Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* SCTP kernel reference Implementation |
| 2 | * (C) Copyright IBM Corp. 2001, 2004 |
| 3 | * Copyright (c) 1999-2000 Cisco, Inc. |
| 4 | * Copyright (c) 1999-2001 Motorola, Inc. |
| 5 | * |
| 6 | * This file is part of the SCTP kernel reference Implementation |
| 7 | * |
| 8 | * These functions handle output processing. |
| 9 | * |
| 10 | * The SCTP reference implementation is free software; |
| 11 | * you can redistribute it and/or modify it under the terms of |
| 12 | * the GNU General Public License as published by |
| 13 | * the Free Software Foundation; either version 2, or (at your option) |
| 14 | * any later version. |
| 15 | * |
| 16 | * The SCTP reference implementation is distributed in the hope that it |
| 17 | * will be useful, but WITHOUT ANY WARRANTY; without even the implied |
| 18 | * ************************ |
| 19 | * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. |
| 20 | * See the GNU General Public License for more details. |
| 21 | * |
| 22 | * You should have received a copy of the GNU General Public License |
| 23 | * along with GNU CC; see the file COPYING. If not, write to |
| 24 | * the Free Software Foundation, 59 Temple Place - Suite 330, |
| 25 | * Boston, MA 02111-1307, USA. |
| 26 | * |
| 27 | * Please send any bug reports or fixes you make to the |
| 28 | * email address(es): |
| 29 | * lksctp developers <lksctp-developers@lists.sourceforge.net> |
| 30 | * |
| 31 | * Or submit a bug report through the following website: |
| 32 | * http://www.sf.net/projects/lksctp |
| 33 | * |
| 34 | * Written or modified by: |
| 35 | * La Monte H.P. Yarroll <piggy@acm.org> |
| 36 | * Karl Knutson <karl@athena.chicago.il.us> |
| 37 | * Jon Grimm <jgrimm@austin.ibm.com> |
| 38 | * Sridhar Samudrala <sri@us.ibm.com> |
| 39 | * |
| 40 | * Any bugs reported given to us we will try to fix... any fixes shared will |
| 41 | * be incorporated into the next SCTP release. |
| 42 | */ |
| 43 | |
| 44 | #include <linux/types.h> |
| 45 | #include <linux/kernel.h> |
| 46 | #include <linux/wait.h> |
| 47 | #include <linux/time.h> |
| 48 | #include <linux/ip.h> |
| 49 | #include <linux/ipv6.h> |
| 50 | #include <linux/init.h> |
| 51 | #include <net/inet_ecn.h> |
| 52 | #include <net/icmp.h> |
| 53 | |
| 54 | #ifndef TEST_FRAME |
| 55 | #include <net/tcp.h> |
| 56 | #endif /* TEST_FRAME (not defined) */ |
| 57 | |
| 58 | #include <linux/socket.h> /* for sa_family_t */ |
| 59 | #include <net/sock.h> |
| 60 | |
| 61 | #include <net/sctp/sctp.h> |
| 62 | #include <net/sctp/sm.h> |
| 63 | |
| 64 | /* Forward declarations for private helpers. */ |
| 65 | static sctp_xmit_t sctp_packet_append_data(struct sctp_packet *packet, |
| 66 | struct sctp_chunk *chunk); |
| 67 | |
| 68 | /* Config a packet. |
| 69 | * This appears to be a followup set of initializations. |
| 70 | */ |
| 71 | struct sctp_packet *sctp_packet_config(struct sctp_packet *packet, |
| 72 | __u32 vtag, int ecn_capable) |
| 73 | { |
| 74 | struct sctp_chunk *chunk = NULL; |
| 75 | |
| 76 | SCTP_DEBUG_PRINTK("%s: packet:%p vtag:0x%x\n", __FUNCTION__, |
| 77 | packet, vtag); |
| 78 | |
| 79 | packet->vtag = vtag; |
| 80 | packet->has_cookie_echo = 0; |
| 81 | packet->has_sack = 0; |
| 82 | packet->ipfragok = 0; |
| 83 | |
| 84 | if (ecn_capable && sctp_packet_empty(packet)) { |
| 85 | chunk = sctp_get_ecne_prepend(packet->transport->asoc); |
| 86 | |
| 87 | /* If there a is a prepend chunk stick it on the list before |
