associola.c 48 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711
  1. // SPDX-License-Identifier: GPL-2.0-or-later
  2. /* SCTP kernel implementation
  3. * (C) Copyright IBM Corp. 2001, 2004
  4. * Copyright (c) 1999-2000 Cisco, Inc.
  5. * Copyright (c) 1999-2001 Motorola, Inc.
  6. * Copyright (c) 2001 Intel Corp.
  7. * Copyright (c) 2001 La Monte H.P. Yarroll
  8. *
  9. * This file is part of the SCTP kernel implementation
  10. *
  11. * This module provides the abstraction for an SCTP association.
  12. *
  13. * Please send any bug reports or fixes you make to the
  14. * email address(es):
  15. * lksctp developers <linux-sctp@vger.kernel.org>
  16. *
  17. * Written or modified by:
  18. * La Monte H.P. Yarroll <piggy@acm.org>
  19. * Karl Knutson <karl@athena.chicago.il.us>
  20. * Jon Grimm <jgrimm@us.ibm.com>
  21. * Xingang Guo <xingang.guo@intel.com>
  22. * Hui Huang <hui.huang@nokia.com>
  23. * Sridhar Samudrala <sri@us.ibm.com>
  24. * Daisy Chang <daisyc@us.ibm.com>
  25. * Ryan Layer <rmlayer@us.ibm.com>
  26. * Kevin Gao <kevin.gao@intel.com>
  27. */
  28. #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  29. #include <linux/types.h>
  30. #include <linux/fcntl.h>
  31. #include <linux/poll.h>
  32. #include <linux/init.h>
  33. #include <linux/slab.h>
  34. #include <linux/in.h>
  35. #include <net/ipv6.h>
  36. #include <net/sctp/sctp.h>
  37. #include <net/sctp/sm.h>
  38. /* Forward declarations for internal functions. */
  39. static void sctp_select_active_and_retran_path(struct sctp_association *asoc);
  40. static void sctp_assoc_bh_rcv(struct work_struct *work);
  41. static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc);
  42. static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc);
  43. /* 1st Level Abstractions. */
  44. /* Initialize a new association from provided memory. */
  45. static struct sctp_association *sctp_association_init(
  46. struct sctp_association *asoc,
  47. const struct sctp_endpoint *ep,
  48. const struct sock *sk,
  49. enum sctp_scope scope, gfp_t gfp)
  50. {
  51. struct sctp_sock *sp;
  52. struct sctp_paramhdr *p;
  53. int i;
  54. /* Retrieve the SCTP per socket area. */
  55. sp = sctp_sk((struct sock *)sk);
  56. /* Discarding const is appropriate here. */
  57. asoc->ep = (struct sctp_endpoint *)ep;
  58. asoc->base.sk = (struct sock *)sk;
  59. asoc->base.net = sock_net(sk);
  60. sctp_endpoint_hold(asoc->ep);
  61. sock_hold(asoc->base.sk);
  62. /* Initialize the common base substructure. */
  63. asoc->base.type = SCTP_EP_TYPE_ASSOCIATION;
  64. /* Initialize the object handling fields. */
  65. refcount_set(&asoc->base.refcnt, 1);
  66. /* Initialize the bind addr area. */
  67. sctp_bind_addr_init(&asoc->base.bind_addr, ep->base.bind_addr.port);
  68. asoc->state = SCTP_STATE_CLOSED;
  69. asoc->cookie_life = ms_to_ktime(sp->assocparams.sasoc_cookie_life);
  70. asoc->user_frag = sp->user_frag;
  71. /* Set the association max_retrans and RTO values from the
  72. * socket values.
  73. */
  74. asoc->max_retrans = sp->assocparams.sasoc_asocmaxrxt;
  75. asoc->pf_retrans = sp->pf_retrans;
  76. asoc->ps_retrans = sp->ps_retrans;
  77. asoc->pf_expose = sp->pf_expose;
  78. asoc->rto_initial = msecs_to_jiffies(sp->rtoinfo.srto_initial);
  79. asoc->rto_max = msecs_to_jiffies(sp->rtoinfo.srto_max);
  80. asoc->rto_min = msecs_to_jiffies(sp->rtoinfo.srto_min);
  81. /* Initialize the association's heartbeat interval based on the
  82. * sock configured value.
  83. */
  84. asoc->hbinterval = msecs_to_jiffies(sp->hbinterval);
  85. asoc->probe_interval = msecs_to_jiffies(sp->probe_interval);
  86. asoc->encap_port = sp->encap_port;
  87. /* Initialize path max retrans value. */
  88. asoc->pathmaxrxt = sp->pathmaxrxt;
  89. asoc->flowlabel = sp->flowlabel;
  90. asoc->dscp = sp->dscp;
  91. /* Set association default SACK delay */
  92. asoc->sackdelay = msecs_to_jiffies(sp->sackdelay);
  93. asoc->sackfreq = sp->sackfreq;
  94. /* Set the association default flags controlling
  95. * Heartbeat, SACK delay, and Path MTU Discovery.
  96. */
  97. asoc->param_flags = sp->param_flags;
  98. /* Initialize the maximum number of new data packets that can be sent
  99. * in a burst.
  100. */
  101. asoc->max_burst = sp->max_burst;
  102. asoc->subscribe = sp->subscribe;
  103. /* initialize association timers */
  104. asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_COOKIE] = asoc->rto_initial;
  105. asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_INIT] = asoc->rto_initial;
  106. asoc->timeouts[SCTP_EVENT_TIMEOUT_T2_SHUTDOWN] = asoc->rto_initial;
  107. /* sctpimpguide Section 2.12.2
  108. * If the 'T5-shutdown-guard' timer is used, it SHOULD be set to the
  109. * recommended value of 5 times 'RTO.Max'.
  110. */
  111. asoc->timeouts[SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]
  112. = 5 * asoc->rto_max;
  113. asoc->timeouts[SCTP_EVENT_TIMEOUT_SACK] = asoc->sackdelay;
  114. asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE] =
  115. (unsigned long)sp->autoclose * HZ;
  116. /* Initializes the timers */
  117. for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i)
  118. timer_setup(&asoc->timers[i], sctp_timer_events[i], 0);
  119. /* Pull default initialization values from the sock options.
  120. * Note: This assumes that the values have already been
  121. * validated in the sock.
  122. */
  123. asoc->c.sinit_max_instreams = sp->initmsg.sinit_max_instreams;
  124. asoc->c.sinit_num_ostreams = sp->initmsg.sinit_num_ostreams;
  125. asoc->max_init_attempts = sp->initmsg.sinit_max_attempts;
  126. asoc->max_init_timeo =
  127. msecs_to_jiffies(sp->initmsg.sinit_max_init_timeo);
  128. /* Set the local window size for receive.
  129. * This is also the rcvbuf space per association.
  130. * RFC 6 - A SCTP receiver MUST be able to receive a minimum of
  131. * 1500 bytes in one SCTP packet.
  132. */
  133. if ((sk->sk_rcvbuf/2) < SCTP_DEFAULT_MINWINDOW)
  134. asoc->rwnd = SCTP_DEFAULT_MINWINDOW;
  135. else
  136. asoc->rwnd = sk->sk_rcvbuf/2;
  137. asoc->a_rwnd = asoc->rwnd;
  138. /* Use my own max window until I learn something better. */
  139. asoc->peer.rwnd = SCTP_DEFAULT_MAXWINDOW;
  140. /* Initialize the receive memory counter */
  141. atomic_set(&asoc->rmem_alloc, 0);
  142. init_waitqueue_head(&asoc->wait);
  143. asoc->c.my_vtag = sctp_generate_tag(ep);
  144. asoc->c.my_port = ep->base.bind_addr.port;
  145. asoc->c.initial_tsn = sctp_generate_tsn(ep);
  146. asoc->next_tsn = asoc->c.initial_tsn;
  147. asoc->ctsn_ack_point = asoc->next_tsn - 1;
  148. asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
  149. asoc->highest_sacked = asoc->ctsn_ack_point;
  150. asoc->last_cwr_tsn = asoc->ctsn_ack_point;
  151. /* ADDIP Section 4.1 Asconf Chunk Procedures
  152. *
  153. * When an endpoint has an ASCONF signaled change to be sent to the
  154. * remote endpoint it should do the following:
  155. * ...
