1 /* SCTP kernel implementation 2 * (C) Copyright IBM Corp. 2001, 2004 3 * Copyright (c) 1999-2000 Cisco, Inc. 4 * Copyright (c) 1999-2001 Motorola, Inc. 5 * Copyright (c) 2001-2003 Intel Corp. 6 * 7 * This file is part of the SCTP kernel implementation 8 * 9 * These functions implement the sctp_outq class. The outqueue handles 10 * bundling and queueing of outgoing SCTP chunks. 11 * 12 * This SCTP implementation is free software; 13 * you can redistribute it and/or modify it under the terms of 14 * the GNU General Public License as published by 15 * the Free Software Foundation; either version 2, or (at your option) 16 * any later version. 17 * 18 * This SCTP implementation is distributed in the hope that it 19 * will be useful, but WITHOUT ANY WARRANTY; without even the implied 20 * ************************ 21 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. 22 * See the GNU General Public License for more details. 23 * 24 * You should have received a copy of the GNU General Public License 25 * along with GNU CC; see the file COPYING. If not, see 26 * <http://www.gnu.org/licenses/>. 27 * 28 * Please send any bug reports or fixes you make to the 29 * email address(es): 30 * lksctp developers <linux-sctp@vger.kernel.org> 31 * 32 * Written or modified by: 33 * La Monte H.P. Yarroll <piggy@acm.org> 34 * Karl Knutson <karl@athena.chicago.il.us> 35 * Perry Melange <pmelange@null.cc.uic.edu> 36 * Xingang Guo <xingang.guo@intel.com> 37 * Hui Huang <hui.huang@nokia.com> 38 * Sridhar Samudrala <sri@us.ibm.com> 39 * Jon Grimm <jgrimm@us.ibm.com> 40 */ 41 42 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 43 44 #include <linux/types.h> 45 #include <linux/list.h> /* For struct list_head */ 46 #include <linux/socket.h> 47 #include <linux/ip.h> 48 #include <linux/slab.h> 49 #include <net/sock.h> /* For skb_set_owner_w */ 50 51 #include <net/sctp/sctp.h> 52 #include <net/sctp/sm.h> 53 54 /* Declare internal functions here. */ 55 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn); 56 static void sctp_check_transmitted(struct sctp_outq *q, 57 struct list_head *transmitted_queue, 58 struct sctp_transport *transport, 59 union sctp_addr *saddr, 60 struct sctp_sackhdr *sack, 61 __u32 *highest_new_tsn); 62 63 static void sctp_mark_missing(struct sctp_outq *q, 64 struct list_head *transmitted_queue, 65 struct sctp_transport *transport, 66 __u32 highest_new_tsn, 67 int count_of_newacks); 68 69 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn); 70 71 static void sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp); 72 73 /* Add data to the front of the queue. */ 74 static inline void sctp_outq_head_data(struct sctp_outq *q, 75 struct sctp_chunk *ch) 76 { 77 list_add(&ch->list, &q->out_chunk_list); 78 q->out_qlen += ch->skb->len; 79 } 80 81 /* Take data from the front of the queue. */ 82 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q) 83 { 84 struct sctp_chunk *ch = NULL; 85 86 if (!list_empty(&q->out_chunk_list)) { 87 struct list_head *entry = q->out_chunk_list.next; 88 89 ch = list_entry(entry, struct sctp_chunk, list); 90 list_del_init(entry); 91 q->out_qlen -= ch->skb->len; 92 } 93 return ch; 94 } 95 /* Add data chunk to the end of the queue. */ 96 static inline void sctp_outq_tail_data(struct sctp_outq *q, 97 struct sctp_chunk *ch) 98 { 99 list_add_tail(&ch->list, &q->out_chunk_list); 100 q->out_qlen += ch->skb->len; 101 } 102 103 /* 104 * SFR-CACC algorithm: 105 * D) If count_of_newacks is greater than or equal to 2 106 * and t was not sent to the current primary then the 107 * sender MUST NOT increment missing report count for t. 108 */ 109 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary, 110 struct sctp_transport *transport, 111 int count_of_newacks) 112 { 113 if (count_of_newacks >= 2 && transport != primary) 114 return 1; 115 return 0; 116 } 117 118 /* 119 * SFR-CACC algorithm: 120 * F) If count_of_newacks is less than 2, let d be the 121 * destination to which t was sent. If cacc_saw_newack 122 * is 0 for destination d, then the sender MUST NOT 123 * increment missing report count for t. 124 */ 125 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport, 126 int count_of_newacks) 127 { 128 if (count_of_newacks < 2 && 129 (transport && !transport->cacc.cacc_saw_newack)) 130 return 1; 131 return 0; 132 } 133 134 /* 135 * SFR-CACC algorithm: 136 * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD 137 * execute steps C, D, F. 138 * 139 * C has been implemented in sctp_outq_sack 140 */ 141 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary, 142 struct sctp_transport *transport, 143 int count_of_newacks) 144 { 145 if (!primary->cacc.cycling_changeover) { 146 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks)) 147 return 1; 148 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks)) 149 return 1; 150 return 0; 151 } 152 return 0; 153 } 154 155 /* 156 * SFR-CACC algorithm: 157 * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less 158 * than next_tsn_at_change of the current primary, then 159 * the sender MUST NOT increment missing report count 160 * for t. 161 */ 162 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn) 163 { 164 if (primary->cacc.cycling_changeover && 165 TSN_lt(tsn, primary->cacc.next_tsn_at_change)) 166 return 1; 167 return 0; 168 } 169 170 /* 171 * SFR-CACC algorithm: 172 * 3) If the missing report count for TSN t is to be 173 * incremented according to [RFC2960] and 174 * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set, 175 * then the sender MUST further execute steps 3.1 and 176 * 3.2 to determine if the missing report count for 177 * TSN t SHOULD NOT be incremented. 178 * 179 * 3.3) If 3.1 and 3.2 do not dictate that the missing 180 * report count for t should not be incremented, then 181 * the sender SHOULD increment missing report count for 182 * t (according to [RFC2960] and [SCTP_STEWART_2002]). 183 */ 184 static inline int sctp_cacc_skip(struct sctp_transport *primary, 185 struct sctp_transport *transport, 186 int count_of_newacks, 187 __u32 tsn) 188 { 189 if (primary->cacc.changeover_active && 190 (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) || 191 sctp_cacc_skip_3_2(primary, tsn))) 192 return 1; 193 return 0; 194 } 195 196 /* Initialize an existing sctp_outq. This does the boring stuff. 197 * You still need to define handlers if you really want to DO 198 * something with this structure... 199 */ 200 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q) 201 { 202 memset(q, 0, sizeof(struct sctp_outq)); 203 204 q->asoc = asoc; 205 INIT_LIST_HEAD(&q->out_chunk_list); 206 INIT_LIST_HEAD(&q->control_chunk_list); 207 INIT_LIST_HEAD(&q->retransmit); 208 INIT_LIST_HEAD(&q->sacked); 209 INIT_LIST_HEAD(&q->abandoned); 210 } 211 212 /* Free the outqueue structure and any related pending chunks. 