Line data Source code
1 : /*
2 : * Copyright (c) 2016-2019 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 <vnet/tcp/tcp.h>
17 : #include <vnet/tcp/tcp_inlines.h>
18 : #include <math.h>
19 : #include <vnet/ip/ip4_inlines.h>
20 : #include <vnet/ip/ip6_inlines.h>
21 :
22 : typedef enum _tcp_output_next
23 : {
24 : TCP_OUTPUT_NEXT_DROP,
25 : TCP_OUTPUT_NEXT_IP_LOOKUP,
26 : TCP_OUTPUT_NEXT_IP_REWRITE,
27 : TCP_OUTPUT_NEXT_IP_ARP,
28 : TCP_OUTPUT_N_NEXT
29 : } tcp_output_next_t;
30 :
31 : #define foreach_tcp4_output_next \
32 : _ (DROP, "error-drop") \
33 : _ (IP_LOOKUP, "ip4-lookup") \
34 : _ (IP_REWRITE, "ip4-rewrite") \
35 : _ (IP_ARP, "ip4-arp")
36 :
37 : #define foreach_tcp6_output_next \
38 : _ (DROP, "error-drop") \
39 : _ (IP_LOOKUP, "ip6-lookup") \
40 : _ (IP_REWRITE, "ip6-rewrite") \
41 : _ (IP_ARP, "ip6-discover-neighbor")
42 :
43 : static vlib_error_desc_t tcp_output_error_counters[] = {
44 : #define tcp_error(f, n, s, d) { #n, d, VL_COUNTER_SEVERITY_##s },
45 : #include <vnet/tcp/tcp_error.def>
46 : #undef tcp_error
47 : };
48 :
49 : typedef struct
50 : {
51 : tcp_header_t tcp_header;
52 : tcp_connection_t tcp_connection;
53 : } tcp_tx_trace_t;
54 :
55 : static u8 *
56 0 : format_tcp_tx_trace (u8 * s, va_list * args)
57 : {
58 0 : CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
59 0 : CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
60 0 : tcp_tx_trace_t *t = va_arg (*args, tcp_tx_trace_t *);
61 0 : tcp_connection_t *tc = &t->tcp_connection;
62 0 : u32 indent = format_get_indent (s);
63 :
64 0 : s = format (s, "%U state %U\n%U%U", format_tcp_connection_id, tc,
65 0 : format_tcp_state, tc->state, format_white_space, indent,
66 : format_tcp_header, &t->tcp_header, 128);
67 :
68 0 : return s;
69 : }
70 :
71 : #ifndef CLIB_MARCH_VARIANT
72 : static u8
73 264 : tcp_window_compute_scale (u32 window)
74 : {
75 264 : u8 wnd_scale = 0;
76 2904 : while (wnd_scale < TCP_MAX_WND_SCALE && (window >> wnd_scale) > TCP_WND_MAX)
77 2640 : wnd_scale++;
78 264 : return wnd_scale;
79 : }
80 :
81 : /**
82 : * TCP's initial window
83 : */
84 : always_inline u32
85 279 : tcp_initial_wnd_unscaled (tcp_connection_t * tc)
86 : {
87 : /* RFC 6928 recommends the value lower. However at the time our connections
88 : * are initialized, fifos may not be allocated. Therefore, advertise the
89 : * smallest possible unscaled window size and update once fifos are
90 : * assigned to the session.
91 : */
92 : /*
93 : tcp_update_rcv_mss (tc);
94 : TCP_IW_N_SEGMENTS * tc->mss;
95 : */
96 279 : return tcp_cfg.min_rx_fifo;
97 : }
98 :
99 : /**
100 : * Compute initial window and scale factor. As per RFC1323, window field in
101 : * SYN and SYN-ACK segments is never scaled.
102 : */
103 : u32
104 279 : tcp_initial_window_to_advertise (tcp_connection_t * tc)
105 : {
106 : /* Compute rcv wscale only if peer advertised support for it */
107 279 : if (tc->state != TCP_STATE_SYN_RCVD || tcp_opts_wscale (&tc->rcv_opts))
108 264 : tc->rcv_wscale = tcp_window_compute_scale (tcp_cfg.max_rx_fifo);
109 :
110 279 : tc->rcv_wnd = tcp_initial_wnd_unscaled (tc);
111 :
112 279 : return clib_min (tc->rcv_wnd, TCP_WND_MAX);
113 : }
114 :
115 : static inline void
116 95327 : tcp_update_rcv_wnd (tcp_connection_t * tc)
117 : {
118 : u32 available_space, wnd;
119 : i32 observed_wnd;
120 :
121 : /*
122 : * Figure out how much space we have available
123 : */
124 95327 : available_space = transport_max_rx_enqueue (&tc->connection);
125 :
126 : /*
127 : * Use the above and what we know about what we've previously advertised
128 : * to compute the new window
129 : */
130 95327 : observed_wnd = (i32) tc->rcv_wnd - (tc->rcv_nxt - tc->rcv_las);
131 :
132 : /* Check if we are about to retract the window. Do the comparison before
133 : * rounding to avoid errors. Per RFC7323 sec. 2.4 we could remove this */
134 95327 : if (PREDICT_FALSE ((i32) available_space < observed_wnd))
135 : {
136 0 : wnd = round_down_pow2 (clib_max (observed_wnd, 0), 1 << tc->rcv_wscale);
137 : TCP_EVT (TCP_EVT_RCV_WND_SHRUNK, tc, observed_wnd, available_space);
138 : }
139 : else
140 : {
141 : /* Make sure we have a multiple of 1 << rcv_wscale. We round down to
142 : * avoid advertising a window larger than what can be buffered */
143 95327 : wnd = round_down_pow2 (available_space, 1 << tc->rcv_wscale);
144 : }
145 :
146 95327 : if (PREDICT_FALSE (wnd < tc->rcv_opts.mss))
147 0 : wnd = 0;
148 :
149 95327 : tc->rcv_wnd = clib_min (wnd, TCP_WND_MAX << tc->rcv_wscale);
150 95327 : }
151 :
152 : /**
153 : * Compute and return window to advertise, scaled as per RFC1323
154 : */
155 : static inline u32
156 24495 : tcp_window_to_advertise (tcp_connection_t * tc, tcp_state_t state)
157 : {
158 24495 : if (state < TCP_STATE_ESTABLISHED)
159 0 : return tcp_initial_window_to_advertise (tc);
160 :
161 24495 : tcp_update_rcv_wnd (tc);
162 24495 : return tc->rcv_wnd >> tc->rcv_wscale;
163 : }
164 :
165 : static int
166 132 : tcp_make_syn_options (tcp_connection_t * tc, tcp_options_t * opts)
167 : {
168 132 : u8 len = 0;
169 :
170 132 : opts->flags |= TCP_OPTS_FLAG_MSS;
171 132 : opts->mss = tc->mss;
172 132 : len += TCP_OPTION_LEN_MSS;
173 :
174 132 : opts->flags |= TCP_OPTS_FLAG_WSCALE;
175 132 : opts->wscale = tc->rcv_wscale;
176 132 : len += TCP_OPTION_LEN_WINDOW_SCALE;
177 :
178 132 : opts->flags |= TCP_OPTS_FLAG_TSTAMP;
179 132 : opts->tsval = tcp_time_tstamp (tc->c_thread_index);
180 132 : opts->tsecr = 0;
181 132 : len += TCP_OPTION_LEN_TIMESTAMP;
182 :
183 : if (TCP_USE_SACKS)
184 : {
185 132 : opts->flags |= TCP_OPTS_FLAG_SACK_PERMITTED;
186 132 : len += TCP_OPTION_LEN_SACK_PERMITTED;
187 : }
188 :
189 : /* Align to needed boundary */
190 132 : len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN;
191 132 : return len;
192 : }
193 :
194 : static int
195 147 : tcp_make_synack_options (tcp_connection_t * tc, tcp_options_t * opts)
196 : {
197 147 : u8 len = 0;
198 :
199 147 : opts->flags |= TCP_OPTS_FLAG_MSS;
200 147 : opts->mss = tc->mss;
201 147 : len += TCP_OPTION_LEN_MSS;
202 :
203 147 : if (tcp_opts_wscale (&tc->rcv_opts))
204 : {
205 132 : opts->flags |= TCP_OPTS_FLAG_WSCALE;
206 132 : opts->wscale = tc->rcv_wscale;
207 132 : len += TCP_OPTION_LEN_WINDOW_SCALE;
208 : }
209 :
210 147 : if (tcp_opts_tstamp (&tc->rcv_opts))
211 : {
212 132 : opts->flags |= TCP_OPTS_FLAG_TSTAMP;
213 132 : opts->tsval = tcp_time_tstamp (tc->c_thread_index);
214 132 : opts->tsecr = tc->tsval_recent;
215 132 : len += TCP_OPTION_LEN_TIMESTAMP;
216 : }
217 :
218 147 : if (tcp_opts_sack_permitted (&tc->rcv_opts))
219 : {
220 132 : opts->flags |= TCP_OPTS_FLAG_SACK_PERMITTED;
221 132 : len += TCP_OPTION_LEN_SACK_PERMITTED;
222 : }
223 :
224 : /* Align to needed boundary */
225 147 : len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN;
226 147 : return len;
227 : }
228 :
229 : static int
230 95333 : tcp_make_established_options (tcp_connection_t * tc, tcp_options_t * opts)
231 : {
232 95333 : u8 len = 0;
233 :
234 95333 : opts->flags = 0;
235 :
236 95333 : if (tcp_opts_tstamp (&tc->rcv_opts))
237 : {
238 95333 : opts->flags |= TCP_OPTS_FLAG_TSTAMP;
239 95333 : opts->tsval = tcp_tstamp (tc);
240 95333 : opts->tsecr = tc->tsval_recent;
241 95333 : len += TCP_OPTION_LEN_TIMESTAMP;
242 : }
243 95333 : if (tcp_opts_sack_permitted (&tc->rcv_opts))
244 : {
245 95333 : if (vec_len (tc->snd_sacks))
246 : {
247 0 : opts->flags |= TCP_OPTS_FLAG_SACK;
248 0 : if (tc->snd_sack_pos >= vec_len (tc->snd_sacks))
249 0 : tc->snd_sack_pos = 0;
250 0 : opts->sacks = &tc->snd_sacks[tc->snd_sack_pos];
251 0 : opts->n_sack_blocks = vec_len (tc->snd_sacks) - tc->snd_sack_pos;
252 0 : opts->n_sack_blocks = clib_min (opts->n_sack_blocks,
253 : TCP_OPTS_MAX_SACK_BLOCKS);
254 0 : tc->snd_sack_pos += opts->n_sack_blocks;
255 0 : len += 2 + TCP_OPTION_LEN_SACK_BLOCK * opts->n_sack_blocks;
256 : }
257 : }
258 :
259 : /* Align to needed boundary */
260 95333 : len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN;
261 95333 : return len;
262 : }
263 :
264 : always_inline int
265 70838 : tcp_make_options (tcp_connection_t * tc, tcp_options_t * opts,
266 : tcp_state_t state)
267 : {
268 70838 : switch (state)
269 : {
270 70838 : case TCP_STATE_ESTABLISHED:
271 : case TCP_STATE_CLOSE_WAIT:
272 : case TCP_STATE_FIN_WAIT_1:
273 : case TCP_STATE_LAST_ACK:
274 : case TCP_STATE_CLOSING:
275 : case TCP_STATE_FIN_WAIT_2:
276 : case TCP_STATE_TIME_WAIT:
277 : case TCP_STATE_CLOSED:
278 70838 : return tcp_make_established_options (tc, opts);
279 0 : case TCP_STATE_SYN_RCVD:
280 0 : return tcp_make_synack_options (tc, opts);
281 0 : case TCP_STATE_SYN_SENT:
282 0 : return tcp_make_syn_options (tc, opts);
283 0 : default:
284 0 : clib_warning ("State not handled! %d", state);
285 0 : return 0;
286 : }
287 : }
288 :
289 : /**
290 : * Update burst send vars
291 : *
292 : * - Updates snd_mss to reflect the effective segment size that we can send
293 : * by taking into account all TCP options, including SACKs.
294 : * - Cache 'on the wire' options for reuse
295 : * - Updates receive window which can be reused for a burst.