YOSHIFUJI Hideaki | d808ad9 | 2007-02-09 23:25:18 +0900 | [diff] [blame^] | 88 | * any other chunks get appended. |
| 89 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 90 | if (chunk) |
| 91 | sctp_packet_append_chunk(packet, chunk); |
| 92 | } |
| 93 | |
| 94 | return packet; |
| 95 | } |
| 96 | |
| 97 | /* Initialize the packet structure. */ |
| 98 | struct sctp_packet *sctp_packet_init(struct sctp_packet *packet, |
| 99 | struct sctp_transport *transport, |
| 100 | __u16 sport, __u16 dport) |
| 101 | { |
| 102 | struct sctp_association *asoc = transport->asoc; |
| 103 | size_t overhead; |
| 104 | |
| 105 | SCTP_DEBUG_PRINTK("%s: packet:%p transport:%p\n", __FUNCTION__, |
| 106 | packet, transport); |
| 107 | |
| 108 | packet->transport = transport; |
| 109 | packet->source_port = sport; |
| 110 | packet->destination_port = dport; |
David S. Miller | 79af02c | 2005-07-08 21:47:49 -0700 | [diff] [blame] | 111 | INIT_LIST_HEAD(&packet->chunk_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | if (asoc) { |
YOSHIFUJI Hideaki | d808ad9 | 2007-02-09 23:25:18 +0900 | [diff] [blame^] | 113 | struct sctp_sock *sp = sctp_sk(asoc->base.sk); |
| 114 | overhead = sp->pf->af->net_header_len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 115 | } else { |
| 116 | overhead = sizeof(struct ipv6hdr); |
| 117 | } |
| 118 | overhead += sizeof(struct sctphdr); |
| 119 | packet->overhead = overhead; |
| 120 | packet->size = overhead; |
| 121 | packet->vtag = 0; |
| 122 | packet->has_cookie_echo = 0; |
| 123 | packet->has_sack = 0; |
| 124 | packet->ipfragok = 0; |
| 125 | packet->malloced = 0; |
| 126 | return packet; |
| 127 | } |
| 128 | |
| 129 | /* Free a packet. */ |
| 130 | void sctp_packet_free(struct sctp_packet *packet) |
| 131 | { |
David S. Miller | 79af02c | 2005-07-08 21:47:49 -0700 | [diff] [blame] | 132 | struct sctp_chunk *chunk, *tmp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 133 | |
| 134 | SCTP_DEBUG_PRINTK("%s: packet:%p\n", __FUNCTION__, packet); |
| 135 | |
David S. Miller | 79af02c | 2005-07-08 21:47:49 -0700 | [diff] [blame] | 136 | list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) { |
| 137 | list_del_init(&chunk->list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 138 | sctp_chunk_free(chunk); |
David S. Miller | 79af02c | 2005-07-08 21:47:49 -0700 | [diff] [blame] | 139 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 140 | |
| 141 | if (packet->malloced) |
| 142 | kfree(packet); |
| 143 | } |
| 144 | |
| 145 | /* This routine tries to append the chunk to the offered packet. If adding |
| 146 | * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk |
| 147 | * is not present in the packet, it transmits the input packet. |
| 148 | * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long |
| 149 | * as it can fit in the packet, but any more data that does not fit in this |
| 150 | * packet can be sent only after receiving the COOKIE_ACK. |
| 151 | */ |
| 152 | sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet, |
| 153 | struct sctp_chunk *chunk) |
| 154 | { |
| 155 | sctp_xmit_t retval; |
| 156 | int error = 0; |
| 157 | |
| 158 | SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __FUNCTION__, |
| 159 | packet, chunk); |
| 160 | |
| 161 | switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) { |
| 162 | case SCTP_XMIT_PMTU_FULL: |
| 163 | if (!packet->has_cookie_echo) { |
| 164 | error = sctp_packet_transmit(packet); |
| 165 | if (error < 0) |
| 166 | chunk->skb->sk->sk_err = -error; |
| 167 | |
| 168 | /* If we have an empty packet, then we can NOT ever |
