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, write to 26 * the Free Software Foundation, 59 Temple Place - Suite 330, 27 * Boston, MA 02111-1307, USA. 28 * 29 * Please send any bug reports or fixes you make to the 30 * email address(es): 31 * lksctp developers <lksctp-developers@lists.sourceforge.net> 32 * 33 * Or submit a bug report through the following website: 34 * http://www.sf.net/projects/lksctp 35 * 36 * Written or modified by: 37 * La Monte H.P. Yarroll <piggy@acm.org> 38 * Karl Knutson <karl@athena.chicago.il.us> 39 * Perry Melange <pmelange@null.cc.uic.edu> 40 * Xingang Guo <xingang.guo@intel.com> 41 * Hui Huang <hui.huang@nokia.com> 42 * Sridhar Samudrala <sri@us.ibm.com> 43 * Jon Grimm <jgrimm@us.ibm.com> 44 * 45 * Any bugs reported given to us we will try to fix... any fixes shared will 46 * be incorporated into the next SCTP release. 47 */ 48 49 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 50 51 #include <linux/types.h> 52 #include <linux/list.h> /* For struct list_head */ 53 #include <linux/socket.h> 54 #include <linux/ip.h> 55 #include <linux/slab.h> 56 #include <net/sock.h> /* For skb_set_owner_w */ 57 58 #include <net/sctp/sctp.h> 59 #include <net/sctp/sm.h> 60 61 /* Declare internal functions here. */ 62 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn); 63 static void sctp_check_transmitted(struct sctp_outq *q, 64 struct list_head *transmitted_queue, 65 struct sctp_transport *transport, 66 union sctp_addr *saddr, 67 struct sctp_sackhdr *sack, 68 __u32 *highest_new_tsn); 69 70 static void sctp_mark_missing(struct sctp_outq *q, 71 struct list_head *transmitted_queue, 72 struct sctp_transport *transport, 73 __u32 highest_new_tsn, 74 int count_of_newacks); 75 76 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn); 77 78 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout); 79 80 /* Add data to the front of the queue. */ 81 static inline void sctp_outq_head_data(struct sctp_outq *q, 82 struct sctp_chunk *ch) 83 { 84 list_add(&ch->list, &q->out_chunk_list); 85 q->out_qlen += ch->skb->len; 86 } 87 88 /* Take data from the front of the queue. */ 89 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q) 90 { 91 struct sctp_chunk *ch = NULL; 92 93 if (!list_empty(&q->out_chunk_list)) { 94 struct list_head *entry = q->out_chunk_list.next; 95 96 ch = list_entry(entry, struct sctp_chunk, list); 97 list_del_init(entry); 98 q->out_qlen -= ch->skb->len; 99 } 100 return ch; 101 } 102 /* Add data chunk to the end of the queue. */ 103 static inline void sctp_outq_tail_data(struct sctp_outq *q, 104 struct sctp_chunk *ch) 105 { 106 list_add_tail(&ch->list, &q->out_chunk_list); 107 q->out_qlen += ch->skb->len; 108 } 109 110 /* 111 * SFR-CACC algorithm: 112 * D) If count_of_newacks is greater than or equal to 2 113 * and t was not sent to the current primary then the 114 * sender MUST NOT increment missing report count for t. 115 */ 116 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary, 117 struct sctp_transport *transport, 118 int count_of_newacks) 119 { 120 if (count_of_newacks >=2 && transport != primary) 121 return 1; 122 return 0; 123 } 124 125 /* 126 * SFR-CACC algorithm: 127 * F) If count_of_newacks is less than 2, let d be the 128 * destination to which t was sent. If cacc_saw_newack 129 * is 0 for destination d, then the sender MUST NOT 130 * increment missing report count for t. 131 */ 132 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport, 133 int count_of_newacks) 134 { 135 if (count_of_newacks < 2 && 136 (transport && !transport->cacc.cacc_saw_newack)) 137 return 1; 138 return 0; 139 } 140 141 /* 142 * SFR-CACC algorithm: 143 * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD 144 * execute steps C, D, F. 145 * 146 * C has been implemented in sctp_outq_sack 147 */ 148 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary, 149 struct sctp_transport *transport, 150 int count_of_newacks) 151 { 152 if (!primary->cacc.cycling_changeover) { 153 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks)) 154 return 1; 155 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks)) 156 return 1; 157 return 0; 158 } 159 return 0; 160 } 161 162 /* 163 * SFR-CACC algorithm: 164 * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less 165 * than next_tsn_at_change of the current primary, then 166 * the sender MUST NOT increment missing report count 167 * for t. 168 */ 169 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn) 170 { 171 if (primary->cacc.cycling_changeover && 172 TSN_lt(tsn, primary->cacc.next_tsn_at_change)) 173 return 1; 174 return 0; 175 } 176 177 /* 178 * SFR-CACC algorithm: 179 * 3) If the missing report count for TSN t is to be 180 * incremented according to [RFC2960] and 181 * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set, 182 * then the sender MUST further execute steps 3.1 and 183 * 3.2 to determine if the missing report count for 184 * TSN t SHOULD NOT be incremented. 185 * 186 * 3.3) If 3.1 and 3.2 do not dictate that the missing 187 * report count for t should not be incremented, then 188 * the sender SHOULD increment missing report count for 189 * t (according to [RFC2960] and [SCTP_STEWART_2002]). 190 */ 191 static inline int sctp_cacc_skip(struct sctp_transport *primary, 192 struct sctp_transport *transport, 193 int count_of_newacks, 194 __u32 tsn) 195 { 196 if (primary->cacc.changeover_active && 197 (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) || 198 sctp_cacc_skip_3_2(primary, tsn))) 199 return 1; 200 return 0; 201 } 202 203 /* Initialize an existing sctp_outq. This does the boring stuff. 204 * You still need to define handlers if you really want to DO 205 * something with this structure... 206 */ 207 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q) 208 { 209 q->asoc = asoc; 210 INIT_LIST_HEAD(&q->out_chunk_list); 211 INIT_LIST_HEAD(&q->control_chunk_list); 212 INIT_LIST_HEAD(&q->retransmit); 213 INIT_LIST_HEAD(&q->sacked); 214 INIT_LIST_HEAD(&q->abandoned); 215 216 q->fast_rtx = 0; 217 q->outstanding_bytes = 0; 218 q->empty = 1; 219 q->cork = 0; 220 221 q->malloced = 0; 222 q->out_qlen = 0; 223 } 224 225 /* Free the outqueue structure and any related pending chunks. 226 */ 227 void sctp_outq_teardown(struct sctp_outq *q) 228 { 229 struct sctp_transport *transport; 230 struct list_head *lchunk, *temp; 231 struct sctp_chunk *chunk, *tmp; 232 233 /* Throw away unacknowledged chunks. */ 234 list_for_each_entry(transport, &q->asoc->peer.transport_addr_list, 235 transports) { 236 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) { 237 chunk = list_entry(lchunk, struct sctp_chunk, 238 transmitted_list); 239 /* Mark as part of a failed message. */ 240 sctp_chunk_fail(chunk, q->error); 241 sctp_chunk_free(chunk); 242 } 243 } 244 245 /* Throw away chunks that have been gap ACKed. */ 246 list_for_each_safe(lchunk, temp, &q->sacked) { 247 list_del_init(lchunk); 248 chunk = list_entry(lchunk, struct sctp_chunk, 249 transmitted_list); 250 sctp_chunk_fail(chunk, q->error); 251 sctp_chunk_free(chunk); 252 } 253 254 /* Throw away any chunks in the retransmit queue. */ 255 list_for_each_safe(lchunk, temp, &q->retransmit) { 256 list_del_init(lchunk); 257 chunk = list_entry(lchunk, struct sctp_chunk, 258 transmitted_list); 259 sctp_chunk_fail(chunk, q->error); 260 sctp_chunk_free(chunk); 261 } 