output.c 23 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835
  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. *
  6. * This file is part of the SCTP kernel implementation
  7. *
  8. * These functions handle output processing.
  9. *
  10. * This SCTP 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. * This SCTP 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, see
  24. * <http://www.gnu.org/licenses/>.
  25. *
  26. * Please send any bug reports or fixes you make to the
  27. * email address(es):
  28. * lksctp developers <linux-sctp@vger.kernel.org>
  29. *
  30. * Written or modified by:
  31. * La Monte H.P. Yarroll <piggy@acm.org>
  32. * Karl Knutson <karl@athena.chicago.il.us>
  33. * Jon Grimm <jgrimm@austin.ibm.com>
  34. * Sridhar Samudrala <sri@us.ibm.com>
  35. */
  36. #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  37. #include <linux/types.h>
  38. #include <linux/kernel.h>
  39. #include <linux/wait.h>
  40. #include <linux/time.h>
  41. #include <linux/ip.h>
  42. #include <linux/ipv6.h>
  43. #include <linux/init.h>
  44. #include <linux/slab.h>
  45. #include <net/inet_ecn.h>
  46. #include <net/ip.h>
  47. #include <net/icmp.h>
  48. #include <net/net_namespace.h>
  49. #include <linux/socket.h> /* for sa_family_t */
  50. #include <net/sock.h>
  51. #include <net/sctp/sctp.h>
  52. #include <net/sctp/sm.h>
  53. #include <net/sctp/checksum.h>
  54. /* Forward declarations for private helpers. */
  55. static sctp_xmit_t __sctp_packet_append_chunk(struct sctp_packet *packet,
  56. struct sctp_chunk *chunk);
  57. static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
  58. struct sctp_chunk *chunk);
  59. static void sctp_packet_append_data(struct sctp_packet *packet,
  60. struct sctp_chunk *chunk);
  61. static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
  62. struct sctp_chunk *chunk,
  63. u16 chunk_len);
  64. static void sctp_packet_reset(struct sctp_packet *packet)
  65. {
  66. packet->size = packet->overhead;
  67. packet->has_cookie_echo = 0;
  68. packet->has_sack = 0;
  69. packet->has_data = 0;
  70. packet->has_auth = 0;
  71. packet->ipfragok = 0;
  72. packet->auth = NULL;
  73. }
  74. /* Config a packet.
  75. * This appears to be a followup set of initializations.
  76. */
  77. void sctp_packet_config(struct sctp_packet *packet, __u32 vtag,
  78. int ecn_capable)
  79. {
  80. struct sctp_transport *tp = packet->transport;
  81. struct sctp_association *asoc = tp->asoc;
  82. pr_debug("%s: packet:%p vtag:0x%x\n", __func__, packet, vtag);
  83. packet->vtag = vtag;
  84. if (asoc && tp->dst) {
  85. struct sock *sk = asoc->base.sk;
  86. rcu_read_lock();
  87. if (__sk_dst_get(sk) != tp->dst) {
  88. dst_hold(tp->dst);
  89. sk_setup_caps(sk, tp->dst);
  90. }
  91. if (sk_can_gso(sk)) {
  92. struct net_device *dev = tp->dst->dev;
  93. packet->max_size = dev->gso_max_size;
  94. } else {
  95. packet->max_size = asoc->pathmtu;
  96. }
  97. rcu_read_unlock();
  98. } else {
  99. packet->max_size = tp->pathmtu;
  100. }
  101. if (ecn_capable && sctp_packet_empty(packet)) {
  102. struct sctp_chunk *chunk;
  103. /* If there a is a prepend chunk stick it on the list before
  104. * any other chunks get appended.