  156. * A2) a serial number should be assigned to the chunk. The serial
  157. * number SHOULD be a monotonically increasing number. The serial
  158. * numbers SHOULD be initialized at the start of the
  159. * association to the same value as the initial TSN.
  160. */
  161. asoc->addip_serial = asoc->c.initial_tsn;
  162. asoc->strreset_outseq = asoc->c.initial_tsn;
  163. INIT_LIST_HEAD(&asoc->addip_chunk_list);
  164. INIT_LIST_HEAD(&asoc->asconf_ack_list);
  165. /* Make an empty list of remote transport addresses. */
  166. INIT_LIST_HEAD(&asoc->peer.transport_addr_list);
  167. /* RFC 2960 5.1 Normal Establishment of an Association
  168. *
  169. * After the reception of the first data chunk in an
  170. * association the endpoint must immediately respond with a
  171. * sack to acknowledge the data chunk. Subsequent
  172. * acknowledgements should be done as described in Section
  173. * 6.2.
  174. *
  175. * [We implement this by telling a new association that it
  176. * already received one packet.]
  177. */
  178. asoc->peer.sack_needed = 1;
  179. asoc->peer.sack_generation = 1;
  180. /* Create an input queue. */
  181. sctp_inq_init(&asoc->base.inqueue);
  182. sctp_inq_set_th_handler(&asoc->base.inqueue, sctp_assoc_bh_rcv);
  183. /* Create an output queue. */
  184. sctp_outq_init(asoc, &asoc->outqueue);
  185. sctp_ulpq_init(&asoc->ulpq, asoc);
  186. if (sctp_stream_init(&asoc->stream, asoc->c.sinit_num_ostreams, 0, gfp))
  187. goto stream_free;
  188. /* Initialize default path MTU. */
  189. asoc->pathmtu = sp->pathmtu;
  190. sctp_assoc_update_frag_point(asoc);
  191. /* Assume that peer would support both address types unless we are
  192. * told otherwise.
  193. */
  194. asoc->peer.ipv4_address = 1;
  195. if (asoc->base.sk->sk_family == PF_INET6)
  196. asoc->peer.ipv6_address = 1;
  197. INIT_LIST_HEAD(&asoc->asocs);
  198. asoc->default_stream = sp->default_stream;
  199. asoc->default_ppid = sp->default_ppid;
  200. asoc->default_flags = sp->default_flags;
  201. asoc->default_context = sp->default_context;
  202. asoc->default_timetolive = sp->default_timetolive;
  203. asoc->default_rcv_context = sp->default_rcv_context;
  204. /* AUTH related initializations */
  205. INIT_LIST_HEAD(&asoc->endpoint_shared_keys);
  206. if (sctp_auth_asoc_copy_shkeys(ep, asoc, gfp))
  207. goto stream_free;
  208. asoc->active_key_id = ep->active_key_id;
  209. asoc->strreset_enable = ep->strreset_enable;
  210. /* Save the hmacs and chunks list into this association */
  211. if (ep->auth_hmacs_list)
  212. memcpy(asoc->c.auth_hmacs, ep->auth_hmacs_list,
  213. ntohs(ep->auth_hmacs_list->param_hdr.length));
  214. if (ep->auth_chunk_list)
  215. memcpy(asoc->c.auth_chunks, ep->auth_chunk_list,
  216. ntohs(ep->auth_chunk_list->param_hdr.length));
  217. /* Get the AUTH random number for this association */
  218. p = (struct sctp_paramhdr *)asoc->c.auth_random;
  219. p->type = SCTP_PARAM_RANDOM;
  220. p->length = htons(sizeof(*p) + SCTP_AUTH_RANDOM_LENGTH);
  221. get_random_bytes(p+1, SCTP_AUTH_RANDOM_LENGTH);
  222. return asoc;
  223. stream_free:
  224. sctp_stream_free(&asoc->stream);
  225. sock_put(asoc->base.sk);
  226. sctp_endpoint_put(asoc->ep);
  227. return NULL;
  228. }
  229. /* Allocate and initialize a new association */
  230. struct sctp_association *sctp_association_new(const struct sctp_endpoint *ep,
  231. const struct sock *sk,
  232. enum sctp_scope scope, gfp_t gfp)
  233. {
  234. struct sctp_association *asoc;
  235. asoc = kzalloc_obj(*asoc, gfp);
  236. if (!asoc)
  237. goto fail;
  238. if (!sctp_association_init(asoc, ep, sk, scope, gfp))
  239. goto fail_init;
  240. SCTP_DBG_OBJCNT_INC(assoc);
  241. pr_debug("Created asoc %p\n", asoc);
  242. return asoc;
  243. fail_init:
  244. kfree(asoc);
  245. fail:
  246. return NULL;
  247. }
  248. /* Free this association if possible. There may still be users, so
  249. * the actual deallocation may be delayed.
  250. */
  251. void sctp_association_free(struct sctp_association *asoc)
  252. {
  253. struct sock *sk = asoc->base.sk;
  254. struct sctp_transport *transport;
  255. struct list_head *pos, *temp;
  256. int i;
  257. /* Only real associations count against the endpoint, so
  258. * don't bother for if this is a temporary association.
  259. */
  260. if (!list_empty(&asoc->asocs)) {
  261. list_del(&asoc->asocs);
  262. /* Decrement the backlog value for a TCP-style listening
  263. * socket.
  264. */
  265. if (sctp_style(sk, TCP) && sctp_sstate(sk, LISTENING))
  266. sk_acceptq_removed(sk);
  267. }
  268. /* Mark as dead, so other users can know this structure is
  269. * going away.
  270. */
  271. asoc->base.dead = true;
  272. /* Dispose of any data lying around in the outqueue. */
  273. sctp_outq_free(&asoc->outqueue);
  274. /* Dispose of any pending messages for the upper layer. */
  275. sctp_ulpq_free(&asoc->ulpq);
  276. /* Dispose of any pending chunks on the inqueue. */
  277. sctp_inq_free(&asoc->base.inqueue);
  278. sctp_tsnmap_free(&asoc->peer.tsn_map);
  279. /* Free stream information. */
  280. sctp_stream_free(&asoc->stream);
  281. if (asoc->strreset_chunk)
  282. sctp_chunk_free(asoc->strreset_chunk);
  283. /* Clean up the bound address list. */
  284. sctp_bind_addr_free(&asoc->base.bind_addr);
  285. /* Do we need to go through all of our timers and
  286. * delete them? To be safe we will try to delete all, but we
  287. * should be able to go through and make a guess based
  288. * on our state.
  289. */
  290. for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) {
  291. if (timer_delete(&asoc->timers[i]))
  292. sctp_association_put(asoc);
  293. }
  294. /* Free peer's cached cookie. */
  295. kfree(asoc->peer.cookie);
  296. kfree(asoc->peer.peer_random);
  297. kfree(asoc->peer.peer_chunks);
  298. kfree(asoc->peer.peer_hmacs);
  299. /* Release the transport structures. */
  300. list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
  301. transport = list_entry(pos, struct sctp_transport, transports);
  302. list_del_rcu(pos);
  303. sctp_unhash_transport(transport);
  304. sctp_transport_free(transport);
  305. }
  306. asoc->peer.transport_count = 0;
  307. sctp_asconf_queue_teardown(asoc);
  308. /* Free pending address space being deleted */
  309. kfree(asoc->asconf_addr_del_pending);
  310. /* AUTH - Free the endpoint shared keys */
  311. sctp_auth_destroy_keys(&asoc->endpoint_shared_keys);
  312. /* AUTH - Free the association shared key */
  313. sctp_auth_key_put(asoc->asoc_shared_key);
  314. sctp_association_put(asoc);
  315. }
  316. /* Cleanup and free up an association. */
  317. static void sctp_association_destroy(struct sctp_association *asoc)
  318. {
  319. if (unlikely(!asoc->base.dead)) {
  320. WARN(1, "Attempt to destroy undead association %p!\n", asoc);
  321. return;
  322. }
  323. sctp_endpoint_put(asoc->ep);
  324. sock_put(asoc->base.sk);
  325. if (asoc->assoc_id != 0) {
  326. spin_lock_bh(&sctp_assocs_id_lock);
  327. idr_remove(&sctp_assocs_id, asoc->assoc_id);
  328. spin_unlock_bh(&sctp_assocs_id_lock);
  329. }
  330. WARN_ON(atomic_read(&asoc->rmem_alloc));
  331. kfree_rcu(asoc, rcu);
  332. SCTP_DBG_OBJCNT_DEC(assoc);
  333. }
  334. /* Change the primary destination address for the peer. */
  335. void sctp_assoc_set_primary(struct sctp_association *asoc,
  336. struct sctp_transport *transport)
  337. {
  338. int changeover = 0;
  339. /* it's a changeover only if we already have a primary path
  340. * that we are changing
  341. */
  342. if (asoc->peer.primary_path != NULL &&
  343. asoc->peer.primary_path != transport)
  344. changeover = 1 ;
  345. asoc->peer.primary_path = transport;
  346. sctp_ulpevent_notify_peer_addr_change(transport,
  347. SCTP_ADDR_MADE_PRIM, 0);
  348. /* Set a default msg_name for events. */
  349. memcpy(&asoc->peer.primary_addr, &transport->ipaddr,
  350. sizeof(union sctp_addr));
  351. /* If the primary path is changing, assume that the
  352. * user wants to use this new path.
