diff options
Diffstat (limited to 'src/vnet/tcp/tcp_input.c')
-rwxr-xr-x | src/vnet/tcp/tcp_input.c | 864 |
1 files changed, 1 insertions, 863 deletions
diff --git a/src/vnet/tcp/tcp_input.c b/src/vnet/tcp/tcp_input.c index 86158b9d6c6..e27cffb9444 100755 --- a/src/vnet/tcp/tcp_input.c +++ b/src/vnet/tcp/tcp_input.c @@ -16,8 +16,8 @@ #include <vppinfra/sparse_vec.h> #include <vnet/fib/ip4_fib.h> #include <vnet/fib/ip6_fib.h> -#include <vnet/tcp/tcp_packet.h> #include <vnet/tcp/tcp.h> +#include <vnet/tcp/tcp_inlines.h> #include <vnet/session/session.h> #include <math.h> @@ -116,119 +116,6 @@ tcp_segment_in_rcv_wnd (tcp_connection_t * tc, u32 seq, u32 end_seq) } /** - * Parse TCP header options. - * - * @param th TCP header - * @param to TCP options data structure to be populated - * @param is_syn set if packet is syn - * @return -1 if parsing failed - */ -static inline int -tcp_options_parse (tcp_header_t * th, tcp_options_t * to, u8 is_syn) -{ - const u8 *data; - u8 opt_len, opts_len, kind; - int j; - sack_block_t b; - - opts_len = (tcp_doff (th) << 2) - sizeof (tcp_header_t); - data = (const u8 *) (th + 1); - - /* Zero out all flags but those set in SYN */ - to->flags &= (TCP_OPTS_FLAG_SACK_PERMITTED | TCP_OPTS_FLAG_WSCALE - | TCP_OPTS_FLAG_TSTAMP | TCP_OPTS_FLAG_MSS); - - for (; opts_len > 0; opts_len -= opt_len, data += opt_len) - { - kind = data[0]; - - /* Get options length */ - if (kind == TCP_OPTION_EOL) - break; - else if (kind == TCP_OPTION_NOOP) - { - opt_len = 1; - continue; - } - else - { - /* broken options */ - if (opts_len < 2) - return -1; - opt_len = data[1]; - - /* weird option length */ - if (opt_len < 2 || opt_len > opts_len) - return -1; - } - - /* Parse options */ - switch (kind) - { - case TCP_OPTION_MSS: - if (!is_syn) - break; - if ((opt_len == TCP_OPTION_LEN_MSS) && tcp_syn (th)) - { - to->flags |= TCP_OPTS_FLAG_MSS; - to->mss = clib_net_to_host_u16 (*(u16 *) (data + 2)); - } - break; - case TCP_OPTION_WINDOW_SCALE: - if (!is_syn) - break; - if ((opt_len == TCP_OPTION_LEN_WINDOW_SCALE) && tcp_syn (th)) - { - to->flags |= TCP_OPTS_FLAG_WSCALE; - to->wscale = data[2]; - if (to->wscale > TCP_MAX_WND_SCALE) - to->wscale = TCP_MAX_WND_SCALE; - } - break; - case TCP_OPTION_TIMESTAMP: - if (is_syn) - to->flags |= TCP_OPTS_FLAG_TSTAMP; - if ((to->flags & TCP_OPTS_FLAG_TSTAMP) - && opt_len == TCP_OPTION_LEN_TIMESTAMP) - { - to->tsval = clib_net_to_host_u32 (*(u32 *) (data + 2)); - to->tsecr = clib_net_to_host_u32 (*(u32 *) (data + 6)); - } - break; - case TCP_OPTION_SACK_PERMITTED: - if (!is_syn) - break; - if (opt_len == TCP_OPTION_LEN_SACK_PERMITTED && tcp_syn (th)) - to->flags |= TCP_OPTS_FLAG_SACK_PERMITTED; - break; - case TCP_OPTION_SACK_BLOCK: - /* If SACK permitted was not advertised or a SYN, break */ - if ((to->flags & TCP_OPTS_FLAG_SACK_PERMITTED) == 0 || tcp_syn (th)) - break; - - /* If too short or not correctly formatted, break */ - if (opt_len < 10 || ((opt_len - 2) % TCP_OPTION_LEN_SACK_BLOCK)) - break; - - to->flags |= TCP_OPTS_FLAG_SACK; - to->n_sack_blocks = (opt_len - 2) / TCP_OPTION_LEN_SACK_BLOCK; - vec_reset_length (to->sacks); - for (j = 0; j < to->n_sack_blocks; j++) - { - b.start = clib_net_to_host_u32 (*(u32 *) (data + 2 + 8 * j)); - b.end = clib_net_to_host_u32 (*(u32 *) (data + 6 + 8 * j)); - vec_add1 (to->sacks, b); - } - break; - default: - /* Nothing to see