  105. */
  106. chunk = sctp_get_ecne_prepend(asoc);
  107. if (chunk)
  108. sctp_packet_append_chunk(packet, chunk);
  109. }
  110. }
  111. /* Initialize the packet structure. */
  112. void sctp_packet_init(struct sctp_packet *packet,
  113. struct sctp_transport *transport,
  114. __u16 sport, __u16 dport)
  115. {
  116. struct sctp_association *asoc = transport->asoc;
  117. size_t overhead;
  118. pr_debug("%s: packet:%p transport:%p\n", __func__, packet, transport);
  119. packet->transport = transport;
  120. packet->source_port = sport;
  121. packet->destination_port = dport;
  122. INIT_LIST_HEAD(&packet->chunk_list);
  123. if (asoc) {
  124. struct sctp_sock *sp = sctp_sk(asoc->base.sk);
  125. overhead = sp->pf->af->net_header_len;
  126. } else {
  127. overhead = sizeof(struct ipv6hdr);
  128. }
  129. overhead += sizeof(struct sctphdr);
  130. packet->overhead = overhead;
  131. sctp_packet_reset(packet);
  132. packet->vtag = 0;
  133. }
  134. /* Free a packet. */
  135. void sctp_packet_free(struct sctp_packet *packet)
  136. {
  137. struct sctp_chunk *chunk, *tmp;
  138. pr_debug("%s: packet:%p\n", __func__, packet);
  139. list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
  140. list_del_init(&chunk->list);
  141. sctp_chunk_free(chunk);
  142. }
  143. }
  144. /* This routine tries to append the chunk to the offered packet. If adding
  145. * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
  146. * is not present in the packet, it transmits the input packet.
  147. * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
  148. * as it can fit in the packet, but any more data that does not fit in this
  149. * packet can be sent only after receiving the COOKIE_ACK.
  150. */
  151. sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
  152. struct sctp_chunk *chunk,
  153. int one_packet, gfp_t gfp)
  154. {
  155. sctp_xmit_t retval;
  156. pr_debug("%s: packet:%p size:%zu chunk:%p size:%d\n", __func__,
  157. packet, packet->size, chunk, chunk->skb ? chunk->skb->len : -1);
  158. switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
  159. case SCTP_XMIT_PMTU_FULL:
  160. if (!packet->has_cookie_echo) {
  161. int error = 0;
  162. error = sctp_packet_transmit(packet, gfp);
  163. if (error < 0)
  164. chunk->skb->sk->sk_err = -error;
  165. /* If we have an empty packet, then we can NOT ever
  166. * return PMTU_FULL.
  167. */
  168. if (!one_packet)
  169. retval = sctp_packet_append_chunk(packet,
  170. chunk);
  171. }
  172. break;
  173. case SCTP_XMIT_RWND_FULL:
  174. case SCTP_XMIT_OK:
  175. case SCTP_XMIT_DELAY:
  176. break;
  177. }
  178. return retval;
  179. }
  180. /* Try to bundle an auth chunk into the packet. */
  181. static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
  182. struct sctp_chunk *chunk)
  183. {
  184. struct sctp_association *asoc = pkt->transport->asoc;
  185. struct sctp_chunk *auth;
  186. sctp_xmit_t retval = SCTP_XMIT_OK;
  187. /* if we don't have an association, we can't do authentication */
  188. if (!asoc)
  189. return retval;
  190. /* See if this is an auth chunk we are bundling or if
  191. * auth is already bundled.
  192. */
  193. if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth)
  194. return retval;
  195. /* if the peer did not request this chunk to be authenticated,
  196. * don't do it
  197. */
  198. if (!chunk->auth)
  199. return retval;
  200. auth = sctp_make_auth(asoc);
  201. if (!auth)
  202. return retval;
  203. retval = __sctp_packet_append_chunk(pkt, auth);
  204. if (retval != SCTP_XMIT_OK)
  205. sctp_chunk_free(auth);
  206. return retval;
  207. }
  208. /* Try to bundle a SACK with the packet. */
  209. static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
  210. struct sctp_chunk *chunk)
  211. {
  212. sctp_xmit_t retval = SCTP_XMIT_OK;
  213. /* If sending DATA and haven't aleady bundled a SACK, try to
  214. * bundle one in to the packet.
  215. */
  216. if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
  217. !pkt->has_cookie_echo) {
  218. struct sctp_association *asoc;
  219. struct timer_list *timer;
  220. asoc = pkt->transport->asoc;
  221. timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
  222. /* If the SACK timer is running, we have a pending SACK */
  223. if (timer_pending(timer)) {
  224. struct sctp_chunk *sack;
  225. if (pkt->transport->sack_generation !=
  226. pkt->transport->asoc->peer.sack_generation)
  227. return retval;
  228. asoc->a_rwnd = asoc->rwnd;
  229. sack = sctp_make_sack(asoc);
  230. if (sack) {
  231. retval = __sctp_packet_append_chunk(pkt, sack);
  232. if (retval != SCTP_XMIT_OK) {
  233. sctp_chunk_free(sack);
  234. goto out;
  235. }
  236. asoc->peer.sack_needed = 0;
  237. if (del_timer(timer))
  238. sctp_association_put(asoc);
  239. }
  240. }
  241. }
  242. out:
  243. return retval;
  244. }
  245. /* Append a chunk to the offered packet reporting back any inability to do
  246. * so.