| 169 | * return PMTU_FULL. |
| 170 | */ |
| 171 | retval = sctp_packet_append_chunk(packet, chunk); |
| 172 | } |
| 173 | break; |
| 174 | |
| 175 | case SCTP_XMIT_RWND_FULL: |
| 176 | case SCTP_XMIT_OK: |
| 177 | case SCTP_XMIT_NAGLE_DELAY: |
| 178 | break; |
| 179 | }; |
| 180 | |
| 181 | return retval; |
| 182 | } |
| 183 | |
| 184 | /* Try to bundle a SACK with the packet. */ |
| 185 | static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt, |
| 186 | struct sctp_chunk *chunk) |
| 187 | { |
| 188 | sctp_xmit_t retval = SCTP_XMIT_OK; |
| 189 | |
| 190 | /* If sending DATA and haven't aleady bundled a SACK, try to |
| 191 | * bundle one in to the packet. |
| 192 | */ |
| 193 | if (sctp_chunk_is_data(chunk) && !pkt->has_sack && |
| 194 | !pkt->has_cookie_echo) { |
| 195 | struct sctp_association *asoc; |
| 196 | asoc = pkt->transport->asoc; |
| 197 | |
| 198 | if (asoc->a_rwnd > asoc->rwnd) { |
| 199 | struct sctp_chunk *sack; |
| 200 | asoc->a_rwnd = asoc->rwnd; |
| 201 | sack = sctp_make_sack(asoc); |
| 202 | if (sack) { |
| 203 | struct timer_list *timer; |
| 204 | retval = sctp_packet_append_chunk(pkt, sack); |
| 205 | asoc->peer.sack_needed = 0; |
| 206 | timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK]; |
| 207 | if (timer_pending(timer) && del_timer(timer)) |
| 208 | sctp_association_put(asoc); |
| 209 | } |
| 210 | } |
| 211 | } |
| 212 | return retval; |
| 213 | } |
| 214 | |
| 215 | /* Append a chunk to the offered packet reporting back any inability to do |
| 216 | * so. |
| 217 | */ |
| 218 | sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet, |
| 219 | struct sctp_chunk *chunk) |
| 220 | { |
| 221 | sctp_xmit_t retval = SCTP_XMIT_OK; |
| 222 | __u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length)); |
| 223 | size_t psize; |
| 224 | size_t pmtu; |
| 225 | int too_big; |
| 226 | |
| 227 | SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __FUNCTION__, packet, |
| 228 | chunk); |
| 229 | |
| 230 | retval = sctp_packet_bundle_sack(packet, chunk); |
| 231 | psize = packet->size; |
| 232 | |
| 233 | if (retval != SCTP_XMIT_OK) |
| 234 | goto finish; |
| 235 | |
| 236 | pmtu = ((packet->transport->asoc) ? |
Frank Filz | 52ccb8e | 2005-12-22 11:36:46 -0800 | [diff] [blame] | 237 | (packet->transport->asoc->pathmtu) : |
| 238 | (packet->transport->pathmtu)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 239 | |
| 240 | too_big = (psize + chunk_len > pmtu); |
| 241 | |
| 242 | /* Decide if we need to fragment or resubmit later. */ |
| 243 | if (too_big) { |
| 244 | /* Both control chunks and data chunks with TSNs are |
| 245 | * non-fragmentable. |
| 246 | */ |
| 247 | if (sctp_packet_empty(packet) || !sctp_chunk_is_data(chunk)) { |
| 248 | /* We no longer do re-fragmentation. |
| 249 | * Just fragment at the IP layer, if we |
| 250 | * actually hit this condition |
| 251 | */ |
| 252 | packet->ipfragok = 1; |
| 253 | goto append; |
| 254 | |
| 255 | } else { |
| 256 | retval = SCTP_XMIT_PMTU_FULL; |
| 257 | goto finish; |
| 258 | } |
| 259 | } |
| 260 | |
| 261 | append: |
| 262 | /* We believe that this chunk is OK to add to the packet (as |
| 263 | * long as we have the cwnd for it). |
| 264 | */ |
| 265 | |
| 266 | /* DATA is a special case since we must examine both rwnd and cwnd |
| 267 | * before we send DATA. |
| 268 | */ |
| 269 | if (sctp_chunk_is_data(chunk)) { |
| 270 | retval = sctp_packet_append_data(packet, chunk); |
| 271 | /* Disallow SACK bundling after DATA. */ |
| 272 | packet->has_sack = 1; |
| 273 | if (SCTP_XMIT_OK != retval) |