296 : *
297 : * This should *only* be called when doing bursts
298 : */
299 : void
300 70832 : tcp_update_burst_snd_vars (tcp_connection_t * tc)
301 : {
302 70832 : tcp_main_t *tm = &tcp_main;
303 :
304 : /* Compute options to be used for connection. These may be reused when
305 : * sending data or to compute the effective mss (snd_mss) */
306 70832 : tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts,
307 : TCP_STATE_ESTABLISHED);
308 :
309 : /* XXX check if MTU has been updated */
310 70832 : tc->snd_mss = clib_min (tc->mss, tc->rcv_opts.mss) - tc->snd_opts_len;
311 70832 : ASSERT (tc->snd_mss > 0);
312 :
313 70832 : tcp_options_write (tm->wrk_ctx[tc->c_thread_index].cached_opts,
314 : &tc->snd_opts);
315 :
316 70832 : tcp_update_rcv_wnd (tc);
317 :
318 70832 : if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
319 0 : tcp_bt_check_app_limited (tc);
320 :
321 70832 : if (tc->snd_una == tc->snd_nxt)
322 : {
323 25861 : tcp_cc_event (tc, TCP_CC_EVT_START_TX);
324 : }
325 :
326 70832 : if (tc->flags & TCP_CONN_PSH_PENDING)
327 : {
328 23937 : u32 max_deq = transport_max_tx_dequeue (&tc->connection);
329 : /* Last byte marked for push */
330 23937 : tc->psh_seq = tc->snd_una + max_deq - 1;
331 : }
332 70832 : }
333 :
334 : static void *
335 24780 : tcp_init_buffer (vlib_main_t * vm, vlib_buffer_t * b)
336 : {
337 24780 : ASSERT ((b->flags & VLIB_BUFFER_NEXT_PRESENT) == 0);
338 24780 : b->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED;
339 24780 : b->total_length_not_including_first_buffer = 0;
340 24780 : b->current_data = 0;
341 24780 : vnet_buffer (b)->tcp.flags = 0;
342 : /* Leave enough space for headers */
343 24780 : return vlib_buffer_make_headroom (b, TRANSPORT_MAX_HDRS_LEN);
344 : }
345 :
346 : /* Compute TCP checksum in software when offloading is disabled for a connection */
347 : u16
348 0 : ip6_tcp_compute_checksum_custom (vlib_main_t * vm, vlib_buffer_t * p0,
349 : ip46_address_t * src, ip46_address_t * dst)
350 : {
351 : ip_csum_t sum0;
352 : u16 payload_length_host_byte_order;
353 : u32 i;
354 :
355 : /* Initialize checksum with ip header. */
356 0 : sum0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, p0)) +
357 0 : clib_host_to_net_u16 (IP_PROTOCOL_TCP);
358 0 : payload_length_host_byte_order = vlib_buffer_length_in_chain (vm, p0);
359 :
360 0 : for (i = 0; i < ARRAY_LEN (src->ip6.as_uword); i++)
361 : {
362 0 : sum0 = ip_csum_with_carry
363 0 : (sum0, clib_mem_unaligned (&src->ip6.as_uword[i], uword));
364 0 : sum0 = ip_csum_with_carry
365 0 : (sum0, clib_mem_unaligned (&dst->ip6.as_uword[i], uword));
366 : }
367 :
368 0 : return ip_calculate_l4_checksum (vm, p0, sum0,
369 : payload_length_host_byte_order, NULL, 0,
370 : NULL);
371 : }
372 :
373 : u16
374 0 : ip4_tcp_compute_checksum_custom (vlib_main_t * vm, vlib_buffer_t * p0,
375 : ip46_address_t * src, ip46_address_t * dst)
376 : {
377 : ip_csum_t sum0;
378 : u32 payload_length_host_byte_order;
379 :
380 0 : payload_length_host_byte_order = vlib_buffer_length_in_chain (vm, p0);
381 0 : sum0 =
382 0 : clib_host_to_net_u32 (payload_length_host_byte_order +
383 : (IP_PROTOCOL_TCP << 16));
384 :
385 0 : sum0 = ip_csum_with_carry (sum0, clib_mem_unaligned (&src->ip4, u32));
386 0 : sum0 = ip_csum_with_carry (sum0, clib_mem_unaligned (&dst->ip4, u32));
387 :
388 0 : return ip_calculate_l4_checksum (vm, p0, sum0,
389 : payload_length_host_byte_order, NULL, 0,
390 : NULL);
391 : }
392 :
393 : static inline u16
394 1020810 : tcp_compute_checksum (tcp_connection_t * tc, vlib_buffer_t * b)
395 : {
396 1020810 : u16 checksum = 0;
397 1020810 : if (PREDICT_FALSE (tc->cfg_flags & TCP_CFG_F_NO_CSUM_OFFLOAD))
398 : {
399 0 : tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
400 0 : vlib_main_t *vm = wrk->vm;
401 :
402 0 : if (tc->c_is_ip4)
403 0 : checksum = ip4_tcp_compute_checksum_custom
404 : (vm, b, &tc->c_lcl_ip, &tc->c_rmt_ip);
405 : else
406 0 : checksum = ip6_tcp_compute_checksum_custom
407 : (vm, b, &tc->c_lcl_ip, &tc->c_rmt_ip);
408 : }
409 : else
410 : {
411 1020810 : vnet_buffer_offload_flags_set (b, VNET_BUFFER_OFFLOAD_F_TCP_CKSUM);
412 : }
413 1020810 : return checksum;
414 : }
415 :
416 : /**
417 : * Prepare ACK
418 : */
419 : static inline void
420 24495 : tcp_make_ack_i (tcp_connection_t * tc, vlib_buffer_t * b, tcp_state_t state,
421 : u8 flags)
422 : {
423 24495 : tcp_options_t _snd_opts, *snd_opts = &_snd_opts;
424 : u8 tcp_opts_len, tcp_hdr_opts_len;
425 : tcp_header_t *th;
426 : u16 wnd;
427 :
428 24495 : wnd = tcp_window_to_advertise (tc, state);
429 :
430 : /* Make and write options */
431 24495 : tcp_opts_len = tcp_make_established_options (tc, snd_opts);
432 24495 : tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t);
433 :
434 24495 : th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->snd_nxt,
435 : tc->rcv_nxt, tcp_hdr_opts_len, flags, wnd);
436 :
437 24495 : tcp_options_write ((u8 *) (th + 1), snd_opts);
438 :
439 24495 : th->checksum = tcp_compute_checksum (tc, b);
440 :
441 24495 : vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
442 :
443 24495 : if (wnd == 0)
444 : {
445 0 : transport_rx_fifo_req_deq_ntf (&tc->connection);
446 0 : tcp_zero_rwnd_sent_on (tc);
447 : }
448 : else
449 24495 : tcp_zero_rwnd_sent_off (tc);
450 24495 : }
451 :
452 : /**
453 : * Convert buffer to ACK
454 : */
455 : static inline void
456 24239 : tcp_make_ack (tcp_connection_t * tc, vlib_buffer_t * b)
457 : {
458 24239 : tcp_make_ack_i (tc, b, TCP_STATE_ESTABLISHED, TCP_FLAG_ACK);
459 : TCP_EVT (TCP_EVT_ACK_SENT, tc);
460 24239 : tc->rcv_las = tc->rcv_nxt;
461 24239 : }
462 :
463 : /**
464 : * Convert buffer to FIN-ACK
465 : */
466 : static void
467 256 : tcp_make_fin (tcp_connection_t * tc, vlib_buffer_t * b)
468 : {
469 256 : tcp_make_ack_i (tc, b, TCP_STATE_ESTABLISHED, TCP_FLAG_FIN | TCP_FLAG_ACK);
470 256 : }
471 :
472 : /**
473 : * Convert buffer to SYN
474 : */
475 : void
476 132 : tcp_make_syn (tcp_connection_t * tc, vlib_buffer_t * b)
477 : {
478 : u8 tcp_hdr_opts_len, tcp_opts_len;
479 : tcp_header_t *th;
480 : u16 initial_wnd;
481 : tcp_options_t snd_opts;
482 :
483 132 : initial_wnd = tcp_initial_window_to_advertise (tc);
484 :
485 : /* Make and write options */
486 132 : clib_memset (&snd_opts, 0, sizeof (snd_opts));
487 132 : tcp_opts_len = tcp_make_syn_options (tc, &snd_opts);
488 132 : tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t);
489 :
490 132 : th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->iss,
491 : tc->rcv_nxt, tcp_hdr_opts_len, TCP_FLAG_SYN,
492 : initial_wnd);
493 132 : vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
494 132 : tcp_options_write ((u8 *) (th + 1), &snd_opts);
495 132 : th->checksum = tcp_compute_checksum (tc, b);
496 132 : }
497 :
498 : /**
499 : * Convert buffer to SYN-ACK
500 : */
501 : static void
502 147 : tcp_make_synack (tcp_connection_t * tc, vlib_buffer_t * b)
503 : {
504 147 : tcp_options_t _snd_opts, *snd_opts = &_snd_opts;
505 : u8 tcp_opts_len, tcp_hdr_opts_len;
506 : tcp_header_t *th;
507 : u16 initial_wnd;
508 :
509 147 : clib_memset (snd_opts, 0, sizeof (*snd_opts));
510 147 : initial_wnd = tcp_initial_window_to_advertise (tc);
511 147 : tcp_opts_len = tcp_make_synack_options (tc, snd_opts);
512 147 : tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t);
513 :
514 147 : th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->iss,
515 : tc->rcv_nxt, tcp_hdr_opts_len,
516 : TCP_FLAG_SYN | TCP_FLAG_ACK, initial_wnd);
517 147 : tcp_options_write ((u8 *) (th + 1), snd_opts);
518 :
519 147 : vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
520 147 : th->checksum = tcp_compute_checksum (tc, b);
521 147 : }
522 :
523 : static void
524 132 : tcp_enqueue_half_open (tcp_worker_ctx_t *wrk, tcp_connection_t *tc,
525 : vlib_buffer_t *b, u32 bi)
526 : {
527 132 : vlib_main_t *vm = wrk->vm;
528 :
529 132 : b->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED;
530 132 : b->error = 0;
531 :
532 132 : session_add_pending_tx_buffer (vm->thread_index, bi,
533 132 : wrk->tco_next_node[!tc->c_is_ip4]);
534 :
535 132 : if (vm->thread_index == 0 && vlib_num_workers ())
536 0 : session_queue_run_on_main_thread (vm);
537 132 : }
538 :
539 : static void
540 24648 : tcp_enqueue_to_output (tcp_worker_ctx_t * wrk, vlib_buffer_t * b, u32 bi,
541 : u8 is_ip4)
542 : {
543 24648 : b->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED;
544 24648 : b->error = 0;
545 :
546 24648 : session_add_pending_tx_buffer (wrk->vm->thread_index, bi,
547 24648 : wrk->tco_next_node[!is_ip4]);
548 24648 : }
549 :
550 : int
551 86 : tcp_buffer_make_reset (vlib_main_t *vm, vlib_buffer_t *b, u8 is_ip4)
552 : {
553 86 : ip4_address_t src_ip4 = {}, dst_ip4 = {};
554 : ip6_address_t src_ip6, dst_ip6;
555 : u16 src_port, dst_port;
556 : u32 tmp, len, seq, ack;
557 : ip4_header_t *ih4;
558 : ip6_header_t *ih6;
559 : tcp_header_t *th;
560 : u8 flags;
561 :
562 : /*
563 : * Find IP and TCP headers and glean information from them. Assumes
564 : * buffer was parsed by something like @ref tcp_input_lookup_buffer
565 : */
566 86 : th = tcp_buffer_hdr (b);
567 :
568 86 : if (is_ip4)
569 : {
570 86 : ih4 = vlib_buffer_get_current (b);
571 86 : ASSERT ((ih4->ip_version_and_header_length & 0xF0) == 0x40);
572 86 : src_ip4.as_u32 = ih4->src_address.as_u32;
573 86 : dst_ip4.as_u32 = ih4->dst_address.as_u32;
574 : }
575 : else
576 : {
577 0 : ih6 = vlib_buffer_get_current (b);
578 0 : ASSERT ((ih6->ip_version_traffic_class_and_flow_label & 0xF0) == 0x60);
579 0 : clib_memcpy_fast (&src_ip6, &ih6->src_address, sizeof (ip6_address_t));
580 0 : clib_memcpy_fast (&dst_ip6, &ih6->dst_address, sizeof (ip6_address_t));
581 : }
582 :
583 86 : src_port = th->src_port;
584 86 : dst_port = th->dst_port;
585 86 : flags = TCP_FLAG_RST;
586 :
587 : /*
