From 93992a9048cb6e5dcd22de5091e72de778122627 Mon Sep 17 00:00:00 2001 From: Florin Coras Date: Wed, 24 May 2017 18:03:56 -0700 Subject: Implement sack based tcp loss recovery (RFC 6675) - refactor existing congestion control code (RFC 6582/5681). Handling of ack feedback now consists of: ack parsing, cc event detection, event handling, congestion control update - extend sack scoreboard to support sack based retransmissions - basic implementation of Eifel detection algorithm (RFC 3522) for detecting spurious retransmissions - actually initialize the per-thread frame freelist hash tables - increase worker stack size to 2mb - fix session queue node out-of-buffer handling - ensure that the local buffer cache vec_len matches reality - avoid 2x spurious event requeues when short of buffers - count out-of-buffer events - make the builtin server thread-safe - fix bihash template threading issue: need to paint -1 across uninitialized working_copy_length vector elements (via rebase from master) Change-Id: I646cb9f1add9a67d08f4a87badbcb117980ebfc4 Signed-off-by: Florin Coras Signed-off-by: Dave Barach --- src/vnet/tcp/tcp_input.c | 671 ++++++++++++++++++++++++++++++++++------------- 1 file changed, 485 insertions(+), 186 deletions(-) (limited to 'src/vnet/tcp/tcp_input.c') diff --git a/src/vnet/tcp/tcp_input.c b/src/vnet/tcp/tcp_input.c index 35bc9094..ff2229b3 100644 --- a/src/vnet/tcp/tcp_input.c +++ b/src/vnet/tcp/tcp_input.c @@ -231,8 +231,8 @@ tcp_options_parse (tcp_header_t * th, tcp_options_t * to) always_inline int tcp_segment_check_paws (tcp_connection_t * tc) { - return tcp_opts_tstamp (&tc->opt) && tc->tsval_recent - && timestamp_lt (tc->opt.tsval, tc->tsval_recent); + return tcp_opts_tstamp (&tc->rcv_opts) && tc->tsval_recent + && timestamp_lt (tc->rcv_opts.tsval, tc->tsval_recent); } /** @@ -248,10 +248,10 @@ tcp_update_timestamp (tcp_connection_t * tc, u32 seq, u32 seq_end) * then the TSval from the segment is copied to TS.Recent; * otherwise, the TSval is ignored. */ - if (tcp_opts_tstamp (&tc->opt) && tc->tsval_recent + if (tcp_opts_tstamp (&tc->rcv_opts) && tc->tsval_recent && seq_leq (seq, tc->rcv_las) && seq_leq (tc->rcv_las, seq_end)) { - tc->tsval_recent = tc->opt.tsval; + tc->tsval_recent = tc->rcv_opts.tsval; tc->tsval_recent_age = tcp_time_now (); } } @@ -272,14 +272,21 @@ tcp_segment_validate (vlib_main_t * vm, tcp_connection_t * tc0, if (PREDICT_FALSE (!tcp_ack (th0) && !tcp_rst (th0) && !tcp_syn (th0))) return -1; - if (PREDICT_FALSE (tcp_options_parse (th0, &tc0->opt))) + if (PREDICT_FALSE (tcp_options_parse (th0, &tc0->rcv_opts))) { return -1; } if (tcp_segment_check_paws (tc0)) { - clib_warning ("paws failed"); + if (CLIB_DEBUG > 2) + { + clib_warning ("paws failed\n%U", format_tcp_connection, tc0, 2); + clib_warning ("seq %u seq_end %u ack %u", + vnet_buffer (b0)->tcp.seq_number - tc0->irs, + vnet_buffer (b0)->tcp.seq_end - tc0->irs, + vnet_buffer (b0)->tcp.ack_number - tc0->iss); + } TCP_EVT_DBG (TCP_EVT_PAWS_FAIL, tc0, vnet_buffer (b0)->tcp.seq_number, vnet_buffer (b0)->tcp.seq_end); @@ -348,7 +355,6 @@ tcp_segment_validate (vlib_main_t * vm, tcp_connection_t * tc0, /* If segment in window, save timestamp */ tcp_update_timestamp (tc0, vnet_buffer (b0)->tcp.seq_number, vnet_buffer (b0)->tcp.seq_end); - return 0; } @@ -391,6 +397,12 @@ tcp_estimate_rtt (tcp_connection_t * tc, u32 mrtt) } } +void +tcp_update_rto (tcp_connection_t * tc) +{ + tc->rto = clib_min (tc->srtt + (tc->rttvar << 2), TCP_RTO_MAX); +} + /** Update RTT estimate and RTO timer * * Measure RTT: We have two sources of RTT measurements: TSOPT and ACK @@ -405,7 +417,7 @@ tcp_update_rtt (tcp_connection_t * tc, u32 ack) u32 mrtt = 0; u8 rtx_acked; - /* Determine if only rtx bytes are acked. TODO fast retransmit */ + /* Determine if only rtx bytes are acked. TODO XXX fast retransmit */ rtx_acked = tc->rto_boff && (tc->bytes_acked <= tc->snd_mss); /* Karn's rule, part 1. Don't use retransmitted segments to estimate @@ -418,9 +430,10 @@ tcp_update_rtt (tcp_connection_t * tc, u32 ack) * snd_una, i.e., the left side of the send window: * seq_lt (tc->snd_una, ack). Note: last condition could be dropped, we don't * try to update rtt for dupacks */ - else if (tcp_opts_tstamp (&tc->opt) && tc->opt.tsecr && tc->bytes_acked) + else if (tcp_opts_tstamp (&tc->rcv_opts) && tc->rcv_opts.tsecr + && tc->bytes_acked) { - mrtt = tcp_time_now () - tc->opt.tsecr; + mrtt = tcp_time_now () - tc->rcv_opts.tsecr; } /* Allow measuring of a new RTT */ @@ -436,7 +449,7 @@ tcp_update_rtt (tcp_connection_t * tc, u32 ack) return 0; tcp_estimate_rtt (tc, mrtt); - tc->rto = clib_min (tc->srtt + (tc->rttvar << 2), TCP_RTO_MAX); + tcp_update_rto (tc); return 0; } @@ -447,25 +460,46 @@ tcp_update_rtt (tcp_connection_t * tc, u32 ack) static void tcp_dequeue_acked (tcp_connection_t * tc, u32 ack) { - /* Dequeue the newly ACKed bytes */ - stream_session_dequeue_drop (&tc->connection, tc->bytes_acked); + /* Dequeue the newly ACKed add SACKed bytes */ + stream_session_dequeue_drop (&tc->connection, + tc->bytes_acked + tc->sack_sb.snd_una_adv); + + tcp_validate_txf_size (tc, tc->snd_una_max - tc->snd_una); /* Update rtt and rto */ tcp_update_rtt (tc, ack); + + /* If everything has been acked, stop retransmit timer + * otherwise update. */ + tcp_retransmit_timer_update (tc); } /** - * Check if dupack as per RFC5681 Sec. 2 - * - * This works only if called before updating snd_wnd. - * */ -always_inline u8 -tcp_ack_is_dupack (tcp_connection_t * tc, vlib_buffer_t * b, u32 new_snd_wnd) + * Check if duplicate ack as per RFC5681 Sec. 2 + */ +static u8 +tcp_ack_is_dupack (tcp_connection_t * tc, vlib_buffer_t * b, u32 prev_snd_wnd, + u32 prev_snd_una) { - return ((vnet_buffer (b)->tcp.ack_number == tc->snd_una) + return ((vnet_buffer (b)->tcp.ack_number == prev_snd_una) && seq_gt (tc->snd_una_max, tc->snd_una) && (vnet_buffer (b)->tcp.seq_end == vnet_buffer (b)->tcp.seq_number) - && (new_snd_wnd == tc->snd_wnd)); + && (prev_snd_wnd == tc->snd_wnd)); +} + +/** + * Checks if ack is a congestion control event. + */ +static u8 +tcp_ack_is_cc_event (tcp_connection_t * tc, vlib_buffer_t * b, + u32 prev_snd_wnd, u32 prev_snd_una, u8 * is_dack) +{ + /* Check if ack is duplicate. Per RFC 6675, ACKs that SACK new data are + * defined to be 'duplicate' */ + *is_dack = tc->sack_sb.last_sacked_bytes + || tcp_ack_is_dupack (tc, b, prev_snd_wnd, prev_snd_una); + + return (*is_dack || tcp_in_cong_recovery (tc)); } void @@ -478,6 +512,10 @@ scoreboard_remove_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole) 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) { @@ -489,6 +527,9 @@ scoreboard_remove_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole) sb->head = hole->next; } + if (scoreboard_hole_index (sb, hole) == sb->cur_rxt_hole) + sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX; + pool_put (sb->holes, hole); } @@ -527,26 +568,131 @@ scoreboard_insert_hole (sack_scoreboard_t * sb, u32 prev_index, return hole; } +void +scoreboard_update_lost (tcp_connection_t * tc, sack_scoreboard_t * sb) +{ + sack_scoreboard_hole_t *hole, *prev; + u32 bytes = 0, blks = 0; + + sb->lost_bytes = 0; + hole = scoreboard_last_hole (sb); + if (!hole) + return; + + if (seq_gt (sb->high_sacked, hole->end)) + { + bytes = sb->high_sacked - hole->end; + blks = 1; + } + + while ((prev = scoreboard_prev_hole (sb, hole)) + && (bytes < (TCP_DUPACK_THRESHOLD - 1) * tc->snd_mss + && blks < TCP_DUPACK_THRESHOLD)) + { + bytes += hole->start - prev->end; + blks++; + hole = prev; + } + + hole = prev; + while (hole) + { + sb->lost_bytes += scoreboard_hole_bytes (hole); + hole->is_lost = 1; + hole = scoreboard_prev_hole (sb, hole); + } +} + +/** + * 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_sent_1_smss, + 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): output takes care of transmitting new data */ + if (!have_sent_1_smss) + { + hole = 0; + sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX; + } + /* Rule (3): if hole not lost */ + else if (seq_lt (hole->start, sb->high_sacked)) + { + *snd_limited = 1; + 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_high_rxt (sack_scoreboard_t * sb) +{ + sack_scoreboard_hole_t *hole; + hole = scoreboard_first_hole (sb); + sb->high_rxt = hole->start; + sb->cur_rxt_hole = sb->head; +} + void tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) { sack_scoreboard_t *sb = &tc->sack_sb; sack_block_t *blk, tmp; sack_scoreboard_hole_t *hole, *next_hole, *last_hole, *new_hole; - u32 blk_index = 0, old_sacked_bytes, delivered_bytes, hole_index; + u32 blk_index = 0, old_sacked_bytes, hole_index; int i, j; sb->last_sacked_bytes = 0; sb->snd_una_adv = 0; old_sacked_bytes = sb->sacked_bytes; - delivered_bytes = 0; + sb->last_bytes_delivered = 0; - if (!tcp_opts_sack (&tc->opt) && sb->head == TCP_INVALID_SACK_HOLE_INDEX) + if (!tcp_opts_sack (&tc->rcv_opts) + && sb->head == TCP_INVALID_SACK_HOLE_INDEX) return; /* Remove invalid blocks */ - blk = tc->opt.sacks; - while (blk < vec_end (tc->opt.sacks)) + 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) @@ -555,7 +701,7 @@ tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) blk++; continue; } - vec_del1 (tc->opt.sacks, blk - tc->opt.sacks); + vec_del1 (tc->rcv_opts.sacks, blk - tc->rcv_opts.sacks); } /* Add block for cumulative ack */ @@ -563,20 +709,20 @@ tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) { tmp.start = tc->snd_una; tmp.end = ack; - vec_add1 (tc->opt.sacks, tmp); + vec_add1 (tc->rcv_opts.sacks, tmp); } - if (vec_len (tc->opt.sacks) == 0) + if (vec_len (tc->rcv_opts.sacks) == 0) return; /* Make sure blocks are ordered */ - for (i = 0; i < vec_len (tc->opt.sacks); i++) - for (j = i + 1; j < vec_len (tc->opt.sacks); j++) - if (seq_lt (tc->opt.sacks[j].start, tc->opt.sacks[i].start)) + for (i = 0; i < vec_len (tc->rcv_opts.sacks); i++) + for (j = i + 1; j < vec_len (tc->rcv_opts.sacks); j++) + if (seq_lt (tc->rcv_opts.sacks[j].start, tc->rcv_opts.sacks[i].start)) { - tmp = tc->opt.sacks[i]; - tc->opt.sacks[i] = tc->opt.sacks[j]; - tc->opt.sacks[j] = tmp; + tmp = tc->rcv_opts.sacks[i]; + tc->rcv_opts.sacks[i] = tc->rcv_opts.sacks[j]; + tc->rcv_opts.sacks[j] = tmp; } if (sb->head == TCP_INVALID_SACK_HOLE_INDEX) @@ -585,25 +731,25 @@ tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) last_hole = scoreboard_insert_hole (sb, TCP_INVALID_SACK_HOLE_INDEX, tc->snd_una, tc->snd_una_max); sb->tail = scoreboard_hole_index (sb, last_hole); - tmp = tc->opt.sacks[vec_len (tc->opt.sacks) - 1]; - sb->max_byte_sacked = tmp.end; + tmp = tc->rcv_opts.sacks[vec_len (tc->rcv_opts.sacks) - 1]; + sb->high_sacked = tmp.end; } else { /* If we have holes but snd_una_max is beyond the last hole, update * last hole end */ - tmp = tc->opt.sacks[vec_len (tc->opt.sacks) - 1]; + tmp = tc->rcv_opts.sacks[vec_len (tc->rcv_opts.sacks) - 1]; last_hole = scoreboard_last_hole (sb); - if (seq_gt (tc->snd_una_max, sb->max_byte_sacked) + if (seq_gt (tc->snd_una_max, sb->high_sacked) && seq_gt (tc->snd_una_max, last_hole->end)) last_hole->end = tc->snd_una_max; } /* Walk the holes with the SACK blocks */ hole = pool_elt_at_index (sb->holes, sb->head); - while (hole && blk_index < vec_len (tc->opt.sacks)) + while (hole && blk_index < vec_len (tc->rcv_opts.sacks)) { - blk = &tc->opt.sacks[blk_index]; + blk = &tc->rcv_opts.sacks[blk_index]; if (seq_leq (blk->start, hole->start)) { @@ -617,9 +763,9 @@ tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) { /* Bytes lost because snd_wnd left edge advances */ if (next_hole && seq_leq (next_hole->start, ack)) - delivered_bytes += next_hole->start - hole->end; + sb->last_bytes_delivered += next_hole->start - hole->end; else - delivered_bytes += ack - hole->end; + sb->last_bytes_delivered += ack - hole->end; } else { @@ -633,8 +779,8 @@ tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) last_hole = scoreboard_last_hole (sb); /* keep track of max byte sacked for when