  353. */
  354. if ((transport->state == SCTP_ACTIVE) ||
  355. (transport->state == SCTP_UNKNOWN))
  356. asoc->peer.active_path = transport;
  357. /*
  358. * SFR-CACC algorithm:
  359. * Upon the receipt of a request to change the primary
  360. * destination address, on the data structure for the new
  361. * primary destination, the sender MUST do the following:
  362. *
  363. * 1) If CHANGEOVER_ACTIVE is set, then there was a switch
  364. * to this destination address earlier. The sender MUST set
  365. * CYCLING_CHANGEOVER to indicate that this switch is a
  366. * double switch to the same destination address.
  367. *
  368. * Really, only bother is we have data queued or outstanding on
  369. * the association.
  370. */
  371. if (!asoc->outqueue.outstanding_bytes && !asoc->outqueue.out_qlen)
  372. return;
  373. if (transport->cacc.changeover_active)
  374. transport->cacc.cycling_changeover = changeover;
  375. /* 2) The sender MUST set CHANGEOVER_ACTIVE to indicate that
  376. * a changeover has occurred.
  377. */
  378. transport->cacc.changeover_active = changeover;
  379. /* 3) The sender MUST store the next TSN to be sent in
  380. * next_tsn_at_change.
  381. */
  382. transport->cacc.next_tsn_at_change = asoc->next_tsn;
  383. }
  384. /* Remove a transport from an association. */
  385. void sctp_assoc_rm_peer(struct sctp_association *asoc,
  386. struct sctp_transport *peer)
  387. {
  388. struct sctp_transport *transport;
  389. struct list_head *pos;
  390. struct sctp_chunk *ch;
  391. pr_debug("%s: association:%p addr:%pISpc\n",
  392. __func__, asoc, &peer->ipaddr.sa);
  393. /* If we are to remove the current retran_path, update it
  394. * to the next peer before removing this peer from the list.
  395. */
  396. if (asoc->peer.retran_path == peer)
  397. sctp_assoc_update_retran_path(asoc);
  398. /* Remove this peer from the list. */
  399. list_del_rcu(&peer->transports);
  400. /* Remove this peer from the transport hashtable */
  401. sctp_unhash_transport(peer);
  402. /* Get the first transport of asoc. */
  403. pos = asoc->peer.transport_addr_list.next;
  404. transport = list_entry(pos, struct sctp_transport, transports);
  405. /* Update any entries that match the peer to be deleted. */
  406. if (asoc->peer.primary_path == peer)
  407. sctp_assoc_set_primary(asoc, transport);
  408. if (asoc->peer.active_path == peer)
  409. asoc->peer.active_path = transport;
  410. if (asoc->peer.retran_path == peer)
  411. asoc->peer.retran_path = transport;
  412. if (asoc->peer.last_data_from == peer)
  413. asoc->peer.last_data_from = transport;
  414. if (asoc->strreset_chunk &&
  415. asoc->strreset_chunk->transport == peer) {
  416. asoc->strreset_chunk->transport = transport;
  417. sctp_transport_reset_reconf_timer(transport);
  418. }
  419. /* If we remove the transport an INIT was last sent to, set it to
  420. * NULL. Combined with the update of the retran path above, this
  421. * will cause the next INIT to be sent to the next available
  422. * transport, maintaining the cycle.
  423. */
  424. if (asoc->init_last_sent_to == peer)
  425. asoc->init_last_sent_to = NULL;
  426. /* If we remove the transport an SHUTDOWN was last sent to, set it
  427. * to NULL. Combined with the update of the retran path above, this
  428. * will cause the next SHUTDOWN to be sent to the next available
  429. * transport, maintaining the cycle.
  430. */
  431. if (asoc->shutdown_last_sent_to == peer)
  432. asoc->shutdown_last_sent_to = NULL;
  433. /* If we remove the transport an ASCONF was last sent to, set it to
  434. * NULL.
  435. */
  436. if (asoc->addip_last_asconf &&
  437. asoc->addip_last_asconf->transport == peer)
  438. asoc->addip_last_asconf->transport = NULL;
  439. /* If we have something on the transmitted list, we have to
  440. * save it off. The best place is the active path.
  441. */
  442. if (!list_empty(&peer->transmitted)) {
  443. struct sctp_transport *active = asoc->peer.active_path;
  444. /* Reset the transport of each chunk on this list */
  445. list_for_each_entry(ch, &peer->transmitted,
  446. transmitted_list) {
  447. ch->transport = NULL;
  448. ch->rtt_in_progress = 0;
  449. }
  450. list_splice_tail_init(&peer->transmitted,
  451. &active->transmitted);
  452. /* Start a T3 timer here in case it wasn't running so
  453. * that these migrated packets have a chance to get
  454. * retransmitted.
  455. */
  456. if (!timer_pending(&active->T3_rtx_timer))
  457. if (!mod_timer(&active->T3_rtx_timer,
  458. jiffies + active->rto))
  459. sctp_transport_hold(active);
  460. }
  461. list_for_each_entry(ch, &asoc->outqueue.out_chunk_list, list)
  462. if (ch->transport == peer)
  463. ch->transport = NULL;
  464. asoc->peer.transport_count--;
  465. sctp_ulpevent_notify_peer_addr_change(peer, SCTP_ADDR_REMOVED, 0);
  466. sctp_transport_free(peer);
  467. }
  468. /* Add a transport address to an association. */
  469. struct sctp_transport *sctp_assoc_add_peer(struct sctp_association *asoc,
  470. const union sctp_addr *addr,
  471. const gfp_t gfp,
  472. const int peer_state)
  473. {
  474. struct sctp_transport *peer;
  475. struct sctp_sock *sp;
  476. unsigned short port;
  477. sp = sctp_sk(asoc->base.sk);
  478. /* AF_INET and AF_INET6 share common port field. */
  479. port = ntohs(addr->v4.sin_port);
  480. pr_debug("%s: association:%p addr:%pISpc state:%d\n", __func__,
  481. asoc, &addr->sa, peer_state);
  482. /* Set the port if it has not been set yet. */
  483. if (0 == asoc->peer.port)
  484. asoc->peer.port = port;
  485. /* Check to see if this is a duplicate. */
  486. peer = sctp_assoc_lookup_paddr(asoc, addr);
  487. if (peer) {
  488. /* An UNKNOWN state is only set on transports added by
  489. * user in sctp_connectx() call. Such transports should be
  490. * considered CONFIRMED per RFC 4960, Section 5.4.
  491. */
  492. if (peer->state == SCTP_UNKNOWN) {
  493. peer->state = SCTP_ACTIVE;
  494. }
  495. return peer;
  496. }
  497. peer = sctp_transport_new(asoc->base.net, addr, gfp);
  498. if (!peer)
  499. return NULL;
  500. sctp_transport_set_owner(peer, asoc);
  501. /* Initialize the peer's heartbeat interval based on the
  502. * association configured value.
