1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* SCTP kernel implementation 3 * (C) Copyright IBM Corp. 2001, 2004 4 * Copyright (c) 1999-2000 Cisco, Inc. 5 * Copyright (c) 1999-2001 Motorola, Inc. 6 * Copyright (c) 2001-2003 Intel Corp. 7 * 8 * This file is part of the SCTP kernel implementation 9 * 10 * These functions implement the sctp_outq class. The outqueue handles 11 * bundling and queueing of outgoing SCTP chunks. 12 * 13 * Please send any bug reports or fixes you make to the 14 * email address(es): 15 * lksctp developers <linux-sctp@vger.kernel.org> 16 * 17 * Written or modified by: 18 * La Monte H.P. Yarroll <piggy@acm.org> 19 * Karl Knutson <karl@athena.chicago.il.us> 20 * Perry Melange <pmelange@null.cc.uic.edu> 21 * Xingang Guo <xingang.guo@intel.com> 22 * Hui Huang <hui.huang@nokia.com> 23 * Sridhar Samudrala <sri@us.ibm.com> 24 * Jon Grimm <jgrimm@us.ibm.com> 25 */ 26 27 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 28 29 #include <linux/types.h> 30 #include <linux/list.h> /* For struct list_head */ 31 #include <linux/socket.h> 32 #include <linux/ip.h> 33 #include <linux/slab.h> 34 #include <net/sock.h> /* For skb_set_owner_w */ 35 36 #include <net/sctp/sctp.h> 37 #include <net/sctp/sm.h> 38 #include <net/sctp/stream_sched.h> 39 #include <trace/events/sctp.h> 40 41 /* Declare internal functions here. */ 42 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn); 43 static void sctp_check_transmitted(struct sctp_outq *q, 44 struct list_head *transmitted_queue, 45 struct sctp_transport *transport, 46 union sctp_addr *saddr, 47 struct sctp_sackhdr *sack, 48 __u32 *highest_new_tsn); 49 50 static void sctp_mark_missing(struct sctp_outq *q, 51 struct list_head *transmitted_queue, 52 struct sctp_transport *transport, 53 __u32 highest_new_tsn, 54 int count_of_newacks); 55 56 static void sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp); 57 58 /* Add data to the front of the queue. */ 59 static inline void sctp_outq_head_data(struct sctp_outq *q, 60 struct sctp_chunk *ch) 61 { 62 struct sctp_stream_out_ext *oute; 63 __u16 stream; 64 65 list_add(&ch->list, &q->out_chunk_list); 66 q->out_qlen += ch->skb->len; 67 68 stream = sctp_chunk_stream_no(ch); 69 oute = SCTP_SO(&q->asoc->stream, stream)->ext; 70 list_add(&ch->stream_list, &oute->outq); 71 } 72 73 /* Take data from the front of the queue. */ 74 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q) 75 { 76 return q->sched->dequeue(q); 77 } 78 79 /* Add data chunk to the end of the queue. */ 80 static inline void sctp_outq_tail_data(struct sctp_outq *q, 81 struct sctp_chunk *ch) 82 { 83 struct sctp_stream_out_ext *oute; 84 __u16 stream; 85 86 list_add_tail(&ch->list, &q->out_chunk_list); 87 q->out_qlen += ch->skb->len; 88 89 stream = sctp_chunk_stream_no(ch); 90 oute = SCTP_SO(&q->asoc->stream, stream)->ext; 91 list_add_tail(&ch->stream_list, &oute->outq); 92 } 93 94 /* 95 * SFR-CACC algorithm: 96 * D) If count_of_newacks is greater than or equal to 2 97 * and t was not sent to the current primary then the 98 * sender MUST NOT increment missing report count for t. 99 */ 100 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary, 101 struct sctp_transport *transport, 102 int count_of_newacks) 103 { 104 if (count_of_newacks >= 2 && transport != primary) 105 return 1; 106 return 0; 107 } 108 109 /* 110 * SFR-CACC algorithm: 111 * F) If count_of_newacks is less than 2, let d be the 112 * destination to which t was sent. If cacc_saw_newack 113 * is 0 for destination d, then the sender MUST NOT 114 * increment missing report count for t. 115 */ 116 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport, 117 int count_of_newacks) 118 { 119 if (count_of_newacks < 2 && 120 (transport && !transport->cacc.cacc_saw_newack)) 121 return 1; 122 return 0; 123 } 124 125 /* 126 * SFR-CACC algorithm: 127 * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD 128 * execute steps C, D, F. 129 * 130 * C has been implemented in sctp_outq_sack 131 */ 132 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary, 133 struct sctp_transport *transport, 134 int count_of_newacks) 135 { 136 if (!primary->cacc.cycling_changeover) { 137 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks)) 138 return 1; 139 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks)) 140 return 1; 141 return 0; 142 } 143 return 0; 144 } 145 146 /* 147 * SFR-CACC algorithm: 148 * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less 149 * than next_tsn_at_change of the current primary, then 150 * the sender MUST NOT increment missing report count 151 * for t. 152 */ 153 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn) 154 { 155 if (primary->cacc.cycling_changeover && 156 TSN_lt(tsn, primary->cacc.next_tsn_at_change)) 157 return 1; 158 return 0; 159 } 160 161 /* 162 * SFR-CACC algorithm: 163 * 3) If the missing report count for TSN t is to be 164 * incremented according to [RFC2960] and 165 * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set, 166 * then the sender MUST further execute steps 3.1 and 167 * 3.2 to determine if the missing report count for 168 * TSN t SHOULD NOT be incremented. 169 * 170 * 3.3) If 3.1 and 3.2 do not dictate that the missing 171 * report count for t should not be incremented, then 172 * the sender SHOULD increment missing report count for 173 * t (according to [RFC2960] and [SCTP_STEWART_2002]). 174 */ 175 static inline int sctp_cacc_skip(struct sctp_transport *primary, 176 struct sctp_transport *transport, 177 int count_of_newacks, 178 __u32 tsn) 179 { 180 if (primary->cacc.changeover_active && 181 (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) || 182 sctp_cacc_skip_3_2(primary, tsn))) 183 return 1; 184 return 0; 185 } 186 187 /* Initialize an existing sctp_outq. This does the boring stuff. 188 * You still need to define handlers if you really want to DO 189 * something with this structure... 190 */ 191 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q) 192 { 193 memset(q, 0, sizeof(struct sctp_outq)); 194 195 q->asoc = asoc; 196 INIT_LIST_HEAD(&q->out_chunk_list); 197 INIT_LIST_HEAD(&q->control_chunk_list); 198 INIT_LIST_HEAD(&q->retransmit); 199 INIT_LIST_HEAD(&q->sacked); 200 INIT_LIST_HEAD(&q->abandoned); 201 sctp_sched_set_sched(asoc, sctp_sk(asoc->base.sk)->default_ss); 202 } 203 204 /* Free the outqueue structure and any related pending chunks. 