here */ - continue; - } - } - return 0; -} - -/** * RFC1323: Check against wrapped sequence numbers (PAWS). If we have * timestamp to echo and it's less than tsval_recent, drop segment * but still send an ACK in order to retain TCP's mechanism for detecting @@ -565,15 +452,6 @@ tcp_estimate_rtt (tcp_connection_t * tc, u32 mrtt) } } -#ifndef CLIB_MARCH_VARIANT -void -tcp_update_rto (tcp_connection_t * tc) -{ - tc->rto = clib_min (tc->srtt + (tc->rttvar << 2), TCP_RTO_MAX); - tc->rto = clib_max (tc->rto, TCP_RTO_MIN); -} -#endif /* CLIB_MARCH_VARIANT */ - /** * Update RTT estimate and RTO timer * @@ -726,567 +604,6 @@ tcp_program_dequeue (tcp_worker_ctx_t * wrk, tcp_connection_t * tc) tc->burst_acked += tc->bytes_acked; } -#ifndef CLIB_MARCH_VARIANT -static u32 -scoreboard_hole_index (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole) -{ - ASSERT (!pool_is_free_index (sb->holes, hole - sb->holes)); - return hole - sb->holes; -} - -static u32 -scoreboard_hole_bytes (sack_scoreboard_hole_t * hole) -{ - return hole->end - hole->start; -} - -sack_scoreboard_hole_t * -scoreboard_get_hole (sack_scoreboard_t * sb, u32 index) -{ - if (index != TCP_INVALID_SACK_HOLE_INDEX) - return pool_elt_at_index (sb->holes, index); - return 0; -} - -sack_scoreboard_hole_t * -scoreboard_next_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole) -{ - if (hole->next != TCP_INVALID_SACK_HOLE_INDEX) - return pool_elt_at_index (sb->holes, hole->next); - return 0; -} - -sack_scoreboard_hole_t * -scoreboard_prev_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole) -{ - if (hole->prev != TCP_INVALID_SACK_HOLE_INDEX) - return pool_elt_at_index (sb->holes, hole->prev); - return 0; -} - -sack_scoreboard_hole_t * -scoreboard_first_hole (sack_scoreboard_t * sb) -{ - if (sb->head != TCP_INVALID_SACK_HOLE_INDEX) - return pool_elt_at_index (sb->holes, sb->head); - return 0; -} - -sack_scoreboard_hole_t * -scoreboard_last_hole (sack_scoreboard_t * sb) -{ - if (sb->tail != TCP_INVALID_SACK_HOLE_INDEX) - return pool_elt_at_index (sb->holes, sb->tail); - return 0; -} - -static void -scoreboard_remove_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole) -{ - sack_scoreboard_hole_t *next, *prev; - - if (hole->next != TCP_INVALID_SACK_HOLE_INDEX) - { - next = pool_elt_at_index (sb->holes, hole->next); - next->prev = hole->prev; - } - else - { - sb->tail = hole->prev; - } - - if (hole->prev != TCP_INVALID_SACK_HOLE_INDEX) - { - prev = pool_elt_at_index (sb->holes, hole->prev); - prev->next = hole->next; - } - else - { - sb->head = hole->next; - } - - if (scoreboard_hole_index (sb, hole) == sb->cur_rxt_hole) - sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX; - - /* Poison the entry */ - if (CLIB_DEBUG > 0) - clib_memset (hole, 0xfe, sizeof (*hole)); - - pool_put (sb->holes, hole); -} - -static sack_scoreboard_hole_t * -scoreboard_insert_hole (sack_scoreboard_t * sb, u32 prev_index, - u32 start, u32 end) -{ - sack_scoreboard_hole_t *hole, *next, *prev; - u32 hole_index; - - pool_get (sb->holes, hole); - clib_memset (hole, 0, sizeof (*hole)); - - hole->start = start; - hole->end = end; - hole_index = scoreboard_hole_index (sb, hole); - - prev = scoreboard_get_hole (sb, prev_index); - if (prev) - { - hole->prev = prev_index; - hole->next = prev->next; - - if ((next = scoreboard_next_hole (sb, hole))) - next->prev = hole_index; - else - sb->tail = hole_index; - - prev->next = hole_index; - } - else - { - sb->head = hole_index; - hole->prev = TCP_INVALID_SACK_HOLE_INDEX; - hole->next = TCP_INVALID_SACK_HOLE_INDEX; - } - - return hole; -} - -always_inline void -scoreboard_update_sacked_rxt (sack_scoreboard_t * sb, u32 start, u32 end, - u8 has_rxt) -{ - if (!has_rxt || seq_geq (start, sb->high_rxt)) - return; - - sb->rxt_sacked += - seq_lt (end, sb->high_rxt) ? (end - start) : (sb->high_rxt - start); -} - -always_inline void -scoreboard_update_bytes (sack_scoreboard_t * sb, u32 ack, u32 snd_mss) -{ - sack_scoreboard_hole_t *left, *right; - u32 sacked = 0, blks = 0, old_sacked; - - old_sacked = sb->sacked_bytes; - - sb->last_lost_bytes = 0; - sb->lost_bytes = 0; - sb->sacked_bytes = 0; - - right = scoreboard_last_hole (sb); - if (!right) - { - sb->sacked_bytes = sb->high_sacked - ack; - sb->last_sacked_bytes = sb->sacked_bytes - - (old_sacked - sb->last_bytes_delivered); - return; - } - - if (seq_gt (sb->high_sacked, right->end)) - { - sacked = sb->high_sacked - right->end; - blks = 1; - } - - while (sacked < (TCP_DUPACK_THRESHOLD - 1) * snd_mss - && blks < TCP_DUPACK_THRESHOLD) - { - if (right->is_lost) - sb->lost_bytes += scoreboard_hole_bytes (right); - - left = scoreboard_prev_hole (sb, right); - if (!left) - { - ASSERT (right->start == ack || sb->is_reneging); - sacked += right->start - ack; - right = 0; - break; - } - - sacked += right->start - left->end; - blks++; - right = left; - } - - /* right is first lost */ - while (right) - { - sb->lost_bytes += scoreboard_hole_bytes (right); - sb->last_lost_bytes += right->is_lost ? 0 : (right->end - right->start); - right->is_lost = 1; - left = scoreboard_prev_hole (sb, right); - if (!left) - { - ASSERT (right->start == ack || sb->is_reneging); - sacked += right->start - ack; - break; - } - sacked += right->start - left->end; - right = left; - } - - sb->sacked_bytes = sacked; - sb->last_sacked_bytes = sacked - (old_sacked - sb->last_bytes_delivered); -} - -/** - * Figure out the next hole to retransmit - * - * Follows logic proposed in RFC6675 Sec. 4, NextSeg() - */ -sack_scoreboard_hole_t * -scoreboard_next_rxt_hole (sack_scoreboard_t * sb, - sack_scoreboard_hole_t * start, - u8 have_unsent, u8 * can_rescue, u8 * snd_limited) -{ - sack_scoreboard_hole_t *hole = 0; - - hole = start ? start : scoreboard_first_hole (sb); - while (hole && seq_leq (hole->end, sb->high_rxt) && hole->is_lost) - hole = scoreboard_next_hole (sb, hole); - - /* Nothing, return */ - if (!hole) - { - sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX; - return 0; - } - - /* Rule (1): if higher than rxt, less than high_sacked and lost */ - if (hole->is_lost && seq_lt (hole->start, sb->high_sacked)) - { - sb->cur_rxt_hole = scoreboard_hole_index (sb, hole); - } - else - { - /* Rule (2): available unsent data */ - if (have_unsent) - { - sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX; - return 0; - } - /* Rule (3): if hole not lost */ - else if (seq_lt (hole->start, sb->high_sacked)) - { - /* And we didn't already retransmit it */ - if (seq_leq (hole->end, sb->high_rxt)) - { - sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX; - return 0; - } - *snd_limited = 0; - sb->cur_rxt_hole = scoreboard_hole_index (sb, hole); - } - /* Rule (4): if hole beyond high_sacked */ - else - { - ASSERT (seq_geq (hole->start, sb->high_sacked)); - *snd_limited = 1; - *can_rescue = 1; - /* HighRxt MUST NOT be updated */ - return 0; - } - } - - if (hole && seq_lt (sb->high_rxt, hole->start)) - sb->high_rxt = hole->start; - - return hole; -} - -void -scoreboard_init_rxt (sack_scoreboard_t * sb, u32 snd_una) -{ - sack_scoreboard_hole_t *hole; - hole = scoreboard_first_hole (sb); - if (hole) - { - snd_una = seq_gt (snd_una, hole->start) ? snd_una : hole->start; - sb->cur_rxt_hole = sb->head; - } - sb->high_rxt = snd_una; - sb->rescue_rxt = snd_una - 1; -} - -void -scoreboard_init (sack_scoreboard_t * sb) -{ - sb->head = TCP_INVALID_SACK_HOLE_INDEX; - sb->tail = TCP_INVALID_SACK_HOLE_INDEX; - sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX; -} - -void -scoreboard_clear (sack_scoreboard_t * sb) -{ - sack_scoreboard_hole_t *hole; - while ((hole = scoreboard_first_hole (sb))) - { - scoreboard_remove_hole (sb, hole); - } - ASSERT (sb->head == sb->tail && sb->head == TCP_INVALID_SACK_HOLE_INDEX); - ASSERT (pool_elts (sb->holes) == 0); - sb->sacked_bytes = 0; - sb->last_sacked_bytes = 0; - sb->last_bytes_delivered = 0; - sb->lost_bytes = 0; - sb->last_lost_bytes = 0; - sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX; - sb->is_reneging = 0; -} - -void -scoreboard_clear_reneging (sack_scoreboard_t * sb, u32 start, u32 end) -{ - sack_scoreboard_hole_t *last_hole; - - clib_warning ("sack reneging"); - - scoreboard_clear (sb); - last_hole = scoreboard_insert_hole (sb, TCP_INVALID_SACK_HOLE_INDEX, - start, end); - last_hole->is_lost = 1; - sb->tail = scoreboard_hole_index (sb, last_hole); - sb->high_sacked = start; - scoreboard_init_rxt (sb, start); -} - -#endif /* CLIB_MARCH_VARIANT */ - -/** - * Test that scoreboard is sane after recovery - * - * Returns 1 if scoreboard is empty or if first hole beyond - * snd_una. - */ -static u8 -tcp_scoreboard_is_sane_post_recovery (tcp_connection_t * tc) -{ - sack_scoreboard_hole_t *hole; - hole = scoreboard_first_hole (&tc->sack_sb); - return (!hole || (seq_geq (hole->start, tc->snd_una) - && seq_lt (hole->end, tc->snd_nxt))); -} - -#ifndef CLIB_MARCH_VARIANT - -void -tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) -{ - sack_scoreboard_hole_t *hole, *next_hole; - sack_scoreboard_t *sb = &tc->sack_sb; - sack_block_t *blk, *rcv_sacks; - u32 blk_index = 0, i, j; - u8 has_rxt; - - sb->last_sacked_bytes = 0; - sb->last_bytes_delivered = 0; - sb->rxt_sacked = 0; - - if (!tcp_opts_sack (&tc->rcv_opts) && !sb->sacked_bytes - && sb->head == TCP_INVALID_SACK_HOLE_INDEX) - return; - - has_rxt = tcp_in_cong_recovery (tc); - - /* Remove invalid blocks */ - blk = tc->rcv_opts.sacks; - while (blk < vec_end (tc->rcv_opts.sacks)) - { - if (seq_lt (blk->start, blk->end) - && seq_gt (blk->start, tc->snd_una) - && seq_gt (blk->start, ack) && seq_leq (blk->end, tc->snd_nxt)) - { - blk++; - continue; - } - vec_del1 (tc->rcv_opts.sacks, blk - tc->rcv_opts.sacks); - } - - /* Add block for cumulative ack */ - if (seq_gt (ack, tc->snd_una)) - { - vec_add2 (tc->rcv_opts.sacks, blk, 1); - blk->start = tc->snd_una; - blk->end = ack; - } - - if (vec_len (tc->rcv_opts.sacks) == 0) - return; - - tcp_scoreboard_trace_add (tc, ack); - - /* Make sure blocks are ordered */ - rcv_sacks = tc->rcv_opts.sacks; - for (i = 0; i < vec_len (rcv_sacks); i++) - for (j = i + 1; j < vec_len (rcv_sacks); j++) - if (seq_lt (rcv_sacks[j].start, rcv_sacks[i].start)) - { - sack_block_t tmp = rcv_sacks[i]; - rcv_sacks[i] = rcv_sacks[j]; - rcv_sacks[j] = tmp; - } - - if (sb->head == TCP_INVALID_SACK_HOLE_INDEX) - { - /* Handle reneging as a special case */ - if (PREDICT_FALSE (sb->is_reneging)) - { - /* No holes, only sacked bytes */ - if (seq_leq (tc->snd_nxt, sb->high_sacked)) - { - /* No progress made so return */ - if (seq_leq (ack, tc->snd_una)) - return; - - /* Update sacked bytes delivered and return */ - sb->last_bytes_delivered = ack - tc->snd_una; - sb->sacked_bytes -= sb->last_bytes_delivered; - sb->is_reneging = seq_lt (ack, sb->high_sacked); - return; - } - - /* New hole above high sacked. Add it and process normally */ - hole = scoreboard_insert_hole (sb, TCP_INVALID_SACK_HOLE_INDEX, - sb->high_sacked, tc->snd_nxt); - sb->tail = scoreboard_hole_index (sb, hole); - } - /* Not reneging and no holes. Insert the first that covers all - * outstanding bytes */ - else - { - hole = scoreboard_insert_hole (sb, TCP_INVALID_SACK_HOLE_INDEX, - tc->snd_una, tc->snd_nxt); - sb->tail = scoreboard_hole_index (sb, hole); - } - sb->high_sacked = rcv_sacks[vec_len (rcv_sacks) - 1].end; - } - else - { - /* If we have holes but snd_nxt is beyond the last hole, update - * last hole end or add new hole after high sacked */ - hole = scoreboard_last_hole (sb); - if (seq_gt (tc->snd_nxt, hole->end)) - { - if (seq_geq (hole->start, sb->high_sacked)) - { - hole->end = tc->snd_nxt; - } - /* New hole after high sacked block */ - else if (seq_lt (sb->high_sacked, tc->snd_nxt)) - { - scoreboard_insert_hole (sb, sb->tail, sb->high_sacked, - tc->snd_nxt); - } - } - - /* Keep track of max byte sacked for when the last hole - * is acked */ - sb->high_sacked = seq_max (rcv_sacks[vec_len (rcv_sacks) - 1].end, - sb->high_sacked); - } - - /* Walk the holes with the SACK blocks */ - hole = pool_elt_at_index (sb->holes, sb->head); - - if (PREDICT_FALSE (sb->is_reneging)) - { - sb->last_bytes_delivered += clib_min (hole->start - tc->snd_una, - ack - tc->snd_una); - sb->is_reneging = seq_lt (ack, hole->start); - } - - while (hole && blk_index < vec_len (rcv_sacks)) - { - blk = &rcv_sacks[blk_index]; - if (seq_leq (blk->start, hole->start)) - { - /* Block covers hole. Remove hole */ - if (seq_geq (blk->end, hole->end)) - { - next_hole = scoreboard_next_hole (sb, hole); - - /* If covered by ack, compute delivered bytes */ - if (blk->end == ack) - { - u32 sacked = next_hole ? next_hole->start : sb->high_sacked; - if (PREDICT_FALSE (seq_lt (ack, sacked))) - { - sb->last_bytes_delivered += ack - hole->end; - sb->is_reneging = 1; - } - else - { - sb->last_bytes_delivered += sacked - hole->end; - sb->is_reneging = 0; - } - } - scoreboard_update_sacked_rxt (sb, hole->start, hole->end, - has_rxt); - scoreboard_remove_hole (sb, hole); - hole = next_hole; - } - /* Partial 'head' overlap */ - else - { - if (seq_gt (blk->end, hole->start)) - { - scoreboard_update_sacked_rxt (sb, hole->start, blk->end, - has_rxt); - hole->start = blk->end; - } - blk_index++; - } - } - else - { - /* Hole must be split */ - if (seq_lt (blk->end, hole->end)) - { - u32 hole_index = scoreboard_hole_index (sb, hole); - next_hole = scoreboard_insert_hole (sb, hole_index, blk->end, - hole->end); - /* Pool might've moved */ - hole = scoreboard_get_hole (sb, hole_index); - hole->end = blk->start; - - scoreboard_update_sacked_rxt (sb, blk->start, blk->end, - has_rxt); - - blk_index++; - ASSERT (hole->next == scoreboard_hole_index (sb, next_hole)); - } - else if (seq_lt (blk->start, hole->end)) - { - scoreboard_update_sacked_rxt (sb, blk->start, hole->end, - has_rxt); - hole->end = blk->start; - } - hole = scoreboard_next_hole (sb, hole); - } - } - - scoreboard_update_bytes (sb, ack, tc->snd_mss); - - ASSERT (sb->last_sacked_bytes <= sb->sacked_bytes || tcp_in_recovery (tc)); - ASSERT (sb->sacked_bytes == 0 || tcp_in_recovery (tc) - || sb->sacked_bytes <= tc->snd_nxt - seq_max (tc->snd_una, ack)); - ASSERT (sb->last_sacked_bytes + sb->lost_bytes <= tc->snd_nxt - - seq_max (tc->snd_una, ack) || tcp_in_recovery (tc)); - ASSERT (sb->head == TCP_INVALID_SACK_HOLE_INDEX || tcp_in_recovery (tc) - || sb->is_reneging || sb->holes[sb->head].start == ack); - ASSERT (sb->last_lost_bytes <= sb->lost_bytes); - ASSERT ((ack - tc->snd_una) + sb->last_sacked_bytes - - sb->last_bytes_delivered >= sb->rxt_sacked); - ASSERT ((ack - tc->snd_una) >= tc->sack_sb.last_bytes_delivered - || (tc->flags & TCP_CONN_FINSNT)); - - TCP_EVT (TCP_EVT_CC_SCOREBOARD, tc); -} -#endif /* CLIB_MARCH_VARIANT */ - /** * Try to update snd_wnd based on feedback received from peer. * @@ -1825,89 +1142,6 @@ tcp_rcv_fin (tcp_worker_ctx_t * wrk, tcp_connection_t * tc, vlib_buffer_t * b, *error = TCP_ERROR_FIN_RCVD; } -#ifndef CLIB_MARCH_VARIANT -static u8 -tcp_sack_vector_is_sane (sack_block_t * sacks) -{ - int i; - for (i = 1; i < vec_len (sacks); i++) - { - if (sacks[i - 1].end == sacks[i].start) - return 0; - } - return 1; -} - -/** - * Build SACK list as per RFC2018. - * - * Makes sure the first block contains the segment that generated the current - * ACK and the following ones are the ones most recently reported in SACK - * blocks. - * - * @param tc TCP connection for which the SACK list is updated - * @param start Start sequence number of the newest SACK block - * @param end End sequence of the newest SACK block - */ -void -tcp_update_sack_list (tcp_connection_t * tc, u32 start, u32 end) -{ - sack_block_t *new_list = tc->snd_sacks_fl, *block = 0; - int i; - - /* If the first segment is ooo add it to the list. Last write might've moved - * rcv_nxt over the first segment. */ - if (seq_lt (tc->rcv_nxt, start)) - { - vec_add2 (new_list, block, 1); - block->start = start; - block->end = end; - } - - /* Find the blocks still worth keeping. */ - for (i = 0; i < vec_len (tc->snd_sacks); i++) - { - /* Discard if rcv_nxt advanced beyond current block */ - if (seq_leq (tc->snd_sacks[i].start, tc->rcv_nxt)) - continue; - - /* Merge or drop if segment overlapped by the new segment */ - if (block && (seq_geq (tc->snd_sacks[i].end, new_list[0].start) - && seq_leq (tc->snd_sacks[i].start, new_list[0].end))) - { - if (seq_lt (tc->snd_sacks[i].start, new_list[0].start)) - new_list[0].start = tc->snd_sacks[i].start; - if (seq_lt (new_list[0].end, tc->snd_sacks[i].end)) - new_list[0].end = tc->snd_sacks[i].end; - continue; - } - - /* Save to new SACK list if we have space. */ - if (vec_len (new_list) < TCP_MAX_SACK_BLOCKS) - vec_add1 (new_list, tc->snd_sacks[i]); - } - - ASSERT (vec_len (new_list) <= TCP_MAX_SACK_BLOCKS); - - /* Replace old vector with new one */ - vec_reset_length (tc->snd_sacks); - tc->snd_sacks_fl = tc->snd_sacks; - tc->snd_sacks = new_list; - - /* Segments