588 : * RFC 793. If the ACK bit is off, sequence number zero is used,
589 : * <SEQ=0><ACK=SEG.SEQ+SEG.LEN><CTL=RST,ACK>
590 : * If the ACK bit is on,
591 : * <SEQ=SEG.ACK><CTL=RST>
592 : */
593 86 : if (tcp_ack (th))
594 : {
595 86 : seq = th->ack_number;
596 86 : ack = 0;
597 : }
598 : else
599 : {
600 0 : flags |= TCP_FLAG_ACK;
601 0 : tmp = clib_net_to_host_u32 (th->seq_number);
602 0 : len = vnet_buffer (b)->tcp.data_len + tcp_is_syn (th) + tcp_is_fin (th);
603 0 : ack = clib_host_to_net_u32 (tmp + len);
604 0 : seq = 0;
605 : }
606 :
607 : /*
608 : * Clear and reuse current buffer for reset
609 : */
610 86 : if (b->flags & VLIB_BUFFER_NEXT_PRESENT)
611 0 : vlib_buffer_free_one (vm, b->next_buffer);
612 :
613 : /* Zero all flags but free list index and trace flag */
614 86 : b->flags &= VLIB_BUFFER_NEXT_PRESENT - 1;
615 : /* Make sure new tcp header comes after current ip */
616 86 : b->current_data = ((u8 *) th - b->data) + sizeof (tcp_header_t);
617 86 : b->current_length = 0;
618 86 : b->total_length_not_including_first_buffer = 0;
619 86 : vnet_buffer (b)->tcp.flags = 0;
620 :
621 : /*
622 : * Add TCP and IP headers
623 : */
624 86 : th = vlib_buffer_push_tcp_net_order (b, dst_port, src_port, seq, ack,
625 : sizeof (tcp_header_t), flags, 0);
626 :
627 86 : if (is_ip4)
628 : {
629 86 : ih4 = vlib_buffer_push_ip4 (vm, b, &dst_ip4, &src_ip4,
630 : IP_PROTOCOL_TCP, 1);
631 86 : th->checksum = ip4_tcp_udp_compute_checksum (vm, b, ih4);
632 : }
633 : else
634 : {
635 0 : int bogus = ~0;
636 0 : ih6 = vlib_buffer_push_ip6 (vm, b, &dst_ip6, &src_ip6, IP_PROTOCOL_TCP);
637 0 : th->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b, ih6, &bogus);
638 0 : ASSERT (!bogus);
639 : }
640 :
641 86 : return 0;
642 : }
643 :
644 : /**
645 : * Send reset without reusing existing buffer
646 : *
647 : * It extracts connection info out of original packet
648 : */
649 : void
650 0 : tcp_send_reset_w_pkt (tcp_connection_t * tc, vlib_buffer_t * pkt,
651 : u32 thread_index, u8 is_ip4)
652 : {
653 0 : tcp_worker_ctx_t *wrk = tcp_get_worker (thread_index);
654 0 : vlib_main_t *vm = wrk->vm;
655 : vlib_buffer_t *b;
656 0 : u8 tcp_hdr_len, flags = 0;
657 : tcp_header_t *th, *pkt_th;
658 : u32 seq, ack, bi;
659 : ip4_header_t *ih4, *pkt_ih4;
660 : ip6_header_t *ih6, *pkt_ih6;
661 :
662 0 : if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
663 : {
664 0 : tcp_worker_stats_inc (wrk, no_buffer, 1);
665 0 : return;
666 : }
667 :
668 0 : b = vlib_get_buffer (vm, bi);
669 0 : tcp_init_buffer (vm, b);
670 :
671 : /* Make and write options */
672 0 : tcp_hdr_len = sizeof (tcp_header_t);
673 :
674 0 : if (is_ip4)
675 : {
676 0 : pkt_ih4 = vlib_buffer_get_current (pkt);
677 0 : pkt_th = ip4_next_header (pkt_ih4);
678 : }
679 : else
680 : {
681 0 : pkt_ih6 = vlib_buffer_get_current (pkt);
682 0 : pkt_th = ip6_next_header (pkt_ih6);
683 : }
684 :
685 0 : if (tcp_ack (pkt_th))
686 : {
687 0 : flags = TCP_FLAG_RST;
688 0 : seq = pkt_th->ack_number;
689 0 : ack = (tc->state >= TCP_STATE_SYN_RCVD) ? tc->rcv_nxt : 0;
690 0 : ack = clib_host_to_net_u32 (ack);
691 : }
692 : else
693 : {
694 0 : flags = TCP_FLAG_RST | TCP_FLAG_ACK;
695 0 : seq = 0;
696 0 : ack = clib_host_to_net_u32 (vnet_buffer (pkt)->tcp.seq_end);
697 : }
698 :
699 0 : th = vlib_buffer_push_tcp_net_order (b, pkt_th->dst_port, pkt_th->src_port,
700 : seq, ack, tcp_hdr_len, flags, 0);
701 :
702 : /* Swap src and dst ip */
703 0 : if (is_ip4)
704 : {
705 0 : ASSERT ((pkt_ih4->ip_version_and_header_length & 0xF0) == 0x40);
706 0 : ih4 = vlib_buffer_push_ip4 (vm, b, &pkt_ih4->dst_address,
707 : &pkt_ih4->src_address, IP_PROTOCOL_TCP,
708 0 : tcp_csum_offload (tc));
709 0 : th->checksum = ip4_tcp_udp_compute_checksum (vm, b, ih4);
710 : }
711 : else
712 : {
713 0 : int bogus = ~0;
714 0 : ASSERT ((pkt_ih6->ip_version_traffic_class_and_flow_label & 0xF0) ==
715 : 0x60);
716 0 : ih6 = vlib_buffer_push_ip6_custom (vm, b, &pkt_ih6->dst_address,
717 : &pkt_ih6->src_address,
718 : IP_PROTOCOL_TCP,
719 : tc->ipv6_flow_label);
720 0 : th->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b, ih6, &bogus);
721 0 : ASSERT (!bogus);
722 : }
723 :
724 0 : tcp_enqueue_half_open (wrk, tc, b, bi);
725 : TCP_EVT (TCP_EVT_RST_SENT, tc);
726 0 : vlib_node_increment_counter (vm, tcp_node_index (output, tc->c_is_ip4),
727 : TCP_ERROR_RST_SENT, 1);
728 : }
729 :
730 : /**
731 : * Build and set reset packet for connection
732 : */
733 : void
734 6 : tcp_send_reset (tcp_connection_t * tc)
735 : {
736 6 : tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
737 6 : vlib_main_t *vm = wrk->vm;
738 : vlib_buffer_t *b;
739 : u32 bi;
740 : tcp_header_t *th;
741 : u16 tcp_hdr_opts_len, advertise_wnd, opts_write_len;
742 : u8 flags;
743 :
744 6 : if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
745 : {
746 0 : tcp_worker_stats_inc (wrk, no_buffer, 1);
747 0 : return;
748 : }
749 6 : b = vlib_get_buffer (vm, bi);
750 6 : tcp_init_buffer (vm, b);
751 :
752 6 : tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state);
753 6 : tcp_hdr_opts_len = tc->snd_opts_len + sizeof (tcp_header_t);
754 6 : advertise_wnd = tc->rcv_wnd >> tc->rcv_wscale;
755 6 : flags = TCP_FLAG_RST | TCP_FLAG_ACK;
756 6 : th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->snd_nxt,
757 : tc->rcv_nxt, tcp_hdr_opts_len, flags,
758 : advertise_wnd);
759 6 : opts_write_len = tcp_options_write ((u8 *) (th + 1), &tc->snd_opts);
760 6 : th->checksum = tcp_compute_checksum (tc, b);
761 6 : ASSERT (opts_write_len == tc->snd_opts_len);
762 6 : vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
763 6 : tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
764 : TCP_EVT (TCP_EVT_RST_SENT, tc);
765 6 : vlib_node_increment_counter (vm, tcp_node_index (output, tc->c_is_ip4),
766 : TCP_ERROR_RST_SENT, 1);
767 : }
768 :
769 : /**
770 : * Send SYN
771 : *
772 : * Builds a SYN packet for a half-open connection and sends it to tcp-output.
773 : * The packet is handled by main thread and because half-open and established
774 : * connections use the same pool the connection can be retrieved without
775 : * additional logic.
776 : */
777 : void
778 132 : tcp_send_syn (tcp_connection_t * tc)
779 : {
780 132 : tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
781 132 : vlib_main_t *vm = wrk->vm;
782 : vlib_buffer_t *b;
783 : u32 bi;
784 :
785 : /*
786 : * Setup retransmit and establish timers before requesting buffer
787 : * such that we can return if we've ran out.
788 : */
789 132 : tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT_SYN,
790 132 : (u32) tc->rto * TCP_TO_TIMER_TICK);
791 :
792 132 : if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
793 : {
794 0 : tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT_SYN,
795 : tcp_cfg.alloc_err_timeout);
796 0 : tcp_worker_stats_inc (wrk, no_buffer, 1);
797 0 : return;
798 : }
799 :
800 132 : b = vlib_get_buffer (vm, bi);
801 132 : tcp_init_buffer (vm, b);
802 132 : tcp_make_syn (tc, b);
803 :
804 : /* Measure RTT with this */
805 132 : tc->rtt_ts = tcp_time_now_us (vlib_num_workers ()? 1 : 0);
806 132 : tc->rtt_seq = tc->snd_nxt;
807 132 : tc->rto_boff = 0;
808 :
809 132 : tcp_enqueue_half_open (wrk, tc, b, bi);
810 : TCP_EVT (TCP_EVT_SYN_SENT, tc);
811 : }
812 :
813 : void
814 135 : tcp_send_synack (tcp_connection_t * tc)
815 : {
816 135 : tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
817 135 : vlib_main_t *vm = wrk->vm;
818 : vlib_buffer_t *b;
819 : u32 bi;
820 :
821 135 : ASSERT (tc->snd_una != tc->snd_nxt);
822 135 : tcp_retransmit_timer_update (&wrk->timer_wheel, tc);
823 :
824 135 : if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
825 : {
826 0 : tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT,
827 : tcp_cfg.alloc_err_timeout);
828 0 : tcp_worker_stats_inc (wrk, no_buffer, 1);
829 0 : return;
830 : }
831 :
832 135 : tc->rtt_ts = tcp_time_now_us (tc->c_thread_index);
833 135 : b = vlib_get_buffer (vm, bi);
834 135 : tcp_init_buffer (vm, b);
835 135 : tcp_make_synack (tc, b);
836 135 : tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
837 : TCP_EVT (TCP_EVT_SYNACK_SENT, tc);
838 : }
839 :
840 : /**
841 : * Send FIN
842 : */
843 : void
844 256 : tcp_send_fin (tcp_connection_t * tc)
845 : {
846 256 : tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
847 256 : vlib_main_t *vm = wrk->vm;
848 : vlib_buffer_t *b;
849 : u32 bi;
850 256 : u8 fin_snt = 0;
851 :
852 256 : fin_snt = tc->flags & TCP_CONN_FINSNT;
853 256 : if (fin_snt)
854 0 : tc->snd_nxt -= 1;
855 :
856 256 : if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
857 : {
858 : /* Out of buffers so program fin retransmit ASAP */
859 0 : tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT,
860 : tcp_cfg.alloc_err_timeout);
861 0 : if (fin_snt)
862 0 : tc->snd_nxt += 1;
863 : else
864 : /* Make sure retransmit retries a fin not data */
865 0 : tc->flags |= TCP_CONN_FINSNT;
866 0 : tcp_worker_stats_inc (wrk, no_buffer, 1);
867 0 : return;
868 : }
869 :
870 : /* If we have non-dupacks programmed, no need to send them */
871 256 : if ((tc->flags & TCP_CONN_SNDACK) && !tc->pending_dupacks)
872 205 : tc->flags &= ~TCP_CONN_SNDACK;
873 :
874 256 : b = vlib_get_buffer (vm, bi);
875 256 : tcp_init_buffer (vm, b);
876 256 : tcp_make_fin (tc, b);
877 256 : tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
878 : TCP_EVT (TCP_EVT_FIN_SENT, tc);
879 : /* Account for the FIN */
880 256 : tc->snd_nxt += 1;
881 256 : tcp_retransmit_timer_update (&wrk->timer_wheel, tc);
882 256 : if (!fin_snt)
883 : {
884 256 : tc->flags |= TCP_CONN_FINSNT;
885 256 : tc->flags &= ~TCP_CONN_FINPNDG;
886 : }
887 : }
888 :
889 : /**
890 : * Push TCP header and update connection variables. Should only be called
891 : * for segments with data, not for 'control' packets.