205 */ 206 static void __sctp_outq_teardown(struct sctp_outq *q) 207 { 208 struct sctp_transport *transport; 209 struct list_head *lchunk, *temp; 210 struct sctp_chunk *chunk, *tmp; 211 212 /* Throw away unacknowledged chunks. */ 213 list_for_each_entry(transport, &q->asoc->peer.transport_addr_list, 214 transports) { 215 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) { 216 chunk = list_entry(lchunk, struct sctp_chunk, 217 transmitted_list); 218 /* Mark as part of a failed message. */ 219 sctp_chunk_fail(chunk, q->error); 220 sctp_chunk_free(chunk); 221 } 222 } 223 224 /* Throw away chunks that have been gap ACKed. */ 225 list_for_each_safe(lchunk, temp, &q->sacked) { 226 list_del_init(lchunk); 227 chunk = list_entry(lchunk, struct sctp_chunk, 228 transmitted_list); 229 sctp_chunk_fail(chunk, q->error); 230 sctp_chunk_free(chunk); 231 } 232 233 /* Throw away any chunks in the retransmit queue. */ 234 list_for_each_safe(lchunk, temp, &q->retransmit) { 235 list_del_init(lchunk); 236 chunk = list_entry(lchunk, struct sctp_chunk, 237 transmitted_list); 238 sctp_chunk_fail(chunk, q->error); 239 sctp_chunk_free(chunk); 240 } 241 242 /* Throw away any chunks that are in the abandoned queue. */ 243 list_for_each_safe(lchunk, temp, &q->abandoned) { 244 list_del_init(lchunk); 245 chunk = list_entry(lchunk, struct sctp_chunk, 246 transmitted_list); 247 sctp_chunk_fail(chunk, q->error); 248 sctp_chunk_free(chunk); 249 } 250 251 /* Throw away any leftover data chunks. */ 252 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) { 253 sctp_sched_dequeue_done(q, chunk); 254 255 /* Mark as send failure. */ 256 sctp_chunk_fail(chunk, q->error); 257 sctp_chunk_free(chunk); 258 } 259 260 /* Throw away any leftover control chunks. */ 261 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) { 262 list_del_init(&chunk->list); 263 sctp_chunk_free(chunk); 264 } 265 } 266 267 void sctp_outq_teardown(struct sctp_outq *q) 268 { 269 __sctp_outq_teardown(q); 270 sctp_outq_init(q->asoc, q); 271 } 272 273 /* Free the outqueue structure and any related pending chunks. */ 274 void sctp_outq_free(struct sctp_outq *q) 275 { 276 /* Throw away leftover chunks. */ 277 __sctp_outq_teardown(q); 278 } 279 280 /* Put a new chunk in an sctp_outq. */ 281 void sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk, gfp_t gfp) 282 { 283 struct net *net = q->asoc->base.net; 284 285 pr_debug("%s: outq:%p, chunk:%p[%s]\n", __func__, q, chunk, 286 chunk && chunk->chunk_hdr ? 287 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) : 288 "illegal chunk"); 289 290 /* If it is data, queue it up, otherwise, send it 291 * immediately. 292 */ 293 if (sctp_chunk_is_data(chunk)) { 294 pr_debug("%s: outqueueing: outq:%p, chunk:%p[%s])\n", 295 __func__, q, chunk, chunk && chunk->chunk_hdr ? 296 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) : 297 "illegal chunk"); 298 299 sctp_outq_tail_data(q, chunk); 300 if (chunk->asoc->peer.prsctp_capable && 301 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags)) 302 chunk->asoc->sent_cnt_removable++; 303 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED) 304 SCTP_INC_STATS(net, SCTP_MIB_OUTUNORDERCHUNKS); 305 else 306 SCTP_INC_STATS(net, SCTP_MIB_OUTORDERCHUNKS); 307 } else { 308 list_add_tail(&chunk->list, &q->control_chunk_list); 309 SCTP_INC_STATS(net, SCTP_MIB_OUTCTRLCHUNKS); 310 } 311 312 if (!q->cork) 313 sctp_outq_flush(q, 0, gfp); 314 } 315 316 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list 317 * and the abandoned list are in ascending order. 318 */ 319 static void sctp_insert_list(struct list_head *head, struct list_head *new) 320 { 321 struct list_head *pos; 322 struct sctp_chunk *nchunk, *lchunk; 323 __u32 ntsn, ltsn; 324 int done = 0; 325 326 nchunk = list_entry(new, struct sctp_chunk, transmitted_list); 327 ntsn = ntohl(nchunk->subh.data_hdr->tsn); 328 329 list_for_each(pos, head) { 330 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list); 331 ltsn = ntohl(lchunk->subh.data_hdr->tsn); 332 if (TSN_lt(ntsn, ltsn)) { 333 list_add(new, pos->prev); 334 done = 1; 335 break; 336 } 337 } 338 if (!done) 339 list_add_tail(new, head); 340 } 341 342 static int sctp_prsctp_prune_sent(struct sctp_association *asoc, 343 struct sctp_sndrcvinfo *sinfo, 344 struct list_head *queue, int msg_len) 345 { 346 struct sctp_chunk *chk, *temp; 347 348 list_for_each_entry_safe(chk, temp, queue, transmitted_list) { 349 struct sctp_stream_out *streamout; 350 351 if (!chk->msg->abandoned && 352 (!SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) || 353 chk->sinfo.sinfo_timetolive <= sinfo->sinfo_timetolive)) 354 continue; 355 356 chk->msg->abandoned = 1; 357 list_del_init(&chk->transmitted_list); 358 sctp_insert_list(&asoc->outqueue.abandoned, 359 &chk->transmitted_list); 360 361 streamout = SCTP_SO(&asoc->stream, chk->sinfo.sinfo_stream); 362 asoc->sent_cnt_removable--; 363 asoc->abandoned_sent[SCTP_PR_INDEX(PRIO)]++; 364 streamout->ext->abandoned_sent[SCTP_PR_INDEX(PRIO)]++; 365 366 if (queue != &asoc->outqueue.retransmit && 367 !chk->tsn_gap_acked) { 368 if (chk->transport) 369 chk->transport->flight_size -= 370 sctp_data_size(chk); 371 asoc->outqueue.outstanding_bytes -= sctp_data_size(chk); 372 } 373 374 msg_len -= chk->skb->truesize + sizeof(struct sctp_chunk); 375 if (msg_len <= 0) 376 break; 377 } 378 379 return msg_len; 380 } 381 382 static int sctp_prsctp_prune_unsent(struct sctp_association *asoc, 383 struct sctp_sndrcvinfo *sinfo, int msg_len) 384 { 385 struct sctp_outq *q = &asoc->outqueue; 386 struct sctp_chunk *chk, *temp; 387 388 q->sched->unsched_all(&asoc->stream); 389 390 list_for_each_entry_safe(chk, temp, &q->out_chunk_list, list) { 391 if (!chk->msg->abandoned && 392 (!(chk->chunk_hdr->flags & SCTP_DATA_FIRST_FRAG) || 393 !SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) || 394 chk->sinfo.sinfo_timetolive <= sinfo->sinfo_timetolive)) 395 continue; 396 397 chk->msg->abandoned = 1; 398 sctp_sched_dequeue_common(q, chk); 399 asoc->sent_cnt_removable--; 400 asoc->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++; 401 if (chk->sinfo.sinfo_stream < asoc->stream.outcnt) { 402 struct sctp_stream_out *streamout = 403 SCTP_SO(&asoc->stream, chk->sinfo.sinfo_stream); 404 405 streamout->ext->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++; 406 } 407 408 msg_len -= chk->skb->truesize + sizeof(struct sctp_chunk); 409 sctp_chunk_free(chk); 410 if (msg_len <= 0) 411 break; 412 } 413 414 q->sched->sched_all(&asoc->stream); 415 416 return msg_len; 417 } 418 419 /* Abandon the chunks according their priorities */ 420 void sctp_prsctp_prune(struct sctp_association *asoc, 421 struct sctp_sndrcvinfo *sinfo, int msg_len) 422 { 423 struct sctp_transport *transport; 424 425 if (!asoc->peer.prsctp_capable || !asoc->sent_cnt_removable) 426 return; 427 428 msg_len = sctp_prsctp_prune_sent(asoc, sinfo, 429 &asoc->outqueue.retransmit, 430 msg_len); 431 if (msg_len <= 0) 432 return; 433 434 list_for_each_entry(transport, &asoc->peer.transport_addr_list, 435 transports) { 436 msg_len = sctp_prsctp_prune_sent(asoc, sinfo, 437 &transport->transmitted, 438 msg_len); 439 if (msg_len <= 0) 440 return; 441 } 442 443 sctp_prsctp_prune_unsent(asoc, sinfo, msg_len); 444 } 445 446 /* Mark all the eligible packets on a transport for retransmission. */ 447 void sctp_retransmit_mark(struct sctp_outq *q, 448 struct sctp_transport *transport, 449 __u8 reason) 450 { 451 struct list_head *lchunk, *ltemp; 452 struct sctp_chunk *chunk; 453 454 /* Walk through the specified transmitted queue. */ 455 list_for_each_safe(lchunk, ltemp, &transport->transmitted) { 456 chunk = list_entry(lchunk, struct sctp_chunk, 457 transmitted_list); 458 459 /* If the chunk is abandoned, move it to abandoned list. */ 460 if (sctp_chunk_abandoned(chunk)) { 461 list_del_init(lchunk); 462 sctp_insert_list(&q->abandoned, lchunk); 463 464 /* If this chunk has not been previousely acked, 465 * stop considering it 'outstanding'. Our peer 466 * will most likely never see it since it will 467 * not be retransmitted 468 */ 469 if (!chunk->tsn_gap_acked) { 470 if (chunk->transport) 471 chunk->transport->flight_size -= 472 sctp_data_size(chunk); 473 q->outstanding_bytes -= sctp_data_size(chunk); 474 q->asoc->peer.rwnd += sctp_data_size(chunk); 475 } 476 continue; 477 } 478 479 /* If we are doing retransmission due to a timeout or pmtu 480 * discovery, only the chunks that are not yet acked should 481 * be added to the retransmit queue. 