blob: 154b9ac836382c46c366757571de1d39f6a445a3 [file] [log] [blame]
Dave Barach68b0fb02017-02-28 15:15:56 -05001/*
2 * Copyright (c) 2016 Cisco and/or its affiliates.
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
6 *
7 * http://www.apache.org/licenses/LICENSE-2.0
8 *
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
14 */
15
16#include <vppinfra/sparse_vec.h>
17#include <vnet/tcp/tcp_packet.h>
18#include <vnet/tcp/tcp.h>
19#include <vnet/session/session.h>
20#include <math.h>
21
22static char *tcp_error_strings[] = {
23#define tcp_error(n,s) s,
24#include <vnet/tcp/tcp_error.def>
25#undef tcp_error
26};
27
28/* All TCP nodes have the same outgoing arcs */
29#define foreach_tcp_state_next \
Vijayabhaskar Katamreddyce074122017-11-15 13:50:26 -080030 _ (DROP4, "ip4-drop") \
31 _ (DROP6, "ip6-drop") \
Dave Barach68b0fb02017-02-28 15:15:56 -050032 _ (TCP4_OUTPUT, "tcp4-output") \
33 _ (TCP6_OUTPUT, "tcp6-output")
34
35typedef enum _tcp_established_next
36{
37#define _(s,n) TCP_ESTABLISHED_NEXT_##s,
38 foreach_tcp_state_next
39#undef _
40 TCP_ESTABLISHED_N_NEXT,
41} tcp_established_next_t;
42
43typedef enum _tcp_rcv_process_next
44{
45#define _(s,n) TCP_RCV_PROCESS_NEXT_##s,
46 foreach_tcp_state_next
47#undef _
48 TCP_RCV_PROCESS_N_NEXT,
49} tcp_rcv_process_next_t;
50
51typedef enum _tcp_syn_sent_next
52{
53#define _(s,n) TCP_SYN_SENT_NEXT_##s,
54 foreach_tcp_state_next
55#undef _
56 TCP_SYN_SENT_N_NEXT,
57} tcp_syn_sent_next_t;
58
59typedef enum _tcp_listen_next
60{
61#define _(s,n) TCP_LISTEN_NEXT_##s,
62 foreach_tcp_state_next
63#undef _
64 TCP_LISTEN_N_NEXT,
65} tcp_listen_next_t;
66
67/* Generic, state independent indices */
68typedef enum _tcp_state_next
69{
70#define _(s,n) TCP_NEXT_##s,
71 foreach_tcp_state_next
72#undef _
73 TCP_STATE_N_NEXT,
74} tcp_state_next_t;
75
76#define tcp_next_output(is_ip4) (is_ip4 ? TCP_NEXT_TCP4_OUTPUT \
77 : TCP_NEXT_TCP6_OUTPUT)
78
Vijayabhaskar Katamreddyce074122017-11-15 13:50:26 -080079#define tcp_next_drop(is_ip4) (is_ip4 ? TCP_NEXT_DROP4 \
80 : TCP_NEXT_DROP6)
81
Dave Barach68b0fb02017-02-28 15:15:56 -050082vlib_node_registration_t tcp4_established_node;
83vlib_node_registration_t tcp6_established_node;
84
85/**
86 * Validate segment sequence number. As per RFC793:
87 *
88 * Segment Receive Test
89 * Length Window
90 * ------- ------- -------------------------------------------
91 * 0 0 SEG.SEQ = RCV.NXT
92 * 0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND
93 * >0 0 not acceptable
94 * >0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND
95 * or RCV.NXT =< SEG.SEQ+SEG.LEN-1 < RCV.NXT+RCV.WND
96 *
97 * This ultimately consists in checking if segment falls within the window.
98 * The one important difference compared to RFC793 is that we use rcv_las,
99 * or the rcv_nxt at last ack sent instead of rcv_nxt since that's the
100 * peer's reference when computing our receive window.
101 *
Florin Coras6792ec02017-03-13 03:49:51 -0700102 * This:
103 * seq_leq (end_seq, tc->rcv_las + tc->rcv_wnd) && seq_geq (seq, tc->rcv_las)
104 * however, is too strict when we have retransmits. Instead we just check that
105 * the seq is not beyond the right edge and that the end of the segment is not
106 * less than the left edge.
107 *
108 * N.B. rcv_nxt and rcv_wnd are both updated in this node if acks are sent, so
109 * use rcv_nxt in the right edge window test instead of rcv_las.
110 *
Dave Barach68b0fb02017-02-28 15:15:56 -0500111 */
112always_inline u8
113tcp_segment_in_rcv_wnd (tcp_connection_t * tc, u32 seq, u32 end_seq)
114{
Florin Coras6792ec02017-03-13 03:49:51 -0700115 return (seq_geq (end_seq, tc->rcv_las)
116 && seq_leq (seq, tc->rcv_nxt + tc->rcv_wnd));
Dave Barach68b0fb02017-02-28 15:15:56 -0500117}
118
Florin Corasdb84e572017-05-09 18:54:52 -0700119/**
120 * Parse TCP header options.
121 *
122 * @param th TCP header
123 * @param to TCP options data structure to be populated
124 * @return -1 if parsing failed
125 */
Florin Coras0dbd5172018-06-25 16:19:34 -0700126static int
Dave Barach68b0fb02017-02-28 15:15:56 -0500127tcp_options_parse (tcp_header_t * th, tcp_options_t * to)
128{
129 const u8 *data;
130 u8 opt_len, opts_len, kind;
131 int j;
132 sack_block_t b;
133
134 opts_len = (tcp_doff (th) << 2) - sizeof (tcp_header_t);
135 data = (const u8 *) (th + 1);
136
137 /* Zero out all flags but those set in SYN */
Florin Corasd6fe5bd2018-10-16 19:52:10 -0700138 to->flags &= (TCP_OPTS_FLAG_SACK_PERMITTED | TCP_OPTS_FLAG_WSCALE
139 | TCP_OPTS_FLAG_SACK);
Dave Barach68b0fb02017-02-28 15:15:56 -0500140
141 for (; opts_len > 0; opts_len -= opt_len, data += opt_len)
142 {
143 kind = data[0];
144
145 /* Get options length */
146 if (kind == TCP_OPTION_EOL)
147 break;
148 else if (kind == TCP_OPTION_NOOP)
Florin Corasdb84e572017-05-09 18:54:52 -0700149 {
150 opt_len = 1;
151 continue;
152 }
Dave Barach68b0fb02017-02-28 15:15:56 -0500153 else
154 {
155 /* broken options */
156 if (opts_len < 2)
Florin Corasdb84e572017-05-09 18:54:52 -0700157 return -1;
Dave Barach68b0fb02017-02-28 15:15:56 -0500158 opt_len = data[1];
159
160 /* weird option length */
161 if (opt_len < 2 || opt_len > opts_len)
Florin Corasdb84e572017-05-09 18:54:52 -0700162 return -1;
Dave Barach68b0fb02017-02-28 15:15:56 -0500163 }
164
165 /* Parse options */
166 switch (kind)
167 {
168 case TCP_OPTION_MSS:
169 if ((opt_len == TCP_OPTION_LEN_MSS) && tcp_syn (th))
170 {
171 to->flags |= TCP_OPTS_FLAG_MSS;
172 to->mss = clib_net_to_host_u16 (*(u16 *) (data + 2));
173 }
174 break;
175 case TCP_OPTION_WINDOW_SCALE:
176 if ((opt_len == TCP_OPTION_LEN_WINDOW_SCALE) && tcp_syn (th))
177 {
178 to->flags |= TCP_OPTS_FLAG_WSCALE;
179 to->wscale = data[2];
180 if (to->wscale > TCP_MAX_WND_SCALE)
181 {
182 clib_warning ("Illegal window scaling value: %d",
183 to->wscale);
184 to->wscale = TCP_MAX_WND_SCALE;
185 }
186 }
187 break;
188 case TCP_OPTION_TIMESTAMP:
189 if (opt_len == TCP_OPTION_LEN_TIMESTAMP)
190 {
191 to->flags |= TCP_OPTS_FLAG_TSTAMP;
192 to->tsval = clib_net_to_host_u32 (*(u32 *) (data + 2));
193 to->tsecr = clib_net_to_host_u32 (*(u32 *) (data + 6));
194 }
195 break;
196 case TCP_OPTION_SACK_PERMITTED:
197 if (opt_len == TCP_OPTION_LEN_SACK_PERMITTED && tcp_syn (th))
198 to->flags |= TCP_OPTS_FLAG_SACK_PERMITTED;
199 break;
200 case TCP_OPTION_SACK_BLOCK:
201 /* If SACK permitted was not advertised or a SYN, break */
202 if ((to->flags & TCP_OPTS_FLAG_SACK_PERMITTED) == 0 || tcp_syn (th))
203 break;
204
205 /* If too short or not correctly formatted, break */
206 if (opt_len < 10 || ((opt_len - 2) % TCP_OPTION_LEN_SACK_BLOCK))
207 break;
208
209 to->flags |= TCP_OPTS_FLAG_SACK;
210 to->n_sack_blocks = (opt_len - 2) / TCP_OPTION_LEN_SACK_BLOCK;
211 vec_reset_length (to->sacks);
212 for (j = 0; j < to->n_sack_blocks; j++)
213 {
Florin Coras3eb50622017-07-13 01:24:57 -0400214 b.start = clib_net_to_host_u32 (*(u32 *) (data + 2 + 8 * j));
215 b.end = clib_net_to_host_u32 (*(u32 *) (data + 6 + 8 * j));
Dave Barach68b0fb02017-02-28 15:15:56 -0500216 vec_add1 (to->sacks, b);
217 }
218 break;
219 default:
220 /* Nothing to see here */
221 continue;
222 }
223 }
Florin Corasdb84e572017-05-09 18:54:52 -0700224 return 0;
Dave Barach68b0fb02017-02-28 15:15:56 -0500225}
226
Florin Corasc28764f2017-04-26 00:08:42 -0700227/**
228 * RFC1323: Check against wrapped sequence numbers (PAWS). If we have
229 * timestamp to echo and it's less than tsval_recent, drop segment
230 * but still send an ACK in order to retain TCP's mechanism for detecting
231 * and recovering from half-open connections
232 *
233 * Or at least that's what the theory says. It seems that this might not work
234 * very well with packet reordering and fast retransmit. XXX
235 */
Dave Barach68b0fb02017-02-28 15:15:56 -0500236always_inline int
237tcp_segment_check_paws (tcp_connection_t * tc)
238{
Florin Coras93992a92017-05-24 18:03:56 -0700239 return tcp_opts_tstamp (&tc->rcv_opts) && tc->tsval_recent
240 && timestamp_lt (tc->rcv_opts.tsval, tc->tsval_recent);
Dave Barach68b0fb02017-02-28 15:15:56 -0500241}
242
243/**
Florin Corasc28764f2017-04-26 00:08:42 -0700244 * Update tsval recent
245 */
246always_inline void
247tcp_update_timestamp (tcp_connection_t * tc, u32 seq, u32 seq_end)
248{
249 /*
250 * RFC1323: If Last.ACK.sent falls within the range of sequence numbers
251 * of an incoming segment:
252 * SEG.SEQ <= Last.ACK.sent < SEG.SEQ + SEG.LEN
253 * then the TSval from the segment is copied to TS.Recent;
254 * otherwise, the TSval is ignored.
255 */
Florin Corasf1762d62017-09-24 19:43:08 -0400256 if (tcp_opts_tstamp (&tc->rcv_opts) && seq_leq (seq, tc->rcv_las)
257 && seq_leq (tc->rcv_las, seq_end))
Florin Corasc28764f2017-04-26 00:08:42 -0700258 {
Dave Barach2c25a622017-06-26 11:35:07 -0400259 ASSERT (timestamp_leq (tc->tsval_recent, tc->rcv_opts.tsval));
Florin Coras93992a92017-05-24 18:03:56 -0700260 tc->tsval_recent = tc->rcv_opts.tsval;
Florin Corasc28764f2017-04-26 00:08:42 -0700261 tc->tsval_recent_age = tcp_time_now ();
262 }
263}
264
265/**
Dave Barach68b0fb02017-02-28 15:15:56 -0500266 * Validate incoming segment as per RFC793 p. 69 and RFC1323 p. 19
267 *
268 * It first verifies if segment has a wrapped sequence number (PAWS) and then
269 * does the processing associated to the first four steps (ignoring security
270 * and precedence): sequence number, rst bit and syn bit checks.
271 *
272 * @return 0 if segments passes validation.
273 */
274static int
275tcp_segment_validate (vlib_main_t * vm, tcp_connection_t * tc0,
Florin Coras00cd22d2018-04-18 13:20:18 -0700276 vlib_buffer_t * b0, tcp_header_t * th0,
277 u32 * next0, u32 * error0)
Dave Barach68b0fb02017-02-28 15:15:56 -0500278{
Florin Corasca1c8f32018-05-23 21:01:30 -0700279 /* We could get a burst of RSTs interleaved with acks */
280 if (PREDICT_FALSE (tc0->state == TCP_STATE_CLOSED))
281 {
282 tcp_send_reset (tc0);
283 *error0 = TCP_ERROR_CONNECTION_CLOSED;
284 goto drop;
285 }
286
Dave Barach68b0fb02017-02-28 15:15:56 -0500287 if (PREDICT_FALSE (!tcp_ack (th0) && !tcp_rst (th0) && !tcp_syn (th0)))
Florin Coras00cd22d2018-04-18 13:20:18 -0700288 {
289 *error0 = TCP_ERROR_SEGMENT_INVALID;
290 goto drop;
291 }
Dave Barach68b0fb02017-02-28 15:15:56 -0500292
Florin Coras93992a92017-05-24 18:03:56 -0700293 if (PREDICT_FALSE (tcp_options_parse (th0, &tc0->rcv_opts)))
Florin Corasdb84e572017-05-09 18:54:52 -0700294 {
Florin Coras4eeeaaf2017-09-05 14:03:37 -0400295 clib_warning ("options parse error");
Florin Coras00cd22d2018-04-18 13:20:18 -0700296 *error0 = TCP_ERROR_OPTIONS;
297 goto drop;
Florin Corasdb84e572017-05-09 18:54:52 -0700298 }
Dave Barach68b0fb02017-02-28 15:15:56 -0500299
Florin Coras00cd22d2018-04-18 13:20:18 -0700300 if (PREDICT_FALSE (tcp_segment_check_paws (tc0)))
Dave Barach68b0fb02017-02-28 15:15:56 -0500301 {
Florin Coras00cd22d2018-04-18 13:20:18 -0700302 *error0 = TCP_ERROR_PAWS;
Florin Coras93992a92017-05-24 18:03:56 -0700303 if (CLIB_DEBUG > 2)
Florin Corasca1c8f32018-05-23 21:01:30 -0700304 clib_warning ("paws failed\n%U", format_tcp_connection, tc0, 2);
Florin Corasc28764f2017-04-26 00:08:42 -0700305 TCP_EVT_DBG (TCP_EVT_PAWS_FAIL, tc0, vnet_buffer (b0)->tcp.seq_number,
306 vnet_buffer (b0)->tcp.seq_end);
Dave Barach68b0fb02017-02-28 15:15:56 -0500307
308 /* If it just so happens that a segment updates tsval_recent for a
309 * segment over 24 days old, invalidate tsval_recent. */
310 if (timestamp_lt (tc0->tsval_recent_age + TCP_PAWS_IDLE,
311 tcp_time_now ()))
312 {
313 /* Age isn't reset until we get a valid tsval (bsd inspired) */
314 tc0->tsval_recent = 0;
Florin Corasc28764f2017-04-26 00:08:42 -0700315 clib_warning ("paws failed - really old segment. REALLY?");
Dave Barach68b0fb02017-02-28 15:15:56 -0500316 }
317 else
318 {
319 /* Drop after ack if not rst */
320 if (!tcp_rst (th0))
321 {
322 tcp_make_ack (tc0, b0);
Florin Corasca1c8f32018-05-23 21:01:30 -0700323 TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc0, vnet_buffer (b0)->tcp);
Florin Coras00cd22d2018-04-18 13:20:18 -0700324 goto error;
Dave Barach68b0fb02017-02-28 15:15:56 -0500325 }
326 }
Florin Coras00cd22d2018-04-18 13:20:18 -0700327 goto drop;
Dave Barach68b0fb02017-02-28 15:15:56 -0500328 }
329
330 /* 1st: check sequence number */
331 if (!tcp_segment_in_rcv_wnd (tc0, vnet_buffer (b0)->tcp.seq_number,
332 vnet_buffer (b0)->tcp.seq_end))
333 {
Florin Coras00cd22d2018-04-18 13:20:18 -0700334 *error0 = TCP_ERROR_RCV_WND;
Florin Coras6792ec02017-03-13 03:49:51 -0700335 /* If our window is 0 and the packet is in sequence, let it pass
Florin Coras00cd22d2018-04-18 13:20:18 -0700336 * through for ack processing. It should be dropped later. */
337 if (!(tc0->rcv_wnd == 0
338 && tc0->rcv_nxt == vnet_buffer (b0)->tcp.seq_number))
Florin Coras6792ec02017-03-13 03:49:51 -0700339 {
340 /* If not RST, send dup ack */
341 if (!tcp_rst (th0))
342 {
343 tcp_make_ack (tc0, b0);
Florin Corasca1c8f32018-05-23 21:01:30 -0700344 TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc0, vnet_buffer (b0)->tcp);
Florin Coras00cd22d2018-04-18 13:20:18 -0700345 goto error;
Florin Coras6792ec02017-03-13 03:49:51 -0700346 }
Florin Coras00cd22d2018-04-18 13:20:18 -0700347 goto drop;
Florin Coras6792ec02017-03-13 03:49:51 -0700348 }
Dave Barach68b0fb02017-02-28 15:15:56 -0500349 }
350
351 /* 2nd: check the RST bit */
Florin Coras00cd22d2018-04-18 13:20:18 -0700352 if (PREDICT_FALSE (tcp_rst (th0)))
Dave Barach68b0fb02017-02-28 15:15:56 -0500353 {
Florin Corasd79b41e2017-03-04 05:37:52 -0800354 tcp_connection_reset (tc0);
Florin Coras00cd22d2018-04-18 13:20:18 -0700355 *error0 = TCP_ERROR_RST_RCVD;
356 goto drop;
Dave Barach68b0fb02017-02-28 15:15:56 -0500357 }
358
359 /* 3rd: check security and precedence (skip) */
360
361 /* 4th: check the SYN bit */
Florin Coras00cd22d2018-04-18 13:20:18 -0700362 if (PREDICT_FALSE (tcp_syn (th0)))
Dave Barach68b0fb02017-02-28 15:15:56 -0500363 {
Florin Coras6534b7a2017-07-18 05:38:03 -0400364 /* TODO implement RFC 5961 */
Florin Coras9d063042017-09-14 03:08:00 -0400365 if (tc0->state == TCP_STATE_SYN_RCVD)
366 {
367 tcp_make_synack (tc0, b0);
368 TCP_EVT_DBG (TCP_EVT_SYN_RCVD, tc0, 0);
369 }
Florin Coras4eeeaaf2017-09-05 14:03:37 -0400370 else
Florin Coras9d063042017-09-14 03:08:00 -0400371 {
372 tcp_make_ack (tc0, b0);
373 TCP_EVT_DBG (TCP_EVT_SYNACK_RCVD, tc0);
374 }
Florin Coras00cd22d2018-04-18 13:20:18 -0700375 goto error;
Dave Barach68b0fb02017-02-28 15:15:56 -0500376 }
377
Florin Corasc28764f2017-04-26 00:08:42 -0700378 /* If segment in window, save timestamp */
379 tcp_update_timestamp (tc0, vnet_buffer (b0)->tcp.seq_number,
380 vnet_buffer (b0)->tcp.seq_end);
Dave Barach68b0fb02017-02-28 15:15:56 -0500381 return 0;
Florin Coras00cd22d2018-04-18 13:20:18 -0700382
383drop:
384 *next0 = tcp_next_drop (tc0->c_is_ip4);
385 return -1;
386error:
387 *next0 = tcp_next_output (tc0->c_is_ip4);
388 return -1;
Dave Barach68b0fb02017-02-28 15:15:56 -0500389}
390
391always_inline int
392tcp_rcv_ack_is_acceptable (tcp_connection_t * tc0, vlib_buffer_t * tb0)
393{
394 /* SND.UNA =< SEG.ACK =< SND.NXT */
395 return (seq_leq (tc0->snd_una, vnet_buffer (tb0)->tcp.ack_number)
396 && seq_leq (vnet_buffer (tb0)->tcp.ack_number, tc0->snd_nxt));
397}
398
399/**
400 * Compute smoothed RTT as per VJ's '88 SIGCOMM and RFC6298
401 *
402 * Note that although the original article, srtt and rttvar are scaled
403 * to minimize round-off errors, here we don't. Instead, we rely on
404 * better precision time measurements.
405 *
406 * TODO support us rtt resolution
407 */
408static void
409tcp_estimate_rtt (tcp_connection_t * tc, u32 mrtt)
410{
Florin Corasf03a59a2017-06-09 21:07:32 -0700411 int err, diff;
Dave Barach68b0fb02017-02-28 15:15:56 -0500412
413 if (tc->srtt != 0)
414 {
415 err = mrtt - tc->srtt;
Dave Barach68b0fb02017-02-28 15:15:56 -0500416
417 /* XXX Drop in RTT results in RTTVAR increase and bigger RTO.
418 * The increase should be bound */
Florin Corasf03a59a2017-06-09 21:07:32 -0700419 tc->srtt = clib_max ((int) tc->srtt + (err >> 3), 1);
420 diff = (clib_abs (err) - (int) tc->rttvar) >> 2;
421 tc->rttvar = clib_max ((int) tc->rttvar + diff, 1);
Dave Barach68b0fb02017-02-28 15:15:56 -0500422 }
423 else
424 {
425 /* First measurement. */
426 tc->srtt = mrtt;
Florin Coras6792ec02017-03-13 03:49:51 -0700427 tc->rttvar = mrtt >> 1;
Dave Barach68b0fb02017-02-28 15:15:56 -0500428 }
429}
430
Florin Coras93992a92017-05-24 18:03:56 -0700431void
432tcp_update_rto (tcp_connection_t * tc)
433{
434 tc->rto = clib_min (tc->srtt + (tc->rttvar << 2), TCP_RTO_MAX);
Florin Corasf03a59a2017-06-09 21:07:32 -0700435 tc->rto = clib_max (tc->rto, TCP_RTO_MIN);
Florin Coras93992a92017-05-24 18:03:56 -0700436}
437
Florin Corasf1762d62017-09-24 19:43:08 -0400438/**
439 * Update RTT estimate and RTO timer
Dave Barach68b0fb02017-02-28 15:15:56 -0500440 *
441 * Measure RTT: We have two sources of RTT measurements: TSOPT and ACK
442 * timing. Middle boxes are known to fiddle with TCP options so we
443 * should give higher priority to ACK timing.
444 *
Florin Corasf1762d62017-09-24 19:43:08 -0400445 * This should be called only if previously sent bytes have been acked.
446 *
Dave Barach68b0fb02017-02-28 15:15:56 -0500447 * return 1 if valid rtt 0 otherwise
448 */
449static int
450tcp_update_rtt (tcp_connection_t * tc, u32 ack)
451{
452 u32 mrtt = 0;
453
454 /* Karn's rule, part 1. Don't use retransmitted segments to estimate
455 * RTT because they're ambiguous. */
Florin Corasf1762d62017-09-24 19:43:08 -0400456 if (tcp_in_cong_recovery (tc) || tc->sack_sb.sacked_bytes)
Florin Coras3ec66b02018-08-23 16:27:05 -0700457 {
458 if (tcp_in_recovery (tc))
459 return 0;
460 goto done;
461 }
Florin Corasf1762d62017-09-24 19:43:08 -0400462
463 if (tc->rtt_ts && seq_geq (ack, tc->rtt_seq))
Dave Barach68b0fb02017-02-28 15:15:56 -0500464 {
Florin Corasd67f1122018-05-21 17:47:40 -0700465 tc->mrtt_us = tcp_time_now_us (tc->c_thread_index) - tc->rtt_ts;
466 mrtt = clib_max ((u32) (tc->mrtt_us * THZ), 1);
Dave Barach68b0fb02017-02-28 15:15:56 -0500467 }
Dave Barach68b0fb02017-02-28 15:15:56 -0500468 /* As per RFC7323 TSecr can be used for RTTM only if the segment advances
469 * snd_una, i.e., the left side of the send window:
Florin Corasf1762d62017-09-24 19:43:08 -0400470 * seq_lt (tc->snd_una, ack). This is a condition for calling update_rtt */
471 else if (tcp_opts_tstamp (&tc->rcv_opts) && tc->rcv_opts.tsecr)
Dave Barach68b0fb02017-02-28 15:15:56 -0500472 {
Florin Corasd67f1122018-05-21 17:47:40 -0700473 mrtt = clib_max (tcp_time_now () - tc->rcv_opts.tsecr, 1);
Dave Barach68b0fb02017-02-28 15:15:56 -0500474 }
475
Florin Corasf1762d62017-09-24 19:43:08 -0400476 /* Ignore dubious measurements */
477 if (mrtt == 0 || mrtt > TCP_RTT_MAX)
478 goto done;
479
480 tcp_estimate_rtt (tc, mrtt);
481
482done:
483
Florin Coras3af90fc2017-05-03 21:09:42 -0700484 /* Allow measuring of a new RTT */
485 tc->rtt_ts = 0;
486
Florin Corasf1762d62017-09-24 19:43:08 -0400487 /* If we got here something must've been ACKed so make sure boff is 0,
Florin Coras3ec66b02018-08-23 16:27:05 -0700488 * even if mrtt is not valid since we update the rto lower */
Florin Corasf1762d62017-09-24 19:43:08 -0400489 tc->rto_boff = 0;
Florin Coras93992a92017-05-24 18:03:56 -0700490 tcp_update_rto (tc);
Dave Barach68b0fb02017-02-28 15:15:56 -0500491
Florin Coras3af90fc2017-05-03 21:09:42 -0700492 return 0;
Dave Barach68b0fb02017-02-28 15:15:56 -0500493}
494
495/**
496 * Dequeue bytes that have been acked and while at it update RTT estimates.