213 */ 214 static void __sctp_outq_teardown(struct sctp_outq *q) 215 { 216 struct sctp_transport *transport; 217 struct list_head *lchunk, *temp; 218 struct sctp_chunk *chunk, *tmp; 219 220 /* Throw away unacknowledged chunks. */ 221 list_for_each_entry(transport, &q->asoc->peer.transport_addr_list, 222 transports) { 223 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) { 224 chunk = list_entry(lchunk, struct sctp_chunk, 225 transmitted_list); 226 /* Mark as part of a failed message. */ 227 sctp_chunk_fail(chunk, q->error); 228 sctp_chunk_free(chunk); 229 } 230 } 231 232 /* Throw away chunks that have been gap ACKed. */ 233 list_for_each_safe(lchunk, temp, &q->sacked) { 234 list_del_init(lchunk); 235 chunk = list_entry(lchunk, struct sctp_chunk, 236 transmitted_list); 237 sctp_chunk_fail(chunk, q->error); 238 sctp_chunk_free(chunk); 239 } 240 241 /* Throw away any chunks in the retransmit queue. */ 242 list_for_each_safe(lchunk, temp, &q->retransmit) { 243 list_del_init(lchunk); 244 chunk = list_entry(lchunk, struct sctp_chunk, 245 transmitted_list); 246 sctp_chunk_fail(chunk, q->error); 247 sctp_chunk_free(chunk); 248 } 249 250 /* Throw away any chunks that are in the abandoned queue. */ 251 list_for_each_safe(lchunk, temp, &q->abandoned) { 252 list_del_init(lchunk); 253 chunk = list_entry(lchunk, struct sctp_chunk, 254 transmitted_list); 255 sctp_chunk_fail(chunk, q->error); 256 sctp_chunk_free(chunk); 257 } 258 259 /* Throw away any leftover data chunks. */ 260 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) { 261 262 /* Mark as send failure. */ 263 sctp_chunk_fail(chunk, q->error); 264 sctp_chunk_free(chunk); 265 } 266 267 /* Throw away any leftover control chunks. */ 268 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) { 269 list_del_init(&chunk->list); 270 sctp_chunk_free(chunk); 271 } 272 } 273 274 void sctp_outq_teardown(struct sctp_outq *q) 275 { 276 __sctp_outq_teardown(q); 277 sctp_outq_init(q->asoc, q); 278 } 279 280 /* Free the outqueue structure and any related pending chunks. */ 281 void sctp_outq_free(struct sctp_outq *q) 282 { 283 /* Throw away leftover chunks. */ 284 __sctp_outq_teardown(q); 285 } 286 287 /* Put a new chunk in an sctp_outq. */ 288 void sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk, gfp_t gfp) 289 { 290 struct net *net = sock_net(q->asoc->base.sk); 291 292 pr_debug("%s: outq:%p, chunk:%p[%s]\n", __func__, q, chunk, 293 chunk && chunk->chunk_hdr ? 294 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) : 295 "illegal chunk"); 296 297 /* If it is data, queue it up, otherwise, send it 298 * immediately. 299 */ 300 if (sctp_chunk_is_data(chunk)) { 301 pr_debug("%s: outqueueing: outq:%p, chunk:%p[%s])\n", 302 __func__, q, chunk, chunk && chunk->chunk_hdr ? 303 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) : 304 "illegal chunk"); 305 306 sctp_outq_tail_data(q, chunk); 307 if (chunk->asoc->peer.prsctp_capable && 308 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags)) 309 chunk->asoc->sent_cnt_removable++; 310 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED) 311 SCTP_INC_STATS(net, SCTP_MIB_OUTUNORDERCHUNKS); 312 else 313 SCTP_INC_STATS(net, SCTP_MIB_OUTORDERCHUNKS); 314 } else { 315 list_add_tail(&chunk->list, &q->control_chunk_list); 316 SCTP_INC_STATS(net, SCTP_MIB_OUTCTRLCHUNKS); 317 } 318 319 if (!q->cork) 320 sctp_outq_flush(q, 0, gfp); 321 } 322 323 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list 324 * and the abandoned list are in ascending order. 325 */ 326 static void sctp_insert_list(struct list_head *head, struct list_head *new) 327 { 328 struct list_head *pos; 329 struct sctp_chunk *nchunk, *lchunk; 330 __u32 ntsn, ltsn; 331 int done = 0; 332 333 nchunk = list_entry(new, struct sctp_chunk, transmitted_list); 334 ntsn = ntohl(nchunk->subh.data_hdr->tsn); 335 336 list_for_each(pos, head) { 337 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list); 338 ltsn = ntohl(lchunk->subh.data_hdr->tsn); 339 if (TSN_lt(ntsn, ltsn)) { 340 list_add(new, pos->prev); 341 done = 1; 342 break; 343 } 344 } 345 if (!done) 346 list_add_tail(new, head); 347 } 348 349 static int sctp_prsctp_prune_sent(struct sctp_association *asoc, 350 struct sctp_sndrcvinfo *sinfo, 351 struct list_head *queue, int msg_len) 352 { 353 struct sctp_chunk *chk, *temp; 354 355 list_for_each_entry_safe(chk, temp, queue, transmitted_list) { 356 struct sctp_stream_out *streamout; 357 358 if (!SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) || 359 chk->sinfo.sinfo_timetolive <= sinfo->sinfo_timetolive) 360 continue; 361 362 list_del_init(&chk->transmitted_list); 363 sctp_insert_list(&asoc->outqueue.abandoned, 364 &chk->transmitted_list); 365 366 streamout = &asoc->stream.out[chk->sinfo.sinfo_stream]; 367 asoc->sent_cnt_removable--; 368 asoc->abandoned_sent[SCTP_PR_INDEX(PRIO)]++; 369 streamout->abandoned_sent[SCTP_PR_INDEX(PRIO)]++; 370 371 if (!chk->tsn_gap_acked) { 372 if (chk->transport) 373 chk->transport->flight_size -= 374 sctp_data_size(chk); 375 asoc->outqueue.outstanding_bytes -= sctp_data_size(chk); 376 } 377 378 msg_len -= SCTP_DATA_SNDSIZE(chk) + 379 sizeof(struct sk_buff) + 380 sizeof(struct sctp_chunk); 381 if (msg_len <= 0) 382 break; 383 } 384 385 return msg_len; 386 } 387 388 static int sctp_prsctp_prune_unsent(struct sctp_association *asoc, 389 struct sctp_sndrcvinfo *sinfo, int msg_len) 390 { 391 struct sctp_outq *q = &asoc->outqueue; 392 struct sctp_chunk *chk, *temp; 393 394 list_for_each_entry_safe(chk, temp, &q->out_chunk_list, list) { 395 if (!SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) || 396 chk->sinfo.sinfo_timetolive <= sinfo->sinfo_timetolive) 397 continue; 398 399 list_del_init(&chk->list); 400 q->out_qlen -= chk->skb->len; 401 asoc->sent_cnt_removable--; 402 asoc->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++; 403 if (chk->sinfo.sinfo_stream < asoc->stream.outcnt) { 404 struct sctp_stream_out *streamout = 405 &asoc->stream.out[chk->sinfo.sinfo_stream]; 406 407 streamout->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++; 408 } 409 410 msg_len -= SCTP_DATA_SNDSIZE(chk) + 411 sizeof(struct sk_buff) + 412 sizeof(struct sctp_chunk); 413 sctp_chunk_free(chk); 414 if (msg_len <= 0) 415 break; 416 } 417 418 return msg_len; 419 } 420 421 /* Abandon the chunks according their priorities */ 422 void sctp_prsctp_prune(struct sctp_association *asoc, 423 struct sctp_sndrcvinfo *sinfo, int msg_len) 424 { 425 struct sctp_transport *transport; 426 427 if (!asoc->peer.prsctp_capable || !asoc->sent_cnt_removable) 428 return; 429 430 msg_len = sctp_prsctp_prune_sent(asoc, sinfo, 431 &asoc->outqueue.retransmit, 432 msg_len); 433 if (msg_len <= 0) 434 return; 435 436 list_for_each_entry(transport, &asoc->peer.transport_addr_list, 437 transports) { 438 msg_len = sctp_prsctp_prune_sent(asoc, sinfo, 439 &transport->transmitted, 440 msg_len); 441 if (msg_len <= 0) 442 return; 443 } 444 445 sctp_prsctp_prune_unsent(asoc, sinfo, msg_len); 446 } 447 448 /* Mark all the eligible packets on a transport for retransmission. */ 449 void sctp_retransmit_mark(struct sctp_outq *q, 450 struct sctp_transport *transport, 451 __u8 reason) 452 { 453 struct list_head *lchunk, *ltemp; 454 struct sctp_chunk *chunk; 455 456 /* Walk through the specified transmitted queue. */ 457 list_for_each_safe(lchunk, ltemp, &transport->transmitted) { 458 chunk = list_entry(lchunk, struct sctp_chunk, 459 transmitted_list); 460 461 /* If the chunk is abandoned, move it to abandoned list. */ 462 if (sctp_chunk_abandoned(chunk)) { 463 list_del_init(lchunk); 464 sctp_insert_list(&q->abandoned, lchunk); 465 466 /* If this chunk has not been previousely acked, 467 * stop considering it 'outstanding'. Our peer 468 * will most likely never see it since it will 469 * not be retransmitted 470 */ 471 if (!chunk->tsn_gap_acked) { 472 if (chunk->transport) 473 chunk->transport->flight_size -= 474 sctp_data_size(chunk); 475 q->outstanding_bytes -= sctp_data_size(chunk); 476 q->asoc->peer.rwnd += sctp_data_size(chunk); 477 } 478 continue; 479 } 480 481 /* If we are doing retransmission due to a timeout or pmtu 482 * discovery, only the chunks that are not yet acked should 483 * be added to the retransmit queue. 