262 263 /* Throw away any chunks that are in the abandoned queue. */ 264 list_for_each_safe(lchunk, temp, &q->abandoned) { 265 list_del_init(lchunk); 266 chunk = list_entry(lchunk, struct sctp_chunk, 267 transmitted_list); 268 sctp_chunk_fail(chunk, q->error); 269 sctp_chunk_free(chunk); 270 } 271 272 /* Throw away any leftover data chunks. */ 273 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) { 274 275 /* Mark as send failure. */ 276 sctp_chunk_fail(chunk, q->error); 277 sctp_chunk_free(chunk); 278 } 279 280 q->error = 0; 281 282 /* Throw away any leftover control chunks. */ 283 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) { 284 list_del_init(&chunk->list); 285 sctp_chunk_free(chunk); 286 } 287 } 288 289 /* Free the outqueue structure and any related pending chunks. */ 290 void sctp_outq_free(struct sctp_outq *q) 291 { 292 /* Throw away leftover chunks. */ 293 sctp_outq_teardown(q); 294 295 /* If we were kmalloc()'d, free the memory. */ 296 if (q->malloced) 297 kfree(q); 298 } 299 300 /* Put a new chunk in an sctp_outq. */ 301 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk) 302 { 303 struct net *net = sock_net(q->asoc->base.sk); 304 int error = 0; 305 306 SCTP_DEBUG_PRINTK("sctp_outq_tail(%p, %p[%s])\n", 307 q, chunk, chunk && chunk->chunk_hdr ? 308 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) 309 : "Illegal Chunk"); 310 311 /* If it is data, queue it up, otherwise, send it 312 * immediately. 313 */ 314 if (sctp_chunk_is_data(chunk)) { 315 /* Is it OK to queue data chunks? */ 316 /* From 9. Termination of Association 317 * 318 * When either endpoint performs a shutdown, the 319 * association on each peer will stop accepting new 320 * data from its user and only deliver data in queue 321 * at the time of sending or receiving the SHUTDOWN 322 * chunk. 323 */ 324 switch (q->asoc->state) { 325 case SCTP_STATE_CLOSED: 326 case SCTP_STATE_SHUTDOWN_PENDING: 327 case SCTP_STATE_SHUTDOWN_SENT: 328 case SCTP_STATE_SHUTDOWN_RECEIVED: 329 case SCTP_STATE_SHUTDOWN_ACK_SENT: 330 /* Cannot send after transport endpoint shutdown */ 331 error = -ESHUTDOWN; 332 break; 333 334 default: 335 SCTP_DEBUG_PRINTK("outqueueing (%p, %p[%s])\n", 336 q, chunk, chunk && chunk->chunk_hdr ? 337 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) 338 : "Illegal Chunk"); 339 340 sctp_outq_tail_data(q, chunk); 341 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED) 342 SCTP_INC_STATS(net, SCTP_MIB_OUTUNORDERCHUNKS); 343 else 344 SCTP_INC_STATS(net, SCTP_MIB_OUTORDERCHUNKS); 345 q->empty = 0; 346 break; 347 } 348 } else { 349 list_add_tail(&chunk->list, &q->control_chunk_list); 350 SCTP_INC_STATS(net, SCTP_MIB_OUTCTRLCHUNKS); 351 } 352 353 if (error < 0) 354 return error; 355 356 if (!q->cork) 357 error = sctp_outq_flush(q, 0); 358 359 return error; 360 } 361 362 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list 363 * and the abandoned list are in ascending order. 364 */ 365 static void sctp_insert_list(struct list_head *head, struct list_head *new) 366 { 367 struct list_head *pos; 368 struct sctp_chunk *nchunk, *lchunk; 369 __u32 ntsn, ltsn; 370 int done = 0; 371 372 nchunk = list_entry(new, struct sctp_chunk, transmitted_list); 373 ntsn = ntohl(nchunk->subh.data_hdr->tsn); 374 375 list_for_each(pos, head) { 376 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list); 377 ltsn = ntohl(lchunk->subh.data_hdr->tsn); 378 if (TSN_lt(ntsn, ltsn)) { 379 list_add(new, pos->prev); 380 done = 1; 381 break; 382 } 383 } 384 if (!done) 385 list_add_tail(new, head); 386 } 387 388 /* Mark all the eligible packets on a transport for retransmission. */ 389 void sctp_retransmit_mark(struct sctp_outq *q, 390 struct sctp_transport *transport, 391 __u8 reason) 392 { 393 struct list_head *lchunk, *ltemp; 394 struct sctp_chunk *chunk; 395 396 /* Walk through the specified transmitted queue. */ 397 list_for_each_safe(lchunk, ltemp, &transport->transmitted) { 398 chunk = list_entry(lchunk, struct sctp_chunk, 399 transmitted_list); 400 401 /* If the chunk is abandoned, move it to abandoned list. */ 402 if (sctp_chunk_abandoned(chunk)) { 403 list_del_init(lchunk); 404 sctp_insert_list(&q->abandoned, lchunk); 405 406 /* If this chunk has not been previousely acked, 407 * stop considering it 'outstanding'. Our peer 408 * will most likely never see it since it will 409 * not be retransmitted 410 */ 411 if (!chunk->tsn_gap_acked) { 412 if (chunk->transport) 413 chunk->transport->flight_size -= 414 sctp_data_size(chunk); 415 q->outstanding_bytes -= sctp_data_size(chunk); 416 q->asoc->peer.rwnd += sctp_data_size(chunk); 417 } 418 continue; 419 } 420 421 /* If we are doing retransmission due to a timeout or pmtu 422 * discovery, only the chunks that are not yet acked should 423 * be added to the retransmit queue. 424 */ 425 if ((reason == SCTP_RTXR_FAST_RTX && 426 (chunk->fast_retransmit == SCTP_NEED_FRTX)) || 427 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) { 428 /* RFC 2960 6.2.1 Processing a Received SACK 429 * 430 * C) Any time a DATA chunk is marked for 431 * retransmission (via either T3-rtx timer expiration 432 * (Section 6.3.3) or via fast retransmit 433 * (Section 7.2.4)), add the data size of those 434 * chunks to the rwnd. 435 */ 436 q->asoc->peer.rwnd += sctp_data_size(chunk); 437 q->outstanding_bytes -= sctp_data_size(chunk); 438 if (chunk->transport) 439 transport->flight_size -= sctp_data_size(chunk); 440 441 /* sctpimpguide-05 Section 2.8.2 442 * M5) If a T3-rtx timer expires, the 443 * 'TSN.Missing.Report' of all affected TSNs is set 444 * to 0. 445 */ 446 chunk->tsn_missing_report = 0; 447 448 /* If a chunk that is being used for RTT measurement 449 * has to be retransmitted, we cannot use this chunk 450 * anymore for RTT measurements. Reset rto_pending so 451 * that a new RTT measurement is started when a new 452 * data chunk is sent. 453 */ 454 if (chunk->rtt_in_progress) { 455 chunk->rtt_in_progress = 0; 456 transport->rto_pending = 0; 457 } 458 459 /* Move the chunk to the retransmit queue. The chunks 460 * on the retransmit queue are always kept in order. 461 */ 462 list_del_init(lchunk); 463 sctp_insert_list(&q->retransmit, lchunk); 464 } 465 } 466 467 SCTP_DEBUG_PRINTK("%s: transport: %p, reason: %d, " 468 "cwnd: %d, ssthresh: %d, flight_size: %d, " 469 "pba: %d\n", __func__, 470 transport, reason, 471 transport->cwnd, transport->ssthresh, 472 transport->flight_size, 473 transport->partial_bytes_acked); 474 475 } 476 477 /* Mark all the eligible packets on a transport for retransmission and force 478 * one packet out. 479 */ 480 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport, 481 sctp_retransmit_reason_t reason) 482 { 483 struct net *net = sock_net(q->asoc->base.sk); 484 int error = 0; 485 486 switch(reason) { 487 case SCTP_RTXR_T3_RTX: 488 SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS); 489 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX); 490 /* Update the retran path if the T3-rtx timer has expired for 491 * the current retran path. 492 */ 493 if (transport == transport->asoc->peer.retran_path) 494 sctp_assoc_update_retran_path(transport->asoc); 495 transport->asoc->rtx_data_chunks += 496 transport->asoc->unack_data; 497 break; 498 case SCTP_RTXR_FAST_RTX: 499 SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS); 500 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX); 501 q->fast_rtx = 1; 502 break; 503 case SCTP_RTXR_PMTUD: 504 SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS); 505 break; 506 case SCTP_RTXR_T1_RTX: 507 SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS); 508 transport->asoc->init_retries++; 509 break; 510 default: 511 BUG(); 512 } 513 514 sctp_retransmit_mark(q, transport, reason); 515 516 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination, 517 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by 518 * following the procedures outlined in C1 - C5. 