diff options
author | Florin Coras <fcoras@cisco.com> | 2017-05-24 18:03:56 -0700 |
---|---|---|
committer | Dave Barach <openvpp@barachs.net> | 2017-06-09 16:47:19 +0000 |
commit | 93992a9048cb6e5dcd22de5091e72de778122627 (patch) | |
tree | f295bb870ad1614d9327828dd036fb03f60e455e /src/vnet/tcp/tcp_output.c | |
parent | 583dc8d3e23a780c85ebe48ea59f0338aad4df17 (diff) |
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 <fcoras@cisco.com>
Signed-off-by: Dave Barach <dbarach@cisco.com>
Diffstat (limited to 'src/vnet/tcp/tcp_output.c')
-rw-r--r-- | src/vnet/tcp/tcp_output.c | 287 |
1 files changed, 189 insertions, 98 deletions
diff --git a/src/vnet/tcp/tcp_output.c b/src/vnet/tcp/tcp_output.c index 49fd6beffa3..47c94e6daca 100644 --- a/src/vnet/tcp/tcp_output.c +++ b/src/vnet/tcp/tcp_output.c @@ -136,10 +136,10 @@ tcp_update_rcv_wnd (tcp_connection_t * tc) * Figure out how much space we have available */ available_space = stream_session_max_rx_enqueue (&tc->connection); - max_fifo = stream_session_fifo_size (&tc->connection); + max_fifo = stream_session_rx_fifo_size (&tc->connection); - ASSERT (tc->opt.mss < max_fifo); - if (available_space < tc->opt.mss && available_space < max_fifo >> 3) + ASSERT (tc->rcv_opts.mss < max_fifo); + if (available_space < tc->rcv_opts.mss && available_space < max_fifo >> 3) available_space = 0; /* @@ -276,8 +276,11 @@ tcp_make_syn_options (tcp_options_t * opts, u8 wnd_scale) opts->tsecr = 0; len += TCP_OPTION_LEN_TIMESTAMP; - opts->flags |= TCP_OPTS_FLAG_SACK_PERMITTED; - len += TCP_OPTION_LEN_SACK_PERMITTED; + if (TCP_USE_SACKS) + { + opts->flags |= TCP_OPTS_FLAG_SACK_PERMITTED; + len += TCP_OPTION_LEN_SACK_PERMITTED; + } /* Align to needed boundary */ len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN; @@ -293,14 +296,14 @@ tcp_make_synack_options (tcp_connection_t * tc, tcp_options_t * opts) opts->mss = tc->mss; len += TCP_OPTION_LEN_MSS; - if (tcp_opts_wscale (&tc->opt)) + if (tcp_opts_wscale (&tc->rcv_opts)) { opts->flags |= TCP_OPTS_FLAG_WSCALE; opts->wscale = tc->rcv_wscale; len += TCP_OPTION_LEN_WINDOW_SCALE; } - if (tcp_opts_tstamp (&tc->opt)) + if (tcp_opts_tstamp (&tc->rcv_opts)) { opts->flags |= TCP_OPTS_FLAG_TSTAMP; opts->tsval = tcp_time_now (); @@ -308,7 +311,7 @@ tcp_make_synack_options (tcp_connection_t * tc, tcp_options_t * opts) len += TCP_OPTION_LEN_TIMESTAMP; } - if (tcp_opts_sack_permitted (&tc->opt)) + if (tcp_opts_sack_permitted (&tc->rcv_opts)) { opts->flags |= TCP_OPTS_FLAG_SACK_PERMITTED; len += TCP_OPTION_LEN_SACK_PERMITTED; @@ -326,14 +329,14 @@ tcp_make_established_options (tcp_connection_t * tc, tcp_options_t * opts) opts->flags = 0; - if (tcp_opts_tstamp (&tc->opt)) + if (tcp_opts_tstamp (&tc->rcv_opts)) { opts->flags |= TCP_OPTS_FLAG_TSTAMP; opts->tsval = tcp_time_now (); opts->tsecr = tc->tsval_recent; len += TCP_OPTION_LEN_TIMESTAMP; } - if (tcp_opts_sack_permitted (&tc->opt)) + if (tcp_opts_sack_permitted (&tc->rcv_opts)) { if (vec_len (tc->snd_sacks)) { @@ -395,7 +398,7 @@ tcp_update_snd_mss (tcp_connection_t * tc) tcp_make_options (tc, &tc->snd_opts, TCP_STATE_ESTABLISHED); /* XXX check if MTU has been updated */ - tc->snd_mss = clib_min (tc->mss, tc->opt.mss) - tc->snd_opts_len; + tc->snd_mss = clib_min (tc->mss, tc->rcv_opts.mss) - tc->snd_opts_len; ASSERT (tc->snd_mss > 0); } @@ -406,21 +409,21 @@ tcp_init_mss (tcp_connection_t * tc) tcp_update_rcv_mss (tc); /* TODO cache mss and consider PMTU discovery */ - tc->snd_mss = clib_min (tc->opt.mss, tc->mss); + tc->snd_mss = clib_min (tc->rcv_opts.mss, tc->mss); if (tc->snd_mss < 45) { clib_warning ("snd mss is 0"); /* Assume that at least the min default mss works */ tc->snd_mss = default_min_mss; - tc->opt.mss = default_min_mss; + tc->rcv_opts.mss = default_min_mss; } /* We should have enough space for 40 bytes of options */ ASSERT (tc->snd_mss > 45); /* If we use timestamp option, account for it */ - if (tcp_opts_tstamp (&tc->opt)) + if (tcp_opts_tstamp (&tc->rcv_opts)) tc->snd_mss -= TCP_OPTION_LEN_TIMESTAMP; } @@ -879,6 +882,7 @@ tcp_send_fin (tcp_connection_t * tc) tcp_make_fin (tc, b); tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4); tc->flags |= TCP_CONN_FINSNT; + tcp_retransmit_timer_force_update (tc); TCP_EVT_DBG (TCP_EVT_FIN_SENT, tc); } @@ -919,10 +923,7 @@ tcp_push_hdr_i (tcp_connection_t * tc, vlib_buffer_t * b, if (compute_opts) tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state); - /* Write pre-computed options */ tcp_hdr_opts_len = tc->snd_opts_len + sizeof (tcp_header_t); - - /* Get rcv window to advertise */ advertise_wnd = tcp_window_to_advertise (tc, next_state); flags = tcp_make_state_flags (next_state); @@ -930,26 +931,25 @@ tcp_push_hdr_i (tcp_connection_t * tc, vlib_buffer_t * b, th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->snd_nxt, tc->rcv_nxt, tcp_hdr_opts_len, flags, advertise_wnd); - opts_write_len = tcp_options_write ((u8 *) (th + 1), &tc->snd_opts); ASSERT (opts_write_len == tc->snd_opts_len); - - /* Tag the buffer with the connection index */ vnet_buffer (b)->tcp.connection_index = tc->c_c_index; + /* + * Update connection variables + */ + tc->snd_nxt += data_len; tc->rcv_las = tc->rcv_nxt; /* TODO this is updated in output as well ... */ - if (tc->snd_nxt > tc->snd_una_max) - tc->snd_una_max = tc->snd_nxt; - - if (tc->rtt_ts == 0) + if (seq_gt (tc->snd_nxt, tc->snd_una_max)) { - tc->rtt_ts = tcp_time_now (); - tc->rtt_seq = tc->snd_nxt; + tc->snd_una_max = tc->snd_nxt; + tcp_validate_txf_size (tc, tc->snd_una_max - tc->snd_una); } + TCP_EVT_DBG (TCP_EVT_PKTIZE, tc); } @@ -987,13 +987,14 @@ tcp_timer_delack_handler (u32 index) * * @return the number of bytes in the segment or 0 if there's nothing to * retransmit - * */ + */ u32 tcp_prepare_retransmit_segment (tcp_connection_t * tc, vlib_buffer_t * b, u32 offset, u32 max_bytes) { vlib_main_t *vm = vlib_get_main (); - u32 n_bytes = 0; + int n_bytes = 0; + u32 start; tcp_reuse_buffer (vm, b); @@ -1001,15 +1002,16 @@ tcp_prepare_retransmit_segment (tcp_connection_t * tc, vlib_buffer_t * b, ASSERT (max_bytes != 0); max_bytes = clib_min (tc->snd_mss, max_bytes); + start = tc->snd_una + offset; /* Start is beyond snd_congestion */ - if (seq_geq (tc->snd_una + offset, tc->snd_congestion)) + if (seq_geq (start, tc->snd_congestion)) goto done; /* Don't overshoot snd_congestion */ - if (seq_gt (tc->snd_nxt + max_bytes, tc->snd_congestion)) + if (seq_gt (start + max_bytes, tc->snd_congestion)) { - max_bytes = tc->snd_congestion - tc->snd_nxt; + max_bytes = tc->snd_congestion - start; if (max_bytes == 0) goto done; } @@ -1021,15 +1023,12 @@ tcp_prepare_retransmit_segment (tcp_connection_t * tc, vlib_buffer_t * b, n_bytes = stream_session_peek_bytes (&tc->connection, vlib_buffer_get_current (b), offset, max_bytes); - ASSERT (n_bytes != 0); + ASSERT (n_bytes > 0); b->current_length = n_bytes; tcp_push_hdr_i (tc, b, tc->state, 0); - /* Don't count multiple retransmits of the same segment */ - if (tc->rto_boff > 1) - goto done; - - tc->rtx_bytes += n_bytes; + if (tcp_in_fastrecovery (tc)) + tc->snd_rxt_bytes += n_bytes; done: TCP_EVT_DBG (TCP_EVT_CC_RTX, tc, offset, n_bytes); @@ -1042,18 +1041,15 @@ done: static void tcp_rtx_timeout_cc (tcp_connection_t * tc) { + tc->prev_ssthresh = tc->ssthresh; + tc->prev_cwnd = tc->cwnd; + /* Cleanly recover cc (also clears up fast retransmit) */ if (tcp_in_fastrecovery (tc)) - { - tcp_cc_recover (tc); - } - else - { - tc->ssthresh = clib_max (tcp_flight_size (tc) / 2, 2 * tc->snd_mss); - } + tcp_cc_fastrecovery_exit (tc); /* Start again from the beginning */ - + tc->ssthresh = clib_max (tcp_flight_size (tc) / 2, 2 * tc->snd_mss); tc->cwnd = tcp_loss_wnd (tc); tc->snd_congestion = tc->snd_una_max; tcp_recovery_on (tc); @@ -1081,18 +1077,31 @@ tcp_timer_retransmit_handler_i (u32 index, u8 is_syn) /* Make sure timer handle is set to invalid */ tc->timers[TCP_TIMER_RETRANSMIT] = TCP_TIMER_HANDLE_INVALID; + if (!tcp_in_recovery (tc) && tc->rto_boff > 0 + && tc->state >= TCP_STATE_ESTABLISHED) + { + tc->rto_boff = 0; + tcp_update_rto (tc); + } + /* Increment RTO backoff (also equal to number of retries) */ tc->rto_boff += 1; /* Go back to first un-acked byte */ tc->snd_nxt = tc->snd_una; - /* Get buffer */ tcp_get_free_buffer_index (tm, &bi); b = vlib_get_buffer (vm, bi); if (tc->state >= TCP_STATE_ESTABLISHED) { + /* Lost FIN, retransmit and return */ + if (tc->flags & TCP_CONN_FINSNT) + { + tcp_send_fin (tc); + return; + } + /* First retransmit timeout */ if (tc->rto_boff == 1) tcp_rtx_timeout_cc (tc); @@ -1102,24 +1111,30 @@ tcp_timer_retransmit_handler_i (u32 index, u8 is_syn) TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 1); - /* Send one segment. No fancy recovery for now! */ + /* Send one segment */ n_bytes = tcp_prepare_retransmit_segment (tc, b, 0, tc->snd_mss); + /* TODO be less aggressive about this */ scoreboard_clear (&tc->sack_sb); if (n_bytes == 0) { clib_warning ("could not retransmit anything"); + clib_warning ("%U", format_tcp_connection, tc, 2); + /* Try again eventually */ tcp_retransmit_timer_set (tc); + ASSERT (0 || (tc->rto_boff > 1 + && tc->snd_una == tc->snd_congestion)); return; } + + /* For first retransmit, record timestamp (Eifel detection RFC3522) */ + if (tc->rto_boff == 1) + tc->snd_rxt_ts = tcp_time_now (); } - else + /* Retransmit for SYN/SYNACK */ + else if (tc->state == TCP_STATE_SYN_RCVD || tc->state == TCP_STATE_SYN_SENT) { - /* Retransmit for SYN/SYNACK */ - ASSERT (tc->state == TCP_STATE_SYN_RCVD - || tc->state == TCP_STATE_SYN_SENT); - /* Try without increasing RTO a number of times. If this fails, * start growing RTO exponentially */ if (tc->rto_boff > TCP_RTO_SYN_RETRIES) @@ -1132,6 +1147,12 @@ tcp_timer_retransmit_handler_i (u32 index, u8 is_syn) /* Account for the SYN */ tc->snd_nxt += 1; } + else + { + ASSERT (tc->state == TCP_STATE_CLOSED); + clib_warning ("connection closed ..."); + return; + } if (!is_syn) { @@ -1180,7 +1201,8 @@ tcp_timer_persist_handler (u32 index) u32 thread_index = vlib_get_thread_index (); tcp_connection_t *tc; vlib_buffer_t *b; - u32 bi, n_bytes; + u32 bi, old_snd_nxt; + int n_bytes = 0; tc = tcp_connection_get_if_valid (index, thread_index); @@ -1202,13 +1224,15 @@ tcp_timer_persist_handler (u32 index) /* Try to force the first unsent segment */ tcp_get_free_buffer_index (tm, &bi); b = vlib_get_buffer (vm, bi); + + tcp_validate_txf_size (tc, tc->snd_una_max - tc->snd_una); tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state); n_bytes = stream_session_peek_bytes (&tc->connection, vlib_buffer_get_current (b), tc->snd_una_max - tc->snd_una, tc->snd_mss); /* Nothing to send */ - if (n_bytes == 0) + if (n_bytes <= 0) { clib_warning ("persist found nothing to send"); tcp_return_buffer (tm); @@ -1216,7 +1240,13 @@ tcp_timer_persist_handler (u32 index) } b->current_length = n_bytes; + ASSERT (tc->snd_nxt == tc->snd_una_max || tc->rto_boff > 1 + || tcp_timer_is_active (tc, TCP_TIMER_RETRANSMIT)); + + /* Allow updating of snd_una_max but don't update snd_nxt */ + old_snd_nxt = tc->snd_nxt; tcp_push_hdr_i (tc, b, tc->state, 0); + tc->snd_nxt = old_snd_nxt; tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4); /* Re-enable persist timer */ @@ -1232,8 +1262,9 @@ tcp_retransmit_first_unacked (tcp_connection_t * tc) tcp_main_t *tm = vnet_get_tcp_main (); vlib_main_t *vm = vlib_get_main (); vlib_buffer_t *b; - u32 bi, n_bytes; + u32 bi, n_bytes, old_snd_nxt; + old_snd_nxt = tc->snd_nxt; tc->snd_nxt = tc->snd_una; /* Get buffer */ @@ -1244,75 +1275,117 @@ tcp_retransmit_first_unacked (tcp_connection_t * tc) n_bytes = tcp_prepare_retransmit_segment (tc, b, 0, tc->snd_mss); if (n_bytes == 0) - goto done; + { + tcp_return_buffer (tm); + goto done; + } tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4); done: - tc->snd_nxt = tc->snd_una_max; + tc->snd_nxt = old_snd_nxt; } -sack_scoreboard_hole_t * -scoreboard_first_rtx_hole (sack_scoreboard_t * sb) +/** + * Do fast retransmit with SACKs + */ +void +tcp_fast_retransmit_sack (tcp_connection_t * tc) { - sack_scoreboard_hole_t *hole = 0; - -// hole = scoreboard_first_hole (&tc->sack_sb); -// if (hole) -// { -// -// offset = hole->start - tc->snd_una; -// hole_size = hole->end - hole->start; -// -// ASSERT(hole_size); -// -// if (hole_size < max_bytes) -// max_bytes = hole_size; -// } - return hole; + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = vlib_get_main (); + u32 n_written = 0, offset = 0, max_bytes; + vlib_buffer_t *b; + sack_scoreboard_hole_t *hole; + sack_scoreboard_t *sb; + u32 bi, old_snd_nxt; + int snd_space; + u8 snd_limited = 0, can_rescue = 0; + + ASSERT (tcp_in_fastrecovery (tc)); + TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 0); + + old_snd_nxt = tc->snd_nxt; + sb = &tc->sack_sb; + snd_space = tcp_available_snd_space (tc); + + hole = scoreboard_get_hole (sb, sb->cur_rxt_hole); + while (hole && snd_space > 0) + { + tcp_get_free_buffer_index (tm, &bi); + b = vlib_get_buffer (vm, bi); + + hole = scoreboard_next_rxt_hole (sb, hole, + tcp_fastrecovery_sent_1_smss (tc), + &can_rescue, &snd_limited); + if (!hole) + { + if (!can_rescue || !