| 274 | goto finish; |
| 275 | } else if (SCTP_CID_COOKIE_ECHO == chunk->chunk_hdr->type) |
| 276 | packet->has_cookie_echo = 1; |
| 277 | else if (SCTP_CID_SACK == chunk->chunk_hdr->type) |
| 278 | packet->has_sack = 1; |
| 279 | |
| 280 | /* It is OK to send this chunk. */ |
David S. Miller | 79af02c | 2005-07-08 21:47:49 -0700 | [diff] [blame] | 281 | list_add_tail(&chunk->list, &packet->chunk_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 282 | packet->size += chunk_len; |
| 283 | chunk->transport = packet->transport; |
| 284 | finish: |
| 285 | return retval; |
| 286 | } |
| 287 | |
| 288 | /* All packets are sent to the network through this function from |
| 289 | * sctp_outq_tail(). |
| 290 | * |
| 291 | * The return value is a normal kernel error return value. |
| 292 | */ |
| 293 | int sctp_packet_transmit(struct sctp_packet *packet) |
| 294 | { |
| 295 | struct sctp_transport *tp = packet->transport; |
| 296 | struct sctp_association *asoc = tp->asoc; |
| 297 | struct sctphdr *sh; |
Sridhar Samudrala | 503b55f | 2006-06-17 22:57:28 -0700 | [diff] [blame] | 298 | __u32 crc32 = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 299 | struct sk_buff *nskb; |
David S. Miller | 79af02c | 2005-07-08 21:47:49 -0700 | [diff] [blame] | 300 | struct sctp_chunk *chunk, *tmp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 301 | struct sock *sk; |
| 302 | int err = 0; |
| 303 | int padding; /* How much padding do we need? */ |
| 304 | __u8 has_data = 0; |
Sridhar Samudrala | 503b55f | 2006-06-17 22:57:28 -0700 | [diff] [blame] | 305 | struct dst_entry *dst = tp->dst; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 306 | |
| 307 | SCTP_DEBUG_PRINTK("%s: packet:%p\n", __FUNCTION__, packet); |
| 308 | |
| 309 | /* Do NOT generate a chunkless packet. */ |
David S. Miller | 79af02c | 2005-07-08 21:47:49 -0700 | [diff] [blame] | 310 | if (list_empty(&packet->chunk_list)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 311 | return err; |
| 312 | |
| 313 | /* Set up convenience variables... */ |
David S. Miller | 79af02c | 2005-07-08 21:47:49 -0700 | [diff] [blame] | 314 | chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 315 | sk = chunk->skb->sk; |
| 316 | |
| 317 | /* Allocate the new skb. */ |
Sridhar Samudrala | 594ccc1 | 2005-04-28 12:00:23 -0700 | [diff] [blame] | 318 | nskb = alloc_skb(packet->size + LL_MAX_HEADER, GFP_ATOMIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 319 | if (!nskb) |
| 320 | goto nomem; |
| 321 | |
| 322 | /* Make sure the outbound skb has enough header room reserved. */ |
Sridhar Samudrala | 594ccc1 | 2005-04-28 12:00:23 -0700 | [diff] [blame] | 323 | skb_reserve(nskb, packet->overhead + LL_MAX_HEADER); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 324 | |
| 325 | /* Set the owning socket so that we know where to get the |
| 326 | * destination IP address. |
| 327 | */ |
| 328 | skb_set_owner_w(nskb, sk); |
| 329 | |
Sridhar Samudrala | 503b55f | 2006-06-17 22:57:28 -0700 | [diff] [blame] | 330 | /* The 'obsolete' field of dst is set to 2 when a dst is freed. */ |
| 331 | if (!dst || (dst->obsolete > 1)) { |
| 332 | dst_release(dst); |
| 333 | sctp_transport_route(tp, NULL, sctp_sk(sk)); |
| 334 | if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) { |
| 335 | sctp_assoc_sync_pmtu(asoc); |
| 336 | } |
| 337 | } |
| 338 | nskb->dst = dst_clone(tp->dst); |
| 339 | if (!nskb->dst) |
| 340 | goto no_route; |
| 341 | dst = nskb->dst; |
| 342 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 343 | /* Build the SCTP header. */ |
