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. */ 42static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn); 43static 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 50static 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 56static void sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp); 57 58/* Add data to the front of the queue. */ 59static 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. */ 74static 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. */ 80static 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 */ 100static 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 */ 116static 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 */ 132static 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 */ 153static 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 */ 175static 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 */ 191void 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 */ 206static 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 267void 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. */ 274void 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. */ 281void 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 */ 319static 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 342static 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 382static 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 struct sctp_stream_out *sout; 388 389 q->sched->unsched_all(&asoc->stream); 390 391 list_for_each_entry_safe(chk, temp, &q->out_chunk_list, list) { 392 if (!chk->msg->abandoned && 393 (!(chk->chunk_hdr->flags & SCTP_DATA_FIRST_FRAG) || 394 !SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) || 395 chk->sinfo.sinfo_timetolive <= sinfo->sinfo_timetolive)) 396 continue; 397 398 chk->msg->abandoned = 1; 399 sctp_sched_dequeue_common(q, chk); 400 asoc->sent_cnt_removable--; 401 asoc->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++; 402 403 sout = SCTP_SO(&asoc->stream, chk->sinfo.sinfo_stream); 404 sout->ext->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++; 405 406 /* clear out_curr if all frag chunks are pruned */ 407 if (asoc->stream.out_curr == sout && 408 list_is_last(&chk->frag_list, &chk->msg->chunks)) 409 asoc->stream.out_curr = NULL; 410 411 msg_len -= chk->skb->truesize + sizeof(struct sctp_chunk); 412 sctp_chunk_free(chk); 413 if (msg_len <= 0) 414 break; 415 } 416 417 q->sched->sched_all(&asoc->stream); 418 419 return msg_len; 420} 421 422/* Abandon the chunks according their priorities */ 423void sctp_prsctp_prune(struct sctp_association *asoc, 424 struct sctp_sndrcvinfo *sinfo, int msg_len) 425{ 426 struct sctp_transport *transport; 427 428 if (!asoc->peer.prsctp_capable || !asoc->sent_cnt_removable) 429 return; 430 431 msg_len = sctp_prsctp_prune_sent(asoc, sinfo, 432 &asoc->outqueue.retransmit, 433 msg_len); 434 if (msg_len <= 0) 435 return; 436 437 list_for_each_entry(transport, &asoc->peer.transport_addr_list, 438 transports) { 439 msg_len = sctp_prsctp_prune_sent(asoc, sinfo, 440 &transport->transmitted, 441 msg_len); 442 if (msg_len <= 0) 443 return; 444 } 445 446 sctp_prsctp_prune_unsent(asoc, sinfo, msg_len); 447} 448 449/* Mark all the eligible packets on a transport for retransmission. */ 450void sctp_retransmit_mark(struct sctp_outq *q, 451 struct sctp_transport *transport, 452 __u8 reason) 453{ 454 struct list_head *lchunk, *ltemp; 455 struct sctp_chunk *chunk; 456 457 /* Walk through the specified transmitted queue. */ 458 list_for_each_safe(lchunk, ltemp, &transport->transmitted) { 459 chunk = list_entry(lchunk, struct sctp_chunk, 460 transmitted_list); 461 462 /* If the chunk is abandoned, move it to abandoned list. */ 463 if (sctp_chunk_abandoned(chunk)) { 464 list_del_init(lchunk); 465 sctp_insert_list(&q->abandoned, lchunk); 466 467 /* If this chunk has not been previousely acked, 468 * stop considering it 'outstanding'. Our peer 469 * will most likely never see it since it will 470 * not be retransmitted 471 */ 472 if (!chunk->tsn_gap_acked) { 473 if (chunk->transport) 474 chunk->transport->flight_size -= 475 sctp_data_size(chunk); 476 q->outstanding_bytes -= sctp_data_size(chunk); 477 q->asoc->peer.rwnd += sctp_data_size(chunk); 478 } 479 continue; 480 } 481 482 /* If we are doing retransmission due to a timeout or pmtu 483 * discovery, only the chunks that are not yet acked should 484 * be added to the retransmit queue. 485 */ 486 if ((reason == SCTP_RTXR_FAST_RTX && 487 (chunk->fast_retransmit == SCTP_NEED_FRTX)) || 488 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) { 489 /* RFC 2960 6.2.1 Processing a Received SACK 490 * 491 * C) Any time a DATA chunk is marked for 492 * retransmission (via either T3-rtx timer expiration 493 * (Section 6.3.3) or via fast retransmit 494 * (Section 7.2.4)), add the data size of those 495 * chunks to the rwnd. 496 */ 497 q->asoc->peer.rwnd += sctp_data_size(chunk); 498 q->outstanding_bytes -= sctp_data_size(chunk); 499 if (chunk->transport) 500 transport->flight_size -= sctp_data_size(chunk); 501 502 /* sctpimpguide-05 Section 2.8.2 503 * M5) If a T3-rtx timer expires, the 504 * 'TSN.Missing.Report' of all affected TSNs is set 505 * to 0. 506 */ 507 chunk->tsn_missing_report = 0; 508 509 /* If a chunk that is being used for RTT measurement 510 * has to be retransmitted, we cannot use this chunk 511 * anymore for RTT measurements. Reset rto_pending so 512 * that a new RTT measurement is started when a new 513 * data chunk is sent. 514 */ 515 if (chunk->rtt_in_progress) { 516 chunk->rtt_in_progress = 0; 517 transport->rto_pending = 0; 518 } 519 520 /* Move the chunk to the retransmit queue. The chunks 521 * on the retransmit queue are always kept in order. 