497 */
498static void
499tcp_dequeue_acked (tcp_connection_t * tc, u32 ack)
500{
Florin Coras93992a92017-05-24 18:03:56 -0700501 /* Dequeue the newly ACKed add SACKed bytes */
502 stream_session_dequeue_drop (&tc->connection,
503 tc->bytes_acked + tc->sack_sb.snd_una_adv);
504
505 tcp_validate_txf_size (tc, tc->snd_una_max - tc->snd_una);
Dave Barach68b0fb02017-02-28 15:15:56 -0500506
507 /* Update rtt and rto */
Florin Coras3e350af2017-03-30 02:54:28 -0700508 tcp_update_rtt (tc, ack);
Florin Coras93992a92017-05-24 18:03:56 -0700509
510 /* If everything has been acked, stop retransmit timer
511 * otherwise update. */
512 tcp_retransmit_timer_update (tc);
Dave Barach68b0fb02017-02-28 15:15:56 -0500513}
514
Florin Coras6792ec02017-03-13 03:49:51 -0700515/**
Florin Coras93992a92017-05-24 18:03:56 -0700516 * Check if duplicate ack as per RFC5681 Sec. 2
517 */
518static u8
519tcp_ack_is_dupack (tcp_connection_t * tc, vlib_buffer_t * b, u32 prev_snd_wnd,
520 u32 prev_snd_una)
Dave Barach68b0fb02017-02-28 15:15:56 -0500521{
Florin Coras93992a92017-05-24 18:03:56 -0700522 return ((vnet_buffer (b)->tcp.ack_number == prev_snd_una)
Dave Barach68b0fb02017-02-28 15:15:56 -0500523 && seq_gt (tc->snd_una_max, tc->snd_una)
524 && (vnet_buffer (b)->tcp.seq_end == vnet_buffer (b)->tcp.seq_number)
Florin Coras93992a92017-05-24 18:03:56 -0700525 && (prev_snd_wnd == tc->snd_wnd));
526}
527
528/**
529 * Checks if ack is a congestion control event.
530 */
531static u8
532tcp_ack_is_cc_event (tcp_connection_t * tc, vlib_buffer_t * b,
533 u32 prev_snd_wnd, u32 prev_snd_una, u8 * is_dack)
534{
535 /* Check if ack is duplicate. Per RFC 6675, ACKs that SACK new data are
536 * defined to be 'duplicate' */
537 *is_dack = tc->sack_sb.last_sacked_bytes
538 || tcp_ack_is_dupack (tc, b, prev_snd_wnd, prev_snd_una);
539
Dave Barach2c25a622017-06-26 11:35:07 -0400540 return ((*is_dack || tcp_in_cong_recovery (tc)) && !tcp_is_lost_fin (tc));
Dave Barach68b0fb02017-02-28 15:15:56 -0500541}
542
Florin Coras0dbd5172018-06-25 16:19:34 -0700543static u32
544scoreboard_hole_index (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole)
545{
546 ASSERT (!pool_is_free_index (sb->holes, hole - sb->holes));
547 return hole - sb->holes;
548}
549
550static u32
551scoreboard_hole_bytes (sack_scoreboard_hole_t * hole)
552{
553 return hole->end - hole->start;
554}
555
556sack_scoreboard_hole_t *
557scoreboard_get_hole (sack_scoreboard_t * sb, u32 index)
558{
559 if (index != TCP_INVALID_SACK_HOLE_INDEX)
560 return pool_elt_at_index (sb->holes, index);
561 return 0;
562}
563
564sack_scoreboard_hole_t *
565scoreboard_next_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole)
566{
567 if (hole->next != TCP_INVALID_SACK_HOLE_INDEX)
568 return pool_elt_at_index (sb->holes, hole->next);
569 return 0;
570}
571
572sack_scoreboard_hole_t *
573scoreboard_prev_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole)
574{
575 if (hole->prev != TCP_INVALID_SACK_HOLE_INDEX)
576 return pool_elt_at_index (sb->holes, hole->prev);
577 return 0;
578}
579
580sack_scoreboard_hole_t *
581scoreboard_first_hole (sack_scoreboard_t * sb)
582{
583 if (sb->head != TCP_INVALID_SACK_HOLE_INDEX)
584 return pool_elt_at_index (sb->holes, sb->head);
585 return 0;
586}
587
588sack_scoreboard_hole_t *
589scoreboard_last_hole (sack_scoreboard_t * sb)
590{
591 if (sb->tail != TCP_INVALID_SACK_HOLE_INDEX)
592 return pool_elt_at_index (sb->holes, sb->tail);
593 return 0;
594}
595
596static void
Dave Barach68b0fb02017-02-28 15:15:56 -0500597scoreboard_remove_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole)
598{
599 sack_scoreboard_hole_t *next, *prev;
600
601 if (hole->next != TCP_INVALID_SACK_HOLE_INDEX)
602 {
603 next = pool_elt_at_index (sb->holes, hole->next);
604 next->prev = hole->prev;
605 }
Florin Coras93992a92017-05-24 18:03:56 -0700606 else
607 {
608 sb->tail = hole->prev;
609 }
Dave Barach68b0fb02017-02-28 15:15:56 -0500610
611 if (hole->prev != TCP_INVALID_SACK_HOLE_INDEX)
612 {
613 prev = pool_elt_at_index (sb->holes, hole->prev);
614 prev->next = hole->next;
615 }
616 else
617 {
618 sb->head = hole->next;
619 }
620
Florin Coras93992a92017-05-24 18:03:56 -0700621 if (scoreboard_hole_index (sb, hole) == sb->cur_rxt_hole)
622 sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX;
623
Florin Coras3eb50622017-07-13 01:24:57 -0400624 /* Poison the entry */
625 if (CLIB_DEBUG > 0)
Dave Barachb7b92992018-10-17 10:38:51 -0400626 clib_memset (hole, 0xfe, sizeof (*hole));
Florin Coras3eb50622017-07-13 01:24:57 -0400627
Dave Barach68b0fb02017-02-28 15:15:56 -0500628 pool_put (sb->holes, hole);
629}
630
Florin Coras0dbd5172018-06-25 16:19:34 -0700631static sack_scoreboard_hole_t *
Florin Coras6792ec02017-03-13 03:49:51 -0700632scoreboard_insert_hole (sack_scoreboard_t * sb, u32 prev_index,
Dave Barach68b0fb02017-02-28 15:15:56 -0500633 u32 start, u32 end)
634{
Florin Coras6792ec02017-03-13 03:49:51 -0700635 sack_scoreboard_hole_t *hole, *next, *prev;
Dave Barach68b0fb02017-02-28 15:15:56 -0500636 u32 hole_index;
637
638 pool_get (sb->holes, hole);
Dave Barachb7b92992018-10-17 10:38:51 -0400639 clib_memset (hole, 0, sizeof (*hole));
Dave Barach68b0fb02017-02-28 15:15:56 -0500640
641 hole->start = start;
642 hole->end = end;
Florin Coras3eb50622017-07-13 01:24:57 -0400643 hole_index = scoreboard_hole_index (sb, hole);
Dave Barach68b0fb02017-02-28 15:15:56 -0500644
Florin Coras6792ec02017-03-13 03:49:51 -0700645 prev = scoreboard_get_hole (sb, prev_index);
Dave Barach68b0fb02017-02-28 15:15:56 -0500646 if (prev)
647 {
Florin Corasf03a59a2017-06-09 21:07:32 -0700648 hole->prev = prev_index;
Dave Barach68b0fb02017-02-28 15:15:56 -0500649 hole->next = prev->next;
650
651 if ((next = scoreboard_next_hole (sb, hole)))
652 next->prev = hole_index;
Florin Corasf03a59a2017-06-09 21:07:32 -0700653 else
654 sb->tail = hole_index;
Dave Barach68b0fb02017-02-28 15:15:56 -0500655
656 prev->next = hole_index;
657 }
658 else
659 {
660 sb->head = hole_index;
661 hole->prev = TCP_INVALID_SACK_HOLE_INDEX;
662 hole->next = TCP_INVALID_SACK_HOLE_INDEX;
663 }
664
665 return hole;
666}
667
Florin Coras0dbd5172018-06-25 16:19:34 -0700668static void
Florin Corasf03a59a2017-06-09 21:07:32 -0700669scoreboard_update_bytes (tcp_connection_t * tc, sack_scoreboard_t * sb)
Florin Coras93992a92017-05-24 18:03:56 -0700670{
Florin Corasecbd20b2018-10-17 23:34:54 -0700671 sack_scoreboard_hole_t *left, *right;
Florin Coras93992a92017-05-24 18:03:56 -0700672 u32 bytes = 0, blks = 0;
673
674 sb->lost_bytes = 0;
Florin Corasf03a59a2017-06-09 21:07:32 -0700675 sb->sacked_bytes = 0;
Florin Corasecbd20b2018-10-17 23:34:54 -0700676 left = scoreboard_last_hole (sb);
677 if (!left)
Florin Coras93992a92017-05-24 18:03:56 -0700678 return;
679
Florin Corasecbd20b2018-10-17 23:34:54 -0700680 if (seq_gt (sb->high_sacked, left->end))
Florin Coras93992a92017-05-24 18:03:56 -0700681 {
Florin Corasecbd20b2018-10-17 23:34:54 -0700682 bytes = sb->high_sacked - left->end;
Florin Coras93992a92017-05-24 18:03:56 -0700683 blks = 1;
684 }
685
Florin Coras9f9e9692018-10-19 17:49:00 -0700686 while ((right = left)
687 && bytes < (TCP_DUPACK_THRESHOLD - 1) * tc->snd_mss
688 && blks < TCP_DUPACK_THRESHOLD
689 /* left not updated if above conditions fail */
690 && (left = scoreboard_prev_hole (sb, right)))
Florin Coras93992a92017-05-24 18:03:56 -0700691 {
Florin Corasecbd20b2018-10-17 23:34:54 -0700692 bytes += right->start - left->end;
Florin Coras93992a92017-05-24 18:03:56 -0700693 blks++;
Florin Coras93992a92017-05-24 18:03:56 -0700694 }
695
Florin Coras9f9e9692018-10-19 17:49:00 -0700696 /* left is first lost */
697 if (left)
Florin Coras93992a92017-05-24 18:03:56 -0700698 {
Florin Coras9f9e9692018-10-19 17:49:00 -0700699 do
700 {
701 sb->lost_bytes += scoreboard_hole_bytes (right);
702 left->is_lost = 1;
703 left = scoreboard_prev_hole (sb, right);
704 if (left)
705 bytes += right->start - left->end;
706 }
707 while ((right = left));
Florin Coras93992a92017-05-24 18:03:56 -0700708 }
Florin Coras9f9e9692018-10-19 17:49:00 -0700709
Florin Corasf03a59a2017-06-09 21:07:32 -0700710 sb->sacked_bytes = bytes;
Florin Coras93992a92017-05-24 18:03:56 -0700711}
712
713/**
714 * Figure out the next hole to retransmit
715 *
716 * Follows logic proposed in RFC6675 Sec. 4, NextSeg()
717 */
718sack_scoreboard_hole_t *
719scoreboard_next_rxt_hole (sack_scoreboard_t * sb,
720 sack_scoreboard_hole_t * start,
721 u8 have_sent_1_smss,
722 u8 * can_rescue, u8 * snd_limited)
723{
724 sack_scoreboard_hole_t *hole = 0;
725
726 hole = start ? start : scoreboard_first_hole (sb);
727 while (hole && seq_leq (hole->end, sb->high_rxt) && hole->is_lost)
728 hole = scoreboard_next_hole (sb, hole);
729
730 /* Nothing, return */
731 if (!hole)
732 {
733 sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX;
734 return 0;
735 }
736
737 /* Rule (1): if higher than rxt, less than high_sacked and lost */
738 if (hole->is_lost && seq_lt (hole->start, sb->high_sacked))
739 {
740 sb->cur_rxt_hole = scoreboard_hole_index (sb, hole);
741 }
742 else
743 {
744 /* Rule (2): output takes care of transmitting new data */
745 if (!have_sent_1_smss)
746 {
747 hole = 0;
748 sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX;
749 }
750 /* Rule (3): if hole not lost */
751 else if (seq_lt (hole->start, sb->high_sacked))
752 {
Florin Corasbf4d5ce2018-10-19 16:26:24 -0700753 *snd_limited = 0;
Florin Coras93992a92017-05-24 18:03:56 -0700754 sb->cur_rxt_hole = scoreboard_hole_index (sb, hole);
755 }
756 /* Rule (4): if hole beyond high_sacked */
757 else
758 {
759 ASSERT (seq_geq (hole->start, sb->high_sacked));
760 *snd_limited = 1;
761 *can_rescue = 1;
762 /* HighRxt MUST NOT be updated */
763 return 0;
764 }
765 }
766
767 if (hole && seq_lt (sb->high_rxt, hole->start))
768 sb->high_rxt = hole->start;
769
770 return hole;
771}
772
Florin Coras0dbd5172018-06-25 16:19:34 -0700773static void
Florin Coras3eb50622017-07-13 01:24:57 -0400774scoreboard_init_high_rxt (sack_scoreboard_t * sb, u32 seq)
Florin Coras93992a92017-05-24 18:03:56 -0700775{
776 sack_scoreboard_hole_t *hole;
777 hole = scoreboard_first_hole (sb);
Florin Coras3eb50622017-07-13 01:24:57 -0400778 if (hole)
779 {
780 seq = seq_gt (seq, hole->start) ? seq : hole->start;
781 sb->cur_rxt_hole = sb->head;
782 }
783 sb->high_rxt = seq;
784}
785
Florin Coras0dbd5172018-06-25 16:19:34 -0700786void
787scoreboard_init (sack_scoreboard_t * sb)
788{
789 sb->head = TCP_INVALID_SACK_HOLE_INDEX;
790 sb->tail = TCP_INVALID_SACK_HOLE_INDEX;
791 sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX;
792}
793
794void
795scoreboard_clear (sack_scoreboard_t * sb)
796{
797 sack_scoreboard_hole_t *hole;
798 while ((hole = scoreboard_first_hole (sb)))
799 {
800 scoreboard_remove_hole (sb, hole);
801 }
802 ASSERT (sb->head == sb->tail && sb->head == TCP_INVALID_SACK_HOLE_INDEX);
803 ASSERT (pool_elts (sb->holes) == 0);
804 sb->sacked_bytes = 0;
805 sb->last_sacked_bytes = 0;
806 sb->last_bytes_delivered = 0;
807 sb->snd_una_adv = 0;
808 sb->high_sacked = 0;
809 sb->high_rxt = 0;
810 sb->lost_bytes = 0;
811 sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX;
812}
813
Florin Coras3eb50622017-07-13 01:24:57 -0400814/**
815 * Test that scoreboard is sane after recovery
816 *
817 * Returns 1 if scoreboard is empty or if first hole beyond
818 * snd_una.
819 */
Florin Coras0dbd5172018-06-25 16:19:34 -0700820static u8
Florin Coras3eb50622017-07-13 01:24:57 -0400821tcp_scoreboard_is_sane_post_recovery (tcp_connection_t * tc)
822{
823 sack_scoreboard_hole_t *hole;
824 hole = scoreboard_first_hole (&tc->sack_sb);
Florin Corasecbd20b2018-10-17 23:34:54 -0700825 return (!hole || (seq_geq (hole->start, tc->snd_una)
826 && seq_lt (hole->end, tc->snd_una_max)));
Florin Coras93992a92017-05-24 18:03:56 -0700827}
828
829void
Dave Barach68b0fb02017-02-28 15:15:56 -0500830tcp_rcv_sacks (tcp_connection_t * tc, u32 ack)
831{
832 sack_scoreboard_t *sb = &tc->sack_sb;
833 sack_block_t *blk, tmp;
Florin Corasf03a59a2017-06-09 21:07:32 -0700834 sack_scoreboard_hole_t *hole, *next_hole, *last_hole;
Florin Coras93992a92017-05-24 18:03:56 -0700835 u32 blk_index = 0, old_sacked_bytes, hole_index;
Dave Barach68b0fb02017-02-28 15:15:56 -0500836 int i, j;
837
Florin Coras6792ec02017-03-13 03:49:51 -0700838 sb->last_sacked_bytes = 0;
Florin Coras93992a92017-05-24 18:03:56 -0700839 sb->last_bytes_delivered = 0;
Florin Coras352c2c42018-06-26 01:22:41 -0700840 sb->snd_una_adv = 0;
Florin Coras6792ec02017-03-13 03:49:51 -0700841
Florin Coras93992a92017-05-24 18:03:56 -0700842 if (!tcp_opts_sack (&tc->rcv_opts)
843 && sb->head == TCP_INVALID_SACK_HOLE_INDEX)
Dave Barach68b0fb02017-02-28 15:15:56 -0500844 return;
845
Florin Coras352c2c42018-06-26 01:22:41 -0700846 old_sacked_bytes = sb->sacked_bytes;
847
Dave Barach68b0fb02017-02-28 15:15:56 -0500848 /* Remove invalid blocks */
Florin Coras93992a92017-05-24 18:03:56 -0700849 blk = tc->rcv_opts.sacks;
850 while (blk < vec_end (tc->rcv_opts.sacks))
Florin Coras6792ec02017-03-13 03:49:51 -0700851 {
852 if (seq_lt (blk->start, blk->end)
853 && seq_gt (blk->start, tc->snd_una)
Florin Coras3eb50622017-07-13 01:24:57 -0400854 && seq_gt (blk->start, ack) && seq_leq (blk->end, tc->snd_una_max))
Florin Coras6792ec02017-03-13 03:49:51 -0700855 {
856 blk++;
857 continue;
858 }
Florin Coras93992a92017-05-24 18:03:56 -0700859 vec_del1 (tc->rcv_opts.sacks, blk - tc->rcv_opts.sacks);
Florin Coras6792ec02017-03-13 03:49:51 -0700860 }
Dave Barach68b0fb02017-02-28 15:15:56 -0500861
862 /* Add block for cumulative ack */
863 if (seq_gt (ack, tc->snd_una))
864 {
865 tmp.start = tc->snd_una;
866 tmp.end = ack;
Florin Coras93992a92017-05-24 18:03:56 -0700867 vec_add1 (tc->rcv_opts.sacks, tmp);
Dave Barach68b0fb02017-02-28 15:15:56 -0500868 }
869
Florin Coras93992a92017-05-24 18:03:56 -0700870 if (vec_len (tc->rcv_opts.sacks) == 0)
Dave Barach68b0fb02017-02-28 15:15:56 -0500871 return;
872
Florin Coras3eb50622017-07-13 01:24:57 -0400873 tcp_scoreboard_trace_add (tc, ack);
874
Dave Barach68b0fb02017-02-28 15:15:56 -0500875 /* Make sure blocks are ordered */
Florin Coras93992a92017-05-24 18:03:56 -0700876 for (i = 0; i < vec_len (tc->rcv_opts.sacks); i++)
877 for (j = i + 1; j < vec_len (tc->rcv_opts.sacks); j++)
878 if (seq_lt (tc->rcv_opts.sacks[j].start, tc->rcv_opts.sacks[i].start))
Dave Barach68b0fb02017-02-28 15:15:56 -0500879 {
Florin Coras93992a92017-05-24 18:03:56 -0700880 tmp = tc->rcv_opts.sacks[i];
881 tc->rcv_opts.sacks[i] = tc->rcv_opts.sacks[j];
882 tc->rcv_opts.sacks[j] = tmp;
Dave Barach68b0fb02017-02-28 15:15:56 -0500883 }
884
Dave Barach68b0fb02017-02-28 15:15:56 -0500885 if (sb->head == TCP_INVALID_SACK_HOLE_INDEX)
886 {
Florin Coras6792ec02017-03-13 03:49:51 -0700887 /* If no holes, insert the first that covers all outstanding bytes */
888 last_hole = scoreboard_insert_hole (sb, TCP_INVALID_SACK_HOLE_INDEX,
889 tc->snd_una, tc->snd_una_max);
890 sb->tail = scoreboard_hole_index (sb, last_hole);
Florin Coras93992a92017-05-24 18:03:56 -0700891 tmp = tc->rcv_opts.sacks[vec_len (tc->rcv_opts.sacks) - 1];
892 sb->high_sacked = tmp.end;
Florin Coras6792ec02017-03-13 03:49:51 -0700893 }
894 else
895 {
896 /* If we have holes but snd_una_max is beyond the last hole, update
897 * last hole end */
Florin Coras93992a92017-05-24 18:03:56 -0700898 tmp = tc->rcv_opts.sacks[vec_len (tc->rcv_opts.sacks) - 1];
Florin Coras6792ec02017-03-13 03:49:51 -0700899 last_hole = scoreboard_last_hole (sb);
Dave Barach2c25a622017-06-26 11:35:07 -0400900 if (seq_gt (tc->snd_una_max, last_hole->end))
901 {
902 if (seq_geq (last_hole->start, sb->high_sacked))
903 {
904 last_hole->end = tc->snd_una_max;
905 }
906 /* New hole after high sacked block */
907 else if (seq_lt (sb->high_sacked, tc->snd_una_max))
908 {
909 scoreboard_insert_hole (sb, sb->tail, sb->high_sacked,
910 tc->snd_una_max);
911 }
912 }
913 /* Keep track of max byte sacked for when the last hole
Florin Corasf03a59a2017-06-09 21:07:32 -0700914 * is acked */
915 if (seq_gt (tmp.end, sb->high_sacked))
916 sb->high_sacked = tmp.end;
Dave Barach68b0fb02017-02-28 15:15:56 -0500917 }
918
919 /* Walk the holes with the SACK blocks */
920 hole = pool_elt_at_index (sb->holes, sb->head);
Florin Coras93992a92017-05-24 18:03:56 -0700921 while (hole && blk_index < vec_len (tc->rcv_opts.sacks))
Dave Barach68b0fb02017-02-28 15:15:56 -0500922 {
Florin Coras93992a92017-05-24 18:03:56 -0700923 blk = &tc->rcv_opts.sacks[blk_index];
Dave Barach68b0fb02017-02-28 15:15:56 -0500924 if (seq_leq (blk->start, hole->start))
925 {
926 /* Block covers hole. Remove hole */
927 if (seq_geq (blk->end, hole->end))
928 {
929 next_hole = scoreboard_next_hole (sb, hole);
930
Florin Corasf03a59a2017-06-09 21:07:32 -0700931 /* Byte accounting: snd_una needs to be advanced */
932 if (blk->end == ack)
Dave Barach68b0fb02017-02-28 15:15:56 -0500933 {
Florin Corasf03a59a2017-06-09 21:07:32 -0700934 if (next_hole)
Florin Coras06d11012017-05-17 14:21:51 -0700935 {
Florin Corasf03a59a2017-06-09 21:07:32 -0700936 if (seq_lt (ack, next_hole->start))
937 sb->snd_una_adv = next_hole->start - ack;
938 sb->last_bytes_delivered +=
939 next_hole->start - hole->end;
Florin Coras06d11012017-05-17 14:21:51 -0700940 }
Florin Coras3eb50622017-07-13 01:24:57 -0400941 else
Florin Coras06d11012017-05-17 14:21:51 -0700942 {
Dave Barach2c25a622017-06-26 11:35:07 -0400943 ASSERT (seq_geq (sb->high_sacked, ack));
Florin Coras93992a92017-05-24 18:03:56 -0700944 sb->snd_una_adv = sb->high_sacked - ack;
Florin Corasf03a59a2017-06-09 21:07:32 -0700945 sb->last_bytes_delivered += sb->high_sacked - hole->end;
Florin Coras06d11012017-05-17 14:21:51 -0700946 }
947 }
948
Dave Barach68b0fb02017-02-28 15:15:56 -0500949 scoreboard_remove_hole (sb, hole);
950 hole = next_hole;
951 }
Florin Coras6792ec02017-03-13 03:49:51 -0700952 /* Partial 'head' overlap */
Dave Barach68b0fb02017-02-28 15:15:56 -0500953 else
954 {
Florin Coras6792ec02017-03-13 03:49:51 -0700955 if (seq_gt (blk->end, hole->start))
956 {
Florin Coras6792ec02017-03-13 03:49:51 -0700957 hole->start = blk->end;
958 }
Dave Barach68b0fb02017-02-28 15:15:56 -0500959 blk_index++;
960 }
961 }
962 else
963 {
964 /* Hole must be split */
Florin Coras6792ec02017-03-13 03:49:51 -0700965 if (seq_lt (blk->end, hole->end))
Dave Barach68b0fb02017-02-28 15:15:56 -0500966 {
Florin Coras6792ec02017-03-13 03:49:51 -0700967 hole_index = scoreboard_hole_index (sb, hole);
Florin Coras3eb50622017-07-13 01:24:57 -0400968 next_hole = scoreboard_insert_hole (sb, hole_index, blk->end,
969 hole->end);
Florin Coras6792ec02017-03-13 03:49:51 -0700970
971 /* Pool might've moved */
972 hole = scoreboard_get_hole (sb, hole_index);
973 hole->end = blk->start;
Dave Barach68b0fb02017-02-28 15:15:56 -0500974 blk_index++;
Florin Coras3eb50622017-07-13 01:24:57 -0400975 ASSERT (hole->next == scoreboard_hole_index (sb, next_hole));
Dave Barach68b0fb02017-02-28 15:15:56 -0500976 }
Florin Corasf03a59a2017-06-09 21:07:32 -0700977 else if (seq_lt (blk->start, hole->end))
Dave Barach68b0fb02017-02-28 15:15:56 -0500978 {
Florin Coras6792ec02017-03-13 03:49:51 -0700979 hole->end = blk->start;
Dave Barach68b0fb02017-02-28 15:15:56 -0500980 }
Florin Coras93992a92017-05-24 18:03:56 -0700981 hole = scoreboard_next_hole (sb, hole);
Dave Barach68b0fb02017-02-28 15:15:56 -0500982 }
983 }
Florin Coras6792ec02017-03-13 03:49:51 -0700984
Florin Corasecbd20b2018-10-17 23:34:54 -0700985 if (pool_elts (sb->holes) == 1)
986 {
987 hole = scoreboard_first_hole (sb);
988 if (hole->start == ack + sb->snd_una_adv
989 && hole->end == tc->snd_una_max)
990 scoreboard_remove_hole (sb, hole);
991 }
992
Florin Corasf03a59a2017-06-09 21:07:32 -0700993 scoreboard_update_bytes (tc, sb);
994 sb->last_sacked_bytes = sb->sacked_bytes
995 - (old_sacked_bytes - sb->last_bytes_delivered);
Florin Corasca1c8f32018-05-23 21:01:30 -0700996 ASSERT (sb->last_sacked_bytes <= sb->sacked_bytes || tcp_in_recovery (tc));
Florin Corasbf4d5ce2018-10-19 16:26:24 -0700997 ASSERT (sb->sacked_bytes == 0 || tcp_in_recovery (tc)
Florin Corasf03a59a2017-06-09 21:07:32 -0700998 || sb->sacked_bytes < tc->snd_una_max - seq_max (tc->snd_una, ack));
999 ASSERT (sb->last_sacked_bytes + sb->lost_bytes <= tc->snd_una_max
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001000 - seq_max (tc->snd_una, ack) || tcp_in_recovery (tc));
Dave Barach2c25a622017-06-26 11:35:07 -04001001 ASSERT (sb->head == TCP_INVALID_SACK_HOLE_INDEX || tcp_in_recovery (tc)
1002 || sb->holes[sb->head].start == ack + sb->snd_una_adv);
Florin Corasca1c8f32018-05-23 21:01:30 -07001003 TCP_EVT_DBG (TCP_EVT_CC_SCOREBOARD, tc);
Dave Barach68b0fb02017-02-28 15:15:56 -05001004}
1005
Florin Coras93992a92017-05-24 18:03:56 -07001006/**
1007 * Try to update snd_wnd based on feedback received from peer.
