output.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698
  1. /*
  2. * net/dccp/output.c
  3. *
  4. * An implementation of the DCCP protocol
  5. * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
  6. *
  7. * This program is free software; you can redistribute it and/or
  8. * modify it under the terms of the GNU General Public License
  9. * as published by the Free Software Foundation; either version
  10. * 2 of the License, or (at your option) any later version.
  11. */
  12. #include <linux/dccp.h>
  13. #include <linux/kernel.h>
  14. #include <linux/skbuff.h>
  15. #include <linux/slab.h>
  16. #include <net/inet_sock.h>
  17. #include <net/sock.h>
  18. #include "ackvec.h"
  19. #include "ccid.h"
  20. #include "dccp.h"
  21. static inline void dccp_event_ack_sent(struct sock *sk)
  22. {
  23. inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
  24. }
  25. /* enqueue @skb on sk_send_head for retransmission, return clone to send now */
  26. static struct sk_buff *dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
  27. {
  28. skb_set_owner_w(skb, sk);
  29. WARN_ON(sk->sk_send_head);
  30. sk->sk_send_head = skb;
  31. return skb_clone(sk->sk_send_head, gfp_any());
  32. }
  33. /*
  34. * All SKB's seen here are completely headerless. It is our
  35. * job to build the DCCP header, and pass the packet down to
  36. * IP so it can do the same plus pass the packet off to the
  37. * device.
  38. */
  39. static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
  40. {
  41. if (likely(skb != NULL)) {
  42. struct inet_sock *inet = inet_sk(sk);
  43. const struct inet_connection_sock *icsk = inet_csk(sk);
  44. struct dccp_sock *dp = dccp_sk(sk);
  45. struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
  46. struct dccp_hdr *dh;
  47. /* XXX For now we're using only 48 bits sequence numbers */
  48. const u32 dccp_header_size = sizeof(*dh) +
  49. sizeof(struct dccp_hdr_ext) +
  50. dccp_packet_hdr_len(dcb->dccpd_type);
  51. int err, set_ack = 1;
  52. u64 ackno = dp->dccps_gsr;
  53. /*
  54. * Increment GSS here already in case the option code needs it.
  55. * Update GSS for real only if option processing below succeeds.
  56. */
  57. dcb->dccpd_seq = ADD48(dp->dccps_gss, 1);
  58. switch (dcb->dccpd_type) {
  59. case DCCP_PKT_DATA:
  60. set_ack = 0;
  61. /* fall through */
  62. case DCCP_PKT_DATAACK:
  63. case DCCP_PKT_RESET:
  64. break;
  65. case DCCP_PKT_REQUEST:
  66. set_ack = 0;
  67. /* Use ISS on the first (non-retransmitted) Request. */
  68. if (icsk->icsk_retransmits == 0)
  69. dcb->dccpd_seq = dp->dccps_iss;
  70. /* fall through */
  71. case DCCP_PKT_SYNC:
  72. case DCCP_PKT_SYNCACK:
  73. ackno = dcb->dccpd_ack_seq;
  74. /* fall through */
  75. default:
  76. /*
  77. * Set owner/destructor: some skbs are allocated via
  78. * alloc_skb (e.g. when retransmission may happen).
  79. * Only Data, DataAck, and Reset packets should come
  80. * through here with skb->sk set.