  503. */
  504. peer->hbinterval = asoc->hbinterval;
  505. peer->probe_interval = asoc->probe_interval;
  506. peer->encap_port = asoc->encap_port;
  507. /* Set the path max_retrans. */
  508. peer->pathmaxrxt = asoc->pathmaxrxt;
  509. /* And the partial failure retrans threshold */
  510. peer->pf_retrans = asoc->pf_retrans;
  511. /* And the primary path switchover retrans threshold */
  512. peer->ps_retrans = asoc->ps_retrans;
  513. /* Initialize the peer's SACK delay timeout based on the
  514. * association configured value.
  515. */
  516. peer->sackdelay = asoc->sackdelay;
  517. peer->sackfreq = asoc->sackfreq;
  518. if (addr->sa.sa_family == AF_INET6) {
  519. __be32 info = addr->v6.sin6_flowinfo;
  520. if (info) {
  521. peer->flowlabel = ntohl(info & IPV6_FLOWLABEL_MASK);
  522. peer->flowlabel |= SCTP_FLOWLABEL_SET_MASK;
  523. } else {
  524. peer->flowlabel = asoc->flowlabel;
  525. }
  526. }
  527. peer->dscp = asoc->dscp;
  528. /* Enable/disable heartbeat, SACK delay, and path MTU discovery
  529. * based on association setting.
  530. */
  531. peer->param_flags = asoc->param_flags;
  532. /* Initialize the pmtu of the transport. */
  533. sctp_transport_route(peer, NULL, sp);
  534. /* If this is the first transport addr on this association,
  535. * initialize the association PMTU to the peer's PMTU.
  536. * If not and the current association PMTU is higher than the new
  537. * peer's PMTU, reset the association PMTU to the new peer's PMTU.
  538. */
  539. sctp_assoc_set_pmtu(asoc, asoc->pathmtu ?
  540. min_t(int, peer->pathmtu, asoc->pathmtu) :
  541. peer->pathmtu);
  542. peer->pmtu_pending = 0;
  543. /* The asoc->peer.port might not be meaningful yet, but
  544. * initialize the packet structure anyway.
  545. */
  546. sctp_packet_init(&peer->packet, peer, asoc->base.bind_addr.port,
  547. asoc->peer.port);
  548. /* 7.2.1 Slow-Start
  549. *
  550. * o The initial cwnd before DATA transmission or after a sufficiently
  551. * long idle period MUST be set to
  552. * min(4*MTU, max(2*MTU, 4380 bytes))
  553. *
  554. * o The initial value of ssthresh MAY be arbitrarily high
  555. * (for example, implementations MAY use the size of the
  556. * receiver advertised window).
  557. */
  558. peer->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
  559. /* At this point, we may not have the receiver's advertised window,
  560. * so initialize ssthresh to the default value and it will be set
  561. * later when we process the INIT.
  562. */
  563. peer->ssthresh = SCTP_DEFAULT_MAXWINDOW;
  564. peer->partial_bytes_acked = 0;
  565. peer->flight_size = 0;
  566. peer->burst_limited = 0;
  567. /* Set the transport's RTO.initial value */
  568. peer->rto = asoc->rto_initial;
  569. sctp_max_rto(asoc, peer);
  570. /* Set the peer's active state. */
  571. peer->state = peer_state;
  572. /* Add this peer into the transport hashtable */
  573. if (sctp_hash_transport(peer)) {
  574. sctp_transport_free(peer);
  575. return NULL;
  576. }
  577. sctp_transport_pl_reset(peer);
  578. /* Attach the remote transport to our asoc. */
  579. list_add_tail_rcu(&peer->transports, &asoc->peer.transport_addr_list);
  580. asoc->peer.transport_count++;
  581. sctp_ulpevent_notify_peer_addr_change(peer, SCTP_ADDR_ADDED, 0);
  582. /* If we do not yet have a primary path, set one. */
  583. if (!asoc->peer.primary_path) {
  584. sctp_assoc_set_primary(asoc, peer);
  585. asoc->peer.retran_path = peer;
  586. }
  587. if (asoc->peer.active_path == asoc->peer.retran_path &&
  588. peer->state != SCTP_UNCONFIRMED) {
  589. asoc->peer.retran_path = peer;
  590. }
  591. return peer;
  592. }
  593. /* Lookup a transport by address. */
  594. struct sctp_transport *sctp_assoc_lookup_paddr(
  595. const struct sctp_association *asoc,
  596. const union sctp_addr *address)
  597. {
  598. struct sctp_transport *t;
  599. /* Cycle through all transports searching for a peer address. */
  600. list_for_each_entry(t, &asoc->peer.transport_addr_list,
  601. transports) {
  602. if (sctp_cmp_addr_exact(address, &t->ipaddr))
  603. return t;
  604. }
  605. return NULL;
  606. }
  607. /* Remove all transports except a give one */
  608. void sctp_assoc_del_nonprimary_peers(struct sctp_association *asoc,
  609. struct sctp_transport *primary)
  610. {
  611. struct sctp_transport *temp;
  612. struct sctp_transport *t;
  613. list_for_each_entry_safe(t, temp, &asoc->peer.transport_addr_list,
  614. transports) {
  615. /* if the current transport is not the primary one, delete it */
  616. if (t != primary)
  617. sctp_assoc_rm_peer(asoc, t);
  618. }
  619. }
  620. /* Engage in transport control operations.
  621. * Mark the transport up or down and send a notification to the user.
  622. * Select and update the new active and retran paths.
  623. */
  624. void sctp_assoc_control_transport(struct sctp_association *asoc,
  625. struct sctp_transport *transport,
  626. enum sctp_transport_cmd command,
  627. sctp_sn_error_t error)
  628. {
  629. int spc_state = SCTP_ADDR_AVAILABLE;
  630. bool ulp_notify = true;
  631. /* Record the transition on the transport. */
  632. switch (command) {
  633. case SCTP_TRANSPORT_UP:
  634. /* If we are moving from UNCONFIRMED state due
  635. * to heartbeat success, report the SCTP_ADDR_CONFIRMED
  636. * state to the user, otherwise report SCTP_ADDR_AVAILABLE.
  637. */
  638. if (transport->state == SCTP_PF &&
  639. asoc->pf_expose != SCTP_PF_EXPOSE_ENABLE)
  640. ulp_notify = false;
  641. else if (transport->state == SCTP_UNCONFIRMED &&
  642. error == SCTP_HEARTBEAT_SUCCESS)
  643. spc_state = SCTP_ADDR_CONFIRMED;
  644. transport->state = SCTP_ACTIVE;
  645. sctp_transport_pl_reset(transport);
  646. break;
  647. case SCTP_TRANSPORT_DOWN:
  648. /* If the transport was never confirmed, do not transition it
  649. * to inactive state. Also, release the cached route since
  650. * there may be a better route next time.
  651. */
  652. if (transport->state != SCTP_UNCONFIRMED) {
  653. transport->state = SCTP_INACTIVE;
  654. sctp_transport_pl_reset(transport);
  655. spc_state = SCTP_ADDR_UNREACHABLE;
  656. } else {
  657. sctp_transport_dst_release(transport);
  658. ulp_notify = false;
  659. }
  660. break;
  661. case SCTP_TRANSPORT_PF:
  662. transport->state = SCTP_PF;
  663. if (asoc->pf_expose != SCTP_PF_EXPOSE_ENABLE)
  664. ulp_notify = false;
  665. else
  666. spc_state = SCTP_ADDR_POTENTIALLY_FAILED;
  667. break;
  668. default:
  669. return;
  670. }
  671. /* Generate and send a SCTP_PEER_ADDR_CHANGE notification
  672. * to the user.
  673. */
  674. if (ulp_notify)
  675. sctp_ulpevent_notify_peer_addr_change(transport,
  676. spc_state, error);
  677. /* Select new active and retran paths. */
  678. sctp_select_active_and_retran_path(asoc);
  679. }
  680. /* Hold a reference to an association. */
  681. void sctp_association_hold(struct sctp_association *asoc)
  682. {
  683. refcount_inc(&asoc->base.refcnt);
  684. }
  685. /* Release a reference to an association and cleanup
  686. * if there are no more references.
  687. */
  688. void sctp_association_put(struct sctp_association *asoc)
  689. {
  690. if (refcount_dec_and_test(&asoc->base.refcnt))
  691. sctp_association_destroy(asoc);
  692. }
  693. /* Allocate the next TSN, Transmission Sequence Number, for the given
  694. * association.