Dave Barach68b0fb02017-02-28 15:15:56 -05001008 *
Florin Coras93992a92017-05-24 18:03:56 -07001009 * If successful, and new window is 'effectively' 0, activate persist
1010 * timer.
1011 */
Dave Barach68b0fb02017-02-28 15:15:56 -05001012static void
1013tcp_update_snd_wnd (tcp_connection_t * tc, u32 seq, u32 ack, u32 snd_wnd)
1014{
Florin Coras93992a92017-05-24 18:03:56 -07001015 /* If (SND.WL1 < SEG.SEQ or (SND.WL1 = SEG.SEQ and SND.WL2 =< SEG.ACK)), set
1016 * SND.WND <- SEG.WND, set SND.WL1 <- SEG.SEQ, and set SND.WL2 <- SEG.ACK */
Florin Coras6792ec02017-03-13 03:49:51 -07001017 if (seq_lt (tc->snd_wl1, seq)
1018 || (tc->snd_wl1 == seq && seq_leq (tc->snd_wl2, ack)))
Dave Barach68b0fb02017-02-28 15:15:56 -05001019 {
1020 tc->snd_wnd = snd_wnd;
1021 tc->snd_wl1 = seq;
1022 tc->snd_wl2 = ack;
Florin Coras6792ec02017-03-13 03:49:51 -07001023 TCP_EVT_DBG (TCP_EVT_SND_WND, tc);
Florin Coras3e350af2017-03-30 02:54:28 -07001024
Florin Corasbb292f42017-05-19 09:49:19 -07001025 if (tc->snd_wnd < tc->snd_mss)
1026 {
Florin Coras93992a92017-05-24 18:03:56 -07001027 /* Set persist timer if not set and we just got 0 wnd */
1028 if (!tcp_timer_is_active (tc, TCP_TIMER_PERSIST)
1029 && !tcp_timer_is_active (tc, TCP_TIMER_RETRANSMIT))
Florin Corasbb292f42017-05-19 09:49:19 -07001030 tcp_persist_timer_set (tc);
1031 }
Florin Coras3e350af2017-03-30 02:54:28 -07001032 else
Florin Coras93992a92017-05-24 18:03:56 -07001033 {
1034 tcp_persist_timer_reset (tc);
1035 if (!tcp_in_recovery (tc) && tc->rto_boff > 0)
1036 {
1037 tc->rto_boff = 0;
1038 tcp_update_rto (tc);
1039 }
1040 }
Dave Barach68b0fb02017-02-28 15:15:56 -05001041 }
1042}
1043
Florin Corasd2aab832018-05-22 11:39:59 -07001044/**
1045 * Init loss recovery/fast recovery.
1046 *
1047 * Triggered by dup acks as opposed to timer timeout. Note that cwnd is
1048 * updated in @ref tcp_cc_handle_event after fast retransmit
1049 */
Florin Coras6792ec02017-03-13 03:49:51 -07001050void
Florin Coras93992a92017-05-24 18:03:56 -07001051tcp_cc_init_congestion (tcp_connection_t * tc)
Dave Barach68b0fb02017-02-28 15:15:56 -05001052{
Florin Coras93992a92017-05-24 18:03:56 -07001053 tcp_fastrecovery_on (tc);
1054 tc->snd_congestion = tc->snd_una_max;
Florin Coras62166002018-04-18 16:40:55 -07001055 tc->cwnd_acc_bytes = 0;
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001056 tc->snd_rxt_bytes = 0;
1057 tc->prev_ssthresh = tc->ssthresh;
1058 tc->prev_cwnd = tc->cwnd;
Dave Barach68b0fb02017-02-28 15:15:56 -05001059 tc->cc_algo->congestion (tc);
Florin Coras6792ec02017-03-13 03:49:51 -07001060 TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 4);
Dave Barach68b0fb02017-02-28 15:15:56 -05001061}
1062
Florin Coras93992a92017-05-24 18:03:56 -07001063static void
1064tcp_cc_recovery_exit (tcp_connection_t * tc)
Dave Barach68b0fb02017-02-28 15:15:56 -05001065{
Florin Coras93992a92017-05-24 18:03:56 -07001066 tc->rto_boff = 0;
Florin Corasf1762d62017-09-24 19:43:08 -04001067 tcp_update_rto (tc);
Florin Coras93992a92017-05-24 18:03:56 -07001068 tc->snd_rxt_ts = 0;
Florin Coras84275e92017-09-26 12:30:40 -04001069 tc->snd_nxt = tc->snd_una_max;
Florin Coras93992a92017-05-24 18:03:56 -07001070 tcp_recovery_off (tc);
Florin Corasf1762d62017-09-24 19:43:08 -04001071 TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 3);
Florin Coras93992a92017-05-24 18:03:56 -07001072}
Florin Coras3e350af2017-03-30 02:54:28 -07001073
Florin Coras93992a92017-05-24 18:03:56 -07001074void
1075tcp_cc_fastrecovery_exit (tcp_connection_t * tc)
1076{
Florin Coras6792ec02017-03-13 03:49:51 -07001077 tc->cc_algo->recovered (tc);
Florin Coras93992a92017-05-24 18:03:56 -07001078 tc->snd_rxt_bytes = 0;
Florin Coras6792ec02017-03-13 03:49:51 -07001079 tc->rcv_dupacks = 0;
Florin Coras84275e92017-09-26 12:30:40 -04001080 tc->snd_nxt = tc->snd_una_max;
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001081 tc->snd_rxt_bytes = 0;
1082
Florin Coras93992a92017-05-24 18:03:56 -07001083 tcp_fastrecovery_off (tc);
1084 tcp_fastrecovery_1_smss_off (tc);
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001085 tcp_fastrecovery_first_off (tc);
Florin Corasd67f1122018-05-21 17:47:40 -07001086
1087 /* Update pacer because our cwnd changed. Also makes sure
1088 * that we recompute the max burst size */
1089 tcp_update_pacer (tc);
1090
Florin Corasf1762d62017-09-24 19:43:08 -04001091 TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 3);
Dave Barach68b0fb02017-02-28 15:15:56 -05001092}
1093
1094static void
Florin Coras93992a92017-05-24 18:03:56 -07001095tcp_cc_congestion_undo (tcp_connection_t * tc)
Dave Barach68b0fb02017-02-28 15:15:56 -05001096{
Florin Coras93992a92017-05-24 18:03:56 -07001097 tc->cwnd = tc->prev_cwnd;
1098 tc->ssthresh = tc->prev_ssthresh;
1099 tc->snd_nxt = tc->snd_una_max;
1100 tc->rcv_dupacks = 0;
1101 if (tcp_in_recovery (tc))
1102 tcp_cc_recovery_exit (tc);
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001103 else if (tcp_in_fastrecovery (tc))
1104 tcp_cc_fastrecovery_exit (tc);
Florin Coras93992a92017-05-24 18:03:56 -07001105 ASSERT (tc->rto_boff == 0);
Florin Corasf1762d62017-09-24 19:43:08 -04001106 TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 5);
Florin Coras93992a92017-05-24 18:03:56 -07001107}
Dave Barach68b0fb02017-02-28 15:15:56 -05001108
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001109static inline u8
1110tcp_cc_is_spurious_timeout_rxt (tcp_connection_t * tc)
Florin Coras93992a92017-05-24 18:03:56 -07001111{
Florin Corasf1762d62017-09-24 19:43:08 -04001112 return (tcp_in_recovery (tc) && tc->rto_boff == 1
Dave Barach2c25a622017-06-26 11:35:07 -04001113 && tc->snd_rxt_ts
Florin Coras93992a92017-05-24 18:03:56 -07001114 && tcp_opts_tstamp (&tc->rcv_opts)
1115 && timestamp_lt (tc->rcv_opts.tsecr, tc->snd_rxt_ts));
1116}
1117
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001118static inline u8
1119tcp_cc_is_spurious_fast_rxt (tcp_connection_t * tc)
1120{
1121 return (tcp_in_fastrecovery (tc)
1122 && tc->cwnd > tc->ssthresh + 3 * tc->snd_mss);
1123}
1124
1125static u8
1126tcp_cc_is_spurious_retransmit (tcp_connection_t * tc)
1127{
1128 return (tcp_cc_is_spurious_timeout_rxt (tc)
1129 || tcp_cc_is_spurious_fast_rxt (tc));
1130}
1131
Florin Coras0dbd5172018-06-25 16:19:34 -07001132static int
Florin Coras93992a92017-05-24 18:03:56 -07001133tcp_cc_recover (tcp_connection_t * tc)
1134{
1135 ASSERT (tcp_in_cong_recovery (tc));
1136 if (tcp_cc_is_spurious_retransmit (tc))
Dave Barach68b0fb02017-02-28 15:15:56 -05001137 {
Florin Coras93992a92017-05-24 18:03:56 -07001138 tcp_cc_congestion_undo (tc);
1139 return 1;
1140 }
1141
1142 if (tcp_in_recovery (tc))
1143 tcp_cc_recovery_exit (tc);
1144 else if (tcp_in_fastrecovery (tc))
1145 tcp_cc_fastrecovery_exit (tc);
1146
1147 ASSERT (tc->rto_boff == 0);
1148 ASSERT (!tcp_in_cong_recovery (tc));
Florin Coras3eb50622017-07-13 01:24:57 -04001149 ASSERT (tcp_scoreboard_is_sane_post_recovery (tc));
Florin Coras93992a92017-05-24 18:03:56 -07001150 return 0;
1151}
1152
1153static void
1154tcp_cc_update (tcp_connection_t * tc, vlib_buffer_t * b)
1155{
Florin Coras3eb50622017-07-13 01:24:57 -04001156 ASSERT (!tcp_in_cong_recovery (tc) || tcp_is_lost_fin (tc));
Florin Coras93992a92017-05-24 18:03:56 -07001157
1158 /* Congestion avoidance */
Florin Corasd67f1122018-05-21 17:47:40 -07001159 tcp_cc_rcv_ack (tc);
Florin Coras93992a92017-05-24 18:03:56 -07001160
1161 /* If a cumulative ack, make sure dupacks is 0 */
1162 tc->rcv_dupacks = 0;
1163
1164 /* When dupacks hits the threshold we only enter fast retransmit if
1165 * cumulative ack covers more than snd_congestion. Should snd_una
1166 * wrap this test may fail under otherwise valid circumstances.
1167 * Therefore, proactively update snd_congestion when wrap detected. */
1168 if (PREDICT_FALSE
1169 (seq_leq (tc->snd_congestion, tc->snd_una - tc->bytes_acked)
1170 && seq_gt (tc->snd_congestion, tc->snd_una)))
1171 tc->snd_congestion = tc->snd_una - 1;
1172}
1173
1174static u8
1175tcp_should_fastrecover_sack (tcp_connection_t * tc)
1176{
1177 return (TCP_DUPACK_THRESHOLD - 1) * tc->snd_mss < tc->sack_sb.sacked_bytes;
1178}
1179
1180static u8
1181tcp_should_fastrecover (tcp_connection_t * tc)
1182{
1183 return (tc->rcv_dupacks == TCP_DUPACK_THRESHOLD
1184 || tcp_should_fastrecover_sack (tc));
1185}
1186
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001187void
1188tcp_program_fastretransmit (tcp_connection_t * tc)
1189{
1190 tcp_worker_ctx_t *wrk = &tcp_main.wrk_ctx[tc->c_thread_index];
1191 if (!(tc->flags & TCP_CONN_FRXT_PENDING))
1192 {
1193 vec_add1 (wrk->pending_fast_rxt, tc->c_c_index);
1194 tc->flags |= TCP_CONN_FRXT_PENDING;
1195 }
1196}
1197
1198void
1199tcp_do_fastretransmits (u32 thread_index)
1200{
1201 tcp_worker_ctx_t *wrk = &tcp_main.wrk_ctx[thread_index];
Florin Corase55a6d72018-10-31 23:09:22 -07001202 u32 max_burst_size, burst_size, n_segs = 0, n_segs_now;
1203 u32 *ongoing_fast_rxt, burst_bytes, sent_bytes;
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001204 tcp_connection_t *tc;
Florin Corase55a6d72018-10-31 23:09:22 -07001205 u64 last_cpu_time;
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001206 int i;
1207
Florin Corase55a6d72018-10-31 23:09:22 -07001208 if (vec_len (wrk->pending_fast_rxt) == 0
1209 && vec_len (wrk->postponed_fast_rxt) == 0)
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001210 return;
1211
Florin Corase55a6d72018-10-31 23:09:22 -07001212 last_cpu_time = wrk->vm->clib_time.last_cpu_time;
1213 ongoing_fast_rxt = wrk->ongoing_fast_rxt;
1214 vec_append (ongoing_fast_rxt, wrk->postponed_fast_rxt);
1215 vec_append (ongoing_fast_rxt, wrk->pending_fast_rxt);
1216
1217 _vec_len (wrk->postponed_fast_rxt) = 0;
1218 _vec_len (wrk->pending_fast_rxt) = 0;
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001219
1220 max_burst_size = VLIB_FRAME_SIZE / vec_len (wrk->ongoing_fast_rxt);
1221 max_burst_size = clib_max (max_burst_size, 1);
1222
Florin Corase55a6d72018-10-31 23:09:22 -07001223 for (i = 0; i < vec_len (ongoing_fast_rxt); i++)
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001224 {
Florin Corase55a6d72018-10-31 23:09:22 -07001225 if (n_segs >= VLIB_FRAME_SIZE)
1226 {
1227 vec_add1 (wrk->postponed_fast_rxt, ongoing_fast_rxt[i]);
1228 continue;
1229 }
1230
1231 tc = tcp_connection_get (ongoing_fast_rxt[i], thread_index);
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001232 tc->flags &= ~TCP_CONN_FRXT_PENDING;
1233
1234 if (!tcp_in_fastrecovery (tc))
1235 continue;
1236
Florin Corase55a6d72018-10-31 23:09:22 -07001237 burst_size = clib_min (max_burst_size, VLIB_FRAME_SIZE - n_segs);
1238 burst_bytes = transport_connection_tx_pacer_burst (&tc->connection,
1239 last_cpu_time);
1240 burst_size = clib_min (burst_size, burst_bytes / tc->snd_mss);
1241 if (!burst_size)
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001242 {
1243 tcp_program_fastretransmit (tc);
1244 continue;
1245 }
1246
Florin Corase55a6d72018-10-31 23:09:22 -07001247 n_segs_now = tcp_fast_retransmit (tc, burst_size);
1248 sent_bytes = clib_min (n_segs_now * tc->snd_mss, burst_bytes);
1249 transport_connection_tx_pacer_update_bytes (&tc->connection,
1250 sent_bytes);
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001251
Florin Corase55a6d72018-10-31 23:09:22 -07001252 n_segs += n_segs_now;
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001253 }
Florin Corase55a6d72018-10-31 23:09:22 -07001254 _vec_len (ongoing_fast_rxt) = 0;
1255 wrk->ongoing_fast_rxt = ongoing_fast_rxt;
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001256}
1257
Florin Corasf03a59a2017-06-09 21:07:32 -07001258/**
1259 * One function to rule them all ... and in the darkness bind them
1260 */
Florin Coras93992a92017-05-24 18:03:56 -07001261static void
1262tcp_cc_handle_event (tcp_connection_t * tc, u32 is_dack)
1263{
Florin Corasf03a59a2017-06-09 21:07:32 -07001264 u32 rxt_delivered;
1265
Florin Corasca1c8f32018-05-23 21:01:30 -07001266 if (tcp_in_fastrecovery (tc) && tcp_opts_sack_permitted (&tc->rcv_opts))
1267 {
1268 if (tc->bytes_acked)
1269 goto partial_ack;
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001270 tcp_program_fastretransmit (tc);
Florin Corasca1c8f32018-05-23 21:01:30 -07001271 return;
1272 }
Florin Coras93992a92017-05-24 18:03:56 -07001273 /*
1274 * Duplicate ACK. Check if we should enter fast recovery, or if already in
1275 * it account for the bytes that left the network.
1276 */
Florin Corasca1c8f32018-05-23 21:01:30 -07001277 else if (is_dack && !tcp_in_recovery (tc))
Florin Coras93992a92017-05-24 18:03:56 -07001278 {
Florin Corasd2aab832018-05-22 11:39:59 -07001279 TCP_EVT_DBG (TCP_EVT_DUPACK_RCVD, tc, 1);
Florin Coras93992a92017-05-24 18:03:56 -07001280 ASSERT (tc->snd_una != tc->snd_una_max
1281 || tc->sack_sb.last_sacked_bytes);
Dave Barach2c25a622017-06-26 11:35:07 -04001282
Florin Coras93992a92017-05-24 18:03:56 -07001283 tc->rcv_dupacks++;
1284
Florin Corasd2aab832018-05-22 11:39:59 -07001285 /* Pure duplicate ack. If some data got acked, it's handled lower */
Florin Coras93992a92017-05-24 18:03:56 -07001286 if (tc->rcv_dupacks > TCP_DUPACK_THRESHOLD && !tc->bytes_acked)
Dave Barach68b0fb02017-02-28 15:15:56 -05001287 {
Florin Coras93992a92017-05-24 18:03:56 -07001288 ASSERT (tcp_in_fastrecovery (tc));
Florin Coras93992a92017-05-24 18:03:56 -07001289 tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK);
1290 return;
Dave Barach68b0fb02017-02-28 15:15:56 -05001291 }
Florin Coras93992a92017-05-24 18:03:56 -07001292 else if (tcp_should_fastrecover (tc))
Dave Barach68b0fb02017-02-28 15:15:56 -05001293 {
Florin Corase55a6d72018-10-31 23:09:22 -07001294 u32 byte_rate;
Florin Corasd2aab832018-05-22 11:39:59 -07001295 ASSERT (!tcp_in_fastrecovery (tc));
Florin Coras93992a92017-05-24 18:03:56 -07001296
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001297 /* Heuristic to catch potential late dupacks
1298 * after fast retransmit exits */
1299 if (is_dack && tc->snd_una == tc->snd_congestion
1300 && timestamp_leq (tc->rcv_opts.tsecr, tc->tsecr_last_ack))
Florin Coras93992a92017-05-24 18:03:56 -07001301 {
1302 tc->rcv_dupacks = 0;
1303 return;
1304 }
1305
1306 tcp_cc_init_congestion (tc);
1307 tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK);
1308
Florin Coras3eb50622017-07-13 01:24:57 -04001309 if (tcp_opts_sack_permitted (&tc->rcv_opts))
Florin Corase55a6d72018-10-31 23:09:22 -07001310 {
1311 tc->cwnd = tc->ssthresh;
1312 scoreboard_init_high_rxt (&tc->sack_sb, tc->snd_una);
1313 tc->sack_sb.rescue_rxt = tc->snd_una - 1;
1314 }
1315 else
1316 {
1317 /* Post retransmit update cwnd to ssthresh and account for the
1318 * three segments that have left the network and should've been
1319 * buffered at the receiver XXX */
1320 tc->cwnd = tc->ssthresh + 3 * tc->snd_mss;
1321 }
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001322
Florin Corase55a6d72018-10-31 23:09:22 -07001323 byte_rate = (0.3 * tc->cwnd) / ((f64) TCP_TICK * tc->srtt);
1324 transport_connection_tx_pacer_init (&tc->connection, byte_rate, 0);
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001325 tcp_program_fastretransmit (tc);
Florin Coras93992a92017-05-24 18:03:56 -07001326 return;
Dave Barach68b0fb02017-02-28 15:15:56 -05001327 }
Florin Coras93992a92017-05-24 18:03:56 -07001328 else if (!tc->bytes_acked
1329 || (tc->bytes_acked && !tcp_in_cong_recovery (tc)))
1330 {
1331 tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK);
1332 return;
1333 }
1334 else
1335 goto partial_ack;
1336 }
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001337 /* Don't allow entry in fast recovery if still in recovery, for now */
1338 else if (0 && is_dack && tcp_in_recovery (tc))
1339 {
1340 /* If of of the two conditions lower hold, reset dupacks because
1341 * we're probably after timeout (RFC6582 heuristics).