522 */ 523 list_del_init(lchunk); 524 sctp_insert_list(&q->retransmit, lchunk); 525 } 526 } 527 528 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, " 529 "flight_size:%d, pba:%d\n", __func__, transport, reason, 530 transport->cwnd, transport->ssthresh, transport->flight_size, 531 transport->partial_bytes_acked); 532} 533 534/* Mark all the eligible packets on a transport for retransmission and force 535 * one packet out. 536 */ 537void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport, 538 enum sctp_retransmit_reason reason) 539{ 540 struct net *net = q->asoc->base.net; 541 542 switch (reason) { 543 case SCTP_RTXR_T3_RTX: 544 SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS); 545 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX); 546 /* Update the retran path if the T3-rtx timer has expired for 547 * the current retran path. 548 */ 549 if (transport == transport->asoc->peer.retran_path) 550 sctp_assoc_update_retran_path(transport->asoc); 551 transport->asoc->rtx_data_chunks += 552 transport->asoc->unack_data; 553 break; 554 case SCTP_RTXR_FAST_RTX: 555 SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS); 556 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX); 557 q->fast_rtx = 1; 558 break; 559 case SCTP_RTXR_PMTUD: 560 SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS); 561 break; 562 case SCTP_RTXR_T1_RTX: 563 SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS); 564 transport->asoc->init_retries++; 565 break; 566 default: 567 BUG(); 568 } 569 570 sctp_retransmit_mark(q, transport, reason); 571 572 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination, 573 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by 574 * following the procedures outlined in C1 - C5. 575 */ 576 if (reason == SCTP_RTXR_T3_RTX) 577 q->asoc->stream.si->generate_ftsn(q, q->asoc->ctsn_ack_point); 578 579 /* Flush the queues only on timeout, since fast_rtx is only 580 * triggered during sack processing and the queue 581 * will be flushed at the end. 582 */ 583 if (reason != SCTP_RTXR_FAST_RTX) 584 sctp_outq_flush(q, /* rtx_timeout */ 1, GFP_ATOMIC); 585} 586 587/* 588 * Transmit DATA chunks on the retransmit queue. Upon return from 589 * __sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which 590 * need to be transmitted by the caller. 591 * We assume that pkt->transport has already been set. 592 * 593 * The return value is a normal kernel error return value. 594 */ 595static int __sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt, 596 int rtx_timeout, int *start_timer, gfp_t gfp) 597{ 598 struct sctp_transport *transport = pkt->transport; 599 struct sctp_chunk *chunk, *chunk1; 600 struct list_head *lqueue; 601 enum sctp_xmit status; 602 int error = 0; 603 int timer = 0; 604 int done = 0; 605 int fast_rtx; 606 607 lqueue = &q->retransmit; 608 fast_rtx = q->fast_rtx; 609 610 /* This loop handles time-out retransmissions, fast retransmissions, 611 * and retransmissions due to opening of whindow. 612 * 613 * RFC 2960 6.3.3 Handle T3-rtx Expiration 614 * 615 * E3) Determine how many of the earliest (i.e., lowest TSN) 616 * outstanding DATA chunks for the address for which the 617 * T3-rtx has expired will fit into a single packet, subject 618 * to the MTU constraint for the path corresponding to the 619 * destination transport address to which the retransmission 620 * is being sent (this may be different from the address for 621 * which the timer expires [see Section 6.4]). Call this value 622 * K. Bundle and retransmit those K DATA chunks in a single 623 * packet to the destination endpoint. 624 * 625 * [Just to be painfully clear, if we are retransmitting 626 * because a timeout just happened, we should send only ONE 627 * packet of retransmitted data.] 628 * 629 * For fast retransmissions we also send only ONE packet. However, 630 * if we are just flushing the queue due to open window, we'll 631 * try to send as much as possible. 632 */ 633 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) { 634 /* If the chunk is abandoned, move it to abandoned list. */ 635 if (sctp_chunk_abandoned(chunk)) { 636 list_del_init(&chunk->transmitted_list); 637 sctp_insert_list(&q->abandoned, 638 &chunk->transmitted_list); 639 continue; 640 } 641 642 /* Make sure that Gap Acked TSNs are not retransmitted. A 643 * simple approach is just to move such TSNs out of the 644 * way and into a 'transmitted' queue and skip to the 645 * next chunk. 646 */ 647 if (chunk->tsn_gap_acked) { 648 list_move_tail(&chunk->transmitted_list, 649 &transport->transmitted); 650 continue; 651 } 652 653 /* If we are doing fast retransmit, ignore non-fast_rtransmit 654 * chunks 655 */ 656 if (fast_rtx && !chunk->fast_retransmit) 657 continue; 658 659redo: 660 /* Attempt to append this chunk to the packet. */ 661 status = sctp_packet_append_chunk(pkt, chunk); 662 663 switch (status) { 664 case SCTP_XMIT_PMTU_FULL: 665 if (!pkt->has_data && !pkt->has_cookie_echo) { 666 /* If this packet did not contain DATA then 667 * retransmission did not happen, so do it 668 * again. We'll ignore the error here since 669 * control chunks are already freed so there 670 * is nothing we can do. 671 */ 672 sctp_packet_transmit(pkt, gfp); 673 goto redo; 674 } 675 676 /* Send this packet. */ 677 error = sctp_packet_transmit(pkt, gfp); 678 679 /* If we are retransmitting, we should only 680 * send a single packet. 681 * Otherwise, try appending this chunk again. 682 */ 683 if (rtx_timeout || fast_rtx) 684 done = 1; 685 else 686 goto redo; 687 688 /* Bundle next chunk in the next round. */ 689 break; 690 691 case SCTP_XMIT_RWND_FULL: 692 /* Send this packet. */ 693 error = sctp_packet_transmit(pkt, gfp); 694 695 /* Stop sending DATA as there is no more room 696 * at the receiver. 697 */ 698 done = 1; 699 break; 700 701 case SCTP_XMIT_DELAY: 702 /* Send this packet. */ 703 error = sctp_packet_transmit(pkt, gfp); 704 705 /* Stop sending DATA because of nagle delay. */ 706 done = 1; 707 break; 708 709 default: 710 /* The append was successful, so add this chunk to 711 * the transmitted list. 712 */ 713 list_move_tail(&chunk->transmitted_list, 714 &transport->transmitted); 715 716 /* Mark the chunk as ineligible for fast retransmit 717 * after it is retransmitted. 718 */ 719 if (chunk->fast_retransmit == SCTP_NEED_FRTX) 720 chunk->fast_retransmit = SCTP_DONT_FRTX; 721 722 q->asoc->stats.rtxchunks++; 723 break; 724 } 725 726 /* Set the timer if there were no errors */ 727 if (!error && !timer) 728 timer = 1; 729 730 if (done) 731 break; 732 } 733 734 /* If we are here due to a retransmit timeout or a fast 735 * retransmit and if there are any chunks left in the retransmit 736 * queue that could not fit in the PMTU sized packet, they need 737 * to be marked as ineligible for a subsequent fast retransmit. 