519 */ 520 if (reason == SCTP_RTXR_T3_RTX) 521 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point); 522 523 /* Flush the queues only on timeout, since fast_rtx is only 524 * triggered during sack processing and the queue 525 * will be flushed at the end. 526 */ 527 if (reason != SCTP_RTXR_FAST_RTX) 528 error = sctp_outq_flush(q, /* rtx_timeout */ 1); 529 530 if (error) 531 q->asoc->base.sk->sk_err = -error; 532 } 533 534 /* 535 * Transmit DATA chunks on the retransmit queue. Upon return from 536 * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which 537 * need to be transmitted by the caller. 538 * We assume that pkt->transport has already been set. 539 * 540 * The return value is a normal kernel error return value. 541 */ 542 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt, 543 int rtx_timeout, int *start_timer) 544 { 545 struct list_head *lqueue; 546 struct sctp_transport *transport = pkt->transport; 547 sctp_xmit_t status; 548 struct sctp_chunk *chunk, *chunk1; 549 int fast_rtx; 550 int error = 0; 551 int timer = 0; 552 int done = 0; 553 554 lqueue = &q->retransmit; 555 fast_rtx = q->fast_rtx; 556 557 /* This loop handles time-out retransmissions, fast retransmissions, 558 * and retransmissions due to opening of whindow. 559 * 560 * RFC 2960 6.3.3 Handle T3-rtx Expiration 561 * 562 * E3) Determine how many of the earliest (i.e., lowest TSN) 563 * outstanding DATA chunks for the address for which the 564 * T3-rtx has expired will fit into a single packet, subject 565 * to the MTU constraint for the path corresponding to the 566 * destination transport address to which the retransmission 567 * is being sent (this may be different from the address for 568 * which the timer expires [see Section 6.4]). Call this value 569 * K. Bundle and retransmit those K DATA chunks in a single 570 * packet to the destination endpoint. 571 * 572 * [Just to be painfully clear, if we are retransmitting 573 * because a timeout just happened, we should send only ONE 574 * packet of retransmitted data.] 575 * 576 * For fast retransmissions we also send only ONE packet. However, 577 * if we are just flushing the queue due to open window, we'll 578 * try to send as much as possible. 579 */ 580 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) { 581 /* If the chunk is abandoned, move it to abandoned list. */ 582 if (sctp_chunk_abandoned(chunk)) { 583 list_del_init(&chunk->transmitted_list); 584 sctp_insert_list(&q->abandoned, 585 &chunk->transmitted_list); 586 continue; 587 } 588 589 /* Make sure that Gap Acked TSNs are not retransmitted. A 590 * simple approach is just to move such TSNs out of the 591 * way and into a 'transmitted' queue and skip to the 592 * next chunk. 593 */ 594 if (chunk->tsn_gap_acked) { 595 list_move_tail(&chunk->transmitted_list, 596 &transport->transmitted); 597 continue; 598 } 599 600 /* If we are doing fast retransmit, ignore non-fast_rtransmit 601 * chunks 602 */ 603 if (fast_rtx && !chunk->fast_retransmit) 604 continue; 605 606 redo: 607 /* Attempt to append this chunk to the packet. */ 608 status = sctp_packet_append_chunk(pkt, chunk); 609 610 switch (status) { 611 case SCTP_XMIT_PMTU_FULL: 612 if (!pkt->has_data && !pkt->has_cookie_echo) { 613 /* If this packet did not contain DATA then 614 * retransmission did not happen, so do it 615 * again. We'll ignore the error here since 616 * control chunks are already freed so there 617 * is nothing we can do. 618 */ 619 sctp_packet_transmit(pkt); 620 goto redo; 621 } 622 623 /* Send this packet. */ 624 error = sctp_packet_transmit(pkt); 625 626 /* If we are retransmitting, we should only 627 * send a single packet. 628 * Otherwise, try appending this chunk again. 629 */ 630 if (rtx_timeout || fast_rtx) 631 done = 1; 632 else 633 goto redo; 634 635 /* Bundle next chunk in the next round. */ 636 break; 637 638 case SCTP_XMIT_RWND_FULL: 639 /* Send this packet. */ 640 error = sctp_packet_transmit(pkt); 641 642 /* Stop sending DATA as there is no more room 643 * at the receiver. 644 */ 645 done = 1; 646 break; 647 648 case SCTP_XMIT_NAGLE_DELAY: 649 /* Send this packet. */ 650 error = sctp_packet_transmit(pkt); 651 652 /* Stop sending DATA because of nagle delay. */ 653 done = 1; 654 break; 655 656 default: 657 /* The append was successful, so add this chunk to 658 * the transmitted list. 659 */ 660 list_move_tail(&chunk->transmitted_list, 661 &transport->transmitted); 662 663 /* Mark the chunk as ineligible for fast retransmit 664 * after it is retransmitted. 665 */ 666 if (chunk->fast_retransmit == SCTP_NEED_FRTX) 667 chunk->fast_retransmit = SCTP_DONT_FRTX; 668 669 q->empty = 0; 670 q->asoc->stats.rtxchunks++; 671 break; 672 } 673 674 /* Set the timer if there were no errors */ 675 if (!error && !timer) 676 timer = 1; 677 678 if (done) 679 break; 680 } 681 682 /* If we are here due to a retransmit timeout or a fast 683 * retransmit and if there are any chunks left in the retransmit 684 * queue that could not fit in the PMTU sized packet, they need 685 * to be marked as ineligible for a subsequent fast retransmit. 686 */ 687 if (rtx_timeout || fast_rtx) { 688 list_for_each_entry(chunk1, lqueue, transmitted_list) { 689 if (chunk1->fast_retransmit == SCTP_NEED_FRTX) 690 chunk1->fast_retransmit = SCTP_DONT_FRTX; 691 } 692 } 693 694 *start_timer = timer; 695 696 /* Clear fast retransmit hint */ 697 if (fast_rtx) 698 q->fast_rtx = 0; 699 700 return error; 701 } 702 703 /* Cork the outqueue so queued chunks are really queued. */ 704 int sctp_outq_uncork(struct sctp_outq *q) 705 { 706 int error = 0; 707 if (q->cork) 708 q->cork = 0; 709 error = sctp_outq_flush(q, 0); 710 return error; 711 } 712 713 714 /* 715 * Try to flush an outqueue. 716 * 717 * Description: Send everything in q which we legally can, subject to 718 * congestion limitations. 719 * * Note: This function can be called from multiple contexts so appropriate 720 * locking concerns must be made. Today we use the sock lock to protect 721 * this function. 722 */ 723 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout) 724 { 725 struct sctp_packet *packet; 726 struct sctp_packet singleton; 727 struct sctp_association *asoc = q->asoc; 728 __u16 sport = asoc->base.bind_addr.port; 729 __u16 dport = asoc->peer.port; 730 __u32 vtag = asoc->peer.i.init_tag; 731 struct sctp_transport *transport = NULL; 732 struct sctp_transport *new_transport; 733 struct sctp_chunk *chunk, *tmp; 734 sctp_xmit_t status; 735 int error = 0; 736 int start_timer = 0; 737 int one_packet = 0; 738 739 /* These transports have chunks to send. */ 740 struct list_head transport_list; 741 struct list_head *ltransport; 742 743 INIT_LIST_HEAD(&transport_list); 744 packet = NULL; 745 746 /* 747 * 6.10 Bundling 748 * ... 