1342 * If Cumulative ack does not cover more than congestion threshold,
1343 * and:
1344 * 1) The following doesn't hold: The congestion window is greater
1345 * than SMSS bytes and the difference between highest_ack
1346 * and prev_highest_ack is at most 4*SMSS bytes
1347 * 2) Echoed timestamp in the last non-dup ack does not equal the
1348 * stored timestamp
1349 */
1350 if (seq_leq (tc->snd_una, tc->snd_congestion)
1351 && ((!(tc->cwnd > tc->snd_mss
1352 && tc->bytes_acked <= 4 * tc->snd_mss))
1353 || (tc->rcv_opts.tsecr != tc->tsecr_last_ack)))
1354 {
1355 tc->rcv_dupacks = 0;
1356 return;
1357 }
1358 }
Florin Coras93992a92017-05-24 18:03:56 -07001359
Florin Coras93992a92017-05-24 18:03:56 -07001360 if (!tc->bytes_acked)
1361 return;
1362
1363partial_ack:
Florin Corasd2aab832018-05-22 11:39:59 -07001364 TCP_EVT_DBG (TCP_EVT_CC_PACK, tc);
1365
Florin Coras93992a92017-05-24 18:03:56 -07001366 /*
1367 * Legitimate ACK. 1) See if we can exit recovery
1368 */
Florin Coras93992a92017-05-24 18:03:56 -07001369
1370 if (seq_geq (tc->snd_una, tc->snd_congestion))
1371 {
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001372 tcp_retransmit_timer_update (tc);
1373
Florin Coras93992a92017-05-24 18:03:56 -07001374 /* If spurious return, we've already updated everything */
1375 if (tcp_cc_recover (tc))
Florin Corasf03a59a2017-06-09 21:07:32 -07001376 {
1377 tc->tsecr_last_ack = tc->rcv_opts.tsecr;
1378 return;
1379 }
Florin Coras93992a92017-05-24 18:03:56 -07001380
1381 tc->snd_nxt = tc->snd_una_max;
1382
1383 /* Treat as congestion avoidance ack */
Florin Corasd67f1122018-05-21 17:47:40 -07001384 tcp_cc_rcv_ack (tc);
Florin Coras93992a92017-05-24 18:03:56 -07001385 return;
1386 }
1387
1388 /*
1389 * Legitimate ACK. 2) If PARTIAL ACK try to retransmit
1390 */
Florin Coras93992a92017-05-24 18:03:56 -07001391
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001392 /* XXX limit this only to first partial ack? */
1393 tcp_retransmit_timer_force_update (tc);
1394
Florin Coras93992a92017-05-24 18:03:56 -07001395 /* RFC6675: If the incoming ACK is a cumulative acknowledgment,
Florin Corasd2aab832018-05-22 11:39:59 -07001396 * reset dupacks to 0. Also needed if in congestion recovery */
Florin Coras93992a92017-05-24 18:03:56 -07001397 tc->rcv_dupacks = 0;
1398
Florin Coras93992a92017-05-24 18:03:56 -07001399 /* Post RTO timeout don't try anything fancy */
1400 if (tcp_in_recovery (tc))
Florin Corasd2aab832018-05-22 11:39:59 -07001401 {
Florin Corasd67f1122018-05-21 17:47:40 -07001402 tcp_cc_rcv_ack (tc);
Florin Coras3ec66b02018-08-23 16:27:05 -07001403 transport_add_tx_event (&tc->connection);
Florin Corasd2aab832018-05-22 11:39:59 -07001404 return;
1405 }
Florin Coras93992a92017-05-24 18:03:56 -07001406
1407 /* Remove retransmitted bytes that have been delivered */
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001408 if (tcp_opts_sack_permitted (&tc->rcv_opts))
Florin Coras93992a92017-05-24 18:03:56 -07001409 {
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001410 ASSERT (tc->bytes_acked + tc->sack_sb.snd_una_adv
1411 >= tc->sack_sb.last_bytes_delivered
1412 || (tc->flags & TCP_CONN_FINSNT));
1413
Florin Coras93992a92017-05-24 18:03:56 -07001414 /* If we have sacks and we haven't gotten an ack beyond high_rxt,
1415 * remove sacked bytes delivered */
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001416 if (seq_lt (tc->snd_una, tc->sack_sb.high_rxt))
1417 {
1418 rxt_delivered = tc->bytes_acked + tc->sack_sb.snd_una_adv
1419 - tc->sack_sb.last_bytes_delivered;
1420 ASSERT (tc->snd_rxt_bytes >= rxt_delivered);
1421 tc->snd_rxt_bytes -= rxt_delivered;
1422 }
1423 else
1424 {
1425 /* Apparently all retransmitted holes have been acked */
1426 tc->snd_rxt_bytes = 0;
1427 }
Dave Barach68b0fb02017-02-28 15:15:56 -05001428 }
1429 else
1430 {
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001431 if (tc->snd_rxt_bytes > tc->bytes_acked)
1432 tc->snd_rxt_bytes -= tc->bytes_acked;
1433 else
1434 tc->snd_rxt_bytes = 0;
Dave Barach68b0fb02017-02-28 15:15:56 -05001435 }
Dave Barach68b0fb02017-02-28 15:15:56 -05001436
Florin Coras93992a92017-05-24 18:03:56 -07001437 tc->cc_algo->rcv_cong_ack (tc, TCP_CC_PARTIALACK);
Dave Barach68b0fb02017-02-28 15:15:56 -05001438
Florin Coras93992a92017-05-24 18:03:56 -07001439 /*
1440 * Since this was a partial ack, try to retransmit some more data
1441 */
Florin Corasbf4d5ce2018-10-19 16:26:24 -07001442 tcp_program_fastretransmit (tc);
Dave Barach68b0fb02017-02-28 15:15:56 -05001443}
1444
Florin Coras93992a92017-05-24 18:03:56 -07001445/**
1446 * Process incoming ACK
1447 */
Dave Barach68b0fb02017-02-28 15:15:56 -05001448static int
1449tcp_rcv_ack (tcp_connection_t * tc, vlib_buffer_t * b,
1450 tcp_header_t * th, u32 * next, u32 * error)
1451{
Florin Coras93992a92017-05-24 18:03:56 -07001452 u32 prev_snd_wnd, prev_snd_una;
1453 u8 is_dack;
Dave Barach68b0fb02017-02-28 15:15:56 -05001454
Florin Corasf03a59a2017-06-09 21:07:32 -07001455 TCP_EVT_DBG (TCP_EVT_CC_STAT, tc);
1456
Florin Coras6792ec02017-03-13 03:49:51 -07001457 /* If the ACK acks something not yet sent (SEG.ACK > SND.NXT) */
Florin Coras93992a92017-05-24 18:03:56 -07001458 if (PREDICT_FALSE (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_nxt)))
Dave Barach68b0fb02017-02-28 15:15:56 -05001459 {
Florin Corasca1c8f32018-05-23 21:01:30 -07001460 /* When we entered recovery, we reset snd_nxt to snd_una. Seems peer
1461 * still has the data so accept the ack */
1462 if (tcp_in_recovery (tc)
Florin Coras3ec66b02018-08-23 16:27:05 -07001463 && seq_leq (vnet_buffer (b)->tcp.ack_number, tc->snd_congestion))
Florin Corasca1c8f32018-05-23 21:01:30 -07001464 {
Florin Coras3ec66b02018-08-23 16:27:05 -07001465 tc->snd_nxt = vnet_buffer (b)->tcp.ack_number;
1466 if (seq_gt (tc->snd_nxt, tc->snd_una_max))
1467 tc->snd_una_max = tc->snd_nxt;
Florin Corasca1c8f32018-05-23 21:01:30 -07001468 goto process_ack;
1469 }
1470
Florin Coras6792ec02017-03-13 03:49:51 -07001471 /* If we have outstanding data and this is within the window, accept it,
1472 * probably retransmit has timed out. Otherwise ACK segment and then
1473 * drop it */
1474 if (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_una_max))
1475 {
1476 tcp_make_ack (tc, b);
1477 *next = tcp_next_output (tc->c_is_ip4);
1478 *error = TCP_ERROR_ACK_INVALID;
1479 TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 0,
1480 vnet_buffer (b)->tcp.ack_number);
1481 return -1;
1482 }
1483
Florin Coras6792ec02017-03-13 03:49:51 -07001484 TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 2,
1485 vnet_buffer (b)->tcp.ack_number);
Florin Coras3e350af2017-03-30 02:54:28 -07001486
1487 tc->snd_nxt = vnet_buffer (b)->tcp.ack_number;
1488 *error = TCP_ERROR_ACK_FUTURE;
Dave Barach68b0fb02017-02-28 15:15:56 -05001489 }
1490
Florin Coras6792ec02017-03-13 03:49:51 -07001491 /* If old ACK, probably it's an old dupack */
Florin Coras93992a92017-05-24 18:03:56 -07001492 if (PREDICT_FALSE (seq_lt (vnet_buffer (b)->tcp.ack_number, tc->snd_una)))
Dave Barach68b0fb02017-02-28 15:15:56 -05001493 {
1494 *error = TCP_ERROR_ACK_OLD;
Florin Coras6792ec02017-03-13 03:49:51 -07001495 TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 1,
1496 vnet_buffer (b)->tcp.ack_number);
1497 if (tcp_in_fastrecovery (tc) && tc->rcv_dupacks == TCP_DUPACK_THRESHOLD)
Florin Corasca1c8f32018-05-23 21:01:30 -07001498 tcp_cc_handle_event (tc, 1);
Florin Corasc28764f2017-04-26 00:08:42 -07001499 /* Don't drop yet */
1500 return 0;
Dave Barach68b0fb02017-02-28 15:15:56 -05001501 }
1502
Florin Coras93992a92017-05-24 18:03:56 -07001503 /*
1504 * Looks okay, process feedback
1505 */
Florin Corasca1c8f32018-05-23 21:01:30 -07001506process_ack:
Florin Coras93992a92017-05-24 18:03:56 -07001507 if (tcp_opts_sack_permitted (&tc->rcv_opts))
Dave Barach68b0fb02017-02-28 15:15:56 -05001508 tcp_rcv_sacks (tc, vnet_buffer (b)->tcp.ack_number);
1509
Florin Coras93992a92017-05-24 18:03:56 -07001510 prev_snd_wnd = tc->snd_wnd;
1511 prev_snd_una = tc->snd_una;
1512 tcp_update_snd_wnd (tc, vnet_buffer (b)->tcp.seq_number,
1513 vnet_buffer (b)->tcp.ack_number,
1514 clib_net_to_host_u16 (th->window) << tc->snd_wscale);
1515 tc->bytes_acked = vnet_buffer (b)->tcp.ack_number - tc->snd_una;
1516 tc->snd_una = vnet_buffer (b)->tcp.ack_number + tc->sack_sb.snd_una_adv;
1517 tcp_validate_txf_size (tc, tc->bytes_acked);
Dave Barach68b0fb02017-02-28 15:15:56 -05001518
Florin Coras93992a92017-05-24 18:03:56 -07001519 if (tc->bytes_acked)
1520 tcp_dequeue_acked (tc, vnet_buffer (b)->tcp.ack_number);
1521
Florin Coras6534b7a2017-07-18 05:38:03 -04001522 TCP_EVT_DBG (TCP_EVT_ACK_RCVD, tc);
1523
Florin Coras93992a92017-05-24 18:03:56 -07001524 /*
1525 * Check if we have congestion event
1526 */
1527
1528 if (tcp_ack_is_cc_event (tc, b, prev_snd_wnd, prev_snd_una, &is_dack))
Dave Barach68b0fb02017-02-28 15:15:56 -05001529 {
Florin Coras93992a92017-05-24 18:03:56 -07001530 tcp_cc_handle_event (tc, is_dack);
Florin Corasb2215d62017-08-01 16:56:58 -07001531 if (!tcp_in_cong_recovery (tc))
1532 return 0;
Dave Barach68b0fb02017-02-28 15:15:56 -05001533 *error = TCP_ERROR_ACK_DUP;
Florin Corasca09d072018-10-01 18:31:02 -07001534 if (vnet_buffer (b)->tcp.data_len || tcp_is_fin (th))
1535 return 0;
1536 return -1;
Dave Barach68b0fb02017-02-28 15:15:56 -05001537 }
1538
Florin Coras6792ec02017-03-13 03:49:51 -07001539 /*
Florin Coras93992a92017-05-24 18:03:56 -07001540 * Update congestion control (slow start/congestion avoidance)
Florin Coras6792ec02017-03-13 03:49:51 -07001541 */
Florin Coras93992a92017-05-24 18:03:56 -07001542 tcp_cc_update (tc, b);
Florin Coras00cd22d2018-04-18 13:20:18 -07001543 *error = TCP_ERROR_ACK_OK;
Dave Barach68b0fb02017-02-28 15:15:56 -05001544 return 0;
1545}
1546
Florin Coras3eb50622017-07-13 01:24:57 -04001547static u8
1548tcp_sack_vector_is_sane (sack_block_t * sacks)
1549{
1550 int i;
1551 for (i = 1; i < vec_len (sacks); i++)
1552 {
1553 if (sacks[i - 1].end == sacks[i].start)
1554 return 0;
1555 }
1556 return 1;
1557}
1558
Dave Barach68b0fb02017-02-28 15:15:56 -05001559/**
1560 * Build SACK list as per RFC2018.
1561 *
1562 * Makes sure the first block contains the segment that generated the current
1563 * ACK and the following ones are the ones most recently reported in SACK
1564 * blocks.
1565 *
1566 * @param tc TCP connection for which the SACK list is updated
1567 * @param start Start sequence number of the newest SACK block
1568 * @param end End sequence of the newest SACK block
1569 */
Florin Coras45d34962017-04-25 00:05:27 -07001570void
Dave Barach68b0fb02017-02-28 15:15:56 -05001571tcp_update_sack_list (tcp_connection_t * tc, u32 start, u32 end)
1572{
Florin Coras45d34962017-04-25 00:05:27 -07001573 sack_block_t *new_list = 0, *block = 0;
Dave Barach68b0fb02017-02-28 15:15:56 -05001574 int i;
Dave Barach68b0fb02017-02-28 15:15:56 -05001575
1576 /* If the first segment is ooo add it to the list. Last write might've moved
1577 * rcv_nxt over the first segment. */
1578 if (seq_lt (tc->rcv_nxt, start))
1579 {
Florin Coras45d34962017-04-25 00:05:27 -07001580 vec_add2 (new_list, block, 1);
1581 block->start = start;
1582 block->end = end;
Dave Barach68b0fb02017-02-28 15:15:56 -05001583 }
1584
1585 /* Find the blocks still worth keeping. */
1586 for (i = 0; i < vec_len (tc->snd_sacks); i++)
1587 {
Florin Coras45d34962017-04-25 00:05:27 -07001588 /* Discard if rcv_nxt advanced beyond current block */
1589 if (seq_leq (tc->snd_sacks[i].start, tc->rcv_nxt))
Dave Barach68b0fb02017-02-28 15:15:56 -05001590 continue;
1591
Florin Coras45d34962017-04-25 00:05:27 -07001592 /* Merge or drop if segment overlapped by the new segment */
1593 if (block && (seq_geq (tc->snd_sacks[i].end, new_list[0].start)
1594 && seq_leq (tc->snd_sacks[i].start, new_list[0].end)))
1595 {
1596 if (seq_lt (tc->snd_sacks[i].start, new_list[0].start))
1597 new_list[0].start = tc->snd_sacks[i].start;
1598 if (seq_lt (new_list[0].end, tc->snd_sacks[i].end))
1599 new_list[0].end = tc->snd_sacks[i].end;
1600 continue;
1601 }
1602
1603 /* Save to new SACK list if we have space. */
1604 if (vec_len (new_list) < TCP_MAX_SACK_BLOCKS)
1605 {
1606 vec_add1 (new_list, tc->snd_sacks[i]);
1607 }
Florin Corasf03a59a2017-06-09 21:07:32 -07001608 else
1609 {
1610 clib_warning ("sack discarded");
1611 }
Dave Barach68b0fb02017-02-28 15:15:56 -05001612 }
1613
Florin Coras45d34962017-04-25 00:05:27 -07001614 ASSERT (vec_len (new_list) <= TCP_MAX_SACK_BLOCKS);
Florin Coras6792ec02017-03-13 03:49:51 -07001615
Dave Barach68b0fb02017-02-28 15:15:56 -05001616 /* Replace old vector with new one */
1617 vec_free (tc->snd_sacks);
1618 tc->snd_sacks = new_list;
Florin Coras3eb50622017-07-13 01:24:57 -04001619
1620 /* Segments should not 'touch' */
1621 ASSERT (tcp_sack_vector_is_sane (tc->snd_sacks));
Dave Barach68b0fb02017-02-28 15:15:56 -05001622}
1623
Florin Corasca1c8f32018-05-23 21:01:30 -07001624u32
1625tcp_sack_list_bytes (tcp_connection_t * tc)
1626{
1627 u32 bytes = 0, i;
1628 for (i = 0; i < vec_len (tc->snd_sacks); i++)
1629 bytes += tc->snd_sacks[i].end - tc->snd_sacks[i].start;
1630 return bytes;
1631}
1632
Dave Barach68b0fb02017-02-28 15:15:56 -05001633/** Enqueue data for delivery to application */
Florin Coras0dbd5172018-06-25 16:19:34 -07001634static int
Dave Barach68b0fb02017-02-28 15:15:56 -05001635tcp_session_enqueue_data (tcp_connection_t * tc, vlib_buffer_t * b,
1636 u16 data_len)
1637{
Florin Coras1f152cd2017-08-18 19:28:03 -07001638 int written, error = TCP_ERROR_ENQUEUED;
Dave Barach68b0fb02017-02-28 15:15:56 -05001639
Dave Barach2c25a622017-06-26 11:35:07 -04001640 ASSERT (seq_geq (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt));
Florin Coras00cd22d2018-04-18 13:20:18 -07001641 ASSERT (data_len);
Florin Coras3cbc04b2017-10-02 00:18:51 -07001642 written = session_enqueue_stream_connection (&tc->connection, b, 0,
1643 1 /* queue event */ , 1);
Dave Barach68b0fb02017-02-28 15:15:56 -05001644
Florin Coras6792ec02017-03-13 03:49:51 -07001645 TCP_EVT_DBG (TCP_EVT_INPUT, tc, 0, data_len, written);
1646
Dave Barach68b0fb02017-02-28 15:15:56 -05001647 /* Update rcv_nxt */
1648 if (PREDICT_TRUE (written == data_len))
1649 {
Florin Coras1f152cd2017-08-18 19:28:03 -07001650 tc->rcv_nxt += written;
Dave Barach68b0fb02017-02-28 15:15:56 -05001651 }
1652 /* If more data written than expected, account for out-of-order bytes. */
1653 else if (written > data_len)
1654 {
Florin Coras1f152cd2017-08-18 19:28:03 -07001655 tc->rcv_nxt += written;
Dave Barach68b0fb02017-02-28 15:15:56 -05001656
1657 /* Send ACK confirming the update */
1658 tc->flags |= TCP_CONN_SNDACK;
Florin Corasca1c8f32018-05-23 21:01:30 -07001659 TCP_EVT_DBG (TCP_EVT_CC_INPUT, tc, data_len, written);
Florin Coras6792ec02017-03-13 03:49:51 -07001660 }
1661 else if (written > 0)
1662 {
1663 /* We've written something but FIFO is probably full now */
1664 tc->rcv_nxt += written;
Dave Barach68b0fb02017-02-28 15:15:56 -05001665
Florin Coras6792ec02017-03-13 03:49:51 -07001666 /* Depending on how fast the app is, all remaining buffers in burst will
Florin Coras3e350af2017-03-30 02:54:28 -07001667 * not be enqueued. Inform peer */
1668 tc->flags |= TCP_CONN_SNDACK;
1669
Florin Coras1f152cd2017-08-18 19:28:03 -07001670 error = TCP_ERROR_PARTIALLY_ENQUEUED;
Dave Barach68b0fb02017-02-28 15:15:56 -05001671 }
1672 else
1673 {
Florin Coras3e350af2017-03-30 02:54:28 -07001674 tc->flags |= TCP_CONN_SNDACK;
Dave Barach68b0fb02017-02-28 15:15:56 -05001675 return TCP_ERROR_FIFO_FULL;
1676 }
1677
Florin Coras6792ec02017-03-13 03:49:51 -07001678 /* Update SACK list if need be */
Florin Coras93992a92017-05-24 18:03:56 -07001679 if (tcp_opts_sack_permitted (&tc->rcv_opts))
Florin Coras6792ec02017-03-13 03:49:51 -07001680 {
1681 /* Remove SACK blocks that have been delivered */
1682 tcp_update_sack_list (tc, tc->rcv_nxt, tc->rcv_nxt);
1683 }
1684
Florin Coras1f152cd2017-08-18 19:28:03 -07001685 return error;
Dave Barach68b0fb02017-02-28 15:15:56 -05001686}
1687
1688/** Enqueue out-of-order data */
Florin Coras0dbd5172018-06-25 16:19:34 -07001689static int
Dave Barach68b0fb02017-02-28 15:15:56 -05001690tcp_session_enqueue_ooo (tcp_connection_t * tc, vlib_buffer_t * b,
1691 u16 data_len)
1692{
1693 stream_session_t *s0;
Florin Coras3eb50622017-07-13 01:24:57 -04001694 int rv, offset;
Florin Coras6792ec02017-03-13 03:49:51 -07001695
Florin Corasf03a59a2017-06-09 21:07:32 -07001696 ASSERT (seq_gt (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt));
Florin Coras00cd22d2018-04-18 13:20:18 -07001697 ASSERT (data_len);
Dave Barach68b0fb02017-02-28 15:15:56 -05001698
Florin Corasf03a59a2017-06-09 21:07:32 -07001699 /* Enqueue out-of-order data with relative offset */
Florin Coras3cbc04b2017-10-02 00:18:51 -07001700 rv = session_enqueue_stream_connection (&tc->connection, b,
1701 vnet_buffer (b)->tcp.seq_number -
1702 tc->rcv_nxt, 0 /* queue event */ ,
1703 0);
Florin Coras6792ec02017-03-13 03:49:51 -07001704
1705 /* Nothing written */
1706 if (rv)
1707 {
1708 TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, 0);
1709 return TCP_ERROR_FIFO_FULL;
1710 }
1711
1712 TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, data_len);
Dave Barach68b0fb02017-02-28 15:15:56 -05001713
1714 /* Update SACK list if in use */
Florin Coras93992a92017-05-24 18:03:56 -07001715 if (tcp_opts_sack_permitted (&tc->rcv_opts))
Dave Barach68b0fb02017-02-28 15:15:56 -05001716 {
1717 ooo_segment_t *newest;
1718 u32 start, end;
1719
Florin Corascea194d2017-10-02 00:18:51 -07001720 s0 = session_get (tc->c_s_index, tc->c_thread_index);
Florin Corasf6d68ed2017-05-07 19:12:02 -07001721
Dave Barach68b0fb02017-02-28 15:15:56 -05001722 /* Get the newest segment from the fifo */
1723 newest = svm_fifo_newest_ooo_segment (s0->server_rx_fifo);
Florin Corasf03a59a2017-06-09 21:07:32 -07001724 if (newest)
1725 {
Florin Coras3eb50622017-07-13 01:24:57 -04001726 offset = ooo_segment_offset (s0->server_rx_fifo, newest);
1727 ASSERT (offset <= vnet_buffer (b)->tcp.seq_number - tc->rcv_nxt);
1728 start = tc->rcv_nxt + offset;
Florin Corasf03a59a2017-06-09 21:07:32 -07001729 end = start + ooo_segment_length (s0->server_rx_fifo, newest);
1730 tcp_update_sack_list (tc, start, end);
Florin Coras3eb50622017-07-13 01:24:57 -04001731 svm_fifo_newest_ooo_segment_reset (s0->server_rx_fifo);
Florin Corasca1c8f32018-05-23 21:01:30 -07001732 TCP_EVT_DBG (TCP_EVT_CC_SACKS, tc);
Florin Corasf03a59a2017-06-09 21:07:32 -07001733 }
Dave Barach68b0fb02017-02-28 15:15:56 -05001734 }
1735
Florin Coras00cd22d2018-04-18 13:20:18 -07001736 return TCP_ERROR_ENQUEUED_OOO;
Dave Barach68b0fb02017-02-28 15:15:56 -05001737}
1738
1739/**
Florin Coras6792ec02017-03-13 03:49:51 -07001740 * Check if ACK could be delayed. If ack can be delayed, it should return
1741 * true for a full frame. If we're always acking return 0.
Dave Barach68b0fb02017-02-28 15:15:56 -05001742 */
1743always_inline int
1744tcp_can_delack (tcp_connection_t * tc)
1745{
Florin Coras6792ec02017-03-13 03:49:51 -07001746 /* Send ack if ... */
1747 if (TCP_ALWAYS_ACK
1748 /* just sent a rcv wnd 0 */
1749 || (tc->flags & TCP_CONN_SENT_RCV_WND0) != 0
1750 /* constrained to send ack */
1751 || (tc->flags & TCP_CONN_SNDACK) != 0
1752 /* we're almost out of tx wnd */
Florin Corasca1c8f32018-05-23 21:01:30 -07001753 || tcp_available_cc_snd_space (tc) < 4 * tc->snd_mss)
Florin Coras6792ec02017-03-13 03:49:51 -07001754 return 0;
Dave Barach68b0fb02017-02-28 15:15:56 -05001755
Florin Coras6792ec02017-03-13 03:49:51 -07001756 return 1;
Dave Barach68b0fb02017-02-28 15:15:56 -05001757}
1758
1759static int
Florin Corasb2215d62017-08-01 16:56:58 -07001760tcp_buffer_discard_bytes (vlib_buffer_t * b, u32 n_bytes_to_drop)
1761{
Florin Coras1f152cd2017-08-18 19:28:03 -07001762 u32 discard, first = b->current_length;
Florin Corasb2215d62017-08-01 16:56:58 -07001763 vlib_main_t *vm = vlib_get_main ();
1764
Florin Coras1f152cd2017-08-18 19:28:03 -07001765 /* Handle multi-buffer segments */
Florin Corasb2215d62017-08-01 16:56:58 -07001766 if (n_bytes_to_drop > b->current_length)
1767 {
1768 if (!(b->flags & VLIB_BUFFER_NEXT_PRESENT))
1769 return -1;
1770 do
1771 {
1772 discard = clib_min (n_bytes_to_drop, b->current_length);
1773 vlib_buffer_advance (b, discard);
1774 b = vlib_get_buffer (vm, b->next_buffer);
1775 n_bytes_to_drop -= discard;
1776 }
1777 while (n_bytes_to_drop);
Florin Coras1f152cd2017-08-18 19:28:03 -07001778 if (n_bytes_to_drop > first)
1779 b->total_length_not_including_first_buffer -= n_bytes_to_drop - first;
Florin Corasb2215d62017-08-01 16:56:58 -07001780 }
Florin Coras1f152cd2017-08-18 19:28:03 -07001781 else
1782 vlib_buffer_advance (b, n_bytes_to_drop);
1783 vnet_buffer (b)->tcp.data_len -= n_bytes_to_drop;
Florin Corasb2215d62017-08-01 16:56:58 -07001784 return 0;
1785}
1786
Florin Coras00cd22d2018-04-18 13:20:18 -07001787/**
1788 * Receive buffer for connection and handle acks
1789 *
1790 * It handles both in order or out-of-order data.