  695. */
  696. __u32 sctp_association_get_next_tsn(struct sctp_association *asoc)
  697. {
  698. /* From Section 1.6 Serial Number Arithmetic:
  699. * Transmission Sequence Numbers wrap around when they reach
  700. * 2**32 - 1. That is, the next TSN a DATA chunk MUST use
  701. * after transmitting TSN = 2*32 - 1 is TSN = 0.
  702. */
  703. __u32 retval = asoc->next_tsn;
  704. asoc->next_tsn++;
  705. asoc->unack_data++;
  706. return retval;
  707. }
  708. /* Compare two addresses to see if they match. Wildcard addresses
  709. * only match themselves.
  710. */
  711. int sctp_cmp_addr_exact(const union sctp_addr *ss1,
  712. const union sctp_addr *ss2)
  713. {
  714. struct sctp_af *af;
  715. af = sctp_get_af_specific(ss1->sa.sa_family);
  716. if (unlikely(!af))
  717. return 0;
  718. return af->cmp_addr(ss1, ss2);
  719. }
  720. /* Return an ecne chunk to get prepended to a packet.
  721. * Note: We are sly and return a shared, prealloced chunk. FIXME:
  722. * No we don't, but we could/should.
  723. */
  724. struct sctp_chunk *sctp_get_ecne_prepend(struct sctp_association *asoc)
  725. {
  726. if (!asoc->need_ecne)
  727. return NULL;
  728. /* Send ECNE if needed.
  729. * Not being able to allocate a chunk here is not deadly.
  730. */
  731. return sctp_make_ecne(asoc, asoc->last_ecne_tsn);
  732. }
  733. /*
  734. * Find which transport this TSN was sent on.
  735. */
  736. struct sctp_transport *sctp_assoc_lookup_tsn(struct sctp_association *asoc,
  737. __u32 tsn)
  738. {
  739. struct sctp_transport *active;
  740. struct sctp_transport *match;
  741. struct sctp_transport *transport;
  742. struct sctp_chunk *chunk;
  743. __be32 key = htonl(tsn);
  744. match = NULL;
  745. /*
  746. * FIXME: In general, find a more efficient data structure for
  747. * searching.
  748. */
  749. /*
  750. * The general strategy is to search each transport's transmitted
  751. * list. Return which transport this TSN lives on.
  752. *
  753. * Let's be hopeful and check the active_path first.
  754. * Another optimization would be to know if there is only one
  755. * outbound path and not have to look for the TSN at all.
  756. *
  757. */
  758. active = asoc->peer.active_path;
  759. list_for_each_entry(chunk, &active->transmitted,
  760. transmitted_list) {
  761. if (key == chunk->subh.data_hdr->tsn) {
  762. match = active;
  763. goto out;
  764. }
  765. }
  766. /* If not found, go search all the other transports. */
  767. list_for_each_entry(transport, &asoc->peer.transport_addr_list,
  768. transports) {
  769. if (transport == active)
  770. continue;
  771. list_for_each_entry(chunk, &transport->transmitted,
  772. transmitted_list) {
  773. if (key == chunk->subh.data_hdr->tsn) {
  774. match = transport;
  775. goto out;
  776. }
  777. }
  778. }
  779. out:
  780. return match;
  781. }
  782. /* Do delayed input processing. This is scheduled by sctp_rcv(). */
  783. static void sctp_assoc_bh_rcv(struct work_struct *work)
  784. {
  785. struct sctp_association *asoc =
  786. container_of(work, struct sctp_association,
  787. base.inqueue.immediate);
  788. struct net *net = asoc->base.net;
  789. union sctp_subtype subtype;
  790. struct sctp_endpoint *ep;
  791. struct sctp_chunk *chunk;
  792. struct sctp_inq *inqueue;
  793. int first_time = 1; /* is this the first time through the loop */
  794. int error = 0;
  795. int state;
  796. /* The association should be held so we should be safe. */
  797. ep = asoc->ep;
  798. inqueue = &asoc->base.inqueue;
  799. sctp_association_hold(asoc);
  800. while (NULL != (chunk = sctp_inq_pop(inqueue))) {
  801. state = asoc->state;
  802. subtype = SCTP_ST_CHUNK(chunk->chunk_hdr->type);
  803. /* If the first chunk in the packet is AUTH, do special
  804. * processing specified in Section 6.3 of SCTP-AUTH spec
  805. */
  806. if (first_time && subtype.chunk == SCTP_CID_AUTH) {
  807. struct sctp_chunkhdr *next_hdr;
  808. next_hdr = sctp_inq_peek(inqueue);
  809. if (!next_hdr)
  810. goto normal;
  811. /* If the next chunk is COOKIE-ECHO, skip the AUTH
  812. * chunk while saving a pointer to it so we can do
  813. * Authentication later (during cookie-echo
  814. * processing).
  815. */
  816. if (next_hdr->type == SCTP_CID_COOKIE_ECHO) {
  817. chunk->auth_chunk = skb_clone(chunk->skb,
  818. GFP_ATOMIC);
  819. chunk->auth = 1;
  820. continue;
  821. }
  822. }
  823. normal:
  824. /* SCTP-AUTH, Section 6.3:
  825. * The receiver has a list of chunk types which it expects
  826. * to be received only after an AUTH-chunk. This list has
  827. * been sent to the peer during the association setup. It
  828. * MUST silently discard these chunks if they are not placed
  829. * after an AUTH chunk in the packet.
  830. */
  831. if (sctp_auth_recv_cid(subtype.chunk, asoc) && !chunk->auth)
  832. continue;
  833. /* Remember where the last DATA chunk came from so we
  834. * know where to send the SACK.
  835. */
  836. if (sctp_chunk_is_data(chunk))
  837. asoc->peer.last_data_from = chunk->transport;
  838. else {
  839. SCTP_INC_STATS(net, SCTP_MIB_INCTRLCHUNKS);
  840. asoc->stats.ictrlchunks++;
  841. if (chunk->chunk_hdr->type == SCTP_CID_SACK)
  842. asoc->stats.isacks++;
  843. }
  844. if (chunk->transport)
  845. chunk->transport->last_time_heard = ktime_get();
  846. /* Run through the state machine. */
  847. error = sctp_do_sm(net, SCTP_EVENT_T_CHUNK, subtype,
  848. state, ep, asoc, chunk, GFP_ATOMIC);
  849. /* Check to see if the association is freed in response to
  850. * the incoming chunk. If so, get out of the while loop.
  851. */
  852. if (asoc->base.dead)
  853. break;
  854. /* If there is an error on chunk, discard this packet. */
  855. if (error && chunk)
  856. chunk->pdiscard = 1;
  857. if (first_time)
  858. first_time = 0;
  859. }
  860. sctp_association_put(asoc);
  861. }
  862. /* This routine moves an association from its old sk to a new sk. */
  863. void sctp_assoc_migrate(struct sctp_association *assoc, struct sock *newsk)
  864. {
  865. struct sctp_sock *newsp = sctp_sk(newsk);
  866. struct sock *oldsk = assoc->base.sk;
  867. /* Delete the association from the old endpoint's list of
  868. * associations.
  869. */
  870. list_del_init(&assoc->asocs);
  871. /* Decrement the backlog value for a TCP-style socket. */
  872. if (sctp_style(oldsk, TCP))
  873. sk_acceptq_removed(oldsk);
  874. /* Release references to the old endpoint and the sock. */
  875. sctp_endpoint_put(assoc->ep);
  876. sock_put(assoc->base.sk);
  877. /* Get a reference to the new endpoint. */
  878. assoc->ep = newsp->ep;
  879. sctp_endpoint_hold(assoc->ep);
  880. /* Get a reference to the new sock. */
  881. assoc->base.sk = newsk;
  882. sock_hold(assoc->base.sk);
  883. /* Add the association to the new endpoint's list of associations. */
  884. sctp_endpoint_add_asoc(newsp->ep, assoc);
  885. }
  886. /* Update an association (possibly from unexpected COOKIE-ECHO processing). */
  887. int sctp_assoc_update(struct sctp_association *asoc,
  888. struct sctp_association *new)
  889. {
  890. struct sctp_transport *trans;
  891. struct list_head *pos, *temp;
  892. /* Copy in new parameters of peer. */
  893. asoc->c = new->c;
  894. asoc->peer.rwnd = new->peer.rwnd;
  895. asoc->peer.sack_needed = new->peer.sack_needed;
  896. asoc->peer.auth_capable = new->peer.auth_capable;
  897. asoc->peer.i = new->peer.i;
  898. if (!sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_INITIAL,
  899. asoc->peer.i.initial_tsn, GFP_ATOMIC))
  900. return -ENOMEM;
  901. /* Remove any peer addresses not present in the new association. */
  902. list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
  903. trans = list_entry(pos, struct sctp_transport, transports);
  904. if (!sctp_assoc_lookup_paddr(new, &trans->ipaddr)) {
  905. sctp_assoc_rm_peer(asoc, trans);
  906. continue;
  907. }
  908. if (asoc->state >= SCTP_STATE_ESTABLISHED)
  909. sctp_transport_reset(trans);
  910. }
  911. /* If the case is A (association restart), use
  912. * initial_tsn as next_tsn. If the case is B, use
  913. * current next_tsn in case data sent to peer
  914. * has been discarded and needs retransmission.