the last hole * is acked */ - if (seq_gt (hole->end, sb->max_byte_sacked)) - sb->max_byte_sacked = hole->end; + if (seq_gt (hole->end, sb->high_sacked)) + sb->high_sacked = hole->end; } /* snd_una needs to be advanced */ @@ -645,12 +791,12 @@ tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) sb->snd_una_adv = next_hole->start - ack; /* all these can be delivered */ - delivered_bytes += sb->snd_una_adv; + sb->last_bytes_delivered += sb->snd_una_adv; } else if (!next_hole) { - sb->snd_una_adv = sb->max_byte_sacked - ack; - delivered_bytes += sb->snd_una_adv; + sb->snd_una_adv = sb->high_sacked - ack; + sb->last_bytes_delivered += sb->snd_una_adv; } } @@ -691,28 +837,33 @@ tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) } blk_index++; - hole = scoreboard_next_hole (sb, hole); } - else + else if (seq_leq (blk->start, hole->end)) { sb->sacked_bytes += hole->end - blk->start; hole->end = blk->start; - hole = scoreboard_next_hole (sb, hole); } + + hole = scoreboard_next_hole (sb, hole); } } sb->last_sacked_bytes = sb->sacked_bytes - old_sacked_bytes; - sb->sacked_bytes -= delivered_bytes; + sb->sacked_bytes -= sb->last_bytes_delivered; + scoreboard_update_lost (tc, sb); } -/** Update snd_wnd +/** + * Try to update snd_wnd based on feedback received from peer. * - * If (SND.WL1 < SEG.SEQ or (SND.WL1 = SEG.SEQ and SND.WL2 =< SEG.ACK)), set - * SND.WND <- SEG.WND, set SND.WL1 <- SEG.SEQ, and set SND.WL2 <- SEG.ACK */ + * If successful, and new window is 'effectively' 0, activate persist + * timer. + */ static void tcp_update_snd_wnd (tcp_connection_t * tc, u32 seq, u32 ack, u32 snd_wnd) { + /* If (SND.WL1 < SEG.SEQ or (SND.WL1 = SEG.SEQ and SND.WL2 =< SEG.ACK)), set + * SND.WND <- SEG.WND, set SND.WL1 <- SEG.SEQ, and set SND.WL2 <- SEG.ACK */ if (seq_lt (tc->snd_wl1, seq) || (tc->snd_wl1 == seq && seq_leq (tc->snd_wl2, ack))) { @@ -721,138 +872,269 @@ tcp_update_snd_wnd (tcp_connection_t * tc, u32 seq, u32 ack, u32 snd_wnd) tc->snd_wl2 = ack; TCP_EVT_DBG (TCP_EVT_SND_WND, tc); - /* Set probe timer if we just got 0 wnd */ if (tc->snd_wnd < tc->snd_mss) { - if (!tcp_timer_is_active (tc, TCP_TIMER_PERSIST)) + /* Set persist timer if not set and we just got 0 wnd */ + if (!tcp_timer_is_active (tc, TCP_TIMER_PERSIST) + && !tcp_timer_is_active (tc, TCP_TIMER_RETRANSMIT)) tcp_persist_timer_set (tc); } else - tcp_persist_timer_reset (tc); + { + tcp_persist_timer_reset (tc); + if (!tcp_in_recovery (tc) && tc->rto_boff > 0) + { + tc->rto_boff = 0; + tcp_update_rto (tc); + } + } } } void -tcp_cc_congestion (tcp_connection_t * tc) +tcp_cc_init_congestion (tcp_connection_t * tc) { - tc->snd_congestion = tc->snd_nxt; + tcp_fastrecovery_on (tc); + tc->snd_congestion = tc->snd_una_max; tc->cc_algo->congestion (tc); TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 4); } -void -tcp_cc_recover (tcp_connection_t * tc) +static void +tcp_cc_recovery_exit (tcp_connection_t * tc) { - /* TODO: check if time to recover was small. It might be that RTO popped - * too soon. - */ + /* Deflate rto */ + tcp_update_rto (tc); + tc->rto_boff = 0; + tc->snd_rxt_ts = 0; + tcp_recovery_off (tc); +} +void +tcp_cc_fastrecovery_exit (tcp_connection_t * tc) +{ tc->cc_algo->recovered (tc); + tc->snd_rxt_bytes = 0; + tc->rcv_dupacks = 0; + tcp_fastrecovery_off (tc); + tcp_fastrecovery_1_smss_off (tc); +} - tc->rtx_bytes = 0; +static void +tcp_cc_congestion_undo (tcp_connection_t * tc) +{ + tc->cwnd = tc->prev_cwnd; + tc->ssthresh = tc->prev_ssthresh; + tc->snd_nxt = tc->snd_una_max; tc->rcv_dupacks = 0; - tc->snd_nxt = tc->snd_una; + if (tcp_in_recovery (tc)) + tcp_cc_recovery_exit (tc); + ASSERT (tc->rto_boff == 0); + /* TODO extend for fastrecovery */ +} - tc->cc_algo->rcv_ack (tc); - tc->tsecr_last_ack = tc->opt.tsecr; +static u8 +tcp_cc_is_spurious_retransmit (tcp_connection_t * tc) +{ + return (tc->snd_rxt_ts + && tcp_opts_tstamp (&tc->rcv_opts) + && timestamp_lt (tc->rcv_opts.tsecr, tc->snd_rxt_ts)); +} - tcp_cong_recovery_off (tc); +int +tcp_cc_recover (tcp_connection_t * tc) +{ + ASSERT (tcp_in_cong_recovery (tc)); + if (tcp_cc_is_spurious_retransmit (tc)) + { + tcp_cc_congestion_undo (tc); + return 1; + } + + if (tcp_in_recovery (tc)) + tcp_cc_recovery_exit (tc); + else if (tcp_in_fastrecovery (tc)) + tcp_cc_fastrecovery_exit (tc); + + ASSERT (tc->rto_boff == 0); + ASSERT (!tcp_in_cong_recovery (tc)); TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 3); + return 0; } static void -tcp_cc_rcv_ack (tcp_connection_t * tc, vlib_buffer_t * b) +tcp_cc_update (tcp_connection_t * tc, vlib_buffer_t * b) +{ + ASSERT (!tcp_in_cong_recovery (tc)); + + /* Congestion avoidance */ + tc->cc_algo->rcv_ack (tc); + tc->tsecr_last_ack = tc->rcv_opts.tsecr; + + /* If a cumulative ack, make sure dupacks is 0 */ + tc->rcv_dupacks = 0; + + /* When dupacks hits the threshold we only enter fast retransmit if + * cumulative ack covers more than snd_congestion. Should snd_una + * wrap this test may fail under otherwise valid circumstances. + * Therefore, proactively update snd_congestion when wrap detected. */ + if (PREDICT_FALSE + (seq_leq (tc->snd_congestion, tc->snd_una - tc->bytes_acked) + && seq_gt (tc->snd_congestion, tc->snd_una))) + tc->snd_congestion = tc->snd_una - 1; +} + +static u8 +tcp_should_fastrecover_sack (tcp_connection_t * tc) { - u8 partial_ack; - u32 bytes_advanced; + return (TCP_DUPACK_THRESHOLD - 1) * tc->snd_mss < tc->sack_sb.sacked_bytes; +} - if (tcp_in_fastrecovery (tc)) +static u8 +tcp_should_fastrecover (tcp_connection_t * tc) +{ + return (tc->rcv_dupacks == TCP_DUPACK_THRESHOLD + || tcp_should_fastrecover_sack (tc)); +} + +static void +tcp_cc_handle_event (tcp_connection_t * tc, u32 is_dack) +{ + /* + * Duplicate ACK. Check if we should enter fast recovery, or if already in + * it account for the bytes that left the network. + */ + if (is_dack) { - partial_ack = seq_lt (tc->snd_una, tc->snd_congestion); - if (!partial_ack) + ASSERT (tc->snd_una != tc->snd_una_max + || tc->sack_sb.last_sacked_bytes); + tc->rcv_dupacks++; + + if (tc->rcv_dupacks > TCP_DUPACK_THRESHOLD && !tc->bytes_acked) { - /* Clear retransmitted bytes. */ - tcp_cc_recover (tc); + ASSERT (tcp_in_fastrecovery (tc)); + /* Pure duplicate ack. If some data got acked, it's handled lower */ + tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK); + return; } - else + else if (tcp_should_fastrecover (tc)) { - TCP_EVT_DBG (TCP_EVT_CC_PACK, tc); + /* Things are already bad */ + if (tcp_in_cong_recovery (tc)) + { + tc->rcv_dupacks = 0; + goto partial_ack_test; + } - /* Clear retransmitted bytes. XXX should we clear all? */ - tc->rtx_bytes = 0; + /* If of of the two conditions lower hold, reset dupacks + * 1) Cumulative ack does not cover more than congestion threshold + * 2) RFC6582 heuristic to avoid multiple fast retransmits + */ + if (seq_leq (tc->snd_una, tc->snd_congestion) + || tc->rcv_opts.tsecr != tc->tsecr_last_ack) + { + tc->rcv_dupacks = 0; + return; + } + + tcp_cc_init_congestion (tc); + tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK); - tc->cc_algo->rcv_cong_ack (tc, TCP_CC_PARTIALACK); + /* The first segment MUST be retransmitted */ + tcp_retransmit_first_unacked (tc); - /* In case snd_nxt is still in the past and output tries to - * shove some new bytes */ - tc->snd_nxt = tc->snd_una_max; + /* Post retransmit update cwnd to ssthresh and account for the + * three segments that have left the network and should've been + * buffered at the receiver XXX */ + tc->cwnd = tc->ssthresh + tc->rcv_dupacks * tc->snd_mss; - /* XXX need proper RFC6675 support */ - if (tc->sack_sb.last_sacked_bytes && !tcp_in_recovery (tc)) + /* If cwnd allows, send more data */ + if (tcp_opts_sack_permitted (&tc->rcv_opts) + && scoreboard_first_hole (&tc->sack_sb)) { - tcp_fast_retransmit (tc); + scoreboard_init_high_rxt (&tc->sack_sb); + tcp_fast_retransmit_sack (tc); } else { - /* Retransmit first unacked segment */ - tcp_retransmit_first_unacked (tc); + tcp_fast_retransmit_no_sack (tc); } + + return; } - } - else - { - tc->cc_algo->rcv_ack (tc); - tc->tsecr_last_ack = tc->opt.tsecr; - tc->rcv_dupacks = 0; - if (tcp_in_recovery (tc)) + else if (!tc->bytes_acked + || (tc->bytes_acked && !tcp_in_cong_recovery (tc))) { - bytes_advanced = tc->bytes_acked + tc->sack_sb.snd_una_adv; - tc->rtx_bytes -= clib_min (bytes_advanced, tc->rtx_bytes); - tc->rto = clib_min (tc->srtt + (tc->rttvar << 2), TCP_RTO_MAX); - if (seq_geq (tc->snd_una, tc->snd_congestion)) - { - tc->rtx_bytes = 0; - tcp_recovery_off (tc); - } + tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK); + return; } + else + goto partial_ack; } -} -static void -tcp_cc_rcv_dupack (tcp_connection_t * tc, u32 ack) -{ -// ASSERT (seq_geq(tc->snd_una, ack)); +partial_ack_test: + + if (!tc->bytes_acked) + return; + +partial_ack: + /* + * Legitimate ACK. 1) See if we can exit recovery + */ + /* XXX limit this only to first partial ack? */ + tcp_retransmit_timer_update (tc); - tc->rcv_dupacks++; - if (tc->rcv_dupacks == TCP_DUPACK_THRESHOLD) + if (seq_geq (tc->snd_una, tc->snd_congestion)) { - /* RFC6582 NewReno heuristic to avoid multiple fast retransmits */ - if (tc->opt.tsecr != tc->tsecr_last_ack) - { - tc->rcv_dupacks = 0; - return; - } + /* If spurious return, we've already updated everything */ + if (tcp_cc_recover (tc)) + return; + + tc->snd_nxt = tc->snd_una_max; - tcp_fastrecovery_on (tc); + /* Treat as congestion avoidance ack */ + tc->cc_algo->rcv_ack (tc); + tc->tsecr_last_ack = tc->rcv_opts.tsecr; + return; + } + + /* + * Legitimate ACK. 2) If PARTIAL ACK try to retransmit + */ + TCP_EVT_DBG (TCP_EVT_CC_PACK, tc); + + /* RFC6675: If the incoming ACK is a cumulative acknowledgment, + * reset dupacks to 0 */ + tc->rcv_dupacks = 0; - /* Handle congestion and dupack */ - tcp_cc_congestion (tc); - tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK); + tcp_retransmit_first_unacked (tc); - tcp_fast_retransmit (tc); + /* Post RTO timeout don't try anything fancy */ + if (tcp_in_recovery (tc)) + return; - /* Post retransmit update cwnd to ssthresh and account for the - * three segments that have left the network and should've been - * buffered at the receiver */ - tc->cwnd = tc->ssthresh + TCP_DUPACK_THRESHOLD * tc->snd_mss; + /* Remove retransmitted bytes that have been delivered */ + if (tc->sack_sb.last_bytes_delivered + && seq_gt (tc->sack_sb.high_rxt, tc->snd_una)) + { + /* If we have sacks and we haven't gotten an ack beyond high_rxt, + * remove sacked bytes delivered */ + tc->snd_rxt_bytes -= tc->sack_sb.last_bytes_delivered; } - else if (tc->rcv_dupacks > TCP_DUPACK_THRESHOLD) + else { - ASSERT (tcp_in_fastrecovery (tc)); - - tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK); + /* Either all retransmitted holes have been acked, or we're + * "in the blind" and retransmitting segment by segment */ + tc->snd_rxt_bytes = 0; } + + tc->cc_algo->rcv_cong_ack (tc, TCP_CC_PARTIALACK); + + /* + * Since this was a partial ack, try to retransmit some more data + */ + tcp_fast_retransmit (tc); } void @@ -862,14 +1144,18 @@ tcp_cc_init (tcp_connection_t * tc) tc->cc_algo->init (tc); } +/** + * Process incoming ACK + */ static int tcp_rcv_ack (tcp_connection_t * tc, vlib_buffer_t * b, tcp_header_t * th, u32 * next, u32 * error) { - u32 new_snd_wnd; + u32 prev_snd_wnd, prev_snd_una; + u8 is_dack; /* If the ACK acks something not yet sent (SEG.ACK > SND.NXT) */ - if (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_nxt)) + if (PREDICT_FALSE (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_nxt))) { /* If we have outstanding data and this is within the window, accept it, * probably retransmit has timed out. Otherwise ACK segment and then @@ -892,7 +1178,7 @@ tcp_rcv_ack (tcp_connection_t * tc, vlib_buffer_t * b, } /* If old ACK, probably it's an old dupack */ - if (seq_lt (vnet_buffer (b)->tcp.ack_number, tc->snd_una)) + if (PREDICT_FALSE (seq_lt (vnet_buffer (b)->tcp.ack_number, tc->snd_una))) { *error = TCP_ERROR_ACK_OLD; TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 1, @@ -900,54 +1186,50 @@ tcp_rcv_ack (tcp_connection_t * tc, vlib_buffer_t * b, if (tcp_in_fastrecovery (tc) && tc->rcv_dupacks == TCP_DUPACK_THRESHOLD) { TCP_EVT_DBG (TCP_EVT_DUPACK_RCVD, tc); - tcp_cc_rcv_dupack (tc, vnet_buffer (b)->tcp.ack_number); + tcp_cc_handle_event (tc, 1); } /* Don't drop yet */ return 0; } - if (tcp_opts_sack_permitted (&tc->opt)) - tcp_rcv_sacks (tc, vnet_buffer (b)->tcp.ack_number); - - new_snd_wnd = clib_net_to_host_u16 (th->window) << tc->snd_wscale; - - if (tcp_ack_is_dupack (tc, b, new_snd_wnd)) - { - TCP_EVT_DBG (TCP_EVT_DUPACK_RCVD, tc, 1); - tcp_cc_rcv_dupack (tc, vnet_buffer (b)->tcp.ack_number); - *error = TCP_ERROR_ACK_DUP; - return -1; - } - /* - * Valid ACK + * Looks okay, process feedback */ - tc->bytes_acked = vnet_buffer (b)->tcp.ack_number - tc->snd_una; - tc->snd_una = vnet_buffer (b)->tcp.ack_number + tc->sack_sb.snd_una_adv; + TCP_EVT_DBG (TCP_EVT_ACK_RCVD, tc); + + if (tcp_opts_sack_permitted (&tc->rcv_opts)) + tcp_rcv_sacks (tc, vnet_buffer (b)->tcp.ack_number); - /* Dequeue ACKed data and update RTT */ - tcp_dequeue_acked (tc, vnet_buffer (b)->tcp.ack_number); + prev_snd_wnd = tc->snd_wnd; + prev_snd_una = tc->snd_una; tcp_update_snd_wnd (tc, vnet_buffer (b)->tcp.seq_number, - vnet_buffer (b)->tcp.ack_number, new_snd_wnd); + vnet_buffer (b)->tcp.ack_number, + clib_net_to_host_u16 (th->window) << tc->snd_wscale); + tc->bytes_acked = vnet_buffer (b)->tcp.ack_number - tc->snd_una; + tc->snd_una = vnet_buffer (b)->tcp.ack_number + tc->sack_sb.snd_una_adv; + tcp_validate_txf_size (tc, tc->bytes_acked); - /* If some of our sent bytes have been acked, update cc and retransmit - * timer. */ if (tc->bytes_acked) - { - TCP_EVT_DBG (TCP_EVT_ACK_RCVD, tc); + tcp_dequeue_acked (tc, vnet_buffer (b)->tcp.ack_number); - /* Updates congestion control (slow start/congestion avoidance) */ - tcp_cc_rcv_ack (tc, b); + /* + * Check if we have congestion event + */ - /* If everything has been acked, stop retransmit timer - * otherwise update. */ - if (tc->snd_una == tc->snd_una_max) - tcp_retransmit_timer_reset (tc); - else - tcp_retransmit_timer_update (tc); + if (tcp_ack_is_cc_event (tc, b, prev_snd_wnd, prev_snd_una, &is_dack)) + { + tcp_cc_handle_event (tc, is_dack); + *error = TCP_ERROR_ACK_DUP; + TCP_EVT_DBG (TCP_EVT_DUPACK_RCVD, tc, 1); + return vnet_buffer (b)->tcp.data_len ? 0 : -1; } + /* + * Update congestion control (slow start/congestion avoidance) + */ + tcp_cc_update (tc, b); + return 0; } @@ -1059,7 +1341,7 @@ tcp_session_enqueue_data (tcp_connection_t * tc, vlib_buffer_t * b, } /* Update SACK list if need be */ - if (tcp_opts_sack_permitted (&tc->opt)) + if (tcp_opts_sack_permitted (&tc->rcv_opts)) { /* Remove SACK blocks that have been delivered */ tcp_update_sack_list (tc, tc->rcv_nxt, tc->rcv_nxt); @@ -1097,7 +1379,7 @@ tcp_session_enqueue_ooo (tcp_connection_t * tc, vlib_buffer_t * b, TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, data_len); /* Update SACK list if in use */ - if (tcp_opts_sack_permitted (&tc->opt)) + if (tcp_opts_sack_permitted (&tc->rcv_opts)) { ooo_segment_t *newest; u32 start, end; @@ -1294,7 +1576,6 @@ tcp46_established_inline (vlib_main_t * vm, vlib_node_runtime_t * node, u32 n_left_to_next; vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); - while (n_left_from > 0 && n_left_to_next > 0) { u32 bi0; @@ -1321,7 +1602,6 @@ tcp46_established_inline (vlib_main_t * vm, vlib_node_runtime_t * node, } th0 = tcp_buffer_hdr (b0); - is_fin = (th0->flags & TCP_FLAG_FIN) != 0; /* SYNs, FINs and data consume sequence numbers */ @@ -1387,7 +1667,6 @@ tcp46_established_inline (vlib_main_t * vm, vlib_node_runtime_t * node, errors = session_manager_flush_enqueue_events (my_thread_index); tcp_established_inc_counter (vm, is_ip4, TCP_ERROR_EVENT_FIFO_FULL, errors); - return from_frame->n_vectors; } @@ -1582,17 +1861,17 @@ tcp46_syn_sent_inline (vlib_main_t * vm, vlib_node_runtime_t * node, new_tc0->irs = seq0; /* Parse options */ - if (tcp_options_parse (tcp0, &new_tc0->opt)) + if (tcp_options_parse (tcp0, &new_tc0->rcv_opts)) goto drop; - if (tcp_opts_tstamp (&new_tc0->opt)) + if (tcp_opts_tstamp (&new_tc0->rcv_opts)) { - new_tc0->tsval_recent = new_tc0->opt.tsval; + new_tc0->tsval_recent = new_tc0->rcv_opts.tsval; new_tc0->tsval_recent_age = tcp_time_now (); } - if (tcp_opts_wscale (&new_tc0->opt)) - new_tc0->snd_wscale = new_tc0->opt.wscale; + if (tcp_opts_wscale (&new_tc0->rcv_opts)) + new_tc0->snd_wscale = new_tc0->rcv_opts.wscale; /* No scaling */ new_tc0->snd_wnd = clib_net_to_host_u16 (tcp0->window); @@ -1845,7 +2124,7 @@ tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node, /* Initialize session variables */ tc0->snd_una = vnet_buffer (b0)->tcp.ack_number; tc0->snd_wnd = clib_net_to_host_u16 (tcp0->window) - << tc0->opt.wscale; + << tc0->rcv_opts.wscale; tc0->snd_wl1 = vnet_buffer (b0)->tcp.seq_number; tc0->snd_wl2 = vnet_buffer (b0)->tcp.ack_number; @@ -1903,13 +2182,21 @@ tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node, break; case TCP_STATE_LAST_ACK: - /* The only thing that can arrive in this state is an + /* The only thing that [should] arrive in this state is an * acknowledgment of our FIN. If our FIN is now acknowledged, * delete the TCB, enter the CLOSED state, and return. */ if (!tcp_rcv_ack_is_acceptable (tc0, b0)) goto drop; + /* Apparently our FIN was lost */ + if (tcp_fin (tcp0)) + { + /* Don't "make" fin since that increments snd_nxt */ + tcp_send_fin (tc0); + goto drop; + } + tc0->state = TCP_STATE_CLOSED; /* Don't delete the connection/session yet. Instead, wait a @@ -1929,8 +2216,15 @@ tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node, * retransmission of the remote FIN. Acknowledge it, and restart * the 2 MSL timeout. */ - /* TODO */ + if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) + goto drop; + + tcp_make_ack (tc0, b0); + tcp_timer_reset (tc0, TCP_TIMER_WAITCLOSE); + tcp_timer_set (tc0, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME); + goto drop; + break; default: ASSERT (0); @@ -2194,7 +2488,7 @@ tcp46_listen_inline (vlib_main_t * vm, vlib_node_runtime_t * node, goto drop; } - if (tcp_options_parse (th0, &child0->opt)) + if (tcp_options_parse (th0, &child0->rcv_opts)) { goto drop; } @@ -2205,14 +2499,14 @@ tcp46_listen_inline (vlib_main_t * vm, vlib_node_runtime_t * node, /* RFC1323: TSval timestamps sent on {SYN} and {SYN,ACK} * segments are used to initialize PAWS. */ - if (tcp_opts_tstamp (&child0->opt)) + if (tcp_opts_tstamp (&child0->rcv_opts)) { - child0->tsval_recent = child0->opt.tsval; + child0->tsval_recent = child0->rcv_opts.tsval; child0->tsval_recent_age = tcp_time_now (); } - if (tcp_opts_wscale (&child0->opt)) - child0->snd_wscale = child0->opt.wscale; + if (tcp_opts_wscale (&child0->rcv_opts)) + child0->snd_wscale = child0->rcv_opts.wscale; /* No scaling */ child0->snd_wnd = clib_net_to_host_u16 (th0->window); @@ -2477,7 +2771,6 @@ tcp46_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_add_trace (vm, node, b0, sizeof (*t0)); tcp_set_rx_trace_data (t0, tc0, tcp0, b0, is_ip4); } - vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, n_left_to_next, bi0, next0); } @@ -2600,7 +2893,13 @@ do { \ _(FIN_WAIT_2, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); _(LAST_ACK, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); + _(LAST_ACK, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); + _(LAST_ACK, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, + TCP_ERROR_NONE); _(LAST_ACK, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); + _(TIME_WAIT, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); + _(TIME_WAIT, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, + TCP_ERROR_NONE); _(CLOSED, TCP_FLAG_ACK, TCP_INPUT_NEXT_RESET, TCP_ERROR_CONNECTION_CLOSED); _(CLOSED, TCP_FLAG_RST, TCP_INPUT_NEXT_DROP, TCP_ERROR_CONNECTION_CLOSED); #undef _ -- cgit 1.2.3-korg