1791 */
Florin Corasb2215d62017-08-01 16:56:58 -07001792static int
Florin Coras00cd22d2018-04-18 13:20:18 -07001793tcp_segment_rcv (tcp_connection_t * tc, vlib_buffer_t * b, u32 * next0)
Dave Barach68b0fb02017-02-28 15:15:56 -05001794{
Florin Coras00cd22d2018-04-18 13:20:18 -07001795 u32 error, n_bytes_to_drop, n_data_bytes;
Florin Coras6534b7a2017-07-18 05:38:03 -04001796
1797 vlib_buffer_advance (b, vnet_buffer (b)->tcp.data_offset);
1798 n_data_bytes = vnet_buffer (b)->tcp.data_len;
1799 ASSERT (n_data_bytes);
Dave Barach68b0fb02017-02-28 15:15:56 -05001800
1801 /* Handle out-of-order data */
1802 if (PREDICT_FALSE (vnet_buffer (b)->tcp.seq_number != tc->rcv_nxt))
1803 {
Florin Coras6792ec02017-03-13 03:49:51 -07001804 /* Old sequence numbers allowed through because they overlapped
1805 * the rx window */
1806 if (seq_lt (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt))
Dave Barach68b0fb02017-02-28 15:15:56 -05001807 {
Florin Coras00cd22d2018-04-18 13:20:18 -07001808 /* Completely in the past (possible retransmit). Ack
1809 * retransmissions since we may not have any data to send */
Florin Corasf03a59a2017-06-09 21:07:32 -07001810 if (seq_leq (vnet_buffer (b)->tcp.seq_end, tc->rcv_nxt))
Florin Coras6534b7a2017-07-18 05:38:03 -04001811 {
Florin Coras6534b7a2017-07-18 05:38:03 -04001812 tcp_make_ack (tc, b);
Florin Coras00cd22d2018-04-18 13:20:18 -07001813 error = TCP_ERROR_SEGMENT_OLD;
Florin Coras6534b7a2017-07-18 05:38:03 -04001814 *next0 = tcp_next_output (tc->c_is_ip4);
1815 goto done;
1816 }
Dave Barach259cdae2017-05-15 16:27:05 -04001817
Florin Coras00cd22d2018-04-18 13:20:18 -07001818 /* Chop off the bytes in the past and see if what is left
1819 * can be enqueued in order */
Florin Corasdb84e572017-05-09 18:54:52 -07001820 n_bytes_to_drop = tc->rcv_nxt - vnet_buffer (b)->tcp.seq_number;
1821 n_data_bytes -= n_bytes_to_drop;
Dave Barach2c25a622017-06-26 11:35:07 -04001822 vnet_buffer (b)->tcp.seq_number = tc->rcv_nxt;
Florin Corasb2215d62017-08-01 16:56:58 -07001823 if (tcp_buffer_discard_bytes (b, n_bytes_to_drop))
Florin Coras00cd22d2018-04-18 13:20:18 -07001824 {
1825 error = TCP_ERROR_SEGMENT_OLD;
1826 *next0 = tcp_next_drop (tc->c_is_ip4);
1827 goto done;
1828 }
Florin Corasdb84e572017-05-09 18:54:52 -07001829 goto in_order;
Dave Barach68b0fb02017-02-28 15:15:56 -05001830 }
1831
Florin Coras00cd22d2018-04-18 13:20:18 -07001832 /* RFC2581: Enqueue and send DUPACK for fast retransmit */
Florin Coras6792ec02017-03-13 03:49:51 -07001833 error = tcp_session_enqueue_ooo (tc, b, n_data_bytes);
Florin Coras6792ec02017-03-13 03:49:51 -07001834 *next0 = tcp_next_output (tc->c_is_ip4);
Florin Coras00cd22d2018-04-18 13:20:18 -07001835 tcp_make_ack (tc, b);
1836 vnet_buffer (b)->tcp.flags = TCP_BUF_FLAG_DUPACK;
Florin Corasca1c8f32018-05-23 21:01:30 -07001837 TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc, vnet_buffer (b)->tcp);
Dave Barach68b0fb02017-02-28 15:15:56 -05001838 goto done;
1839 }
1840
Florin Corasdb84e572017-05-09 18:54:52 -07001841in_order:
1842
Dave Barach68b0fb02017-02-28 15:15:56 -05001843 /* In order data, enqueue. Fifo figures out by itself if any out-of-order
1844 * segments can be enqueued after fifo tail offset changes. */
1845 error = tcp_session_enqueue_data (tc, b, n_data_bytes);
Florin Coras3e350af2017-03-30 02:54:28 -07001846 if (tcp_can_delack (tc))
Dave Barach68b0fb02017-02-28 15:15:56 -05001847 {
Florin Coras00cd22d2018-04-18 13:20:18 -07001848 *next0 = tcp_next_drop (tc->c_is_ip4);
Florin Coras6792ec02017-03-13 03:49:51 -07001849 if (!tcp_timer_is_active (tc, TCP_TIMER_DELACK))
1850 tcp_timer_set (tc, TCP_TIMER_DELACK, TCP_DELACK_TIME);
Florin Coras3e350af2017-03-30 02:54:28 -07001851 goto done;
Dave Barach68b0fb02017-02-28 15:15:56 -05001852 }
1853
Florin Coras3e350af2017-03-30 02:54:28 -07001854 *next0 = tcp_next_output (tc->c_is_ip4);
1855 tcp_make_ack (tc, b);
1856
Dave Barach68b0fb02017-02-28 15:15:56 -05001857done:
1858 return error;
1859}
1860
Clement Durand6cf260c2017-04-13 13:27:04 +02001861typedef struct
1862{
1863 tcp_header_t tcp_header;
1864 tcp_connection_t tcp_connection;
1865} tcp_rx_trace_t;
1866
Florin Coras0dbd5172018-06-25 16:19:34 -07001867static u8 *
Clement Durand6cf260c2017-04-13 13:27:04 +02001868format_tcp_rx_trace (u8 * s, va_list * args)
1869{
1870 CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
1871 CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
1872 tcp_rx_trace_t *t = va_arg (*args, tcp_rx_trace_t *);
Christophe Fontained3c008d2017-10-02 18:10:54 +02001873 u32 indent = format_get_indent (s);
Clement Durand6cf260c2017-04-13 13:27:04 +02001874
1875 s = format (s, "%U\n%U%U",
1876 format_tcp_header, &t->tcp_header, 128,
1877 format_white_space, indent,
Florin Corasbb292f42017-05-19 09:49:19 -07001878 format_tcp_connection, &t->tcp_connection, 1);
Clement Durand6cf260c2017-04-13 13:27:04 +02001879
1880 return s;
1881}
1882
Florin Coras0dbd5172018-06-25 16:19:34 -07001883static u8 *
Clement Durand6cf260c2017-04-13 13:27:04 +02001884format_tcp_rx_trace_short (u8 * s, va_list * args)
1885{
1886 CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
1887 CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
1888 tcp_rx_trace_t *t = va_arg (*args, tcp_rx_trace_t *);
1889
1890 s = format (s, "%d -> %d (%U)",
Florin Coras4df38712018-06-20 12:44:16 -07001891 clib_net_to_host_u16 (t->tcp_header.dst_port),
1892 clib_net_to_host_u16 (t->tcp_header.src_port), format_tcp_state,
Florin Corasbb292f42017-05-19 09:49:19 -07001893 t->tcp_connection.state);
Clement Durand6cf260c2017-04-13 13:27:04 +02001894
1895 return s;
1896}
1897
Florin Coras4df38712018-06-20 12:44:16 -07001898static void
Florin Coras82b13a82017-04-25 11:58:06 -07001899tcp_set_rx_trace_data (tcp_rx_trace_t * t0, tcp_connection_t * tc0,
1900 tcp_header_t * th0, vlib_buffer_t * b0, u8 is_ip4)
1901{
1902 if (tc0)
1903 {
1904 clib_memcpy (&t0->tcp_connection, tc0, sizeof (t0->tcp_connection));
1905 }
1906 else
1907 {
1908 th0 = tcp_buffer_hdr (b0);
1909 }
1910 clib_memcpy (&t0->tcp_header, th0, sizeof (t0->tcp_header));
1911}
1912
Florin Coras4df38712018-06-20 12:44:16 -07001913static void
1914tcp_established_trace_frame (vlib_main_t * vm, vlib_node_runtime_t * node,
1915 vlib_frame_t * frame, u8 is_ip4)
1916{
1917 u32 *from, n_left;
1918
1919 n_left = frame->n_vectors;
1920 from = vlib_frame_vector_args (frame);
1921
1922 while (n_left >= 1)
1923 {
1924 tcp_connection_t *tc0;
1925 tcp_rx_trace_t *t0;
1926 tcp_header_t *th0;
1927 vlib_buffer_t *b0;
1928 u32 bi0;
1929
1930 bi0 = from[0];
1931 b0 = vlib_get_buffer (vm, bi0);
1932
1933 if (b0->flags & VLIB_BUFFER_IS_TRACED)
1934 {
1935 t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
1936 tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index,
1937 vm->thread_index);
1938 th0 = tcp_buffer_hdr (b0);
1939 tcp_set_rx_trace_data (t0, tc0, th0, b0, is_ip4);
1940 }
1941
1942 from += 1;
1943 n_left -= 1;
1944 }
1945}
1946
Florin Coras6792ec02017-03-13 03:49:51 -07001947always_inline void
Florin Coras00cd22d2018-04-18 13:20:18 -07001948tcp_node_inc_counter_i (vlib_main_t * vm, u32 tcp4_node, u32 tcp6_node,
1949 u8 is_ip4, u32 evt, u32 val)
Dave Barach68b0fb02017-02-28 15:15:56 -05001950{
Florin Coras6792ec02017-03-13 03:49:51 -07001951 if (is_ip4)
Florin Coras3cbc04b2017-10-02 00:18:51 -07001952 vlib_node_increment_counter (vm, tcp4_node, evt, val);
Florin Coras6792ec02017-03-13 03:49:51 -07001953 else
Florin Coras3cbc04b2017-10-02 00:18:51 -07001954 vlib_node_increment_counter (vm, tcp6_node, evt, val);
Dave Barach68b0fb02017-02-28 15:15:56 -05001955}
1956
Florin Coras00cd22d2018-04-18 13:20:18 -07001957#define tcp_maybe_inc_counter(node_id, err, count) \
1958{ \
1959 if (next0 != tcp_next_drop (is_ip4)) \
1960 tcp_node_inc_counter_i (vm, tcp4_##node_id##_node.index, \
1961 tcp6_##node_id##_node.index, is_ip4, err, \
1962 1); \
1963}
1964#define tcp_inc_counter(node_id, err, count) \
1965 tcp_node_inc_counter_i (vm, tcp4_##node_id##_node.index, \
1966 tcp6_##node_id##_node.index, is_ip4, \
1967 err, count)
1968#define tcp_maybe_inc_err_counter(cnts, err) \
1969{ \
1970 cnts[err] += (next0 != tcp_next_drop (is_ip4)); \
1971}
1972#define tcp_inc_err_counter(cnts, err, val) \
1973{ \
1974 cnts[err] += val; \
1975}
1976#define tcp_store_err_counters(node_id, cnts) \
1977{ \
1978 int i; \
1979 for (i = 0; i < TCP_N_ERROR; i++) \
1980 if (cnts[i]) \
1981 tcp_inc_counter(node_id, i, cnts[i]); \
1982}
1983
1984
Dave Barach68b0fb02017-02-28 15:15:56 -05001985always_inline uword
1986tcp46_established_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
Florin Coras4df38712018-06-20 12:44:16 -07001987 vlib_frame_t * frame, int is_ip4)
Dave Barach68b0fb02017-02-28 15:15:56 -05001988{
Florin Coras4df38712018-06-20 12:44:16 -07001989 u32 thread_index = vm->thread_index, errors = 0;
Florin Coras00cd22d2018-04-18 13:20:18 -07001990 u32 n_left_from, next_index, *from, *to_next;
1991 u16 err_counters[TCP_N_ERROR] = { 0 };
Dave Barach1f75cfd2017-04-14 16:46:44 -04001992 u8 is_fin = 0;
Dave Barach68b0fb02017-02-28 15:15:56 -05001993
Florin Coras4df38712018-06-20 12:44:16 -07001994 if (node->flags & VLIB_NODE_FLAG_TRACE)
1995 tcp_established_trace_frame (vm, node, frame, is_ip4);
1996
1997 from = vlib_frame_vector_args (frame);
1998 n_left_from = frame->n_vectors;
Dave Barach68b0fb02017-02-28 15:15:56 -05001999 next_index = node->cached_next_index;
2000
2001 while (n_left_from > 0)
2002 {
2003 u32 n_left_to_next;
2004
2005 vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
Dave Barach68b0fb02017-02-28 15:15:56 -05002006 while (n_left_from > 0 && n_left_to_next > 0)
2007 {
2008 u32 bi0;
2009 vlib_buffer_t *b0;
2010 tcp_header_t *th0 = 0;
2011 tcp_connection_t *tc0;
Florin Coras00cd22d2018-04-18 13:20:18 -07002012 u32 next0 = tcp_next_drop (is_ip4), error0 = TCP_ERROR_ACK_OK;
Dave Barach68b0fb02017-02-28 15:15:56 -05002013
Florin Coras81a13db2018-03-16 08:48:31 -07002014 if (n_left_from > 1)
2015 {
2016 vlib_buffer_t *pb;
2017 pb = vlib_get_buffer (vm, from[1]);
2018 vlib_prefetch_buffer_header (pb, LOAD);
2019 CLIB_PREFETCH (pb->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
2020 }
2021
Dave Barach68b0fb02017-02-28 15:15:56 -05002022 bi0 = from[0];
2023 to_next[0] = bi0;
2024 from += 1;
2025 to_next += 1;
2026 n_left_from -= 1;
2027 n_left_to_next -= 1;
2028
2029 b0 = vlib_get_buffer (vm, bi0);
2030 tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index,
Florin Coras4df38712018-06-20 12:44:16 -07002031 thread_index);
Dave Barach68b0fb02017-02-28 15:15:56 -05002032
Florin Corasd79b41e2017-03-04 05:37:52 -08002033 if (PREDICT_FALSE (tc0 == 0))
2034 {
2035 error0 = TCP_ERROR_INVALID_CONNECTION;
Florin Coras6792ec02017-03-13 03:49:51 -07002036 goto done;
Florin Corasd79b41e2017-03-04 05:37:52 -08002037 }
2038
Florin Coras82b13a82017-04-25 11:58:06 -07002039 th0 = tcp_buffer_hdr (b0);
Florin Coras6534b7a2017-07-18 05:38:03 -04002040 /* N.B. buffer is rewritten if segment is ooo. Thus, th0 becomes a
2041 * dangling reference. */
2042 is_fin = tcp_is_fin (th0);
Dave Barach1f75cfd2017-04-14 16:46:44 -04002043
Dave Barach68b0fb02017-02-28 15:15:56 -05002044 /* SYNs, FINs and data consume sequence numbers */
2045 vnet_buffer (b0)->tcp.seq_end = vnet_buffer (b0)->tcp.seq_number
Florin Coras82b13a82017-04-25 11:58:06 -07002046 + tcp_is_syn (th0) + is_fin + vnet_buffer (b0)->tcp.data_len;
Dave Barach68b0fb02017-02-28 15:15:56 -05002047
2048 /* TODO header prediction fast path */
2049
2050 /* 1-4: check SEQ, RST, SYN */
Florin Coras00cd22d2018-04-18 13:20:18 -07002051 if (PREDICT_FALSE (tcp_segment_validate (vm, tc0, b0, th0, &next0,
2052 &error0)))
Dave Barach68b0fb02017-02-28 15:15:56 -05002053 {
Florin Coras00cd22d2018-04-18 13:20:18 -07002054 tcp_maybe_inc_err_counter (err_counters, error0);
Florin Corasca1c8f32018-05-23 21:01:30 -07002055 TCP_EVT_DBG (TCP_EVT_SEG_INVALID, tc0, vnet_buffer (b0)->tcp);
Florin Coras6792ec02017-03-13 03:49:51 -07002056 goto done;
Dave Barach68b0fb02017-02-28 15:15:56 -05002057 }
2058
2059 /* 5: check the ACK field */
Florin Coras00cd22d2018-04-18 13:20:18 -07002060 if (PREDICT_FALSE (tcp_rcv_ack (tc0, b0, th0, &next0, &error0)))
2061 {
2062 tcp_maybe_inc_err_counter (err_counters, error0);
2063 goto done;
2064 }
Dave Barach68b0fb02017-02-28 15:15:56 -05002065
2066 /* 6: check the URG bit TODO */
2067
2068 /* 7: process the segment text */
Florin Coras6534b7a2017-07-18 05:38:03 -04002069 if (vnet_buffer (b0)->tcp.data_len)
Florin Coras00cd22d2018-04-18 13:20:18 -07002070 {
2071 error0 = tcp_segment_rcv (tc0, b0, &next0);
2072 tcp_maybe_inc_err_counter (err_counters, error0);
2073 }
Dave Barach1f75cfd2017-04-14 16:46:44 -04002074
Dave Barach68b0fb02017-02-28 15:15:56 -05002075 /* 8: check the FIN bit */
Florin Coras6534b7a2017-07-18 05:38:03 -04002076 if (PREDICT_FALSE (is_fin))
Dave Barach68b0fb02017-02-28 15:15:56 -05002077 {
Florin Coras9d063042017-09-14 03:08:00 -04002078 /* Enter CLOSE-WAIT and notify session. To avoid lingering
Florin Corasd79b41e2017-03-04 05:37:52 -08002079 * in CLOSE-WAIT, set timer (reuse WAITCLOSE). */
Florin Coras9d063042017-09-14 03:08:00 -04002080 /* Account for the FIN if nothing else was received */
Florin Coras68810622017-07-24 17:40:28 -07002081 if (vnet_buffer (b0)->tcp.data_len == 0)
Florin Coras9d063042017-09-14 03:08:00 -04002082 tc0->rcv_nxt += 1;
2083 tcp_make_ack (tc0, b0);
2084 next0 = tcp_next_output (tc0->c_is_ip4);
2085 tc0->state = TCP_STATE_CLOSE_WAIT;
Dave Barach68b0fb02017-02-28 15:15:56 -05002086 stream_session_disconnect_notify (&tc0->connection);
Florin Coras68810622017-07-24 17:40:28 -07002087 tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_CLOSEWAIT_TIME);
Florin Coras9d063042017-09-14 03:08:00 -04002088 TCP_EVT_DBG (TCP_EVT_FIN_RCVD, tc0);
Florin Coras00cd22d2018-04-18 13:20:18 -07002089 tcp_inc_err_counter (err_counters, TCP_ERROR_FIN_RCVD, 1);
Dave Barach68b0fb02017-02-28 15:15:56 -05002090 }
2091
Florin Coras6792ec02017-03-13 03:49:51 -07002092 done:
Dave Barach68b0fb02017-02-28 15:15:56 -05002093 b0->error = node->errors[error0];
Dave Barach68b0fb02017-02-28 15:15:56 -05002094 vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
2095 n_left_to_next, bi0, next0);
2096 }
2097
2098 vlib_put_next_frame (vm, node, next_index, n_left_to_next);
2099 }
2100
Florin Coras3cbc04b2017-10-02 00:18:51 -07002101 errors = session_manager_flush_enqueue_events (TRANSPORT_PROTO_TCP,
Florin Coras4df38712018-06-20 12:44:16 -07002102 thread_index);
Florin Coras00cd22d2018-04-18 13:20:18 -07002103 err_counters[TCP_ERROR_EVENT_FIFO_FULL] = errors;
2104 tcp_store_err_counters (established, err_counters);
Florin Coras4df38712018-06-20 12:44:16 -07002105 tcp_flush_frame_to_output (vm, thread_index, is_ip4);
2106
2107 return frame->n_vectors;
Dave Barach68b0fb02017-02-28 15:15:56 -05002108}
2109
2110static uword
2111tcp4_established (vlib_main_t * vm, vlib_node_runtime_t * node,
2112 vlib_frame_t * from_frame)
2113{
2114 return tcp46_established_inline (vm, node, from_frame, 1 /* is_ip4 */ );
2115}
2116
2117static uword
2118tcp6_established (vlib_main_t * vm, vlib_node_runtime_t * node,
2119 vlib_frame_t * from_frame)
2120{
2121 return tcp46_established_inline (vm, node, from_frame, 0 /* is_ip4 */ );
2122}
2123
2124/* *INDENT-OFF* */
2125VLIB_REGISTER_NODE (tcp4_established_node) =
2126{
2127 .function = tcp4_established,
2128 .name = "tcp4-established",
2129 /* Takes a vector of packets. */
2130 .vector_size = sizeof (u32),
Florin Corase69f4952017-03-07 10:06:24 -08002131 .n_errors = TCP_N_ERROR,
2132 .error_strings = tcp_error_strings,
Dave Barach68b0fb02017-02-28 15:15:56 -05002133 .n_next_nodes = TCP_ESTABLISHED_N_NEXT,
2134 .next_nodes =
2135 {
2136#define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n,
2137 foreach_tcp_state_next
2138#undef _
2139 },
Clement Durand6cf260c2017-04-13 13:27:04 +02002140 .format_trace = format_tcp_rx_trace_short,
Dave Barach68b0fb02017-02-28 15:15:56 -05002141};
2142/* *INDENT-ON* */
2143
2144VLIB_NODE_FUNCTION_MULTIARCH (tcp4_established_node, tcp4_established);
2145
2146/* *INDENT-OFF* */
2147VLIB_REGISTER_NODE (tcp6_established_node) =
2148{
2149 .function = tcp6_established,
2150 .name = "tcp6-established",
2151 /* Takes a vector of packets. */
2152 .vector_size = sizeof (u32),
2153 .n_errors = TCP_N_ERROR,
2154 .error_strings = tcp_error_strings,
2155 .n_next_nodes = TCP_ESTABLISHED_N_NEXT,
2156 .next_nodes =
2157 {
2158#define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n,
2159 foreach_tcp_state_next
2160#undef _
2161 },
Clement Durand6cf260c2017-04-13 13:27:04 +02002162 .format_trace = format_tcp_rx_trace_short,
Dave Barach68b0fb02017-02-28 15:15:56 -05002163};
2164/* *INDENT-ON* */
2165
2166
2167VLIB_NODE_FUNCTION_MULTIARCH (tcp6_established_node, tcp6_established);
2168
2169vlib_node_registration_t tcp4_syn_sent_node;
2170vlib_node_registration_t tcp6_syn_sent_node;
2171
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002172static u8
2173tcp_lookup_is_valid (tcp_connection_t * tc, tcp_header_t * hdr)
2174{
Florin Corascea194d2017-10-02 00:18:51 -07002175 transport_connection_t *tmp = 0;
2176 u64 handle;
2177
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002178 if (!tc)
2179 return 1;
2180
Florin Coras70131132017-11-27 02:43:30 -08002181 /* Proxy case */
2182 if (tc->c_lcl_port == 0 && tc->state == TCP_STATE_LISTEN)
2183 return 1;
2184
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002185 u8 is_valid = (tc->c_lcl_port == hdr->dst_port
2186 && (tc->state == TCP_STATE_LISTEN
2187 || tc->c_rmt_port == hdr->src_port));
2188
2189 if (!is_valid)
2190 {
Florin Corascea194d2017-10-02 00:18:51 -07002191 handle = session_lookup_half_open_handle (&tc->connection);
2192 tmp = session_lookup_half_open_connection (handle & 0xFFFFFFFF,
Florin Coras3cbc04b2017-10-02 00:18:51 -07002193 tc->c_proto, tc->c_is_ip4);
Florin Corascea194d2017-10-02 00:18:51 -07002194
2195 if (tmp)
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002196 {
2197 if (tmp->lcl_port == hdr->dst_port
2198 && tmp->rmt_port == hdr->src_port)
2199 {
Florin Corascea194d2017-10-02 00:18:51 -07002200 TCP_DBG ("half-open is valid!");
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002201 }
2202 }
2203 }
2204 return is_valid;
2205}
2206
2207/**
2208 * Lookup transport connection
2209 */
2210static tcp_connection_t *
Florin Corascea194d2017-10-02 00:18:51 -07002211tcp_lookup_connection (u32 fib_index, vlib_buffer_t * b, u8 thread_index,
2212 u8 is_ip4)
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002213{
2214 tcp_header_t *tcp;
2215 transport_connection_t *tconn;
2216 tcp_connection_t *tc;
Florin Corasdff48db2017-11-19 18:06:58 -08002217 u8 is_filtered = 0;
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002218 if (is_ip4)
2219 {
2220 ip4_header_t *ip4;
2221 ip4 = vlib_buffer_get_current (b);
2222 tcp = ip4_next_header (ip4);
Florin Corascea194d2017-10-02 00:18:51 -07002223 tconn = session_lookup_connection_wt4 (fib_index,
2224 &ip4->dst_address,
2225 &ip4->src_address,
2226 tcp->dst_port,
2227 tcp->src_port,
2228 TRANSPORT_PROTO_TCP,
Florin Corasdff48db2017-11-19 18:06:58 -08002229 thread_index, &is_filtered);
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002230 tc = tcp_get_connection_from_transport (tconn);
2231 ASSERT (tcp_lookup_is_valid (tc, tcp));
2232 }
2233 else
2234 {
2235 ip6_header_t *ip6;
2236 ip6 = vlib_buffer_get_current (b);
2237 tcp = ip6_next_header (ip6);
Florin Corascea194d2017-10-02 00:18:51 -07002238 tconn = session_lookup_connection_wt6 (fib_index,
2239 &ip6->dst_address,
2240 &ip6->src_address,
2241 tcp->dst_port,
2242 tcp->src_port,
2243 TRANSPORT_PROTO_TCP,
Florin Corasdff48db2017-11-19 18:06:58 -08002244 thread_index, &is_filtered);
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002245 tc = tcp_get_connection_from_transport (tconn);
2246 ASSERT (tcp_lookup_is_valid (tc, tcp));
2247 }
2248 return tc;
2249}
2250
Dave Barach68b0fb02017-02-28 15:15:56 -05002251always_inline uword
2252tcp46_syn_sent_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
2253 vlib_frame_t * from_frame, int is_ip4)
2254{
2255 tcp_main_t *tm = vnet_get_tcp_main ();
2256 u32 n_left_from, next_index, *from, *to_next;
Damjan Marion586afd72017-04-05 19:18:20 +02002257 u32 my_thread_index = vm->thread_index, errors = 0;
Dave Barach68b0fb02017-02-28 15:15:56 -05002258
2259 from = vlib_frame_vector_args (from_frame);
2260 n_left_from = from_frame->n_vectors;
2261
2262 next_index = node->cached_next_index;
2263
2264 while (n_left_from > 0)
2265 {
2266 u32 n_left_to_next;
2267
2268 vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
2269
2270 while (n_left_from > 0 && n_left_to_next > 0)
2271 {
2272 u32 bi0, ack0, seq0;
2273 vlib_buffer_t *b0;
Clement Durand6cf260c2017-04-13 13:27:04 +02002274 tcp_rx_trace_t *t0;
Dave Barach68b0fb02017-02-28 15:15:56 -05002275 tcp_header_t *tcp0 = 0;
2276 tcp_connection_t *tc0;
Dave Barach68b0fb02017-02-28 15:15:56 -05002277 tcp_connection_t *new_tc0;
Vijayabhaskar Katamreddyce074122017-11-15 13:50:26 -08002278 u32 next0 = tcp_next_drop (is_ip4), error0 = TCP_ERROR_ENQUEUED;
Dave Barach68b0fb02017-02-28 15:15:56 -05002279
2280 bi0 = from[0];
2281 to_next[0] = bi0;
2282 from += 1;
2283 to_next += 1;
2284 n_left_from -= 1;
2285 n_left_to_next -= 1;
2286
2287 b0 = vlib_get_buffer (vm, bi0);
2288 tc0 =
2289 tcp_half_open_connection_get (vnet_buffer (b0)->
2290 tcp.connection_index);
Dave Barachd84ba852017-08-22 17:56:46 -04002291 if (PREDICT_FALSE (tc0 == 0))
2292 {
2293 error0 = TCP_ERROR_INVALID_CONNECTION;
2294 goto drop;
2295 }
Dave Barach68b0fb02017-02-28 15:15:56 -05002296
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002297 /* Half-open completed recently but the connection was't removed
2298 * yet by the owning thread */
2299 if (PREDICT_FALSE (tc0->flags & TCP_CONN_HALF_OPEN_DONE))
2300 {
2301 /* Make sure the connection actually exists */
Florin Corascea194d2017-10-02 00:18:51 -07002302 ASSERT (tcp_lookup_connection (tc0->c_fib_index, b0,
2303 my_thread_index, is_ip4));
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002304 goto drop;
2305 }
2306
Dave Barach68b0fb02017-02-28 15:15:56 -05002307 ack0 = vnet_buffer (b0)->tcp.ack_number;
2308 seq0 = vnet_buffer (b0)->tcp.seq_number;
Florin Coras82b13a82017-04-25 11:58:06 -07002309 tcp0 = tcp_buffer_hdr (b0);
Dave Barach68b0fb02017-02-28 15:15:56 -05002310
Florin Coras9d063042017-09-14 03:08:00 -04002311 /* Crude check to see if the connection handle does not match
2312 * the packet. Probably connection just switched to established */
2313 if (PREDICT_FALSE (tcp0->dst_port != tc0->c_lcl_port
2314 || tcp0->src_port != tc0->c_rmt_port))
2315 goto drop;
2316
Dave Barach68b0fb02017-02-28 15:15:56 -05002317 if (PREDICT_FALSE
2318 (!tcp_ack (tcp0) && !tcp_rst (tcp0) && !tcp_syn (tcp0)))
2319 goto drop;
2320
2321 /* SYNs, FINs and data consume sequence numbers */
2322 vnet_buffer (b0)->tcp.seq_end = seq0 + tcp_is_syn (tcp0)
Florin Coras82b13a82017-04-25 11:58:06 -07002323 + tcp_is_fin (tcp0) + vnet_buffer (b0)->tcp.data_len;
Dave Barach68b0fb02017-02-28 15:15:56 -05002324
2325 /*
2326 * 1. check the ACK bit
2327 */
2328
2329 /*
2330 * If the ACK bit is set
2331 * If SEG.ACK =< ISS, or SEG.ACK > SND.NXT, send a reset (unless
2332 * the RST bit is set, if so drop the segment and return)
2333 * <SEQ=SEG.ACK><CTL=RST>
2334 * and discard the segment. Return.