749 * When bundling control chunks with DATA chunks, an 750 * endpoint MUST place control chunks first in the outbound 751 * SCTP packet. The transmitter MUST transmit DATA chunks 752 * within a SCTP packet in increasing order of TSN. 753 * ... 754 */ 755 756 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) { 757 /* RFC 5061, 5.3 758 * F1) This means that until such time as the ASCONF 759 * containing the add is acknowledged, the sender MUST 760 * NOT use the new IP address as a source for ANY SCTP 761 * packet except on carrying an ASCONF Chunk. 762 */ 763 if (asoc->src_out_of_asoc_ok && 764 chunk->chunk_hdr->type != SCTP_CID_ASCONF) 765 continue; 766 767 list_del_init(&chunk->list); 768 769 /* Pick the right transport to use. */ 770 new_transport = chunk->transport; 771 772 if (!new_transport) { 773 /* 774 * If we have a prior transport pointer, see if 775 * the destination address of the chunk 776 * matches the destination address of the 777 * current transport. If not a match, then 778 * try to look up the transport with a given 779 * destination address. We do this because 780 * after processing ASCONFs, we may have new 781 * transports created. 782 */ 783 if (transport && 784 sctp_cmp_addr_exact(&chunk->dest, 785 &transport->ipaddr)) 786 new_transport = transport; 787 else 788 new_transport = sctp_assoc_lookup_paddr(asoc, 789 &chunk->dest); 790 791 /* if we still don't have a new transport, then 792 * use the current active path. 793 */ 794 if (!new_transport) 795 new_transport = asoc->peer.active_path; 796 } else if ((new_transport->state == SCTP_INACTIVE) || 797 (new_transport->state == SCTP_UNCONFIRMED) || 798 (new_transport->state == SCTP_PF)) { 799 /* If the chunk is Heartbeat or Heartbeat Ack, 800 * send it to chunk->transport, even if it's 801 * inactive. 802 * 803 * 3.3.6 Heartbeat Acknowledgement: 804 * ... 805 * A HEARTBEAT ACK is always sent to the source IP 806 * address of the IP datagram containing the 807 * HEARTBEAT chunk to which this ack is responding. 808 * ... 809 * 810 * ASCONF_ACKs also must be sent to the source. 811 */ 812 if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT && 813 chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK && 814 chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK) 815 new_transport = asoc->peer.active_path; 816 } 817 818 /* Are we switching transports? 819 * Take care of transport locks. 820 */ 821 if (new_transport != transport) { 822 transport = new_transport; 823 if (list_empty(&transport->send_ready)) { 824 list_add_tail(&transport->send_ready, 825 &transport_list); 826 } 827 packet = &transport->packet; 828 sctp_packet_config(packet, vtag, 829 asoc->peer.ecn_capable); 830 } 831 832 switch (chunk->chunk_hdr->type) { 833 /* 834 * 6.10 Bundling 835 * ... 836 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN 837 * COMPLETE with any other chunks. [Send them immediately.] 838 */ 839 case SCTP_CID_INIT: 840 case SCTP_CID_INIT_ACK: 841 case SCTP_CID_SHUTDOWN_COMPLETE: 842 sctp_packet_init(&singleton, transport, sport, dport); 843 sctp_packet_config(&singleton, vtag, 0); 844 sctp_packet_append_chunk(&singleton, chunk); 845 error = sctp_packet_transmit(&singleton); 846 if (error < 0) 847 return error; 848 break; 849 850 case SCTP_CID_ABORT: 851 if (sctp_test_T_bit(chunk)) { 852 packet->vtag = asoc->c.my_vtag; 853 } 854 /* The following chunks are "response" chunks, i.e. 855 * they are generated in response to something we 856 * received. If we are sending these, then we can 857 * send only 1 packet containing these chunks. 858 */ 859 case SCTP_CID_HEARTBEAT_ACK: 860 case SCTP_CID_SHUTDOWN_ACK: 861 case SCTP_CID_COOKIE_ACK: 862 case SCTP_CID_COOKIE_ECHO: 863 case SCTP_CID_ERROR: 864 case SCTP_CID_ECN_CWR: 865 case SCTP_CID_ASCONF_ACK: 866 one_packet = 1; 867 /* Fall through */ 868 869 case SCTP_CID_SACK: 870 case SCTP_CID_HEARTBEAT: 871 case SCTP_CID_SHUTDOWN: 872 case SCTP_CID_ECN_ECNE: 873 case SCTP_CID_ASCONF: 874 case SCTP_CID_FWD_TSN: 875 status = sctp_packet_transmit_chunk(packet, chunk, 876 one_packet); 877 if (status != SCTP_XMIT_OK) { 878 /* put the chunk back */ 879 list_add(&chunk->list, &q->control_chunk_list); 880 } else { 881 asoc->stats.octrlchunks++; 882 /* PR-SCTP C5) If a FORWARD TSN is sent, the 883 * sender MUST assure that at least one T3-rtx 884 * timer is running. 885 */ 886 if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN) 887 sctp_transport_reset_timers(transport); 888 } 889 break; 890 891 default: 892 /* We built a chunk with an illegal type! */ 893 BUG(); 894 } 895 } 896 897 if (q->asoc->src_out_of_asoc_ok) 898 goto sctp_flush_out; 899 900 /* Is it OK to send data chunks? */ 901 switch (asoc->state) { 902 case SCTP_STATE_COOKIE_ECHOED: 903 /* Only allow bundling when this packet has a COOKIE-ECHO 904 * chunk. 905 */ 906 if (!packet || !packet->has_cookie_echo) 907 break; 908 909 /* fallthru */ 910 case SCTP_STATE_ESTABLISHED: 911 case SCTP_STATE_SHUTDOWN_PENDING: 912 case SCTP_STATE_SHUTDOWN_RECEIVED: 913 /* 914 * RFC 2960 6.1 Transmission of DATA Chunks 915 * 916 * C) When the time comes for the sender to transmit, 917 * before sending new DATA chunks, the sender MUST 918 * first transmit any outstanding DATA chunks which 919 * are marked for retransmission (limited by the 920 * current cwnd). 921 */ 922 if (!list_empty(&q->retransmit)) { 923 if (asoc->peer.retran_path->state == SCTP_UNCONFIRMED) 924 goto sctp_flush_out; 925 if (transport == asoc->peer.retran_path) 926 goto retran; 927 928 /* Switch transports & prepare the packet. */ 929 930 transport = asoc->peer.retran_path; 931 932 if (list_empty(&transport->send_ready)) { 933 list_add_tail(&transport->send_ready, 934 &transport_list); 935 } 936 937 packet = &transport->packet; 938 sctp_packet_config(packet, vtag, 939 asoc->peer.ecn_capable); 940 retran: 941 error = sctp_outq_flush_rtx(q, packet, 942 rtx_timeout, &start_timer); 943 944 if (start_timer) 945 sctp_transport_reset_timers(transport); 946 947 /* This can happen on COOKIE-ECHO resend. Only 948 * one chunk can get bundled with a COOKIE-ECHO. 949 */ 950 if (packet->has_cookie_echo) 951 goto sctp_flush_out; 952 953 /* Don't send new data if there is still data 954 * waiting to retransmit. 955 */ 956 if (!list_empty(&q->retransmit)) 957 goto sctp_flush_out; 958 } 959 960 /* Apply Max.Burst limitation to the current transport in 961 * case it will be used for new data. We are going to 962 * rest it before we return, but we want to apply the limit 963 * to the currently queued data. 964 */ 965 if (transport) 966 sctp_transport_burst_limited(transport); 967 968 /* Finally, transmit new packets. */ 969 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) { 970 /* RFC 2960 6.5 Every DATA chunk MUST carry a valid 971 * stream identifier. 972 */ 973 if (chunk->sinfo.sinfo_stream >= 974 asoc->c.sinit_num_ostreams) { 975 976 /* Mark as failed send. */ 977 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM); 978 sctp_chunk_free(chunk); 979 continue; 980 } 981 982 /* Has this chunk expired? */ 983 if (sctp_chunk_abandoned(chunk)) { 984 sctp_chunk_fail(chunk, 0); 985 sctp_chunk_free(chunk); 986 continue; 987 } 988 989 /* If there is a specified transport, use it. 990 * Otherwise, we want to use the active path. 991 */ 992 new_transport = chunk->transport; 993 if (!new_transport || 994 ((new_transport->state == SCTP_INACTIVE) || 995 (new_transport->state == SCTP_UNCONFIRMED) || 996 (new_transport->state == SCTP_PF))) 997 new_transport = asoc->peer.active_path; 998 if (new_transport->state == SCTP_UNCONFIRMED) 999 continue; 1000 1001 /* Change packets if necessary. */ 1002 if (new_transport != transport) { 1003 transport = new_transport; 1004 1005 /* Schedule to have this transport's 1006 * packet flushed. 