738 */ 739 if (rtx_timeout || fast_rtx) { 740 list_for_each_entry(chunk1, lqueue, transmitted_list) { 741 if (chunk1->fast_retransmit == SCTP_NEED_FRTX) 742 chunk1->fast_retransmit = SCTP_DONT_FRTX; 743 } 744 } 745 746 *start_timer = timer; 747 748 /* Clear fast retransmit hint */ 749 if (fast_rtx) 750 q->fast_rtx = 0; 751 752 return error; 753} 754 755/* Cork the outqueue so queued chunks are really queued. */ 756void sctp_outq_uncork(struct sctp_outq *q, gfp_t gfp) 757{ 758 if (q->cork) 759 q->cork = 0; 760 761 sctp_outq_flush(q, 0, gfp); 762} 763 764static int sctp_packet_singleton(struct sctp_transport *transport, 765 struct sctp_chunk *chunk, gfp_t gfp) 766{ 767 const struct sctp_association *asoc = transport->asoc; 768 const __u16 sport = asoc->base.bind_addr.port; 769 const __u16 dport = asoc->peer.port; 770 const __u32 vtag = asoc->peer.i.init_tag; 771 struct sctp_packet singleton; 772 773 sctp_packet_init(&singleton, transport, sport, dport); 774 sctp_packet_config(&singleton, vtag, 0); 775 sctp_packet_append_chunk(&singleton, chunk); 776 return sctp_packet_transmit(&singleton, gfp); 777} 778 779/* Struct to hold the context during sctp outq flush */ 780struct sctp_flush_ctx { 781 struct sctp_outq *q; 782 /* Current transport being used. It's NOT the same as curr active one */ 783 struct sctp_transport *transport; 784 /* These transports have chunks to send. */ 785 struct list_head transport_list; 786 struct sctp_association *asoc; 787 /* Packet on the current transport above */ 788 struct sctp_packet *packet; 789 gfp_t gfp; 790}; 791 792/* transport: current transport */ 793static void sctp_outq_select_transport(struct sctp_flush_ctx *ctx, 794 struct sctp_chunk *chunk) 795{ 796 struct sctp_transport *new_transport = chunk->transport; 797 798 if (!new_transport) { 799 if (!sctp_chunk_is_data(chunk)) { 800 /* If we have a prior transport pointer, see if 801 * the destination address of the chunk 802 * matches the destination address of the 803 * current transport. If not a match, then 804 * try to look up the transport with a given 805 * destination address. We do this because 806 * after processing ASCONFs, we may have new 807 * transports created. 808 */ 809 if (ctx->transport && sctp_cmp_addr_exact(&chunk->dest, 810 &ctx->transport->ipaddr)) 811 new_transport = ctx->transport; 812 else 813 new_transport = sctp_assoc_lookup_paddr(ctx->asoc, 814 &chunk->dest); 815 } 816 817 /* if we still don't have a new transport, then 818 * use the current active path. 819 */ 820 if (!new_transport) 821 new_transport = ctx->asoc->peer.active_path; 822 } else { 823 __u8 type; 824 825 switch (new_transport->state) { 826 case SCTP_INACTIVE: 827 case SCTP_UNCONFIRMED: 828 case SCTP_PF: 829 /* If the chunk is Heartbeat or Heartbeat Ack, 830 * send it to chunk->transport, even if it's 831 * inactive. 832 * 833 * 3.3.6 Heartbeat Acknowledgement: 834 * ... 835 * A HEARTBEAT ACK is always sent to the source IP 836 * address of the IP datagram containing the 837 * HEARTBEAT chunk to which this ack is responding. 838 * ... 839 * 840 * ASCONF_ACKs also must be sent to the source. 841 */ 842 type = chunk->chunk_hdr->type; 843 if (type != SCTP_CID_HEARTBEAT && 844 type != SCTP_CID_HEARTBEAT_ACK && 845 type != SCTP_CID_ASCONF_ACK) 846 new_transport = ctx->asoc->peer.active_path; 847 break; 848 default: 849 break; 850 } 851 } 852 853 /* Are we switching transports? Take care of transport locks. */ 854 if (new_transport != ctx->transport) { 855 ctx->transport = new_transport; 856 ctx->packet = &ctx->transport->packet; 857 858 if (list_empty(&ctx->transport->send_ready)) 859 list_add_tail(&ctx->transport->send_ready, 860 &ctx->transport_list); 861 862 sctp_packet_config(ctx->packet, 863 ctx->asoc->peer.i.init_tag, 864 ctx->asoc->peer.ecn_capable); 865 /* We've switched transports, so apply the 866 * Burst limit to the new transport. 867 */ 868 sctp_transport_burst_limited(ctx->transport); 869 } 870} 871 872static void sctp_outq_flush_ctrl(struct sctp_flush_ctx *ctx) 873{ 874 struct sctp_chunk *chunk, *tmp; 875 enum sctp_xmit status; 876 int one_packet, error; 877 878 list_for_each_entry_safe(chunk, tmp, &ctx->q->control_chunk_list, list) { 879 one_packet = 0; 880 881 /* RFC 5061, 5.3 882 * F1) This means that until such time as the ASCONF 883 * containing the add is acknowledged, the sender MUST 884 * NOT use the new IP address as a source for ANY SCTP 885 * packet except on carrying an ASCONF Chunk. 886 */ 887 if (ctx->asoc->src_out_of_asoc_ok && 888 chunk->chunk_hdr->type != SCTP_CID_ASCONF) 889 continue; 890 891 list_del_init(&chunk->list); 892 893 /* Pick the right transport to use. Should always be true for 894 * the first chunk as we don't have a transport by then. 895 */ 896 sctp_outq_select_transport(ctx, chunk); 897 898 switch (chunk->chunk_hdr->type) { 899 /* 6.10 Bundling 900 * ... 901 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN 902 * COMPLETE with any other chunks. [Send them immediately.] 903 */ 904 case SCTP_CID_INIT: 905 case SCTP_CID_INIT_ACK: 906 case SCTP_CID_SHUTDOWN_COMPLETE: 907 error = sctp_packet_singleton(ctx->transport, chunk, 908 ctx->gfp); 909 if (error < 0) { 910 ctx->asoc->base.sk->sk_err = -error; 911 return; 912 } 913 break; 914 915 case SCTP_CID_ABORT: 916 if (sctp_test_T_bit(chunk)) 917 ctx->packet->vtag = ctx->asoc->c.my_vtag; 918 fallthrough; 919 920 /* The following chunks are "response" chunks, i.e. 921 * they are generated in response to something we 922 * received. If we are sending these, then we can 923 * send only 1 packet containing these chunks. 