2335 * If SND.UNA =< SEG.ACK =< SND.NXT then the ACK is acceptable.
2336 */
2337 if (tcp_ack (tcp0))
2338 {
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002339 if (seq_leq (ack0, tc0->iss) || seq_gt (ack0, tc0->snd_nxt))
Dave Barach68b0fb02017-02-28 15:15:56 -05002340 {
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002341 clib_warning ("ack not in rcv wnd");
Dave Barach68b0fb02017-02-28 15:15:56 -05002342 if (!tcp_rst (tcp0))
Florin Coras1f152cd2017-08-18 19:28:03 -07002343 tcp_send_reset_w_pkt (tc0, b0, is_ip4);
Dave Barach68b0fb02017-02-28 15:15:56 -05002344 goto drop;
2345 }
2346
2347 /* Make sure ACK is valid */
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002348 if (seq_gt (tc0->snd_una, ack0))
2349 {
2350 clib_warning ("ack invalid");
2351 goto drop;
2352 }
Dave Barach68b0fb02017-02-28 15:15:56 -05002353 }
2354
2355 /*
2356 * 2. check the RST bit
2357 */
2358
2359 if (tcp_rst (tcp0))
2360 {
2361 /* If ACK is acceptable, signal client that peer is not
2362 * willing to accept connection and drop connection*/
2363 if (tcp_ack (tcp0))
Florin Coras6534b7a2017-07-18 05:38:03 -04002364 tcp_connection_reset (tc0);
Dave Barach68b0fb02017-02-28 15:15:56 -05002365 goto drop;
2366 }
2367
2368 /*
2369 * 3. check the security and precedence (skipped)
2370 */
2371
2372 /*
2373 * 4. check the SYN bit
2374 */
2375
2376 /* No SYN flag. Drop. */
2377 if (!tcp_syn (tcp0))
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002378 {
2379 clib_warning ("not synack");
2380 goto drop;
2381 }
Dave Barach68b0fb02017-02-28 15:15:56 -05002382
Florin Coras6534b7a2017-07-18 05:38:03 -04002383 /* Parse options */
2384 if (tcp_options_parse (tcp0, &tc0->rcv_opts))
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002385 {
2386 clib_warning ("options parse fail");
2387 goto drop;
2388 }
Florin Coras6534b7a2017-07-18 05:38:03 -04002389
Dave Barach68b0fb02017-02-28 15:15:56 -05002390 /* Valid SYN or SYN-ACK. Move connection from half-open pool to
2391 * current thread pool. */
2392 pool_get (tm->connections[my_thread_index], new_tc0);
2393 clib_memcpy (new_tc0, tc0, sizeof (*new_tc0));
Dave Barach259cdae2017-05-15 16:27:05 -04002394 new_tc0->c_c_index = new_tc0 - tm->connections[my_thread_index];
Florin Coras6534b7a2017-07-18 05:38:03 -04002395 new_tc0->c_thread_index = my_thread_index;
Dave Barach68b0fb02017-02-28 15:15:56 -05002396 new_tc0->rcv_nxt = vnet_buffer (b0)->tcp.seq_end;
2397 new_tc0->irs = seq0;
Florin Coras68810622017-07-24 17:40:28 -07002398 new_tc0->timers[TCP_TIMER_ESTABLISH] = TCP_TIMER_HANDLE_INVALID;
2399 new_tc0->timers[TCP_TIMER_RETRANSMIT_SYN] =
2400 TCP_TIMER_HANDLE_INVALID;
Florin Corasf9d05682018-04-26 08:26:52 -07002401 new_tc0->sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_RX];
Florin Coras68810622017-07-24 17:40:28 -07002402
2403 /* If this is not the owning thread, wait for syn retransmit to
2404 * expire and cleanup then */
2405 if (tcp_half_open_connection_cleanup (tc0))
2406 tc0->flags |= TCP_CONN_HALF_OPEN_DONE;
Dave Barach68b0fb02017-02-28 15:15:56 -05002407
Florin Coras93992a92017-05-24 18:03:56 -07002408 if (tcp_opts_tstamp (&new_tc0->rcv_opts))
Dave Barach68b0fb02017-02-28 15:15:56 -05002409 {
Florin Coras93992a92017-05-24 18:03:56 -07002410 new_tc0->tsval_recent = new_tc0->rcv_opts.tsval;
Dave Barach68b0fb02017-02-28 15:15:56 -05002411 new_tc0->tsval_recent_age = tcp_time_now ();
2412 }
2413
Florin Coras93992a92017-05-24 18:03:56 -07002414 if (tcp_opts_wscale (&new_tc0->rcv_opts))
2415 new_tc0->snd_wscale = new_tc0->rcv_opts.wscale;
Dave Barach68b0fb02017-02-28 15:15:56 -05002416
Florin Coras45ca73f2018-09-27 09:19:29 -07002417 new_tc0->snd_wnd = clib_net_to_host_u16 (tcp0->window)
2418 << new_tc0->snd_wscale;
Dave Barach68b0fb02017-02-28 15:15:56 -05002419 new_tc0->snd_wl1 = seq0;
2420 new_tc0->snd_wl2 = ack0;
2421
Florin Corase04c2992017-03-01 08:17:34 -08002422 tcp_connection_init_vars (new_tc0);
2423
Dave Barach68b0fb02017-02-28 15:15:56 -05002424 /* SYN-ACK: See if we can switch to ESTABLISHED state */
Florin Coras6534b7a2017-07-18 05:38:03 -04002425 if (PREDICT_TRUE (tcp_ack (tcp0)))
Dave Barach68b0fb02017-02-28 15:15:56 -05002426 {
2427 /* Our SYN is ACKed: we have iss < ack = snd_una */
2428
2429 /* TODO Dequeue acknowledged segments if we support Fast Open */
2430 new_tc0->snd_una = ack0;
2431 new_tc0->state = TCP_STATE_ESTABLISHED;
2432
Florin Corase04c2992017-03-01 08:17:34 -08002433 /* Make sure las is initialized for the wnd computation */
2434 new_tc0->rcv_las = new_tc0->rcv_nxt;
2435
Florin Corasf03a59a2017-06-09 21:07:32 -07002436 /* Notify app that we have connection. If session layer can't
2437 * allocate session send reset */
Florin Coras3cbc04b2017-10-02 00:18:51 -07002438 if (session_stream_connect_notify (&new_tc0->connection, 0))
Florin Corasf03a59a2017-06-09 21:07:32 -07002439 {
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002440 clib_warning ("connect notify fail");
Florin Coras1f152cd2017-08-18 19:28:03 -07002441 tcp_send_reset_w_pkt (new_tc0, b0, is_ip4);
Florin Corasf03a59a2017-06-09 21:07:32 -07002442 tcp_connection_cleanup (new_tc0);
Florin Corasf03a59a2017-06-09 21:07:32 -07002443 goto drop;
2444 }
Dave Barach68b0fb02017-02-28 15:15:56 -05002445
2446 /* Make sure after data segment processing ACK is sent */
2447 new_tc0->flags |= TCP_CONN_SNDACK;
Dave Barach2c25a622017-06-26 11:35:07 -04002448
2449 /* Update rtt with the syn-ack sample */
Dave Barach2c25a622017-06-26 11:35:07 -04002450 tcp_update_rtt (new_tc0, vnet_buffer (b0)->tcp.ack_number);
Florin Coras6534b7a2017-07-18 05:38:03 -04002451 TCP_EVT_DBG (TCP_EVT_SYNACK_RCVD, new_tc0);
Dave Barach68b0fb02017-02-28 15:15:56 -05002452 }
2453 /* SYN: Simultaneous open. Change state to SYN-RCVD and send SYN-ACK */
2454 else
2455 {
2456 new_tc0->state = TCP_STATE_SYN_RCVD;
2457
Florin Corase69f4952017-03-07 10:06:24 -08002458 /* Notify app that we have connection */
Florin Coras3cbc04b2017-10-02 00:18:51 -07002459 if (session_stream_connect_notify (&new_tc0->connection, 0))
Florin Corasf03a59a2017-06-09 21:07:32 -07002460 {
2461 tcp_connection_cleanup (new_tc0);
Florin Coras1f152cd2017-08-18 19:28:03 -07002462 tcp_send_reset_w_pkt (tc0, b0, is_ip4);
Florin Coras6534b7a2017-07-18 05:38:03 -04002463 TCP_EVT_DBG (TCP_EVT_RST_SENT, tc0);
Florin Corasf03a59a2017-06-09 21:07:32 -07002464 goto drop;
2465 }
2466
Dave Barach2c25a622017-06-26 11:35:07 -04002467 tc0->rtt_ts = 0;
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002468 tcp_init_snd_vars (tc0);
Dave Barach68b0fb02017-02-28 15:15:56 -05002469 tcp_make_synack (new_tc0, b0);
2470 next0 = tcp_next_output (is_ip4);
2471
2472 goto drop;
2473 }
2474
2475 /* Read data, if any */
Florin Coras6534b7a2017-07-18 05:38:03 -04002476 if (PREDICT_FALSE (vnet_buffer (b0)->tcp.data_len))
Dave Barach68b0fb02017-02-28 15:15:56 -05002477 {
Florin Coras00cd22d2018-04-18 13:20:18 -07002478 clib_warning ("rcvd data in syn-sent");
2479 error0 = tcp_segment_rcv (new_tc0, b0, &next0);
2480 if (error0 == TCP_ERROR_ACK_OK)
Dave Barach68b0fb02017-02-28 15:15:56 -05002481 error0 = TCP_ERROR_SYN_ACKS_RCVD;
Florin Coras00cd22d2018-04-18 13:20:18 -07002482 tcp_maybe_inc_counter (syn_sent, error0, 1);
Dave Barach68b0fb02017-02-28 15:15:56 -05002483 }
2484 else
2485 {
2486 tcp_make_ack (new_tc0, b0);
2487 next0 = tcp_next_output (new_tc0->c_is_ip4);
2488 }
2489
2490 drop:
2491
2492 b0->error = error0 ? node->errors[error0] : 0;
Chris Luke879ace32017-09-26 13:15:16 -04002493 if (PREDICT_FALSE
2494 ((b0->flags & VLIB_BUFFER_IS_TRACED) && tcp0 != 0))
Dave Barach68b0fb02017-02-28 15:15:56 -05002495 {
Clement Durand6cf260c2017-04-13 13:27:04 +02002496 t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
2497 clib_memcpy (&t0->tcp_header, tcp0, sizeof (t0->tcp_header));
2498 clib_memcpy (&t0->tcp_connection, tc0,
2499 sizeof (t0->tcp_connection));
Dave Barach68b0fb02017-02-28 15:15:56 -05002500 }
2501
2502 vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
2503 n_left_to_next, bi0, next0);
2504 }
2505
2506 vlib_put_next_frame (vm, node, next_index, n_left_to_next);
2507 }
2508
Florin Coras3cbc04b2017-10-02 00:18:51 -07002509 errors = session_manager_flush_enqueue_events (TRANSPORT_PROTO_TCP,
2510 my_thread_index);
Florin Coras00cd22d2018-04-18 13:20:18 -07002511 tcp_inc_counter (syn_sent, TCP_ERROR_EVENT_FIFO_FULL, errors);
Dave Barach68b0fb02017-02-28 15:15:56 -05002512 return from_frame->n_vectors;
2513}
2514
2515static uword
2516tcp4_syn_sent (vlib_main_t * vm, vlib_node_runtime_t * node,
2517 vlib_frame_t * from_frame)
2518{
2519 return tcp46_syn_sent_inline (vm, node, from_frame, 1 /* is_ip4 */ );
2520}
2521
2522static uword
2523tcp6_syn_sent_rcv (vlib_main_t * vm, vlib_node_runtime_t * node,
2524 vlib_frame_t * from_frame)
2525{
2526 return tcp46_syn_sent_inline (vm, node, from_frame, 0 /* is_ip4 */ );
2527}
2528
2529/* *INDENT-OFF* */
2530VLIB_REGISTER_NODE (tcp4_syn_sent_node) =
2531{
2532 .function = tcp4_syn_sent,
2533 .name = "tcp4-syn-sent",
2534 /* Takes a vector of packets. */
2535 .vector_size = sizeof (u32),
2536 .n_errors = TCP_N_ERROR,
2537 .error_strings = tcp_error_strings,
2538 .n_next_nodes = TCP_SYN_SENT_N_NEXT,
2539 .next_nodes =
2540 {
2541#define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n,
2542 foreach_tcp_state_next
2543#undef _
2544 },
Clement Durand6cf260c2017-04-13 13:27:04 +02002545 .format_trace = format_tcp_rx_trace_short,
Dave Barach68b0fb02017-02-28 15:15:56 -05002546};
2547/* *INDENT-ON* */
2548
2549VLIB_NODE_FUNCTION_MULTIARCH (tcp4_syn_sent_node, tcp4_syn_sent);
2550
2551/* *INDENT-OFF* */
2552VLIB_REGISTER_NODE (tcp6_syn_sent_node) =
2553{
2554 .function = tcp6_syn_sent_rcv,
2555 .name = "tcp6-syn-sent",
2556 /* Takes a vector of packets. */
2557 .vector_size = sizeof (u32),
2558 .n_errors = TCP_N_ERROR,
2559 .error_strings = tcp_error_strings,
2560 .n_next_nodes = TCP_SYN_SENT_N_NEXT,
2561 .next_nodes =
2562 {
2563#define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n,
2564 foreach_tcp_state_next
2565#undef _
Clement Durand6cf260c2017-04-13 13:27:04 +02002566 },
2567 .format_trace = format_tcp_rx_trace_short,
2568};
Dave Barach68b0fb02017-02-28 15:15:56 -05002569/* *INDENT-ON* */
2570
2571VLIB_NODE_FUNCTION_MULTIARCH (tcp6_syn_sent_node, tcp6_syn_sent_rcv);
Florin Coras6534b7a2017-07-18 05:38:03 -04002572
Florin Coras3cbc04b2017-10-02 00:18:51 -07002573vlib_node_registration_t tcp4_rcv_process_node;
2574vlib_node_registration_t tcp6_rcv_process_node;
2575
Dave Barach68b0fb02017-02-28 15:15:56 -05002576/**
Florin Corasd79b41e2017-03-04 05:37:52 -08002577 * Handles reception for all states except LISTEN, SYN-SENT and ESTABLISHED
Dave Barach68b0fb02017-02-28 15:15:56 -05002578 * as per RFC793 p. 64
2579 */
2580always_inline uword
2581tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
2582 vlib_frame_t * from_frame, int is_ip4)
2583{
Florin Coras00cd22d2018-04-18 13:20:18 -07002584 u32 n_left_from, next_index, *from, *to_next, n_fins = 0;
Damjan Marion586afd72017-04-05 19:18:20 +02002585 u32 my_thread_index = vm->thread_index, errors = 0;
Dave Barach68b0fb02017-02-28 15:15:56 -05002586
2587 from = vlib_frame_vector_args (from_frame);
2588 n_left_from = from_frame->n_vectors;
Dave Barach68b0fb02017-02-28 15:15:56 -05002589 next_index = node->cached_next_index;
2590
2591 while (n_left_from > 0)
2592 {
2593 u32 n_left_to_next;
2594
2595 vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
2596
2597 while (n_left_from > 0 && n_left_to_next > 0)
2598 {
2599 u32 bi0;
2600 vlib_buffer_t *b0;
2601 tcp_header_t *tcp0 = 0;
2602 tcp_connection_t *tc0;
Florin Coras00cd22d2018-04-18 13:20:18 -07002603 u32 next0 = tcp_next_drop (is_ip4), error0 = TCP_ERROR_NONE;
Florin Coras9d063042017-09-14 03:08:00 -04002604 u8 is_fin0;
Dave Barach68b0fb02017-02-28 15:15:56 -05002605
2606 bi0 = from[0];
2607 to_next[0] = bi0;
2608 from += 1;
2609 to_next += 1;
2610 n_left_from -= 1;
2611 n_left_to_next -= 1;
2612
2613 b0 = vlib_get_buffer (vm, bi0);
2614 tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index,
2615 my_thread_index);
Florin Corasd79b41e2017-03-04 05:37:52 -08002616 if (PREDICT_FALSE (tc0 == 0))
2617 {
2618 error0 = TCP_ERROR_INVALID_CONNECTION;
2619 goto drop;
2620 }
Dave Barach68b0fb02017-02-28 15:15:56 -05002621
Florin Coras82b13a82017-04-25 11:58:06 -07002622 tcp0 = tcp_buffer_hdr (b0);
Florin Coras9d063042017-09-14 03:08:00 -04002623 is_fin0 = tcp_is_fin (tcp0);
Dave Barach68b0fb02017-02-28 15:15:56 -05002624
2625 /* SYNs, FINs and data consume sequence numbers */
2626 vnet_buffer (b0)->tcp.seq_end = vnet_buffer (b0)->tcp.seq_number
Florin Coras9d063042017-09-14 03:08:00 -04002627 + tcp_is_syn (tcp0) + is_fin0 + vnet_buffer (b0)->tcp.data_len;
Dave Barach68b0fb02017-02-28 15:15:56 -05002628
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002629 if (CLIB_DEBUG)
2630 {
2631 tcp_connection_t *tmp;
Florin Coras00cd22d2018-04-18 13:20:18 -07002632 tmp = tcp_lookup_connection (tc0->c_fib_index, b0,
2633 my_thread_index, is_ip4);
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002634 if (tmp->state != tc0->state)
2635 {
2636 clib_warning ("state changed");
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002637 goto drop;
2638 }
2639 }
2640
Dave Barach68b0fb02017-02-28 15:15:56 -05002641 /*
2642 * Special treatment for CLOSED
2643 */
Florin Coras00cd22d2018-04-18 13:20:18 -07002644 if (PREDICT_FALSE (tc0->state == TCP_STATE_CLOSED))
Dave Barach68b0fb02017-02-28 15:15:56 -05002645 {
Florin Coras00cd22d2018-04-18 13:20:18 -07002646 error0 = TCP_ERROR_CONNECTION_CLOSED;
Dave Barach68b0fb02017-02-28 15:15:56 -05002647 goto drop;
Dave Barach68b0fb02017-02-28 15:15:56 -05002648 }
2649
2650 /*
2651 * For all other states (except LISTEN)
2652 */
2653
2654 /* 1-4: check SEQ, RST, SYN */
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002655 if (PREDICT_FALSE (tcp_segment_validate (vm, tc0, b0, tcp0,
Florin Coras00cd22d2018-04-18 13:20:18 -07002656 &next0, &error0)))
Dave Barach68b0fb02017-02-28 15:15:56 -05002657 {
Florin Coras00cd22d2018-04-18 13:20:18 -07002658 tcp_maybe_inc_counter (rcv_process, error0, 1);
Dave Barach68b0fb02017-02-28 15:15:56 -05002659 goto drop;
2660 }
2661
2662 /* 5: check the ACK field */
2663 switch (tc0->state)
2664 {
2665 case TCP_STATE_SYN_RCVD:
2666 /*
2667 * If the segment acknowledgment is not acceptable, form a
2668 * reset segment,
2669 * <SEQ=SEG.ACK><CTL=RST>
2670 * and send it.