1007 */ 1008 if (list_empty(&transport->send_ready)) { 1009 list_add_tail(&transport->send_ready, 1010 &transport_list); 1011 } 1012 1013 packet = &transport->packet; 1014 sctp_packet_config(packet, vtag, 1015 asoc->peer.ecn_capable); 1016 /* We've switched transports, so apply the 1017 * Burst limit to the new transport. 1018 */ 1019 sctp_transport_burst_limited(transport); 1020 } 1021 1022 SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ", 1023 q, chunk, 1024 chunk && chunk->chunk_hdr ? 1025 sctp_cname(SCTP_ST_CHUNK( 1026 chunk->chunk_hdr->type)) 1027 : "Illegal Chunk"); 1028 1029 SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head " 1030 "%p skb->users %d.\n", 1031 ntohl(chunk->subh.data_hdr->tsn), 1032 chunk->skb ?chunk->skb->head : NULL, 1033 chunk->skb ? 1034 atomic_read(&chunk->skb->users) : -1); 1035 1036 /* Add the chunk to the packet. */ 1037 status = sctp_packet_transmit_chunk(packet, chunk, 0); 1038 1039 switch (status) { 1040 case SCTP_XMIT_PMTU_FULL: 1041 case SCTP_XMIT_RWND_FULL: 1042 case SCTP_XMIT_NAGLE_DELAY: 1043 /* We could not append this chunk, so put 1044 * the chunk back on the output queue. 1045 */ 1046 SCTP_DEBUG_PRINTK("sctp_outq_flush: could " 1047 "not transmit TSN: 0x%x, status: %d\n", 1048 ntohl(chunk->subh.data_hdr->tsn), 1049 status); 1050 sctp_outq_head_data(q, chunk); 1051 goto sctp_flush_out; 1052 break; 1053 1054 case SCTP_XMIT_OK: 1055 /* The sender is in the SHUTDOWN-PENDING state, 1056 * The sender MAY set the I-bit in the DATA 1057 * chunk header. 1058 */ 1059 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING) 1060 chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM; 1061 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED) 1062 asoc->stats.ouodchunks++; 1063 else 1064 asoc->stats.oodchunks++; 1065 1066 break; 1067 1068 default: 1069 BUG(); 1070 } 1071 1072 /* BUG: We assume that the sctp_packet_transmit() 1073 * call below will succeed all the time and add the 1074 * chunk to the transmitted list and restart the 1075 * timers. 1076 * It is possible that the call can fail under OOM 1077 * conditions. 1078 * 1079 * Is this really a problem? Won't this behave 1080 * like a lost TSN? 1081 */ 1082 list_add_tail(&chunk->transmitted_list, 1083 &transport->transmitted); 1084 1085 sctp_transport_reset_timers(transport); 1086 1087 q->empty = 0; 1088 1089 /* Only let one DATA chunk get bundled with a 1090 * COOKIE-ECHO chunk. 1091 */ 1092 if (packet->has_cookie_echo) 1093 goto sctp_flush_out; 1094 } 1095 break; 1096 1097 default: 1098 /* Do nothing. */ 1099 break; 1100 } 1101 1102 sctp_flush_out: 1103 1104 /* Before returning, examine all the transports touched in 1105 * this call. Right now, we bluntly force clear all the 1106 * transports. Things might change after we implement Nagle. 1107 * But such an examination is still required. 1108 * 1109 * --xguo 1110 */ 1111 while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) { 1112 struct sctp_transport *t = list_entry(ltransport, 1113 struct sctp_transport, 1114 send_ready); 1115 packet = &t->packet; 1116 if (!sctp_packet_empty(packet)) 1117 error = sctp_packet_transmit(packet); 1118 1119 /* Clear the burst limited state, if any */ 1120 sctp_transport_burst_reset(t); 1121 } 1122 1123 return error; 1124 } 1125 1126 /* Update unack_data based on the incoming SACK chunk */ 1127 static void sctp_sack_update_unack_data(struct sctp_association *assoc, 1128 struct sctp_sackhdr *sack) 1129 { 1130 sctp_sack_variable_t *frags; 1131 __u16 unack_data; 1132 int i; 1133 1134 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1; 1135 1136 frags = sack->variable; 1137 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) { 1138 unack_data -= ((ntohs(frags[i].gab.end) - 1139 ntohs(frags[i].gab.start) + 1)); 1140 } 1141 1142 assoc->unack_data = unack_data; 1143 } 1144 1145 /* This is where we REALLY process a SACK. 1146 * 1147 * Process the SACK against the outqueue. Mostly, this just frees 1148 * things off the transmitted queue. 1149 */ 1150 int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk) 1151 { 1152 struct sctp_association *asoc = q->asoc; 1153 struct sctp_sackhdr *sack = chunk->subh.sack_hdr; 1154 struct sctp_transport *transport; 1155 struct sctp_chunk *tchunk = NULL; 1156 struct list_head *lchunk, *transport_list, *temp; 1157 sctp_sack_variable_t *frags = sack->variable; 1158 __u32 sack_ctsn, ctsn, tsn; 1159 __u32 highest_tsn, highest_new_tsn; 1160 __u32 sack_a_rwnd; 1161 unsigned int outstanding; 1162 struct sctp_transport *primary = asoc->peer.primary_path; 1163 int count_of_newacks = 0; 1164 int gap_ack_blocks; 1165 u8 accum_moved = 0; 1166 1167 /* Grab the association's destination address list. */ 1168 transport_list = &asoc->peer.transport_addr_list; 1169 1170 sack_ctsn = ntohl(sack->cum_tsn_ack); 1171 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks); 1172 asoc->stats.gapcnt += gap_ack_blocks; 1173 /* 1174 * SFR-CACC algorithm: 1175 * On receipt of a SACK the sender SHOULD execute the 1176 * following statements. 1177 * 1178 * 1) If the cumulative ack in the SACK passes next tsn_at_change 1179 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be 1180 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for 1181 * all destinations. 1182 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE 1183 * is set the receiver of the SACK MUST take the following actions: 1184 * 1185 * A) Initialize the cacc_saw_newack to 0 for all destination 1186 * addresses. 1187 * 1188 * Only bother if changeover_active is set. Otherwise, this is 1189 * totally suboptimal to do on every SACK. 1190 */ 1191 if (primary->cacc.changeover_active) { 1192 u8 clear_cycling = 0; 1193 1194 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) { 1195 primary->cacc.changeover_active = 0; 1196 clear_cycling = 1; 1197 } 1198 1199 if (clear_cycling || gap_ack_blocks) { 1200 list_for_each_entry(transport, transport_list, 1201 transports) { 1202 if (clear_cycling) 1203 transport->cacc.cycling_changeover = 0; 1204 if (gap_ack_blocks) 1205 transport->cacc.cacc_saw_newack = 0; 1206 } 1207 } 1208 } 1209 1210 /* Get the highest TSN in the sack. */ 1211 highest_tsn = sack_ctsn; 1212 if (gap_ack_blocks) 1213 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end); 1214 1215 if (TSN_lt(asoc->highest_sacked, highest_tsn)) 1216 asoc->highest_sacked = highest_tsn; 1217 1218 highest_new_tsn = sack_ctsn; 1219 1220 /* Run through the retransmit queue. Credit bytes received 1221 * and free those chunks that we can. 1222 */ 1223 sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn); 1224 1225 /* Run through the transmitted queue. 1226 * Credit bytes received and free those chunks which we can. 1227 * 1228 * This is a MASSIVE candidate for optimization. 