924 */ 925 case SCTP_CID_HEARTBEAT_ACK: 926 case SCTP_CID_SHUTDOWN_ACK: 927 case SCTP_CID_COOKIE_ACK: 928 case SCTP_CID_COOKIE_ECHO: 929 case SCTP_CID_ERROR: 930 case SCTP_CID_ECN_CWR: 931 case SCTP_CID_ASCONF_ACK: 932 one_packet = 1; 933 fallthrough; 934 935 case SCTP_CID_SACK: 936 case SCTP_CID_HEARTBEAT: 937 case SCTP_CID_SHUTDOWN: 938 case SCTP_CID_ECN_ECNE: 939 case SCTP_CID_ASCONF: 940 case SCTP_CID_FWD_TSN: 941 case SCTP_CID_I_FWD_TSN: 942 case SCTP_CID_RECONF: 943 status = sctp_packet_transmit_chunk(ctx->packet, chunk, 944 one_packet, ctx->gfp); 945 if (status != SCTP_XMIT_OK) { 946 /* put the chunk back */ 947 list_add(&chunk->list, &ctx->q->control_chunk_list); 948 break; 949 } 950 951 ctx->asoc->stats.octrlchunks++; 952 /* PR-SCTP C5) If a FORWARD TSN is sent, the 953 * sender MUST assure that at least one T3-rtx 954 * timer is running. 955 */ 956 if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN || 957 chunk->chunk_hdr->type == SCTP_CID_I_FWD_TSN) { 958 sctp_transport_reset_t3_rtx(ctx->transport); 959 ctx->transport->last_time_sent = jiffies; 960 } 961 962 if (chunk == ctx->asoc->strreset_chunk) 963 sctp_transport_reset_reconf_timer(ctx->transport); 964 965 break; 966 967 default: 968 /* We built a chunk with an illegal type! */ 969 BUG(); 970 } 971 } 972} 973 974/* Returns false if new data shouldn't be sent */ 975static bool sctp_outq_flush_rtx(struct sctp_flush_ctx *ctx, 976 int rtx_timeout) 977{ 978 int error, start_timer = 0; 979 980 if (ctx->asoc->peer.retran_path->state == SCTP_UNCONFIRMED) 981 return false; 982 983 if (ctx->transport != ctx->asoc->peer.retran_path) { 984 /* Switch transports & prepare the packet. */ 985 ctx->transport = ctx->asoc->peer.retran_path; 986 ctx->packet = &ctx->transport->packet; 987 988 if (list_empty(&ctx->transport->send_ready)) 989 list_add_tail(&ctx->transport->send_ready, 990 &ctx->transport_list); 991 992 sctp_packet_config(ctx->packet, ctx->asoc->peer.i.init_tag, 993 ctx->asoc->peer.ecn_capable); 994 } 995 996 error = __sctp_outq_flush_rtx(ctx->q, ctx->packet, rtx_timeout, 997 &start_timer, ctx->gfp); 998 if (error < 0) 999 ctx->asoc->base.sk->sk_err = -error; 1000 1001 if (start_timer) { 1002 sctp_transport_reset_t3_rtx(ctx->transport); 1003 ctx->transport->last_time_sent = jiffies; 1004 } 1005 1006 /* This can happen on COOKIE-ECHO resend. Only 1007 * one chunk can get bundled with a COOKIE-ECHO. 1008 */ 1009 if (ctx->packet->has_cookie_echo) 1010 return false; 1011 1012 /* Don't send new data if there is still data 1013 * waiting to retransmit. 1014 */ 1015 if (!list_empty(&ctx->q->retransmit)) 1016 return false; 1017 1018 return true; 1019} 1020 1021static void sctp_outq_flush_data(struct sctp_flush_ctx *ctx, 1022 int rtx_timeout) 1023{ 1024 struct sctp_chunk *chunk; 1025 enum sctp_xmit status; 1026 1027 /* Is it OK to send data chunks? */ 1028 switch (ctx->asoc->state) { 1029 case SCTP_STATE_COOKIE_ECHOED: 1030 /* Only allow bundling when this packet has a COOKIE-ECHO 1031 * chunk. 1032 */ 1033 if (!ctx->packet || !ctx->packet->has_cookie_echo) 1034 return; 1035 1036 fallthrough; 1037 case SCTP_STATE_ESTABLISHED: 1038 case SCTP_STATE_SHUTDOWN_PENDING: 1039 case SCTP_STATE_SHUTDOWN_RECEIVED: 1040 break; 1041 1042 default: 1043 /* Do nothing. */ 1044 return; 1045 } 1046 1047 /* RFC 2960 6.1 Transmission of DATA Chunks 1048 * 1049 * C) When the time comes for the sender to transmit, 1050 * before sending new DATA chunks, the sender MUST 1051 * first transmit any outstanding DATA chunks which 1052 * are marked for retransmission (limited by the 1053 * current cwnd). 1054 */ 1055 if (!list_empty(&ctx->q->retransmit) && 1056 !sctp_outq_flush_rtx(ctx, rtx_timeout)) 1057 return; 1058 1059 /* Apply Max.Burst limitation to the current transport in 1060 * case it will be used for new data. We are going to 1061 * rest it before we return, but we want to apply the limit 1062 * to the currently queued data. 1063 */ 1064 if (ctx->transport) 1065 sctp_transport_burst_limited(ctx->transport); 1066 1067 /* Finally, transmit new packets. */ 1068 while ((chunk = sctp_outq_dequeue_data(ctx->q)) != NULL) { 1069 __u32 sid = ntohs(chunk->subh.data_hdr->stream); 1070 __u8 stream_state = SCTP_SO(&ctx->asoc->stream, sid)->state; 1071 1072 /* Has this chunk expired? */ 1073 if (sctp_chunk_abandoned(chunk)) { 1074 sctp_sched_dequeue_done(ctx->q, chunk); 1075 sctp_chunk_fail(chunk, 0); 1076 sctp_chunk_free(chunk); 1077 continue; 1078 } 1079 1080 if (stream_state == SCTP_STREAM_CLOSED) { 1081 sctp_outq_head_data(ctx->q, chunk); 1082 break; 1083 } 1084 1085 sctp_outq_select_transport(ctx, chunk); 1086 1087 pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p skb->users:%d\n", 1088 __func__, ctx->q, chunk, chunk && chunk->chunk_hdr ? 1089 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) : 1090 "illegal chunk", ntohl(chunk->subh.data_hdr->tsn), 1091 chunk->skb ? chunk->skb->head : NULL, chunk->skb ? 1092 refcount_read(&chunk->skb->users) : -1); 1093 1094 /* Add the chunk to the packet. */ 1095 status = sctp_packet_transmit_chunk(ctx->packet, chunk, 0, 1096 ctx->gfp); 1097 if (status != SCTP_XMIT_OK) { 1098 /* We could not append this chunk, so put 1099 * the chunk back on the output queue. 1100 */ 1101 pr_debug("%s: could not transmit tsn:0x%x, status:%d\n", 1102 __func__, ntohl(chunk->subh.data_hdr->tsn), 1103 status); 1104 1105 sctp_outq_head_data(ctx->q, chunk); 1106 break; 1107 } 1108 1109 /* The sender is in the SHUTDOWN-PENDING state, 1110 * The sender MAY set the I-bit in the DATA 1111 * chunk header. 1112 */ 1113 if (ctx->asoc->state == SCTP_STATE_SHUTDOWN_PENDING) 1114 chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM; 1115 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED) 1116 ctx->asoc->stats.ouodchunks++; 1117 else 1118 ctx->asoc->stats.oodchunks++; 1119 1120 /* Only now it's safe to consider this 1121 * chunk as sent, sched-wise. 1122 */ 1123 sctp_sched_dequeue_done(ctx->q, chunk); 1124 1125 list_add_tail(&chunk->transmitted_list, 1126 &ctx->transport->transmitted); 1127 1128 sctp_transport_reset_t3_rtx(ctx->transport); 1129 ctx->transport->last_time_sent = jiffies; 1130 1131 /* Only let one DATA chunk get bundled with a 1132 * COOKIE-ECHO chunk. 