(seq_lt (sb->rescue_rxt, tc->snd_una) + || seq_gt (sb->rescue_rxt, + tc->snd_congestion))) + break; + + /* If rescue rxt undefined or less than snd_una then one segment of + * up to SMSS octets that MUST include the highest outstanding + * unSACKed sequence number SHOULD be returned, and RescueRxt set to + * RecoveryPoint. HighRxt MUST NOT be updated. + */ + max_bytes = clib_min (tc->snd_mss, snd_space); + offset = tc->snd_congestion - tc->snd_una - max_bytes; + sb->rescue_rxt = tc->snd_congestion; + tc->snd_nxt = tc->snd_una + offset; + tcp_prepare_retransmit_segment (tc, b, offset, max_bytes); + tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4); + break; + } + + max_bytes = snd_limited ? tc->snd_mss : hole->end - sb->high_rxt; + offset = sb->high_rxt - tc->snd_una; + tc->snd_nxt = tc->snd_una + offset; + n_written = tcp_prepare_retransmit_segment (tc, b, offset, max_bytes); + + /* Nothing left to retransmit */ + if (n_written == 0) + { + tcp_return_buffer (tm); + break; + } + + sb->high_rxt += n_written; + tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4); + snd_space -= n_written; + } + + /* If window allows, send 1 SMSS of new data */ + tc->snd_nxt = old_snd_nxt; } /** - * Do fast retransmit. + * Fast retransmit without SACK info */ void -tcp_fast_retransmit (tcp_connection_t * tc) +tcp_fast_retransmit_no_sack (tcp_connection_t * tc) { tcp_main_t *tm = vnet_get_tcp_main (); vlib_main_t *vm = vlib_get_main (); - u32 bi; + u32 n_written = 0, offset = 0, bi, old_snd_nxt; int snd_space; - u32 n_written = 0, offset = 0; vlib_buffer_t *b; - u8 use_sacks = 0; ASSERT (tcp_in_fastrecovery (tc)); + TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 0); /* Start resending from first un-acked segment */ + old_snd_nxt = tc->snd_nxt; tc->snd_nxt = tc->snd_una; - snd_space = tcp_available_snd_space (tc); - TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 0); - - /* If we have SACKs use them */ - if (tcp_opts_sack_permitted (&tc->opt) - && scoreboard_first_hole (&tc->sack_sb)) - use_sacks = 0; while (snd_space > 0) { tcp_get_free_buffer_index (tm, &bi); b = vlib_get_buffer (vm, bi); - if (use_sacks) - { - scoreboard_first_rtx_hole (&tc->sack_sb); - } - else - { - offset += n_written; - } - + offset += n_written; n_written = tcp_prepare_retransmit_segment (tc, b, offset, snd_space); /* Nothing left to retransmit */ @@ -1326,9 +1399,21 @@ tcp_fast_retransmit (tcp_connection_t * tc) snd_space -= n_written; } - /* If window allows, send 1 SMSS of new data */ - if (seq_lt (tc->snd_nxt, tc->snd_congestion)) - tc->snd_nxt = tc->snd_congestion; + /* Restore snd_nxt. If window allows, send 1 SMSS of new data */ + tc->snd_nxt = old_snd_nxt; +} + +/** + * Do fast retransmit + */ +void +tcp_fast_retransmit (tcp_connection_t * tc) +{ + if (tcp_opts_sack_permitted (&tc->rcv_opts) + && scoreboard_first_hole (&tc->sack_sb)) + tcp_fast_retransmit_sack (tc); + else + tcp_fast_retransmit_no_sack (tc); } always_inline u32 @@ -1544,6 +1629,12 @@ tcp_push_header (transport_connection_t * tconn, vlib_buffer_t * b) tc = (tcp_connection_t *) tconn; tcp_push_hdr_i (tc, b, TCP_STATE_ESTABLISHED, 0); + + if (tc->rtt_ts == 0) + { + tc->rtt_ts = tcp_time_now (); + tc->rtt_seq = tc->snd_nxt; + } return 0; } |