1229 */ 1230 list_for_each_entry(transport, transport_list, transports) { 1231 sctp_check_transmitted(q, &transport->transmitted, 1232 transport, &chunk->source, sack, 1233 &highest_new_tsn); 1234 /* 1235 * SFR-CACC algorithm: 1236 * C) Let count_of_newacks be the number of 1237 * destinations for which cacc_saw_newack is set. 1238 */ 1239 if (transport->cacc.cacc_saw_newack) 1240 count_of_newacks ++; 1241 } 1242 1243 /* Move the Cumulative TSN Ack Point if appropriate. */ 1244 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) { 1245 asoc->ctsn_ack_point = sack_ctsn; 1246 accum_moved = 1; 1247 } 1248 1249 if (gap_ack_blocks) { 1250 1251 if (asoc->fast_recovery && accum_moved) 1252 highest_new_tsn = highest_tsn; 1253 1254 list_for_each_entry(transport, transport_list, transports) 1255 sctp_mark_missing(q, &transport->transmitted, transport, 1256 highest_new_tsn, count_of_newacks); 1257 } 1258 1259 /* Update unack_data field in the assoc. */ 1260 sctp_sack_update_unack_data(asoc, sack); 1261 1262 ctsn = asoc->ctsn_ack_point; 1263 1264 /* Throw away stuff rotting on the sack queue. */ 1265 list_for_each_safe(lchunk, temp, &q->sacked) { 1266 tchunk = list_entry(lchunk, struct sctp_chunk, 1267 transmitted_list); 1268 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1269 if (TSN_lte(tsn, ctsn)) { 1270 list_del_init(&tchunk->transmitted_list); 1271 sctp_chunk_free(tchunk); 1272 } 1273 } 1274 1275 /* ii) Set rwnd equal to the newly received a_rwnd minus the 1276 * number of bytes still outstanding after processing the 1277 * Cumulative TSN Ack and the Gap Ack Blocks. 1278 */ 1279 1280 sack_a_rwnd = ntohl(sack->a_rwnd); 1281 outstanding = q->outstanding_bytes; 1282 1283 if (outstanding < sack_a_rwnd) 1284 sack_a_rwnd -= outstanding; 1285 else 1286 sack_a_rwnd = 0; 1287 1288 asoc->peer.rwnd = sack_a_rwnd; 1289 1290 sctp_generate_fwdtsn(q, sack_ctsn); 1291 1292 SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n", 1293 __func__, sack_ctsn); 1294 SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, " 1295 "%p is 0x%x. Adv peer ack point: 0x%x\n", 1296 __func__, asoc, ctsn, asoc->adv_peer_ack_point); 1297 1298 /* See if all chunks are acked. 1299 * Make sure the empty queue handler will get run later. 1300 */ 1301 q->empty = (list_empty(&q->out_chunk_list) && 1302 list_empty(&q->retransmit)); 1303 if (!q->empty) 1304 goto finish; 1305 1306 list_for_each_entry(transport, transport_list, transports) { 1307 q->empty = q->empty && list_empty(&transport->transmitted); 1308 if (!q->empty) 1309 goto finish; 1310 } 1311 1312 SCTP_DEBUG_PRINTK("sack queue is empty.\n"); 1313 finish: 1314 return q->empty; 1315 } 1316 1317 /* Is the outqueue empty? */ 1318 int sctp_outq_is_empty(const struct sctp_outq *q) 1319 { 1320 return q->empty; 1321 } 1322 1323 /******************************************************************** 1324 * 2nd Level Abstractions 1325 ********************************************************************/ 1326 1327 /* Go through a transport's transmitted list or the association's retransmit 1328 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked. 1329 * The retransmit list will not have an associated transport. 1330 * 1331 * I added coherent debug information output. --xguo 1332 * 1333 * Instead of printing 'sacked' or 'kept' for each TSN on the 1334 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5. 1335 * KEPT TSN6-TSN7, etc. 1336 */ 1337 static void sctp_check_transmitted(struct sctp_outq *q, 1338 struct list_head *transmitted_queue, 1339 struct sctp_transport *transport, 1340 union sctp_addr *saddr, 1341 struct sctp_sackhdr *sack, 1342 __u32 *highest_new_tsn_in_sack) 1343 { 1344 struct list_head *lchunk; 1345 struct sctp_chunk *tchunk; 1346 struct list_head tlist; 1347 __u32 tsn; 1348 __u32 sack_ctsn; 1349 __u32 rtt; 1350 __u8 restart_timer = 0; 1351 int bytes_acked = 0; 1352 int migrate_bytes = 0; 1353 1354 /* These state variables are for coherent debug output. --xguo */ 1355 1356 #if SCTP_DEBUG 1357 __u32 dbg_ack_tsn = 0; /* An ACKed TSN range starts here... */ 1358 __u32 dbg_last_ack_tsn = 0; /* ...and finishes here. */ 1359 __u32 dbg_kept_tsn = 0; /* An un-ACKed range starts here... */ 1360 __u32 dbg_last_kept_tsn = 0; /* ...and finishes here. */ 1361 1362 /* 0 : The last TSN was ACKed. 1363 * 1 : The last TSN was NOT ACKed (i.e. KEPT). 1364 * -1: We need to initialize. 1365 */ 1366 int dbg_prt_state = -1; 1367 #endif /* SCTP_DEBUG */ 1368 1369 sack_ctsn = ntohl(sack->cum_tsn_ack); 1370 1371 INIT_LIST_HEAD(&tlist); 1372 1373 /* The while loop will skip empty transmitted queues. */ 1374 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) { 1375 tchunk = list_entry(lchunk, struct sctp_chunk, 1376 transmitted_list); 1377 1378 if (sctp_chunk_abandoned(tchunk)) { 1379 /* Move the chunk to abandoned list. */ 1380 sctp_insert_list(&q->abandoned, lchunk); 1381 1382 /* If this chunk has not been acked, stop 1383 * considering it as 'outstanding'. 1384 */ 1385 if (!tchunk->tsn_gap_acked) { 1386 if (tchunk->transport) 1387 tchunk->transport->flight_size -= 1388 sctp_data_size(tchunk); 1389 q->outstanding_bytes -= sctp_data_size(tchunk); 1390 } 1391 continue; 1392 } 1393 1394 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1395 if (sctp_acked(sack, tsn)) { 1396 /* If this queue is the retransmit queue, the 1397 * retransmit timer has already reclaimed 1398 * the outstanding bytes for this chunk, so only 1399 * count bytes associated with a transport. 1400 */ 1401 if (transport) { 1402 /* If this chunk is being used for RTT 1403 * measurement, calculate the RTT and update 1404 * the RTO using this value. 1405 * 1406 * 6.3.1 C5) Karn's algorithm: RTT measurements 1407 * MUST NOT be made using packets that were 1408 * retransmitted (and thus for which it is 1409 * ambiguous whether the reply was for the 1410 * first instance of the packet or a later 1411 * instance). 1412 */ 1413 if (!tchunk->tsn_gap_acked && 1414 tchunk->rtt_in_progress) { 1415 tchunk->rtt_in_progress = 0; 1416 rtt = jiffies - tchunk->sent_at; 1417 sctp_transport_update_rto(transport, 1418 rtt); 1419 } 1420 } 1421 1422 /* If the chunk hasn't been marked as ACKED, 1423 * mark it and account bytes_acked if the 1424 * chunk had a valid transport (it will not 1425 * have a transport if ASCONF had deleted it 1426 * while DATA was outstanding). 1427 */ 1428 if (!tchunk->tsn_gap_acked) { 1429 tchunk->tsn_gap_acked = 1; 1430 *highest_new_tsn_in_sack = tsn; 1431 bytes_acked += sctp_data_size(tchunk); 1432 if (!tchunk->transport) 1433 migrate_bytes += sctp_data_size(tchunk); 1434 } 1435 1436 if (TSN_lte(tsn, sack_ctsn)) { 1437 /* RFC 2960 6.3.2 Retransmission Timer Rules 1438 * 1439 * R3) Whenever a SACK is received 1440 * that acknowledges the DATA chunk 1441 * with the earliest outstanding TSN 1442 * for that address, restart T3-rtx 1443 * timer for that address with its 1444 * current RTO. 1445 */ 1446 restart_timer = 1; 1447 1448 if (!tchunk->tsn_gap_acked) { 1449 /* 1450 * SFR-CACC algorithm: 1451 * 2) If the SACK contains gap acks 1452 * and the flag CHANGEOVER_ACTIVE is 1453 * set the receiver of the SACK MUST 1454 * take the following action: 1455 * 1456 * B) For each TSN t being acked that 1457 * has not been acked in any SACK so 1458 * far, set cacc_saw_newack to 1 for 1459 * the destination that the TSN was 1460 * sent to. 1461 */ 1462 if (transport && 1463 sack->num_gap_ack_blocks && 1464 q->asoc->peer.primary_path->cacc. 1465 changeover_active) 1466 transport->cacc.cacc_saw_newack 1467 = 1; 1468 } 1469 1470 list_add_tail(&tchunk->transmitted_list, 1471 &q->sacked); 1472 } else { 1473 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2 1474 * M2) Each time a SACK arrives reporting 1475 * 'Stray DATA chunk(s)' record the highest TSN 1476 * reported as newly acknowledged, call this 1477 * value 'HighestTSNinSack'. A newly 1478 * acknowledged DATA chunk is one not 1479 * previously acknowledged in a SACK. 1480 * 1481 * When the SCTP sender of data receives a SACK 1482 * chunk that acknowledges, for the first time, 1483 * the receipt of a DATA chunk, all the still 1484 * unacknowledged DATA chunks whose TSN is 1485 * older than that newly acknowledged DATA 1486 * chunk, are qualified as 'Stray DATA chunks'. 