| 344 | sh = (struct sctphdr *)skb_push(nskb, sizeof(struct sctphdr)); |
| 345 | sh->source = htons(packet->source_port); |
| 346 | sh->dest = htons(packet->destination_port); |
| 347 | |
| 348 | /* From 6.8 Adler-32 Checksum Calculation: |
| 349 | * After the packet is constructed (containing the SCTP common |
| 350 | * header and one or more control or DATA chunks), the |
| 351 | * transmitter shall: |
| 352 | * |
| 353 | * 1) Fill in the proper Verification Tag in the SCTP common |
| 354 | * header and initialize the checksum field to 0's. |
| 355 | */ |
| 356 | sh->vtag = htonl(packet->vtag); |
| 357 | sh->checksum = 0; |
| 358 | |
| 359 | /* 2) Calculate the Adler-32 checksum of the whole packet, |
| 360 | * including the SCTP common header and all the |
| 361 | * chunks. |
| 362 | * |
| 363 | * Note: Adler-32 is no longer applicable, as has been replaced |
| 364 | * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>. |
| 365 | */ |
Sridhar Samudrala | 503b55f | 2006-06-17 22:57:28 -0700 | [diff] [blame] | 366 | if (!(dst->dev->features & NETIF_F_NO_CSUM)) |
| 367 | crc32 = sctp_start_cksum((__u8 *)sh, sizeof(struct sctphdr)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | |
| 369 | /** |
| 370 | * 6.10 Bundling |
| 371 | * |
| 372 | * An endpoint bundles chunks by simply including multiple |
| 373 | * chunks in one outbound SCTP packet. ... |
| 374 | */ |
| 375 | |
| 376 | /** |
| 377 | * 3.2 Chunk Field Descriptions |
| 378 | * |
| 379 | * The total length of a chunk (including Type, Length and |
| 380 | * Value fields) MUST be a multiple of 4 bytes. If the length |
| 381 | * of the chunk is not a multiple of 4 bytes, the sender MUST |
| 382 | * pad the chunk with all zero bytes and this padding is not |
| 383 | * included in the chunk length field. The sender should |
| 384 | * never pad with more than 3 bytes. |
| 385 | * |
| 386 | * [This whole comment explains WORD_ROUND() below.] |
| 387 | */ |
| 388 | SCTP_DEBUG_PRINTK("***sctp_transmit_packet***\n"); |
David S. Miller | 79af02c | 2005-07-08 21:47:49 -0700 | [diff] [blame] | 389 | list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) { |
| 390 | list_del_init(&chunk->list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 391 | if (sctp_chunk_is_data(chunk)) { |
| 392 | |
| 393 | if (!chunk->has_tsn) { |
| 394 | sctp_chunk_assign_ssn(chunk); |
| 395 | sctp_chunk_assign_tsn(chunk); |
| 396 | |
| 397 | /* 6.3.1 C4) When data is in flight and when allowed |
| 398 | * by rule C5, a new RTT measurement MUST be made each |
| 399 | * round trip. Furthermore, new RTT measurements |
| 400 | * SHOULD be made no more than once per round-trip |
| 401 | * for a given destination transport address. |
| 402 | */ |
| 403 | |
| 404 | if (!tp->rto_pending) { |
| 405 | chunk->rtt_in_progress = 1; |
| 406 | tp->rto_pending = 1; |
| 407 | } |
| 408 | } else |
| 409 | chunk->resent = 1; |
| 410 | |
| 411 | chunk->sent_at = jiffies; |
| 412 | has_data = 1; |
| 413 | } |
| 414 | |
| 415 | padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len; |
| 416 | if (padding) |
| 417 | memset(skb_put(chunk->skb, padding), 0, padding); |
| 418 | |
Sridhar Samudrala | 503b55f | 2006-06-17 22:57:28 -0700 | [diff] [blame] | 419 | if (dst->dev->features & NETIF_F_NO_CSUM) |
| 420 | memcpy(skb_put(nskb, chunk->skb->len), |
| 421 | chunk->skb->data, chunk->skb->len); |
| 422 | else |
| 423 | crc32 = sctp_update_copy_cksum(skb_put(nskb, |
| 424 | chunk->skb->len), |
| 425 | chunk->skb->data, |