482 */ 483 if ((reason == SCTP_RTXR_FAST_RTX && 484 (chunk->fast_retransmit == SCTP_NEED_FRTX)) || 485 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) { 486 /* RFC 2960 6.2.1 Processing a Received SACK 487 * 488 * C) Any time a DATA chunk is marked for 489 * retransmission (via either T3-rtx timer expiration 490 * (Section 6.3.3) or via fast retransmit 491 * (Section 7.2.4)), add the data size of those 492 * chunks to the rwnd. 493 */ 494 q->asoc->peer.rwnd += sctp_data_size(chunk); 495 q->outstanding_bytes -= sctp_data_size(chunk); 496 if (chunk->transport) 497 transport->flight_size -= sctp_data_size(chunk); 498 499 /* sctpimpguide-05 Section 2.8.2 500 * M5) If a T3-rtx timer expires, the 501 * 'TSN.Missing.Report' of all affected TSNs is set 502 * to 0. 503 */ 504 chunk->tsn_missing_report = 0; 505 506 /* If a chunk that is being used for RTT measurement 507 * has to be retransmitted, we cannot use this chunk 508 * anymore for RTT measurements. Reset rto_pending so 509 * that a new RTT measurement is started when a new 510 * data chunk is sent. 511 */ 512 if (chunk->rtt_in_progress) { 513 chunk->rtt_in_progress = 0; 514 transport->rto_pending = 0; 515 } 516 517 /* Move the chunk to the retransmit queue. The chunks 518 * on the retransmit queue are always kept in order. 519 */ 520 list_del_init(lchunk); 521 sctp_insert_list(&q->retransmit, lchunk); 522 } 523 } 524 525 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, " 526 "flight_size:%d, pba:%d\n", __func__, transport, reason, 527 transport->cwnd, transport->ssthresh, transport->flight_size, 528 transport->partial_bytes_acked); 529 } 530 531 /* Mark all the eligible packets on a transport for retransmission and force 532 * one packet out. 533 */ 534 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport, 535 enum sctp_retransmit_reason reason) 536 { 537 struct net *net = q->asoc->base.net; 538 539 switch (reason) { 540 case SCTP_RTXR_T3_RTX: 541 SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS); 542 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX); 543 /* Update the retran path if the T3-rtx timer has expired for 544 * the current retran path. 545 */ 546 if (transport == transport->asoc->peer.retran_path) 547 sctp_assoc_update_retran_path(transport->asoc); 548 transport->asoc->rtx_data_chunks += 549 transport->asoc->unack_data; 550 break; 551 case SCTP_RTXR_FAST_RTX: 552 SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS); 553 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX); 554 q->fast_rtx = 1; 555 break; 556 case SCTP_RTXR_PMTUD: 557 SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS); 558 break; 559 case SCTP_RTXR_T1_RTX: 560 SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS); 561 transport->asoc->init_retries++; 562 break; 563 default: 564 BUG(); 565 } 566 567 sctp_retransmit_mark(q, transport, reason); 568 569 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination, 570 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by 571 * following the procedures outlined in C1 - C5. 572 */ 573 if (reason == SCTP_RTXR_T3_RTX) 574 q->asoc->stream.si->generate_ftsn(q, q->asoc->ctsn_ack_point); 575 576 /* Flush the queues only on timeout, since fast_rtx is only 577 * triggered during sack processing and the queue 578 * will be flushed at the end. 579 */ 580 if (reason != SCTP_RTXR_FAST_RTX) 581 sctp_outq_flush(q, /* rtx_timeout */ 1, GFP_ATOMIC); 582 } 583 584 /* 585 * Transmit DATA chunks on the retransmit queue. Upon return from 586 * __sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which 587 * need to be transmitted by the caller. 588 * We assume that pkt->transport has already been set. 589 * 590 * The return value is a normal kernel error return value. 591 */ 592 static int __sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt, 593 int rtx_timeout, int *start_timer, gfp_t gfp) 594 { 595 struct sctp_transport *transport = pkt->transport; 596 struct sctp_chunk *chunk, *chunk1; 597 struct list_head *lqueue; 598 enum sctp_xmit status; 599 int error = 0; 600 int timer = 0; 601 int done = 0; 602 int fast_rtx; 603 604 lqueue = &q->retransmit; 605 fast_rtx = q->fast_rtx; 606 607 /* This loop handles time-out retransmissions, fast retransmissions, 608 * and retransmissions due to opening of whindow. 609 * 610 * RFC 2960 6.3.3 Handle T3-rtx Expiration 611 * 612 * E3) Determine how many of the earliest (i.e., lowest TSN) 613 * outstanding DATA chunks for the address for which the 614 * T3-rtx has expired will fit into a single packet, subject 615 * to the MTU constraint for the path corresponding to the 616 * destination transport address to which the retransmission 617 * is being sent (this may be different from the address for 618 * which the timer expires [see Section 6.4]). Call this value 619 * K. Bundle and retransmit those K DATA chunks in a single 620 * packet to the destination endpoint. 621 * 622 * [Just to be painfully clear, if we are retransmitting 623 * because a timeout just happened, we should send only ONE 624 * packet of retransmitted data.] 625 * 626 * For fast retransmissions we also send only ONE packet. However, 627 * if we are just flushing the queue due to open window, we'll 628 * try to send as much as possible. 629 */ 630 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) { 631 /* If the chunk is abandoned, move it to abandoned list. */ 632 if (sctp_chunk_abandoned(chunk)) { 633 list_del_init(&chunk->transmitted_list); 634 sctp_insert_list(&q->abandoned, 635 &chunk->transmitted_list); 636 continue; 637 } 638 639 /* Make sure that Gap Acked TSNs are not retransmitted. A 640 * simple approach is just to move such TSNs out of the 641 * way and into a 'transmitted' queue and skip to the 642 * next chunk. 643 */ 644 if (chunk->tsn_gap_acked) { 645 list_move_tail(&chunk->transmitted_list, 646 &transport->transmitted); 647 continue; 648 } 649 650 /* If we are doing fast retransmit, ignore non-fast_rtransmit 651 * chunks 652 */ 653 if (fast_rtx && !chunk->fast_retransmit) 654 continue; 655 656 redo: 657 /* Attempt to append this chunk to the packet. */ 658 status = sctp_packet_append_chunk(pkt, chunk); 659 660 switch (status) { 661 case SCTP_XMIT_PMTU_FULL: 662 if (!pkt->has_data && !pkt->has_cookie_echo) { 663 /* If this packet did not contain DATA then 664 * retransmission did not happen, so do it 665 * again. We'll ignore the error here since 666 * control chunks are already freed so there 667 * is nothing we can do. 668 */ 669 sctp_packet_transmit(pkt, gfp); 670 goto redo; 671 } 672 673 /* Send this packet. */ 674 error = sctp_packet_transmit(pkt, gfp); 675 676 /* If we are retransmitting, we should only 677 * send a single packet. 678 * Otherwise, try appending this chunk again. 679 */ 680 if (rtx_timeout || fast_rtx) 681 done = 1; 682 else 683 goto redo; 684 685 /* Bundle next chunk in the next round. */ 686 break; 687 688 case SCTP_XMIT_RWND_FULL: 689 /* Send this packet. */ 690 error = sctp_packet_transmit(pkt, gfp); 691 692 /* Stop sending DATA as there is no more room 693 * at the receiver. 694 */ 695 done = 1; 696 break; 697 698 case SCTP_XMIT_DELAY: 699 /* Send this packet. */ 700 error = sctp_packet_transmit(pkt, gfp); 701 702 /* Stop sending DATA because of nagle delay. */ 703 done = 1; 704 break; 705 706 default: 707 /* The append was successful, so add this chunk to 708 * the transmitted list. 709 */ 710 list_move_tail(&chunk->transmitted_list, 711 &transport->transmitted); 712 713 /* Mark the chunk as ineligible for fast retransmit 714 * after it is retransmitted. 