484 */ 485 if ((reason == SCTP_RTXR_FAST_RTX && 486 (chunk->fast_retransmit == SCTP_NEED_FRTX)) || 487 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) { 488 /* RFC 2960 6.2.1 Processing a Received SACK 489 * 490 * C) Any time a DATA chunk is marked for 491 * retransmission (via either T3-rtx timer expiration 492 * (Section 6.3.3) or via fast retransmit 493 * (Section 7.2.4)), add the data size of those 494 * chunks to the rwnd. 495 */ 496 q->asoc->peer.rwnd += sctp_data_size(chunk); 497 q->outstanding_bytes -= sctp_data_size(chunk); 498 if (chunk->transport) 499 transport->flight_size -= sctp_data_size(chunk); 500 501 /* sctpimpguide-05 Section 2.8.2 502 * M5) If a T3-rtx timer expires, the 503 * 'TSN.Missing.Report' of all affected TSNs is set 504 * to 0. 505 */ 506 chunk->tsn_missing_report = 0; 507 508 /* If a chunk that is being used for RTT measurement 509 * has to be retransmitted, we cannot use this chunk 510 * anymore for RTT measurements. Reset rto_pending so 511 * that a new RTT measurement is started when a new 512 * data chunk is sent. 513 */ 514 if (chunk->rtt_in_progress) { 515 chunk->rtt_in_progress = 0; 516 transport->rto_pending = 0; 517 } 518 519 /* Move the chunk to the retransmit queue. The chunks 520 * on the retransmit queue are always kept in order. 521 */ 522 list_del_init(lchunk); 523 sctp_insert_list(&q->retransmit, lchunk); 524 } 525 } 526 527 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, " 528 "flight_size:%d, pba:%d\n", __func__, transport, reason, 529 transport->cwnd, transport->ssthresh, transport->flight_size, 530 transport->partial_bytes_acked); 531 } 532 533 /* Mark all the eligible packets on a transport for retransmission and force 534 * one packet out. 535 */ 536 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport, 537 enum sctp_retransmit_reason reason) 538 { 539 struct net *net = sock_net(q->asoc->base.sk); 540 541 switch (reason) { 542 case SCTP_RTXR_T3_RTX: 543 SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS); 544 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX); 545 /* Update the retran path if the T3-rtx timer has expired for 546 * the current retran path. 547 */ 548 if (transport == transport->asoc->peer.retran_path) 549 sctp_assoc_update_retran_path(transport->asoc); 550 transport->asoc->rtx_data_chunks += 551 transport->asoc->unack_data; 552 break; 553 case SCTP_RTXR_FAST_RTX: 554 SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS); 555 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX); 556 q->fast_rtx = 1; 557 break; 558 case SCTP_RTXR_PMTUD: 559 SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS); 560 break; 561 case SCTP_RTXR_T1_RTX: 562 SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS); 563 transport->asoc->init_retries++; 564 break; 565 default: 566 BUG(); 567 } 568 569 sctp_retransmit_mark(q, transport, reason); 570 571 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination, 572 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by 573 * following the procedures outlined in C1 - C5. 574 */ 575 if (reason == SCTP_RTXR_T3_RTX) 576 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point); 577 578 /* Flush the queues only on timeout, since fast_rtx is only 579 * triggered during sack processing and the queue 580 * will be flushed at the end. 581 */ 582 if (reason != SCTP_RTXR_FAST_RTX) 583 sctp_outq_flush(q, /* rtx_timeout */ 1, GFP_ATOMIC); 584 } 585 586 /* 587 * Transmit DATA chunks on the retransmit queue. Upon return from 588 * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which 589 * need to be transmitted by the caller. 590 * We assume that pkt->transport has already been set. 591 * 592 * The return value is a normal kernel error return value. 593 */ 594 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt, 595 int rtx_timeout, int *start_timer) 596 { 597 struct sctp_transport *transport = pkt->transport; 598 struct sctp_chunk *chunk, *chunk1; 599 struct list_head *lqueue; 600 enum sctp_xmit status; 601 int error = 0; 602 int timer = 0; 603 int done = 0; 604 int fast_rtx; 605 606 lqueue = &q->retransmit; 607 fast_rtx = q->fast_rtx; 608 609 /* This loop handles time-out retransmissions, fast retransmissions, 610 * and retransmissions due to opening of whindow. 611 * 612 * RFC 2960 6.3.3 Handle T3-rtx Expiration 613 * 614 * E3) Determine how many of the earliest (i.e., lowest TSN) 615 * outstanding DATA chunks for the address for which the 616 * T3-rtx has expired will fit into a single packet, subject 617 * to the MTU constraint for the path corresponding to the 618 * destination transport address to which the retransmission 619 * is being sent (this may be different from the address for 620 * which the timer expires [see Section 6.4]). Call this value 621 * K. Bundle and retransmit those K DATA chunks in a single 622 * packet to the destination endpoint. 623 * 624 * [Just to be painfully clear, if we are retransmitting 625 * because a timeout just happened, we should send only ONE 626 * packet of retransmitted data.] 627 * 628 * For fast retransmissions we also send only ONE packet. However, 629 * if we are just flushing the queue due to open window, we'll 630 * try to send as much as possible. 631 */ 632 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) { 633 /* If the chunk is abandoned, move it to abandoned list. */ 634 if (sctp_chunk_abandoned(chunk)) { 635 list_del_init(&chunk->transmitted_list); 636 sctp_insert_list(&q->abandoned, 637 &chunk->transmitted_list); 638 continue; 639 } 640 641 /* Make sure that Gap Acked TSNs are not retransmitted. A 642 * simple approach is just to move such TSNs out of the 643 * way and into a 'transmitted' queue and skip to the 644 * next chunk. 645 */ 646 if (chunk->tsn_gap_acked) { 647 list_move_tail(&chunk->transmitted_list, 648 &transport->transmitted); 649 continue; 650 } 651 652 /* If we are doing fast retransmit, ignore non-fast_rtransmit 653 * chunks 654 */ 655 if (fast_rtx && !chunk->fast_retransmit) 656 continue; 657 658 redo: 659 /* Attempt to append this chunk to the packet. */ 660 status = sctp_packet_append_chunk(pkt, chunk); 661 662 switch (status) { 663 case SCTP_XMIT_PMTU_FULL: 664 if (!pkt->has_data && !pkt->has_cookie_echo) { 665 /* If this packet did not contain DATA then 666 * retransmission did not happen, so do it 667 * again. We'll ignore the error here since 668 * control chunks are already freed so there 669 * is nothing we can do. 670 */ 671 sctp_packet_transmit(pkt, GFP_ATOMIC); 672 goto redo; 673 } 674 675 /* Send this packet. */ 676 error = sctp_packet_transmit(pkt, GFP_ATOMIC); 677 678 /* If we are retransmitting, we should only 679 * send a single packet. 680 * Otherwise, try appending this chunk again. 681 */ 682 if (rtx_timeout || fast_rtx) 683 done = 1; 684 else 685 goto redo; 686 687 /* Bundle next chunk in the next round. */ 688 break; 689 690 case SCTP_XMIT_RWND_FULL: 691 /* Send this packet. */ 692 error = sctp_packet_transmit(pkt, GFP_ATOMIC); 693 694 /* Stop sending DATA as there is no more room 695 * at the receiver. 