  247. */
  248. static sctp_xmit_t __sctp_packet_append_chunk(struct sctp_packet *packet,
  249. struct sctp_chunk *chunk)
  250. {
  251. sctp_xmit_t retval = SCTP_XMIT_OK;
  252. __u16 chunk_len = SCTP_PAD4(ntohs(chunk->chunk_hdr->length));
  253. /* Check to see if this chunk will fit into the packet */
  254. retval = sctp_packet_will_fit(packet, chunk, chunk_len);
  255. if (retval != SCTP_XMIT_OK)
  256. goto finish;
  257. /* We believe that this chunk is OK to add to the packet */
  258. switch (chunk->chunk_hdr->type) {
  259. case SCTP_CID_DATA:
  260. /* Account for the data being in the packet */
  261. sctp_packet_append_data(packet, chunk);
  262. /* Disallow SACK bundling after DATA. */
  263. packet->has_sack = 1;
  264. /* Disallow AUTH bundling after DATA */
  265. packet->has_auth = 1;
  266. /* Let it be knows that packet has DATA in it */
  267. packet->has_data = 1;
  268. /* timestamp the chunk for rtx purposes */
  269. chunk->sent_at = jiffies;
  270. /* Mainly used for prsctp RTX policy */
  271. chunk->sent_count++;
  272. break;
  273. case SCTP_CID_COOKIE_ECHO:
  274. packet->has_cookie_echo = 1;
  275. break;
  276. case SCTP_CID_SACK:
  277. packet->has_sack = 1;
  278. if (chunk->asoc)
  279. chunk->asoc->stats.osacks++;
  280. break;
  281. case SCTP_CID_AUTH:
  282. packet->has_auth = 1;
  283. packet->auth = chunk;
  284. break;
  285. }
  286. /* It is OK to send this chunk. */
  287. list_add_tail(&chunk->list, &packet->chunk_list);
  288. packet->size += chunk_len;
  289. chunk->transport = packet->transport;
  290. finish:
  291. return retval;
  292. }
  293. /* Append a chunk to the offered packet reporting back any inability to do
  294. * so.
  295. */
  296. sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
  297. struct sctp_chunk *chunk)
  298. {
  299. sctp_xmit_t retval = SCTP_XMIT_OK;
  300. pr_debug("%s: packet:%p chunk:%p\n", __func__, packet, chunk);
  301. /* Data chunks are special. Before seeing what else we can
  302. * bundle into this packet, check to see if we are allowed to
  303. * send this DATA.
  304. */
  305. if (sctp_chunk_is_data(chunk)) {
  306. retval = sctp_packet_can_append_data(packet, chunk);
  307. if (retval != SCTP_XMIT_OK)
  308. goto finish;
  309. }
  310. /* Try to bundle AUTH chunk */
  311. retval = sctp_packet_bundle_auth(packet, chunk);
  312. if (retval != SCTP_XMIT_OK)
  313. goto finish;
  314. /* Try to bundle SACK chunk */
  315. retval = sctp_packet_bundle_sack(packet, chunk);
  316. if (retval != SCTP_XMIT_OK)
  317. goto finish;
  318. retval = __sctp_packet_append_chunk(packet, chunk);
  319. finish:
  320. return retval;
  321. }
  322. static void sctp_packet_release_owner(struct sk_buff *skb)
  323. {
  324. sk_free(skb->sk);
  325. }
  326. static void sctp_packet_set_owner_w(struct sk_buff *skb, struct sock *sk)
  327. {
  328. skb_orphan(skb);
  329. skb->sk = sk;
  330. skb->destructor = sctp_packet_release_owner;
  331. /*
  332. * The data chunks have already been accounted for in sctp_sendmsg(),
  333. * therefore only reserve a single byte to keep socket around until
  334. * the packet has been transmitted.