715 */ 716 if (chunk->fast_retransmit == SCTP_NEED_FRTX) 717 chunk->fast_retransmit = SCTP_DONT_FRTX; 718 719 q->asoc->stats.rtxchunks++; 720 break; 721 } 722 723 /* Set the timer if there were no errors */ 724 if (!error && !timer) 725 timer = 1; 726 727 if (done) 728 break; 729 } 730 731 /* If we are here due to a retransmit timeout or a fast 732 * retransmit and if there are any chunks left in the retransmit 733 * queue that could not fit in the PMTU sized packet, they need 734 * to be marked as ineligible for a subsequent fast retransmit. 735 */ 736 if (rtx_timeout || fast_rtx) { 737 list_for_each_entry(chunk1, lqueue, transmitted_list) { 738 if (chunk1->fast_retransmit == SCTP_NEED_FRTX) 739 chunk1->fast_retransmit = SCTP_DONT_FRTX; 740 } 741 } 742 743 *start_timer = timer; 744 745 /* Clear fast retransmit hint */ 746 if (fast_rtx) 747 q->fast_rtx = 0; 748 749 return error; 750 } 751 752 /* Cork the outqueue so queued chunks are really queued. */ 753 void sctp_outq_uncork(struct sctp_outq *q, gfp_t gfp) 754 { 755 if (q->cork) 756 q->cork = 0; 757 758 sctp_outq_flush(q, 0, gfp); 759 } 760 761 static int sctp_packet_singleton(struct sctp_transport *transport, 762 struct sctp_chunk *chunk, gfp_t gfp) 763 { 764 const struct sctp_association *asoc = transport->asoc; 765 const __u16 sport = asoc->base.bind_addr.port; 766 const __u16 dport = asoc->peer.port; 767 const __u32 vtag = asoc->peer.i.init_tag; 768 struct sctp_packet singleton; 769 770 sctp_packet_init(&singleton, transport, sport, dport); 771 sctp_packet_config(&singleton, vtag, 0); 772 sctp_packet_append_chunk(&singleton, chunk); 773 return sctp_packet_transmit(&singleton, gfp); 774 } 775 776 /* Struct to hold the context during sctp outq flush */ 777 struct sctp_flush_ctx { 778 struct sctp_outq *q; 779 /* Current transport being used. It's NOT the same as curr active one */ 780 struct sctp_transport *transport; 781 /* These transports have chunks to send. */ 782 struct list_head transport_list; 783 struct sctp_association *asoc; 784 /* Packet on the current transport above */ 785 struct sctp_packet *packet; 786 gfp_t gfp; 787 }; 788 789 /* transport: current transport */ 790 static void sctp_outq_select_transport(struct sctp_flush_ctx *ctx, 791 struct sctp_chunk *chunk) 792 { 793 struct sctp_transport *new_transport = chunk->transport; 794 795 if (!new_transport) { 796 if (!sctp_chunk_is_data(chunk)) { 797 /* If we have a prior transport pointer, see if 798 * the destination address of the chunk 799 * matches the destination address of the 800 * current transport. If not a match, then 801 * try to look up the transport with a given 802 * destination address. We do this because 803 * after processing ASCONFs, we may have new 804 * transports created. 805 */ 806 if (ctx->transport && sctp_cmp_addr_exact(&chunk->dest, 807 &ctx->transport->ipaddr)) 808 new_transport = ctx->transport; 809 else 810 new_transport = sctp_assoc_lookup_paddr(ctx->asoc, 811 &chunk->dest); 812 } 813 814 /* if we still don't have a new transport, then 815 * use the current active path. 816 */ 817 if (!new_transport) 818 new_transport = ctx->asoc->peer.active_path; 819 } else { 820 __u8 type; 821 822 switch (new_transport->state) { 823 case SCTP_INACTIVE: 824 case SCTP_UNCONFIRMED: 825 case SCTP_PF: 826 /* If the chunk is Heartbeat or Heartbeat Ack, 827 * send it to chunk->transport, even if it's 828 * inactive. 829 * 830 * 3.3.6 Heartbeat Acknowledgement: 831 * ... 832 * A HEARTBEAT ACK is always sent to the source IP 833 * address of the IP datagram containing the 834 * HEARTBEAT chunk to which this ack is responding. 835 * ... 836 * 837 * ASCONF_ACKs also must be sent to the source. 838 */ 839 type = chunk->chunk_hdr->type; 840 if (type != SCTP_CID_HEARTBEAT && 841 type != SCTP_CID_HEARTBEAT_ACK && 842 type != SCTP_CID_ASCONF_ACK) 843 new_transport = ctx->asoc->peer.active_path; 844 break; 845 default: 846 break; 847 } 848 } 849 850 /* Are we switching transports? Take care of transport locks. */ 851 if (new_transport != ctx->transport) { 852 ctx->transport = new_transport; 853 ctx->packet = &ctx->transport->packet; 854 855 if (list_empty(&ctx->transport->send_ready)) 856 list_add_tail(&ctx->transport->send_ready, 857 &ctx->transport_list); 858 859 sctp_packet_config(ctx->packet, 860 ctx->asoc->peer.i.init_tag, 861 ctx->asoc->peer.ecn_capable); 862 /* We've switched transports, so apply the 863 * Burst limit to the new transport. 864 */ 865 sctp_transport_burst_limited(ctx->transport); 866 } 867 } 868 869 static void sctp_outq_flush_ctrl(struct sctp_flush_ctx *ctx) 870 { 871 struct sctp_chunk *chunk, *tmp; 872 enum sctp_xmit status; 873 int one_packet, error; 874 875 list_for_each_entry_safe(chunk, tmp, &ctx->q->control_chunk_list, list) { 876 one_packet = 0; 877 878 /* RFC 5061, 5.3 879 * F1) This means that until such time as the ASCONF 880 * containing the add is acknowledged, the sender MUST 881 * NOT use the new IP address as a source for ANY SCTP 882 * packet except on carrying an ASCONF Chunk. 883 */ 884 if (ctx->asoc->src_out_of_asoc_ok && 885 chunk->chunk_hdr->type != SCTP_CID_ASCONF) 886 continue; 887 888 list_del_init(&chunk->list); 889 890 /* Pick the right transport to use. Should always be true for 891 * the first chunk as we don't have a transport by then. 892 */ 893 sctp_outq_select_transport(ctx, chunk); 894 895 switch (chunk->chunk_hdr->type) { 896 /* 6.10 Bundling 897 * ... 898 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN 899 * COMPLETE with any other chunks. [Send them immediately.] 900 */ 901 case SCTP_CID_INIT: 902 case SCTP_CID_INIT_ACK: 903 case SCTP_CID_SHUTDOWN_COMPLETE: 904 error = sctp_packet_singleton(ctx->transport, chunk, 905 ctx->gfp); 906 if (error < 0) { 907 ctx->asoc->base.sk->sk_err = -error; 908 return; 909 } 910 break; 911 912 case SCTP_CID_ABORT: 913 if (sctp_test_T_bit(chunk)) 914 ctx->packet->vtag = ctx->asoc->c.my_vtag; 915 /* fallthru */ 916 917 /* The following chunks are "response" chunks, i.e. 918 * they are generated in response to something we 919 * received. If we are sending these, then we can 920 * send only 1 packet containing these chunks. 921 */ 922 case SCTP_CID_HEARTBEAT_ACK: 923 case SCTP_CID_SHUTDOWN_ACK: 924 case SCTP_CID_COOKIE_ACK: 925 case SCTP_CID_COOKIE_ECHO: 926 case SCTP_CID_ERROR: 927 case SCTP_CID_ECN_CWR: 928 case SCTP_CID_ASCONF_ACK: 929 one_packet = 1; 930 /* Fall through */ 931 932 case SCTP_CID_SACK: 933 case SCTP_CID_HEARTBEAT: 934 case SCTP_CID_SHUTDOWN: 935 case SCTP_CID_ECN_ECNE: 936 case SCTP_CID_ASCONF: 937 case SCTP_CID_FWD_TSN: 938 case SCTP_CID_I_FWD_TSN: 939 case SCTP_CID_RECONF: 940 status = sctp_packet_transmit_chunk(ctx->packet, chunk, 941 one_packet, ctx->gfp); 942 if (status != SCTP_XMIT_OK) { 943 /* put the chunk back */ 944 list_add(&chunk->list, &ctx->q->control_chunk_list); 945 break; 946 } 947 948 ctx->asoc->stats.octrlchunks++; 949 /* PR-SCTP C5) If a FORWARD TSN is sent, the 950 * sender MUST assure that at least one T3-rtx 951 * timer is running. 