1133 */ 1134 if (ctx->packet->has_cookie_echo) 1135 break; 1136 } 1137} 1138 1139static void sctp_outq_flush_transports(struct sctp_flush_ctx *ctx) 1140{ 1141 struct list_head *ltransport; 1142 struct sctp_packet *packet; 1143 struct sctp_transport *t; 1144 int error = 0; 1145 1146 while ((ltransport = sctp_list_dequeue(&ctx->transport_list)) != NULL) { 1147 t = list_entry(ltransport, struct sctp_transport, send_ready); 1148 packet = &t->packet; 1149 if (!sctp_packet_empty(packet)) { 1150 error = sctp_packet_transmit(packet, ctx->gfp); 1151 if (error < 0) 1152 ctx->q->asoc->base.sk->sk_err = -error; 1153 } 1154 1155 /* Clear the burst limited state, if any */ 1156 sctp_transport_burst_reset(t); 1157 } 1158} 1159 1160/* Try to flush an outqueue. 1161 * 1162 * Description: Send everything in q which we legally can, subject to 1163 * congestion limitations. 1164 * * Note: This function can be called from multiple contexts so appropriate 1165 * locking concerns must be made. Today we use the sock lock to protect 1166 * this function. 1167 */ 1168 1169static void sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp) 1170{ 1171 struct sctp_flush_ctx ctx = { 1172 .q = q, 1173 .transport = NULL, 1174 .transport_list = LIST_HEAD_INIT(ctx.transport_list), 1175 .asoc = q->asoc, 1176 .packet = NULL, 1177 .gfp = gfp, 1178 }; 1179 1180 /* 6.10 Bundling 1181 * ... 1182 * When bundling control chunks with DATA chunks, an 1183 * endpoint MUST place control chunks first in the outbound 1184 * SCTP packet. The transmitter MUST transmit DATA chunks 1185 * within a SCTP packet in increasing order of TSN. 1186 * ... 1187 */ 1188 1189 sctp_outq_flush_ctrl(&ctx); 1190 1191 if (q->asoc->src_out_of_asoc_ok) 1192 goto sctp_flush_out; 1193 1194 sctp_outq_flush_data(&ctx, rtx_timeout); 1195 1196sctp_flush_out: 1197 1198 sctp_outq_flush_transports(&ctx); 1199} 1200 1201/* Update unack_data based on the incoming SACK chunk */ 1202static void sctp_sack_update_unack_data(struct sctp_association *assoc, 1203 struct sctp_sackhdr *sack) 1204{ 1205 union sctp_sack_variable *frags; 1206 __u16 unack_data; 1207 int i; 1208 1209 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1; 1210 1211 frags = sack->variable; 1212 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) { 1213 unack_data -= ((ntohs(frags[i].gab.end) - 1214 ntohs(frags[i].gab.start) + 1)); 1215 } 1216 1217 assoc->unack_data = unack_data; 1218} 1219 1220/* This is where we REALLY process a SACK. 1221 * 1222 * Process the SACK against the outqueue. Mostly, this just frees 1223 * things off the transmitted queue. 1224 */ 1225int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk) 1226{ 1227 struct sctp_association *asoc = q->asoc; 1228 struct sctp_sackhdr *sack = chunk->subh.sack_hdr; 1229 struct sctp_transport *transport; 1230 struct sctp_chunk *tchunk = NULL; 1231 struct list_head *lchunk, *transport_list, *temp; 1232 union sctp_sack_variable *frags = sack->variable; 1233 __u32 sack_ctsn, ctsn, tsn; 1234 __u32 highest_tsn, highest_new_tsn; 1235 __u32 sack_a_rwnd; 1236 unsigned int outstanding; 1237 struct sctp_transport *primary = asoc->peer.primary_path; 1238 int count_of_newacks = 0; 1239 int gap_ack_blocks; 1240 u8 accum_moved = 0; 1241 1242 /* Grab the association's destination address list. */ 1243 transport_list = &asoc->peer.transport_addr_list; 1244 1245 /* SCTP path tracepoint for congestion control debugging. */ 1246 if (trace_sctp_probe_path_enabled()) { 1247 list_for_each_entry(transport, transport_list, transports) 1248 trace_sctp_probe_path(transport, asoc); 1249 } 1250 1251 sack_ctsn = ntohl(sack->cum_tsn_ack); 1252 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks); 1253 asoc->stats.gapcnt += gap_ack_blocks; 1254 /* 1255 * SFR-CACC algorithm: 1256 * On receipt of a SACK the sender SHOULD execute the 1257 * following statements. 1258 * 1259 * 1) If the cumulative ack in the SACK passes next tsn_at_change 1260 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be 1261 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for 1262 * all destinations. 1263 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE 1264 * is set the receiver of the SACK MUST take the following actions: 1265 * 1266 * A) Initialize the cacc_saw_newack to 0 for all destination 1267 * addresses. 1268 * 1269 * Only bother if changeover_active is set. Otherwise, this is 1270 * totally suboptimal to do on every SACK. 1271 */ 1272 if (primary->cacc.changeover_active) { 1273 u8 clear_cycling = 0; 1274 1275 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) { 1276 primary->cacc.changeover_active = 0; 1277 clear_cycling = 1; 1278 } 1279 1280 if (clear_cycling || gap_ack_blocks) { 1281 list_for_each_entry(transport, transport_list, 1282 transports) { 1283 if (clear_cycling) 1284 transport->cacc.cycling_changeover = 0; 1285 if (gap_ack_blocks) 1286 transport->cacc.cacc_saw_newack = 0; 1287 } 1288 } 1289 } 1290 1291 /* Get the highest TSN in the sack. */ 1292 highest_tsn = sack_ctsn; 1293 if (gap_ack_blocks) 1294 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end); 1295 1296 if (TSN_lt(asoc->highest_sacked, highest_tsn)) 1297 asoc->highest_sacked = highest_tsn; 1298 1299 highest_new_tsn = sack_ctsn; 1300 1301 /* Run through the retransmit queue. Credit bytes received 1302 * and free those chunks that we can. 1303 */ 1304 sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn); 1305 1306 /* Run through the transmitted queue. 1307 * Credit bytes received and free those chunks which we can. 1308 * 1309 * This is a MASSIVE candidate for optimization. 1310 */ 1311 list_for_each_entry(transport, transport_list, transports) { 1312 sctp_check_transmitted(q, &transport->transmitted, 1313 transport, &chunk->source, sack, 1314 &highest_new_tsn); 1315 /* 1316 * SFR-CACC algorithm: 1317 * C) Let count_of_newacks be the number of 1318 * destinations for which cacc_saw_newack is set. 1319 */ 1320 if (transport->cacc.cacc_saw_newack) 1321 count_of_newacks++; 1322 } 1323 1324 /* Move the Cumulative TSN Ack Point if appropriate. */ 1325 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) { 1326 asoc->ctsn_ack_point = sack_ctsn; 1327 accum_moved = 1; 1328 } 1329 1330 if (gap_ack_blocks) { 1331 1332 if (asoc->fast_recovery && accum_moved) 1333 highest_new_tsn = highest_tsn; 1334 1335 list_for_each_entry(transport, transport_list, transports) 1336 sctp_mark_missing(q, &transport->transmitted, transport, 1337 highest_new_tsn, count_of_newacks); 1338 } 1339 1340 /* Update unack_data field in the assoc. */ 1341 sctp_sack_update_unack_data(asoc, sack); 1342 1343 ctsn = asoc->ctsn_ack_point; 1344 1345 /* Throw away stuff rotting on the sack queue. */ 1346 list_for_each_safe(lchunk, temp, &q->sacked) { 1347 tchunk = list_entry(lchunk, struct sctp_chunk, 1348 transmitted_list); 1349 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1350 if (TSN_lte(tsn, ctsn)) { 1351 list_del_init(&tchunk->transmitted_list); 1352 if (asoc->peer.prsctp_capable && 1353 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags)) 1354 asoc->sent_cnt_removable--; 1355 sctp_chunk_free(tchunk); 1356 } 1357 } 1358 1359 /* ii) Set rwnd equal to the newly received a_rwnd minus the 1360 * number of bytes still outstanding after processing the 1361 * Cumulative TSN Ack and the Gap Ack Blocks. 