2671 */
2672 if (!tcp_rcv_ack_is_acceptable (tc0, b0))
2673 {
Florin Corasa096f2d2017-09-28 23:49:42 -04002674 TCP_DBG ("connection not accepted");
Florin Coras1f152cd2017-08-18 19:28:03 -07002675 tcp_send_reset_w_pkt (tc0, b0, is_ip4);
Florin Coras00cd22d2018-04-18 13:20:18 -07002676 error0 = TCP_ERROR_ACK_INVALID;
Dave Barach68b0fb02017-02-28 15:15:56 -05002677 goto drop;
2678 }
Florin Coras3af90fc2017-05-03 21:09:42 -07002679
2680 /* Update rtt and rto */
Florin Coras3af90fc2017-05-03 21:09:42 -07002681 tcp_update_rtt (tc0, vnet_buffer (b0)->tcp.ack_number);
2682
Dave Barach68b0fb02017-02-28 15:15:56 -05002683 /* Switch state to ESTABLISHED */
2684 tc0->state = TCP_STATE_ESTABLISHED;
Florin Corasb384b542018-01-15 01:08:33 -08002685 TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
Dave Barach68b0fb02017-02-28 15:15:56 -05002686
2687 /* Initialize session variables */
2688 tc0->snd_una = vnet_buffer (b0)->tcp.ack_number;
Florin Corase04c2992017-03-01 08:17:34 -08002689 tc0->snd_wnd = clib_net_to_host_u16 (tcp0->window)
Florin Coras93992a92017-05-24 18:03:56 -07002690 << tc0->rcv_opts.wscale;
Dave Barach68b0fb02017-02-28 15:15:56 -05002691 tc0->snd_wl1 = vnet_buffer (b0)->tcp.seq_number;
2692 tc0->snd_wl2 = vnet_buffer (b0)->tcp.ack_number;
Dave Barach68b0fb02017-02-28 15:15:56 -05002693
Florin Corasab0289a2017-08-14 11:25:25 -07002694 /* Reset SYN-ACK retransmit and SYN_RCV establish timers */
Florin Coras6792ec02017-03-13 03:49:51 -07002695 tcp_retransmit_timer_reset (tc0);
Florin Corasab0289a2017-08-14 11:25:25 -07002696 tcp_timer_reset (tc0, TCP_TIMER_ESTABLISH);
Florin Corasb384b542018-01-15 01:08:33 -08002697 stream_session_accept_notify (&tc0->connection);
Florin Coras00cd22d2018-04-18 13:20:18 -07002698 error0 = TCP_ERROR_ACK_OK;
Dave Barach68b0fb02017-02-28 15:15:56 -05002699 break;
2700 case TCP_STATE_ESTABLISHED:
2701 /* We can get packets in established state here because they
2702 * were enqueued before state change */
2703 if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
Florin Coras00cd22d2018-04-18 13:20:18 -07002704 {
2705 tcp_maybe_inc_counter (rcv_process, error0, 1);
2706 goto drop;
2707 }
Dave Barach68b0fb02017-02-28 15:15:56 -05002708
2709 break;
2710 case TCP_STATE_FIN_WAIT_1:
2711 /* In addition to the processing for the ESTABLISHED state, if
2712 * our FIN is now acknowledged then enter FIN-WAIT-2 and
2713 * continue processing in that state. */
2714 if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
Florin Coras00cd22d2018-04-18 13:20:18 -07002715 {
2716 tcp_maybe_inc_counter (rcv_process, error0, 1);
2717 goto drop;
2718 }
Florin Corasd79b41e2017-03-04 05:37:52 -08002719
Florin Corasb2215d62017-08-01 16:56:58 -07002720 /* Still have to send the FIN */
2721 if (tc0->flags & TCP_CONN_FINPNDG)
2722 {
2723 /* TX fifo finally drained */
Florin Coras25579b42018-06-06 17:55:02 -07002724 if (!session_tx_fifo_max_dequeue (&tc0->connection))
Florin Corasb2215d62017-08-01 16:56:58 -07002725 tcp_send_fin (tc0);
2726 }
Florin Corasd79b41e2017-03-04 05:37:52 -08002727 /* If FIN is ACKed */
Florin Corasb2215d62017-08-01 16:56:58 -07002728 else if (tc0->snd_una == tc0->snd_una_max)
Florin Corasd79b41e2017-03-04 05:37:52 -08002729 {
2730 tc0->state = TCP_STATE_FIN_WAIT_2;
Florin Coras6534b7a2017-07-18 05:38:03 -04002731 TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
2732
Florin Coras9d063042017-09-14 03:08:00 -04002733 /* Stop all retransmit timers because we have nothing more
2734 * to send. Enable waitclose though because we're willing to
2735 * wait for peer's FIN but not indefinitely. */
2736 tcp_connection_timers_reset (tc0);
2737 tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
Florin Corasd79b41e2017-03-04 05:37:52 -08002738 }
Dave Barach68b0fb02017-02-28 15:15:56 -05002739 break;
2740 case TCP_STATE_FIN_WAIT_2:
2741 /* In addition to the processing for the ESTABLISHED state, if
2742 * the retransmission queue is empty, the user's CLOSE can be
2743 * acknowledged ("ok") but do not delete the TCB. */
2744 if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
Florin Coras00cd22d2018-04-18 13:20:18 -07002745 {
2746 tcp_maybe_inc_counter (rcv_process, error0, 1);
2747 goto drop;
2748 }
Dave Barach68b0fb02017-02-28 15:15:56 -05002749 break;
2750 case TCP_STATE_CLOSE_WAIT:
2751 /* Do the same processing as for the ESTABLISHED state. */
2752 if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
Florin Coras00cd22d2018-04-18 13:20:18 -07002753 {
2754 tcp_maybe_inc_counter (rcv_process, error0, 1);
2755 goto drop;
2756 }
Florin Coras25579b42018-06-06 17:55:02 -07002757 if (tc0->flags & TCP_CONN_FINPNDG)
2758 {
2759 /* TX fifo finally drained */
2760 if (!session_tx_fifo_max_dequeue (&tc0->connection))
2761 {
2762 tcp_send_fin (tc0);
2763 tcp_connection_timers_reset (tc0);
2764 tc0->state = TCP_STATE_LAST_ACK;
2765 tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE,
2766 TCP_2MSL_TIME);
2767 }
2768 }
Dave Barach68b0fb02017-02-28 15:15:56 -05002769 break;
2770 case TCP_STATE_CLOSING:
2771 /* In addition to the processing for the ESTABLISHED state, if
2772 * the ACK acknowledges our FIN then enter the TIME-WAIT state,
2773 * otherwise ignore the segment. */
2774 if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
Florin Coras00cd22d2018-04-18 13:20:18 -07002775 {
2776 tcp_maybe_inc_counter (rcv_process, error0, 1);
2777 goto drop;
2778 }
Dave Barach68b0fb02017-02-28 15:15:56 -05002779
Dave Barach68b0fb02017-02-28 15:15:56 -05002780 tc0->state = TCP_STATE_TIME_WAIT;
Florin Coras6534b7a2017-07-18 05:38:03 -04002781 TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
Florin Corasa096f2d2017-09-28 23:49:42 -04002782 tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME);
Dave Barach68b0fb02017-02-28 15:15:56 -05002783 goto drop;
2784
2785 break;
2786 case TCP_STATE_LAST_ACK:
Florin Coras93992a92017-05-24 18:03:56 -07002787 /* The only thing that [should] arrive in this state is an
Dave Barach68b0fb02017-02-28 15:15:56 -05002788 * acknowledgment of our FIN. If our FIN is now acknowledged,
2789 * delete the TCB, enter the CLOSED state, and return. */
2790
2791 if (!tcp_rcv_ack_is_acceptable (tc0, b0))
Florin Corasa096f2d2017-09-28 23:49:42 -04002792 {
2793 error0 = TCP_ERROR_ACK_INVALID;
2794 goto drop;
2795 }
Florin Coras00cd22d2018-04-18 13:20:18 -07002796 error0 = TCP_ERROR_ACK_OK;
Florin Coras9d063042017-09-14 03:08:00 -04002797 tc0->snd_una = vnet_buffer (b0)->tcp.ack_number;
Florin Corasa096f2d2017-09-28 23:49:42 -04002798 /* Apparently our ACK for the peer's FIN was lost */
2799 if (is_fin0 && tc0->snd_una != tc0->snd_una_max)
Florin Coras93992a92017-05-24 18:03:56 -07002800 {
Florin Coras93992a92017-05-24 18:03:56 -07002801 tcp_send_fin (tc0);
2802 goto drop;
2803 }
2804
Florin Corasd79b41e2017-03-04 05:37:52 -08002805 tc0->state = TCP_STATE_CLOSED;
Florin Coras6534b7a2017-07-18 05:38:03 -04002806 TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
Florin Corasc01d5782018-10-17 14:53:11 -07002807
2808 /* Don't free the connection from the data path since
2809 * we can't ensure that we have no packets already enqueued
2810 * to output. Rely instead on the waitclose timer */
2811 tcp_connection_timers_reset (tc0);
2812 tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, 1);
Florin Corasd79b41e2017-03-04 05:37:52 -08002813
Dave Barach68b0fb02017-02-28 15:15:56 -05002814 goto drop;
2815
2816 break;
2817 case TCP_STATE_TIME_WAIT:
2818 /* The only thing that can arrive in this state is a
2819 * retransmission of the remote FIN. Acknowledge it, and restart
2820 * the 2 MSL timeout. */
2821
Florin Coras93992a92017-05-24 18:03:56 -07002822 if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0))
Florin Coras00cd22d2018-04-18 13:20:18 -07002823 {
2824 tcp_maybe_inc_counter (rcv_process, error0, 1);
2825 goto drop;
2826 }
Florin Coras93992a92017-05-24 18:03:56 -07002827
2828 tcp_make_ack (tc0, b0);
Florin Corasa096f2d2017-09-28 23:49:42 -04002829 next0 = tcp_next_output (is_ip4);
2830 tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME);
Dave Barach68b0fb02017-02-28 15:15:56 -05002831 goto drop;
Florin Coras93992a92017-05-24 18:03:56 -07002832
Dave Barach68b0fb02017-02-28 15:15:56 -05002833 break;
2834 default:
2835 ASSERT (0);
2836 }
2837
2838 /* 6: check the URG bit TODO */
2839
2840 /* 7: process the segment text */
2841 switch (tc0->state)
2842 {
2843 case TCP_STATE_ESTABLISHED:
2844 case TCP_STATE_FIN_WAIT_1:
2845 case TCP_STATE_FIN_WAIT_2:
Florin Coras6534b7a2017-07-18 05:38:03 -04002846 if (vnet_buffer (b0)->tcp.data_len)
Florin Coras00cd22d2018-04-18 13:20:18 -07002847 {
2848 error0 = tcp_segment_rcv (tc0, b0, &next0);
2849 tcp_maybe_inc_counter (rcv_process, error0, 1);
2850 }
Florin Coras9d063042017-09-14 03:08:00 -04002851 else if (is_fin0)
2852 tc0->rcv_nxt += 1;
Dave Barach68b0fb02017-02-28 15:15:56 -05002853 break;
2854 case TCP_STATE_CLOSE_WAIT:
2855 case TCP_STATE_CLOSING:
2856 case TCP_STATE_LAST_ACK:
2857 case TCP_STATE_TIME_WAIT:
2858 /* This should not occur, since a FIN has been received from the
2859 * remote side. Ignore the segment text. */
2860 break;
2861 }
2862
2863 /* 8: check the FIN bit */
Florin Coras9d063042017-09-14 03:08:00 -04002864 if (!is_fin0)
Dave Barach68b0fb02017-02-28 15:15:56 -05002865 goto drop;
2866
2867 switch (tc0->state)
2868 {
2869 case TCP_STATE_ESTABLISHED:
2870 case TCP_STATE_SYN_RCVD:
2871 /* Send FIN-ACK notify app and enter CLOSE-WAIT */
2872 tcp_connection_timers_reset (tc0);
Florin Corasd79b41e2017-03-04 05:37:52 -08002873 tcp_make_fin (tc0, b0);
Florin Coras4eeeaaf2017-09-05 14:03:37 -04002874 tc0->snd_nxt += 1;
Florin Coras8b20bf52018-06-14 14:55:50 -07002875 tc0->snd_una_max = tc0->snd_nxt;
Florin Corasc01c4452018-09-20 18:36:54 -07002876 tcp_retransmit_timer_set (tc0);
Dave Barach68b0fb02017-02-28 15:15:56 -05002877 next0 = tcp_next_output (tc0->c_is_ip4);
2878 stream_session_disconnect_notify (&tc0->connection);
2879 tc0->state = TCP_STATE_CLOSE_WAIT;
Florin Coras6534b7a2017-07-18 05:38:03 -04002880 TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
Dave Barach68b0fb02017-02-28 15:15:56 -05002881 break;
2882 case TCP_STATE_CLOSE_WAIT:
2883 case TCP_STATE_CLOSING:
2884 case TCP_STATE_LAST_ACK:
2885 /* move along .. */
2886 break;
2887 case TCP_STATE_FIN_WAIT_1:
Florin Corasab0289a2017-08-14 11:25:25 -07002888 tc0->state = TCP_STATE_CLOSING;
2889 tcp_make_ack (tc0, b0);
2890 next0 = tcp_next_output (is_ip4);
Florin Coras6534b7a2017-07-18 05:38:03 -04002891 TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
Florin Corasab0289a2017-08-14 11:25:25 -07002892 /* Wait for ACK but not forever */
2893 tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME);
Dave Barach68b0fb02017-02-28 15:15:56 -05002894 break;
2895 case TCP_STATE_FIN_WAIT_2:
Florin Coras9d063042017-09-14 03:08:00 -04002896 /* Got FIN, send ACK! Be more aggressive with resource cleanup */
Dave Barach68b0fb02017-02-28 15:15:56 -05002897 tc0->state = TCP_STATE_TIME_WAIT;
Florin Corasdb84e572017-05-09 18:54:52 -07002898 tcp_connection_timers_reset (tc0);
Florin Coras9d063042017-09-14 03:08:00 -04002899 tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME);
Dave Barach68b0fb02017-02-28 15:15:56 -05002900 tcp_make_ack (tc0, b0);
2901 next0 = tcp_next_output (is_ip4);
Florin Coras6534b7a2017-07-18 05:38:03 -04002902 TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0);
Dave Barach68b0fb02017-02-28 15:15:56 -05002903 break;
2904 case TCP_STATE_TIME_WAIT:
Florin Coras9d063042017-09-14 03:08:00 -04002905 /* Remain in the TIME-WAIT state. Restart the time-wait
Dave Barach68b0fb02017-02-28 15:15:56 -05002906 * timeout.
2907 */
Florin Coras9d063042017-09-14 03:08:00 -04002908 tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME);
Dave Barach68b0fb02017-02-28 15:15:56 -05002909 break;
2910 }
Florin Corase69f4952017-03-07 10:06:24 -08002911 TCP_EVT_DBG (TCP_EVT_FIN_RCVD, tc0);
Florin Coras00cd22d2018-04-18 13:20:18 -07002912 n_fins += 1;
Dave Barach68b0fb02017-02-28 15:15:56 -05002913
Florin Coras82b13a82017-04-25 11:58:06 -07002914 drop:
Dave Barach68b0fb02017-02-28 15:15:56 -05002915 b0->error = error0 ? node->errors[error0] : 0;
2916
Dave Barach68b0fb02017-02-28 15:15:56 -05002917 if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
2918 {
Florin Coras82b13a82017-04-25 11:58:06 -07002919 tcp_rx_trace_t *t0 =
2920 vlib_add_trace (vm, node, b0, sizeof (*t0));
2921 tcp_set_rx_trace_data (t0, tc0, tcp0, b0, is_ip4);
Dave Barach68b0fb02017-02-28 15:15:56 -05002922 }
2923
2924 vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
2925 n_left_to_next, bi0, next0);
2926 }
2927
2928 vlib_put_next_frame (vm, node, next_index, n_left_to_next);
2929 }
2930
Florin Coras3cbc04b2017-10-02 00:18:51 -07002931 errors = session_manager_flush_enqueue_events (TRANSPORT_PROTO_TCP,
2932 my_thread_index);
Florin Coras00cd22d2018-04-18 13:20:18 -07002933 tcp_inc_counter (rcv_process, TCP_ERROR_EVENT_FIFO_FULL, errors);
2934 tcp_inc_counter (rcv_process, TCP_ERROR_FIN_RCVD, n_fins);
Dave Barach68b0fb02017-02-28 15:15:56 -05002935 return from_frame->n_vectors;
2936}
2937
2938static uword
2939tcp4_rcv_process (vlib_main_t * vm, vlib_node_runtime_t * node,
2940 vlib_frame_t * from_frame)
2941{
2942 return tcp46_rcv_process_inline (vm, node, from_frame, 1 /* is_ip4 */ );
2943}
2944
2945static uword
2946tcp6_rcv_process (vlib_main_t * vm, vlib_node_runtime_t * node,
2947 vlib_frame_t * from_frame)
2948{
2949 return tcp46_rcv_process_inline (vm, node, from_frame, 0 /* is_ip4 */ );
2950}
2951
2952/* *INDENT-OFF* */
2953VLIB_REGISTER_NODE (tcp4_rcv_process_node) =
2954{
2955 .function = tcp4_rcv_process,
2956 .name = "tcp4-rcv-process",
2957 /* Takes a vector of packets. */
2958 .vector_size = sizeof (u32),
2959 .n_errors = TCP_N_ERROR,
2960 .error_strings = tcp_error_strings,
2961 .n_next_nodes = TCP_RCV_PROCESS_N_NEXT,
2962 .next_nodes =
2963 {
2964#define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n,
2965 foreach_tcp_state_next
2966#undef _
2967 },
Clement Durand6cf260c2017-04-13 13:27:04 +02002968 .format_trace = format_tcp_rx_trace_short,
Dave Barach68b0fb02017-02-28 15:15:56 -05002969};
2970/* *INDENT-ON* */
2971
2972VLIB_NODE_FUNCTION_MULTIARCH (tcp4_rcv_process_node, tcp4_rcv_process);
2973
2974/* *INDENT-OFF* */
2975VLIB_REGISTER_NODE (tcp6_rcv_process_node) =
2976{
2977 .function = tcp6_rcv_process,
2978 .name = "tcp6-rcv-process",
2979 /* Takes a vector of packets. */
2980 .vector_size = sizeof (u32),
2981 .n_errors = TCP_N_ERROR,
2982 .error_strings = tcp_error_strings,
2983 .n_next_nodes = TCP_RCV_PROCESS_N_NEXT,
2984 .next_nodes =
2985 {
2986#define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n,
2987 foreach_tcp_state_next
2988#undef _
2989 },
Clement Durand6cf260c2017-04-13 13:27:04 +02002990 .format_trace = format_tcp_rx_trace_short,
Dave Barach68b0fb02017-02-28 15:15:56 -05002991};
2992/* *INDENT-ON* */
2993
2994VLIB_NODE_FUNCTION_MULTIARCH (tcp6_rcv_process_node, tcp6_rcv_process);
2995
2996vlib_node_registration_t tcp4_listen_node;
2997vlib_node_registration_t tcp6_listen_node;
2998
2999/**
3000 * LISTEN state processing as per RFC 793 p. 65
3001 */
3002always_inline uword
3003tcp46_listen_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
3004 vlib_frame_t * from_frame, int is_ip4)
3005{
Florin Coras00cd22d2018-04-18 13:20:18 -07003006 u32 n_left_from, next_index, *from, *to_next, n_syns = 0;
Damjan Marion586afd72017-04-05 19:18:20 +02003007 u32 my_thread_index = vm->thread_index;
Dave Barach68b0fb02017-02-28 15:15:56 -05003008
3009 from = vlib_frame_vector_args (from_frame);
3010 n_left_from = from_frame->n_vectors;
3011
3012 next_index = node->cached_next_index;
3013
3014 while (n_left_from > 0)
3015 {
3016 u32 n_left_to_next;
3017
3018 vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
3019
3020 while (n_left_from > 0 && n_left_to_next > 0)
3021 {
3022 u32 bi0;
3023 vlib_buffer_t *b0;
Clement Durand6cf260c2017-04-13 13:27:04 +02003024 tcp_rx_trace_t *t0;
Dave Barach68b0fb02017-02-28 15:15:56 -05003025 tcp_header_t *th0 = 0;
3026 tcp_connection_t *lc0;
3027 ip4_header_t *ip40;
3028 ip6_header_t *ip60;
3029 tcp_connection_t *child0;
Florin Coras00cd22d2018-04-18 13:20:18 -07003030 u32 error0 = TCP_ERROR_NONE, next0 = tcp_next_drop (is_ip4);
Dave Barach68b0fb02017-02-28 15:15:56 -05003031
3032 bi0 = from[0];
3033 to_next[0] = bi0;
3034 from += 1;
3035 to_next += 1;
3036 n_left_from -= 1;
3037 n_left_to_next -= 1;
3038
3039 b0 = vlib_get_buffer (vm, bi0);
3040 lc0 = tcp_listener_get (vnet_buffer (b0)->tcp.connection_index);
3041
3042 if (is_ip4)
3043 {
3044 ip40 = vlib_buffer_get_current (b0);
3045 th0 = ip4_next_header (ip40);
3046 }
3047 else
3048 {
3049 ip60 = vlib_buffer_get_current (b0);
3050 th0 = ip6_next_header (ip60);
3051 }
3052
3053 /* Create child session. For syn-flood protection use filter */
3054
Florin Corasdc629cd2017-05-09 00:52:37 -07003055 /* 1. first check for an RST: handled in dispatch */
3056 /* if (tcp_rst (th0))
3057 goto drop; */
Dave Barach68b0fb02017-02-28 15:15:56 -05003058
Florin Corasdc629cd2017-05-09 00:52:37 -07003059 /* 2. second check for an ACK: handled in dispatch */
3060 /* if (tcp_ack (th0))
3061 {
3062 tcp_send_reset (b0, is_ip4);
3063 goto drop;
3064 } */
Dave Barach68b0fb02017-02-28 15:15:56 -05003065
3066 /* 3. check for a SYN (did that already) */
3067
Florin Coras4eeeaaf2017-09-05 14:03:37 -04003068 /* Make sure connection wasn't just created */
Florin Corasc1a448b2018-04-20 10:51:49 -07003069 child0 = tcp_lookup_connection (lc0->c_fib_index, b0,
3070 my_thread_index, is_ip4);
Florin Coras4eeeaaf2017-09-05 14:03:37 -04003071 if (PREDICT_FALSE (child0->state != TCP_STATE_LISTEN))
3072 {
3073 error0 = TCP_ERROR_CREATE_EXISTS;
3074 goto drop;
3075 }
3076
Dave Barach68b0fb02017-02-28 15:15:56 -05003077 /* Create child session and send SYN-ACK */
Florin Coras6534b7a2017-07-18 05:38:03 -04003078 child0 = tcp_connection_new (my_thread_index);
Florin Coras1c710452017-10-17 00:03:13 -07003079 child0->c_lcl_port = th0->dst_port;
Dave Barach68b0fb02017-02-28 15:15:56 -05003080 child0->c_rmt_port = th0->src_port;
3081 child0->c_is_ip4 = is_ip4;
Florin Corasbb292f42017-05-19 09:49:19 -07003082 child0->state = TCP_STATE_SYN_RCVD;
Florin Coras56b39f62018-03-27 17:29:32 -07003083 child0->c_fib_index = lc0->c_fib_index;
Dave Barach68b0fb02017-02-28 15:15:56 -05003084
3085 if (is_ip4)
3086 {
3087 child0->c_lcl_ip4.as_u32 = ip40->dst_address.as_u32;
3088 child0->c_rmt_ip4.as_u32 = ip40->src_address.as_u32;
3089 }
3090 else
3091 {
3092 clib_memcpy (&child0->c_lcl_ip6, &ip60->dst_address,
3093 sizeof (ip6_address_t));
3094 clib_memcpy (&child0->c_rmt_ip6, &ip60->src_address,
3095 sizeof (ip6_address_t));
3096 }
3097
Florin Coras93992a92017-05-24 18:03:56 -07003098 if (tcp_options_parse (th0, &child0->rcv_opts))
Florin Corasdb84e572017-05-09 18:54:52 -07003099 {
Florin Coras4eeeaaf2017-09-05 14:03:37 -04003100 clib_warning ("options parse fail");
Florin Corasdb84e572017-05-09 18:54:52 -07003101 goto drop;
3102 }
Dave Barach68b0fb02017-02-28 15:15:56 -05003103
3104 child0->irs = vnet_buffer (b0)->tcp.seq_number;
3105 child0->rcv_nxt = vnet_buffer (b0)->tcp.seq_number + 1;
Florin Coras3e350af2017-03-30 02:54:28 -07003106 child0->rcv_las = child0->rcv_nxt;
Florin Corasf9d05682018-04-26 08:26:52 -07003107 child0->sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_RX];
Dave Barach68b0fb02017-02-28 15:15:56 -05003108
3109 /* RFC1323: TSval timestamps sent on {SYN} and {SYN,ACK}
3110 * segments are used to initialize PAWS. */
Florin Coras93992a92017-05-24 18:03:56 -07003111 if (tcp_opts_tstamp (&child0->rcv_opts))
Dave Barach68b0fb02017-02-28 15:15:56 -05003112 {
Florin Coras93992a92017-05-24 18:03:56 -07003113 child0->tsval_recent = child0->rcv_opts.tsval;
Dave Barach68b0fb02017-02-28 15:15:56 -05003114 child0->tsval_recent_age = tcp_time_now ();
3115 }
3116
Florin Coras93992a92017-05-24 18:03:56 -07003117 if (tcp_opts_wscale (&child0->rcv_opts))
3118 child0->snd_wscale = child0->rcv_opts.wscale;
Florin Corase04c2992017-03-01 08:17:34 -08003119
Florin Corasf03a59a2017-06-09 21:07:32 -07003120 child0->snd_wnd = clib_net_to_host_u16 (th0->window)
3121 << child0->snd_wscale;
Florin Corase04c2992017-03-01 08:17:34 -08003122 child0->snd_wl1 = vnet_buffer (b0)->tcp.seq_number;
3123 child0->snd_wl2 = vnet_buffer (b0)->tcp.ack_number;
3124
3125 tcp_connection_init_vars (child0);
Florin Coras4eeeaaf2017-09-05 14:03:37 -04003126 TCP_EVT_DBG (TCP_EVT_SYN_RCVD, child0, 1);
Florin Corase69f4952017-03-07 10:06:24 -08003127
Florin Coras6dfb5ac2017-11-09 16:26:03 -08003128 if (stream_session_accept (&child0->connection, lc0->c_s_index,
3129 0 /* notify */ ))
3130 {
3131 clib_warning ("session accept fail");
3132 tcp_connection_cleanup (child0);
3133 error0 = TCP_ERROR_CREATE_SESSION_FAIL;
3134 goto drop;
3135 }
3136
Dave Barach68b0fb02017-02-28 15:15:56 -05003137 /* Reuse buffer to make syn-ack and send */
3138 tcp_make_synack (child0, b0);
3139 next0 = tcp_next_output (is_ip4);
