* */ #ifndef CLIB_MARCH_VARIANT vnet_api_error_t @@ -1398,25 +1505,30 @@ format_ip4_sv_reass_handoff_trace (u8 * s, va_list * args) } always_inline uword -ip4_sv_reass_handoff_node_inline (vlib_main_t * vm, - vlib_node_runtime_t * node, - vlib_frame_t * frame, bool is_feature) +ip4_sv_reass_handoff_node_inline (vlib_main_t *vm, vlib_node_runtime_t *node, + vlib_frame_t *frame, bool is_feature, + bool is_custom_context) { ip4_sv_reass_main_t *rm = &ip4_sv_reass_main; vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b; - u32 n_enq, n_left_from, *from; + u32 n_enq, n_left_from, *from, *context; u16 thread_indices[VLIB_FRAME_SIZE], *ti; u32 fq_index; from = vlib_frame_vector_args (frame); + if (is_custom_context) + context = vlib_frame_aux_args (frame); + n_left_from = frame->n_vectors; vlib_get_buffers (vm, from, bufs, n_left_from); b = bufs; ti = thread_indices; - fq_index = (is_feature) ? rm->fq_feature_index : rm->fq_index; + fq_index = (is_feature) ? rm->fq_feature_index : + (is_custom_context ? rm->fq_custom_context_index : + rm->fq_index); while (n_left_from > 0) { @@ -1435,8 +1547,12 @@ ip4_sv_reass_handoff_node_inline (vlib_main_t * vm, ti += 1; b += 1; } - n_enq = vlib_buffer_enqueue_to_thread (vm, node, fq_index, from, - thread_indices, frame->n_vectors, 1); + if (is_custom_context) + n_enq = vlib_buffer_enqueue_to_thread_with_aux ( + vm, node, fq_index, from, context, thread_indices, frame->n_vectors, 1); + else + n_enq = vlib_buffer_enqueue_to_thread ( + vm, node, fq_index, from, thread_indices, frame->n_vectors, 1); if (n_enq < frame->n_vectors) vlib_node_increment_counter (vm, node->node_index, @@ -1449,12 +1565,11 @@ VLIB_NODE_FN (ip4_sv_reass_handoff_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip4_sv_reass_handoff_node_inline (vm, node, frame, - false /* is_feature */ ); + return ip4_sv_reass_handoff_node_inline ( + vm, node, frame, false /* is_feature */, false /* is_custom_context */); } -/* *INDENT-OFF* */ VLIB_REGISTER_NODE (ip4_sv_reass_handoff_node) = { .name = "ip4-sv-reassembly-handoff", .vector_size = sizeof (u32), @@ -1468,22 +1583,39 @@ VLIB_REGISTER_NODE (ip4_sv_reass_handoff_node) = { [0] = "error-drop", }, }; -/* *INDENT-ON* */ +VLIB_NODE_FN (ip4_sv_reass_custom_context_handoff_node) +(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame) +{ + return ip4_sv_reass_handoff_node_inline ( + vm, node, frame, false /* is_feature */, true /* is_custom_context */); +} + +VLIB_REGISTER_NODE (ip4_sv_reass_custom_context_handoff_node) = { + .name = "ip4-sv-reassembly-custom-context-handoff", + .vector_size = sizeof (u32), + .aux_size = sizeof (u32), + .n_errors = ARRAY_LEN(ip4_sv_reass_handoff_error_strings), + .error_strings = ip4_sv_reass_handoff_error_strings, + .format_trace = format_ip4_sv_reass_handoff_trace, + + .n_next_nodes = 1, + + .next_nodes = { + [0] = "error-drop", + }, +}; -/* *INDENT-OFF* */ VLIB_NODE_FN (ip4_sv_reass_feature_handoff_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip4_sv_reass_handoff_node_inline (vm, node, frame, - true /* is_feature */ ); + return ip4_sv_reass_handoff_node_inline ( + vm, node, frame, true /* is_feature */, false /* is_custom_context */); } -/* *INDENT-ON* */ -/* *INDENT-OFF* */ VLIB_REGISTER_NODE (ip4_sv_reass_feature_handoff_node) = { .name = "ip4-sv-reass-feature-hoff", .vector_size = sizeof (u32), @@ -1497,7 +1629,6 @@ VLIB_REGISTER_NODE (ip4_sv_reass_feature_handoff_node) = { [0] = "error-drop", }, }; -/* *INDENT-ON* */ #ifndef CLIB_MARCH_VARIANT int @@ -1535,6 +1666,13 @@ ip4_sv_reass_custom_register_next_node (uword node_index) node_index); } +uword +ip4_sv_reass_custom_context_register_next_node (uword node_index) +{ + return vlib_node_add_next ( + vlib_get_main (), ip4_sv_reass_custom_context_node.index, node_index); +} + int ip4_sv_reass_output_enable_disable_with_refcnt (u32 sw_if_index, int is_enable) diff --git a/src/vnet/ip/reass/ip4_sv_reass.h b/src/vnet/ip/reass/ip4_sv_reass.h index e926dbeebcc..3a684eb9809 100644 --- a/src/vnet/ip/reass/ip4_sv_reass.h +++ b/src/vnet/ip/reass/ip4_sv_reass.h @@ -49,6 +49,7 @@ int ip4_sv_reass_output_enable_disable_with_refcnt (u32 sw_if_index, int is_enable); uword ip4_sv_reass_custom_register_next_node (uword node_index); +uword ip4_sv_reass_custom_context_register_next_node (uword node_index); #endif /* __included_ip4_sv_reass_h__ */ diff --git a/src/vnet/ip/reass/ip6_full_reass.c b/src/vnet/ip/reass/ip6_full_reass.c index 9ec40cd347c..27647985877 100644 --- a/src/vnet/ip/reass/ip6_full_reass.c +++ b/src/vnet/ip/reass/ip6_full_reass.c @@ -25,10 +25,14 @@ #include <vnet/ip/ip.h> #include <vppinfra/bihash_48_8.h> #include <vnet/ip/reass/ip6_full_reass.h> +#include <vnet/ip/ip6_inlines.h> #define MSEC_PER_SEC 1000 -#define IP6_FULL_REASS_TIMEOUT_DEFAULT_MS 100 -#define IP6_FULL_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS 10000 // 10 seconds default +#define IP6_FULL_REASS_TIMEOUT_DEFAULT_MS 200 +/* As there are only 1024 reass context per thread, either the DDOS attacks + * or fractions of real timeouts, would consume these contexts quickly and + * running out context space and unable to perform reassembly */ +#define IP6_FULL_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS 50 // 50 ms default #define IP6_FULL_REASS_MAX_REASSEMBLIES_DEFAULT 1024 #define IP6_FULL_REASS_MAX_REASSEMBLY_LENGTH_DEFAULT 3 #define IP6_FULL_REASS_HT_LOAD_FACTOR (0.75) @@ -40,6 +44,8 @@ typedef enum IP6_FULL_REASS_RC_TOO_MANY_FRAGMENTS, IP6_FULL_REASS_RC_NO_BUF, IP6_FULL_REASS_RC_HANDOFF, + IP6_FULL_REASS_RC_INVALID_FRAG_LEN, + IP6_FULL_REASS_RC_OVERLAP, } ip6_full_reass_rc_t; typedef struct @@ -132,6 +138,8 @@ typedef struct ip6_full_reass_t *pool; u32 reass_n; u32 id_counter; + // for pacing the main thread timeouts + u32 last_id; clib_spinlock_t lock; } ip6_full_reass_per_thread_t; @@ -155,17 +163,20 @@ typedef struct // convenience vlib_main_t *vlib_main; - // node index of ip6-drop node - u32 ip6_drop_idx; u32 ip6_icmp_error_idx; u32 ip6_full_reass_expire_node_idx; /** Worker handoff */ u32 fq_index; + u32 fq_local_index; u32 fq_feature_index; + u32 fq_custom_index; // reference count for enabling/disabling feature - per interface u32 *feature_use_refcount_per_intf; + + // whether local fragmented packets are reassembled or not + int is_local_reass_enabled; } ip6_full_reass_main_t; extern ip6_full_reass_main_t ip6_full_reass_main; @@ -185,13 +196,22 @@ typedef enum typedef enum { + NORMAL, + FEATURE, + CUSTOM +} ip6_full_reass_node_type_t; + +typedef enum +{ RANGE_NEW, + RANGE_DISCARD, RANGE_OVERLAP, ICMP_ERROR_RT_EXCEEDED, ICMP_ERROR_FL_TOO_BIG, ICMP_ERROR_FL_NOT_MULT_8, FINALIZE, HANDOFF, + PASSTHROUGH, } ip6_full_reass_trace_operation_e; typedef struct @@ -278,6 +298,10 @@ format_ip6_full_reass_trace (u8 * s, va_list * args) s = format (s, "\n%Unew %U", format_white_space, indent, format_ip6_full_reass_range_trace, &t->trace_range); break; + case RANGE_DISCARD: + s = format (s, "\n%Udiscard %U", format_white_space, indent, + format_ip6_full_reass_range_trace, &t->trace_range); + break; case RANGE_OVERLAP: s = format (s, "\n%Uoverlap %U", format_white_space, indent, format_ip6_full_reass_range_trace, &t->trace_range); @@ -304,6 +328,9 @@ format_ip6_full_reass_trace (u8 * s, va_list * args) format (s, "handoff from thread #%u to thread #%u", t->thread_id, t->thread_id_to); break; + case PASSTHROUGH: + s = format (s, "passthrough - not a fragment"); + break; } return s; } @@ -396,59 +423,69 @@ ip6_full_reass_free (ip6_full_reass_main_t * rm, ip6_full_reass_free_ctx (rt, reass); } +/* n_left_to_next, and to_next are taken as input params, as this function + * could be called from a graphnode, where its managing local copy of these + * variables, and ignoring those and still trying to enqueue the buffers + * with local variables would cause either buffer leak or corruption */ always_inline void ip6_full_reass_drop_all (vlib_main_t *vm, vlib_node_runtime_t *node, - ip6_full_reass_t *reass) + ip6_full_reass_t *reass, u32 *n_left_to_next, + u32 **to_next) { u32 range_bi = reass->first_bi; vlib_buffer_t *range_b; vnet_buffer_opaque_t *range_vnb; u32 *to_free = NULL; + 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) + + if (~0 != range_bi) { - vec_add1 (to_free, 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; - } + vec_add1 (to_free, range_bi); } range_bi = range_vnb->ip.reass.next_range_bi; } + /* send to next_error_index */ - if (~0 != reass->error_next_index) + if (~0 != reass->error_next_index && + reass->error_next_index < node->n_next_nodes) { - u32 n_left_to_next, *to_next, next_index; + u32 next_index; next_index = reass->error_next_index; u32 bi = ~0; + /* record number of packets sent to custom app */ + vlib_node_increment_counter (vm, node->node_index, + IP6_ERROR_REASS_TO_CUSTOM_APP, + vec_len (to_free)); + while (vec_len (to_free) > 0) { - vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + vlib_get_next_frame (vm, node, next_index, *to_next, + (*n_left_to_next)); - while (vec_len (to_free) > 0 && n_left_to_next > 0) + while (vec_len (to_free) > 0 && (*n_left_to_next) > 0) { bi = vec_pop (to_free); if (~0 != bi) { - to_next[0] = bi; - to_next += 1; - n_left_to_next -= 1; + vlib_buffer_t *b = vlib_get_buffer (vm, bi); + if (PREDICT_FALSE (b->flags & VLIB_BUFFER_IS_TRACED)) + { + ip6_full_reass_add_trace (vm, node, reass, bi, NULL, + RANGE_DISCARD, ~0); + } + *to_next[0] = bi; + (*to_next) += 1; + (*n_left_to_next) -= 1; } } - vlib_put_next_frame (vm, node, next_index, n_left_to_next); + vlib_put_next_frame (vm, node, next_index, (*n_left_to_next)); } } else @@ -459,8 +496,65 @@ ip6_full_reass_drop_all (vlib_main_t *vm, vlib_node_runtime_t *node, } always_inline void -ip6_full_reass_on_timeout (vlib_main_t * vm, vlib_node_runtime_t * node, - ip6_full_reass_t * reass, u32 * icmp_bi) +sanitize_reass_buffers_add_missing (vlib_main_t *vm, ip6_full_reass_t *reass, + u32 *bi0) +{ + 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; + if (~0 != bi) + { + if (bi == *bi0) + *bi0 = ~0; + if (range_b->flags & VLIB_BUFFER_NEXT_PRESENT) + { + u32 _bi = bi; + vlib_buffer_t *_b = vlib_get_buffer (vm, _bi); + while (_b->flags & VLIB_BUFFER_NEXT_PRESENT) + { + if (_b->next_buffer != range_vnb->ip.reass.next_range_bi) + { + _bi = _b->next_buffer; + _b = vlib_get_buffer (vm, _bi); + } + else + { + _b->flags &= ~VLIB_BUFFER_NEXT_PRESENT; + break; + } + } + } + range_bi = range_vnb->ip.reass.next_range_bi; + } + } + if (*bi0 != ~0) + { + vlib_buffer_t *fb = vlib_get_buffer (vm, *bi0); + vnet_buffer_opaque_t *fvnb = vnet_buffer (fb); + if (~0 != reass->first_bi) + { + fvnb->ip.reass.next_range_bi = reass->first_bi; + reass->first_bi = *bi0; + } + else + { + reass->first_bi = *bi0; + fvnb->ip.reass.next_range_bi = ~0; + } + *bi0 = ~0; + } +} + +always_inline void +ip6_full_reass_on_timeout (vlib_main_t *vm, vlib_node_runtime_t *node, + ip6_full_reass_t *reass, u32 *icmp_bi, + u32 *n_left_to_next, u32 **to_next) { if (~0 == reass->first_bi) { @@ -493,15 +587,16 @@ ip6_full_reass_on_timeout (vlib_main_t * vm, vlib_node_runtime_t * node, 0); } } - ip6_full_reass_drop_all (vm, node, reass); + ip6_full_reass_drop_all (vm, node, reass, n_left_to_next, to_next); } always_inline ip6_full_reass_t * -ip6_full_reass_find_or_create (vlib_main_t * vm, vlib_node_runtime_t * node, - ip6_full_reass_main_t * rm, - ip6_full_reass_per_thread_t * rt, - ip6_full_reass_kv_t * kv, u32 * icmp_bi, - u8 * do_handoff) +ip6_full_reass_find_or_create (vlib_main_t *vm, vlib_node_runtime_t *node, + ip6_full_reass_main_t *rm, + ip6_full_reass_per_thread_t *rt, + ip6_full_reass_kv_t *kv, u32 *icmp_bi, + u8 *do_handoff, int skip_bihash, + u32 *n_left_to_next, u32 **to_next) { ip6_full_reass_t *reass; f64 now; @@ -511,7 +606,7 @@ again: reass = NULL; now = vlib_time_now (vm); - if (!clib_bihash_search_48_8 (&rm->hash, &kv->kv, &kv->kv)) + if (!skip_bihash && !clib_bihash_search_48_8 (&rm->hash, &kv->kv, &kv->kv)) { if (vm->thread_index != kv->v.memory_owner_thread_index) { @@ -526,7 +621,10 @@ again: if (now > reass->last_heard + rm->timeout) { - ip6_full_reass_on_timeout (vm, node, reass, icmp_bi); + vlib_node_increment_counter (vm, node->node_index, + IP6_ERROR_REASS_TIMEOUT, 1); + ip6_full_reass_on_timeout (vm, node, reass, icmp_bi, n_left_to_next, + to_next); ip6_full_reass_free (rm, rt, reass); reass = NULL; } @@ -554,27 +652,41 @@ again: reass->data_len = 0; reass->next_index = ~0; reass->error_next_index = ~0; + reass->memory_owner_thread_index = vm->thread_index; ++rt->reass_n; } - reass->key.as_u64[0] = kv->kv.key[0]; - reass->key.as_u64[1] = kv->kv.key[1]; - reass->key.as_u64[2] = kv->kv.key[2]; - reass->key.as_u64[3] = kv->kv.key[3]; - reass->key.as_u64[4] = kv->kv.key[4]; - reass->key.as_u64[5] = kv->kv.key[5]; kv->v.reass_index = (reass - rt->pool); kv->v.memory_owner_thread_index = vm->thread_index; reass->last_heard = now; - int rv = clib_bihash_add_del_48_8 (&rm->hash, &kv->kv, 2); - if (rv) + if (!skip_bihash) { - ip6_full_reass_free (rm, rt, reass); - reass = NULL; - // if other worker created a context already work with the other copy - if (-2 == rv) - goto again; + reass->key.as_u64[0] = kv->kv.key[0]; + reass->key.as_u64[1] = kv->kv.key[1]; + reass->key.as_u64[2] = kv->kv.key[2]; + reass->key.as_u64[3] = kv->kv.key[3]; + reass->key.as_u64[4] = kv->kv.key[4]; + reass->key.as_u64[5] = kv->kv.key[5]; + + int rv = clib_bihash_add_del_48_8 (&rm->hash, &kv->kv, 2); + if (rv) + { + ip6_full_reass_free (rm, rt, reass); + reass = NULL; + // if other worker created a context already work with the other copy + if (-2 == rv) + goto again; + } + } + else + { + reass->key.as_u64[0] = ~0; + reass->key.as_u64[1] = ~0; + reass->key.as_u64[2] = ~0; + reass->key.as_u64[3] = ~0; + reass->key.as_u64[4] = ~0; + reass->key.as_u64[5] = ~0; } return reass; @@ -593,8 +705,6 @@ ip6_full_reass_finalize (vlib_main_t * vm, vlib_node_runtime_t * node, 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; u32 *vec_drop_compress = NULL; ip6_full_reass_rc_t rv = IP6_FULL_REASS_RC_OK; do @@ -636,19 +746,18 @@ ip6_full_reass_finalize (vlib_main_t * vm, vlib_node_runtime_t * node, 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); - trim_front -= tmp->current_length; if (!(tmp->flags & VLIB_BUFFER_NEXT_PRESENT)) { rv = IP6_FULL_REASS_RC_INTERNAL_ERROR; goto free_buffers_and_return; } + trim_front -= tmp->current_length; + vec_add1 (vec_drop_compress, tmp_bi); tmp->flags &= ~VLIB_BUFFER_NEXT_PRESENT; tmp_bi = tmp->next_buffer; tmp = vlib_get_buffer (vm, tmp_bi); @@ -686,13 +795,12 @@ ip6_full_reass_finalize (vlib_main_t * vm, vlib_node_runtime_t * node, } else { - vec_add1 (vec_drop_compress, tmp_bi); if (reass->first_bi == tmp_bi) { rv = IP6_FULL_REASS_RC_INTERNAL_ERROR; goto free_buffers_and_return; } - ++dropped_cnt; + vec_add1 (vec_drop_compress, tmp_bi); } if (tmp->flags & VLIB_BUFFER_NEXT_PRESENT) { @@ -729,19 +837,27 @@ ip6_full_reass_finalize (vlib_main_t * vm, vlib_node_runtime_t * node, 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; - frag_hdr = - ip6_ext_header_find (vm, first_b, ip, IP_PROTOCOL_IPV6_FRAGMENTATION, - &prev_hdr); - if (prev_hdr) + ip6_ext_hdr_chain_t hdr_chain; + ip6_ext_header_t *prev_hdr = 0; + int res = ip6_ext_header_walk (first_b, ip, IP_PROTOCOL_IPV6_FRAGMENTATION, + &hdr_chain); + if (res < 0 || + (hdr_chain.eh[res].protocol != IP_PROTOCOL_IPV6_FRAGMENTATION)) { + rv = IP6_FULL_REASS_RC_INTERNAL_ERROR; + goto free_buffers_and_return; + } + frag_hdr = ip6_ext_next_header_offset (ip, hdr_chain.eh[res].offset); + if (res > 0) + { + prev_hdr = ip6_ext_next_header_offset (ip, hdr_chain.eh[res - 1].offset); prev_hdr->next_hdr = frag_hdr->next_hdr; } else { ip->protocol = frag_hdr->next_hdr; } - if (!((u8 *) frag_hdr - (u8 *) ip == ip6_frag_hdr_offset)) + if (hdr_chain.eh[res].offset != ip6_frag_hdr_offset) { rv = IP6_FULL_REASS_RC_INTERNAL_ERROR; goto free_buffers_and_return; @@ -799,6 +915,15 @@ ip6_full_reass_finalize (vlib_main_t * vm, vlib_node_runtime_t * node, *next0 = reass->next_index; } vnet_buffer (first_b)->ip.reass.estimated_mtu = reass->min_fragment_length; + /* Keep track of number of successfully reassembled packets and number of + * fragments reassembled */ + vlib_node_increment_counter (vm, node->node_index, IP6_ERROR_REASS_SUCCESS, + 1); + + vlib_node_increment_counter (vm, node->node_index, + IP6_ERROR_REASS_FRAGMENTS_REASSEMBLED, + reass->fragments_n); + ip6_full_reass_free (rm, rt, reass); reass = NULL; free_buffers_and_return: @@ -834,12 +959,13 @@ ip6_full_reass_insert_range_in_chain (vlib_main_t * vm, } always_inline ip6_full_reass_rc_t -ip6_full_reass_update (vlib_main_t * vm, vlib_node_runtime_t * node, - ip6_full_reass_main_t * rm, - ip6_full_reass_per_thread_t * rt, - ip6_full_reass_t * reass, u32 * bi0, u32 * next0, - u32 * error0, ip6_frag_hdr_t * frag_hdr, - bool is_custom_app, u32 * handoff_thread_idx) +ip6_full_reass_update (vlib_main_t *vm, vlib_node_runtime_t *node, + ip6_full_reass_main_t *rm, + ip6_full_reass_per_thread_t *rt, + ip6_full_reass_t *reass, u32 *bi0, u32 *next0, + u32 *error0, ip6_frag_hdr_t *frag_hdr, + bool is_custom_app, u32 *handoff_thread_idx, + int skip_bihash) { int consumed = 0; vlib_buffer_t *fb = vlib_get_buffer (vm, *bi0); @@ -865,6 +991,10 @@ ip6_full_reass_update (vlib_main_t * vm, vlib_node_runtime_t * node, u32 fragment_length = vlib_buffer_length_in_chain (vm, fb) - (fvnb->ip.reass.ip6_frag_hdr_offset + sizeof (*frag_hdr)); + if (0 == fragment_length) + { + return IP6_FULL_REASS_RC_INVALID_FRAG_LEN; + } u32 fragment_last = fvnb->ip.reass.fragment_last = fragment_first + fragment_length - 1; int more_fragments = ip6_frag_hdr_more (frag_hdr); @@ -929,11 +1059,7 @@ ip6_full_reass_update (vlib_main_t * vm, vlib_node_runtime_t * node, ip6_full_reass_add_trace (vm, node, reass, *bi0, frag_hdr, RANGE_OVERLAP, ~0); } - ip6_full_reass_drop_all (vm, node, reass); - ip6_full_reass_free (rm, rt, reass); - *next0 = IP6_FULL_REASSEMBLY_NEXT_DROP; - *error0 = IP6_ERROR_REASS_OVERLAPPING_FRAGMENT; - return IP6_FULL_REASS_RC_OK; + return IP6_FULL_REASS_RC_OVERLAP; } break; } @@ -947,6 +1073,12 @@ check_if_done_maybe: ~0); } } + else if (skip_bihash) + { + // if this reassembly is not in bihash, then the packet must have been + // consumed + return IP6_FULL_REASS_RC_INTERNAL_ERROR; + } if (~0 != reass->last_packet_octet && reass->data_len == reass->last_packet_octet + 1) { @@ -964,6 +1096,12 @@ check_if_done_maybe: } else { + if (skip_bihash) + { + // if this reassembly is not in bihash, it should've been an atomic + // fragment and thus finalized + return IP6_FULL_REASS_RC_INTERNAL_ERROR; + } if (consumed) { *bi0 = ~0; @@ -982,31 +1120,28 @@ check_if_done_maybe: } always_inline bool -ip6_full_reass_verify_upper_layer_present (vlib_node_runtime_t * node, - vlib_buffer_t * b, - ip6_frag_hdr_t * frag_hdr) +ip6_full_reass_verify_upper_layer_present (vlib_node_runtime_t *node, + vlib_buffer_t *b, + ip6_ext_hdr_chain_t *hc) { - 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) + int nh = hc->eh[hc->length - 1].protocol; + /* Checking to see if it's a terminating header */ + if (ip6_ext_hdr (nh)) { - icmp6_error_set_vnet_buffer (b, ICMP6_parameter_problem, - ICMP6_parameter_problem_first_fragment_has_incomplete_header_chain, - 0); + icmp6_error_set_vnet_buffer ( + b, ICMP6_parameter_problem, + ICMP6_parameter_problem_first_fragment_has_incomplete_header_chain, 0); b->error = node->errors[IP6_ERROR_REASS_MISSING_UPPER]; - return false; } return true; } always_inline bool -ip6_full_reass_verify_fragment_multiple_8 (vlib_main_t * vm, - vlib_buffer_t * b, - ip6_frag_hdr_t * frag_hdr) +ip6_full_reass_verify_fragment_multiple_8 (vlib_main_t *vm, + vlib_node_runtime_t *node, + vlib_buffer_t *b, + ip6_frag_hdr_t *frag_hdr) { vnet_buffer_opaque_t *vnb = vnet_buffer (b); ip6_header_t *ip = vlib_buffer_get_current (b); @@ -1019,15 +1154,17 @@ ip6_full_reass_verify_fragment_multiple_8 (vlib_main_t * vm, icmp6_error_set_vnet_buffer (b, ICMP6_parameter_problem, ICMP6_parameter_problem_erroneous_header_field, (u8 *) & ip->payload_length - (u8 *) ip); + b->error = node->errors[IP6_ERROR_REASS_INVALID_FRAG_SIZE]; return false; } return true; } always_inline bool -ip6_full_reass_verify_packet_size_lt_64k (vlib_main_t * vm, - vlib_buffer_t * b, - ip6_frag_hdr_t * frag_hdr) +ip6_full_reass_verify_packet_size_lt_64k (vlib_main_t *vm, + vlib_node_runtime_t *node, + vlib_buffer_t *b, + ip6_frag_hdr_t *frag_hdr) { vnet_buffer_opaque_t *vnb = vnet_buffer (b); u32 fragment_first = ip6_frag_hdr_offset_bytes (frag_hdr); @@ -1041,16 +1178,16 @@ ip6_full_reass_verify_packet_size_lt_64k (vlib_main_t * vm, ICMP6_parameter_problem_erroneous_header_field, (u8 *) & frag_hdr->fragment_offset_and_more - (u8 *) ip0); + b->error = node->errors[IP6_ERROR_REASS_INVALID_FRAG_SIZE]; return false; } return true; } always_inline uword -ip6_full_reassembly_inline (vlib_main_t * vm, - vlib_node_runtime_t * node, - vlib_frame_t * frame, bool is_feature, - bool is_custom_app) +ip6_full_reassembly_inline (vlib_main_t *vm, vlib_node_runtime_t *node, + vlib_frame_t *frame, bool is_feature, + bool is_custom_app, bool is_local) { u32 *from = vlib_frame_vector_args (frame); u32 n_left_from, n_left_to_next, *to_next, next_index; @@ -1077,55 +1214,95 @@ ip6_full_reassembly_inline (vlib_main_t * vm, ip6_header_t *ip0 = vlib_buffer_get_current (b0); ip6_frag_hdr_t *frag_hdr = NULL; - ip6_ext_header_t *prev_hdr; - if (ip6_ext_hdr (ip0->protocol)) + ip6_ext_hdr_chain_t hdr_chain; + vnet_buffer_opaque_t *fvnb = vnet_buffer (b0); + + int res = ip6_ext_header_walk ( + b0, ip0, IP_PROTOCOL_IPV6_FRAGMENTATION, &hdr_chain); + if (res < 0 || + hdr_chain.eh[res].protocol != IP_PROTOCOL_IPV6_FRAGMENTATION) { - frag_hdr = - ip6_ext_header_find (vm, b0, ip0, - IP_PROTOCOL_IPV6_FRAGMENTATION, - &prev_hdr); + vlib_node_increment_counter (vm, node->node_index, + IP6_ERROR_REASS_NO_FRAG_HDR, 1); + // this is a mangled packet - no fragmentation + next0 = is_custom_app ? fvnb->ip.reass.error_next_index : + IP6_FULL_REASSEMBLY_NEXT_DROP; + ip6_full_reass_add_trace (vm, node, NULL, bi0, NULL, PASSTHROUGH, + ~0); + goto skip_reass; } - if (!frag_hdr) + if (is_local && !rm->is_local_reass_enabled) { - // this is a regular packet - no fragmentation - next0 = IP6_FULL_REASSEMBLY_NEXT_INPUT; + next0 = IP6_FULL_REASSEMBLY_NEXT_DROP; goto skip_reass; } + + /* Keep track of received fragments */ + vlib_node_increment_counter (vm, node->node_index, + IP6_ERROR_REASS_FRAGMENTS_RCVD, 1); + frag_hdr = + ip6_ext_next_header_offset (ip0, hdr_chain.eh[res].offset); vnet_buffer (b0)->ip.reass.ip6_frag_hdr_offset = - (u8 *) frag_hdr - (u8 *) ip0; + hdr_chain.eh[res].offset; if (0 == ip6_frag_hdr_offset (frag_hdr)) { // first fragment - verify upper-layer is present - if (!ip6_full_reass_verify_upper_layer_present - (node, b0, frag_hdr)) + if (!ip6_full_reass_verify_upper_layer_present (node, b0, + &hdr_chain)) { - next0 = IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR; + next0 = is_custom_app ? fvnb->ip.reass.error_next_index : + IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR; goto skip_reass; } } - if (!ip6_full_reass_verify_fragment_multiple_8 (vm, b0, frag_hdr) || - !ip6_full_reass_verify_packet_size_lt_64k (vm, b0, frag_hdr)) + + if (!ip6_full_reass_verify_fragment_multiple_8 (vm, node, b0, + frag_hdr) || + !ip6_full_reass_verify_packet_size_lt_64k (vm, node, b0, + frag_hdr)) { - next0 = IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR; + next0 = is_custom_app ? fvnb->ip.reass.error_next_index : + IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR; goto skip_reass; } + + int skip_bihash = 0; ip6_full_reass_kv_t kv; u8 do_handoff = 0; - kv.k.as_u64[0] = ip0->src_address.as_u64[0]; - kv.k.as_u64[1] = ip0->src_address.as_u64[1]; - kv.k.as_u64[2] = ip0->dst_address.as_u64[0]; - kv.k.as_u64[3] = ip0->dst_address.as_u64[1]; - kv.k.as_u64[4] = - ((u64) vec_elt (ip6_main.fib_index_by_sw_if_index, - vnet_buffer (b0)->sw_if_index[VLIB_RX])) << 32 | - (u64) frag_hdr->identification; - kv.k.as_u64[5] = ip0->protocol; + if (0 == ip6_frag_hdr_offset (frag_hdr) && + !ip6_frag_hdr_more (frag_hdr)) + { + // this is atomic fragment and needs to be processed separately + skip_bihash = 1; + } + else + { + u32 fib_index = + (vnet_buffer (b0)->sw_if_index[VLIB_TX] == (u32) ~0) ? + vec_elt (ip6_main.fib_index_by_sw_if_index, + vnet_buffer (b0)->sw_if_index[VLIB_RX]) : + vnet_buffer (b0)->sw_if_index[VLIB_TX]; + kv.k.as_u64[0] = ip0->src_address.as_u64[0]; + kv.k.as_u64[1] = ip0->src_address.as_u64[1]; + kv.k.as_u64[2] = ip0->dst_address.as_u64[0]; + kv.k.as_u64[3] = ip0->dst_address.as_u64[1]; + kv.k.as_u64[4] = + ((u64) fib_index) << 32 | (u64) frag_hdr->identification; + /* RFC 8200: The Next Header values in the Fragment headers of + * different fragments of the same original packet may differ. + * Only the value from the Offset zero fragment packet is used + * for reassembly. + * + * Also, IPv6 Header doesnt contain the protocol value unlike + * IPv4.*/ + kv.k.as_u64[5] = 0; + } - ip6_full_reass_t *reass = - ip6_full_reass_find_or_create (vm, node, rm, rt, &kv, &icmp_bi, - &do_handoff); + ip6_full_reass_t *reass = ip6_full_reass_find_or_create ( + vm, node, rm, rt, &kv, &icmp_bi, &do_handoff, skip_bihash, + &n_left_to_next, &to_next); if (reass) { @@ -1144,9 +1321,10 @@ ip6_full_reassembly_inline (vlib_main_t * vm, else if (reass) { u32 handoff_thread_idx; - switch (ip6_full_reass_update - (vm, node, rm, rt, reass, &bi0, &next0, &error0, - frag_hdr, is_custom_app, &handoff_thread_idx)) + u32 counter = ~0; + switch (ip6_full_reass_update ( + vm, node, rm, rt, reass, &bi0, &next0, &error0, frag_hdr, + is_custom_app, &handoff_thread_idx, skip_bihash)) { case IP6_FULL_REASS_RC_OK: /* nothing to do here */ @@ -1158,25 +1336,36 @@ ip6_full_reassembly_inline (vlib_main_t * vm, handoff_thread_idx; break; case IP6_FULL_REASS_RC_TOO_MANY_FRAGMENTS: - vlib_node_increment_counter (vm, node->node_index, - IP6_ERROR_REASS_FRAGMENT_CHAIN_TOO_LONG, - 1); - ip6_full_reass_drop_all (vm, node, reass); - ip6_full_reass_free (rm, rt, reass); - goto next_packet; + counter = IP6_ERROR_REASS_FRAGMENT_CHAIN_TOO_LONG; break; case IP6_FULL_REASS_RC_NO_BUF: - vlib_node_increment_counter (vm, node->node_index, - IP6_ERROR_REASS_NO_BUF, 1); - ip6_full_reass_drop_all (vm, node, reass); - ip6_full_reass_free (rm, rt, reass); - goto