| 426 | chunk->skb->len, crc32); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 427 | |
| 428 | SCTP_DEBUG_PRINTK("%s %p[%s] %s 0x%x, %s %d, %s %d, %s %d\n", |
| 429 | "*** Chunk", chunk, |
| 430 | sctp_cname(SCTP_ST_CHUNK( |
| 431 | chunk->chunk_hdr->type)), |
| 432 | chunk->has_tsn ? "TSN" : "No TSN", |
| 433 | chunk->has_tsn ? |
| 434 | ntohl(chunk->subh.data_hdr->tsn) : 0, |
| 435 | "length", ntohs(chunk->chunk_hdr->length), |
| 436 | "chunk->skb->len", chunk->skb->len, |
| 437 | "rtt_in_progress", chunk->rtt_in_progress); |
| 438 | |
| 439 | /* |
| 440 | * If this is a control chunk, this is our last |
| 441 | * reference. Free data chunks after they've been |
| 442 | * acknowledged or have failed. |
| 443 | */ |
| 444 | if (!sctp_chunk_is_data(chunk)) |
YOSHIFUJI Hideaki | d808ad9 | 2007-02-09 23:25:18 +0900 | [diff] [blame^] | 445 | sctp_chunk_free(chunk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 446 | } |
| 447 | |
| 448 | /* Perform final transformation on checksum. */ |
Sridhar Samudrala | 503b55f | 2006-06-17 22:57:28 -0700 | [diff] [blame] | 449 | if (!(dst->dev->features & NETIF_F_NO_CSUM)) |
| 450 | crc32 = sctp_end_cksum(crc32); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 451 | |
| 452 | /* 3) Put the resultant value into the checksum field in the |
| 453 | * common header, and leave the rest of the bits unchanged. |
| 454 | */ |
| 455 | sh->checksum = htonl(crc32); |
| 456 | |
| 457 | /* IP layer ECN support |
| 458 | * From RFC 2481 |
| 459 | * "The ECN-Capable Transport (ECT) bit would be set by the |
| 460 | * data sender to indicate that the end-points of the |
| 461 | * transport protocol are ECN-capable." |
| 462 | * |
| 463 | * Now setting the ECT bit all the time, as it should not cause |
| 464 | * any problems protocol-wise even if our peer ignores it. |
| 465 | * |
| 466 | * Note: The works for IPv6 layer checks this bit too later |
| 467 | * in transmission. See IP6_ECN_flow_xmit(). |
| 468 | */ |
| 469 | INET_ECN_xmit(nskb->sk); |
| 470 | |
| 471 | /* Set up the IP options. */ |
| 472 | /* BUG: not implemented |
| 473 | * For v4 this all lives somewhere in sk->sk_opt... |
| 474 | */ |
| 475 | |
| 476 | /* Dump that on IP! */ |
| 477 | if (asoc && asoc->peer.last_sent_to != tp) { |
| 478 | /* Considering the multiple CPU scenario, this is a |
| 479 | * "correcter" place for last_sent_to. --xguo |
| 480 | */ |
| 481 | asoc->peer.last_sent_to = tp; |
| 482 | } |
| 483 | |
| 484 | if (has_data) { |
| 485 | struct timer_list *timer; |
| 486 | unsigned long timeout; |
| 487 | |
| 488 | tp->last_time_used = jiffies; |
| 489 | |
| 490 | /* Restart the AUTOCLOSE timer when sending data. */ |
| 491 | if (sctp_state(asoc, ESTABLISHED) && asoc->autoclose) { |
| 492 | timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE]; |
| 493 | timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE]; |
| 494 | |
| 495 | if (!mod_timer(timer, jiffies + timeout)) |
| 496 | sctp_association_hold(asoc); |
| 497 | } |
| 498 | } |
| 499 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 500 | SCTP_DEBUG_PRINTK("***sctp_transmit_packet*** skb len %d\n", |
| 501 | nskb->len); |
| 502 | |
Frank Filz | 52ccb8e | 2005-12-22 11:36:46 -0800 | [diff] [blame] | 503 | if (tp->param_flags & SPP_PMTUD_ENABLE) |
| 504 | (*tp->af_specific->sctp_xmit)(nskb, tp, packet->ipfragok); |
| 505 | else |
| 506 | (*tp->af_specific->sctp_xmit)(nskb, tp, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 507 | |
| 508 | out: |
| 509 | packet->size = packet->overhead; |
| 510 | return err; |
| 511 | no_route: |