1362 */ 1363 1364 sack_a_rwnd = ntohl(sack->a_rwnd); 1365 asoc->peer.zero_window_announced = !sack_a_rwnd; 1366 outstanding = q->outstanding_bytes; 1367 1368 if (outstanding < sack_a_rwnd) 1369 sack_a_rwnd -= outstanding; 1370 else 1371 sack_a_rwnd = 0; 1372 1373 asoc->peer.rwnd = sack_a_rwnd; 1374 1375 asoc->stream.si->generate_ftsn(q, sack_ctsn); 1376 1377 pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__, sack_ctsn); 1378 pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, " 1379 "advertised peer ack point:0x%x\n", __func__, asoc, ctsn, 1380 asoc->adv_peer_ack_point); 1381 1382 return sctp_outq_is_empty(q); 1383} 1384 1385/* Is the outqueue empty? 1386 * The queue is empty when we have not pending data, no in-flight data 1387 * and nothing pending retransmissions. 1388 */ 1389int sctp_outq_is_empty(const struct sctp_outq *q) 1390{ 1391 return q->out_qlen == 0 && q->outstanding_bytes == 0 && 1392 list_empty(&q->retransmit); 1393} 1394 1395/******************************************************************** 1396 * 2nd Level Abstractions 1397 ********************************************************************/ 1398 1399/* Go through a transport's transmitted list or the association's retransmit 1400 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked. 1401 * The retransmit list will not have an associated transport. 1402 * 1403 * I added coherent debug information output. --xguo 1404 * 1405 * Instead of printing 'sacked' or 'kept' for each TSN on the 1406 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5. 1407 * KEPT TSN6-TSN7, etc. 1408 */ 1409static void sctp_check_transmitted(struct sctp_outq *q, 1410 struct list_head *transmitted_queue, 1411 struct sctp_transport *transport, 1412 union sctp_addr *saddr, 1413 struct sctp_sackhdr *sack, 1414 __u32 *highest_new_tsn_in_sack) 1415{ 1416 struct list_head *lchunk; 1417 struct sctp_chunk *tchunk; 1418 struct list_head tlist; 1419 __u32 tsn; 1420 __u32 sack_ctsn; 1421 __u32 rtt; 1422 __u8 restart_timer = 0; 1423 int bytes_acked = 0; 1424 int migrate_bytes = 0; 1425 bool forward_progress = false; 1426 1427 sack_ctsn = ntohl(sack->cum_tsn_ack); 1428 1429 INIT_LIST_HEAD(&tlist); 1430 1431 /* The while loop will skip empty transmitted queues. */ 1432 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) { 1433 tchunk = list_entry(lchunk, struct sctp_chunk, 1434 transmitted_list); 1435 1436 if (sctp_chunk_abandoned(tchunk)) { 1437 /* Move the chunk to abandoned list. */ 1438 sctp_insert_list(&q->abandoned, lchunk); 1439 1440 /* If this chunk has not been acked, stop 1441 * considering it as 'outstanding'. 1442 */ 1443 if (transmitted_queue != &q->retransmit && 1444 !tchunk->tsn_gap_acked) { 1445 if (tchunk->transport) 1446 tchunk->transport->flight_size -= 1447 sctp_data_size(tchunk); 1448 q->outstanding_bytes -= sctp_data_size(tchunk); 1449 } 1450 continue; 1451 } 1452 1453 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1454 if (sctp_acked(sack, tsn)) { 1455 /* If this queue is the retransmit queue, the 1456 * retransmit timer has already reclaimed 1457 * the outstanding bytes for this chunk, so only 1458 * count bytes associated with a transport. 1459 */ 1460 if (transport && !tchunk->tsn_gap_acked) { 1461 /* If this chunk is being used for RTT 1462 * measurement, calculate the RTT and update 1463 * the RTO using this value. 1464 * 1465 * 6.3.1 C5) Karn's algorithm: RTT measurements 1466 * MUST NOT be made using packets that were 1467 * retransmitted (and thus for which it is 1468 * ambiguous whether the reply was for the 1469 * first instance of the packet or a later 1470 * instance). 1471 */ 1472 if (!sctp_chunk_retransmitted(tchunk) && 1473 tchunk->rtt_in_progress) { 1474 tchunk->rtt_in_progress = 0; 1475 rtt = jiffies - tchunk->sent_at; 1476 sctp_transport_update_rto(transport, 1477 rtt); 1478 } 1479 1480 if (TSN_lte(tsn, sack_ctsn)) { 1481 /* 1482 * SFR-CACC algorithm: 1483 * 2) If the SACK contains gap acks 1484 * and the flag CHANGEOVER_ACTIVE is 1485 * set the receiver of the SACK MUST 1486 * take the following action: 1487 * 1488 * B) For each TSN t being acked that 1489 * has not been acked in any SACK so 1490 * far, set cacc_saw_newack to 1 for 1491 * the destination that the TSN was 1492 * sent to. 1493 */ 1494 if (sack->num_gap_ack_blocks && 1495 q->asoc->peer.primary_path->cacc. 1496 changeover_active) 1497 transport->cacc.cacc_saw_newack 1498 = 1; 1499 } 1500 } 1501 1502 /* If the chunk hasn't been marked as ACKED, 1503 * mark it and account bytes_acked if the 1504 * chunk had a valid transport (it will not 1505 * have a transport if ASCONF had deleted it 1506 * while DATA was outstanding). 