  915. */
  916. if (asoc->state >= SCTP_STATE_ESTABLISHED) {
  917. asoc->next_tsn = new->next_tsn;
  918. asoc->ctsn_ack_point = new->ctsn_ack_point;
  919. asoc->adv_peer_ack_point = new->adv_peer_ack_point;
  920. /* Reinitialize SSN for both local streams
  921. * and peer's streams.
  922. */
  923. sctp_stream_clear(&asoc->stream);
  924. /* Flush the ULP reassembly and ordered queue.
  925. * Any data there will now be stale and will
  926. * cause problems.
  927. */
  928. sctp_ulpq_flush(&asoc->ulpq);
  929. /* reset the overall association error count so
  930. * that the restarted association doesn't get torn
  931. * down on the next retransmission timer.
  932. */
  933. asoc->overall_error_count = 0;
  934. } else {
  935. /* Add any peer addresses from the new association. */
  936. list_for_each_entry(trans, &new->peer.transport_addr_list,
  937. transports)
  938. if (!sctp_assoc_add_peer(asoc, &trans->ipaddr,
  939. GFP_ATOMIC, trans->state))
  940. return -ENOMEM;
  941. asoc->ctsn_ack_point = asoc->next_tsn - 1;
  942. asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
  943. if (sctp_state(asoc, COOKIE_WAIT))
  944. sctp_stream_update(&asoc->stream, &new->stream);
  945. /* get a new assoc id if we don't have one yet. */
  946. if (sctp_assoc_set_id(asoc, GFP_ATOMIC))
  947. return -ENOMEM;
  948. }
  949. /* SCTP-AUTH: Save the peer parameters from the new associations
  950. * and also move the association shared keys over
  951. */
  952. kfree(asoc->peer.peer_random);
  953. asoc->peer.peer_random = new->peer.peer_random;
  954. new->peer.peer_random = NULL;
  955. kfree(asoc->peer.peer_chunks);
  956. asoc->peer.peer_chunks = new->peer.peer_chunks;
  957. new->peer.peer_chunks = NULL;
  958. kfree(asoc->peer.peer_hmacs);
  959. asoc->peer.peer_hmacs = new->peer.peer_hmacs;
  960. new->peer.peer_hmacs = NULL;
  961. return sctp_auth_asoc_init_active_key(asoc, GFP_ATOMIC);
  962. }
  963. /* Update the retran path for sending a retransmitted packet.
  964. * See also RFC4960, 6.4. Multi-Homed SCTP Endpoints:
  965. *
  966. * When there is outbound data to send and the primary path
  967. * becomes inactive (e.g., due to failures), or where the
  968. * SCTP user explicitly requests to send data to an
  969. * inactive destination transport address, before reporting
  970. * an error to its ULP, the SCTP endpoint should try to send
  971. * the data to an alternate active destination transport
  972. * address if one exists.
  973. *
  974. * When retransmitting data that timed out, if the endpoint
  975. * is multihomed, it should consider each source-destination
  976. * address pair in its retransmission selection policy.
  977. * When retransmitting timed-out data, the endpoint should
  978. * attempt to pick the most divergent source-destination
  979. * pair from the original source-destination pair to which
  980. * the packet was transmitted.
  981. *
  982. * Note: Rules for picking the most divergent source-destination
  983. * pair are an implementation decision and are not specified
  984. * within this document.
  985. *
  986. * Our basic strategy is to round-robin transports in priorities
  987. * according to sctp_trans_score() e.g., if no such
  988. * transport with state SCTP_ACTIVE exists, round-robin through
  989. * SCTP_UNKNOWN, etc. You get the picture.
  990. */
  991. static u8 sctp_trans_score(const struct sctp_transport *trans)
  992. {
  993. switch (trans->state) {
  994. case SCTP_ACTIVE:
  995. return 3; /* best case */
  996. case SCTP_UNKNOWN:
  997. return 2;
  998. case SCTP_PF:
  999. return 1;
  1000. default: /* case SCTP_INACTIVE */
  1001. return 0; /* worst case */
  1002. }
  1003. }
  1004. static struct sctp_transport *sctp_trans_elect_tie(struct sctp_transport *trans1,
  1005. struct sctp_transport *trans2)
  1006. {
  1007. if (trans1->error_count > trans2->error_count) {
  1008. return trans2;
  1009. } else if (trans1->error_count == trans2->error_count &&
  1010. ktime_after(trans2->last_time_heard,
  1011. trans1->last_time_heard)) {
  1012. return trans2;
  1013. } else {
  1014. return trans1;
  1015. }
  1016. }
  1017. static struct sctp_transport *sctp_trans_elect_best(struct sctp_transport *curr,
  1018. struct sctp_transport *best)
  1019. {
  1020. u8 score_curr, score_best;
  1021. if (best == NULL || curr == best)
  1022. return curr;
  1023. score_curr = sctp_trans_score(curr);
  1024. score_best = sctp_trans_score(best);
  1025. /* First, try a score-based selection if both transport states
  1026. * differ. If we're in a tie, lets try to make a more clever
  1027. * decision here based on error counts and last time heard.
  1028. */
  1029. if (score_curr > score_best)
  1030. return curr;
  1031. else if (score_curr == score_best)
  1032. return sctp_trans_elect_tie(best, curr);
  1033. else
  1034. return best;
  1035. }
  1036. void sctp_assoc_update_retran_path(struct sctp_association *asoc)
  1037. {
  1038. struct sctp_transport *trans = asoc->peer.retran_path;
  1039. struct sctp_transport *trans_next = NULL;
  1040. /* We're done as we only have the one and only path. */
  1041. if (asoc->peer.transport_count == 1)
  1042. return;
  1043. /* If active_path and retran_path are the same and active,
  1044. * then this is the only active path. Use it.
  1045. */
  1046. if (asoc->peer.active_path == asoc->peer.retran_path &&
  1047. asoc->peer.active_path->state == SCTP_ACTIVE)
  1048. return;
  1049. /* Iterate from retran_path's successor back to retran_path. */
  1050. for (trans = list_next_entry(trans, transports); 1;
  1051. trans = list_next_entry(trans, transports)) {
  1052. /* Manually skip the head element. */
  1053. if (&trans->transports == &asoc->peer.transport_addr_list)
  1054. continue;
  1055. if (trans->state == SCTP_UNCONFIRMED)
  1056. continue;
  1057. trans_next = sctp_trans_elect_best(trans, trans_next);
  1058. /* Active is good enough for immediate return. */
  1059. if (trans_next->state == SCTP_ACTIVE)
  1060. break;
  1061. /* We've reached the end, time to update path. */
  1062. if (trans == asoc->peer.retran_path)
  1063. break;
  1064. }
  1065. asoc->peer.retran_path = trans_next;
  1066. pr_debug("%s: association:%p updated new path to addr:%pISpc\n",
  1067. __func__, asoc, &asoc->peer.retran_path->ipaddr.sa);
  1068. }
  1069. static void sctp_select_active_and_retran_path(struct sctp_association *asoc)
  1070. {
  1071. struct sctp_transport *trans, *trans_pri = NULL, *trans_sec = NULL;
  1072. struct sctp_transport *trans_pf = NULL;
  1073. /* Look for the two most recently used active transports. */
  1074. list_for_each_entry(trans, &asoc->peer.transport_addr_list,
  1075. transports) {
  1076. /* Skip uninteresting transports. */
  1077. if (trans->state == SCTP_INACTIVE ||
  1078. trans->state == SCTP_UNCONFIRMED)
  1079. continue;
  1080. /* Keep track of the best PF transport from our
  1081. * list in case we don't find an active one.