952 */ 953 if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN || 954 chunk->chunk_hdr->type == SCTP_CID_I_FWD_TSN) { 955 sctp_transport_reset_t3_rtx(ctx->transport); 956 ctx->transport->last_time_sent = jiffies; 957 } 958 959 if (chunk == ctx->asoc->strreset_chunk) 960 sctp_transport_reset_reconf_timer(ctx->transport); 961 962 break; 963 964 default: 965 /* We built a chunk with an illegal type! */ 966 BUG(); 967 } 968 } 969 } 970 971 /* Returns false if new data shouldn't be sent */ 972 static bool sctp_outq_flush_rtx(struct sctp_flush_ctx *ctx, 973 int rtx_timeout) 974 { 975 int error, start_timer = 0; 976 977 if (ctx->asoc->peer.retran_path->state == SCTP_UNCONFIRMED) 978 return false; 979 980 if (ctx->transport != ctx->asoc->peer.retran_path) { 981 /* Switch transports & prepare the packet. */ 982 ctx->transport = ctx->asoc->peer.retran_path; 983 ctx->packet = &ctx->transport->packet; 984 985 if (list_empty(&ctx->transport->send_ready)) 986 list_add_tail(&ctx->transport->send_ready, 987 &ctx->transport_list); 988 989 sctp_packet_config(ctx->packet, ctx->asoc->peer.i.init_tag, 990 ctx->asoc->peer.ecn_capable); 991 } 992 993 error = __sctp_outq_flush_rtx(ctx->q, ctx->packet, rtx_timeout, 994 &start_timer, ctx->gfp); 995 if (error < 0) 996 ctx->asoc->base.sk->sk_err = -error; 997 998 if (start_timer) { 999 sctp_transport_reset_t3_rtx(ctx->transport); 1000 ctx->transport->last_time_sent = jiffies; 1001 } 1002 1003 /* This can happen on COOKIE-ECHO resend. Only 1004 * one chunk can get bundled with a COOKIE-ECHO. 1005 */ 1006 if (ctx->packet->has_cookie_echo) 1007 return false; 1008 1009 /* Don't send new data if there is still data 1010 * waiting to retransmit. 1011 */ 1012 if (!list_empty(&ctx->q->retransmit)) 1013 return false; 1014 1015 return true; 1016 } 1017 1018 static void sctp_outq_flush_data(struct sctp_flush_ctx *ctx, 1019 int rtx_timeout) 1020 { 1021 struct sctp_chunk *chunk; 1022 enum sctp_xmit status; 1023 1024 /* Is it OK to send data chunks? */ 1025 switch (ctx->asoc->state) { 1026 case SCTP_STATE_COOKIE_ECHOED: 1027 /* Only allow bundling when this packet has a COOKIE-ECHO 1028 * chunk. 1029 */ 1030 if (!ctx->packet || !ctx->packet->has_cookie_echo) 1031 return; 1032 1033 /* fall through */ 1034 case SCTP_STATE_ESTABLISHED: 1035 case SCTP_STATE_SHUTDOWN_PENDING: 1036 case SCTP_STATE_SHUTDOWN_RECEIVED: 1037 break; 1038 1039 default: 1040 /* Do nothing. */ 1041 return; 1042 } 1043 1044 /* RFC 2960 6.1 Transmission of DATA Chunks 1045 * 1046 * C) When the time comes for the sender to transmit, 1047 * before sending new DATA chunks, the sender MUST 1048 * first transmit any outstanding DATA chunks which 1049 * are marked for retransmission (limited by the 1050 * current cwnd). 1051 */ 1052 if (!list_empty(&ctx->q->retransmit) && 1053 !sctp_outq_flush_rtx(ctx, rtx_timeout)) 1054 return; 1055 1056 /* Apply Max.Burst limitation to the current transport in 1057 * case it will be used for new data. We are going to 1058 * rest it before we return, but we want to apply the limit 1059 * to the currently queued data. 1060 */ 1061 if (ctx->transport) 1062 sctp_transport_burst_limited(ctx->transport); 1063 1064 /* Finally, transmit new packets. */ 1065 while ((chunk = sctp_outq_dequeue_data(ctx->q)) != NULL) { 1066 __u32 sid = ntohs(chunk->subh.data_hdr->stream); 1067 __u8 stream_state = SCTP_SO(&ctx->asoc->stream, sid)->state; 1068 1069 /* Has this chunk expired? */ 1070 if (sctp_chunk_abandoned(chunk)) { 1071 sctp_sched_dequeue_done(ctx->q, chunk); 1072 sctp_chunk_fail(chunk, 0); 1073 sctp_chunk_free(chunk); 1074 continue; 1075 } 1076 1077 if (stream_state == SCTP_STREAM_CLOSED) { 1078 sctp_outq_head_data(ctx->q, chunk); 1079 break; 1080 } 1081 1082 sctp_outq_select_transport(ctx, chunk); 1083 1084 pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p skb->users:%d\n", 1085 __func__, ctx->q, chunk, chunk && chunk->chunk_hdr ? 1086 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) : 1087 "illegal chunk", ntohl(chunk->subh.data_hdr->tsn), 1088 chunk->skb ? chunk->skb->head : NULL, chunk->skb ? 1089 refcount_read(&chunk->skb->users) : -1); 1090 1091 /* Add the chunk to the packet. */ 1092 status = sctp_packet_transmit_chunk(ctx->packet, chunk, 0, 1093 ctx->gfp); 1094 if (status != SCTP_XMIT_OK) { 1095 /* We could not append this chunk, so put 1096 * the chunk back on the output queue. 1097 */ 1098 pr_debug("%s: could not transmit tsn:0x%x, status:%d\n", 1099 __func__, ntohl(chunk->subh.data_hdr->tsn), 1100 status); 1101 1102 sctp_outq_head_data(ctx->q, chunk); 1103 break; 1104 } 1105 1106 /* The sender is in the SHUTDOWN-PENDING state, 1107 * The sender MAY set the I-bit in the DATA 1108 * chunk header. 1109 */ 1110 if (ctx->asoc->state == SCTP_STATE_SHUTDOWN_PENDING) 1111 chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM; 1112 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED) 1113 ctx->asoc->stats.ouodchunks++; 1114 else 1115 ctx->asoc->stats.oodchunks++; 1116 1117 /* Only now it's safe to consider this 1118 * chunk as sent, sched-wise. 1119 */ 1120 sctp_sched_dequeue_done(ctx->q, chunk); 1121 1122 list_add_tail(&chunk->transmitted_list, 1123 &ctx->transport->transmitted); 1124 1125 sctp_transport_reset_t3_rtx(ctx->transport); 1126 ctx->transport->last_time_sent = jiffies; 1127 1128 /* Only let one DATA chunk get bundled with a 1129 * COOKIE-ECHO chunk. 1130 */ 1131 if (ctx->packet->has_cookie_echo) 1132 break; 1133 } 1134 } 1135 1136 static void sctp_outq_flush_transports(struct sctp_flush_ctx *ctx) 1137 { 1138 struct list_head *ltransport; 1139 struct sctp_packet *packet; 1140 struct sctp_transport *t; 1141 int error = 0; 1142 1143 while ((ltransport = sctp_list_dequeue(&ctx->transport_list)) != NULL) { 1144 t = list_entry(ltransport, struct sctp_transport, send_ready); 1145 packet = &t->packet; 1146 if (!sctp_packet_empty(packet)) { 1147 error = sctp_packet_transmit(packet, ctx->gfp); 1148 if (error < 0) 1149 ctx->q->asoc->base.sk->sk_err = -error; 1150 } 1151 1152 /* Clear the burst limited state, if any */ 1153 sctp_transport_burst_reset(t); 1154 } 1155 } 1156 1157 /* Try to flush an outqueue. 1158 * 1159 * Description: Send everything in q which we legally can, subject to 1160 * congestion limitations. 1161 * * Note: This function can be called from multiple contexts so appropriate 1162 * locking concerns must be made. Today we use the sock lock to protect 1163 * this function. 1164 */ 1165 1166 static void sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp) 1167 { 1168 struct sctp_flush_ctx ctx = { 1169 .q = q, 1170 .transport = NULL, 1171 .transport_list = LIST_HEAD_INIT(ctx.transport_list), 1172 .asoc = q->asoc, 1173 .packet = NULL, 1174 .gfp = gfp, 1175 }; 1176 1177 /* 6.10 Bundling 1178 * ... 1179 * When bundling control chunks with DATA chunks, an 1180 * endpoint MUST place control chunks first in the outbound 1181 * SCTP packet. The transmitter MUST transmit DATA chunks 1182 * within a SCTP packet in increasing order of TSN. 1183 * ... 1184 */ 1185 1186 sctp_outq_flush_ctrl(&ctx); 1187 1188 if (q->asoc->src_out_of_asoc_ok) 1189 goto sctp_flush_out; 1190 1191 sctp_outq_flush_data(&ctx, rtx_timeout); 1192 1193 sctp_flush_out: 1194 1195 sctp_outq_flush_transports(&ctx); 1196 } 1197 1198 /* Update unack_data based on the incoming SACK chunk */ 1199 static void sctp_sack_update_unack_data(struct sctp_association *assoc, 1200 struct sctp_sackhdr *sack) 1201 { 1202 union sctp_sack_variable *frags; 1203 __u16 unack_data; 1204 int i; 1205 1206 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1; 1207 1208 frags = sack->variable; 1209 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) { 1210 unack_data -= ((ntohs(frags[i].gab.end) - 1211 ntohs(frags[i].gab.start) + 1)); 1212 } 1213 1214 assoc->unack_data = unack_data; 1215 } 1216 1217 /* This is where we REALLY process a SACK. 1218 * 1219 * Process the SACK against the outqueue. Mostly, this just frees 1220 * things off the transmitted queue. 