1487 */ 1488 list_add_tail(lchunk, &tlist); 1489 } 1490 1491 #if SCTP_DEBUG 1492 switch (dbg_prt_state) { 1493 case 0: /* last TSN was ACKed */ 1494 if (dbg_last_ack_tsn + 1 == tsn) { 1495 /* This TSN belongs to the 1496 * current ACK range. 1497 */ 1498 break; 1499 } 1500 1501 if (dbg_last_ack_tsn != dbg_ack_tsn) { 1502 /* Display the end of the 1503 * current range. 1504 */ 1505 SCTP_DEBUG_PRINTK_CONT("-%08x", 1506 dbg_last_ack_tsn); 1507 } 1508 1509 /* Start a new range. */ 1510 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn); 1511 dbg_ack_tsn = tsn; 1512 break; 1513 1514 case 1: /* The last TSN was NOT ACKed. */ 1515 if (dbg_last_kept_tsn != dbg_kept_tsn) { 1516 /* Display the end of current range. */ 1517 SCTP_DEBUG_PRINTK_CONT("-%08x", 1518 dbg_last_kept_tsn); 1519 } 1520 1521 SCTP_DEBUG_PRINTK_CONT("\n"); 1522 1523 /* FALL THROUGH... */ 1524 default: 1525 /* This is the first-ever TSN we examined. */ 1526 /* Start a new range of ACK-ed TSNs. */ 1527 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn); 1528 dbg_prt_state = 0; 1529 dbg_ack_tsn = tsn; 1530 } 1531 1532 dbg_last_ack_tsn = tsn; 1533 #endif /* SCTP_DEBUG */ 1534 1535 } else { 1536 if (tchunk->tsn_gap_acked) { 1537 SCTP_DEBUG_PRINTK("%s: Receiver reneged on " 1538 "data TSN: 0x%x\n", 1539 __func__, 1540 tsn); 1541 tchunk->tsn_gap_acked = 0; 1542 1543 if (tchunk->transport) 1544 bytes_acked -= sctp_data_size(tchunk); 1545 1546 /* RFC 2960 6.3.2 Retransmission Timer Rules 1547 * 1548 * R4) Whenever a SACK is received missing a 1549 * TSN that was previously acknowledged via a 1550 * Gap Ack Block, start T3-rtx for the 1551 * destination address to which the DATA 1552 * chunk was originally 1553 * transmitted if it is not already running. 1554 */ 1555 restart_timer = 1; 1556 } 1557 1558 list_add_tail(lchunk, &tlist); 1559 1560 #if SCTP_DEBUG 1561 /* See the above comments on ACK-ed TSNs. */ 1562 switch (dbg_prt_state) { 1563 case 1: 1564 if (dbg_last_kept_tsn + 1 == tsn) 1565 break; 1566 1567 if (dbg_last_kept_tsn != dbg_kept_tsn) 1568 SCTP_DEBUG_PRINTK_CONT("-%08x", 1569 dbg_last_kept_tsn); 1570 1571 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn); 1572 dbg_kept_tsn = tsn; 1573 break; 1574 1575 case 0: 1576 if (dbg_last_ack_tsn != dbg_ack_tsn) 1577 SCTP_DEBUG_PRINTK_CONT("-%08x", 1578 dbg_last_ack_tsn); 1579 SCTP_DEBUG_PRINTK_CONT("\n"); 1580 1581 /* FALL THROUGH... */ 1582 default: 1583 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn); 1584 dbg_prt_state = 1; 1585 dbg_kept_tsn = tsn; 1586 } 1587 1588 dbg_last_kept_tsn = tsn; 1589 #endif /* SCTP_DEBUG */ 1590 } 1591 } 1592 1593 #if SCTP_DEBUG 1594 /* Finish off the last range, displaying its ending TSN. */ 1595 switch (dbg_prt_state) { 1596 case 0: 1597 if (dbg_last_ack_tsn != dbg_ack_tsn) { 1598 SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_ack_tsn); 1599 } else { 1600 SCTP_DEBUG_PRINTK_CONT("\n"); 1601 } 1602 break; 1603 1604 case 1: 1605 if (dbg_last_kept_tsn != dbg_kept_tsn) { 1606 SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_kept_tsn); 1607 } else { 1608 SCTP_DEBUG_PRINTK_CONT("\n"); 1609 } 1610 } 1611 #endif /* SCTP_DEBUG */ 1612 if (transport) { 1613 if (bytes_acked) { 1614 struct sctp_association *asoc = transport->asoc; 1615 1616 /* We may have counted DATA that was migrated 1617 * to this transport due to DEL-IP operation. 1618 * Subtract those bytes, since the were never 1619 * send on this transport and shouldn't be 1620 * credited to this transport. 1621 */ 1622 bytes_acked -= migrate_bytes; 1623 1624 /* 8.2. When an outstanding TSN is acknowledged, 1625 * the endpoint shall clear the error counter of 1626 * the destination transport address to which the 1627 * DATA chunk was last sent. 1628 * The association's overall error counter is 1629 * also cleared. 1630 */ 1631 transport->error_count = 0; 1632 transport->asoc->overall_error_count = 0; 1633 1634 /* 1635 * While in SHUTDOWN PENDING, we may have started 1636 * the T5 shutdown guard timer after reaching the 1637 * retransmission limit. Stop that timer as soon 1638 * as the receiver acknowledged any data. 1639 */ 1640 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING && 1641 del_timer(&asoc->timers 1642 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD])) 1643 sctp_association_put(asoc); 1644 1645 /* Mark the destination transport address as 1646 * active if it is not so marked. 1647 */ 1648 if ((transport->state == SCTP_INACTIVE || 1649 transport->state == SCTP_UNCONFIRMED) && 1650 sctp_cmp_addr_exact(&transport->ipaddr, saddr)) { 1651 sctp_assoc_control_transport( 1652 transport->asoc, 1653 transport, 1654 SCTP_TRANSPORT_UP, 1655 SCTP_RECEIVED_SACK); 1656 } 1657 1658 sctp_transport_raise_cwnd(transport, sack_ctsn, 1659 bytes_acked); 1660 1661 transport->flight_size -= bytes_acked; 1662 if (transport->flight_size == 0) 1663 transport->partial_bytes_acked = 0; 1664 q->outstanding_bytes -= bytes_acked + migrate_bytes; 1665 } else { 1666 /* RFC 2960 6.1, sctpimpguide-06 2.15.2 1667 * When a sender is doing zero window probing, it 1668 * should not timeout the association if it continues 1669 * to receive new packets from the receiver. The 1670 * reason is that the receiver MAY keep its window 1671 * closed for an indefinite time. 1672 * A sender is doing zero window probing when the 1673 * receiver's advertised window is zero, and there is 1674 * only one data chunk in flight to the receiver. 1675 * 1676 * Allow the association to timeout while in SHUTDOWN 1677 * PENDING or SHUTDOWN RECEIVED in case the receiver 1678 * stays in zero window mode forever. 1679 */ 1680 if (!q->asoc->peer.rwnd && 1681 !list_empty(&tlist) && 1682 (sack_ctsn+2 == q->asoc->next_tsn) && 1683 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) { 1684 SCTP_DEBUG_PRINTK("%s: SACK received for zero " 1685 "window probe: %u\n", 1686 __func__, sack_ctsn); 1687 q->asoc->overall_error_count = 0; 1688 transport->error_count = 0; 1689 } 1690 } 1691 1692 /* RFC 2960 6.3.2 Retransmission Timer Rules 1693 * 1694 * R2) Whenever all outstanding data sent to an address have 1695 * been acknowledged, turn off the T3-rtx timer of that 1696 * address. 