696 */ 697 done = 1; 698 break; 699 700 case SCTP_XMIT_DELAY: 701 /* Send this packet. */ 702 error = sctp_packet_transmit(pkt, GFP_ATOMIC); 703 704 /* Stop sending DATA because of nagle delay. */ 705 done = 1; 706 break; 707 708 default: 709 /* The append was successful, so add this chunk to 710 * the transmitted list. 711 */ 712 list_move_tail(&chunk->transmitted_list, 713 &transport->transmitted); 714 715 /* Mark the chunk as ineligible for fast retransmit 716 * after it is retransmitted. 717 */ 718 if (chunk->fast_retransmit == SCTP_NEED_FRTX) 719 chunk->fast_retransmit = SCTP_DONT_FRTX; 720 721 q->asoc->stats.rtxchunks++; 722 break; 723 } 724 725 /* Set the timer if there were no errors */ 726 if (!error && !timer) 727 timer = 1; 728 729 if (done) 730 break; 731 } 732 733 /* If we are here due to a retransmit timeout or a fast 734 * retransmit and if there are any chunks left in the retransmit 735 * queue that could not fit in the PMTU sized packet, they need 736 * to be marked as ineligible for a subsequent fast retransmit. 737 */ 738 if (rtx_timeout || fast_rtx) { 739 list_for_each_entry(chunk1, lqueue, transmitted_list) { 740 if (chunk1->fast_retransmit == SCTP_NEED_FRTX) 741 chunk1->fast_retransmit = SCTP_DONT_FRTX; 742 } 743 } 744 745 *start_timer = timer; 746 747 /* Clear fast retransmit hint */ 748 if (fast_rtx) 749 q->fast_rtx = 0; 750 751 return error; 752 } 753 754 /* Cork the outqueue so queued chunks are really queued. */ 755 void sctp_outq_uncork(struct sctp_outq *q, gfp_t gfp) 756 { 757 if (q->cork) 758 q->cork = 0; 759 760 sctp_outq_flush(q, 0, gfp); 761 } 762 763 764 /* 765 * Try to flush an outqueue. 766 * 767 * Description: Send everything in q which we legally can, subject to 768 * congestion limitations. 769 * * Note: This function can be called from multiple contexts so appropriate 770 * locking concerns must be made. Today we use the sock lock to protect 771 * this function. 772 */ 773 static void sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp) 774 { 775 struct sctp_packet *packet; 776 struct sctp_packet singleton; 777 struct sctp_association *asoc = q->asoc; 778 __u16 sport = asoc->base.bind_addr.port; 779 __u16 dport = asoc->peer.port; 780 __u32 vtag = asoc->peer.i.init_tag; 781 struct sctp_transport *transport = NULL; 782 struct sctp_transport *new_transport; 783 struct sctp_chunk *chunk, *tmp; 784 enum sctp_xmit status; 785 int error = 0; 786 int start_timer = 0; 787 int one_packet = 0; 788 789 /* These transports have chunks to send. */ 790 struct list_head transport_list; 791 struct list_head *ltransport; 792 793 INIT_LIST_HEAD(&transport_list); 794 packet = NULL; 795 796 /* 797 * 6.10 Bundling 798 * ... 799 * When bundling control chunks with DATA chunks, an 800 * endpoint MUST place control chunks first in the outbound 801 * SCTP packet. The transmitter MUST transmit DATA chunks 802 * within a SCTP packet in increasing order of TSN. 803 * ... 804 */ 805 806 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) { 807 /* RFC 5061, 5.3 808 * F1) This means that until such time as the ASCONF 809 * containing the add is acknowledged, the sender MUST 810 * NOT use the new IP address as a source for ANY SCTP 811 * packet except on carrying an ASCONF Chunk. 812 */ 813 if (asoc->src_out_of_asoc_ok && 814 chunk->chunk_hdr->type != SCTP_CID_ASCONF) 815 continue; 816 817 list_del_init(&chunk->list); 818 819 /* Pick the right transport to use. */ 820 new_transport = chunk->transport; 821 822 if (!new_transport) { 823 /* 824 * If we have a prior transport pointer, see if 825 * the destination address of the chunk 826 * matches the destination address of the 827 * current transport. If not a match, then 828 * try to look up the transport with a given 829 * destination address. We do this because 830 * after processing ASCONFs, we may have new 831 * transports created. 832 */ 833 if (transport && 834 sctp_cmp_addr_exact(&chunk->dest, 835 &transport->ipaddr)) 836 new_transport = transport; 837 else 838 new_transport = sctp_assoc_lookup_paddr(asoc, 839 &chunk->dest); 840 841 /* if we still don't have a new transport, then 842 * use the current active path. 843 */ 844 if (!new_transport) 845 new_transport = asoc->peer.active_path; 846 } else if ((new_transport->state == SCTP_INACTIVE) || 847 (new_transport->state == SCTP_UNCONFIRMED) || 848 (new_transport->state == SCTP_PF)) { 849 /* If the chunk is Heartbeat or Heartbeat Ack, 850 * send it to chunk->transport, even if it's 851 * inactive. 852 * 853 * 3.3.6 Heartbeat Acknowledgement: 854 * ... 855 * A HEARTBEAT ACK is always sent to the source IP 856 * address of the IP datagram containing the 857 * HEARTBEAT chunk to which this ack is responding. 858 * ... 859 * 860 * ASCONF_ACKs also must be sent to the source. 861 */ 862 if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT && 863 chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK && 864 chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK) 865 new_transport = asoc->peer.active_path; 866 } 867 868 /* Are we switching transports? 869 * Take care of transport locks. 870 */ 871 if (new_transport != transport) { 872 transport = new_transport; 873 if (list_empty(&transport->send_ready)) { 874 list_add_tail(&transport->send_ready, 875 &transport_list); 876 } 877 packet = &transport->packet; 878 sctp_packet_config(packet, vtag, 879 asoc->peer.ecn_capable); 880 } 881 882 switch (chunk->chunk_hdr->type) { 883 /* 884 * 6.10 Bundling 885 * ... 886 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN 887 * COMPLETE with any other chunks. [Send them immediately.] 888 */ 889 case SCTP_CID_INIT: 890 case SCTP_CID_INIT_ACK: 891 case SCTP_CID_SHUTDOWN_COMPLETE: 892 sctp_packet_init(&singleton, transport, sport, dport); 893 sctp_packet_config(&singleton, vtag, 0); 894 sctp_packet_append_chunk(&singleton, chunk); 895 error = sctp_packet_transmit(&singleton, gfp); 896 if (error < 0) { 897 asoc->base.sk->sk_err = -error; 898 return; 899 } 900 break; 901 902 case SCTP_CID_ABORT: 903 if (sctp_test_T_bit(chunk)) { 904 packet->vtag = asoc->c.my_vtag; 905 } 906 /* The following chunks are "response" chunks, i.e. 907 * they are generated in response to something we 908 * received. If we are sending these, then we can 909 * send only 1 packet containing these chunks. 910 */ 911 case SCTP_CID_HEARTBEAT_ACK: 912 case SCTP_CID_SHUTDOWN_ACK: 913 case SCTP_CID_COOKIE_ACK: 914 case SCTP_CID_COOKIE_ECHO: 915 case SCTP_CID_ERROR: 916 case SCTP_CID_ECN_CWR: 917 case SCTP_CID_ASCONF_ACK: 918 one_packet = 1; 919 /* Fall through */ 920 921 case SCTP_CID_SACK: 922 case SCTP_CID_HEARTBEAT: 923 case SCTP_CID_SHUTDOWN: 924 case SCTP_CID_ECN_ECNE: 925 case SCTP_CID_ASCONF: 926 case SCTP_CID_FWD_TSN: 927 case SCTP_CID_RECONF: 928 status = sctp_packet_transmit_chunk(packet, chunk, 929 one_packet, gfp); 930 if (status != SCTP_XMIT_OK) { 931 /* put the chunk back */ 932 list_add(&chunk->list, &q->control_chunk_list); 933 break; 934 } 935 936 asoc->stats.octrlchunks++; 937 /* PR-SCTP C5) If a FORWARD TSN is sent, the 938 * sender MUST assure that at least one T3-rtx 939 * timer is running. 940 */ 941 if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN) { 942 sctp_transport_reset_t3_rtx(transport); 943 transport->last_time_sent = jiffies; 944 } 945 946 if (chunk == asoc->strreset_chunk) 947 sctp_transport_reset_reconf_timer(transport); 948 949 break; 950 951 default: 952 /* We built a chunk with an illegal type! */ 953 BUG(); 954 } 955 } 956 957 if (q->asoc->src_out_of_asoc_ok) 958 goto sctp_flush_out; 959 960 /* Is it OK to send data chunks? */ 961 switch (asoc->state) { 962 case SCTP_STATE_COOKIE_ECHOED: 963 /* Only allow bundling when this packet has a COOKIE-ECHO 964 * chunk. 