1221 */ 1222 int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk) 1223 { 1224 struct sctp_association *asoc = q->asoc; 1225 struct sctp_sackhdr *sack = chunk->subh.sack_hdr; 1226 struct sctp_transport *transport; 1227 struct sctp_chunk *tchunk = NULL; 1228 struct list_head *lchunk, *transport_list, *temp; 1229 union sctp_sack_variable *frags = sack->variable; 1230 __u32 sack_ctsn, ctsn, tsn; 1231 __u32 highest_tsn, highest_new_tsn; 1232 __u32 sack_a_rwnd; 1233 unsigned int outstanding; 1234 struct sctp_transport *primary = asoc->peer.primary_path; 1235 int count_of_newacks = 0; 1236 int gap_ack_blocks; 1237 u8 accum_moved = 0; 1238 1239 /* Grab the association's destination address list. */ 1240 transport_list = &asoc->peer.transport_addr_list; 1241 1242 /* SCTP path tracepoint for congestion control debugging. */ 1243 if (trace_sctp_probe_path_enabled()) { 1244 list_for_each_entry(transport, transport_list, transports) 1245 trace_sctp_probe_path(transport, asoc); 1246 } 1247 1248 sack_ctsn = ntohl(sack->cum_tsn_ack); 1249 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks); 1250 asoc->stats.gapcnt += gap_ack_blocks; 1251 /* 1252 * SFR-CACC algorithm: 1253 * On receipt of a SACK the sender SHOULD execute the 1254 * following statements. 1255 * 1256 * 1) If the cumulative ack in the SACK passes next tsn_at_change 1257 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be 1258 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for 1259 * all destinations. 1260 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE 1261 * is set the receiver of the SACK MUST take the following actions: 1262 * 1263 * A) Initialize the cacc_saw_newack to 0 for all destination 1264 * addresses. 1265 * 1266 * Only bother if changeover_active is set. Otherwise, this is 1267 * totally suboptimal to do on every SACK. 1268 */ 1269 if (primary->cacc.changeover_active) { 1270 u8 clear_cycling = 0; 1271 1272 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) { 1273 primary->cacc.changeover_active = 0; 1274 clear_cycling = 1; 1275 } 1276 1277 if (clear_cycling || gap_ack_blocks) { 1278 list_for_each_entry(transport, transport_list, 1279 transports) { 1280 if (clear_cycling) 1281 transport->cacc.cycling_changeover = 0; 1282 if (gap_ack_blocks) 1283 transport->cacc.cacc_saw_newack = 0; 1284 } 1285 } 1286 } 1287 1288 /* Get the highest TSN in the sack. */ 1289 highest_tsn = sack_ctsn; 1290 if (gap_ack_blocks) 1291 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end); 1292 1293 if (TSN_lt(asoc->highest_sacked, highest_tsn)) 1294 asoc->highest_sacked = highest_tsn; 1295 1296 highest_new_tsn = sack_ctsn; 1297 1298 /* Run through the retransmit queue. Credit bytes received 1299 * and free those chunks that we can. 1300 */ 1301 sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn); 1302 1303 /* Run through the transmitted queue. 1304 * Credit bytes received and free those chunks which we can. 1305 * 1306 * This is a MASSIVE candidate for optimization. 1307 */ 1308 list_for_each_entry(transport, transport_list, transports) { 1309 sctp_check_transmitted(q, &transport->transmitted, 1310 transport, &chunk->source, sack, 1311 &highest_new_tsn); 1312 /* 1313 * SFR-CACC algorithm: 1314 * C) Let count_of_newacks be the number of 1315 * destinations for which cacc_saw_newack is set. 1316 */ 1317 if (transport->cacc.cacc_saw_newack) 1318 count_of_newacks++; 1319 } 1320 1321 /* Move the Cumulative TSN Ack Point if appropriate. */ 1322 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) { 1323 asoc->ctsn_ack_point = sack_ctsn; 1324 accum_moved = 1; 1325 } 1326 1327 if (gap_ack_blocks) { 1328 1329 if (asoc->fast_recovery && accum_moved) 1330 highest_new_tsn = highest_tsn; 1331 1332 list_for_each_entry(transport, transport_list, transports) 1333 sctp_mark_missing(q, &transport->transmitted, transport, 1334 highest_new_tsn, count_of_newacks); 1335 } 1336 1337 /* Update unack_data field in the assoc. */ 1338 sctp_sack_update_unack_data(asoc, sack); 1339 1340 ctsn = asoc->ctsn_ack_point; 1341 1342 /* Throw away stuff rotting on the sack queue. */ 1343 list_for_each_safe(lchunk, temp, &q->sacked) { 1344 tchunk = list_entry(lchunk, struct sctp_chunk, 1345 transmitted_list); 1346 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1347 if (TSN_lte(tsn, ctsn)) { 1348 list_del_init(&tchunk->transmitted_list); 1349 if (asoc->peer.prsctp_capable && 1350 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags)) 1351 asoc->sent_cnt_removable--; 1352 sctp_chunk_free(tchunk); 1353 } 1354 } 1355 1356 /* ii) Set rwnd equal to the newly received a_rwnd minus the 1357 * number of bytes still outstanding after processing the 1358 * Cumulative TSN Ack and the Gap Ack Blocks. 1359 */ 1360 1361 sack_a_rwnd = ntohl(sack->a_rwnd); 1362 asoc->peer.zero_window_announced = !sack_a_rwnd; 1363 outstanding = q->outstanding_bytes; 1364 1365 if (outstanding < sack_a_rwnd) 1366 sack_a_rwnd -= outstanding; 1367 else 1368 sack_a_rwnd = 0; 1369 1370 asoc->peer.rwnd = sack_a_rwnd; 1371 1372 asoc->stream.si->generate_ftsn(q, sack_ctsn); 1373 1374 pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__, sack_ctsn); 1375 pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, " 1376 "advertised peer ack point:0x%x\n", __func__, asoc, ctsn, 1377 asoc->adv_peer_ack_point); 1378 1379 return sctp_outq_is_empty(q); 1380 } 1381 1382 /* Is the outqueue empty? 1383 * The queue is empty when we have not pending data, no in-flight data 1384 * and nothing pending retransmissions. 1385 */ 1386 int sctp_outq_is_empty(const struct sctp_outq *q) 1387 { 1388 return q->out_qlen == 0 && q->outstanding_bytes == 0 && 1389 list_empty(&q->retransmit); 1390 } 1391 1392 /******************************************************************** 1393 * 2nd Level Abstractions 1394 ********************************************************************/ 1395 1396 /* Go through a transport's transmitted list or the association's retransmit 1397 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked. 1398 * The retransmit list will not have an associated transport. 1399 * 1400 * I added coherent debug information output. --xguo 1401 * 1402 * Instead of printing 'sacked' or 'kept' for each TSN on the 1403 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5. 1404 * KEPT TSN6-TSN7, etc. 1405 */ 1406 static void sctp_check_transmitted(struct sctp_outq *q, 1407 struct list_head *transmitted_queue, 1408 struct sctp_transport *transport, 1409 union sctp_addr *saddr, 1410 struct sctp_sackhdr *sack, 1411 __u32 *highest_new_tsn_in_sack) 1412 { 1413 struct list_head *lchunk; 1414 struct sctp_chunk *tchunk; 1415 struct list_head tlist; 1416 __u32 tsn; 1417 __u32 sack_ctsn; 1418 __u32 rtt; 1419 __u8 restart_timer = 0; 1420 int bytes_acked = 0; 1421 int migrate_bytes = 0; 1422 bool forward_progress = false; 1423 1424 sack_ctsn = ntohl(sack->cum_tsn_ack); 1425 1426 INIT_LIST_HEAD(&tlist); 1427 1428 /* The while loop will skip empty transmitted queues. */ 1429 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) { 1430 tchunk = list_entry(lchunk, struct sctp_chunk, 1431 transmitted_list); 1432 1433 if (sctp_chunk_abandoned(tchunk)) { 1434 /* Move the chunk to abandoned list. */ 1435 sctp_insert_list(&q->abandoned, lchunk); 1436 1437 /* If this chunk has not been acked, stop 1438 * considering it as 'outstanding'. 