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