892 : */
893 : always_inline void
894 996032 : tcp_push_hdr_i (tcp_connection_t * tc, vlib_buffer_t * b, u32 snd_nxt,
895 : u8 compute_opts, u8 maybe_burst, u8 update_snd_nxt)
896 : {
897 996032 : u8 tcp_hdr_opts_len, flags = TCP_FLAG_ACK;
898 : u32 advertise_wnd, data_len;
899 996032 : tcp_main_t *tm = &tcp_main;
900 : tcp_header_t *th;
901 :
902 996032 : data_len = b->current_length;
903 996032 : if (PREDICT_FALSE (b->flags & VLIB_BUFFER_NEXT_PRESENT))
904 0 : data_len += b->total_length_not_including_first_buffer;
905 :
906 996032 : vnet_buffer (b)->tcp.flags = 0;
907 996032 : vnet_buffer (b)->tcp.connection_index = tc->c_c_index;
908 :
909 996032 : if (compute_opts)
910 0 : tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state);
911 :
912 996032 : tcp_hdr_opts_len = tc->snd_opts_len + sizeof (tcp_header_t);
913 :
914 996032 : if (maybe_burst)
915 996032 : advertise_wnd = tc->rcv_wnd >> tc->rcv_wscale;
916 : else
917 0 : advertise_wnd = tcp_window_to_advertise (tc, TCP_STATE_ESTABLISHED);
918 :
919 996032 : if (PREDICT_FALSE (tc->flags & TCP_CONN_PSH_PENDING))
920 : {
921 917020 : if (seq_geq (tc->psh_seq, snd_nxt)
922 917019 : && seq_lt (tc->psh_seq, snd_nxt + data_len))
923 319 : flags |= TCP_FLAG_PSH;
924 : }
925 996032 : th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, snd_nxt,
926 : tc->rcv_nxt, tcp_hdr_opts_len, flags,
927 : advertise_wnd);
928 :
929 996032 : if (maybe_burst)
930 : {
931 996032 : clib_memcpy_fast ((u8 *) (th + 1),
932 996032 : tm->wrk_ctx[tc->c_thread_index].cached_opts,
933 996032 : tc->snd_opts_len);
934 : }
935 : else
936 : {
937 0 : u8 len = tcp_options_write ((u8 *) (th + 1), &tc->snd_opts);
938 0 : ASSERT (len == tc->snd_opts_len);
939 : }
940 :
941 : /*
942 : * Update connection variables
943 : */
944 :
945 996032 : if (update_snd_nxt)
946 996032 : tc->snd_nxt += data_len;
947 996032 : tc->rcv_las = tc->rcv_nxt;
948 :
949 996032 : tc->bytes_out += data_len;
950 996032 : tc->data_segs_out += 1;
951 :
952 996032 : th->checksum = tcp_compute_checksum (tc, b);
953 :
954 : TCP_EVT (TCP_EVT_PKTIZE, tc);
955 996032 : }
956 :
957 : always_inline u32
958 0 : tcp_buffer_len (vlib_buffer_t * b)
959 : {
960 0 : u32 data_len = b->current_length;
961 0 : if (PREDICT_FALSE (b->flags & VLIB_BUFFER_NEXT_PRESENT))
962 0 : data_len += b->total_length_not_including_first_buffer;
963 0 : return data_len;
964 : }
965 :
966 : always_inline u32
967 996032 : tcp_push_one_header (tcp_connection_t *tc, vlib_buffer_t *b)
968 : {
969 996032 : if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
970 0 : tcp_bt_track_tx (tc, tcp_buffer_len (b));
971 :
972 996032 : tcp_push_hdr_i (tc, b, tc->snd_nxt, /* compute opts */ 0, /* burst */ 1,
973 : /* update_snd_nxt */ 1);
974 :
975 996032 : tcp_validate_txf_size (tc, tc->snd_nxt - tc->snd_una);
976 996032 : return 0;
977 : }
978 :
979 : u32
980 42153 : tcp_session_push_header (transport_connection_t *tconn, vlib_buffer_t **bs,
981 : u32 n_bufs)
982 : {
983 42153 : tcp_connection_t *tc = (tcp_connection_t *) tconn;
984 :
985 495416 : while (n_bufs >= 4)
986 : {
987 453263 : vlib_prefetch_buffer_header (bs[2], STORE);
988 453263 : vlib_prefetch_buffer_header (bs[3], STORE);
989 :
990 453263 : tcp_push_one_header (tc, bs[0]);
991 453263 : tcp_push_one_header (tc, bs[1]);
992 :
993 453263 : n_bufs -= 2;
994 453263 : bs += 2;
995 : }
996 131659 : while (n_bufs)
997 : {
998 89506 : if (n_bufs > 1)
999 47353 : vlib_prefetch_buffer_header (bs[1], STORE);
1000 :
1001 89506 : tcp_push_one_header (tc, bs[0]);
1002 :
1003 89506 : n_bufs -= 1;
1004 89506 : bs += 1;
1005 : }
1006 :
1007 : /* If not tracking an ACK, start tracking */
1008 42153 : if (tc->rtt_ts == 0 && !tcp_in_cong_recovery (tc))
1009 : {
1010 21099 : tc->rtt_ts = tcp_time_now_us (tc->c_thread_index);
1011 21099 : tc->rtt_seq = tc->snd_nxt;
1012 : }
1013 42153 : if (PREDICT_FALSE (!tcp_timer_is_active (tc, TCP_TIMER_RETRANSMIT)))
1014 : {
1015 1937 : tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1016 1937 : tcp_retransmit_timer_set (&wrk->timer_wheel, tc);
1017 1937 : tc->rto_boff = 0;
1018 : }
1019 42153 : return 0;
1020 : }
1021 :
1022 : void
1023 24239 : tcp_send_ack (tcp_connection_t * tc)
1024 : {
1025 24239 : tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1026 24239 : vlib_main_t *vm = wrk->vm;
1027 : vlib_buffer_t *b;
1028 : u32 bi;
1029 :
1030 24239 : if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
1031 : {
1032 0 : tcp_update_rcv_wnd (tc);
1033 0 : tcp_worker_stats_inc (wrk, no_buffer, 1);
1034 0 : return;
1035 : }
1036 24239 : b = vlib_get_buffer (vm, bi);
1037 24239 : tcp_init_buffer (vm, b);
1038 24239 : tcp_make_ack (tc, b);
1039 24239 : tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1040 : }
1041 :
1042 : void
1043 996202 : tcp_program_ack (tcp_connection_t * tc)
1044 : {
1045 996202 : if (!(tc->flags & TCP_CONN_SNDACK))
1046 : {
1047 41252 : session_add_self_custom_tx_evt (&tc->connection, 1);
1048 41252 : tc->flags |= TCP_CONN_SNDACK;
1049 : }
1050 996202 : }
1051 :
1052 : void
1053 0 : tcp_program_dupack (tcp_connection_t * tc)
1054 : {
1055 0 : if (!(tc->flags & TCP_CONN_SNDACK))
1056 : {
1057 0 : session_add_self_custom_tx_evt (&tc->connection, 1);
1058 0 : tc->flags |= TCP_CONN_SNDACK;
1059 : }
1060 0 : if (tc->pending_dupacks < 255)
1061 0 : tc->pending_dupacks += 1;
1062 0 : }
1063 :
1064 : void
1065 0 : tcp_program_retransmit (tcp_connection_t * tc)
1066 : {
1067 0 : if (!(tc->flags & TCP_CONN_RXT_PENDING))
1068 : {
1069 0 : session_add_self_custom_tx_evt (&tc->connection, 0);
1070 0 : tc->flags |= TCP_CONN_RXT_PENDING;
1071 : }
1072 0 : }
1073 :
1074 : /**
1075 : * Send window update ack
1076 : *
1077 : * Ensures that it will be sent only once, after a zero rwnd has been
1078 : * advertised in a previous ack, and only if rwnd has grown beyond a
1079 : * configurable value.
1080 : */
1081 : void
1082 0 : tcp_send_window_update_ack (tcp_connection_t * tc)
1083 : {
1084 0 : if (tcp_zero_rwnd_sent (tc))
1085 : {
1086 0 : tcp_update_rcv_wnd (tc);
1087 0 : if (tc->rcv_wnd >= tcp_cfg.rwnd_min_update_ack * tc->snd_mss)
1088 : {
1089 0 : tcp_zero_rwnd_sent_off (tc);
1090 0 : tcp_program_ack (tc);
1091 : }
1092 : }
1093 0 : }
1094 :
1095 : /**
1096 : * Allocate a new buffer and build a new tcp segment
1097 : *
1098 : * @param wrk tcp worker
1099 : * @param tc connection for which the segment will be allocated
1100 : * @param offset offset of the first byte in the tx fifo
1101 : * @param max_deq_byte segment size
1102 : * @param[out] b pointer to buffer allocated
1103 : *
1104 : * @return the number of bytes in the segment or 0 if buffer cannot be
1105 : * allocated or no data available
1106 : */
1107 : static int
1108 0 : tcp_prepare_segment (tcp_worker_ctx_t * wrk, tcp_connection_t * tc,
1109 : u32 offset, u32 max_deq_bytes, vlib_buffer_t ** b)
1110 : {
1111 0 : u32 bytes_per_buffer = vnet_get_tcp_main ()->bytes_per_buffer;
1112 0 : vlib_main_t *vm = wrk->vm;
1113 : u32 bi, seg_size;
1114 0 : int n_bytes = 0;
1115 : u8 *data;
1116 :
1117 0 : seg_size = max_deq_bytes + TRANSPORT_MAX_HDRS_LEN;
1118 :
1119 : /*
1120 : * Prepare options
1121 : */
1122 0 : tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state);
1123 :
1124 : /*
1125 : * Allocate and fill in buffer(s)
1126 : */
1127 :
1128 : /* Easy case, buffer size greater than mss */
1129 0 : if (PREDICT_TRUE (seg_size <= bytes_per_buffer))
1130 : {
1131 0 : if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
1132 : {
1133 0 : tcp_worker_stats_inc (wrk, no_buffer, 1);
1134 0 : return 0;
1135 : }
1136 0 : *b = vlib_get_buffer (vm, bi);
1137 0 : data = tcp_init_buffer (vm, *b);
1138 0 : n_bytes = session_tx_fifo_peek_bytes (&tc->connection, data, offset,
1139 : max_deq_bytes);
1140 0 : ASSERT (n_bytes == max_deq_bytes);
1141 0 : b[0]->current_length = n_bytes;
1142 0 : tcp_push_hdr_i (tc, *b, tc->snd_una + offset, /* compute opts */ 0,
1143 : /* burst */ 0, /* update_snd_nxt */ 0);
1144 : }
1145 : /* Split mss into multiple buffers */
1146 : else
1147 : {
1148 0 : u32 chain_bi = ~0, n_bufs_per_seg, n_bufs;
1149 : u16 n_peeked, len_to_deq;
1150 : vlib_buffer_t *chain_b, *prev_b;
1151 : int i;
1152 :
1153 : /* Make sure we have enough buffers */
1154 0 : n_bufs_per_seg = ceil ((double) seg_size / bytes_per_buffer);
1155 0 : vec_validate_aligned (wrk->tx_buffers, n_bufs_per_seg - 1,
1156 : CLIB_CACHE_LINE_BYTES);
1157 0 : n_bufs = vlib_buffer_alloc (vm, wrk->tx_buffers, n_bufs_per_seg);
1158 0 : if (PREDICT_FALSE (n_bufs != n_bufs_per_seg))
1159 : {
1160 0 : if (n_bufs)
1161 0 : vlib_buffer_free (vm, wrk->tx_buffers, n_bufs);
1162 0 : tcp_worker_stats_inc (wrk, no_buffer, 1);
1163 0 : return 0;
1164 : }
1165 :
1166 0 : *b = vlib_get_buffer (vm, wrk->tx_buffers[--n_bufs]);
1167 0 : data = tcp_init_buffer (vm, *b);
1168 0 : n_bytes = session_tx_fifo_peek_bytes (&tc->connection, data, offset,
1169 : bytes_per_buffer -
1170 : TRANSPORT_MAX_HDRS_LEN);
1171 0 : b[0]->current_length = n_bytes;
1172 0 : b[0]->flags |= VLIB_BUFFER_TOTAL_LENGTH_VALID;
1173 0 : b[0]->total_length_not_including_first_buffer = 0;
1174 0 : max_deq_bytes -= n_bytes;
1175 :
1176 0 : chain_b = *b;
1177 0 : for (i = 1; i < n_bufs_per_seg; i++)
1178 : {
1179 0 : prev_b = chain_b;
1180 0 : len_to_deq = clib_min (max_deq_bytes, bytes_per_buffer);
1181 0 : chain_bi = wrk->tx_buffers[--n_bufs];
1182 0 : chain_b = vlib_get_buffer (vm, chain_bi);
1183 0 : chain_b->current_data = 0;
1184 0 : data = vlib_buffer_get_current (chain_b);
1185 0 : n_peeked = session_tx_fifo_peek_bytes (&tc->connection, data,
1186 : offset + n_bytes,
1187 : len_to_deq);
1188 0 : ASSERT (n_peeked == len_to_deq);
1189 0 : n_bytes += n_peeked;
1190 0 : chain_b->current_length = n_peeked;
1191 0 : chain_b->next_buffer = 0;
1192 :
1193 : /* update previous buffer */
1194 0 : prev_b->next_buffer = chain_bi;
1195 0 : prev_b->flags |= VLIB_BUFFER_NEXT_PRESENT;
1196 :
1197 0 : max_deq_bytes -= n_peeked;
1198 0 : b[0]->total_length_not_including_first_buffer += n_peeked;
1199 : }
1200 :
1201 0 : tcp_push_hdr_i (tc, *b, tc->snd_una + offset, /* compute opts */ 0,
1202 : /* burst */ 0, /* update_snd_nxt */ 0);
1203 :
1204 0 : if (PREDICT_FALSE (n_bufs))
1205 : {
1206 0 : clib_warning ("not all buffers consumed");
1207 0 : vlib_buffer_free (vm, wrk->tx_buffers, n_bufs);
1208 : }
1209 : }
1210 :
1211 0 : ASSERT (n_bytes > 0);
1212 0 : ASSERT (((*b)->current_data + (*b)->current_length) <= bytes_per_buffer);
1213 :
1214 0 : return n_bytes;
1215 : }
1216 :
1217 : /**
1218 : * Build a retransmit segment
1219 : *
1220 : * @return the number of bytes in the segment or 0 if there's nothing to
1221 : * retransmit
1222 : */
1223 : static u32
1224 0 : tcp_prepare_retransmit_segment (tcp_worker_ctx_t * wrk,
1225 : tcp_connection_t * tc, u32 offset,
1226 : u32 max_deq_bytes, vlib_buffer_t ** b)
1227 : {
1228 : u32 start, available_bytes;
1229 0 : int n_bytes = 0;
1230 :
1231 0 : ASSERT (tc->state >= TCP_STATE_ESTABLISHED);
1232 0 : ASSERT (max_deq_bytes != 0);
1233 :
1234 : /*
1235 : * Make sure we can retransmit something
1236 : */
1237 0 : available_bytes = transport_max_tx_dequeue (&tc->connection);
1238 0 : ASSERT (available_bytes >= offset);
1239 0 : available_bytes -= offset;
1240 0 : if (!available_bytes)
1241 0 : return 0;
1242 :
1243 0 : max_deq_bytes = clib_min (tc->snd_mss, max_deq_bytes);
1244 0 : max_deq_bytes = clib_min (available_bytes, max_deq_bytes);
1245 :
1246 0 : start = tc->snd_una + offset;
1247 0 : ASSERT (seq_leq (start + max_deq_bytes, tc->snd_nxt));
1248 :
1249 0 : n_bytes = tcp_prepare_segment (wrk, tc, offset, max_deq_bytes, b);
1250 0 : if (!n_bytes)
1251 0 : return 0;
1252 :
1253 0 : tc->snd_rxt_bytes += n_bytes;
1254 :
1255 0 : if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
1256 0 : tcp_bt_track_rxt (tc, start, start + n_bytes);
1257 :
1258 0 : tc->bytes_retrans += n_bytes;
1259 0 : tc->segs_retrans += 1;
1260 0 : tcp_worker_stats_inc (wrk, rxt_segs, 1);
1261 : TCP_EVT (TCP_EVT_CC_RTX, tc, offset, n_bytes);
1262 :
1263 0 : return n_bytes;
1264 : }
1265 :
1266 : static void
1267 0 : tcp_check_sack_reneging (tcp_connection_t * tc)
1268 : {
1269 0 : sack_scoreboard_t *sb = &tc->sack_sb;
1270 : sack_scoreboard_hole_t *hole;
1271 :
1272 0 : hole = scoreboard_first_hole (sb);
1273 0 : if (!sb->is_reneging && (!hole || hole->start == tc->snd_una))
1274 0 : return;
1275 :
1276 0 : scoreboard_clear_reneging (sb, tc->snd_una, tc->snd_nxt);
1277 : }
1278 :
1279 : /**
1280 : * Reset congestion control, switch cwnd to loss window and try again.