1439 */ 1440 if (transmitted_queue != &q->retransmit && 1441 !tchunk->tsn_gap_acked) { 1442 if (tchunk->transport) 1443 tchunk->transport->flight_size -= 1444 sctp_data_size(tchunk); 1445 q->outstanding_bytes -= sctp_data_size(tchunk); 1446 } 1447 continue; 1448 } 1449 1450 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1451 if (sctp_acked(sack, tsn)) { 1452 /* If this queue is the retransmit queue, the 1453 * retransmit timer has already reclaimed 1454 * the outstanding bytes for this chunk, so only 1455 * count bytes associated with a transport. 1456 */ 1457 if (transport && !tchunk->tsn_gap_acked) { 1458 /* If this chunk is being used for RTT 1459 * measurement, calculate the RTT and update 1460 * the RTO using this value. 1461 * 1462 * 6.3.1 C5) Karn's algorithm: RTT measurements 1463 * MUST NOT be made using packets that were 1464 * retransmitted (and thus for which it is 1465 * ambiguous whether the reply was for the 1466 * first instance of the packet or a later 1467 * instance). 1468 */ 1469 if (!sctp_chunk_retransmitted(tchunk) && 1470 tchunk->rtt_in_progress) { 1471 tchunk->rtt_in_progress = 0; 1472 rtt = jiffies - tchunk->sent_at; 1473 sctp_transport_update_rto(transport, 1474 rtt); 1475 } 1476 1477 if (TSN_lte(tsn, sack_ctsn)) { 1478 /* 1479 * SFR-CACC algorithm: 1480 * 2) If the SACK contains gap acks 1481 * and the flag CHANGEOVER_ACTIVE is 1482 * set the receiver of the SACK MUST 1483 * take the following action: 1484 * 1485 * B) For each TSN t being acked that 1486 * has not been acked in any SACK so 1487 * far, set cacc_saw_newack to 1 for 1488 * the destination that the TSN was 1489 * sent to. 1490 */ 1491 if (sack->num_gap_ack_blocks && 1492 q->asoc->peer.primary_path->cacc. 1493 changeover_active) 1494 transport->cacc.cacc_saw_newack 1495 = 1; 1496 } 1497 } 1498 1499 /* If the chunk hasn't been marked as ACKED, 1500 * mark it and account bytes_acked if the 1501 * chunk had a valid transport (it will not 1502 * have a transport if ASCONF had deleted it 1503 * while DATA was outstanding). 1504 */ 1505 if (!tchunk->tsn_gap_acked) { 1506 tchunk->tsn_gap_acked = 1; 1507 if (TSN_lt(*highest_new_tsn_in_sack, tsn)) 1508 *highest_new_tsn_in_sack = tsn; 1509 bytes_acked += sctp_data_size(tchunk); 1510 if (!tchunk->transport) 1511 migrate_bytes += sctp_data_size(tchunk); 1512 forward_progress = true; 1513 } 1514 1515 if (TSN_lte(tsn, sack_ctsn)) { 1516 /* RFC 2960 6.3.2 Retransmission Timer Rules 1517 * 1518 * R3) Whenever a SACK is received 1519 * that acknowledges the DATA chunk 1520 * with the earliest outstanding TSN 1521 * for that address, restart T3-rtx 1522 * timer for that address with its 1523 * current RTO. 1524 */ 1525 restart_timer = 1; 1526 forward_progress = true; 1527 1528 list_add_tail(&tchunk->transmitted_list, 1529 &q->sacked); 1530 } else { 1531 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2 1532 * M2) Each time a SACK arrives reporting 1533 * 'Stray DATA chunk(s)' record the highest TSN 1534 * reported as newly acknowledged, call this 1535 * value 'HighestTSNinSack'. A newly 1536 * acknowledged DATA chunk is one not 1537 * previously acknowledged in a SACK. 1538 * 1539 * When the SCTP sender of data receives a SACK 1540 * chunk that acknowledges, for the first time, 1541 * the receipt of a DATA chunk, all the still 1542 * unacknowledged DATA chunks whose TSN is 1543 * older than that newly acknowledged DATA 1544 * chunk, are qualified as 'Stray DATA chunks'. 1545 */ 1546 list_add_tail(lchunk, &tlist); 1547 } 1548 } else { 1549 if (tchunk->tsn_gap_acked) { 1550 pr_debug("%s: receiver reneged on data TSN:0x%x\n", 1551 __func__, tsn); 1552 1553 tchunk->tsn_gap_acked = 0; 1554 1555 if (tchunk->transport) 1556 bytes_acked -= sctp_data_size(tchunk); 1557 1558 /* RFC 2960 6.3.2 Retransmission Timer Rules 1559 * 1560 * R4) Whenever a SACK is received missing a 1561 * TSN that was previously acknowledged via a 1562 * Gap Ack Block, start T3-rtx for the 1563 * destination address to which the DATA 1564 * chunk was originally 1565 * transmitted if it is not already running. 1566 */ 1567 restart_timer = 1; 1568 } 1569 1570 list_add_tail(lchunk, &tlist); 1571 } 1572 } 1573 1574 if (transport) { 1575 if (bytes_acked) { 1576 struct sctp_association *asoc = transport->asoc; 1577 1578 /* We may have counted DATA that was migrated 1579 * to this transport due to DEL-IP operation. 1580 * Subtract those bytes, since the were never 1581 * send on this transport and shouldn't be 1582 * credited to this transport. 1583 */ 1584 bytes_acked -= migrate_bytes; 1585 1586 /* 8.2. When an outstanding TSN is acknowledged, 1587 * the endpoint shall clear the error counter of 1588 * the destination transport address to which the 1589 * DATA chunk was last sent. 1590 * The association's overall error counter is 1591 * also cleared. 1592 */ 1593 transport->error_count = 0; 1594 transport->asoc->overall_error_count = 0; 1595 forward_progress = true; 1596 1597 /* 1598 * While in SHUTDOWN PENDING, we may have started 1599 * the T5 shutdown guard timer after reaching the 1600 * retransmission limit. Stop that timer as soon 1601 * as the receiver acknowledged any data. 1602 */ 1603 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING && 1604 del_timer(&asoc->timers 1605 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD])) 1606 sctp_association_put(asoc); 1607 1608 /* Mark the destination transport address as 1609 * active if it is not so marked. 1610 */ 1611 if ((transport->state == SCTP_INACTIVE || 1612 transport->state == SCTP_UNCONFIRMED) && 1613 sctp_cmp_addr_exact(&transport->ipaddr, saddr)) { 1614 sctp_assoc_control_transport( 1615 transport->asoc, 1616 transport, 1617 SCTP_TRANSPORT_UP, 1618 SCTP_RECEIVED_SACK); 1619 } 1620 1621 sctp_transport_raise_cwnd(transport, sack_ctsn, 1622 bytes_acked); 1623 1624 transport->flight_size -= bytes_acked; 1625 if (transport->flight_size == 0) 1626 transport->partial_bytes_acked = 0; 1627 q->outstanding_bytes -= bytes_acked + migrate_bytes; 1628 } else { 1629 /* RFC 2960 6.1, sctpimpguide-06 2.15.2 1630 * When a sender is doing zero window probing, it 1631 * should not timeout the association if it continues 1632 * to receive new packets from the receiver. The 1633 * reason is that the receiver MAY keep its window 1634 * closed for an indefinite time. 1635 * A sender is doing zero window probing when the 1636 * receiver's advertised window is zero, and there is 1637 * only one data chunk in flight to the receiver. 1638 * 1639 * Allow the association to timeout while in SHUTDOWN 1640 * PENDING or SHUTDOWN RECEIVED in case the receiver 1641 * stays in zero window mode forever. 1642 */ 1643 if (!q->asoc->peer.rwnd && 1644 !list_empty(&tlist) && 1645 (sack_ctsn+2 == q->asoc->next_tsn) && 1646 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) { 1647 pr_debug("%s: sack received for zero window " 1648 "probe:%u\n", __func__, sack_ctsn); 1649 1650 q->asoc->overall_error_count = 0; 1651 transport->error_count = 0; 1652 } 1653 } 1654 1655 /* RFC 2960 6.3.2 Retransmission Timer Rules 1656 * 1657 * R2) Whenever all outstanding data sent to an address have 1658 * been acknowledged, turn off the T3-rtx timer of that 1659 * address. 