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