  1082. */
  1083. if (trans->state == SCTP_PF) {
  1084. trans_pf = sctp_trans_elect_best(trans, trans_pf);
  1085. continue;
  1086. }
  1087. /* For active transports, pick the most recent ones. */
  1088. if (trans_pri == NULL ||
  1089. ktime_after(trans->last_time_heard,
  1090. trans_pri->last_time_heard)) {
  1091. trans_sec = trans_pri;
  1092. trans_pri = trans;
  1093. } else if (trans_sec == NULL ||
  1094. ktime_after(trans->last_time_heard,
  1095. trans_sec->last_time_heard)) {
  1096. trans_sec = trans;
  1097. }
  1098. }
  1099. /* RFC 2960 6.4 Multi-Homed SCTP Endpoints
  1100. *
  1101. * By default, an endpoint should always transmit to the primary
  1102. * path, unless the SCTP user explicitly specifies the
  1103. * destination transport address (and possibly source transport
  1104. * address) to use. [If the primary is active but not most recent,
  1105. * bump the most recently used transport.]
  1106. */
  1107. if ((asoc->peer.primary_path->state == SCTP_ACTIVE ||
  1108. asoc->peer.primary_path->state == SCTP_UNKNOWN) &&
  1109. asoc->peer.primary_path != trans_pri) {
  1110. trans_sec = trans_pri;
  1111. trans_pri = asoc->peer.primary_path;
  1112. }
  1113. /* We did not find anything useful for a possible retransmission
  1114. * path; either primary path that we found is the same as
  1115. * the current one, or we didn't generally find an active one.
  1116. */
  1117. if (trans_sec == NULL)
  1118. trans_sec = trans_pri;
  1119. /* If we failed to find a usable transport, just camp on the
  1120. * active or pick a PF iff it's the better choice.
  1121. */
  1122. if (trans_pri == NULL) {
  1123. trans_pri = sctp_trans_elect_best(asoc->peer.active_path, trans_pf);
  1124. trans_sec = trans_pri;
  1125. }
  1126. /* Set the active and retran transports. */
  1127. asoc->peer.active_path = trans_pri;
  1128. asoc->peer.retran_path = trans_sec;
  1129. }
  1130. struct sctp_transport *
  1131. sctp_assoc_choose_alter_transport(struct sctp_association *asoc,
  1132. struct sctp_transport *last_sent_to)
  1133. {
  1134. /* If this is the first time packet is sent, use the active path,
  1135. * else use the retran path. If the last packet was sent over the
  1136. * retran path, update the retran path and use it.
  1137. */
  1138. if (last_sent_to == NULL) {
  1139. return asoc->peer.active_path;
  1140. } else {
  1141. if (last_sent_to == asoc->peer.retran_path)
  1142. sctp_assoc_update_retran_path(asoc);
  1143. return asoc->peer.retran_path;
  1144. }
  1145. }
  1146. void sctp_assoc_update_frag_point(struct sctp_association *asoc)
  1147. {
  1148. int frag = sctp_mtu_payload(sctp_sk(asoc->base.sk), asoc->pathmtu,
  1149. sctp_datachk_len(&asoc->stream));
  1150. if (asoc->user_frag)
  1151. frag = min_t(int, frag, asoc->user_frag);
  1152. frag = min_t(int, frag, SCTP_MAX_CHUNK_LEN -
  1153. sctp_datachk_len(&asoc->stream));
  1154. asoc->frag_point = SCTP_TRUNC4(frag);
  1155. }
  1156. void sctp_assoc_set_pmtu(struct sctp_association *asoc, __u32 pmtu)
  1157. {
  1158. if (asoc->pathmtu != pmtu) {
  1159. asoc->pathmtu = pmtu;
  1160. sctp_assoc_update_frag_point(asoc);
  1161. }
  1162. pr_debug("%s: asoc:%p, pmtu:%d, frag_point:%d\n", __func__, asoc,
  1163. asoc->pathmtu, asoc->frag_point);
  1164. }
  1165. /* Update the association's pmtu and frag_point by going through all the
  1166. * transports. This routine is called when a transport's PMTU has changed.
  1167. */
  1168. void sctp_assoc_sync_pmtu(struct sctp_association *asoc)
  1169. {
  1170. struct sctp_transport *t;
  1171. __u32 pmtu = 0;
  1172. if (!asoc)
  1173. return;
  1174. /* Get the lowest pmtu of all the transports. */
  1175. list_for_each_entry(t, &asoc->peer.transport_addr_list, transports) {
  1176. if (t->pmtu_pending && t->dst) {
  1177. sctp_transport_update_pmtu(t,
  1178. atomic_read(&t->mtu_info));
  1179. t->pmtu_pending = 0;
  1180. }
  1181. if (!pmtu || (t->pathmtu < pmtu))
  1182. pmtu = t->pathmtu;
  1183. }
  1184. sctp_assoc_set_pmtu(asoc, pmtu);
  1185. }
  1186. /* Should we send a SACK to update our peer? */
  1187. static inline bool sctp_peer_needs_update(struct sctp_association *asoc)
  1188. {
  1189. struct net *net = asoc->base.net;
  1190. switch (asoc->state) {
  1191. case SCTP_STATE_ESTABLISHED:
  1192. case SCTP_STATE_SHUTDOWN_PENDING:
  1193. case SCTP_STATE_SHUTDOWN_RECEIVED:
  1194. case SCTP_STATE_SHUTDOWN_SENT:
  1195. if ((asoc->rwnd > asoc->a_rwnd) &&
  1196. ((asoc->rwnd - asoc->a_rwnd) >= max_t(__u32,
  1197. (asoc->base.sk->sk_rcvbuf >> net->sctp.rwnd_upd_shift),
  1198. asoc->pathmtu)))
  1199. return true;
  1200. break;
  1201. default:
  1202. break;
  1203. }
  1204. return false;
  1205. }
  1206. /* Increase asoc's rwnd by len and send any window update SACK if needed. */
  1207. void sctp_assoc_rwnd_increase(struct sctp_association *asoc, unsigned int len)
  1208. {
  1209. struct sctp_chunk *sack;
  1210. struct timer_list *timer;
  1211. if (asoc->rwnd_over) {
  1212. if (asoc->rwnd_over >= len) {
  1213. asoc->rwnd_over -= len;
  1214. } else {
  1215. asoc->rwnd += (len - asoc->rwnd_over);
  1216. asoc->rwnd_over = 0;
  1217. }
  1218. } else {
  1219. asoc->rwnd += len;
  1220. }
  1221. /* If we had window pressure, start recovering it
  1222. * once our rwnd had reached the accumulated pressure
  1223. * threshold. The idea is to recover slowly, but up
  1224. * to the initial advertised window.
  1225. */
  1226. if (asoc->rwnd_press) {
  1227. int change = min(asoc->pathmtu, asoc->rwnd_press);
  1228. asoc->rwnd += change;
  1229. asoc->rwnd_press -= change;
  1230. }
  1231. pr_debug("%s: asoc:%p rwnd increased by %d to (%u, %u) - %u\n",
  1232. __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
  1233. asoc->a_rwnd);
  1234. /* Send a window update SACK if the rwnd has increased by at least the
  1235. * minimum of the association's PMTU and half of the receive buffer.
  1236. * The algorithm used is similar to the one described in
  1237. * Section 4.2.3.3 of RFC 1122.