next_packet; + counter = IP6_ERROR_REASS_NO_BUF; + break; + case IP6_FULL_REASS_RC_INVALID_FRAG_LEN: + counter = IP6_ERROR_REASS_INVALID_FRAG_LEN; + break; + case IP6_FULL_REASS_RC_OVERLAP: + counter = IP6_ERROR_REASS_OVERLAPPING_FRAGMENT; break; case IP6_FULL_REASS_RC_INTERNAL_ERROR: - vlib_node_increment_counter (vm, node->node_index, - IP6_ERROR_REASS_INTERNAL_ERROR, + counter = IP6_ERROR_REASS_INTERNAL_ERROR; + /* Sanitization is needed in internal error cases only, as + * the incoming packet is already dropped in other cases, + * also adding bi0 back to the reassembly list, fixes the + * leaking of buffers during internal errors. + * + * Also it doesnt make sense to send these buffers custom + * app, these fragments are with internal errors */ + sanitize_reass_buffers_add_missing (vm, reass, &bi0); + reass->error_next_index = ~0; + break; + } + if (~0 != counter) + { + vlib_node_increment_counter (vm, node->node_index, counter, 1); - ip6_full_reass_drop_all (vm, node, reass); + ip6_full_reass_drop_all (vm, node, reass, &n_left_to_next, + &to_next); ip6_full_reass_free (rm, rt, reass); goto next_packet; break; @@ -1190,7 +1379,6 @@ ip6_full_reassembly_inline (vlib_main_t * vm, } else { - vnet_buffer_opaque_t *fvnb = vnet_buffer (b0); next0 = fvnb->ip.reass.error_next_index; } error0 = IP6_ERROR_REASS_LIMIT_REACHED; @@ -1223,6 +1411,15 @@ ip6_full_reassembly_inline (vlib_main_t * vm, { vnet_feature_next (&next0, b0); } + + /* Increment the counter to-custom-app also as this fragment is + * also going to application */ + if (is_custom_app) + { + vlib_node_increment_counter ( + vm, node->node_index, IP6_ERROR_REASS_TO_CUSTOM_APP, 1); + } + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, n_left_to_next, bi0, next0); } @@ -1249,26 +1446,21 @@ ip6_full_reassembly_inline (vlib_main_t * vm, return frame->n_vectors; } -static char *ip6_full_reassembly_error_strings[] = { -#define _(sym, string) string, - foreach_ip6_error -#undef _ -}; - VLIB_NODE_FN (ip6_full_reass_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip6_full_reassembly_inline (vm, node, frame, false /* is_feature */ , - false /* is_custom_app */ ); + return ip6_full_reassembly_inline (vm, node, frame, false /* is_feature */, + false /* is_custom_app */, + false /* is_local */); } VLIB_REGISTER_NODE (ip6_full_reass_node) = { .name = "ip6-full-reassembly", .vector_size = sizeof (u32), .format_trace = format_ip6_full_reass_trace, - .n_errors = ARRAY_LEN (ip6_full_reassembly_error_strings), - .error_strings = ip6_full_reassembly_error_strings, + .n_errors = IP6_N_ERROR, + .error_counters = ip6_error_counters, .n_next_nodes = IP6_FULL_REASSEMBLY_N_NEXT, .next_nodes = { @@ -1279,20 +1471,45 @@ VLIB_REGISTER_NODE (ip6_full_reass_node) = { }, }; +VLIB_NODE_FN (ip6_local_full_reass_node) +(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame) +{ + return ip6_full_reassembly_inline (vm, node, frame, false /* is_feature */, + false /* is_custom_app */, + true /* is_local */); +} + +VLIB_REGISTER_NODE (ip6_local_full_reass_node) = { + .name = "ip6-local-full-reassembly", + .vector_size = sizeof (u32), + .format_trace = format_ip6_full_reass_trace, + .n_errors = IP6_N_ERROR, + .error_counters = ip6_error_counters, + .n_next_nodes = IP6_FULL_REASSEMBLY_N_NEXT, + .next_nodes = + { + [IP6_FULL_REASSEMBLY_NEXT_INPUT] = "ip6-input", + [IP6_FULL_REASSEMBLY_NEXT_DROP] = "ip6-drop", + [IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR] = "ip6-icmp-error", + [IP6_FULL_REASSEMBLY_NEXT_HANDOFF] = "ip6-local-full-reassembly-handoff", + }, +}; + VLIB_NODE_FN (ip6_full_reass_node_feature) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip6_full_reassembly_inline (vm, node, frame, true /* is_feature */ , - false /* is_custom_app */ ); + return ip6_full_reassembly_inline (vm, node, frame, true /* is_feature */, + false /* is_custom_app */, + false /* is_local */); } VLIB_REGISTER_NODE (ip6_full_reass_node_feature) = { .name = "ip6-full-reassembly-feature", .vector_size = sizeof (u32), .format_trace = format_ip6_full_reass_trace, - .n_errors = ARRAY_LEN (ip6_full_reassembly_error_strings), - .error_strings = ip6_full_reassembly_error_strings, + .n_errors = IP6_N_ERROR, + .error_counters = ip6_error_counters, .n_next_nodes = IP6_FULL_REASSEMBLY_N_NEXT, .next_nodes = { @@ -1311,6 +1528,30 @@ VNET_FEATURE_INIT (ip6_full_reassembly_feature, static) = { .runs_after = 0, }; +VLIB_NODE_FN (ip6_full_reass_node_custom) +(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame) +{ + return ip6_full_reassembly_inline (vm, node, frame, false /* is_feature */, + true /* is_custom_app */, + false /* is_local */); +} + +VLIB_REGISTER_NODE (ip6_full_reass_node_custom) = { + .name = "ip6-full-reassembly-custom", + .vector_size = sizeof (u32), + .format_trace = format_ip6_full_reass_trace, + .n_errors = IP6_N_ERROR, + .error_counters = ip6_error_counters, + .n_next_nodes = IP6_FULL_REASSEMBLY_N_NEXT, + .next_nodes = + { + [IP6_FULL_REASSEMBLY_NEXT_INPUT] = "ip6-input", + [IP6_FULL_REASSEMBLY_NEXT_DROP] = "ip6-drop", + [IP6_FULL_REASSEMBLY_NEXT_ICMP_ERROR] = "ip6-icmp-error", + [IP6_FULL_REASSEMBLY_NEXT_HANDOFF] = "ip6-full-reass-custom-hoff", + }, +}; + #ifndef CLIB_MARCH_VARIANT static u32 ip6_full_reass_get_nbuckets () @@ -1319,7 +1560,9 @@ ip6_full_reass_get_nbuckets () u32 nbuckets; u8 i; - nbuckets = (u32) (rm->max_reass_n / IP6_FULL_REASS_HT_LOAD_FACTOR); + /* need more mem with more workers */ + nbuckets = (u32) (rm->max_reass_n * (vlib_num_workers () + 1) / + IP6_FULL_REASS_HT_LOAD_FACTOR); for (i = 0; i < 31; i++) if ((1 << i) >= nbuckets) @@ -1446,9 +1689,6 @@ ip6_full_reass_init_function (vlib_main_t * vm) clib_bihash_init_48_8 (&rm->hash, "ip6-full-reass", nbuckets, nbuckets * 1024); - 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-icmp-error"); ASSERT (node); rm->ip6_icmp_error_idx = node->index; @@ -1456,11 +1696,16 @@ ip6_full_reass_init_function (vlib_main_t * vm) if ((error = vlib_call_init_function (vm, ip_main_init))) return error; ip6_register_protocol (IP_PROTOCOL_IPV6_FRAGMENTATION, - ip6_full_reass_node.index); + ip6_local_full_reass_node.index); + rm->is_local_reass_enabled = 1; rm->fq_index = vlib_frame_queue_main_init (ip6_full_reass_node.index, 0); + rm->fq_local_index = + vlib_frame_queue_main_init (ip6_local_full_reass_node.index, 0); rm->fq_feature_index = vlib_frame_queue_main_init (ip6_full_reass_node_feature.index, 0); + rm->fq_custom_index = + vlib_frame_queue_main_init (ip6_full_reass_node_custom.index, 0); rm->feature_use_refcount_per_intf = NULL; return error; @@ -1504,26 +1749,53 @@ ip6_full_reass_walk_expired (vlib_main_t *vm, vlib_node_runtime_t *node, int index; const uword nthreads = vlib_num_workers () + 1; u32 *vec_icmp_bi = NULL; + u32 n_left_to_next, *to_next; + for (thread_index = 0; thread_index < nthreads; ++thread_index) { ip6_full_reass_per_thread_t *rt = &rm->per_thread_data[thread_index]; + u32 reass_timeout_cnt = 0; clib_spinlock_lock (&rt->lock); vec_reset_length (pool_indexes_to_free); - pool_foreach_index (index, rt->pool) { - reass = pool_elt_at_index (rt->pool, index); - if (now > reass->last_heard + rm->timeout) - { - vec_add1 (pool_indexes_to_free, index); - } - } + /* Pace the number of timeouts handled per thread,to avoid barrier + * sync issues in real world scenarios */ + + u32 beg = rt->last_id; + /* to ensure we walk at least once per sec per context */ + u32 end = beg + (IP6_FULL_REASS_MAX_REASSEMBLIES_DEFAULT * + IP6_FULL_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS / + MSEC_PER_SEC + + 1); + if (end > vec_len (rt->pool)) + { + end = vec_len (rt->pool); + rt->last_id = 0; + } + else + { + rt->last_id = end; + } + + pool_foreach_stepping_index (index, beg, end, rt->pool) + { + reass = pool_elt_at_index (rt->pool, index); + if (now > reass->last_heard + rm->timeout) + { + vec_add1 (pool_indexes_to_free, index); + } + } + int *i; vec_foreach (i, pool_indexes_to_free) { ip6_full_reass_t *reass = pool_elt_at_index (rt->pool, i[0]); u32 icmp_bi = ~0; - ip6_full_reass_on_timeout (vm, node, reass, &icmp_bi); + + reass_timeout_cnt += reass->fragments_n; + ip6_full_reass_on_timeout (vm, node, reass, &icmp_bi, + &n_left_to_next, &to_next); if (~0 != icmp_bi) vec_add1 (vec_icmp_bi, icmp_bi); @@ -1531,6 +1803,10 @@ ip6_full_reass_walk_expired (vlib_main_t *vm, vlib_node_runtime_t *node, } clib_spinlock_unlock (&rt->lock); + if (reass_timeout_cnt) + vlib_node_increment_counter (vm, node->node_index, + IP6_ERROR_REASS_TIMEOUT, + reass_timeout_cnt); } while (vec_len (vec_icmp_bi) > 0) @@ -1546,7 +1822,6 @@ ip6_full_reass_walk_expired (vlib_main_t *vm, vlib_node_runtime_t *node, vlib_buffer_t *b = vlib_get_buffer (vm, bi); if (PREDICT_FALSE (b->flags & VLIB_BUFFER_IS_TRACED)) trace_frame = 1; - b->error = node->errors[IP6_ERROR_REASS_TIMEOUT]; to_next[0] = bi; ++f->n_vectors; to_next += 1; @@ -1560,7 +1835,7 @@ ip6_full_reass_walk_expired (vlib_main_t *vm, vlib_node_runtime_t *node, vec_free (vec_icmp_bi); if (event_data) { - _vec_len (event_data) = 0; + vec_set_len (event_data, 0); } } @@ -1568,14 +1843,13 @@ ip6_full_reass_walk_expired (vlib_main_t *vm, vlib_node_runtime_t *node, } VLIB_REGISTER_NODE (ip6_full_reass_expire_node) = { - .function = ip6_full_reass_walk_expired, - .format_trace = format_ip6_full_reass_trace, - .type = VLIB_NODE_TYPE_PROCESS, - .name = "ip6-full-reassembly-expire-walk", - - .n_errors = ARRAY_LEN (ip6_full_reassembly_error_strings), - .error_strings = ip6_full_reassembly_error_strings, + .function = ip6_full_reass_walk_expired, + .format_trace = format_ip6_full_reass_trace, + .type = VLIB_NODE_TYPE_PROCESS, + .name = "ip6-full-reassembly-expire-walk", + .n_errors = IP6_N_ERROR, + .error_counters = ip6_error_counters, }; static u8 * @@ -1733,9 +2007,10 @@ format_ip6_full_reassembly_handoff_trace (u8 * s, va_list * args) } always_inline uword -ip6_full_reassembly_handoff_inline (vlib_main_t * vm, - vlib_node_runtime_t * node, - vlib_frame_t * frame, bool is_feature) +ip6_full_reassembly_handoff_inline (vlib_main_t *vm, vlib_node_runtime_t *node, + vlib_frame_t *frame, + ip6_full_reass_node_type_t type, + bool is_local) { ip6_full_reass_main_t *rm = &ip6_full_reass_main; @@ -1751,8 +2026,28 @@ ip6_full_reassembly_handoff_inline (vlib_main_t * vm, b = bufs; ti = thread_indices; - fq_index = (is_feature) ? rm->fq_feature_index : rm->fq_index; - + switch (type) + { + case NORMAL: + if (is_local) + { + fq_index = rm->fq_local_index; + } + else + { + fq_index = rm->fq_index; + } + break; + case FEATURE: + fq_index = rm->fq_feature_index; + break; + case CUSTOM: + fq_index = rm->fq_custom_index; + break; + default: + clib_warning ("Unexpected `type' (%d)!", type); + ASSERT (0); + } while (n_left_from > 0) { ti[0] = vnet_buffer (b[0])->ip.reass.owner_thread_index; @@ -1784,8 +2079,8 @@ VLIB_NODE_FN (ip6_full_reassembly_handoff_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip6_full_reassembly_handoff_inline (vm, node, frame, - false /* is_feature */ ); + return ip6_full_reassembly_handoff_inline (vm, node, frame, NORMAL, + false /* is_local */); } VLIB_REGISTER_NODE (ip6_full_reassembly_handoff_node) = { @@ -1802,14 +2097,34 @@ VLIB_REGISTER_NODE (ip6_full_reassembly_handoff_node) = { }, }; +VLIB_NODE_FN (ip6_local_full_reassembly_handoff_node) +(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame) +{ + return ip6_full_reassembly_handoff_inline (vm, node, frame, NORMAL, + true /* is_feature */); +} + +VLIB_REGISTER_NODE (ip6_local_full_reassembly_handoff_node) = { + .name = "ip6-local-full-reassembly-handoff", + .vector_size = sizeof (u32), + .n_errors = ARRAY_LEN(ip6_full_reassembly_handoff_error_strings), + .error_strings = ip6_full_reassembly_handoff_error_strings, + .format_trace = format_ip6_full_reassembly_handoff_trace, + + .n_next_nodes = 1, + + .next_nodes = { + [0] = "error-drop", + }, +}; VLIB_NODE_FN (ip6_full_reassembly_feature_handoff_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip6_full_reassembly_handoff_inline (vm, node, frame, true /* is_feature */ ); + return ip6_full_reassembly_handoff_inline (vm, node, frame, FEATURE, + false /* is_local */); } - VLIB_REGISTER_NODE (ip6_full_reassembly_feature_handoff_node) = { .name = "ip6-full-reass-feature-hoff", .vector_size = sizeof (u32), @@ -1824,6 +2139,27 @@ VLIB_REGISTER_NODE (ip6_full_reassembly_feature_handoff_node) = { }, }; +VLIB_NODE_FN (ip6_full_reassembly_custom_handoff_node) +(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame) +{ + return ip6_full_reassembly_handoff_inline (vm, node, frame, CUSTOM, + false /* is_local */); +} + +VLIB_REGISTER_NODE (ip6_full_reassembly_custom_handoff_node) = { + .name = "ip6-full-reass-custom-hoff", + .vector_size = sizeof (u32), + .n_errors = ARRAY_LEN(ip6_full_reassembly_handoff_error_strings), + .error_strings = ip6_full_reassembly_handoff_error_strings, + .format_trace = format_ip6_full_reassembly_handoff_trace, + + .n_next_nodes = 1, + + .next_nodes = { + [0] = "error-drop", + }, +}; + #ifndef CLIB_MARCH_VARIANT int ip6_full_reass_enable_disable_with_refcnt (u32 sw_if_index, int is_enable) @@ -1849,8 +2185,37 @@ ip6_full_reass_enable_disable_with_refcnt (u32 sw_if_index, int is_enable) "ip6-full-reassembly-feature", sw_if_index, 0, 0, 0); } - return -1; + return 0; +} + +void +ip6_local_full_reass_enable_disable (int enable) +{ + if (enable) + { + if (!ip6_full_reass_main.is_local_reass_enabled) + { + ip6_full_reass_main.is_local_reass_enabled = 1; + ip6_register_protocol (IP_PROTOCOL_IPV6_FRAGMENTATION, + ip6_local_full_reass_node.index); + } + } + else + { + if (ip6_full_reass_main.is_local_reass_enabled) + { + ip6_full_reass_main.is_local_reass_enabled = 0; + ip6_unregister_protocol (IP_PROTOCOL_IPV6_FRAGMENTATION); + } + } +} + +int +ip6_local_full_reass_enabled () +{ + return ip6_full_reass_main.is_local_reass_enabled; } + #endif /* diff --git a/src/vnet/ip/reass/ip6_full_reass.h b/src/vnet/ip/reass/ip6_full_reass.h index 546075b04b4..f66cb67d796 100644 --- a/src/vnet/ip/reass/ip6_full_reass.h +++ b/src/vnet/ip/reass/ip6_full_reass.h @@ -46,6 +46,8 @@ vnet_api_error_t ip6_full_reass_enable_disable (u32 sw_if_index, int ip6_full_reass_enable_disable_with_refcnt (u32 sw_if_index, int is_enable); +void ip6_local_full_reass_enable_disable (int enable); +int ip6_local_full_reass_enabled (); #endif /* __included_ip6_full_reass_h */ /* diff --git a/src/vnet/ip/reass/ip6_sv_reass.c b/src/vnet/ip/reass/ip6_sv_reass.c index 28941311f50..fe2ed05555c 100644 --- a/src/vnet/ip/reass/ip6_sv_reass.c +++ b/src/vnet/ip/reass/ip6_sv_reass.c @@ -26,6 +26,7 @@ #include <vnet/ip/ip6_to_ip4.h> #include <vppinfra/bihash_48_8.h> #include <vnet/ip/reass/ip6_sv_reass.h> +#include <vnet/ip/ip6_inlines.h> #define MSEC_PER_SEC 1000 #define IP6_SV_REASS_TIMEOUT_DEFAULT_MS 100 @@ -40,6 +41,7 @@ typedef enum IP6_SV_REASS_RC_TOO_MANY_FRAGMENTS, IP6_SV_REASS_RC_INTERNAL_ERROR, IP6_SV_REASS_RC_UNSUPP_IP_PROTO, + IP6_SV_REASS_RC_INVALID_FRAG_LEN, } ip6_sv_reass_rc_t; typedef struct @@ -50,7 +52,7 @@ typedef struct { ip6_address_t src; ip6_address_t dst; - u32 xx_id; + u32 fib_index; u32 frag_id; u8 unused[7]; u8 proto; @@ -148,6 +150,7 @@ typedef struct /** Worker handoff */ u32 fq_index; u32 fq_feature_index; + u32 fq_custom_context_index; // reference count for enabling/disabling feature - per interface u32 *feature_use_refcount_per_intf; @@ -214,7 +217,7 @@ format_ip6_sv_reass_trace (u8 * s, va_list * args) clib_net_to_host_u16 (t->l4_dst_port)); break; case REASS_PASSTHROUGH: - s = format (s, "[not-fragmented]"); + s = format (s, "[not fragmented or atomic fragment]"); break; } return s; @@ -309,6 +312,8 @@ ip6_sv_reass_find_or_create (vlib_main_t *vm, ip6_sv_reass_main_t *rm, ip6_sv_reass_t *reass = NULL; f64 now = vlib_time_now (vm); +again: + if (!clib_bihash_search_48_8 (&rm->hash, &kv->kv, &kv->kv)) { if (vm->thread_index != kv->v.thread_index) @@ -368,10 +373,14 @@ ip6_sv_reass_find_or_create (vlib_main_t *vm, ip6_sv_reass_main_t *rm, kv->v.thread_index = vm->thread_index; reass->last_heard = now; - if (clib_bihash_add_del_48_8 (&rm->hash, &kv->kv, 1)) + int rv = clib_bihash_add_del_48_8 (&rm->hash, &kv->kv, 2); + if (rv) { ip6_sv_reass_free (vm, rm, rt, reass); reass = NULL; + // if other worker created a context already work with the other copy + if (-2 == rv) + goto again; } return reass; @@ -399,6 +408,10 @@ ip6_sv_reass_update (vlib_main_t *vm, vlib_node_runtime_t *node, u32 fragment_length = vlib_buffer_length_in_chain (vm, fb) - (fvnb->ip.reass.ip6_frag_hdr_offset + sizeof (*frag_hdr)); + if (0 == fragment_length) + { + return IP6_SV_REASS_RC_INVALID_FRAG_LEN; + } u32 fragment_last = fvnb->ip.reass.fragment_last = fragment_first + fragment_length - 1; fvnb->ip.reass.range_first = fragment_first; @@ -440,22 +453,18 @@ ip6_sv_reass_update (vlib_main_t *vm, vlib_node_runtime_t *node, } always_inline bool -ip6_sv_reass_verify_upper_layer_present (vlib_node_runtime_t * node, - vlib_buffer_t * b, - ip6_frag_hdr_t * frag_hdr) +ip6_sv_reass_verify_upper_layer_present (vlib_node_runtime_t *node, + vlib_buffer_t *b, + ip6_ext_hdr_chain_t *hc) { - ip6_ext_header_t *tmp = (ip6_ext_header_t *) frag_hdr; - while (ip6_ext_hdr (tmp->next_hdr)) + int nh = hc->eh[hc->length - 1].protocol; + /* Checking to see if it's a terminating header */ + if (ip6_ext_hdr (nh)) { - tmp = ip6_ext_next_header (tmp); - } - if (IP_PROTOCOL_IP6_NONXT == tmp->next_hdr) - { - icmp6_error_set_vnet_buffer (b, ICMP6_parameter_problem, - ICMP6_parameter_problem_first_fragment_has_incomplete_header_chain, - 0); + icmp6_error_set_vnet_buffer ( + b, ICMP6_parameter_problem, + ICMP6_parameter_problem_first_fragment_has_incomplete_header_chain, 0); b->error = node->errors[IP6_ERROR_REASS_MISSING_UPPER]; - return false; } return true; @@ -505,14 +514,18 @@ ip6_sv_reass_verify_packet_size_lt_64k (vlib_main_t * vm, } always_inline uword -ip6_sv_reassembly_inline (vlib_main_t * vm, - vlib_node_runtime_t * node, - vlib_frame_t * frame, bool is_feature) +ip6_sv_reassembly_inline (vlib_main_t *vm, vlib_node_runtime_t *node, + vlib_frame_t *frame, bool is_feature, + bool custom_next, bool custom_context) { u32 *from = vlib_frame_vector_args (frame); - u32 n_left_from, n_left_to_next, *to_next, next_index; + u32 n_left_from, n_left_to_next, *to_next, *to_next_aux, next_index; ip6_sv_reass_main_t *rm = &ip6_sv_reass_main; ip6_sv_reass_per_thread_t *rt = &rm->per_thread_data[vm->thread_index]; + u32 *context; + if (custom_context) + context = vlib_frame_aux_args (frame); + clib_spinlock_lock (&rt->lock); n_left_from = frame->n_vectors; @@ -520,7 +533,11 @@ ip6_sv_reassembly_inline (vlib_main_t * vm, while (n_left_from > 0) { - vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + if (custom_context) + vlib_get_next_frame_with_aux_safe (vm, node, next_index, to_next, + to_next_aux, n_left_to_next); + else + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); while (n_left_from > 0 && n_left_to_next > 0) { @@ -528,23 +545,31 @@ ip6_sv_reassembly_inline (vlib_main_t * vm, vlib_buffer_t *b0; u32 next0 = IP6_SV_REASSEMBLY_NEXT_DROP; u32 error0 = IP6_ERROR_NONE; - + u8 forward_context = 0; bi0 = from[0]; b0 = vlib_get_buffer (vm, bi0); ip6_header_t *ip0 = vlib_buffer_get_current (b0); - ip6_frag_hdr_t *frag_hdr = NULL; - ip6_ext_header_t *prev_hdr; - if (ip6_ext_hdr (ip0->protocol)) + ip6_frag_hdr_t *frag_hdr; + ip6_ext_hdr_chain_t hdr_chain; + bool is_atomic_fragment = false; + + int res = ip6_ext_header_walk ( + b0, ip0, IP_PROTOCOL_IPV6_FRAGMENTATION, &hdr_chain); + if (res >= 0 && + hdr_chain.eh[res].protocol == IP_PROTOCOL_IPV6_FRAGMENTATION) { frag_hdr = - ip6_ext_header_find (vm, b0, ip0, - IP_PROTOCOL_IPV6_FRAGMENTATION, - &prev_hdr); + ip6_ext_next_header_offset (ip0, hdr_chain.eh[res].offset); + is_atomic_fragment = (0 == ip6_frag_hdr_offset (frag_hdr) && + !ip6_frag_hdr_more (frag_hdr)); } - if (!frag_hdr) + + if (res < 0 || + hdr_chain.eh[res].protocol != IP_PROTOCOL_IPV6_FRAGMENTATION || + is_atomic_fragment) { - // this is a regular packet - no fragmentation + // this is a regular unfragmented packet or an atomic fragment if (!ip6_get_port (vm, b0, ip0, b0->current_length, &(vnet_buffer (b0)->ip.reass.ip_proto), @@ -560,7 +585,8 @@ ip6_sv_reassembly_inline (vlib_main_t * vm, goto packet_enqueue; } vnet_buffer (b0)->ip.reass.is_non_first_fragment = 0; - next0 = IP6_SV_REASSEMBLY_NEXT_INPUT; + next0 = custom_next ? vnet_buffer (b0)->ip.reass.next_index : + IP6_SV_REASSEMBLY_NEXT_INPUT; if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) { ip6_sv_reass_add_trace ( @@ -571,13 +597,15 @@ ip6_sv_reassembly_inline (vlib_main_t * vm, } goto packet_enqueue; } + vnet_buffer (b0)->ip.reass.ip6_frag_hdr_offset = - (u8 *) frag_hdr - (u8 *) ip0; + hdr_chain.eh[res].offset; + if (0 == ip6_frag_hdr_offset (frag_hdr)) { // first fragment - verify upper-layer is present - if (!ip6_sv_reass_verify_upper_layer_present - (node, b0, frag_hdr)) + if (!ip6_sv_reass_verify_upper_layer_present (node, b0, + &hdr_chain)) { next0 = IP6_SV_REASSEMBLY_NEXT_ICMP_ERROR; goto packet_enqueue; @@ -597,10 +625,15 @@ ip6_sv_reassembly_inline (vlib_main_t * vm, kv.k.as_u64[1] = ip0->src_address.as_u64[1]; kv.k.as_u64[2] = ip0->dst_address.as_u64[0]; kv.k.as_u64[3] = ip0->dst_address.as_u64[1]; - kv.k.as_u64[4] = - ((u64) vec_elt (ip6_main.fib_index_by_sw_if_index, - vnet_buffer (b0)->sw_if_index[VLIB_RX])) << 32 | - (u64) frag_hdr->identification; + if (custom_context) + kv.k.as_u64[4] = + (u64) *context << 32 | (u64) frag_hdr->identification; + else + kv.k.as_u64[4] = + ((u64) vec_elt (ip6_main.fib_index_by_sw_if_index, + vnet_buffer (b0)->sw_if_index[VLIB_RX])) + << 32 | + (u64) frag_hdr->identification; kv.k.as_u64[5] = ip0->protocol; ip6_sv_reass_t *reass = @@ -611,6 +644,8 @@ ip6_sv_reassembly_inline (vlib_main_t * vm, next0 = IP6_SV_REASSEMBLY_NEXT_HANDOFF; vnet_buffer (b0)->ip.reass.owner_thread_index = kv.v.thread_index; + if (custom_context) + forward_context = 1; goto packet_enqueue; } @@ -635,7 +670,8 @@ ip6_sv_reassembly_inline (vlib_main_t * vm, reass->tcp_seq_number; vnet_buffer (b0)->ip.reass.l4_src_port = reass->l4_src_port; vnet_buffer (b0)->ip.reass.l4_dst_port = reass->l4_dst_port; - next0 = IP6_SV_REASSEMBLY_NEXT_INPUT; + next0 = custom_next ? vnet_buffer (b0)->ip.reass.next_index : + IP6_SV_REASSEMBLY_NEXT_INPUT; if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) { ip6_sv_reass_add_trace ( @@ -645,31 +681,30 @@ ip6_sv_reassembly_inline (vlib_main_t * vm, goto packet_enqueue; } + u32 counter = ~0; switch (ip6_sv_reass_update (vm, node, rm, reass, bi0, frag_hdr)) { case IP6_SV_REASS_RC_OK: /* nothing to do here */ break; case IP6_SV_REASS_RC_TOO_MANY_FRAGMENTS: - vlib_node_increment_counter (vm, node->node_index, - IP6_ERROR_REASS_FRAGMENT_CHAIN_TOO_LONG, - 1); - ip6_sv_reass_free (vm, rm, rt, reass); - goto next_packet; + counter = IP6_ERROR_REASS_FRAGMENT_CHAIN_TOO_LONG; break; case IP6_SV_REASS_RC_UNSUPP_IP_PROTO: - vlib_node_increment_counter (vm, node->node_index, - IP6_ERROR_REASS_UNSUPP_IP_PROTO, - 1); - ip6_sv_reass_free (vm, rm, rt, reass); - goto next_packet; + counter = IP6_ERROR_REASS_UNSUPP_IP_PROTO; break; case IP6_SV_REASS_RC_INTERNAL_ERROR: - vlib_node_increment_counter (vm, node->node_index, - IP6_ERROR_REASS_INTERNAL_ERROR, 1); + counter = IP6_ERROR_REASS_INTERNAL_ERROR; + break; + case IP6_SV_REASS_RC_INVALID_FRAG_LEN: + counter = IP6_ERROR_REASS_INVALID_FRAG_LEN; + break; + } + if (~0 != counter) + { + vlib_node_increment_counter (vm, node->node_index, counter, 1); ip6_sv_reass_free (vm, rm, rt, reass); goto next_packet; - break; } if (reass->is_complete) @@ -717,7 +752,8 @@ ip6_sv_reassembly_inline (vlib_main_t * vm, to_next, n_left_to_next, bi0, next0); } - _vec_len (reass->cached_buffers) = 0; // buffers are owned by frame now + vec_set_len (reass->cached_buffers, + 0); // buffers are owned by frame now } goto next_packet; @@ -730,11 +766,25 @@ ip6_sv_reassembly_inline (vlib_main_t * vm, b0 = vlib_get_buffer (vm, bi0); vnet_feature_next (&next0, b0); } - vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, - n_left_to_next, bi0, next0); + if (custom_context && forward_context) + { + if (to_next_aux) + { + to_next_aux[0] = *context; + to_next_aux += 1; + } + vlib_validate_buffer_enqueue_with_aux_x1 ( + vm, node, next_index, to_next, to_next_aux, n_left_to_next, + bi0, *context, next0); + } + else + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); next_packet: from += 1; + if (custom_context) + context += 1; n_left_from -= 1; } @@ -745,26 +795,21 @@ ip6_sv_reassembly_inline (vlib_main_t * vm, return frame->n_vectors; } -static char *ip6_sv_reassembly_error_strings[] = { -#define _(sym, string) string, - foreach_ip6_error -#undef _ -}; - VLIB_NODE_FN (ip6_sv_reass_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip6_sv_reassembly_inline (vm, node, frame, false /* is_feature */ ); + return ip6_sv_reassembly_inline (vm, node, frame, false /* is_feature */, + false /* custom next */, + false /* custom context */); } -/* *INDENT-OFF* */ VLIB_REGISTER_NODE (ip6_sv_reass_node) = { .name = "ip6-sv-reassembly", .vector_size = sizeof (u32), .format_trace = format_ip6_sv_reass_trace, - .n_errors = ARRAY_LEN (ip6_sv_reassembly_error_strings), - .error_strings = ip6_sv_reassembly_error_strings, + .n_errors = IP6_N_ERROR, + .error_counters = ip6_error_counters, .n_next_nodes = IP6_SV_REASSEMBLY_N_NEXT, .next_nodes = { @@ -774,22 +819,22 @@ VLIB_REGISTER_NODE (ip6_sv_reass_node) = { [IP6_SV_REASSEMBLY_NEXT_HANDOFF] = "ip6-sv-reassembly-handoff", }, }; -/* *INDENT-ON* */ VLIB_NODE_FN (ip6_sv_reass_node_feature) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip6_sv_reassembly_inline (vm, node, frame, true /* is_feature */ ); + return ip6_sv_reassembly_inline (vm, node, frame, true /* is_feature */, + false /* custom next */, + false /* custom context */); } -/* *INDENT-OFF* */ VLIB_REGISTER_NODE (ip6_sv_reass_node_feature) = { .name = "ip6-sv-reassembly-feature", .vector_size = sizeof (u32), .format_trace = format_ip6_sv_reass_trace, - .n_errors = ARRAY_LEN (ip6_sv_reassembly_error_strings), - .error_strings = ip6_sv_reassembly_error_strings, + .n_errors = IP6_N_ERROR, + .error_counters = ip6_error_counters, .n_next_nodes = IP6_SV_REASSEMBLY_N_NEXT, .next_nodes = { @@ -799,16 +844,38 @@ VLIB_REGISTER_NODE (ip6_sv_reass_node_feature) = { [IP6_SV_REASSEMBLY_NEXT_HANDOFF] = "ip6-sv-reass-feature-hoff", }, }; -/* *INDENT-ON* */ -/* *INDENT-OFF* */ VNET_FEATURE_INIT (ip6_sv_reassembly_feature) = { .arc_name = "ip6-unicast", .node_name = "ip6-sv-reassembly-feature", .runs_before = VNET_FEATURES ("ip6-lookup"), .runs_after = 0, }; -/* *INDENT-ON* */ + +VLIB_NODE_FN (ip6_sv_reass_custom_context_node) +(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame) +{ + return ip6_sv_reassembly_inline (vm, node, frame, false /* is_feature */, + true /* custom next */, + true /* custom context */); +} + +VLIB_REGISTER_NODE (ip6_sv_reass_custom_context_node) = { + .name = "ip6-sv-reassembly-custom-context", + .vector_size = sizeof (u32), + .aux_size = sizeof (u32), + .format_trace = format_ip6_sv_reass_trace, + .n_errors = IP6_N_ERROR, + .error_counters = ip6_error_counters, + .n_next_nodes = IP6_SV_REASSEMBLY_N_NEXT, + .next_nodes = + { + [IP6_SV_REASSEMBLY_NEXT_INPUT] = "ip6-input", + [IP6_SV_REASSEMBLY_NEXT_DROP] = "ip6-drop", + [IP6_SV_REASSEMBLY_NEXT_ICMP_ERROR] = "ip6-icmp-error", + [IP6_SV_REASSEMBLY_NEXT_HANDOFF] = "ip6-sv-reassembly-custom-context-handoff", + }, +}; #ifndef CLIB_MARCH_VARIANT static u32 @@ -959,6 +1026,8 @@ ip6_sv_reass_init_function (vlib_main_t * vm) rm->fq_index = vlib_frame_queue_main_init (ip6_sv_reass_node.index, 0); rm->fq_feature_index = vlib_frame_queue_main_init (ip6_sv_reass_node_feature.index, 0); + rm->fq_custom_context_index = + vlib_frame_queue_main_init (ip6_sv_reass_custom_context_node.index, 0); rm->feature_use_refcount_per_intf = NULL; @@ -1009,7 +1078,6 @@ ip6_sv_reass_walk_expired (vlib_main_t *vm, clib_spinlock_lock (&rt->lock); vec_reset_length (pool_indexes_to_free); - /* *INDENT-OFF* */ pool_foreach_index (index, rt->pool) { reass = pool_elt_at_index (rt->pool, index); if (now > reass->last_heard + rm->timeout) @@ -1017,15 +1085,12 @@ ip6_sv_reass_walk_expired (vlib_main_t *vm, vec_add1 (pool_indexes_to_free, index); } } - /* *INDENT-ON* */ int *i; - /* *INDENT-OFF* */ vec_foreach (i, pool_indexes_to_free) { ip6_sv_reass_t *reass = pool_elt_at_index (rt->pool, i[0]); ip6_sv_reass_free (vm, rm, rt, reass); } - /* *INDENT-ON* */ clib_spinlock_unlock (&rt->lock); } @@ -1033,33 +1098,31 @@ ip6_sv_reass_walk_expired (vlib_main_t *vm, vec_free (pool_indexes_to_free); if (event_data) { - _vec_len (event_data) = 0; + vec_set_len (event_data, 0); } } return 0; } -/* *INDENT-OFF* */ VLIB_REGISTER_NODE (ip6_sv_reass_expire_node) = { - .function = ip6_sv_reass_walk_expired, - .format_trace = format_ip6_sv_reass_trace, - .type = VLIB_NODE_TYPE_PROCESS, - .name = "ip6-sv-reassembly-expire-walk", - - .n_errors = ARRAY_LEN (ip6_sv_reassembly_error_strings), - .error_strings = ip6_sv_reassembly_error_strings, + .function = ip6_sv_reass_walk_expired, + .format_trace = format_ip6_sv_reass_trace, + .type = VLIB_NODE_TYPE_PROCESS, + .name = "ip6-sv-reassembly-expire-walk", + .n_errors = IP6_N_ERROR, + .error_counters = ip6_error_counters, }; -/* *INDENT-ON* */ static u8 * format_ip6_sv_reass_key (u8 * s, va_list * args) { ip6_sv_reass_key_t *key = va_arg (*args, ip6_sv_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); + s = + format (s, "fib_index: %u, src: %U, dst: %U, frag_id: %u, proto: %u", + key->fib_index, format_ip6_address, &key->src, format_ip6_address, + &key->dst, clib_net_to_host_u16 (key->frag_id), key->proto); return s; } @@ -1116,11 +1179,9 @@ show_ip6_sv_reass (vlib_main_t * vm, unformat_input_t * input, clib_spinlock_lock (&rt->lock); if (details) { - /* *INDENT-OFF* */ pool_foreach (reass, rt->pool) { vlib_cli_output (vm, "%U", format_ip6_sv_reass, vm, reass); } - /* *INDENT-ON* */ } sum_reass_n += rt->reass_n; clib_spinlock_unlock (&rt->lock); @@ -1146,13 +1207,11 @@ show_ip6_sv_reass (vlib_main_t * vm, unformat_input_t * input, return 0; } -/* *INDENT-OFF* */ VLIB_CLI_COMMAND (show_ip6_sv_reassembly_cmd, static) = { .path = "show ip6-sv-reassembly", .short_help = "show ip6-sv-reassembly [details]", .function = show_ip6_sv_reass, }; -/* *INDENT-ON* */ #ifndef CLIB_MARCH_VARIANT vnet_api_error_t @@ -1202,25 +1261,29 @@ format_ip6_sv_reassembly_handoff_trace (u8 * s, va_list * args) } always_inline uword -ip6_sv_reassembly_handoff_inline (vlib_main_t * vm, - vlib_node_runtime_t * node, - vlib_frame_t * frame, bool is_feature) +ip6_sv_reassembly_handoff_inline (vlib_main_t *vm, vlib_node_runtime_t *node, + vlib_frame_t *frame, bool is_feature, + bool custom_context) { ip6_sv_reass_main_t *rm = &ip6_sv_reass_main; vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b; - u32 n_enq, n_left_from, *from; + u32 n_enq, n_left_from, *from, *context; u16 thread_indices[VLIB_FRAME_SIZE], *ti; u32 fq_index; from = vlib_frame_vector_args (frame); + if (custom_context) + context = vlib_frame_aux_args (frame); n_left_from = frame->n_vectors; vlib_get_buffers (vm, from, bufs, n_left_from); b = bufs; ti = thread_indices; - fq_index = (is_feature) ? rm->fq_feature_index : rm->fq_index; + fq_index = (is_feature) ? + rm->fq_feature_index : + (custom_context ? rm->fq_custom_context_index : rm->fq_index); while (n_left_from > 0) { @@ -1239,8 +1302,12 @@ ip6_sv_reassembly_handoff_inline (vlib_main_t * vm, ti += 1; b += 1; } - n_enq = vlib_buffer_enqueue_to_thread (vm, node, fq_index, from, - thread_indices, frame->n_vectors, 1); + if (custom_context) + n_enq = vlib_buffer_enqueue_to_thread_with_aux ( + vm, node, fq_index, from, context, thread_indices, frame->n_vectors, 1); + else + n_enq = vlib_buffer_enqueue_to_thread ( + vm, node, fq_index, from, thread_indices, frame->n_vectors, 1); if (n_enq < frame->n_vectors) vlib_node_increment_counter (vm, node->node_index, @@ -1253,11 +1320,10 @@ VLIB_NODE_FN (ip6_sv_reassembly_handoff_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip6_sv_reassembly_handoff_inline (vm, node, frame, - false /* is_feature */ ); + return ip6_sv_reassembly_handoff_inline ( + vm, node, frame, false /* is_feature */, false /* custom_context */); } -/* *INDENT-OFF* */ VLIB_REGISTER_NODE (ip6_sv_reassembly_handoff_node) = { .name = "ip6-sv-reassembly-handoff", .vector_size = sizeof (u32), @@ -1276,11 +1342,11 @@ VLIB_REGISTER_NODE (ip6_sv_reassembly_handoff_node) = { VLIB_NODE_FN (ip6_sv_reassembly_feature_handoff_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame) { - return ip6_sv_reassembly_handoff_inline (vm, node, frame, true /* is_feature */ ); + return ip6_sv_reassembly_handoff_inline ( + vm, node, frame, true /* is_feature */, false /* custom_context */); } -/* *INDENT-OFF* */ VLIB_REGISTER_NODE (ip6_sv_reassembly_feature_handoff_node) = { .name = "ip6-sv-reass-feature-hoff", .vector_size = sizeof (u32), @@ -1294,7 +1360,28 @@ VLIB_REGISTER_NODE (ip6_sv_reassembly_feature_handoff_node) = { [0] = "error-drop", }, }; -/* *INDENT-ON* */ + +VLIB_NODE_FN (ip6_sv_reassembly_custom_context_handoff_node) +(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame) +{ + return ip6_sv_reassembly_handoff_inline ( + vm, node, frame, false /* is_feature */, true /* custom_context */); +} + +VLIB_REGISTER_NODE (ip6_sv_reassembly_custom_context_handoff_node) = { + .name = "ip6-sv-reassembly-custom-context-handoff", + .vector_size = sizeof (u32), + .aux_size = sizeof (u32), + .n_errors = ARRAY_LEN(ip6_sv_reassembly_handoff_error_strings), + .error_strings = ip6_sv_reassembly_handoff_error_strings, + .format_trace = format_ip6_sv_reassembly_handoff_trace, + + .n_next_nodes = 1, + + .next_nodes = { + [0] = "error-drop", + }, +}; #ifndef CLIB_MARCH_VARIANT int @@ -1323,6 +1410,14 @@ ip6_sv_reass_enable_disable_with_refcnt (u32 sw_if_index, int is_enable) } return 0; } + +uword +ip6_sv_reass_custom_context_register_next_node (uword node_index) +{ + return vlib_node_add_next ( + vlib_get_main (), ip6_sv_reassembly_custom_context_handoff_node.index, + node_index); +} #endif /* diff --git a/src/vnet/ip/reass/ip6_sv_reass.h b/src/vnet/ip/reass/ip6_sv_reass.h index 81ac2312bdf..7dc9df132dd 100644 --- a/src/vnet/ip/reass/ip6_sv_reass.h +++ b/src/vnet/ip/reass/ip6_sv_reass.h @@ -44,6 +44,7 @@ vnet_api_error_t ip6_sv_reass_enable_disable (u32 sw_if_index, u8 enable_disable); int ip6_sv_reass_enable_disable_with_refcnt (u32 sw_if_index, int is_enable); +uword ip6_sv_reass_custom_context_register_next_node (uword node_index); #endif /* __included_ip6_sv_reass_h */ diff --git a/src/vnet/ip/reass/reassembly.rst b/src/vnet/ip/reass/reassembly.rst new file mode 100644 index 00000000000..49e0a8de6e6 --- /dev/null +++ b/src/vnet/ip/reass/reassembly.rst @@ -0,0 +1,221 @@ +.. _reassembly: + +IP Reassembly +============= + +Some VPP functions need access to whole packet and/or stream +classification based on L4 headers. Reassembly functionality allows +both former and latter. + +Full reassembly vs shallow (virtual) reassembly +----------------------------------------------- + +There are two kinds of reassembly