1281 : */
1282 : static void
1283 0 : tcp_cc_init_rxt_timeout (tcp_connection_t * tc)
1284 : {
1285 : TCP_EVT (TCP_EVT_CC_EVT, tc, 6);
1286 :
1287 0 : tc->prev_ssthresh = tc->ssthresh;
1288 0 : tc->prev_cwnd = tc->cwnd;
1289 :
1290 : /* If we entrered loss without fast recovery, notify cc algo of the
1291 : * congestion event such that it can update ssthresh and its state */
1292 0 : if (!tcp_in_fastrecovery (tc))
1293 0 : tcp_cc_congestion (tc);
1294 :
1295 : /* Let cc algo decide loss cwnd and ssthresh post unrecovered loss */
1296 0 : tcp_cc_loss (tc);
1297 :
1298 0 : tc->rtt_ts = 0;
1299 0 : tc->cwnd_acc_bytes = 0;
1300 0 : tc->tr_occurences += 1;
1301 0 : tc->sack_sb.reorder = TCP_DUPACK_THRESHOLD;
1302 0 : tcp_recovery_on (tc);
1303 0 : }
1304 :
1305 : void
1306 12 : tcp_timer_retransmit_handler (tcp_connection_t * tc)
1307 : {
1308 12 : tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1309 12 : vlib_main_t *vm = wrk->vm;
1310 12 : vlib_buffer_t *b = 0;
1311 : u32 bi, n_bytes;
1312 :
1313 12 : tcp_worker_stats_inc (wrk, tr_events, 1);
1314 :
1315 : /* Should be handled by a different handler */
1316 12 : if (PREDICT_FALSE (tc->state == TCP_STATE_SYN_SENT))
1317 0 : return;
1318 :
1319 : /* Wait-close and retransmit could pop at the same time */
1320 12 : if (tc->state == TCP_STATE_CLOSED)
1321 0 : return;
1322 :
1323 12 : if (tc->state >= TCP_STATE_ESTABLISHED)
1324 : {
1325 : TCP_EVT (TCP_EVT_CC_EVT, tc, 2);
1326 :
1327 : /* Lost FIN, retransmit and return */
1328 0 : if (tc->flags & TCP_CONN_FINSNT)
1329 : {
1330 0 : tcp_send_fin (tc);
1331 0 : tc->rto_boff += 1;
1332 0 : tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1333 0 : return;
1334 : }
1335 :
1336 : /* Shouldn't be here */
1337 0 : if (tc->snd_una == tc->snd_nxt)
1338 : {
1339 0 : ASSERT (!tcp_in_recovery (tc));
1340 0 : tc->rto_boff = 0;
1341 0 : return;
1342 : }
1343 :
1344 : /* We're not in recovery so make sure rto_boff is 0. Can be non 0 due
1345 : * to persist timer timeout */
1346 0 : if (!tcp_in_recovery (tc) && tc->rto_boff > 0)
1347 : {
1348 0 : tc->rto_boff = 0;
1349 0 : tcp_update_rto (tc);
1350 : }
1351 :
1352 : /* Peer is dead or network connectivity is lost. Close connection.
1353 : * RFC 1122 section 4.2.3.5 recommends a value of at least 100s. For
1354 : * a min rto of 0.2s we need to retry about 8 times. */
1355 0 : if (tc->rto_boff >= TCP_RTO_BOFF_MAX)
1356 : {
1357 0 : tcp_send_reset (tc);
1358 0 : tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1359 0 : session_transport_closing_notify (&tc->connection);
1360 0 : session_transport_closed_notify (&tc->connection);
1361 0 : tcp_connection_timers_reset (tc);
1362 0 : tcp_program_cleanup (wrk, tc);
1363 0 : tcp_worker_stats_inc (wrk, tr_abort, 1);
1364 0 : return;
1365 : }
1366 :
1367 0 : if (tcp_opts_sack_permitted (&tc->rcv_opts))
1368 : {
1369 0 : tcp_check_sack_reneging (tc);
1370 0 : scoreboard_rxt_mark_lost (&tc->sack_sb, tc->snd_una, tc->snd_nxt);
1371 : }
1372 :
1373 : /* Update send congestion to make sure that rxt has data to send */
1374 0 : tc->snd_congestion = tc->snd_nxt;
1375 :
1376 : /* Send the first unacked segment. If we're short on buffers, return
1377 : * as soon as possible */
1378 0 : n_bytes = clib_min (tc->snd_mss, tc->snd_nxt - tc->snd_una);
1379 0 : n_bytes = tcp_prepare_retransmit_segment (wrk, tc, 0, n_bytes, &b);
1380 0 : if (!n_bytes)
1381 : {
1382 0 : tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT,
1383 : tcp_cfg.alloc_err_timeout);
1384 0 : return;
1385 : }
1386 :
1387 0 : bi = vlib_get_buffer_index (vm, b);
1388 0 : tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1389 :
1390 0 : tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1391 0 : tcp_retransmit_timer_update (&wrk->timer_wheel, tc);
1392 :
1393 0 : tc->rto_boff += 1;
1394 0 : if (tc->rto_boff == 1)
1395 : {
1396 0 : tcp_cc_init_rxt_timeout (tc);
1397 : /* Record timestamp. Eifel detection algorithm RFC3522 */
1398 0 : tc->snd_rxt_ts = tcp_tstamp (tc);
1399 : }
1400 :
1401 0 : if (tcp_opts_sack_permitted (&tc->rcv_opts))
1402 0 : scoreboard_init_rxt (&tc->sack_sb, tc->snd_una + n_bytes);
1403 :
1404 0 : tcp_program_retransmit (tc);
1405 : }
1406 : /* Retransmit SYN-ACK */
1407 12 : else if (tc->state == TCP_STATE_SYN_RCVD)
1408 : {
1409 : TCP_EVT (TCP_EVT_CC_EVT, tc, 2);
1410 :
1411 12 : tc->rtt_ts = 0;
1412 :
1413 : /* Passive open establish timeout */
1414 12 : if (tc->rto > TCP_ESTABLISH_TIME >> 1)
1415 : {
1416 0 : tcp_connection_set_state (tc, TCP_STATE_CLOSED);
1417 0 : tcp_connection_timers_reset (tc);
1418 0 : tcp_program_cleanup (wrk, tc);
1419 0 : tcp_worker_stats_inc (wrk, tr_abort, 1);
1420 0 : return;
1421 : }
1422 :
1423 12 : if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
1424 : {
1425 0 : tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT,
1426 : tcp_cfg.alloc_err_timeout);
1427 0 : tcp_worker_stats_inc (wrk, no_buffer, 1);
1428 0 : return;
1429 : }
1430 :
1431 12 : tc->rto_boff += 1;
1432 12 : if (tc->rto_boff > TCP_RTO_SYN_RETRIES)
1433 3 : tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1434 :
1435 12 : ASSERT (tc->snd_una != tc->snd_nxt);
1436 12 : tcp_retransmit_timer_update (&wrk->timer_wheel, tc);
1437 :
1438 12 : b = vlib_get_buffer (vm, bi);
1439 12 : tcp_init_buffer (vm, b);
1440 12 : tcp_make_synack (tc, b);
1441 : TCP_EVT (TCP_EVT_SYN_RXT, tc, 1);
1442 :
1443 : /* Retransmit timer already updated, just enqueue to output */
1444 12 : tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1445 : }
1446 : else
1447 : {
1448 0 : ASSERT (tc->state == TCP_STATE_CLOSED);
1449 0 : return;
1450 : }
1451 : }
1452 :
1453 : /**
1454 : * SYN retransmit timer handler. Active open only.
1455 : */
1456 : void
1457 0 : tcp_timer_retransmit_syn_handler (tcp_connection_t * tc)
1458 : {
1459 0 : tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1460 0 : vlib_main_t *vm = wrk->vm;
1461 0 : vlib_buffer_t *b = 0;
1462 : u32 bi;
1463 :
1464 : /* Note: the connection may have transitioned to ESTABLISHED... */
1465 0 : if (PREDICT_FALSE (tc->state != TCP_STATE_SYN_SENT))
1466 0 : return;
1467 :
1468 : /* Half-open connection actually moved to established but we were
1469 : * waiting for syn retransmit to pop to call cleanup from the right
1470 : * thread. */
1471 0 : if (tc->flags & TCP_CONN_HALF_OPEN_DONE)
1472 : {
1473 0 : if (tcp_half_open_connection_cleanup (tc))
1474 : TCP_DBG ("could not remove half-open connection");
1475 0 : return;
1476 : }
1477 :
1478 : TCP_EVT (TCP_EVT_CC_EVT, tc, 2);
1479 0 : tc->rtt_ts = 0;
1480 :
1481 : /* Active open establish timeout */
1482 0 : if (tc->rto >= TCP_ESTABLISH_TIME >> 1)
1483 : {
1484 0 : session_stream_connect_notify (&tc->connection, SESSION_E_TIMEDOUT);
1485 0 : tcp_connection_cleanup (tc);
1486 0 : return;
1487 : }
1488 :
1489 0 : if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
1490 : {
1491 0 : tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT_SYN,
1492 : tcp_cfg.alloc_err_timeout);
1493 0 : tcp_worker_stats_inc (wrk, no_buffer, 1);
1494 0 : return;
1495 : }
1496 :
1497 : /* Try without increasing RTO a number of times. If this fails,
1498 : * start growing RTO exponentially */
1499 0 : tc->rto_boff += 1;
1500 0 : if (tc->rto_boff > TCP_RTO_SYN_RETRIES)
1501 0 : tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1502 :
1503 0 : b = vlib_get_buffer (vm, bi);
1504 0 : tcp_init_buffer (vm, b);
1505 0 : tcp_make_syn (tc, b);
1506 :
1507 : TCP_EVT (TCP_EVT_SYN_RXT, tc, 0);
1508 :
1509 0 : tcp_enqueue_half_open (wrk, tc, b, bi);
1510 :
1511 0 : tcp_timer_update (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT_SYN,
1512 0 : (u32) tc->rto * TCP_TO_TIMER_TICK);
1513 : }
1514 :
1515 : /**
1516 : * Got 0 snd_wnd from peer, try to do something about it.