965 */ 966 if (!packet || !packet->has_cookie_echo) 967 break; 968 969 /* fallthru */ 970 case SCTP_STATE_ESTABLISHED: 971 case SCTP_STATE_SHUTDOWN_PENDING: 972 case SCTP_STATE_SHUTDOWN_RECEIVED: 973 /* 974 * RFC 2960 6.1 Transmission of DATA Chunks 975 * 976 * C) When the time comes for the sender to transmit, 977 * before sending new DATA chunks, the sender MUST 978 * first transmit any outstanding DATA chunks which 979 * are marked for retransmission (limited by the 980 * current cwnd). 981 */ 982 if (!list_empty(&q->retransmit)) { 983 if (asoc->peer.retran_path->state == SCTP_UNCONFIRMED) 984 goto sctp_flush_out; 985 if (transport == asoc->peer.retran_path) 986 goto retran; 987 988 /* Switch transports & prepare the packet. */ 989 990 transport = asoc->peer.retran_path; 991 992 if (list_empty(&transport->send_ready)) { 993 list_add_tail(&transport->send_ready, 994 &transport_list); 995 } 996 997 packet = &transport->packet; 998 sctp_packet_config(packet, vtag, 999 asoc->peer.ecn_capable); 1000 retran: 1001 error = sctp_outq_flush_rtx(q, packet, 1002 rtx_timeout, &start_timer); 1003 if (error < 0) 1004 asoc->base.sk->sk_err = -error; 1005 1006 if (start_timer) { 1007 sctp_transport_reset_t3_rtx(transport); 1008 transport->last_time_sent = jiffies; 1009 } 1010 1011 /* This can happen on COOKIE-ECHO resend. Only 1012 * one chunk can get bundled with a COOKIE-ECHO. 1013 */ 1014 if (packet->has_cookie_echo) 1015 goto sctp_flush_out; 1016 1017 /* Don't send new data if there is still data 1018 * waiting to retransmit. 1019 */ 1020 if (!list_empty(&q->retransmit)) 1021 goto sctp_flush_out; 1022 } 1023 1024 /* Apply Max.Burst limitation to the current transport in 1025 * case it will be used for new data. We are going to 1026 * rest it before we return, but we want to apply the limit 1027 * to the currently queued data. 1028 */ 1029 if (transport) 1030 sctp_transport_burst_limited(transport); 1031 1032 /* Finally, transmit new packets. */ 1033 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) { 1034 __u32 sid = ntohs(chunk->subh.data_hdr->stream); 1035 1036 /* RFC 2960 6.5 Every DATA chunk MUST carry a valid 1037 * stream identifier. 1038 */ 1039 if (chunk->sinfo.sinfo_stream >= asoc->stream.outcnt) { 1040 1041 /* Mark as failed send. */ 1042 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM); 1043 if (asoc->peer.prsctp_capable && 1044 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags)) 1045 asoc->sent_cnt_removable--; 1046 sctp_chunk_free(chunk); 1047 continue; 1048 } 1049 1050 /* Has this chunk expired? */ 1051 if (sctp_chunk_abandoned(chunk)) { 1052 sctp_chunk_fail(chunk, 0); 1053 sctp_chunk_free(chunk); 1054 continue; 1055 } 1056 1057 if (asoc->stream.out[sid].state == SCTP_STREAM_CLOSED) { 1058 sctp_outq_head_data(q, chunk); 1059 goto sctp_flush_out; 1060 } 1061 1062 /* If there is a specified transport, use it. 1063 * Otherwise, we want to use the active path. 1064 */ 1065 new_transport = chunk->transport; 1066 if (!new_transport || 1067 ((new_transport->state == SCTP_INACTIVE) || 1068 (new_transport->state == SCTP_UNCONFIRMED) || 1069 (new_transport->state == SCTP_PF))) 1070 new_transport = asoc->peer.active_path; 1071 if (new_transport->state == SCTP_UNCONFIRMED) { 1072 WARN_ONCE(1, "Attempt to send packet on unconfirmed path."); 1073 sctp_chunk_fail(chunk, 0); 1074 sctp_chunk_free(chunk); 1075 continue; 1076 } 1077 1078 /* Change packets if necessary. */ 1079 if (new_transport != transport) { 1080 transport = new_transport; 1081 1082 /* Schedule to have this transport's 1083 * packet flushed. 1084 */ 1085 if (list_empty(&transport->send_ready)) { 1086 list_add_tail(&transport->send_ready, 1087 &transport_list); 1088 } 1089 1090 packet = &transport->packet; 1091 sctp_packet_config(packet, vtag, 1092 asoc->peer.ecn_capable); 1093 /* We've switched transports, so apply the 1094 * Burst limit to the new transport. 1095 */ 1096 sctp_transport_burst_limited(transport); 1097 } 1098 1099 pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p " 1100 "skb->users:%d\n", 1101 __func__, q, chunk, chunk && chunk->chunk_hdr ? 1102 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) : 1103 "illegal chunk", ntohl(chunk->subh.data_hdr->tsn), 1104 chunk->skb ? chunk->skb->head : NULL, chunk->skb ? 1105 refcount_read(&chunk->skb->users) : -1); 1106 1107 /* Add the chunk to the packet. */ 1108 status = sctp_packet_transmit_chunk(packet, chunk, 0, gfp); 1109 1110 switch (status) { 1111 case SCTP_XMIT_PMTU_FULL: 1112 case SCTP_XMIT_RWND_FULL: 1113 case SCTP_XMIT_DELAY: 1114 /* We could not append this chunk, so put 1115 * the chunk back on the output queue. 1116 */ 1117 pr_debug("%s: could not transmit tsn:0x%x, status:%d\n", 1118 __func__, ntohl(chunk->subh.data_hdr->tsn), 1119 status); 1120 1121 sctp_outq_head_data(q, chunk); 1122 goto sctp_flush_out; 1123 1124 case SCTP_XMIT_OK: 1125 /* The sender is in the SHUTDOWN-PENDING state, 1126 * The sender MAY set the I-bit in the DATA 1127 * chunk header. 1128 */ 1129 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING) 1130 chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM; 1131 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED) 1132 asoc->stats.ouodchunks++; 1133 else 1134 asoc->stats.oodchunks++; 1135 1136 break; 1137 1138 default: 1139 BUG(); 1140 } 1141 1142 /* BUG: We assume that the sctp_packet_transmit() 1143 * call below will succeed all the time and add the 1144 * chunk to the transmitted list and restart the 1145 * timers. 1146 * It is possible that the call can fail under OOM 1147 * conditions. 1148 * 1149 * Is this really a problem? Won't this behave 1150 * like a lost TSN? 1151 */ 1152 list_add_tail(&chunk->transmitted_list, 1153 &transport->transmitted); 1154 1155 sctp_transport_reset_t3_rtx(transport); 1156 transport->last_time_sent = jiffies; 1157 1158 /* Only let one DATA chunk get bundled with a 1159 * COOKIE-ECHO chunk. 1160 */ 1161 if (packet->has_cookie_echo) 1162 goto sctp_flush_out; 1163 } 1164 break; 1165 1166 default: 1167 /* Do nothing. */ 1168 break; 1169 } 1170 1171 sctp_flush_out: 1172 1173 /* Before returning, examine all the transports touched in 1174 * this call. Right now, we bluntly force clear all the 1175 * transports. Things might change after we implement Nagle. 1176 * But such an examination is still required. 1177 * 1178 * --xguo 1179 */ 1180 while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL) { 1181 struct sctp_transport *t = list_entry(ltransport, 1182 struct sctp_transport, 1183 send_ready); 1184 packet = &t->packet; 1185 if (!sctp_packet_empty(packet)) { 1186 error = sctp_packet_transmit(packet, gfp); 1187 if (error < 0) 1188 asoc->base.sk->sk_err = -error; 1189 } 1190 1191 /* Clear the burst limited state, if any */ 1192 sctp_transport_burst_reset(t); 1193 } 1194 } 1195 1196 /* Update unack_data based on the incoming SACK chunk */ 1197 static void sctp_sack_update_unack_data(struct sctp_association *assoc, 1198 struct sctp_sackhdr *sack) 1199 { 1200 union sctp_sack_variable *frags; 1201 __u16 unack_data; 1202 int i; 1203 1204 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1; 1205 1206 frags = sack->variable; 1207 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) { 1208 unack_data -= ((ntohs(frags[i].gab.end) - 1209 ntohs(frags[i].gab.start) + 1)); 1210 } 1211 1212 assoc->unack_data = unack_data; 1213 } 1214 1215 /* This is where we REALLY process a SACK. 1216 * 1217 * Process the SACK against the outqueue. Mostly, this just frees 1218 * things off the transmitted queue. 