| 512 | kfree_skb(nskb); |
| 513 | IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES); |
| 514 | |
| 515 | /* FIXME: Returning the 'err' will effect all the associations |
| 516 | * associated with a socket, although only one of the paths of the |
| 517 | * association is unreachable. |
| 518 | * The real failure of a transport or association can be passed on |
| 519 | * to the user via notifications. So setting this error may not be |
| 520 | * required. |
| 521 | */ |
| 522 | /* err = -EHOSTUNREACH; */ |
| 523 | err: |
| 524 | /* Control chunks are unreliable so just drop them. DATA chunks |
| 525 | * will get resent or dropped later. |
| 526 | */ |
| 527 | |
David S. Miller | 79af02c | 2005-07-08 21:47:49 -0700 | [diff] [blame] | 528 | list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) { |
| 529 | list_del_init(&chunk->list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 530 | if (!sctp_chunk_is_data(chunk)) |
YOSHIFUJI Hideaki | d808ad9 | 2007-02-09 23:25:18 +0900 | [diff] [blame^] | 531 | sctp_chunk_free(chunk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 532 | } |
| 533 | goto out; |
| 534 | nomem: |
| 535 | err = -ENOMEM; |
| 536 | goto err; |
| 537 | } |
| 538 | |
| 539 | /******************************************************************** |
| 540 | * 2nd Level Abstractions |
| 541 | ********************************************************************/ |
| 542 | |
| 543 | /* This private function handles the specifics of appending DATA chunks. */ |
| 544 | static sctp_xmit_t sctp_packet_append_data(struct sctp_packet *packet, |
| 545 | struct sctp_chunk *chunk) |
| 546 | { |
| 547 | sctp_xmit_t retval = SCTP_XMIT_OK; |
| 548 | size_t datasize, rwnd, inflight; |
| 549 | struct sctp_transport *transport = packet->transport; |
| 550 | __u32 max_burst_bytes; |
| 551 | struct sctp_association *asoc = transport->asoc; |
| 552 | struct sctp_sock *sp = sctp_sk(asoc->base.sk); |
| 553 | struct sctp_outq *q = &asoc->outqueue; |
| 554 | |
| 555 | /* RFC 2960 6.1 Transmission of DATA Chunks |
| 556 | * |
| 557 | * A) At any given time, the data sender MUST NOT transmit new data to |
| 558 | * any destination transport address if its peer's rwnd indicates |
| 559 | * that the peer has no buffer space (i.e. rwnd is 0, see Section |
| 560 | * 6.2.1). However, regardless of the value of rwnd (including if it |
| 561 | * is 0), the data sender can always have one DATA chunk in flight to |
| 562 | * the receiver if allowed by cwnd (see rule B below). This rule |
| 563 | * allows the sender to probe for a change in rwnd that the sender |
| 564 | * missed due to the SACK having been lost in transit from the data |
| 565 | * receiver to the data sender. |
| 566 | */ |
| 567 | |
| 568 | rwnd = asoc->peer.rwnd; |
| 569 | inflight = asoc->outqueue.outstanding_bytes; |
| 570 | |
| 571 | datasize = sctp_data_size(chunk); |
| 572 | |
| 573 | if (datasize > rwnd) { |
| 574 | if (inflight > 0) { |
| 575 | /* We have (at least) one data chunk in flight, |
| 576 | * so we can't fall back to rule 6.1 B). |
| 577 | */ |
| 578 | retval = SCTP_XMIT_RWND_FULL; |
| 579 | goto finish; |
| 580 | } |
| 581 | } |
| 582 | |
| 583 | /* sctpimpguide-05 2.14.2 |
| 584 | * D) When the time comes for the sender to |
| 585 | * transmit new DATA chunks, the protocol parameter Max.Burst MUST |
| 586 | * first be applied to limit how many new DATA chunks may be sent. |
| 587 | * The limit is applied by adjusting cwnd as follows: |
| 588 | * if ((flightsize + Max.Burst * MTU) < cwnd) |