  335. */
  336. atomic_inc(&sk->sk_wmem_alloc);
  337. }
  338. static int sctp_packet_pack(struct sctp_packet *packet,
  339. struct sk_buff *head, int gso, gfp_t gfp)
  340. {
  341. struct sctp_transport *tp = packet->transport;
  342. struct sctp_auth_chunk *auth = NULL;
  343. struct sctp_chunk *chunk, *tmp;
  344. int pkt_count = 0, pkt_size;
  345. struct sock *sk = head->sk;
  346. struct sk_buff *nskb;
  347. int auth_len = 0;
  348. if (gso) {
  349. skb_shinfo(head)->gso_type = sk->sk_gso_type;
  350. NAPI_GRO_CB(head)->last = head;
  351. } else {
  352. nskb = head;
  353. pkt_size = packet->size;
  354. goto merge;
  355. }
  356. do {
  357. /* calculate the pkt_size and alloc nskb */
  358. pkt_size = packet->overhead;
  359. list_for_each_entry_safe(chunk, tmp, &packet->chunk_list,
  360. list) {
  361. int padded = SCTP_PAD4(chunk->skb->len);
  362. if (chunk == packet->auth)
  363. auth_len = padded;
  364. else if (auth_len + padded + packet->overhead >
  365. tp->pathmtu)
  366. return 0;
  367. else if (pkt_size + padded > tp->pathmtu)
  368. break;
  369. pkt_size += padded;
  370. }
  371. nskb = alloc_skb(pkt_size + MAX_HEADER, gfp);
  372. if (!nskb)
  373. return 0;
  374. skb_reserve(nskb, packet->overhead + MAX_HEADER);
  375. merge:
  376. /* merge chunks into nskb and append nskb into head list */
  377. pkt_size -= packet->overhead;
  378. list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
  379. int padding;
  380. list_del_init(&chunk->list);
  381. if (sctp_chunk_is_data(chunk)) {
  382. if (!sctp_chunk_retransmitted(chunk) &&
  383. !tp->rto_pending) {
  384. chunk->rtt_in_progress = 1;
  385. tp->rto_pending = 1;
  386. }
  387. }
  388. padding = SCTP_PAD4(chunk->skb->len) - chunk->skb->len;
  389. if (padding)
  390. memset(skb_put(chunk->skb, padding), 0, padding);
  391. if (chunk == packet->auth)
  392. auth = (struct sctp_auth_chunk *)
  393. skb_tail_pointer(nskb);
  394. memcpy(skb_put(nskb, chunk->skb->len), chunk->skb->data,
  395. chunk->skb->len);
  396. pr_debug("*** Chunk:%p[%s] %s 0x%x, length:%d, chunk->skb->len:%d, rtt_in_progress:%d\n",
  397. chunk,
  398. sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)),
  399. chunk->has_tsn ? "TSN" : "No TSN",
  400. chunk->has_tsn ? ntohl(chunk->subh.data_hdr->tsn) : 0,
  401. ntohs(chunk->chunk_hdr->length), chunk->skb->len,
  402. chunk->rtt_in_progress);
  403. pkt_size -= SCTP_PAD4(chunk->skb->len);
  404. if (!sctp_chunk_is_data(chunk) && chunk != packet->auth)
  405. sctp_chunk_free(chunk);
  406. if (!pkt_size)
  407. break;
  408. }
  409. if (auth) {
  410. sctp_auth_calculate_hmac(tp->asoc, nskb, auth, gfp);
  411. /* free auth if no more chunks, or add it back */
  412. if (list_empty(&packet->chunk_list))
  413. sctp_chunk_free(packet->auth);
  414. else
  415. list_add(&packet->auth->list,
  416. &packet->chunk_list);
  417. }
  418. if (gso) {
  419. if (skb_gro_receive(&head, nskb)) {
  420. kfree_skb(nskb);
  421. return 0;
  422. }
  423. if (WARN_ON_ONCE(skb_shinfo(head)->gso_segs >=
  424. sk->sk_gso_max_segs))
  425. return 0;
  426. }
  427. pkt_count++;
  428. } while (!list_empty(&packet->chunk_list));
  429. if (gso) {
  430. memset(head->cb, 0, max(sizeof(struct inet_skb_parm),
  431. sizeof(struct inet6_skb_parm)));
  432. skb_shinfo(head)->gso_segs = pkt_count;
  433. skb_shinfo(head)->gso_size = GSO_BY_FRAGS;
  434. rcu_read_lock();
  435. if (skb_dst(head) != tp->dst) {
  436. dst_hold(tp->dst);
  437. sk_setup_caps(sk, tp->dst);