1697 */ 1698 if (!transport->flight_size) { 1699 if (timer_pending(&transport->T3_rtx_timer) && 1700 del_timer(&transport->T3_rtx_timer)) { 1701 sctp_transport_put(transport); 1702 } 1703 } else if (restart_timer) { 1704 if (!mod_timer(&transport->T3_rtx_timer, 1705 jiffies + transport->rto)) 1706 sctp_transport_hold(transport); 1707 } 1708 } 1709 1710 list_splice(&tlist, transmitted_queue); 1711 } 1712 1713 /* Mark chunks as missing and consequently may get retransmitted. */ 1714 static void sctp_mark_missing(struct sctp_outq *q, 1715 struct list_head *transmitted_queue, 1716 struct sctp_transport *transport, 1717 __u32 highest_new_tsn_in_sack, 1718 int count_of_newacks) 1719 { 1720 struct sctp_chunk *chunk; 1721 __u32 tsn; 1722 char do_fast_retransmit = 0; 1723 struct sctp_association *asoc = q->asoc; 1724 struct sctp_transport *primary = asoc->peer.primary_path; 1725 1726 list_for_each_entry(chunk, transmitted_queue, transmitted_list) { 1727 1728 tsn = ntohl(chunk->subh.data_hdr->tsn); 1729 1730 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all 1731 * 'Unacknowledged TSN's', if the TSN number of an 1732 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack' 1733 * value, increment the 'TSN.Missing.Report' count on that 1734 * chunk if it has NOT been fast retransmitted or marked for 1735 * fast retransmit already. 1736 */ 1737 if (chunk->fast_retransmit == SCTP_CAN_FRTX && 1738 !chunk->tsn_gap_acked && 1739 TSN_lt(tsn, highest_new_tsn_in_sack)) { 1740 1741 /* SFR-CACC may require us to skip marking 1742 * this chunk as missing. 1743 */ 1744 if (!transport || !sctp_cacc_skip(primary, 1745 chunk->transport, 1746 count_of_newacks, tsn)) { 1747 chunk->tsn_missing_report++; 1748 1749 SCTP_DEBUG_PRINTK( 1750 "%s: TSN 0x%x missing counter: %d\n", 1751 __func__, tsn, 1752 chunk->tsn_missing_report); 1753 } 1754 } 1755 /* 1756 * M4) If any DATA chunk is found to have a 1757 * 'TSN.Missing.Report' 1758 * value larger than or equal to 3, mark that chunk for 1759 * retransmission and start the fast retransmit procedure. 1760 */ 1761 1762 if (chunk->tsn_missing_report >= 3) { 1763 chunk->fast_retransmit = SCTP_NEED_FRTX; 1764 do_fast_retransmit = 1; 1765 } 1766 } 1767 1768 if (transport) { 1769 if (do_fast_retransmit) 1770 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX); 1771 1772 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, " 1773 "ssthresh: %d, flight_size: %d, pba: %d\n", 1774 __func__, transport, transport->cwnd, 1775 transport->ssthresh, transport->flight_size, 1776 transport->partial_bytes_acked); 1777 } 1778 } 1779 1780 /* Is the given TSN acked by this packet? */ 1781 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn) 1782 { 1783 int i; 1784 sctp_sack_variable_t *frags; 1785 __u16 gap; 1786 __u32 ctsn = ntohl(sack->cum_tsn_ack); 1787 1788 if (TSN_lte(tsn, ctsn)) 1789 goto pass; 1790 1791 /* 3.3.4 Selective Acknowledgement (SACK) (3): 1792 * 1793 * Gap Ack Blocks: 1794 * These fields contain the Gap Ack Blocks. They are repeated 1795 * for each Gap Ack Block up to the number of Gap Ack Blocks 1796 * defined in the Number of Gap Ack Blocks field. All DATA 1797 * chunks with TSNs greater than or equal to (Cumulative TSN 1798 * Ack + Gap Ack Block Start) and less than or equal to 1799 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack 1800 * Block are assumed to have been received correctly. 1801 */ 1802 1803 frags = sack->variable; 1804 gap = tsn - ctsn; 1805 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) { 1806 if (TSN_lte(ntohs(frags[i].gab.start), gap) && 1807 TSN_lte(gap, ntohs(frags[i].gab.end))) 1808 goto pass; 1809 } 1810 1811 return 0; 1812 pass: 1813 return 1; 1814 } 1815 1816 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist, 1817 int nskips, __be16 stream) 1818 { 1819 int i; 1820 1821 for (i = 0; i < nskips; i++) { 1822 if (skiplist[i].stream == stream) 1823 return i; 1824 } 1825 return i; 1826 } 1827 1828 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */ 1829 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn) 1830 { 1831 struct sctp_association *asoc = q->asoc; 1832 struct sctp_chunk *ftsn_chunk = NULL; 1833 struct sctp_fwdtsn_skip ftsn_skip_arr[10]; 1834 int nskips = 0; 1835 int skip_pos = 0; 1836 __u32 tsn; 1837 struct sctp_chunk *chunk; 1838 struct list_head *lchunk, *temp; 1839 1840 if (!asoc->peer.prsctp_capable) 1841 return; 1842 1843 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the 1844 * received SACK. 1845 * 1846 * If (Advanced.Peer.Ack.Point < SackCumAck), then update 1847 * Advanced.Peer.Ack.Point to be equal to SackCumAck. 1848 */ 1849 if (TSN_lt(asoc->adv_peer_ack_point, ctsn)) 1850 asoc->adv_peer_ack_point = ctsn; 1851 1852 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point" 1853 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as 1854 * the chunk next in the out-queue space is marked as "abandoned" as 1855 * shown in the following example: 1856 * 1857 * Assuming that a SACK arrived with the Cumulative TSN ACK 102 1858 * and the Advanced.Peer.Ack.Point is updated to this value: 1859 * 1860 * out-queue at the end of ==> out-queue after Adv.Ack.Point 1861 * normal SACK processing local advancement 1862 * ... ... 1863 * Adv.Ack.Pt-> 102 acked 102 acked 1864 * 103 abandoned 103 abandoned 1865 * 104 abandoned Adv.Ack.P-> 104 abandoned 1866 * 105 105 1867 * 106 acked 106 acked 1868 * ... ... 1869 * 1870 * In this example, the data sender successfully advanced the 1871 * "Advanced.Peer.Ack.Point" from 102 to 104 locally. 1872 */ 1873 list_for_each_safe(lchunk, temp, &q->abandoned) { 1874 chunk = list_entry(lchunk, struct sctp_chunk, 1875 transmitted_list); 1876 tsn = ntohl(chunk->subh.data_hdr->tsn); 1877 1878 /* Remove any chunks in the abandoned queue that are acked by 1879 * the ctsn. 1880 */ 1881 if (TSN_lte(tsn, ctsn)) { 1882 list_del_init(lchunk); 1883 sctp_chunk_free(chunk); 1884 } else { 1885 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) { 1886 asoc->adv_peer_ack_point = tsn; 1887 if (chunk->chunk_hdr->flags & 1888 SCTP_DATA_UNORDERED) 1889 continue; 1890 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0], 1891 nskips, 1892 chunk->subh.data_hdr->stream); 1893 ftsn_skip_arr[skip_pos].stream = 1894 chunk->subh.data_hdr->stream; 1895 ftsn_skip_arr[skip_pos].ssn = 1896 chunk->subh.data_hdr->ssn; 1897 if (skip_pos == nskips) 1898 nskips++; 1899 if (nskips == 10) 1900 break; 1901 } else 1902 break; 1903 } 1904 } 1905 1906 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point" 1907 * is greater than the Cumulative TSN ACK carried in the received 1908 * SACK, the data sender MUST send the data receiver a FORWARD TSN 1909 * chunk containing the latest value of the 1910 * "Advanced.Peer.Ack.Point". 1911 * 1912 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD 1913 * list each stream and sequence number in the forwarded TSN. This 1914 * information will enable the receiver to easily find any 1915 * stranded TSN's waiting on stream reorder queues. Each stream 1916 * SHOULD only be reported once; this means that if multiple 1917 * abandoned messages occur in the same stream then only the 1918 * highest abandoned stream sequence number is reported. If the 1919 * total size of the FORWARD TSN does NOT fit in a single MTU then 1920 * the sender of the FORWARD TSN SHOULD lower the 1921 * Advanced.Peer.Ack.Point to the last TSN that will fit in a 1922 * single MTU. 1923 */ 1924 if (asoc->adv_peer_ack_point > ctsn) 1925 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point, 1926 nskips, &ftsn_skip_arr[0]); 1927 1928 if (ftsn_chunk) { 1929 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list); 1930 SCTP_INC_STATS(sock_net(asoc->base.sk), SCTP_MIB_OUTCTRLCHUNKS); 1931 } 1932 } 1933