| 589 | * cwnd = flightsize + Max.Burst * MTU |
| 590 | */ |
Frank Filz | 52ccb8e | 2005-12-22 11:36:46 -0800 | [diff] [blame] | 591 | max_burst_bytes = asoc->max_burst * asoc->pathmtu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 592 | if ((transport->flight_size + max_burst_bytes) < transport->cwnd) { |
| 593 | transport->cwnd = transport->flight_size + max_burst_bytes; |
| 594 | SCTP_DEBUG_PRINTK("%s: cwnd limited by max_burst: " |
| 595 | "transport: %p, cwnd: %d, " |
| 596 | "ssthresh: %d, flight_size: %d, " |
| 597 | "pba: %d\n", |
| 598 | __FUNCTION__, transport, |
| 599 | transport->cwnd, |
| 600 | transport->ssthresh, |
| 601 | transport->flight_size, |
| 602 | transport->partial_bytes_acked); |
| 603 | } |
| 604 | |
| 605 | /* RFC 2960 6.1 Transmission of DATA Chunks |
| 606 | * |
| 607 | * B) At any given time, the sender MUST NOT transmit new data |
| 608 | * to a given transport address if it has cwnd or more bytes |
| 609 | * of data outstanding to that transport address. |
| 610 | */ |
| 611 | /* RFC 7.2.4 & the Implementers Guide 2.8. |
| 612 | * |
| 613 | * 3) ... |
| 614 | * When a Fast Retransmit is being performed the sender SHOULD |
| 615 | * ignore the value of cwnd and SHOULD NOT delay retransmission. |
| 616 | */ |
Vlad Yasevich | 27852c2 | 2006-02-02 16:57:31 -0800 | [diff] [blame] | 617 | if (chunk->fast_retransmit <= 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 618 | if (transport->flight_size >= transport->cwnd) { |
| 619 | retval = SCTP_XMIT_RWND_FULL; |
| 620 | goto finish; |
| 621 | } |
| 622 | |
| 623 | /* Nagle's algorithm to solve small-packet problem: |
| 624 | * Inhibit the sending of new chunks when new outgoing data arrives |
| 625 | * if any previously transmitted data on the connection remains |
| 626 | * unacknowledged. |
| 627 | */ |
| 628 | if (!sp->nodelay && sctp_packet_empty(packet) && |
| 629 | q->outstanding_bytes && sctp_state(asoc, ESTABLISHED)) { |
| 630 | unsigned len = datasize + q->out_qlen; |
| 631 | |
| 632 | /* Check whether this chunk and all the rest of pending |
| 633 | * data will fit or delay in hopes of bundling a full |
| 634 | * sized packet. |
| 635 | */ |
Sridhar Samudrala | cd49788 | 2006-09-29 17:09:05 -0700 | [diff] [blame] | 636 | if (len < asoc->frag_point) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 637 | retval = SCTP_XMIT_NAGLE_DELAY; |
| 638 | goto finish; |
| 639 | } |
| 640 | } |
| 641 | |
| 642 | /* Keep track of how many bytes are in flight over this transport. */ |
| 643 | transport->flight_size += datasize; |
| 644 | |
| 645 | /* Keep track of how many bytes are in flight to the receiver. */ |
| 646 | asoc->outqueue.outstanding_bytes += datasize; |
| 647 | |
Sridhar Samudrala | cd49788 | 2006-09-29 17:09:05 -0700 | [diff] [blame] | 648 | /* Update our view of the receiver's rwnd. Include sk_buff overhead |
| 649 | * while updating peer.rwnd so that it reduces the chances of a |
| 650 | * receiver running out of receive buffer space even when receive |
| 651 | * window is still open. This can happen when a sender is sending |
| 652 | * sending small messages. |
| 653 | */ |
| 654 | datasize += sizeof(struct sk_buff); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 655 | if (datasize < rwnd) |
| 656 | rwnd -= datasize; |
| 657 | else |
| 658 | rwnd = 0; |
| 659 | |
| 660 | asoc->peer.rwnd = rwnd; |
| 661 | /* Has been accepted for transmission. */ |
| 662 | if (!asoc->peer.prsctp_capable) |
| 663 | chunk->msg->can_abandon = 0; |
| 664 | |
| 665 | finish: |
| 666 | return retval; |
| 667 | } |