1507 */ 1508 if (!tchunk->tsn_gap_acked) { 1509 tchunk->tsn_gap_acked = 1; 1510 if (TSN_lt(*highest_new_tsn_in_sack, tsn)) 1511 *highest_new_tsn_in_sack = tsn; 1512 bytes_acked += sctp_data_size(tchunk); 1513 if (!tchunk->transport) 1514 migrate_bytes += sctp_data_size(tchunk); 1515 forward_progress = true; 1516 } 1517 1518 if (TSN_lte(tsn, sack_ctsn)) { 1519 /* RFC 2960 6.3.2 Retransmission Timer Rules 1520 * 1521 * R3) Whenever a SACK is received 1522 * that acknowledges the DATA chunk 1523 * with the earliest outstanding TSN 1524 * for that address, restart T3-rtx 1525 * timer for that address with its 1526 * current RTO. 1527 */ 1528 restart_timer = 1; 1529 forward_progress = true; 1530 1531 list_add_tail(&tchunk->transmitted_list, 1532 &q->sacked); 1533 } else { 1534 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2 1535 * M2) Each time a SACK arrives reporting 1536 * 'Stray DATA chunk(s)' record the highest TSN 1537 * reported as newly acknowledged, call this 1538 * value 'HighestTSNinSack'. A newly 1539 * acknowledged DATA chunk is one not 1540 * previously acknowledged in a SACK. 1541 * 1542 * When the SCTP sender of data receives a SACK 1543 * chunk that acknowledges, for the first time, 1544 * the receipt of a DATA chunk, all the still 1545 * unacknowledged DATA chunks whose TSN is 1546 * older than that newly acknowledged DATA 1547 * chunk, are qualified as 'Stray DATA chunks'. 1548 */ 1549 list_add_tail(lchunk, &tlist); 1550 } 1551 } else { 1552 if (tchunk->tsn_gap_acked) { 1553 pr_debug("%s: receiver reneged on data TSN:0x%x\n", 1554 __func__, tsn); 1555 1556 tchunk->tsn_gap_acked = 0; 1557 1558 if (tchunk->transport) 1559 bytes_acked -= sctp_data_size(tchunk); 1560 1561 /* RFC 2960 6.3.2 Retransmission Timer Rules 1562 * 1563 * R4) Whenever a SACK is received missing a 1564 * TSN that was previously acknowledged via a 1565 * Gap Ack Block, start T3-rtx for the 1566 * destination address to which the DATA 1567 * chunk was originally 1568 * transmitted if it is not already running. 1569 */ 1570 restart_timer = 1; 1571 } 1572 1573 list_add_tail(lchunk, &tlist); 1574 } 1575 } 1576 1577 if (transport) { 1578 if (bytes_acked) { 1579 struct sctp_association *asoc = transport->asoc; 1580 1581 /* We may have counted DATA that was migrated 1582 * to this transport due to DEL-IP operation. 1583 * Subtract those bytes, since the were never 1584 * send on this transport and shouldn't be 1585 * credited to this transport. 1586 */ 1587 bytes_acked -= migrate_bytes; 1588 1589 /* 8.2. When an outstanding TSN is acknowledged, 1590 * the endpoint shall clear the error counter of 1591 * the destination transport address to which the 1592 * DATA chunk was last sent. 1593 * The association's overall error counter is 1594 * also cleared. 1595 */ 1596 transport->error_count = 0; 1597 transport->asoc->overall_error_count = 0; 1598 forward_progress = true; 1599 1600 /* 1601 * While in SHUTDOWN PENDING, we may have started 1602 * the T5 shutdown guard timer after reaching the 1603 * retransmission limit. Stop that timer as soon 1604 * as the receiver acknowledged any data. 1605 */ 1606 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING && 1607 del_timer(&asoc->timers 1608 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD])) 1609 sctp_association_put(asoc); 1610 1611 /* Mark the destination transport address as 1612 * active if it is not so marked. 1613 */ 1614 if ((transport->state == SCTP_INACTIVE || 1615 transport->state == SCTP_UNCONFIRMED) && 1616 sctp_cmp_addr_exact(&transport->ipaddr, saddr)) { 1617 sctp_assoc_control_transport( 1618 transport->asoc, 1619 transport, 1620 SCTP_TRANSPORT_UP, 1621 SCTP_RECEIVED_SACK); 1622 } 1623 1624 sctp_transport_raise_cwnd(transport, sack_ctsn, 1625 bytes_acked); 1626 1627 transport->flight_size -= bytes_acked; 1628 if (transport->flight_size == 0) 1629 transport->partial_bytes_acked = 0; 1630 q->outstanding_bytes -= bytes_acked + migrate_bytes; 1631 } else { 1632 /* RFC 2960 6.1, sctpimpguide-06 2.15.2 1633 * When a sender is doing zero window probing, it 1634 * should not timeout the association if it continues 1635 * to receive new packets from the receiver. The 1636 * reason is that the receiver MAY keep its window 1637 * closed for an indefinite time. 1638 * A sender is doing zero window probing when the 1639 * receiver's advertised window is zero, and there is 1640 * only one data chunk in flight to the receiver. 1641 * 1642 * Allow the association to timeout while in SHUTDOWN 1643 * PENDING or SHUTDOWN RECEIVED in case the receiver 1644 * stays in zero window mode forever. 1645 */ 1646 if (!q->asoc->peer.rwnd && 1647 !list_empty(&tlist) && 1648 (sack_ctsn+2 == q->asoc->next_tsn) && 1649 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) { 1650 pr_debug("%s: sack received for zero window " 1651 "probe:%u\n", __func__, sack_ctsn); 1652 1653 q->asoc->overall_error_count = 0; 1654 transport->error_count = 0; 1655 } 1656 } 1657 1658 /* RFC 2960 6.3.2 Retransmission Timer Rules 1659 * 1660 * R2) Whenever all outstanding data sent to an address have 1661 * been acknowledged, turn off the T3-rtx timer of that 1662 * address. 1663 */ 1664 if (!transport->flight_size) { 1665 if (del_timer(&transport->T3_rtx_timer)) 1666 sctp_transport_put(transport); 1667 } else if (restart_timer) { 1668 if (!mod_timer(&transport->T3_rtx_timer, 1669 jiffies + transport->rto)) 1670 sctp_transport_hold(transport); 1671 } 1672 1673 if (forward_progress) { 1674 if (transport->dst) 1675 sctp_transport_dst_confirm(transport); 1676 } 1677 } 1678 1679 list_splice(&tlist, transmitted_queue); 1680} 1681 1682/* Mark chunks as missing and consequently may get retransmitted. */ 1683static void sctp_mark_missing(struct sctp_outq *q, 1684 struct list_head *transmitted_queue, 1685 struct sctp_transport *transport, 1686 __u32 highest_new_tsn_in_sack, 1687 int count_of_newacks) 1688{ 1689 struct sctp_chunk *chunk; 1690 __u32 tsn; 1691 char do_fast_retransmit = 0; 1692 struct sctp_association *asoc = q->asoc; 1693 struct sctp_transport *primary = asoc->peer.primary_path; 1694 1695 list_for_each_entry(chunk, transmitted_queue, transmitted_list) { 1696 1697 tsn = ntohl(chunk->subh.data_hdr->tsn); 1698 1699 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all 1700 * 'Unacknowledged TSN's', if the TSN number of an 1701 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack' 1702 * value, increment the 'TSN.Missing.Report' count on that 1703 * chunk if it has NOT been fast retransmitted or marked for 1704 * fast retransmit already. 1705 */ 1706 if (chunk->fast_retransmit == SCTP_CAN_FRTX && 1707 !chunk->tsn_gap_acked && 1708 TSN_lt(tsn, highest_new_tsn_in_sack)) { 1709 1710 /* SFR-CACC may require us to skip marking 1711 * this chunk as missing. 