should not 'touch' */ - ASSERT (tcp_sack_vector_is_sane (tc->snd_sacks)); -} - -u32 -tcp_sack_list_bytes (tcp_connection_t * tc) -{ - u32 bytes = 0, i; - for (i = 0; i < vec_len (tc->snd_sacks); i++) - bytes += tc->snd_sacks[i].end - tc->snd_sacks[i].start; - return bytes; -} -#endif /* CLIB_MARCH_VARIANT */ - /** Enqueue data for delivery to application */ static int tcp_session_enqueue_data (tcp_connection_t * tc, vlib_buffer_t * b, @@ -3551,102 +2785,6 @@ tcp_input_set_error_next (tcp_main_t * tm, u16 * next, u32 * error, u8 is_ip4) } } -always_inline tcp_connection_t * -tcp_input_lookup_buffer (vlib_buffer_t * b, u8 thread_index, u32 * error, - u8 is_ip4, u8 is_nolookup) -{ - u32 fib_index = vnet_buffer (b)->ip.fib_index; - int n_advance_bytes, n_data_bytes; - transport_connection_t *tc; - tcp_header_t *tcp; - u8 result = 0; - - if (is_ip4) - { - ip4_header_t *ip4 = vlib_buffer_get_current (b); - int ip_hdr_bytes = ip4_header_bytes (ip4); - if (PREDICT_FALSE (b->current_length < ip_hdr_bytes + sizeof (*tcp))) - { - *error = TCP_ERROR_LENGTH; - return 0; - } - tcp = ip4_next_header (ip4); - vnet_buffer (b)->tcp.hdr_offset = (u8 *) tcp - (u8 *) ip4; - n_advance_bytes = (ip_hdr_bytes + tcp_header_bytes (tcp)); - n_data_bytes = clib_net_to_host_u16 (ip4->length) - n_advance_bytes; - - /* Length check. Checksum computed by ipx_local no need to compute again */ - if (PREDICT_FALSE (n_data_bytes < 0)) - { - *error = TCP_ERROR_LENGTH; - return 0; - } - - if (!is_nolookup) - tc = session_lookup_connection_wt4 (fib_index, &ip4->dst_address, - &ip4->src_address, tcp->dst_port, - tcp->src_port, - TRANSPORT_PROTO_TCP, thread_index, - &result); - } - else - { - ip6_header_t *ip6 = vlib_buffer_get_current (b); - if (PREDICT_FALSE (b->current_length < sizeof (*ip6) + sizeof (*tcp))) - { - *error = TCP_ERROR_LENGTH; - return 0; - } - tcp = ip6_next_header (ip6); - vnet_buffer (b)->tcp.hdr_offset = (u8 *) tcp - (u8 *) ip6; - n_advance_bytes = tcp_header_bytes (tcp); - n_data_bytes = clib_net_to_host_u16 (ip6->payload_length) - - n_advance_bytes; - n_advance_bytes += sizeof (ip6[0]); - - if (PREDICT_FALSE (n_data_bytes < 0)) - { - *error = TCP_ERROR_LENGTH; - return 0; - } - - if (!is_nolookup) - { - if (PREDICT_FALSE - (ip6_address_is_link_local_unicast (&ip6->dst_address))) - { - ip4_main_t *im = &ip4_main; - fib_index = vec_elt (im->fib_index_by_sw_if_index, - vnet_buffer (b)->sw_if_index[VLIB_RX]); - } - - tc = session_lookup_connection_wt6 (fib_index, &ip6->dst_address, - &ip6->src_address, - tcp->dst_port, tcp->src_port, - TRANSPORT_PROTO_TCP, - thread_index, &result); - } - } - - if (is_nolookup) - tc = - (transport_connection_t *) tcp_connection_get (vnet_buffer (b)-> - tcp.connection_index, - thread_index); - - vnet_buffer (b)->tcp.seq_number = clib_net_to_host_u32 (tcp->seq_number); - vnet_buffer (b)->tcp.ack_number = clib_net_to_host_u32 (tcp->ack_number); - vnet_buffer (b)->tcp.data_offset = n_advance_bytes; - vnet_buffer (b)->tcp.data_len = n_data_bytes; - vnet_buffer (b)->tcp.seq_end = vnet_buffer (b)->tcp.seq_number - + n_data_bytes; - vnet_buffer (b)->tcp.flags = 0; - - *error = result ? TCP_ERROR_NONE + result : *error; - - return tcp_get_connection_from_transport (tc); -} - static inline void tcp_input_dispatch_buffer (tcp_main_t * tm, tcp_connection_t * tc, vlib_buffer_t * b, u16 * next, |