available in VPP: + +1. Full reassembly changes a stream of packet fragments into one +packet containing all data reassembled with fragment bits cleared +and fragment header stripped (in case of ip6). Note that resulting +packet may come out of reassembly as a buffer chain. Because it's +impractical to parse headers which are split over multiple vnet +buffers, vnet_buffer_chain_linearize() is called after reassembly so +that L2/L3/L4 headers can be found in first buffer. Full reassembly +is costly and shouldn't be used unless necessary. Full reassembly is by +default enabled for both ipv4 and ipv6 "for us" traffic +- that is packets aimed at VPP addresses. This can be disabled via API +if desired, in which case "for us" fragments are dropped. + +2. Shallow (virtual) reassembly allows various classifying and/or +translating features to work with fragments without having to +understand fragmentation. It works by extracting L4 data and adding +them to vnet_buffer for each packet/fragment passing through SVR +nodes. This operation is performed for both fragments and regular +packets, allowing consuming code to treat all packets in same way. SVR +caches incoming packet fragments (buffers) until first fragment is +seen. Then it extracts L4 data from that first fragment, fills it for +any cached fragments and transmits them in the same order as they were +received. From that point on, any other passing fragments get L4 data +populated in vnet_buffer based on reassembly context. + +Multi-worker behaviour +^^^^^^^^^^^^^^^^^^^^^^ + +Both reassembly types deal with fragments arriving on different workers +via handoff mechanism. All reassembly contexts are stored in pools. +Bihash mapping 5-tuple key to a value containing pool index and thread +index is used for lookups. When a lookup finds an existing reassembly on +a different thread, it hands off the fragment to that thread. If lookup +fails, a new reassembly context is created and current worker becomes +owner of that context. Further fragments received on other worker +threads are then handed off owner worker thread. + +Full reassembly also remembers thread index where first fragment (as in +fragment with fragment offset 0) was seen and uses handoff mechanism to +send the reassembled packet out on that thread even if pool owner is +a different thread. This then requires an additional handoff to free +reassembly context as only pool owner can do that in a thread-safe way. + +Limits +^^^^^^ + +Because reassembly could be an attack vector, there is a configurable +limit on the number of concurrent reassemblies and also maximum +fragments per packet. + +Custom applications +^^^^^^^^^^^^^^^^^^^ + +Both reassembly features allow to be used by custom application which +are not part of VPP source tree. Be it patches or 3rd party plugins, +they can build their own graph paths by using "-custom*" versions of +nodes. Reassembly then reads next_index and error_next_index for each +buffer from vnet_buffer, allowing custom application to steer +both reassembled packets and any packets which are considered an error +in a way the custom application requires. + +Full reassembly +--------------- + +Configuration +^^^^^^^^^^^^^ + +Configuration is via API (``ip_reassembly_enable_disable``) or CLI: + +``set interface reassembly <interface-name> [on|off|ip4|ip6]`` + +here ``on`` means both ip4 and ip6. + +A show command is provided to see reassembly contexts: + +For ip4: + +``show ip4-full-reassembly [details]`` + +For ip6: + +``show ip6-full-reassembly [details]`` + +Global full reassembly parameters can be modified using API +``ip_reassembly_set`` and retrieved using ``ip_reassembly_get``. + +Defaults +"""""""" + +For defaults values, see #defines in + +`ip4_full_reass.c <__REPOSITORY_URL__/src/vnet/ip/reass/ip4_full_reass.c>`_ + +========================================= ========================================== +#define description +----------------------------------------- ------------------------------------------ +IP4_REASS_TIMEOUT_DEFAULT_MS timeout in milliseconds +IP4_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS interval between reaping expired sessions +IP4_REASS_MAX_REASSEMBLIES_DEFAULT maximum number of concurrent reassemblies +IP4_REASS_MAX_REASSEMBLY_LENGTH_DEFAULT maximum number of fragments per reassembly +========================================= ========================================== + +and + +`ip6_full_reass.c <__REPOSITORY_URL__/src/vnet/ip/reass/ip6_full_reass.c>`_ + +========================================= ========================================== +#define description +----------------------------------------- ------------------------------------------ +IP6_REASS_TIMEOUT_DEFAULT_MS timeout in milliseconds +IP6_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS interval between reaping expired sessions +IP6_REASS_MAX_REASSEMBLIES_DEFAULT maximum number of concurrent reassemblies +IP6_REASS_MAX_REASSEMBLY_LENGTH_DEFAULT maximum number of fragments per reassembly +========================================= ========================================== + +Finished/expired contexts +^^^^^^^^^^^^^^^^^^^^^^^^^ + +Reassembly contexts are freed either when reassembly is finished - when +all data has been received or in case of timeout. There is a process +walking all reassemblies, freeing any expired ones. + +Shallow (virtual) reassembly +---------------------------- + +Configuration +^^^^^^^^^^^^^ + +Configuration is via API (``ip_reassembly_enable_disable``) only as +there is no value in turning SVR on by hand without a feature consuming +buffer metadata. SVR is designed to be turned on by a feature requiring +it in a programmatic way. + +A show command is provided to see reassembly contexts: + +For ip4: + +``show ip4-sv-reassembly [details]`` + +For ip6: + +``show ip6-sv-reassembly [details]`` + +Global shallow reassembly parameters can be modified using API +``ip_reassembly_set`` and retrieved using ``ip_reassembly_get``. + +Defaults +"""""""" + +For defaults values, see #defines in + +`ip4_sv_reass.c <__REPOSITORY_URL__/src/vnet/ip/reass/ip4_sv_reass.c>`_ + +============================================ ========================================== +#define description +-------------------------------------------- ------------------------------------------ +IP4_SV_REASS_TIMEOUT_DEFAULT_MS timeout in milliseconds +IP4_SV_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS interval between reaping expired sessions +IP4_SV_REASS_MAX_REASSEMBLIES_DEFAULT maximum number of concurrent reassemblies +IP4_SV_REASS_MAX_REASSEMBLY_LENGTH_DEFAULT maximum number of fragments per reassembly +============================================ ========================================== + +and + +`ip6_sv_reass.c <__REPOSITORY_URL__/src/vnet/ip/reass/ip6_sv_reass.c>`_ + +============================================ ========================================== +#define description +-------------------------------------------- ------------------------------------------ +IP6_SV_REASS_TIMEOUT_DEFAULT_MS timeout in milliseconds +IP6_SV_REASS_EXPIRE_WALK_INTERVAL_DEFAULT_MS interval between reaping expired sessions +IP6_SV_REASS_MAX_REASSEMBLIES_DEFAULT maximum number of concurrent reassemblies +IP6_SV_REASS_MAX_REASSEMBLY_LENGTH_DEFAULT maximum number of fragments per reassembly +============================================ ========================================== + +Expiring contexts +^^^^^^^^^^^^^^^^^ + +There is no way of knowing when a reassembly is finished without +performing (an almost) full reassembly, so contexts in SVR cannot be +freed in the same way as in full reassembly. Instead a different +approach is taken. Least recently used (LRU) list is maintained where +reassembly contexts are ordered based on last update. The oldest +context is then freed whenever SVR hits limit on number of concurrent +reassembly contexts. There is also a process reaping expired sessions +similar as in full reassembly. + +Truncated packets +^^^^^^^^^^^^^^^^^ + +When SVR detects that a packet has been truncated in a way where L4 +headers are not available, it will mark it as such in vnet_buffer, +allowing downstream features to handle such packets as they deem fit. + +Fast path/slow path +^^^^^^^^^^^^^^^^^^^ + +SVR runs is implemented fast path/slow path way. By default, it assumes +that any passing traffic doesn't contain fragments, processing buffers +in a dual-loop. If it sees a fragment, it then jumps to single-loop +processing. + +Feature enabled by other features/reference counting +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + +SVR feature is enabled by some other features, like NAT, when those +features are enabled. For this to work, it implements a reference +counted API for enabling/disabling SVR. |