1219 */ 1220 int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk) 1221 { 1222 struct sctp_association *asoc = q->asoc; 1223 struct sctp_sackhdr *sack = chunk->subh.sack_hdr; 1224 struct sctp_transport *transport; 1225 struct sctp_chunk *tchunk = NULL; 1226 struct list_head *lchunk, *transport_list, *temp; 1227 union sctp_sack_variable *frags = sack->variable; 1228 __u32 sack_ctsn, ctsn, tsn; 1229 __u32 highest_tsn, highest_new_tsn; 1230 __u32 sack_a_rwnd; 1231 unsigned int outstanding; 1232 struct sctp_transport *primary = asoc->peer.primary_path; 1233 int count_of_newacks = 0; 1234 int gap_ack_blocks; 1235 u8 accum_moved = 0; 1236 1237 /* Grab the association's destination address list. */ 1238 transport_list = &asoc->peer.transport_addr_list; 1239 1240 sack_ctsn = ntohl(sack->cum_tsn_ack); 1241 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks); 1242 asoc->stats.gapcnt += gap_ack_blocks; 1243 /* 1244 * SFR-CACC algorithm: 1245 * On receipt of a SACK the sender SHOULD execute the 1246 * following statements. 1247 * 1248 * 1) If the cumulative ack in the SACK passes next tsn_at_change 1249 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be 1250 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for 1251 * all destinations. 1252 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE 1253 * is set the receiver of the SACK MUST take the following actions: 1254 * 1255 * A) Initialize the cacc_saw_newack to 0 for all destination 1256 * addresses. 1257 * 1258 * Only bother if changeover_active is set. Otherwise, this is 1259 * totally suboptimal to do on every SACK. 1260 */ 1261 if (primary->cacc.changeover_active) { 1262 u8 clear_cycling = 0; 1263 1264 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) { 1265 primary->cacc.changeover_active = 0; 1266 clear_cycling = 1; 1267 } 1268 1269 if (clear_cycling || gap_ack_blocks) { 1270 list_for_each_entry(transport, transport_list, 1271 transports) { 1272 if (clear_cycling) 1273 transport->cacc.cycling_changeover = 0; 1274 if (gap_ack_blocks) 1275 transport->cacc.cacc_saw_newack = 0; 1276 } 1277 } 1278 } 1279 1280 /* Get the highest TSN in the sack. */ 1281 highest_tsn = sack_ctsn; 1282 if (gap_ack_blocks) 1283 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end); 1284 1285 if (TSN_lt(asoc->highest_sacked, highest_tsn)) 1286 asoc->highest_sacked = highest_tsn; 1287 1288 highest_new_tsn = sack_ctsn; 1289 1290 /* Run through the retransmit queue. Credit bytes received 1291 * and free those chunks that we can. 1292 */ 1293 sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn); 1294 1295 /* Run through the transmitted queue. 1296 * Credit bytes received and free those chunks which we can. 1297 * 1298 * This is a MASSIVE candidate for optimization. 1299 */ 1300 list_for_each_entry(transport, transport_list, transports) { 1301 sctp_check_transmitted(q, &transport->transmitted, 1302 transport, &chunk->source, sack, 1303 &highest_new_tsn); 1304 /* 1305 * SFR-CACC algorithm: 1306 * C) Let count_of_newacks be the number of 1307 * destinations for which cacc_saw_newack is set. 1308 */ 1309 if (transport->cacc.cacc_saw_newack) 1310 count_of_newacks++; 1311 } 1312 1313 /* Move the Cumulative TSN Ack Point if appropriate. */ 1314 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) { 1315 asoc->ctsn_ack_point = sack_ctsn; 1316 accum_moved = 1; 1317 } 1318 1319 if (gap_ack_blocks) { 1320 1321 if (asoc->fast_recovery && accum_moved) 1322 highest_new_tsn = highest_tsn; 1323 1324 list_for_each_entry(transport, transport_list, transports) 1325 sctp_mark_missing(q, &transport->transmitted, transport, 1326 highest_new_tsn, count_of_newacks); 1327 } 1328 1329 /* Update unack_data field in the assoc. */ 1330 sctp_sack_update_unack_data(asoc, sack); 1331 1332 ctsn = asoc->ctsn_ack_point; 1333 1334 /* Throw away stuff rotting on the sack queue. */ 1335 list_for_each_safe(lchunk, temp, &q->sacked) { 1336 tchunk = list_entry(lchunk, struct sctp_chunk, 1337 transmitted_list); 1338 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1339 if (TSN_lte(tsn, ctsn)) { 1340 list_del_init(&tchunk->transmitted_list); 1341 if (asoc->peer.prsctp_capable && 1342 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags)) 1343 asoc->sent_cnt_removable--; 1344 sctp_chunk_free(tchunk); 1345 } 1346 } 1347 1348 /* ii) Set rwnd equal to the newly received a_rwnd minus the 1349 * number of bytes still outstanding after processing the 1350 * Cumulative TSN Ack and the Gap Ack Blocks. 1351 */ 1352 1353 sack_a_rwnd = ntohl(sack->a_rwnd); 1354 asoc->peer.zero_window_announced = !sack_a_rwnd; 1355 outstanding = q->outstanding_bytes; 1356 1357 if (outstanding < sack_a_rwnd) 1358 sack_a_rwnd -= outstanding; 1359 else 1360 sack_a_rwnd = 0; 1361 1362 asoc->peer.rwnd = sack_a_rwnd; 1363 1364 sctp_generate_fwdtsn(q, sack_ctsn); 1365 1366 pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__, sack_ctsn); 1367 pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, " 1368 "advertised peer ack point:0x%x\n", __func__, asoc, ctsn, 1369 asoc->adv_peer_ack_point); 1370 1371 return sctp_outq_is_empty(q); 1372 } 1373 1374 /* Is the outqueue empty? 1375 * The queue is empty when we have not pending data, no in-flight data 1376 * and nothing pending retransmissions. 1377 */ 1378 int sctp_outq_is_empty(const struct sctp_outq *q) 1379 { 1380 return q->out_qlen == 0 && q->outstanding_bytes == 0 && 1381 list_empty(&q->retransmit); 1382 } 1383 1384 /******************************************************************** 1385 * 2nd Level Abstractions 1386 ********************************************************************/ 1387 1388 /* Go through a transport's transmitted list or the association's retransmit 1389 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked. 1390 * The retransmit list will not have an associated transport. 1391 * 1392 * I added coherent debug information output. --xguo 1393 * 1394 * Instead of printing 'sacked' or 'kept' for each TSN on the 1395 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5. 1396 * KEPT TSN6-TSN7, etc. 1397 */ 1398 static void sctp_check_transmitted(struct sctp_outq *q, 1399 struct list_head *transmitted_queue, 1400 struct sctp_transport *transport, 1401 union sctp_addr *saddr, 1402 struct sctp_sackhdr *sack, 1403 __u32 *highest_new_tsn_in_sack) 1404 { 1405 struct list_head *lchunk; 1406 struct sctp_chunk *tchunk; 1407 struct list_head tlist; 1408 __u32 tsn; 1409 __u32 sack_ctsn; 1410 __u32 rtt; 1411 __u8 restart_timer = 0; 1412 int bytes_acked = 0; 1413 int migrate_bytes = 0; 1414 bool forward_progress = false; 1415 1416 sack_ctsn = ntohl(sack->cum_tsn_ack); 1417 1418 INIT_LIST_HEAD(&tlist); 1419 1420 /* The while loop will skip empty transmitted queues. */ 1421 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) { 1422 tchunk = list_entry(lchunk, struct sctp_chunk, 1423 transmitted_list); 1424 1425 if (sctp_chunk_abandoned(tchunk)) { 1426 /* Move the chunk to abandoned list. */ 1427 sctp_insert_list(&q->abandoned, lchunk); 1428 1429 /* If this chunk has not been acked, stop 1430 * considering it as 'outstanding'. 