1660 */ 1661 if (!transport->flight_size) { 1662 if (del_timer(&transport->T3_rtx_timer)) 1663 sctp_transport_put(transport); 1664 } else if (restart_timer) { 1665 if (!mod_timer(&transport->T3_rtx_timer, 1666 jiffies + transport->rto)) 1667 sctp_transport_hold(transport); 1668 } 1669 1670 if (forward_progress) { 1671 if (transport->dst) 1672 sctp_transport_dst_confirm(transport); 1673 } 1674 } 1675 1676 list_splice(&tlist, transmitted_queue); 1677 } 1678 1679 /* Mark chunks as missing and consequently may get retransmitted. */ 1680 static void sctp_mark_missing(struct sctp_outq *q, 1681 struct list_head *transmitted_queue, 1682 struct sctp_transport *transport, 1683 __u32 highest_new_tsn_in_sack, 1684 int count_of_newacks) 1685 { 1686 struct sctp_chunk *chunk; 1687 __u32 tsn; 1688 char do_fast_retransmit = 0; 1689 struct sctp_association *asoc = q->asoc; 1690 struct sctp_transport *primary = asoc->peer.primary_path; 1691 1692 list_for_each_entry(chunk, transmitted_queue, transmitted_list) { 1693 1694 tsn = ntohl(chunk->subh.data_hdr->tsn); 1695 1696 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all 1697 * 'Unacknowledged TSN's', if the TSN number of an 1698 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack' 1699 * value, increment the 'TSN.Missing.Report' count on that 1700 * chunk if it has NOT been fast retransmitted or marked for 1701 * fast retransmit already. 1702 */ 1703 if (chunk->fast_retransmit == SCTP_CAN_FRTX && 1704 !chunk->tsn_gap_acked && 1705 TSN_lt(tsn, highest_new_tsn_in_sack)) { 1706 1707 /* SFR-CACC may require us to skip marking 1708 * this chunk as missing. 1709 */ 1710 if (!transport || !sctp_cacc_skip(primary, 1711 chunk->transport, 1712 count_of_newacks, tsn)) { 1713 chunk->tsn_missing_report++; 1714 1715 pr_debug("%s: tsn:0x%x missing counter:%d\n", 1716 __func__, tsn, chunk->tsn_missing_report); 1717 } 1718 } 1719 /* 1720 * M4) If any DATA chunk is found to have a 1721 * 'TSN.Missing.Report' 1722 * value larger than or equal to 3, mark that chunk for 1723 * retransmission and start the fast retransmit procedure. 1724 */ 1725 1726 if (chunk->tsn_missing_report >= 3) { 1727 chunk->fast_retransmit = SCTP_NEED_FRTX; 1728 do_fast_retransmit = 1; 1729 } 1730 } 1731 1732 if (transport) { 1733 if (do_fast_retransmit) 1734 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX); 1735 1736 pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, " 1737 "flight_size:%d, pba:%d\n", __func__, transport, 1738 transport->cwnd, transport->ssthresh, 1739 transport->flight_size, transport->partial_bytes_acked); 1740 } 1741 } 1742 1743 /* Is the given TSN acked by this packet? */ 1744 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn) 1745 { 1746 __u32 ctsn = ntohl(sack->cum_tsn_ack); 1747 union sctp_sack_variable *frags; 1748 __u16 tsn_offset, blocks; 1749 int i; 1750 1751 if (TSN_lte(tsn, ctsn)) 1752 goto pass; 1753 1754 /* 3.3.4 Selective Acknowledgment (SACK) (3): 1755 * 1756 * Gap Ack Blocks: 1757 * These fields contain the Gap Ack Blocks. They are repeated 1758 * for each Gap Ack Block up to the number of Gap Ack Blocks 1759 * defined in the Number of Gap Ack Blocks field. All DATA 1760 * chunks with TSNs greater than or equal to (Cumulative TSN 1761 * Ack + Gap Ack Block Start) and less than or equal to 1762 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack 1763 * Block are assumed to have been received correctly. 1764 */ 1765 1766 frags = sack->variable; 1767 blocks = ntohs(sack->num_gap_ack_blocks); 1768 tsn_offset = tsn - ctsn; 1769 for (i = 0; i < blocks; ++i) { 1770 if (tsn_offset >= ntohs(frags[i].gab.start) && 1771 tsn_offset <= ntohs(frags[i].gab.end)) 1772 goto pass; 1773 } 1774 1775 return 0; 1776 pass: 1777 return 1; 1778 } 1779 1780 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist, 1781 int nskips, __be16 stream) 1782 { 1783 int i; 1784 1785 for (i = 0; i < nskips; i++) { 1786 if (skiplist[i].stream == stream) 1787 return i; 1788 } 1789 return i; 1790 } 1791 1792 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */ 1793 void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn) 1794 { 1795 struct sctp_association *asoc = q->asoc; 1796 struct sctp_chunk *ftsn_chunk = NULL; 1797 struct sctp_fwdtsn_skip ftsn_skip_arr[10]; 1798 int nskips = 0; 1799 int skip_pos = 0; 1800 __u32 tsn; 1801 struct sctp_chunk *chunk; 1802 struct list_head *lchunk, *temp; 1803 1804 if (!asoc->peer.prsctp_capable) 1805 return; 1806 1807 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the 1808 * received SACK. 1809 * 1810 * If (Advanced.Peer.Ack.Point < SackCumAck), then update 1811 * Advanced.Peer.Ack.Point to be equal to SackCumAck. 1812 */ 1813 if (TSN_lt(asoc->adv_peer_ack_point, ctsn)) 1814 asoc->adv_peer_ack_point = ctsn; 1815 1816 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point" 1817 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as 1818 * the chunk next in the out-queue space is marked as "abandoned" as 1819 * shown in the following example: 1820 * 1821 * Assuming that a SACK arrived with the Cumulative TSN ACK 102 1822 * and the Advanced.Peer.Ack.Point is updated to this value: 1823 * 1824 * out-queue at the end of ==> out-queue after Adv.Ack.Point 1825 * normal SACK processing local advancement 1826 * ... ... 1827 * Adv.Ack.Pt-> 102 acked 102 acked 1828 * 103 abandoned 103 abandoned 1829 * 104 abandoned Adv.Ack.P-> 104 abandoned 1830 * 105 105 1831 * 106 acked 106 acked 1832 * ... ... 1833 * 1834 * In this example, the data sender successfully advanced the 1835 * "Advanced.Peer.Ack.Point" from 102 to 104 locally. 1836 */ 1837 list_for_each_safe(lchunk, temp, &q->abandoned) { 1838 chunk = list_entry(lchunk, struct sctp_chunk, 1839 transmitted_list); 1840 tsn = ntohl(chunk->subh.data_hdr->tsn); 1841 1842 /* Remove any chunks in the abandoned queue that are acked by 1843 * the ctsn. 1844 */ 1845 if (TSN_lte(tsn, ctsn)) { 1846 list_del_init(lchunk); 1847 sctp_chunk_free(chunk); 1848 } else { 1849 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) { 1850 asoc->adv_peer_ack_point = tsn; 1851 if (chunk->chunk_hdr->flags & 1852 SCTP_DATA_UNORDERED) 1853 continue; 1854 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0], 1855 nskips, 1856 chunk->subh.data_hdr->stream); 1857 ftsn_skip_arr[skip_pos].stream = 1858 chunk->subh.data_hdr->stream; 1859 ftsn_skip_arr[skip_pos].ssn = 1860 chunk->subh.data_hdr->ssn; 1861 if (skip_pos == nskips) 1862 nskips++; 1863 if (nskips == 10) 1864 break; 1865 } else 1866 break; 1867 } 1868 } 1869 1870 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point" 1871 * is greater than the Cumulative TSN ACK carried in the received 1872 * SACK, the data sender MUST send the data receiver a FORWARD TSN 1873 * chunk containing the latest value of the 1874 * "Advanced.Peer.Ack.Point". 1875 * 1876 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD 1877 * list each stream and sequence number in the forwarded TSN. This 1878 * information will enable the receiver to easily find any 1879 * stranded TSN's waiting on stream reorder queues. Each stream 1880 * SHOULD only be reported once; this means that if multiple 1881 * abandoned messages occur in the same stream then only the 1882 * highest abandoned stream sequence number is reported. If the 1883 * total size of the FORWARD TSN does NOT fit in a single MTU then 1884 * the sender of the FORWARD TSN SHOULD lower the 1885 * Advanced.Peer.Ack.Point to the last TSN that will fit in a 1886 * single MTU. 1887 */ 1888 if (asoc->adv_peer_ack_point > ctsn) 1889 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point, 1890 nskips, &ftsn_skip_arr[0]); 1891 1892 if (ftsn_chunk) { 1893 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list); 1894 SCTP_INC_STATS(asoc->base.net, SCTP_MIB_OUTCTRLCHUNKS); 1895 } 1896 } 1897