1517 : *
1518 : */
1519 : void
1520 0 : tcp_timer_persist_handler (tcp_connection_t * tc)
1521 : {
1522 0 : tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index);
1523 : u32 bi, max_snd_bytes, available_bytes, offset;
1524 0 : tcp_main_t *tm = vnet_get_tcp_main ();
1525 0 : vlib_main_t *vm = wrk->vm;
1526 : vlib_buffer_t *b;
1527 0 : int n_bytes = 0;
1528 : u8 *data;
1529 :
1530 : /* Problem already solved or worse */
1531 0 : if (tc->state == TCP_STATE_CLOSED || tc->snd_wnd > tc->snd_mss
1532 0 : || (tc->flags & TCP_CONN_FINSNT))
1533 0 : goto update_scheduler;
1534 :
1535 0 : available_bytes = transport_max_tx_dequeue (&tc->connection);
1536 0 : offset = tc->snd_nxt - tc->snd_una;
1537 :
1538 : /* Reprogram persist if no new bytes available to send. We may have data
1539 : * next time */
1540 0 : if (!available_bytes)
1541 : {
1542 0 : tcp_persist_timer_set (&wrk->timer_wheel, tc);
1543 0 : return;
1544 : }
1545 :
1546 0 : if (available_bytes <= offset)
1547 0 : goto update_scheduler;
1548 :
1549 : /* Increment RTO backoff */
1550 0 : tc->rto_boff += 1;
1551 0 : tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX);
1552 :
1553 : /*
1554 : * Try to force the first unsent segment (or buffer)
1555 : */
1556 0 : if (PREDICT_FALSE (!vlib_buffer_alloc (vm, &bi, 1)))
1557 : {
1558 0 : tcp_persist_timer_set (&wrk->timer_wheel, tc);
1559 0 : tcp_worker_stats_inc (wrk, no_buffer, 1);
1560 0 : return;
1561 : }
1562 :
1563 0 : b = vlib_get_buffer (vm, bi);
1564 0 : data = tcp_init_buffer (vm, b);
1565 :
1566 0 : tcp_validate_txf_size (tc, offset);
1567 0 : tc->snd_opts_len = tcp_make_options (tc, &tc->snd_opts, tc->state);
1568 0 : max_snd_bytes = clib_min (clib_min (tc->snd_mss, available_bytes),
1569 : tm->bytes_per_buffer - TRANSPORT_MAX_HDRS_LEN);
1570 0 : if (tc->snd_wnd > 0)
1571 0 : max_snd_bytes = clib_min (tc->snd_wnd, max_snd_bytes);
1572 0 : n_bytes = session_tx_fifo_peek_bytes (&tc->connection, data, offset,
1573 : max_snd_bytes);
1574 0 : b->current_length = n_bytes;
1575 0 : ASSERT (n_bytes != 0 && (tcp_timer_is_active (tc, TCP_TIMER_RETRANSMIT)
1576 : || tc->snd_una == tc->snd_nxt
1577 : || tc->rto_boff > 1));
1578 :
1579 0 : if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
1580 : {
1581 0 : tcp_bt_check_app_limited (tc);
1582 0 : tcp_bt_track_tx (tc, n_bytes);
1583 : }
1584 :
1585 0 : tcp_push_hdr_i (tc, b, tc->snd_nxt, /* compute opts */ 0,
1586 : /* burst */ 0, /* update_snd_nxt */ 1);
1587 0 : tcp_validate_txf_size (tc, tc->snd_nxt - tc->snd_una);
1588 0 : tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1589 :
1590 : /* Just sent new data, enable retransmit */
1591 0 : tcp_retransmit_timer_update (&wrk->timer_wheel, tc);
1592 :
1593 0 : return;
1594 :
1595 0 : update_scheduler:
1596 :
1597 0 : if (tcp_is_descheduled (tc))
1598 0 : transport_connection_reschedule (&tc->connection);
1599 : }
1600 :
1601 : /**
1602 : * Retransmit first unacked segment
1603 : */
1604 : int
1605 0 : tcp_retransmit_first_unacked (tcp_worker_ctx_t * wrk, tcp_connection_t * tc)
1606 : {
1607 0 : vlib_main_t *vm = wrk->vm;
1608 : vlib_buffer_t *b;
1609 : u32 bi, n_bytes;
1610 :
1611 : TCP_EVT (TCP_EVT_CC_EVT, tc, 1);
1612 :
1613 0 : n_bytes = tcp_prepare_retransmit_segment (wrk, tc, 0, tc->snd_mss, &b);
1614 0 : if (!n_bytes)
1615 0 : return -1;
1616 :
1617 0 : bi = vlib_get_buffer_index (vm, b);
1618 0 : tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1619 :
1620 0 : return 0;
1621 : }
1622 :
1623 : static int
1624 0 : tcp_transmit_unsent (tcp_worker_ctx_t * wrk, tcp_connection_t * tc,
1625 : u32 burst_size)
1626 : {
1627 0 : u32 offset, n_segs = 0, n_written, bi, available_wnd;
1628 0 : vlib_main_t *vm = wrk->vm;
1629 0 : vlib_buffer_t *b = 0;
1630 :
1631 0 : offset = tc->snd_nxt - tc->snd_una;
1632 0 : available_wnd = tc->snd_wnd - offset;
1633 0 : burst_size = clib_min (burst_size, available_wnd / tc->snd_mss);
1634 :
1635 0 : if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
1636 0 : tcp_bt_check_app_limited (tc);
1637 :
1638 0 : while (n_segs < burst_size)
1639 : {
1640 0 : n_written = tcp_prepare_segment (wrk, tc, offset, tc->snd_mss, &b);
1641 0 : if (!n_written)
1642 0 : goto done;
1643 :
1644 0 : bi = vlib_get_buffer_index (vm, b);
1645 0 : tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1646 0 : offset += n_written;
1647 0 : n_segs += 1;
1648 :
1649 0 : if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE)
1650 0 : tcp_bt_track_tx (tc, n_written);
1651 :
1652 0 : tc->snd_nxt += n_written;
1653 : }
1654 :
1655 0 : done:
1656 0 : return n_segs;
1657 : }
1658 :
1659 : /**
1660 : * Estimate send space using proportional rate reduction (RFC6937)
1661 : */
1662 : int
1663 0 : tcp_fastrecovery_prr_snd_space (tcp_connection_t * tc)
1664 : {
1665 : u32 pipe, prr_out;
1666 : int space;
1667 :
1668 0 : pipe = tcp_flight_size (tc);
1669 0 : prr_out = tc->snd_rxt_bytes + (tc->snd_nxt - tc->snd_congestion);
1670 :
1671 0 : if (pipe > tc->ssthresh)
1672 : {
1673 0 : space = ((int) tc->prr_delivered * ((f64) tc->ssthresh / tc->prev_cwnd))
1674 0 : - prr_out;
1675 : }
1676 : else
1677 : {
1678 : int limit;
1679 0 : limit = clib_max ((int) (tc->prr_delivered - prr_out), 0) + tc->snd_mss;
1680 0 : space = clib_min (tc->ssthresh - pipe, limit);
1681 : }
1682 0 : space = clib_max (space, prr_out ? 0 : tc->snd_mss);
1683 0 : return space;
1684 : }
1685 :
1686 : static inline u8
1687 0 : tcp_retransmit_should_retry_head (tcp_connection_t * tc,
1688 : sack_scoreboard_t * sb)
1689 : {
1690 0 : u32 tx_adv_sack = sb->high_sacked - tc->snd_congestion;
1691 0 : f64 rr = (f64) tc->ssthresh / tc->prev_cwnd;
1692 :
1693 0 : if (tcp_fastrecovery_first (tc))
1694 0 : return 1;
1695 :
1696 0 : return (tx_adv_sack > (tc->snd_una - tc->prr_start) * rr);
1697 : }
1698 :
1699 : static inline u8
1700 41047 : tcp_max_tx_deq (tcp_connection_t * tc)
1701 : {
1702 41047 : return (transport_max_tx_dequeue (&tc->connection)
1703 41047 : - (tc->snd_nxt - tc->snd_una));
1704 : }
1705 :
1706 : #define scoreboard_rescue_rxt_valid(_sb, _tc) \
1707 : (seq_geq (_sb->rescue_rxt, _tc->snd_una) \
1708 : && seq_leq (_sb->rescue_rxt, _tc->snd_congestion))
1709 :
1710 : /**
1711 : * Do retransmit with SACKs
1712 : */
1713 : static int
1714 0 : tcp_retransmit_sack (tcp_worker_ctx_t * wrk, tcp_connection_t * tc,
1715 : u32 burst_size)
1716 : {
1717 0 : u32 n_written = 0, offset, max_bytes, n_segs = 0;
1718 0 : u8 snd_limited = 0, can_rescue = 0;
1719 : u32 bi, max_deq, burst_bytes;
1720 : sack_scoreboard_hole_t *hole;
1721 0 : vlib_main_t *vm = wrk->vm;
1722 0 : vlib_buffer_t *b = 0;
1723 : sack_scoreboard_t *sb;
1724 : int snd_space;
1725 :
1726 0 : ASSERT (tcp_in_cong_recovery (tc));
1727 :
1728 0 : burst_bytes = transport_connection_tx_pacer_burst (&tc->connection);
1729 0 : burst_size = clib_min (burst_size, burst_bytes / tc->snd_mss);
1730 0 : if (!burst_size)
1731 : {
1732 0 : tcp_program_retransmit (tc);
1733 0 : return 0;
1734 : }
1735 :
1736 0 : if (tcp_in_recovery (tc))
1737 0 : snd_space = tcp_available_cc_snd_space (tc);
1738 : else
1739 0 : snd_space = tcp_fastrecovery_prr_snd_space (tc);
1740 :
1741 0 : if (snd_space < tc->snd_mss)
1742 0 : goto done;
1743 :
1744 0 : sb = &tc->sack_sb;
1745 :
1746 : /* Check if snd_una is a lost retransmit */
1747 0 : if (pool_elts (sb->holes)
1748 0 : && seq_gt (sb->high_sacked, tc->snd_congestion)
1749 0 : && tc->rxt_head != tc->snd_una
1750 0 : && tcp_retransmit_should_retry_head (tc, sb))
1751 : {
1752 0 : max_bytes = clib_min (tc->snd_mss, tc->snd_nxt - tc->snd_una);
1753 0 : n_written = tcp_prepare_retransmit_segment (wrk, tc, 0, max_bytes, &b);
1754 0 : if (!n_written)
1755 : {
1756 0 : tcp_program_retransmit (tc);
1757 0 : goto done;
1758 : }
1759 0 : bi = vlib_get_buffer_index (vm, b);
1760 0 : tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1761 0 : n_segs = 1;
1762 :
1763 0 : tc->rxt_head = tc->snd_una;
1764 0 : tc->rxt_delivered += n_written;
1765 0 : tc->prr_delivered += n_written;
1766 0 : ASSERT (tc->rxt_delivered <= tc->snd_rxt_bytes);
1767 : }
1768 :
1769 0 : tcp_fastrecovery_first_off (tc);
1770 :
1771 : TCP_EVT (TCP_EVT_CC_EVT, tc, 0);
1772 0 : hole = scoreboard_get_hole (sb, sb->cur_rxt_hole);
1773 :
1774 0 : max_deq = transport_max_tx_dequeue (&tc->connection);
1775 0 : max_deq -= tc->snd_nxt - tc->snd_una;
1776 :
1777 0 : while (snd_space > 0 && n_segs < burst_size)
1778 : {
1779 0 : hole = scoreboard_next_rxt_hole (sb, hole, max_deq != 0, &can_rescue,
1780 : &snd_limited);
1781 0 : if (!hole)
1782 : {
1783 : /* We are out of lost holes to retransmit so send some new data. */
1784 0 : if (max_deq > tc->snd_mss)
1785 : {
1786 : u32 n_segs_new;
1787 : int av_wnd;
1788 :
1789 : /* Make sure we don't exceed available window and leave space
1790 : * for one more packet, to avoid zero window acks */
1791 0 : av_wnd = (int) tc->snd_wnd - (tc->snd_nxt - tc->snd_una);
1792 0 : av_wnd = clib_max (av_wnd - tc->snd_mss, 0);
1793 0 : snd_space = clib_min (snd_space, av_wnd);
1794 0 : snd_space = clib_min (max_deq, snd_space);
1795 0 : burst_size = clib_min (burst_size - n_segs,
1796 : snd_space / tc->snd_mss);
1797 0 : burst_size = clib_min (burst_size, TCP_RXT_MAX_BURST);
1798 0 : n_segs_new = tcp_transmit_unsent (wrk, tc, burst_size);
1799 0 : if (max_deq > n_segs_new * tc->snd_mss)
1800 0 : tcp_program_retransmit (tc);
1801 :
1802 0 : n_segs += n_segs_new;
1803 0 : goto done;
1804 : }
1805 :
1806 0 : if (tcp_in_recovery (tc) || !can_rescue
1807 0 : || scoreboard_rescue_rxt_valid (sb, tc))
1808 : break;
1809 :
1810 : /* If rescue rxt undefined or less than snd_una then one segment of
1811 : * up to SMSS octets that MUST include the highest outstanding
1812 : * unSACKed sequence number SHOULD be returned, and RescueRxt set to
1813 : * RecoveryPoint. HighRxt MUST NOT be updated.
1814 : */
1815 0 : hole = scoreboard_last_hole (sb);
1816 0 : max_bytes = clib_min (tc->snd_mss, hole->end - hole->start);
1817 0 : max_bytes = clib_min (max_bytes, snd_space);
1818 0 : offset = hole->end - tc->snd_una - max_bytes;
1819 0 : n_written = tcp_prepare_retransmit_segment (wrk, tc, offset,
1820 : max_bytes, &b);
1821 0 : if (!n_written)
1822 0 : goto done;
1823 :
1824 0 : sb->rescue_rxt = tc->snd_congestion;
1825 0 : bi = vlib_get_buffer_index (vm, b);
1826 0 : tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1827 0 : n_segs += 1;
1828 0 : break;
1829 : }
1830 :
1831 0 : max_bytes = clib_min (hole->end - sb->high_rxt, snd_space);
1832 0 : max_bytes = snd_limited ? clib_min (max_bytes, tc->snd_mss) : max_bytes;
1833 0 : if (max_bytes == 0)
1834 0 : break;
1835 :
1836 0 : offset = sb->high_rxt - tc->snd_una;
1837 0 : n_written = tcp_prepare_retransmit_segment (wrk, tc, offset, max_bytes,