1431 */ 1432 if (!tchunk->tsn_gap_acked) { 1433 if (tchunk->transport) 1434 tchunk->transport->flight_size -= 1435 sctp_data_size(tchunk); 1436 q->outstanding_bytes -= sctp_data_size(tchunk); 1437 } 1438 continue; 1439 } 1440 1441 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1442 if (sctp_acked(sack, tsn)) { 1443 /* If this queue is the retransmit queue, the 1444 * retransmit timer has already reclaimed 1445 * the outstanding bytes for this chunk, so only 1446 * count bytes associated with a transport. 1447 */ 1448 if (transport) { 1449 /* If this chunk is being used for RTT 1450 * measurement, calculate the RTT and update 1451 * the RTO using this value. 1452 * 1453 * 6.3.1 C5) Karn's algorithm: RTT measurements 1454 * MUST NOT be made using packets that were 1455 * retransmitted (and thus for which it is 1456 * ambiguous whether the reply was for the 1457 * first instance of the packet or a later 1458 * instance). 1459 */ 1460 if (!tchunk->tsn_gap_acked && 1461 !sctp_chunk_retransmitted(tchunk) && 1462 tchunk->rtt_in_progress) { 1463 tchunk->rtt_in_progress = 0; 1464 rtt = jiffies - tchunk->sent_at; 1465 sctp_transport_update_rto(transport, 1466 rtt); 1467 } 1468 } 1469 1470 /* If the chunk hasn't been marked as ACKED, 1471 * mark it and account bytes_acked if the 1472 * chunk had a valid transport (it will not 1473 * have a transport if ASCONF had deleted it 1474 * while DATA was outstanding). 1475 */ 1476 if (!tchunk->tsn_gap_acked) { 1477 tchunk->tsn_gap_acked = 1; 1478 if (TSN_lt(*highest_new_tsn_in_sack, tsn)) 1479 *highest_new_tsn_in_sack = tsn; 1480 bytes_acked += sctp_data_size(tchunk); 1481 if (!tchunk->transport) 1482 migrate_bytes += sctp_data_size(tchunk); 1483 forward_progress = true; 1484 } 1485 1486 if (TSN_lte(tsn, sack_ctsn)) { 1487 /* RFC 2960 6.3.2 Retransmission Timer Rules 1488 * 1489 * R3) Whenever a SACK is received 1490 * that acknowledges the DATA chunk 1491 * with the earliest outstanding TSN 1492 * for that address, restart T3-rtx 1493 * timer for that address with its 1494 * current RTO. 1495 */ 1496 restart_timer = 1; 1497 forward_progress = true; 1498 1499 if (!tchunk->tsn_gap_acked) { 1500 /* 1501 * SFR-CACC algorithm: 1502 * 2) If the SACK contains gap acks 1503 * and the flag CHANGEOVER_ACTIVE is 1504 * set the receiver of the SACK MUST 1505 * take the following action: 1506 * 1507 * B) For each TSN t being acked that 1508 * has not been acked in any SACK so 1509 * far, set cacc_saw_newack to 1 for 1510 * the destination that the TSN was 1511 * sent to. 1512 */ 1513 if (transport && 1514 sack->num_gap_ack_blocks && 1515 q->asoc->peer.primary_path->cacc. 1516 changeover_active) 1517 transport->cacc.cacc_saw_newack 1518 = 1; 1519 } 1520 1521 list_add_tail(&tchunk->transmitted_list, 1522 &q->sacked); 1523 } else { 1524 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2 1525 * M2) Each time a SACK arrives reporting 1526 * 'Stray DATA chunk(s)' record the highest TSN 1527 * reported as newly acknowledged, call this 1528 * value 'HighestTSNinSack'. A newly 1529 * acknowledged DATA chunk is one not 1530 * previously acknowledged in a SACK. 1531 * 1532 * When the SCTP sender of data receives a SACK 1533 * chunk that acknowledges, for the first time, 1534 * the receipt of a DATA chunk, all the still 1535 * unacknowledged DATA chunks whose TSN is 1536 * older than that newly acknowledged DATA 1537 * chunk, are qualified as 'Stray DATA chunks'. 1538 */ 1539 list_add_tail(lchunk, &tlist); 1540 } 1541 } else { 1542 if (tchunk->tsn_gap_acked) { 1543 pr_debug("%s: receiver reneged on data TSN:0x%x\n", 1544 __func__, tsn); 1545 1546 tchunk->tsn_gap_acked = 0; 1547 1548 if (tchunk->transport) 1549 bytes_acked -= sctp_data_size(tchunk); 1550 1551 /* RFC 2960 6.3.2 Retransmission Timer Rules 1552 * 1553 * R4) Whenever a SACK is received missing a 1554 * TSN that was previously acknowledged via a 1555 * Gap Ack Block, start T3-rtx for the 1556 * destination address to which the DATA 1557 * chunk was originally 1558 * transmitted if it is not already running. 1559 */ 1560 restart_timer = 1; 1561 } 1562 1563 list_add_tail(lchunk, &tlist); 1564 } 1565 } 1566 1567 if (transport) { 1568 if (bytes_acked) { 1569 struct sctp_association *asoc = transport->asoc; 1570 1571 /* We may have counted DATA that was migrated 1572 * to this transport due to DEL-IP operation. 1573 * Subtract those bytes, since the were never 1574 * send on this transport and shouldn't be 1575 * credited to this transport. 1576 */ 1577 bytes_acked -= migrate_bytes; 1578 1579 /* 8.2. When an outstanding TSN is acknowledged, 1580 * the endpoint shall clear the error counter of 1581 * the destination transport address to which the 1582 * DATA chunk was last sent. 1583 * The association's overall error counter is 1584 * also cleared. 1585 */ 1586 transport->error_count = 0; 1587 transport->asoc->overall_error_count = 0; 1588 forward_progress = true; 1589 1590 /* 1591 * While in SHUTDOWN PENDING, we may have started 1592 * the T5 shutdown guard timer after reaching the 1593 * retransmission limit. Stop that timer as soon 1594 * as the receiver acknowledged any data. 1595 */ 1596 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING && 1597 del_timer(&asoc->timers 1598 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD])) 1599 sctp_association_put(asoc); 1600 1601 /* Mark the destination transport address as 1602 * active if it is not so marked. 1603 */ 1604 if ((transport->state == SCTP_INACTIVE || 1605 transport->state == SCTP_UNCONFIRMED) && 1606 sctp_cmp_addr_exact(&transport->ipaddr, saddr)) { 1607 sctp_assoc_control_transport( 1608 transport->asoc, 1609 transport, 1610 SCTP_TRANSPORT_UP, 1611 SCTP_RECEIVED_SACK); 1612 } 1613 1614 sctp_transport_raise_cwnd(transport, sack_ctsn, 1615 bytes_acked); 1616 1617 transport->flight_size -= bytes_acked; 1618 if (transport->flight_size == 0) 1619 transport->partial_bytes_acked = 0; 1620 q->outstanding_bytes -= bytes_acked + migrate_bytes; 1621 } else { 1622 /* RFC 2960 6.1, sctpimpguide-06 2.15.2 1623 * When a sender is doing zero window probing, it 1624 * should not timeout the association if it continues 1625 * to receive new packets from the receiver. The 1626 * reason is that the receiver MAY keep its window 1627 * closed for an indefinite time. 1628 * A sender is doing zero window probing when the 1629 * receiver's advertised window is zero, and there is 1630 * only one data chunk in flight to the receiver. 1631 * 1632 * Allow the association to timeout while in SHUTDOWN 1633 * PENDING or SHUTDOWN RECEIVED in case the receiver 1634 * stays in zero window mode forever. 1635 */ 1636 if (!q->asoc->peer.rwnd && 1637 !list_empty(&tlist) && 1638 (sack_ctsn+2 == q->asoc->next_tsn) && 1639 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) { 1640 pr_debug("%s: sack received for zero window " 1641 "probe:%u\n", __func__, sack_ctsn); 1642 1643 q->asoc->overall_error_count = 0; 1644 transport->error_count = 0; 1645 } 1646 } 1647 1648 /* RFC 2960 6.3.2 Retransmission Timer Rules 1649 * 1650 * R2) Whenever all outstanding data sent to an address have 1651 * been acknowledged, turn off the T3-rtx timer of that 1652 * address. 