  438. }
  439. rcu_read_unlock();
  440. goto chksum;
  441. }
  442. if (sctp_checksum_disable)
  443. return 1;
  444. if (!(skb_dst(head)->dev->features & NETIF_F_SCTP_CRC) ||
  445. dst_xfrm(skb_dst(head)) || packet->ipfragok) {
  446. struct sctphdr *sh =
  447. (struct sctphdr *)skb_transport_header(head);
  448. sh->checksum = sctp_compute_cksum(head, 0);
  449. } else {
  450. chksum:
  451. head->ip_summed = CHECKSUM_PARTIAL;
  452. head->csum_start = skb_transport_header(head) - head->head;
  453. head->csum_offset = offsetof(struct sctphdr, checksum);
  454. }
  455. return pkt_count;
  456. }
  457. /* All packets are sent to the network through this function from
  458. * sctp_outq_tail().
  459. *
  460. * The return value is always 0 for now.
  461. */
  462. int sctp_packet_transmit(struct sctp_packet *packet, gfp_t gfp)
  463. {
  464. struct sctp_transport *tp = packet->transport;
  465. struct sctp_association *asoc = tp->asoc;
  466. struct sctp_chunk *chunk, *tmp;
  467. int pkt_count, gso = 0;
  468. int confirm;
  469. struct dst_entry *dst;
  470. struct sk_buff *head;
  471. struct sctphdr *sh;
  472. struct sock *sk;
  473. pr_debug("%s: packet:%p\n", __func__, packet);
  474. if (list_empty(&packet->chunk_list))
  475. return 0;
  476. chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
  477. sk = chunk->skb->sk;
  478. /* check gso */
  479. if (packet->size > tp->pathmtu && !packet->ipfragok) {
  480. if (!sk_can_gso(sk)) {
  481. pr_err_once("Trying to GSO but underlying device doesn't support it.");
  482. goto out;
  483. }
  484. gso = 1;
  485. }
  486. /* alloc head skb */
  487. head = alloc_skb((gso ? packet->overhead : packet->size) +
  488. MAX_HEADER, gfp);
  489. if (!head)
  490. goto out;
  491. skb_reserve(head, packet->overhead + MAX_HEADER);
  492. sctp_packet_set_owner_w(head, sk);
  493. /* set sctp header */
  494. sh = (struct sctphdr *)skb_push(head, sizeof(struct sctphdr));
  495. skb_reset_transport_header(head);
  496. sh->source = htons(packet->source_port);
  497. sh->dest = htons(packet->destination_port);
  498. sh->vtag = htonl(packet->vtag);
  499. sh->checksum = 0;
  500. /* update dst if in need */
  501. if (!sctp_transport_dst_check(tp)) {
  502. sctp_transport_route(tp, NULL, sctp_sk(sk));
  503. if (asoc && asoc->param_flags & SPP_PMTUD_ENABLE)
  504. sctp_assoc_sync_pmtu(sk, asoc);
  505. }
  506. dst = dst_clone(tp->dst);
  507. if (!dst) {
  508. IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
  509. kfree_skb(head);
  510. goto out;
  511. }
  512. skb_dst_set(head, dst);
  513. /* pack up chunks */
  514. pkt_count = sctp_packet_pack(packet, head, gso, gfp);
  515. if (!pkt_count) {
  516. kfree_skb(head);
  517. goto out;
  518. }
  519. pr_debug("***sctp_transmit_packet*** skb->len:%d\n", head->len);
  520. /* start autoclose timer */
  521. if (packet->has_data && sctp_state(asoc, ESTABLISHED) &&
  522. asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE]) {
  523. struct timer_list *timer =
  524. &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
  525. unsigned long timeout =
  526. asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
  527. if (!mod_timer(timer, jiffies + timeout))
  528. sctp_association_hold(asoc);
  529. }
  530. /* sctp xmit */
  531. tp->af_specific->ecn_capable(sk);
  532. if (asoc) {
  533. asoc->stats.opackets += pkt_count;
  534. if (asoc->peer.last_sent_to != tp)
  535. asoc->peer.last_sent_to = tp;
  536. }
  537. head->ignore_df = packet->ipfragok;