  1238. */
  1239. if (sctp_peer_needs_update(asoc)) {
  1240. asoc->a_rwnd = asoc->rwnd;
  1241. pr_debug("%s: sending window update SACK- asoc:%p rwnd:%u "
  1242. "a_rwnd:%u\n", __func__, asoc, asoc->rwnd,
  1243. asoc->a_rwnd);
  1244. sack = sctp_make_sack(asoc);
  1245. if (!sack)
  1246. return;
  1247. asoc->peer.sack_needed = 0;
  1248. sctp_outq_tail(&asoc->outqueue, sack, GFP_ATOMIC);
  1249. /* Stop the SACK timer. */
  1250. timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
  1251. if (timer_delete(timer))
  1252. sctp_association_put(asoc);
  1253. }
  1254. }
  1255. /* Decrease asoc's rwnd by len. */
  1256. void sctp_assoc_rwnd_decrease(struct sctp_association *asoc, unsigned int len)
  1257. {
  1258. int rx_count;
  1259. int over = 0;
  1260. if (unlikely(!asoc->rwnd || asoc->rwnd_over))
  1261. pr_debug("%s: association:%p has asoc->rwnd:%u, "
  1262. "asoc->rwnd_over:%u!\n", __func__, asoc,
  1263. asoc->rwnd, asoc->rwnd_over);
  1264. if (asoc->ep->rcvbuf_policy)
  1265. rx_count = atomic_read(&asoc->rmem_alloc);
  1266. else
  1267. rx_count = atomic_read(&asoc->base.sk->sk_rmem_alloc);
  1268. /* If we've reached or overflowed our receive buffer, announce
  1269. * a 0 rwnd if rwnd would still be positive. Store the
  1270. * potential pressure overflow so that the window can be restored
  1271. * back to original value.
  1272. */
  1273. if (rx_count >= asoc->base.sk->sk_rcvbuf)
  1274. over = 1;
  1275. if (asoc->rwnd >= len) {
  1276. asoc->rwnd -= len;
  1277. if (over) {
  1278. asoc->rwnd_press += asoc->rwnd;
  1279. asoc->rwnd = 0;
  1280. }
  1281. } else {
  1282. asoc->rwnd_over += len - asoc->rwnd;
  1283. asoc->rwnd = 0;
  1284. }
  1285. pr_debug("%s: asoc:%p rwnd decreased by %d to (%u, %u, %u)\n",
  1286. __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
  1287. asoc->rwnd_press);
  1288. }
  1289. /* Build the bind address list for the association based on info from the
  1290. * local endpoint and the remote peer.
  1291. */
  1292. int sctp_assoc_set_bind_addr_from_ep(struct sctp_association *asoc,
  1293. enum sctp_scope scope, gfp_t gfp)
  1294. {
  1295. struct sock *sk = asoc->base.sk;
  1296. int flags;
  1297. /* Use scoping rules to determine the subset of addresses from
  1298. * the endpoint.
  1299. */
  1300. flags = (PF_INET6 == sk->sk_family) ? SCTP_ADDR6_ALLOWED : 0;
  1301. if (!inet_v6_ipv6only(sk))
  1302. flags |= SCTP_ADDR4_ALLOWED;
  1303. if (asoc->peer.ipv4_address)
  1304. flags |= SCTP_ADDR4_PEERSUPP;
  1305. if (asoc->peer.ipv6_address)
  1306. flags |= SCTP_ADDR6_PEERSUPP;
  1307. return sctp_bind_addr_copy(asoc->base.net,
  1308. &asoc->base.bind_addr,
  1309. &asoc->ep->base.bind_addr,
  1310. scope, gfp, flags);
  1311. }
  1312. /* Build the association's bind address list from the cookie. */
  1313. int sctp_assoc_set_bind_addr_from_cookie(struct sctp_association *asoc,
  1314. struct sctp_cookie *cookie,
  1315. gfp_t gfp)
  1316. {
  1317. struct sctp_init_chunk *peer_init = (struct sctp_init_chunk *)(cookie + 1);
  1318. int var_size2 = ntohs(peer_init->chunk_hdr.length);
  1319. int var_size3 = cookie->raw_addr_list_len;
  1320. __u8 *raw = (__u8 *)peer_init + var_size2;
  1321. return sctp_raw_to_bind_addrs(&asoc->base.bind_addr, raw, var_size3,
  1322. asoc->ep->base.bind_addr.port, gfp);
  1323. }
  1324. /* Lookup laddr in the bind address list of an association. */
  1325. int sctp_assoc_lookup_laddr(struct sctp_association *asoc,
  1326. const union sctp_addr *laddr)
  1327. {
  1328. int found = 0;
  1329. if ((asoc->base.bind_addr.port == ntohs(laddr->v4.sin_port)) &&
  1330. sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
  1331. sctp_sk(asoc->base.sk)))
  1332. found = 1;
  1333. return found;
  1334. }
  1335. /* Set an association id for a given association */
  1336. int sctp_assoc_set_id(struct sctp_association *asoc, gfp_t gfp)
  1337. {
  1338. bool preload = gfpflags_allow_blocking(gfp);
  1339. int ret;
  1340. /* If the id is already assigned, keep it. */
  1341. if (asoc->assoc_id)
  1342. return 0;
  1343. if (preload)
  1344. idr_preload(gfp);
  1345. spin_lock_bh(&sctp_assocs_id_lock);
  1346. /* 0, 1, 2 are used as SCTP_FUTURE_ASSOC, SCTP_CURRENT_ASSOC and
  1347. * SCTP_ALL_ASSOC, so an available id must be > SCTP_ALL_ASSOC.
  1348. */
  1349. ret = idr_alloc_cyclic(&sctp_assocs_id, asoc, SCTP_ALL_ASSOC + 1, 0,
  1350. GFP_NOWAIT);
  1351. spin_unlock_bh(&sctp_assocs_id_lock);
  1352. if (preload)
  1353. idr_preload_end();
  1354. if (ret < 0)
  1355. return ret;
  1356. asoc->assoc_id = (sctp_assoc_t)ret;
  1357. return 0;
  1358. }
  1359. /* Free the ASCONF queue */
  1360. static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc)
  1361. {
  1362. struct sctp_chunk *asconf;
  1363. struct sctp_chunk *tmp;
  1364. list_for_each_entry_safe(asconf, tmp, &asoc->addip_chunk_list, list) {
  1365. list_del_init(&asconf->list);
  1366. sctp_chunk_free(asconf);
  1367. }
  1368. }
  1369. /* Free asconf_ack cache */
  1370. static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc)
  1371. {
  1372. struct sctp_chunk *ack;
  1373. struct sctp_chunk *tmp;
  1374. list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
  1375. transmitted_list) {
  1376. list_del_init(&ack->transmitted_list);
  1377. sctp_chunk_free(ack);
  1378. }
  1379. }
  1380. /* Clean up the ASCONF_ACK queue */
  1381. void sctp_assoc_clean_asconf_ack_cache(const struct sctp_association *asoc)
  1382. {
  1383. struct sctp_chunk *ack;
  1384. struct sctp_chunk *tmp;
  1385. /* We can remove all the entries from the queue up to
  1386. * the "Peer-Sequence-Number".
  1387. */
  1388. list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
  1389. transmitted_list) {
  1390. if (ack->subh.addip_hdr->serial ==
  1391. htonl(asoc->peer.addip_serial))
  1392. break;
  1393. list_del_init(&ack->transmitted_list);
  1394. sctp_chunk_free(ack);
  1395. }
  1396. }
  1397. /* Find the ASCONF_ACK whose serial number matches ASCONF */
  1398. struct sctp_chunk *sctp_assoc_lookup_asconf_ack(
  1399. const struct sctp_association *asoc,
  1400. __be32 serial)
  1401. {
  1402. struct sctp_chunk *ack;
  1403. /* Walk through the list of cached ASCONF-ACKs and find the
  1404. * ack chunk whose serial number matches that of the request.
  1405. */
  1406. list_for_each_entry(ack, &asoc->asconf_ack_list, transmitted_list) {
  1407. if (sctp_chunk_pending(ack))
  1408. continue;
  1409. if (ack->subh.addip_hdr->serial == serial) {
  1410. sctp_chunk_hold(ack);
  1411. return ack;
  1412. }
  1413. }
  1414. return NULL;
  1415. }
  1416. void sctp_asconf_queue_teardown(struct sctp_association *asoc)
  1417. {
  1418. /* Free any cached ASCONF_ACK chunk. */
  1419. sctp_assoc_free_asconf_acks(asoc);
  1420. /* Free the ASCONF queue. */
  1421. sctp_assoc_free_asconf_queue(asoc);
  1422. /* Free any cached ASCONF chunk. */
  1423. if (asoc->addip_last_asconf)
  1424. sctp_chunk_free(asoc->addip_last_asconf);
  1425. }