1653 */ 1654 if (!transport->flight_size) { 1655 if (del_timer(&transport->T3_rtx_timer)) 1656 sctp_transport_put(transport); 1657 } else if (restart_timer) { 1658 if (!mod_timer(&transport->T3_rtx_timer, 1659 jiffies + transport->rto)) 1660 sctp_transport_hold(transport); 1661 } 1662 1663 if (forward_progress) { 1664 if (transport->dst) 1665 sctp_transport_dst_confirm(transport); 1666 } 1667 } 1668 1669 list_splice(&tlist, transmitted_queue); 1670 } 1671 1672 /* Mark chunks as missing and consequently may get retransmitted. */ 1673 static void sctp_mark_missing(struct sctp_outq *q, 1674 struct list_head *transmitted_queue, 1675 struct sctp_transport *transport, 1676 __u32 highest_new_tsn_in_sack, 1677 int count_of_newacks) 1678 { 1679 struct sctp_chunk *chunk; 1680 __u32 tsn; 1681 char do_fast_retransmit = 0; 1682 struct sctp_association *asoc = q->asoc; 1683 struct sctp_transport *primary = asoc->peer.primary_path; 1684 1685 list_for_each_entry(chunk, transmitted_queue, transmitted_list) { 1686 1687 tsn = ntohl(chunk->subh.data_hdr->tsn); 1688 1689 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all 1690 * 'Unacknowledged TSN's', if the TSN number of an 1691 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack' 1692 * value, increment the 'TSN.Missing.Report' count on that 1693 * chunk if it has NOT been fast retransmitted or marked for 1694 * fast retransmit already. 1695 */ 1696 if (chunk->fast_retransmit == SCTP_CAN_FRTX && 1697 !chunk->tsn_gap_acked && 1698 TSN_lt(tsn, highest_new_tsn_in_sack)) { 1699 1700 /* SFR-CACC may require us to skip marking 1701 * this chunk as missing. 1702 */ 1703 if (!transport || !sctp_cacc_skip(primary, 1704 chunk->transport, 1705 count_of_newacks, tsn)) { 1706 chunk->tsn_missing_report++; 1707 1708 pr_debug("%s: tsn:0x%x missing counter:%d\n", 1709 __func__, tsn, chunk->tsn_missing_report); 1710 } 1711 } 1712 /* 1713 * M4) If any DATA chunk is found to have a 1714 * 'TSN.Missing.Report' 1715 * value larger than or equal to 3, mark that chunk for 1716 * retransmission and start the fast retransmit procedure. 1717 */ 1718 1719 if (chunk->tsn_missing_report >= 3) { 1720 chunk->fast_retransmit = SCTP_NEED_FRTX; 1721 do_fast_retransmit = 1; 1722 } 1723 } 1724 1725 if (transport) { 1726 if (do_fast_retransmit) 1727 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX); 1728 1729 pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, " 1730 "flight_size:%d, pba:%d\n", __func__, transport, 1731 transport->cwnd, transport->ssthresh, 1732 transport->flight_size, transport->partial_bytes_acked); 1733 } 1734 } 1735 1736 /* Is the given TSN acked by this packet? */ 1737 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn) 1738 { 1739 __u32 ctsn = ntohl(sack->cum_tsn_ack); 1740 union sctp_sack_variable *frags; 1741 __u16 tsn_offset, blocks; 1742 int i; 1743 1744 if (TSN_lte(tsn, ctsn)) 1745 goto pass; 1746 1747 /* 3.3.4 Selective Acknowledgement (SACK) (3): 1748 * 1749 * Gap Ack Blocks: 1750 * These fields contain the Gap Ack Blocks. They are repeated 1751 * for each Gap Ack Block up to the number of Gap Ack Blocks 1752 * defined in the Number of Gap Ack Blocks field. All DATA 1753 * chunks with TSNs greater than or equal to (Cumulative TSN 1754 * Ack + Gap Ack Block Start) and less than or equal to 1755 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack 1756 * Block are assumed to have been received correctly. 1757 */ 1758 1759 frags = sack->variable; 1760 blocks = ntohs(sack->num_gap_ack_blocks); 1761 tsn_offset = tsn - ctsn; 1762 for (i = 0; i < blocks; ++i) { 1763 if (tsn_offset >= ntohs(frags[i].gab.start) && 1764 tsn_offset <= ntohs(frags[i].gab.end)) 1765 goto pass; 1766 } 1767 1768 return 0; 1769 pass: 1770 return 1; 1771 } 1772 1773 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist, 1774 int nskips, __be16 stream) 1775 { 1776 int i; 1777 1778 for (i = 0; i < nskips; i++) { 1779 if (skiplist[i].stream == stream) 1780 return i; 1781 } 1782 return i; 1783 } 1784 1785 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */ 1786 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn) 1787 { 1788 struct sctp_association *asoc = q->asoc; 1789 struct sctp_chunk *ftsn_chunk = NULL; 1790 struct sctp_fwdtsn_skip ftsn_skip_arr[10]; 1791 int nskips = 0; 1792 int skip_pos = 0; 1793 __u32 tsn; 1794 struct sctp_chunk *chunk; 1795 struct list_head *lchunk, *temp; 1796 1797 if (!asoc->peer.prsctp_capable) 1798 return; 1799 1800 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the 1801 * received SACK. 1802 * 1803 * If (Advanced.Peer.Ack.Point < SackCumAck), then update 1804 * Advanced.Peer.Ack.Point to be equal to SackCumAck. 1805 */ 1806 if (TSN_lt(asoc->adv_peer_ack_point, ctsn)) 1807 asoc->adv_peer_ack_point = ctsn; 1808 1809 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point" 1810 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as 1811 * the chunk next in the out-queue space is marked as "abandoned" as 1812 * shown in the following example: 1813 * 1814 * Assuming that a SACK arrived with the Cumulative TSN ACK 102 1815 * and the Advanced.Peer.Ack.Point is updated to this value: 1816 * 1817 * out-queue at the end of ==> out-queue after Adv.Ack.Point 1818 * normal SACK processing local advancement 1819 * ... ... 1820 * Adv.Ack.Pt-> 102 acked 102 acked 1821 * 103 abandoned 103 abandoned 1822 * 104 abandoned Adv.Ack.P-> 104 abandoned 1823 * 105 105 1824 * 106 acked 106 acked 1825 * ... ... 1826 * 1827 * In this example, the data sender successfully advanced the 1828 * "Advanced.Peer.Ack.Point" from 102 to 104 locally. 1829 */ 1830 list_for_each_safe(lchunk, temp, &q->abandoned) { 1831 chunk = list_entry(lchunk, struct sctp_chunk, 1832 transmitted_list); 1833 tsn = ntohl(chunk->subh.data_hdr->tsn); 1834 1835 /* Remove any chunks in the abandoned queue that are acked by 1836 * the ctsn. 1837 */ 1838 if (TSN_lte(tsn, ctsn)) { 1839 list_del_init(lchunk); 1840 sctp_chunk_free(chunk); 1841 } else { 1842 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) { 1843 asoc->adv_peer_ack_point = tsn; 1844 if (chunk->chunk_hdr->flags & 1845 SCTP_DATA_UNORDERED) 1846 continue; 1847 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0], 1848 nskips, 1849 chunk->subh.data_hdr->stream); 1850 ftsn_skip_arr[skip_pos].stream = 1851 chunk->subh.data_hdr->stream; 1852 ftsn_skip_arr[skip_pos].ssn = 1853 chunk->subh.data_hdr->ssn; 1854 if (skip_pos == nskips) 1855 nskips++; 1856 if (nskips == 10) 1857 break; 1858 } else 1859 break; 1860 } 1861 } 1862 1863 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point" 1864 * is greater than the Cumulative TSN ACK carried in the received 1865 * SACK, the data sender MUST send the data receiver a FORWARD TSN 1866 * chunk containing the latest value of the 1867 * "Advanced.Peer.Ack.Point". 1868 * 1869 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD 1870 * list each stream and sequence number in the forwarded TSN. This 1871 * information will enable the receiver to easily find any 1872 * stranded TSN's waiting on stream reorder queues. Each stream 1873 * SHOULD only be reported once; this means that if multiple 1874 * abandoned messages occur in the same stream then only the 1875 * highest abandoned stream sequence number is reported. If the 1876 * total size of the FORWARD TSN does NOT fit in a single MTU then 1877 * the sender of the FORWARD TSN SHOULD lower the 1878 * Advanced.Peer.Ack.Point to the last TSN that will fit in a 1879 * single MTU. 1880 */ 1881 if (asoc->adv_peer_ack_point > ctsn) 1882 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point, 1883 nskips, &ftsn_skip_arr[0]); 1884 1885 if (ftsn_chunk) { 1886 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list); 1887 SCTP_INC_STATS(sock_net(asoc->base.sk), SCTP_MIB_OUTCTRLCHUNKS); 1888 } 1889 } 1890