1838 : &b);
1839 0 : ASSERT (n_written <= snd_space);
1840 :
1841 : /* Nothing left to retransmit */
1842 0 : if (n_written == 0)
1843 0 : break;
1844 :
1845 0 : bi = vlib_get_buffer_index (vm, b);
1846 0 : tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1847 :
1848 0 : sb->high_rxt += n_written;
1849 0 : ASSERT (seq_leq (sb->high_rxt, tc->snd_nxt));
1850 :
1851 0 : snd_space -= n_written;
1852 0 : n_segs += 1;
1853 : }
1854 :
1855 0 : if (hole)
1856 0 : tcp_program_retransmit (tc);
1857 :
1858 0 : done:
1859 :
1860 0 : transport_connection_tx_pacer_reset_bucket (&tc->connection, 0);
1861 0 : return n_segs;
1862 : }
1863 :
1864 : /**
1865 : * Fast retransmit without SACK info
1866 : */
1867 : static int
1868 0 : tcp_retransmit_no_sack (tcp_worker_ctx_t * wrk, tcp_connection_t * tc,
1869 : u32 burst_size)
1870 : {
1871 0 : u32 n_written = 0, offset = 0, bi, max_deq, n_segs_now, max_bytes;
1872 : u32 burst_bytes, sent_bytes;
1873 0 : vlib_main_t *vm = wrk->vm;
1874 0 : int snd_space, n_segs = 0;
1875 0 : u8 cc_limited = 0;
1876 : vlib_buffer_t *b;
1877 :
1878 0 : ASSERT (tcp_in_cong_recovery (tc));
1879 : TCP_EVT (TCP_EVT_CC_EVT, tc, 0);
1880 :
1881 0 : burst_bytes = transport_connection_tx_pacer_burst (&tc->connection);
1882 0 : burst_size = clib_min (burst_size, burst_bytes / tc->snd_mss);
1883 0 : if (!burst_size)
1884 : {
1885 0 : tcp_program_retransmit (tc);
1886 0 : return 0;
1887 : }
1888 :
1889 0 : snd_space = tcp_available_cc_snd_space (tc);
1890 0 : cc_limited = snd_space < burst_bytes;
1891 :
1892 0 : if (!tcp_fastrecovery_first (tc))
1893 0 : goto send_unsent;
1894 :
1895 : /* RFC 6582: [If a partial ack], retransmit the first unacknowledged
1896 : * segment. */
1897 0 : while (snd_space > 0 && n_segs < burst_size)
1898 : {
1899 0 : max_bytes = clib_min (tc->snd_mss,
1900 : tc->snd_congestion - tc->snd_una - offset);
1901 0 : if (!max_bytes)
1902 0 : break;
1903 0 : n_written = tcp_prepare_retransmit_segment (wrk, tc, offset, max_bytes,
1904 : &b);
1905 :
1906 : /* Nothing left to retransmit */
1907 0 : if (n_written == 0)
1908 0 : break;
1909 :
1910 0 : bi = vlib_get_buffer_index (vm, b);
1911 0 : tcp_enqueue_to_output (wrk, b, bi, tc->c_is_ip4);
1912 0 : snd_space -= n_written;
1913 0 : offset += n_written;
1914 0 : n_segs += 1;
1915 : }
1916 :
1917 0 : if (n_segs == burst_size)
1918 0 : goto done;
1919 :
1920 0 : send_unsent:
1921 :
1922 : /* RFC 6582: Send a new segment if permitted by the new value of cwnd. */
1923 0 : if (snd_space < tc->snd_mss || tc->snd_mss == 0)
1924 0 : goto done;
1925 :
1926 0 : max_deq = transport_max_tx_dequeue (&tc->connection);
1927 0 : max_deq -= tc->snd_nxt - tc->snd_una;
1928 0 : if (max_deq)
1929 : {
1930 0 : snd_space = clib_min (max_deq, snd_space);
1931 0 : burst_size = clib_min (burst_size - n_segs, snd_space / tc->snd_mss);
1932 0 : n_segs_now = tcp_transmit_unsent (wrk, tc, burst_size);
1933 0 : if (n_segs_now && max_deq > n_segs_now * tc->snd_mss)
1934 0 : tcp_program_retransmit (tc);
1935 0 : n_segs += n_segs_now;
1936 : }
1937 :
1938 0 : done:
1939 0 : tcp_fastrecovery_first_off (tc);
1940 :
1941 0 : sent_bytes = clib_min (n_segs * tc->snd_mss, burst_bytes);
1942 0 : sent_bytes = cc_limited ? burst_bytes : sent_bytes;
1943 0 : transport_connection_tx_pacer_update_bytes (&tc->connection, sent_bytes);
1944 :
1945 0 : return n_segs;
1946 : }
1947 :
1948 : static int
1949 41047 : tcp_send_acks (tcp_connection_t * tc, u32 max_burst_size)
1950 : {
1951 : int j, n_acks;
1952 :
1953 41047 : if (!tc->pending_dupacks)
1954 : {
1955 41047 : if (tcp_in_cong_recovery (tc) || !tcp_max_tx_deq (tc)
1956 16943 : || tc->state != TCP_STATE_ESTABLISHED)
1957 : {
1958 24107 : tcp_send_ack (tc);
1959 24107 : return 1;
1960 : }
1961 16940 : return 0;
1962 : }
1963 :
1964 : /* If we're supposed to send dupacks but have no ooo data
1965 : * send only one ack */
1966 0 : if (!vec_len (tc->snd_sacks))
1967 : {
1968 0 : tcp_send_ack (tc);
1969 0 : tc->dupacks_out += 1;
1970 0 : tc->pending_dupacks = 0;
1971 0 : return 1;
1972 : }
1973 :
1974 : /* Start with first sack block */
1975 0 : tc->snd_sack_pos = 0;
1976 :
1977 : /* Generate enough dupacks to cover all sack blocks. Do not generate
1978 : * more sacks than the number of packets received. But do generate at
1979 : * least 3, i.e., the number needed to signal congestion, if needed. */
1980 0 : n_acks = vec_len (tc->snd_sacks) / TCP_OPTS_MAX_SACK_BLOCKS;
1981 0 : n_acks = clib_min (n_acks, tc->pending_dupacks);
1982 0 : n_acks = clib_max (n_acks, clib_min (tc->pending_dupacks, 3));
1983 0 : for (j = 0; j < clib_min (n_acks, max_burst_size); j++)
1984 0 : tcp_send_ack (tc);
1985 :
1986 0 : if (n_acks < max_burst_size)
1987 : {
1988 0 : tc->pending_dupacks = 0;
1989 0 : tc->snd_sack_pos = 0;
1990 0 : tc->dupacks_out += n_acks;
1991 0 : return n_acks;
1992 : }
1993 : else
1994 : {
1995 : TCP_DBG ("constrained by burst size");
1996 0 : tc->pending_dupacks = n_acks - max_burst_size;
1997 0 : tc->dupacks_out += max_burst_size;
1998 0 : tcp_program_dupack (tc);
1999 0 : return max_burst_size;
2000 : }
2001 : }
2002 :
2003 : static int
2004 0 : tcp_do_retransmit (tcp_connection_t * tc, u32 max_burst_size)
2005 : {
2006 : tcp_worker_ctx_t *wrk;
2007 : u32 n_segs;
2008 :
2009 0 : if (PREDICT_FALSE (tc->state == TCP_STATE_CLOSED))
2010 0 : return 0;
2011 :
2012 0 : wrk = tcp_get_worker (tc->c_thread_index);
2013 :
2014 0 : if (tcp_opts_sack_permitted (&tc->rcv_opts))
2015 0 : n_segs = tcp_retransmit_sack (wrk, tc, max_burst_size);
2016 : else
2017 0 : n_segs = tcp_retransmit_no_sack (wrk, tc, max_burst_size);
2018 :
2019 0 : return n_segs;
2020 : }
2021 :
2022 : int
2023 41252 : tcp_session_custom_tx (void *conn, transport_send_params_t * sp)
2024 : {
2025 41252 : tcp_connection_t *tc = (tcp_connection_t *) conn;
2026 41252 : u32 n_segs = 0;
2027 :
2028 41252 : if (tcp_in_cong_recovery (tc) && (tc->flags & TCP_CONN_RXT_PENDING))
2029 : {
2030 0 : tc->flags &= ~TCP_CONN_RXT_PENDING;
2031 0 : n_segs = tcp_do_retransmit (tc, sp->max_burst_size);
2032 : }
2033 :
2034 41252 : if (!(tc->flags & TCP_CONN_SNDACK))
2035 205 : return n_segs;
2036 :
2037 41047 : tc->flags &= ~TCP_CONN_SNDACK;
2038 :
2039 : /* We have retransmitted packets and no dupack */
2040 41047 : if (n_segs && !tc->pending_dupacks)
2041 0 : return n_segs;
2042 :
2043 41047 : if (sp->max_burst_size <= n_segs)
2044 : {
2045 0 : tcp_program_ack (tc);
2046 0 : return n_segs;
2047 : }
2048 :
2049 41047 : n_segs += tcp_send_acks (tc, sp->max_burst_size - n_segs);
2050 :
2051 41047 : return n_segs;
2052 : }
2053 : #endif /* CLIB_MARCH_VARIANT */
2054 :
2055 : static void
2056 0 : tcp_output_handle_link_local (tcp_connection_t * tc0, vlib_buffer_t * b0,
2057 : u16 * next0, u32 * error0)
2058 : {
2059 : ip_adjacency_t *adj;
2060 : adj_index_t ai;
2061 :
2062 : /* Not thread safe but as long as the connection exists the adj should
2063 : * not be removed */
2064 0 : ai = adj_nbr_find (FIB_PROTOCOL_IP6, VNET_LINK_IP6, &tc0->c_rmt_ip,
2065 : tc0->sw_if_index);
2066 0 : if (ai == ADJ_INDEX_INVALID)
2067 : {
2068 0 : vnet_buffer (b0)->sw_if_index[VLIB_TX] = ~0;
2069 0 : *next0 = TCP_OUTPUT_NEXT_DROP;
2070 0 : *error0 = TCP_ERROR_LINK_LOCAL_RW;
2071 0 : return;
2072 : }
2073 :
2074 0 : adj = adj_get (ai);
2075 0 : if (PREDICT_TRUE (adj->lookup_next_index == IP_LOOKUP_NEXT_REWRITE))
2076 0 : *next0 = TCP_OUTPUT_NEXT_IP_REWRITE;
2077 0 : else if (adj->lookup_next_index == IP_LOOKUP_NEXT_ARP)
2078 0 : *next0 = TCP_OUTPUT_NEXT_IP_ARP;
2079 : else
2080 : {
2081 0 : *next0 = TCP_OUTPUT_NEXT_DROP;
2082 0 : *error0 = TCP_ERROR_LINK_LOCAL_RW;
2083 : }
2084 0 : vnet_buffer (b0)->ip.adj_index[VLIB_TX] = ai;
2085 : }
2086 :
2087 : static void
2088 0 : tcp46_output_trace_frame (vlib_main_t * vm, vlib_node_runtime_t * node,
2089 : u32 * to_next, u32 n_bufs)
2090 : {
2091 : tcp_connection_t *tc;
2092 : tcp_tx_trace_t *t;
2093 : vlib_buffer_t *b;
2094 : tcp_header_t *th;
2095 : int i;
2096 :
2097 0 : for (i = 0; i < n_bufs; i++)
2098 : {
2099 0 : b = vlib_get_buffer (vm, to_next[i]);
2100 0 : if (!(b->flags & VLIB_BUFFER_IS_TRACED))
2101 0 : continue;
2102 0 : th = vlib_buffer_get_current (b);
2103 0 : tc = tcp_connection_get (vnet_buffer (b)->tcp.connection_index,
2104 : vm->thread_index);
2105 0 : t = vlib_add_trace (vm, node, b, sizeof (*t));
2106 0 : clib_memcpy_fast (&t->tcp_header, th, sizeof (t->tcp_header));
2107 0 : clib_memcpy_fast (&t->tcp_connection, tc, sizeof (t->tcp_connection));
2108 : }
2109 0 : }
2110 :
2111 : always_inline void
2112 1020810 : tcp_output_push_ip (vlib_main_t * vm, vlib_buffer_t * b0,
2113 : tcp_connection_t * tc0, u8 is_ip4)
2114 : {
2115 : TCP_EVT (TCP_EVT_OUTPUT, tc0,
2116 : ((tcp_header_t *) vlib_buffer_get_current (b0))->flags,
2117 : b0->current_length);
2118 :
2119 1020810 : if (is_ip4)
2120 1020800 : vlib_buffer_push_ip4 (vm, b0, &tc0->c_lcl_ip4, &tc0->c_rmt_ip4,
2121 1020800 : IP_PROTOCOL_TCP, tcp_csum_offload (tc0));
2122 : else
2123 17 : vlib_buffer_push_ip6_custom (vm, b0, &tc0->c_lcl_ip6, &tc0->c_rmt_ip6,
2124 : IP_PROTOCOL_TCP, tc0->ipv6_flow_label);
2125 1020810 : }
2126 :
2127 : always_inline void
2128 1020810 : tcp_check_if_gso (tcp_connection_t * tc, vlib_buffer_t * b)
2129 : {
2130 1020810 : if (PREDICT_TRUE (!(tc->cfg_flags & TCP_CFG_F_TSO)))
2131 1020810 : return;
2132 :
2133 0 : u16 data_len = b->current_length - sizeof (tcp_header_t) - tc->snd_opts_len;
2134 :
2135 0 : if (PREDICT_FALSE (b->flags & VLIB_BUFFER_TOTAL_LENGTH_VALID))
2136 0 : data_len += b->total_length_not_including_first_buffer;
2137 :
2138 0 : if (PREDICT_TRUE (data_len <= tc->snd_mss))
2139 0 : return;
2140 : else
2141 : {
2142 0 : ASSERT ((b->flags & VNET_BUFFER_F_L3_HDR_OFFSET_VALID) != 0);
2143 0 : ASSERT ((b->flags & VNET_BUFFER_F_L4_HDR_OFFSET_VALID) != 0);
2144 0 : b->flags |= VNET_BUFFER_F_GSO;
2145 0 : vnet_buffer2 (b)->gso_l4_hdr_sz =
2146 0 : sizeof (tcp_header_t) + tc->snd_opts_len;
2147 0 : vnet_buffer2 (b)->gso_size = tc->snd_mss;
2148 : }
2149 : }
2150 :
2151 : always_inline void
2152 1020810 : tcp_output_handle_packet (tcp_connection_t * tc0, vlib_buffer_t * b0,
2153 : vlib_node_runtime_t * error_node, u16 * next0,
2154 : u8 is_ip4)
2155 : {
2156 : /* If next_index is not drop use it */
2157 1020810 : if (tc0->next_node_index)
2158 : {
2159 0 : *next0 = tc0->next_node_index;
2160 0 : vnet_buffer (b0)->tcp.next_node_opaque = tc0->next_node_opaque;
2161 : }
2162 : else
2163 : {
2164 1020810 : *next0 = TCP_OUTPUT_NEXT_IP_LOOKUP;
2165 : }
2166 :
2167 1020810 : vnet_buffer (b0)->sw_if_index[VLIB_TX] = tc0->c_fib_index;
2168 1020810 : vnet_buffer (b0)->sw_if_index[VLIB_RX] = tc0->sw_if_index;
2169 :
2170 1020810 : if (!is_ip4)