  538. confirm = tp->dst_pending_confirm;
  539. if (confirm)
  540. skb_set_dst_pending_confirm(head, 1);
  541. /* neighbour should be confirmed on successful transmission or
  542. * positive error
  543. */
  544. if (tp->af_specific->sctp_xmit(head, tp) >= 0 && confirm)
  545. tp->dst_pending_confirm = 0;
  546. out:
  547. list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
  548. list_del_init(&chunk->list);
  549. if (!sctp_chunk_is_data(chunk))
  550. sctp_chunk_free(chunk);
  551. }
  552. sctp_packet_reset(packet);
  553. return 0;
  554. }
  555. /********************************************************************
  556. * 2nd Level Abstractions
  557. ********************************************************************/
  558. /* This private function check to see if a chunk can be added */
  559. static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
  560. struct sctp_chunk *chunk)
  561. {
  562. size_t datasize, rwnd, inflight, flight_size;
  563. struct sctp_transport *transport = packet->transport;
  564. struct sctp_association *asoc = transport->asoc;
  565. struct sctp_outq *q = &asoc->outqueue;
  566. /* RFC 2960 6.1 Transmission of DATA Chunks
  567. *
  568. * A) At any given time, the data sender MUST NOT transmit new data to
  569. * any destination transport address if its peer's rwnd indicates
  570. * that the peer has no buffer space (i.e. rwnd is 0, see Section
  571. * 6.2.1). However, regardless of the value of rwnd (including if it
  572. * is 0), the data sender can always have one DATA chunk in flight to
  573. * the receiver if allowed by cwnd (see rule B below). This rule
  574. * allows the sender to probe for a change in rwnd that the sender
  575. * missed due to the SACK having been lost in transit from the data
  576. * receiver to the data sender.
  577. */
  578. rwnd = asoc->peer.rwnd;
  579. inflight = q->outstanding_bytes;
  580. flight_size = transport->flight_size;
  581. datasize = sctp_data_size(chunk);
  582. if (datasize > rwnd && inflight > 0)
  583. /* We have (at least) one data chunk in flight,
  584. * so we can't fall back to rule 6.1 B).
  585. */
  586. return SCTP_XMIT_RWND_FULL;
  587. /* RFC 2960 6.1 Transmission of DATA Chunks
  588. *
  589. * B) At any given time, the sender MUST NOT transmit new data
  590. * to a given transport address if it has cwnd or more bytes
  591. * of data outstanding to that transport address.
  592. */
  593. /* RFC 7.2.4 & the Implementers Guide 2.8.
  594. *
  595. * 3) ...
  596. * When a Fast Retransmit is being performed the sender SHOULD
  597. * ignore the value of cwnd and SHOULD NOT delay retransmission.
  598. */
  599. if (chunk->fast_retransmit != SCTP_NEED_FRTX &&
  600. flight_size >= transport->cwnd)
  601. return SCTP_XMIT_RWND_FULL;
  602. /* Nagle's algorithm to solve small-packet problem:
  603. * Inhibit the sending of new chunks when new outgoing data arrives
  604. * if any previously transmitted data on the connection remains
  605. * unacknowledged.
  606. */
  607. if ((sctp_sk(asoc->base.sk)->nodelay || inflight == 0) &&
  608. !chunk->msg->force_delay)
  609. /* Nothing unacked */
  610. return SCTP_XMIT_OK;
  611. if (!sctp_packet_empty(packet))
  612. /* Append to packet */
  613. return SCTP_XMIT_OK;
  614. if (!sctp_state(asoc, ESTABLISHED))
  615. return SCTP_XMIT_OK;
  616. /* Check whether this chunk and all the rest of pending data will fit
  617. * or delay in hopes of bundling a full sized packet.