Florin Corasab0289a2017-08-14 11:25:25 -07003140 tcp_timer_set (child0, TCP_TIMER_ESTABLISH, TCP_SYN_RCVD_TIME);
Dave Barach68b0fb02017-02-28 15:15:56 -05003141
3142 drop:
3143 if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
3144 {
Clement Durand6cf260c2017-04-13 13:27:04 +02003145 t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
3146 clib_memcpy (&t0->tcp_header, th0, sizeof (t0->tcp_header));
3147 clib_memcpy (&t0->tcp_connection, lc0,
3148 sizeof (t0->tcp_connection));
Dave Barach68b0fb02017-02-28 15:15:56 -05003149 }
3150
Florin Coras00cd22d2018-04-18 13:20:18 -07003151 n_syns += (error0 == TCP_ERROR_NONE);
Florin Corase04c2992017-03-01 08:17:34 -08003152 b0->error = node->errors[error0];
Dave Barach68b0fb02017-02-28 15:15:56 -05003153
3154 vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
3155 n_left_to_next, bi0, next0);
3156 }
3157
3158 vlib_put_next_frame (vm, node, next_index, n_left_to_next);
3159 }
Florin Coras00cd22d2018-04-18 13:20:18 -07003160
3161 tcp_inc_counter (listen, TCP_ERROR_SYNS_RCVD, n_syns);
Dave Barach68b0fb02017-02-28 15:15:56 -05003162 return from_frame->n_vectors;
3163}
3164
3165static uword
3166tcp4_listen (vlib_main_t * vm, vlib_node_runtime_t * node,
3167 vlib_frame_t * from_frame)
3168{
3169 return tcp46_listen_inline (vm, node, from_frame, 1 /* is_ip4 */ );
3170}
3171
3172static uword
3173tcp6_listen (vlib_main_t * vm, vlib_node_runtime_t * node,
3174 vlib_frame_t * from_frame)
3175{
3176 return tcp46_listen_inline (vm, node, from_frame, 0 /* is_ip4 */ );
3177}
3178
3179/* *INDENT-OFF* */
3180VLIB_REGISTER_NODE (tcp4_listen_node) =
3181{
3182 .function = tcp4_listen,
3183 .name = "tcp4-listen",
3184 /* Takes a vector of packets. */
3185 .vector_size = sizeof (u32),
3186 .n_errors = TCP_N_ERROR,
3187 .error_strings = tcp_error_strings,
3188 .n_next_nodes = TCP_LISTEN_N_NEXT,
3189 .next_nodes =
3190 {
3191#define _(s,n) [TCP_LISTEN_NEXT_##s] = n,
3192 foreach_tcp_state_next
3193#undef _
3194 },
Clement Durand6cf260c2017-04-13 13:27:04 +02003195 .format_trace = format_tcp_rx_trace_short,
Dave Barach68b0fb02017-02-28 15:15:56 -05003196};
3197/* *INDENT-ON* */
3198
3199VLIB_NODE_FUNCTION_MULTIARCH (tcp4_listen_node, tcp4_listen);
3200
3201/* *INDENT-OFF* */
3202VLIB_REGISTER_NODE (tcp6_listen_node) =
3203{
3204 .function = tcp6_listen,
3205 .name = "tcp6-listen",
3206 /* Takes a vector of packets. */
3207 .vector_size = sizeof (u32),
3208 .n_errors = TCP_N_ERROR,
3209 .error_strings = tcp_error_strings,
3210 .n_next_nodes = TCP_LISTEN_N_NEXT,
3211 .next_nodes =
3212 {
3213#define _(s,n) [TCP_LISTEN_NEXT_##s] = n,
3214 foreach_tcp_state_next
3215#undef _
3216 },
Clement Durand6cf260c2017-04-13 13:27:04 +02003217 .format_trace = format_tcp_rx_trace_short,
Dave Barach68b0fb02017-02-28 15:15:56 -05003218};
3219/* *INDENT-ON* */
3220
3221VLIB_NODE_FUNCTION_MULTIARCH (tcp6_listen_node, tcp6_listen);
3222
3223vlib_node_registration_t tcp4_input_node;
3224vlib_node_registration_t tcp6_input_node;
3225
3226typedef enum _tcp_input_next
3227{
3228 TCP_INPUT_NEXT_DROP,
3229 TCP_INPUT_NEXT_LISTEN,
3230 TCP_INPUT_NEXT_RCV_PROCESS,
3231 TCP_INPUT_NEXT_SYN_SENT,
3232 TCP_INPUT_NEXT_ESTABLISHED,
3233 TCP_INPUT_NEXT_RESET,
Pierre Pfister7fe51f32017-09-20 08:48:36 +02003234 TCP_INPUT_NEXT_PUNT,
Dave Barach68b0fb02017-02-28 15:15:56 -05003235 TCP_INPUT_N_NEXT
3236} tcp_input_next_t;
3237
3238#define foreach_tcp4_input_next \
Vijayabhaskar Katamreddyce074122017-11-15 13:50:26 -08003239 _ (DROP, "ip4-drop") \
Dave Barach68b0fb02017-02-28 15:15:56 -05003240 _ (LISTEN, "tcp4-listen") \
3241 _ (RCV_PROCESS, "tcp4-rcv-process") \
3242 _ (SYN_SENT, "tcp4-syn-sent") \
3243 _ (ESTABLISHED, "tcp4-established") \
Pierre Pfister7fe51f32017-09-20 08:48:36 +02003244 _ (RESET, "tcp4-reset") \
Vijayabhaskar Katamreddyce074122017-11-15 13:50:26 -08003245 _ (PUNT, "ip4-punt")
Dave Barach68b0fb02017-02-28 15:15:56 -05003246
3247#define foreach_tcp6_input_next \
Vijayabhaskar Katamreddyce074122017-11-15 13:50:26 -08003248 _ (DROP, "ip6-drop") \
Dave Barach68b0fb02017-02-28 15:15:56 -05003249 _ (LISTEN, "tcp6-listen") \
3250 _ (RCV_PROCESS, "tcp6-rcv-process") \
3251 _ (SYN_SENT, "tcp6-syn-sent") \
3252 _ (ESTABLISHED, "tcp6-established") \
Pierre Pfister7fe51f32017-09-20 08:48:36 +02003253 _ (RESET, "tcp6-reset") \
Vijayabhaskar Katamreddyce074122017-11-15 13:50:26 -08003254 _ (PUNT, "ip6-punt")
Dave Barach68b0fb02017-02-28 15:15:56 -05003255
Dave Barach68b0fb02017-02-28 15:15:56 -05003256#define filter_flags (TCP_FLAG_SYN|TCP_FLAG_ACK|TCP_FLAG_RST|TCP_FLAG_FIN)
3257
Florin Coras0c8a3bc2018-06-14 17:11:56 -07003258static void
3259tcp_input_trace_frame (vlib_main_t * vm, vlib_node_runtime_t * node,
Florin Coras4df38712018-06-20 12:44:16 -07003260 vlib_buffer_t ** bs, u32 n_bufs, u8 is_ip4)
Dave Barach68b0fb02017-02-28 15:15:56 -05003261{
Florin Coras0c8a3bc2018-06-14 17:11:56 -07003262 tcp_connection_t *tc;
3263 tcp_header_t *tcp;
3264 tcp_rx_trace_t *t;
Florin Coras0c8a3bc2018-06-14 17:11:56 -07003265 int i;
Dave Barach68b0fb02017-02-28 15:15:56 -05003266
Florin Coras4df38712018-06-20 12:44:16 -07003267 for (i = 0; i < n_bufs; i++)
Dave Barach68b0fb02017-02-28 15:15:56 -05003268 {
Florin Coras4df38712018-06-20 12:44:16 -07003269 if (bs[i]->flags & VLIB_BUFFER_IS_TRACED)
3270 {
3271 t = vlib_add_trace (vm, node, bs[i], sizeof (*t));
3272 tc = tcp_connection_get (vnet_buffer (bs[i])->tcp.connection_index,
3273 vm->thread_index);
3274 tcp = vlib_buffer_get_current (bs[i]);
3275 tcp_set_rx_trace_data (t, tc, tcp, bs[i], is_ip4);
3276 }
Florin Coras0c8a3bc2018-06-14 17:11:56 -07003277 }
3278}
Dave Barach68b0fb02017-02-28 15:15:56 -05003279
Florin Coras0c8a3bc2018-06-14 17:11:56 -07003280static void
3281tcp_input_set_error_next (tcp_main_t * tm, u16 * next, u32 * error, u8 is_ip4)
3282{
3283 if (*error == TCP_ERROR_FILTERED)
3284 {
3285 *next = TCP_INPUT_NEXT_DROP;
3286 }
3287 else if ((is_ip4 && tm->punt_unknown4) || (!is_ip4 && tm->punt_unknown6))
3288 {
3289 *next = TCP_INPUT_NEXT_PUNT;
3290 *error = TCP_ERROR_PUNT;
3291 }
3292 else
3293 {
3294 *next = TCP_INPUT_NEXT_RESET;
3295 *error = TCP_ERROR_NO_LISTENER;
3296 }
3297}
Dave Barach68b0fb02017-02-28 15:15:56 -05003298
Florin Coras0c8a3bc2018-06-14 17:11:56 -07003299static inline tcp_connection_t *
3300tcp_input_lookup_buffer (vlib_buffer_t * b, u8 thread_index, u32 * error,
3301 u8 is_ip4)
3302{
3303 u32 fib_index = vnet_buffer (b)->ip.fib_index;
3304 int n_advance_bytes, n_data_bytes;
3305 transport_connection_t *tc;
3306 tcp_header_t *tcp;
3307 u8 is_filtered = 0;
3308
3309 if (is_ip4)
3310 {
3311 ip4_header_t *ip4 = vlib_buffer_get_current (b);
3312 tcp = ip4_next_header (ip4);
3313 vnet_buffer (b)->tcp.hdr_offset = (u8 *) tcp - (u8 *) ip4;
3314 n_advance_bytes = (ip4_header_bytes (ip4) + tcp_header_bytes (tcp));
3315 n_data_bytes = clib_net_to_host_u16 (ip4->length) - n_advance_bytes;
3316
3317 /* Length check. Checksum computed by ipx_local no need to compute again */
3318 if (PREDICT_FALSE (n_advance_bytes < 0))
Dave Barach68b0fb02017-02-28 15:15:56 -05003319 {
Florin Coras0c8a3bc2018-06-14 17:11:56 -07003320 *error = TCP_ERROR_LENGTH;
3321 return 0;
Dave Barach68b0fb02017-02-28 15:15:56 -05003322 }
3323
Florin Coras0c8a3bc2018-06-14 17:11:56 -07003324 tc = session_lookup_connection_wt4 (fib_index, &ip4->dst_address,
3325 &ip4->src_address, tcp->dst_port,
3326 tcp->src_port, TRANSPORT_PROTO_TCP,
3327 thread_index, &is_filtered);
3328 }
3329 else
3330 {
3331 ip6_header_t *ip6 = vlib_buffer_get_current (b);
3332 tcp = ip6_next_header (ip6);
3333 vnet_buffer (b)->tcp.hdr_offset = (u8 *) tcp - (u8 *) ip6;
3334 n_advance_bytes = tcp_header_bytes (tcp);
3335 n_data_bytes = clib_net_to_host_u16 (ip6->payload_length)
3336 - n_advance_bytes;
3337 n_advance_bytes += sizeof (ip6[0]);
3338
3339 if (PREDICT_FALSE (n_advance_bytes < 0))
3340 {
3341 *error = TCP_ERROR_LENGTH;
3342 return 0;
3343 }
3344
3345 tc = session_lookup_connection_wt6 (fib_index, &ip6->dst_address,
3346 &ip6->src_address, tcp->dst_port,
3347 tcp->src_port, TRANSPORT_PROTO_TCP,
3348 thread_index, &is_filtered);
Dave Barach68b0fb02017-02-28 15:15:56 -05003349 }
3350
Florin Coras0c8a3bc2018-06-14 17:11:56 -07003351 vnet_buffer (b)->tcp.seq_number = clib_net_to_host_u32 (tcp->seq_number);
3352 vnet_buffer (b)->tcp.ack_number = clib_net_to_host_u32 (tcp->ack_number);
3353 vnet_buffer (b)->tcp.data_offset = n_advance_bytes;
3354 vnet_buffer (b)->tcp.data_len = n_data_bytes;
3355 vnet_buffer (b)->tcp.flags = 0;
3356
3357 *error = is_filtered ? TCP_ERROR_FILTERED : *error;
3358
3359 return tcp_get_connection_from_transport (tc);
3360}
3361
3362static inline void
3363tcp_input_dispatch_buffer (tcp_main_t * tm, tcp_connection_t * tc,
3364 vlib_buffer_t * b, u16 * next, u32 * error)
3365{
3366 tcp_header_t *tcp;
3367 u8 flags;
3368
3369 tcp = tcp_buffer_hdr (b);
3370 flags = tcp->flags & filter_flags;
3371 *next = tm->dispatch_table[tc->state][flags].next;
3372 *error = tm->dispatch_table[tc->state][flags].error;
3373
3374 if (PREDICT_FALSE (*error == TCP_ERROR_DISPATCH
3375 || *next == TCP_INPUT_NEXT_RESET))
3376 {
3377 /* Overload tcp flags to store state */
3378 tcp_state_t state = tc->state;
3379 vnet_buffer (b)->tcp.flags = tc->state;
3380
3381 if (*error == TCP_ERROR_DISPATCH)
3382 clib_warning ("disp error state %U flags %U", format_tcp_state,
3383 state, format_tcp_flags, (int) flags);
3384 }
3385}
3386
3387always_inline uword
3388tcp46_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
3389 vlib_frame_t * frame, int is_ip4)
3390{
3391 u32 n_left_from, *from, thread_index = vm->thread_index;
3392 tcp_main_t *tm = vnet_get_tcp_main ();
3393 vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
3394 u16 nexts[VLIB_FRAME_SIZE], *next;
3395
3396 tcp_set_time_now (thread_index);
3397
3398 from = vlib_frame_vector_args (frame);
3399 n_left_from = frame->n_vectors;
3400 vlib_get_buffers (vm, from, bufs, n_left_from);
3401
3402 b = bufs;
3403 next = nexts;
3404
3405 while (n_left_from >= 4)
3406 {
3407 u32 error0 = TCP_ERROR_NO_LISTENER, error1 = TCP_ERROR_NO_LISTENER;
3408 tcp_connection_t *tc0, *tc1;
3409
3410 {
3411 vlib_prefetch_buffer_header (b[2], STORE);
3412 CLIB_PREFETCH (b[2]->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
3413
3414 vlib_prefetch_buffer_header (b[3], STORE);
3415 CLIB_PREFETCH (b[3]->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
3416 }
3417
3418 next[0] = next[1] = TCP_INPUT_NEXT_DROP;
3419
3420 tc0 = tcp_input_lookup_buffer (b[0], thread_index, &error0, is_ip4);
3421 tc1 = tcp_input_lookup_buffer (b[1], thread_index, &error1, is_ip4);
3422
3423 if (PREDICT_TRUE (!tc0 + !tc1 == 0))
3424 {
3425 ASSERT (tcp_lookup_is_valid (tc0, tcp_buffer_hdr (b[0])));
3426 ASSERT (tcp_lookup_is_valid (tc1, tcp_buffer_hdr (b[1])));
3427
3428 vnet_buffer (b[0])->tcp.connection_index = tc0->c_c_index;
3429 vnet_buffer (b[1])->tcp.connection_index = tc1->c_c_index;
3430
3431 tcp_input_dispatch_buffer (tm, tc0, b[0], &next[0], &error0);
3432 tcp_input_dispatch_buffer (tm, tc1, b[1], &next[1], &error1);
3433 }
3434 else
3435 {
3436 if (PREDICT_TRUE (tc0 != 0))
3437 {
3438 ASSERT (tcp_lookup_is_valid (tc0, tcp_buffer_hdr (b[0])));
3439 vnet_buffer (b[0])->tcp.connection_index = tc0->c_c_index;
3440 tcp_input_dispatch_buffer (tm, tc0, b[0], &next[0], &error0);
3441 }
3442 else
3443 tcp_input_set_error_next (tm, &next[0], &error0, is_ip4);
3444
3445 if (PREDICT_TRUE (tc1 != 0))
3446 {
3447 ASSERT (tcp_lookup_is_valid (tc1, tcp_buffer_hdr (b[1])));
3448 vnet_buffer (b[1])->tcp.connection_index = tc1->c_c_index;
3449 tcp_input_dispatch_buffer (tm, tc1, b[1], &next[1], &error1);
3450 }
3451 else
3452 tcp_input_set_error_next (tm, &next[1], &error1, is_ip4);
3453 }
3454
3455 b += 2;
3456 next += 2;
3457 n_left_from -= 2;
3458 }
3459 while (n_left_from > 0)
3460 {
3461 tcp_connection_t *tc0;
3462 u32 error0 = TCP_ERROR_NO_LISTENER;
3463
3464 if (n_left_from > 1)
3465 {
3466 vlib_prefetch_buffer_header (b[1], STORE);
3467 CLIB_PREFETCH (b[1]->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
3468 }
3469
3470 next[0] = TCP_INPUT_NEXT_DROP;
3471 tc0 = tcp_input_lookup_buffer (b[0], thread_index, &error0, is_ip4);
3472 if (PREDICT_TRUE (tc0 != 0))
3473 {
3474 ASSERT (tcp_lookup_is_valid (tc0, tcp_buffer_hdr (b[0])));
3475 vnet_buffer (b[0])->tcp.connection_index = tc0->c_c_index;
3476 tcp_input_dispatch_buffer (tm, tc0, b[0], &next[0], &error0);
3477 }
3478 else
3479 tcp_input_set_error_next (tm, &next[0], &error0, is_ip4);
3480
3481 b += 1;
3482 next += 1;
3483 n_left_from -= 1;
3484 }
3485
3486 if (PREDICT_FALSE (node->flags & VLIB_NODE_FLAG_TRACE))
3487 tcp_input_trace_frame (vm, node, bufs, frame->n_vectors, is_ip4);
3488
3489 vlib_buffer_enqueue_to_next (vm, node, from, nexts, frame->n_vectors);
3490 return frame->n_vectors;
Dave Barach68b0fb02017-02-28 15:15:56 -05003491}
3492
3493static uword
3494tcp4_input (vlib_main_t * vm, vlib_node_runtime_t * node,
3495 vlib_frame_t * from_frame)
3496{
3497 return tcp46_input_inline (vm, node, from_frame, 1 /* is_ip4 */ );
3498}
3499
3500static uword
3501tcp6_input (vlib_main_t * vm, vlib_node_runtime_t * node,
3502 vlib_frame_t * from_frame)
3503{
3504 return tcp46_input_inline (vm, node, from_frame, 0 /* is_ip4 */ );
3505}
3506
3507/* *INDENT-OFF* */
3508VLIB_REGISTER_NODE (tcp4_input_node) =
3509{
3510 .function = tcp4_input,
3511 .name = "tcp4-input",
3512 /* Takes a vector of packets. */
3513 .vector_size = sizeof (u32),
3514 .n_errors = TCP_N_ERROR,
3515 .error_strings = tcp_error_strings,
3516 .n_next_nodes = TCP_INPUT_N_NEXT,
3517 .next_nodes =
3518 {
3519#define _(s,n) [TCP_INPUT_NEXT_##s] = n,
3520 foreach_tcp4_input_next
3521#undef _
3522 },
3523 .format_buffer = format_tcp_header,
3524 .format_trace = format_tcp_rx_trace,
3525};
3526/* *INDENT-ON* */
3527
3528VLIB_NODE_FUNCTION_MULTIARCH (tcp4_input_node, tcp4_input);
3529
3530/* *INDENT-OFF* */
3531VLIB_REGISTER_NODE (tcp6_input_node) =
3532{
3533 .function = tcp6_input,
3534 .name = "tcp6-input",
3535 /* Takes a vector of packets. */
3536 .vector_size = sizeof (u32),
3537 .n_errors = TCP_N_ERROR,
3538 .error_strings = tcp_error_strings,
3539 .n_next_nodes = TCP_INPUT_N_NEXT,
3540 .next_nodes =
3541 {
3542#define _(s,n) [TCP_INPUT_NEXT_##s] = n,
3543 foreach_tcp6_input_next
3544#undef _
3545 },
3546 .format_buffer = format_tcp_header,
3547 .format_trace = format_tcp_rx_trace,
3548};
3549/* *INDENT-ON* */
3550
3551VLIB_NODE_FUNCTION_MULTIARCH (tcp6_input_node, tcp6_input);
Dave Barach68b0fb02017-02-28 15:15:56 -05003552
3553static void
3554tcp_dispatch_table_init (tcp_main_t * tm)
3555{
3556 int i, j;
3557 for (i = 0; i < ARRAY_LEN (tm->dispatch_table); i++)
3558 for (j = 0; j < ARRAY_LEN (tm->dispatch_table[i]); j++)
3559 {
3560 tm->dispatch_table[i][j].next = TCP_INPUT_NEXT_DROP;
3561 tm->dispatch_table[i][j].error = TCP_ERROR_DISPATCH;
3562 }
3563
3564#define _(t,f,n,e) \
3565do { \
3566 tm->dispatch_table[TCP_STATE_##t][f].next = (n); \
3567 tm->dispatch_table[TCP_STATE_##t][f].error = (e); \
3568} while (0)
3569
3570 /* SYNs for new connections -> tcp-listen. */
3571 _(LISTEN, TCP_FLAG_SYN, TCP_INPUT_NEXT_LISTEN, TCP_ERROR_NONE);
Florin Corasdc629cd2017-05-09 00:52:37 -07003572 _(LISTEN, TCP_FLAG_ACK, TCP_INPUT_NEXT_RESET, TCP_ERROR_NONE);
Florin Coras00cd22d2018-04-18 13:20:18 -07003573 _(LISTEN, TCP_FLAG_RST, TCP_INPUT_NEXT_DROP, TCP_ERROR_RST_RCVD);
Dave Barach2c25a622017-06-26 11:35:07 -04003574 _(LISTEN, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RESET,
3575 TCP_ERROR_NONE);
Dave Barach68b0fb02017-02-28 15:15:56 -05003576 /* ACK for for a SYN-ACK -> tcp-rcv-process. */
3577 _(SYN_RCVD, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
Florin Corasc28764f2017-04-26 00:08:42 -07003578 _(SYN_RCVD, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
Florin Corase8460c72018-09-24 14:40:40 -07003579 _(SYN_RCVD, TCP_FLAG_RST | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS,
3580 TCP_ERROR_NONE);
Dave Barach2c25a622017-06-26 11:35:07 -04003581 _(SYN_RCVD, TCP_FLAG_SYN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
Florin Corasc01c4452018-09-20 18:36:54 -07003582 _(SYN_RCVD, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS,
3583 TCP_ERROR_NONE);
Dave Barach68b0fb02017-02-28 15:15:56 -05003584 /* SYN-ACK for a SYN */
3585 _(SYN_SENT, TCP_FLAG_SYN | TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT,
3586 TCP_ERROR_NONE);
3587 _(SYN_SENT, TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE);
3588 _(SYN_SENT, TCP_FLAG_RST, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE);
3589 _(SYN_SENT, TCP_FLAG_RST | TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT,
3590 TCP_ERROR_NONE);
3591 /* ACK for for established connection -> tcp-established. */
3592 _(ESTABLISHED, TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3593 /* FIN for for established connection -> tcp-established. */
3594 _(ESTABLISHED, TCP_FLAG_FIN, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
3595 _(ESTABLISHED, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED,
3596 TCP_ERROR_NONE);
Florin Corasd79b41e2017-03-04 05:37:52 -08003597 _(ESTABLISHED, TCP_FLAG_RST, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
Florin Corasdc629cd2017-05-09 00:52:37 -07003598 _(ESTABLISHED, TCP_FLAG_RST | TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED,
3599 TCP_ERROR_NONE);
Dave Barach2c25a622017-06-26 11:35:07 -04003600 _(ESTABLISHED, TCP_FLAG_SYN, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE);
Florin Coras3eb50622017-07-13 01:24:57 -04003601 _(ESTABLISHED, TCP_FLAG_SYN | TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED,
3602 TCP_ERROR_NONE);
Dave Barach68b0fb02017-02-28 15:15:56 -05003603 /* ACK or FIN-ACK to our FIN */
3604 _(FIN_WAIT_1, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3605 _(FIN_WAIT_1, TCP_FLAG_ACK | TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS,
3606 TCP_ERROR_NONE);
3607 /* FIN in reply to our FIN from the other side */
3608 _(FIN_WAIT_1, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
Dave Barach2c25a622017-06-26 11:35:07 -04003609 _(FIN_WAIT_1, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
Florin Coras56318932018-05-23 20:44:12 -07003610 _(CLOSING, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
Dave Barach68b0fb02017-02-28 15:15:56 -05003611 /* FIN confirming that the peer (app) has closed */
3612 _(FIN_WAIT_2, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
Florin Corasf6d68ed2017-05-07 19:12:02 -07003613 _(FIN_WAIT_2, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
Dave Barach68b0fb02017-02-28 15:15:56 -05003614 _(FIN_WAIT_2, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS,
3615 TCP_ERROR_NONE);
Florin Corasf03a59a2017-06-09 21:07:32 -07003616 _(CLOSE_WAIT, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3617 _(CLOSE_WAIT, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS,
3618 TCP_ERROR_NONE);
Dave Barach68b0fb02017-02-28 15:15:56 -05003619 _(LAST_ACK, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
Florin Coras93992a92017-05-24 18:03:56 -07003620 _(LAST_ACK, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3621 _(LAST_ACK, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS,
3622 TCP_ERROR_NONE);
Florin Corasdb84e572017-05-09 18:54:52 -07003623 _(LAST_ACK, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
Florin Corasc01d5782018-10-17 14:53:11 -07003624 _(LAST_ACK, TCP_FLAG_SYN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
Florin Coras93992a92017-05-24 18:03:56 -07003625 _(TIME_WAIT, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
3626 _(TIME_WAIT, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS,
3627 TCP_ERROR_NONE);
Florin Coras3eb50622017-07-13 01:24:57 -04003628 _(TIME_WAIT, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
Florin Coras50958952017-08-29 14:50:13 -07003629 _(TIME_WAIT, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE);
Florin Coras9d063042017-09-14 03:08:00 -04003630 _(CLOSED, TCP_FLAG_ACK, TCP_INPUT_NEXT_DROP, TCP_ERROR_CONNECTION_CLOSED);
Florin Corasdc629cd2017-05-09 00:52:37 -07003631 _(CLOSED, TCP_FLAG_RST, TCP_INPUT_NEXT_DROP, TCP_ERROR_CONNECTION_CLOSED);
Florin Coras9d063042017-09-14 03:08:00 -04003632 _(CLOSED, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_DROP,
Dave Barach2c25a622017-06-26 11:35:07 -04003633 TCP_ERROR_CONNECTION_CLOSED);
Dave Barach68b0fb02017-02-28 15:15:56 -05003634#undef _
3635}
3636
Florin Coras0dbd5172018-06-25 16:19:34 -07003637static clib_error_t *
Dave Barach68b0fb02017-02-28 15:15:56 -05003638tcp_input_init (vlib_main_t * vm)
3639{
3640 clib_error_t *error = 0;
3641 tcp_main_t *tm = vnet_get_tcp_main ();
3642
3643 if ((error = vlib_call_init_function (vm, tcp_init)))
3644 return error;
3645
3646 /* Initialize dispatch table. */
3647 tcp_dispatch_table_init (tm);
3648
3649 return error;
3650}
3651
3652VLIB_INIT_FUNCTION (tcp_input_init);
3653
3654/*
3655 * fd.io coding-style-patch-verification: ON
3656 *
3657 * Local Variables:
3658 * eval: (c-set-style "gnu")
3659 * End:
3660 */