1712 */ 1713 if (!transport || !sctp_cacc_skip(primary, 1714 chunk->transport, 1715 count_of_newacks, tsn)) { 1716 chunk->tsn_missing_report++; 1717 1718 pr_debug("%s: tsn:0x%x missing counter:%d\n", 1719 __func__, tsn, chunk->tsn_missing_report); 1720 } 1721 } 1722 /* 1723 * M4) If any DATA chunk is found to have a 1724 * 'TSN.Missing.Report' 1725 * value larger than or equal to 3, mark that chunk for 1726 * retransmission and start the fast retransmit procedure. 1727 */ 1728 1729 if (chunk->tsn_missing_report >= 3) { 1730 chunk->fast_retransmit = SCTP_NEED_FRTX; 1731 do_fast_retransmit = 1; 1732 } 1733 } 1734 1735 if (transport) { 1736 if (do_fast_retransmit) 1737 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX); 1738 1739 pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, " 1740 "flight_size:%d, pba:%d\n", __func__, transport, 1741 transport->cwnd, transport->ssthresh, 1742 transport->flight_size, transport->partial_bytes_acked); 1743 } 1744} 1745 1746/* Is the given TSN acked by this packet? */ 1747static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn) 1748{ 1749 __u32 ctsn = ntohl(sack->cum_tsn_ack); 1750 union sctp_sack_variable *frags; 1751 __u16 tsn_offset, blocks; 1752 int i; 1753 1754 if (TSN_lte(tsn, ctsn)) 1755 goto pass; 1756 1757 /* 3.3.4 Selective Acknowledgment (SACK) (3): 1758 * 1759 * Gap Ack Blocks: 1760 * These fields contain the Gap Ack Blocks. They are repeated 1761 * for each Gap Ack Block up to the number of Gap Ack Blocks 1762 * defined in the Number of Gap Ack Blocks field. All DATA 1763 * chunks with TSNs greater than or equal to (Cumulative TSN 1764 * Ack + Gap Ack Block Start) and less than or equal to 1765 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack 1766 * Block are assumed to have been received correctly. 1767 */ 1768 1769 frags = sack->variable; 1770 blocks = ntohs(sack->num_gap_ack_blocks); 1771 tsn_offset = tsn - ctsn; 1772 for (i = 0; i < blocks; ++i) { 1773 if (tsn_offset >= ntohs(frags[i].gab.start) && 1774 tsn_offset <= ntohs(frags[i].gab.end)) 1775 goto pass; 1776 } 1777 1778 return 0; 1779pass: 1780 return 1; 1781} 1782 1783static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist, 1784 int nskips, __be16 stream) 1785{ 1786 int i; 1787 1788 for (i = 0; i < nskips; i++) { 1789 if (skiplist[i].stream == stream) 1790 return i; 1791 } 1792 return i; 1793} 1794 1795/* Create and add a fwdtsn chunk to the outq's control queue if needed. */ 1796void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn) 1797{ 1798 struct sctp_association *asoc = q->asoc; 1799 struct sctp_chunk *ftsn_chunk = NULL; 1800 struct sctp_fwdtsn_skip ftsn_skip_arr[10]; 1801 int nskips = 0; 1802 int skip_pos = 0; 1803 __u32 tsn; 1804 struct sctp_chunk *chunk; 1805 struct list_head *lchunk, *temp; 1806 1807 if (!asoc->peer.prsctp_capable) 1808 return; 1809 1810 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the 1811 * received SACK. 1812 * 1813 * If (Advanced.Peer.Ack.Point < SackCumAck), then update 1814 * Advanced.Peer.Ack.Point to be equal to SackCumAck. 1815 */ 1816 if (TSN_lt(asoc->adv_peer_ack_point, ctsn)) 1817 asoc->adv_peer_ack_point = ctsn; 1818 1819 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point" 1820 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as 1821 * the chunk next in the out-queue space is marked as "abandoned" as 1822 * shown in the following example: 1823 * 1824 * Assuming that a SACK arrived with the Cumulative TSN ACK 102 1825 * and the Advanced.Peer.Ack.Point is updated to this value: 1826 * 1827 * out-queue at the end of ==> out-queue after Adv.Ack.Point 1828 * normal SACK processing local advancement 1829 * ... ... 1830 * Adv.Ack.Pt-> 102 acked 102 acked 1831 * 103 abandoned 103 abandoned 1832 * 104 abandoned Adv.Ack.P-> 104 abandoned 1833 * 105 105 1834 * 106 acked 106 acked 1835 * ... ... 1836 * 1837 * In this example, the data sender successfully advanced the 1838 * "Advanced.Peer.Ack.Point" from 102 to 104 locally. 1839 */ 1840 list_for_each_safe(lchunk, temp, &q->abandoned) { 1841 chunk = list_entry(lchunk, struct sctp_chunk, 1842 transmitted_list); 1843 tsn = ntohl(chunk->subh.data_hdr->tsn); 1844 1845 /* Remove any chunks in the abandoned queue that are acked by 1846 * the ctsn. 1847 */ 1848 if (TSN_lte(tsn, ctsn)) { 1849 list_del_init(lchunk); 1850 sctp_chunk_free(chunk); 1851 } else { 1852 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) { 1853 asoc->adv_peer_ack_point = tsn; 1854 if (chunk->chunk_hdr->flags & 1855 SCTP_DATA_UNORDERED) 1856 continue; 1857 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0], 1858 nskips, 1859 chunk->subh.data_hdr->stream); 1860 ftsn_skip_arr[skip_pos].stream = 1861 chunk->subh.data_hdr->stream; 1862 ftsn_skip_arr[skip_pos].ssn = 1863 chunk->subh.data_hdr->ssn; 1864 if (skip_pos == nskips) 1865 nskips++; 1866 if (nskips == 10) 1867 break; 1868 } else 1869 break; 1870 } 1871 } 1872 1873 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point" 1874 * is greater than the Cumulative TSN ACK carried in the received 1875 * SACK, the data sender MUST send the data receiver a FORWARD TSN 1876 * chunk containing the latest value of the 1877 * "Advanced.Peer.Ack.Point". 1878 * 1879 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD 1880 * list each stream and sequence number in the forwarded TSN. This 1881 * information will enable the receiver to easily find any 1882 * stranded TSN's waiting on stream reorder queues. Each stream 1883 * SHOULD only be reported once; this means that if multiple 1884 * abandoned messages occur in the same stream then only the 1885 * highest abandoned stream sequence number is reported. If the 1886 * total size of the FORWARD TSN does NOT fit in a single MTU then 1887 * the sender of the FORWARD TSN SHOULD lower the 1888 * Advanced.Peer.Ack.Point to the last TSN that will fit in a 1889 * single MTU. 1890 */ 1891 if (asoc->adv_peer_ack_point > ctsn) 1892 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point, 1893 nskips, &ftsn_skip_arr[0]); 1894 1895 if (ftsn_chunk) { 1896 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list); 1897 SCTP_INC_STATS(asoc->base.net, SCTP_MIB_OUTCTRLCHUNKS); 1898 } 1899} 1900