2171 : {
2172 17 : u32 error0 = 0;
2173 :
2174 17 : if (PREDICT_FALSE (ip6_address_is_link_local_unicast (&tc0->c_rmt_ip6)))
2175 0 : tcp_output_handle_link_local (tc0, b0, next0, &error0);
2176 :
2177 17 : if (PREDICT_FALSE (error0))
2178 : {
2179 0 : b0->error = error_node->errors[error0];
2180 0 : return;
2181 : }
2182 : }
2183 :
2184 1020810 : tc0->segs_out += 1;
2185 : }
2186 :
2187 : always_inline uword
2188 30293 : tcp46_output_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
2189 : vlib_frame_t * frame, int is_ip4)
2190 : {
2191 30293 : u32 n_left_from, *from, thread_index = vm->thread_index;
2192 : vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
2193 : u16 nexts[VLIB_FRAME_SIZE], *next;
2194 30293 : u16 err_counters[TCP_N_ERROR] = { 0 };
2195 :
2196 30293 : from = vlib_frame_vector_args (frame);
2197 30293 : n_left_from = frame->n_vectors;
2198 30293 : tcp_update_time_now (tcp_get_worker (thread_index));
2199 :
2200 30293 : if (PREDICT_FALSE (node->flags & VLIB_NODE_FLAG_TRACE))
2201 0 : tcp46_output_trace_frame (vm, node, from, n_left_from);
2202 :
2203 30293 : vlib_get_buffers (vm, from, bufs, n_left_from);
2204 30293 : b = bufs;
2205 30293 : next = nexts;
2206 :
2207 510000 : while (n_left_from >= 4)
2208 : {
2209 : tcp_connection_t *tc0, *tc1;
2210 :
2211 : {
2212 479707 : vlib_prefetch_buffer_header (b[2], STORE);
2213 479707 : CLIB_PREFETCH (b[2]->data, 2 * CLIB_CACHE_LINE_BYTES, STORE);
2214 :
2215 479707 : vlib_prefetch_buffer_header (b[3], STORE);
2216 479707 : CLIB_PREFETCH (b[3]->data, 2 * CLIB_CACHE_LINE_BYTES, STORE);
2217 : }
2218 :
2219 479707 : tc0 = tcp_connection_get (vnet_buffer (b[0])->tcp.connection_index,
2220 : thread_index);
2221 479707 : tc1 = tcp_connection_get (vnet_buffer (b[1])->tcp.connection_index,
2222 : thread_index);
2223 :
2224 479707 : if (PREDICT_TRUE (!tc0 + !tc1 == 0))
2225 : {
2226 479707 : tcp_output_push_ip (vm, b[0], tc0, is_ip4);
2227 479707 : tcp_output_push_ip (vm, b[1], tc1, is_ip4);
2228 :
2229 479707 : tcp_check_if_gso (tc0, b[0]);
2230 479707 : tcp_check_if_gso (tc1, b[1]);
2231 :
2232 479707 : tcp_output_handle_packet (tc0, b[0], node, &next[0], is_ip4);
2233 479707 : tcp_output_handle_packet (tc1, b[1], node, &next[1], is_ip4);
2234 : }
2235 : else
2236 : {
2237 0 : if (tc0 != 0)
2238 : {
2239 0 : tcp_output_push_ip (vm, b[0], tc0, is_ip4);
2240 0 : tcp_check_if_gso (tc0, b[0]);
2241 0 : tcp_output_handle_packet (tc0, b[0], node, &next[0], is_ip4);
2242 : }
2243 : else
2244 : {
2245 0 : tcp_inc_err_counter (err_counters, TCP_ERROR_INVALID_CONNECTION,
2246 : 1);
2247 0 : next[0] = TCP_OUTPUT_NEXT_DROP;
2248 : }
2249 0 : if (tc1 != 0)
2250 : {
2251 0 : tcp_output_push_ip (vm, b[1], tc1, is_ip4);
2252 0 : tcp_check_if_gso (tc1, b[1]);
2253 0 : tcp_output_handle_packet (tc1, b[1], node, &next[1], is_ip4);
2254 : }
2255 : else
2256 : {
2257 0 : tcp_inc_err_counter (err_counters, TCP_ERROR_INVALID_CONNECTION,
2258 : 1);
2259 0 : next[1] = TCP_OUTPUT_NEXT_DROP;
2260 : }
2261 : }
2262 :
2263 479707 : b += 2;
2264 479707 : next += 2;
2265 479707 : n_left_from -= 2;
2266 : }
2267 91691 : while (n_left_from > 0)
2268 : {
2269 : tcp_connection_t *tc0;
2270 :
2271 61398 : if (n_left_from > 1)
2272 : {
2273 31105 : vlib_prefetch_buffer_header (b[1], STORE);
2274 31105 : CLIB_PREFETCH (b[1]->data, 2 * CLIB_CACHE_LINE_BYTES, STORE);
2275 : }
2276 :
2277 61398 : tc0 = tcp_connection_get (vnet_buffer (b[0])->tcp.connection_index,
2278 : thread_index);
2279 :
2280 61398 : if (PREDICT_TRUE (tc0 != 0))
2281 : {
2282 61398 : tcp_output_push_ip (vm, b[0], tc0, is_ip4);
2283 61398 : tcp_check_if_gso (tc0, b[0]);
2284 61398 : tcp_output_handle_packet (tc0, b[0], node, &next[0], is_ip4);
2285 : }
2286 : else
2287 : {
2288 0 : tcp_inc_err_counter (err_counters, TCP_ERROR_INVALID_CONNECTION, 1);
2289 0 : next[0] = TCP_OUTPUT_NEXT_DROP;
2290 : }
2291 :
2292 61398 : b += 1;
2293 61398 : next += 1;
2294 61398 : n_left_from -= 1;
2295 : }
2296 :
2297 1151130 : tcp_store_err_counters (output, err_counters);
2298 30293 : vlib_buffer_enqueue_to_next (vm, node, from, nexts, frame->n_vectors);
2299 30293 : vlib_node_increment_counter (vm, tcp_node_index (output, is_ip4),
2300 30293 : TCP_ERROR_PKTS_SENT, frame->n_vectors);
2301 30293 : return frame->n_vectors;
2302 : }
2303 :
2304 32512 : VLIB_NODE_FN (tcp4_output_node) (vlib_main_t * vm, vlib_node_runtime_t * node,
2305 : vlib_frame_t * from_frame)
2306 : {
2307 30276 : return tcp46_output_inline (vm, node, from_frame, 1 /* is_ip4 */ );
2308 : }
2309 :
2310 2253 : VLIB_NODE_FN (tcp6_output_node) (vlib_main_t * vm, vlib_node_runtime_t * node,
2311 : vlib_frame_t * from_frame)
2312 : {
2313 17 : return tcp46_output_inline (vm, node, from_frame, 0 /* is_ip4 */ );
2314 : }
2315 :
2316 : /* *INDENT-OFF* */
2317 178120 : VLIB_REGISTER_NODE (tcp4_output_node) =
2318 : {
2319 : .name = "tcp4-output",
2320 : /* Takes a vector of packets. */
2321 : .vector_size = sizeof (u32),
2322 : .n_errors = TCP_N_ERROR,
2323 : .protocol_hint = VLIB_NODE_PROTO_HINT_TCP,
2324 : .error_counters = tcp_output_error_counters,
2325 : .n_next_nodes = TCP_OUTPUT_N_NEXT,
2326 : .next_nodes = {
2327 : #define _(s,n) [TCP_OUTPUT_NEXT_##s] = n,
2328 : foreach_tcp4_output_next
2329 : #undef _
2330 : },
2331 : .format_buffer = format_tcp_header,
2332 : .format_trace = format_tcp_tx_trace,
2333 : };
2334 : /* *INDENT-ON* */
2335 :
2336 : /* *INDENT-OFF* */
2337 178120 : VLIB_REGISTER_NODE (tcp6_output_node) =
2338 : {
2339 : .name = "tcp6-output",
2340 : /* Takes a vector of packets. */
2341 : .vector_size = sizeof (u32),
2342 : .n_errors = TCP_N_ERROR,
2343 : .protocol_hint = VLIB_NODE_PROTO_HINT_TCP,
2344 : .error_counters = tcp_output_error_counters,
2345 : .n_next_nodes = TCP_OUTPUT_N_NEXT,
2346 : .next_nodes = {
2347 : #define _(s,n) [TCP_OUTPUT_NEXT_##s] = n,
2348 : foreach_tcp6_output_next
2349 : #undef _
2350 : },
2351 : .format_buffer = format_tcp_header,
2352 : .format_trace = format_tcp_tx_trace,
2353 : };
2354 : /* *INDENT-ON* */
2355 :
2356 : typedef enum _tcp_reset_next
2357 : {
2358 : TCP_RESET_NEXT_DROP,
2359 : TCP_RESET_NEXT_IP_LOOKUP,
2360 : TCP_RESET_N_NEXT
2361 : } tcp_reset_next_t;
2362 :
2363 : #define foreach_tcp4_reset_next \
2364 : _(DROP, "error-drop") \
2365 : _(IP_LOOKUP, "ip4-lookup")
2366 :
2367 : #define foreach_tcp6_reset_next \
2368 : _(DROP, "error-drop") \
2369 : _(IP_LOOKUP, "ip6-lookup")
2370 :
2371 : static void
2372 0 : tcp_reset_trace_frame (vlib_main_t *vm, vlib_node_runtime_t *node,
2373 : vlib_buffer_t **bs, u32 n_bufs, u8 is_ip4)
2374 : {
2375 : tcp_header_t *tcp;
2376 : tcp_tx_trace_t *t;
2377 : int i;
2378 :
2379 0 : for (i = 0; i < n_bufs; i++)
2380 : {
2381 0 : if (bs[i]->flags & VLIB_BUFFER_IS_TRACED)
2382 : {
2383 0 : tcp = vlib_buffer_get_current (bs[i]);
2384 0 : t = vlib_add_trace (vm, node, bs[i], sizeof (*t));
2385 :
2386 0 : if (is_ip4)
2387 : {
2388 0 : ip4_header_t *ih4 = vlib_buffer_get_current (bs[i]);
2389 0 : tcp = ip4_next_header (ih4);
2390 0 : t->tcp_connection.c_lcl_ip.ip4 = ih4->dst_address;
2391 0 : t->tcp_connection.c_rmt_ip.ip4 = ih4->src_address;
2392 0 : t->tcp_connection.c_is_ip4 = 1;
2393 : }
2394 : else
2395 : {
2396 0 : ip6_header_t *ih6 = vlib_buffer_get_current (bs[i]);
2397 0 : tcp = ip6_next_header (ih6);
2398 0 : t->tcp_connection.c_lcl_ip.ip6 = ih6->dst_address;
2399 0 : t->tcp_connection.c_rmt_ip.ip6 = ih6->src_address;
2400 : }
2401 0 : t->tcp_connection.c_lcl_port = tcp->dst_port;
2402 0 : t->tcp_connection.c_rmt_port = tcp->src_port;
2403 0 : t->tcp_connection.c_proto = TRANSPORT_PROTO_TCP;
2404 0 : clib_memcpy_fast (&t->tcp_header, tcp, sizeof (t->tcp_header));
2405 : }
2406 : }
2407 0 : }
2408 :
2409 : static uword
2410 2 : tcp46_reset_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
2411 : vlib_frame_t *frame, u8 is_ip4)
2412 : {
2413 : vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
2414 : u16 nexts[VLIB_FRAME_SIZE], *next;
2415 : u32 n_left_from, *from;
2416 :
2417 2 : from = vlib_frame_vector_args (frame);
2418 2 : n_left_from = frame->n_vectors;
2419 2 : vlib_get_buffers (vm, from, bufs, n_left_from);
2420 :
2421 2 : b = bufs;
2422 2 : next = nexts;
2423 :
2424 88 : while (n_left_from > 0)
2425 : {
2426 86 : tcp_buffer_make_reset (vm, b[0], is_ip4);
2427 :
2428 : /* IP lookup in fib where it was received. Previous value
2429 : * was overwritten by tcp-input */
2430 86 : vnet_buffer (b[0])->sw_if_index[VLIB_TX] =
2431 86 : vec_elt (ip4_main.fib_index_by_sw_if_index,
2432 : vnet_buffer (b[0])->sw_if_index[VLIB_RX]);
2433 :
2434 86 : b[0]->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED;
2435 86 : next[0] = TCP_RESET_NEXT_IP_LOOKUP;
2436 :
2437 86 : b += 1;
2438 86 : next += 1;
2439 86 : n_left_from -= 1;
2440 : }
2441 :
2442 2 : if (PREDICT_FALSE (node->flags & VLIB_NODE_FLAG_TRACE))
2443 0 : tcp_reset_trace_frame (vm, node, bufs, frame->n_vectors, is_ip4);
2444 :
2445 2 : vlib_buffer_enqueue_to_next (vm, node, from, nexts, frame->n_vectors);
2446 :
2447 2 : vlib_node_increment_counter (vm, node->node_index, TCP_ERROR_RST_SENT,
2448 2 : frame->n_vectors);
2449 :
2450 2 : return frame->n_vectors;
2451 : }
2452 :
2453 2238 : VLIB_NODE_FN (tcp4_reset_node) (vlib_main_t * vm, vlib_node_runtime_t * node,
2454 : vlib_frame_t * from_frame)
2455 : {
2456 2 : return tcp46_reset_inline (vm, node, from_frame, 1);
2457 : }
2458 :
2459 2236 : VLIB_NODE_FN (tcp6_reset_node) (vlib_main_t * vm, vlib_node_runtime_t * node,
2460 : vlib_frame_t * from_frame)
2461 : {
2462 0 : return tcp46_reset_inline (vm, node, from_frame, 0);
2463 : }
2464 :
2465 : /* *INDENT-OFF* */
2466 178120 : VLIB_REGISTER_NODE (tcp4_reset_node) = {
2467 : .name = "tcp4-reset",
2468 : .vector_size = sizeof (u32),
2469 : .n_errors = TCP_N_ERROR,
2470 : .error_counters = tcp_output_error_counters,
2471 : .n_next_nodes = TCP_RESET_N_NEXT,
2472 : .next_nodes = {
2473 : #define _(s,n) [TCP_RESET_NEXT_##s] = n,
2474 : foreach_tcp4_reset_next
2475 : #undef _
2476 : },
2477 : .format_trace = format_tcp_tx_trace,
2478 : };
2479 : /* *INDENT-ON* */
2480 :
2481 : /* *INDENT-OFF* */
2482 178120 : VLIB_REGISTER_NODE (tcp6_reset_node) = {
2483 : .name = "tcp6-reset",
2484 : .vector_size = sizeof (u32),
2485 : .n_errors = TCP_N_ERROR,
2486 : .error_counters = tcp_output_error_counters,
2487 : .n_next_nodes = TCP_RESET_N_NEXT,
2488 : .next_nodes = {
2489 : #define _(s,n) [TCP_RESET_NEXT_##s] = n,
2490 : foreach_tcp6_reset_next
2491 : #undef _
2492 : },
2493 : .format_trace = format_tcp_tx_trace,
2494 : };
2495 : /* *INDENT-ON* */
2496 :
2497 : /*
2498 : * fd.io coding-style-patch-verification: ON
2499 : *
2500 : * Local Variables:
2501 : * eval: (c-set-style "gnu")
2502 : * End:
2503 : */
|