  618. */
  619. if (chunk->skb->len + q->out_qlen >
  620. transport->pathmtu - packet->overhead - sizeof(sctp_data_chunk_t) - 4)
  621. /* Enough data queued to fill a packet */
  622. return SCTP_XMIT_OK;
  623. /* Don't delay large message writes that may have been fragmented */
  624. if (!chunk->msg->can_delay)
  625. return SCTP_XMIT_OK;
  626. /* Defer until all data acked or packet full */
  627. return SCTP_XMIT_DELAY;
  628. }
  629. /* This private function does management things when adding DATA chunk */
  630. static void sctp_packet_append_data(struct sctp_packet *packet,
  631. struct sctp_chunk *chunk)
  632. {
  633. struct sctp_transport *transport = packet->transport;
  634. size_t datasize = sctp_data_size(chunk);
  635. struct sctp_association *asoc = transport->asoc;
  636. u32 rwnd = asoc->peer.rwnd;
  637. /* Keep track of how many bytes are in flight over this transport. */
  638. transport->flight_size += datasize;
  639. /* Keep track of how many bytes are in flight to the receiver. */
  640. asoc->outqueue.outstanding_bytes += datasize;
  641. /* Update our view of the receiver's rwnd. */
  642. if (datasize < rwnd)
  643. rwnd -= datasize;
  644. else
  645. rwnd = 0;
  646. asoc->peer.rwnd = rwnd;
  647. sctp_chunk_assign_tsn(chunk);
  648. sctp_chunk_assign_ssn(chunk);
  649. }
  650. static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
  651. struct sctp_chunk *chunk,
  652. u16 chunk_len)
  653. {
  654. size_t psize, pmtu, maxsize;
  655. sctp_xmit_t retval = SCTP_XMIT_OK;
  656. psize = packet->size;
  657. if (packet->transport->asoc)
  658. pmtu = packet->transport->asoc->pathmtu;
  659. else
  660. pmtu = packet->transport->pathmtu;
  661. /* Decide if we need to fragment or resubmit later. */
  662. if (psize + chunk_len > pmtu) {
  663. /* It's OK to fragment at IP level if any one of the following
  664. * is true:
  665. * 1. The packet is empty (meaning this chunk is greater
  666. * the MTU)
  667. * 2. The packet doesn't have any data in it yet and data
  668. * requires authentication.
  669. */
  670. if (sctp_packet_empty(packet) ||
  671. (!packet->has_data && chunk->auth)) {
  672. /* We no longer do re-fragmentation.
  673. * Just fragment at the IP layer, if we
  674. * actually hit this condition
  675. */
  676. packet->ipfragok = 1;
  677. goto out;
  678. }
  679. /* Similarly, if this chunk was built before a PMTU
  680. * reduction, we have to fragment it at IP level now. So
  681. * if the packet already contains something, we need to
  682. * flush.
  683. */
  684. maxsize = pmtu - packet->overhead;
  685. if (packet->auth)
  686. maxsize -= SCTP_PAD4(packet->auth->skb->len);
  687. if (chunk_len > maxsize)
  688. retval = SCTP_XMIT_PMTU_FULL;
  689. /* It is also okay to fragment if the chunk we are
  690. * adding is a control chunk, but only if current packet
  691. * is not a GSO one otherwise it causes fragmentation of
  692. * a large frame. So in this case we allow the
  693. * fragmentation by forcing it to be in a new packet.
  694. */
  695. if (!sctp_chunk_is_data(chunk) && packet->has_data)
  696. retval = SCTP_XMIT_PMTU_FULL;
  697. if (psize + chunk_len > packet->max_size)
  698. /* Hit GSO/PMTU limit, gotta flush */
  699. retval = SCTP_XMIT_PMTU_FULL;
  700. if (!packet->transport->burst_limited &&
  701. psize + chunk_len > (packet->transport->cwnd >> 1))
  702. /* Do not allow a single GSO packet to use more
  703. * than half of cwnd.
  704. */
  705. retval = SCTP_XMIT_PMTU_FULL;
  706. if (packet->transport->burst_limited &&
  707. psize + chunk_len > (packet->transport->burst_limited >> 1))
  708. /* Do not allow a single GSO packet to use more
  709. * than half of original cwnd.
  710. */
  711. retval = SCTP_XMIT_PMTU_FULL;
  712. /* Otherwise it will fit in the GSO packet */
  713. }
  714. out:
  715. return retval;
  716. }