  81. */
  82. WARN_ON(skb->sk);
  83. skb_set_owner_w(skb, sk);
  84. break;
  85. }
  86. if (dccp_insert_options(sk, skb)) {
  87. kfree_skb(skb);
  88. return -EPROTO;
  89. }
  90. /* Build DCCP header and checksum it. */
  91. dh = dccp_zeroed_hdr(skb, dccp_header_size);
  92. dh->dccph_type = dcb->dccpd_type;
  93. dh->dccph_sport = inet->inet_sport;
  94. dh->dccph_dport = inet->inet_dport;
  95. dh->dccph_doff = (dccp_header_size + dcb->dccpd_opt_len) / 4;
  96. dh->dccph_ccval = dcb->dccpd_ccval;
  97. dh->dccph_cscov = dp->dccps_pcslen;
  98. /* XXX For now we're using only 48 bits sequence numbers */
  99. dh->dccph_x = 1;
  100. dccp_update_gss(sk, dcb->dccpd_seq);
  101. dccp_hdr_set_seq(dh, dp->dccps_gss);
  102. if (set_ack)
  103. dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
  104. switch (dcb->dccpd_type) {
  105. case DCCP_PKT_REQUEST:
  106. dccp_hdr_request(skb)->dccph_req_service =
  107. dp->dccps_service;
  108. /*
  109. * Limit Ack window to ISS <= P.ackno <= GSS, so that
  110. * only Responses to Requests we sent are considered.
  111. */
  112. dp->dccps_awl = dp->dccps_iss;
  113. break;
  114. case DCCP_PKT_RESET:
  115. dccp_hdr_reset(skb)->dccph_reset_code =
  116. dcb->dccpd_reset_code;
  117. break;
  118. }
  119. icsk->icsk_af_ops->send_check(sk, skb);
  120. if (set_ack)
  121. dccp_event_ack_sent(sk);
  122. DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
  123. err = icsk->icsk_af_ops->queue_xmit(skb, &inet->cork.fl);
  124. return net_xmit_eval(err);
  125. }
  126. return -ENOBUFS;
  127. }
  128. /**
  129. * dccp_determine_ccmps - Find out about CCID-specific packet-size limits
  130. * We only consider the HC-sender CCID for setting the CCMPS (RFC 4340, 14.),
  131. * since the RX CCID is restricted to feedback packets (Acks), which are small
  132. * in comparison with the data traffic. A value of 0 means "no current CCMPS".
  133. */
  134. static u32 dccp_determine_ccmps(const struct dccp_sock *dp)
  135. {
  136. const struct ccid *tx_ccid = dp->dccps_hc_tx_ccid;
  137. if (tx_ccid == NULL || tx_ccid->ccid_ops == NULL)
  138. return 0;
  139. return tx_ccid->ccid_ops->ccid_ccmps;
  140. }
  141. unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
  142. {
  143. struct inet_connection_sock *icsk = inet_csk(sk);
  144. struct dccp_sock *dp = dccp_sk(sk);
  145. u32 ccmps = dccp_determine_ccmps(dp);
  146. u32 cur_mps = ccmps ? min(pmtu, ccmps) : pmtu;
  147. /* Account for header lengths and IPv4/v6 option overhead */
  148. cur_mps -= (icsk->icsk_af_ops->net_header_len + icsk->icsk_ext_hdr_len +
  149. sizeof(struct dccp_hdr) + sizeof(struct dccp_hdr_ext));
  150. /*
  151. * Leave enough headroom for common DCCP header options.
  152. * This only considers options which may appear on DCCP-Data packets, as
  153. * per table 3 in RFC 4340, 5.8. When running out of space for other
  154. * options (eg. Ack Vector which can take up to 255 bytes), it is better
  155. * to schedule a separate Ack. Thus we leave headroom for the following:
  156. * - 1 byte for Slow Receiver (11.6)
  157. * - 6 bytes for Timestamp (13.1)
  158. * - 10 bytes for Timestamp Echo (13.3)
  159. * - 8 bytes for NDP count (7.7, when activated)
  160. * - 6 bytes for Data Checksum (9.3)
  161. * - %DCCPAV_MIN_OPTLEN bytes for Ack Vector size (11.4, when enabled)
  162. */
  163. cur_mps -= roundup(1 + 6 + 10 + dp->dccps_send_ndp_count * 8 + 6 +
  164. (dp->dccps_hc_rx_ackvec ? DCCPAV_MIN_OPTLEN : 0), 4);
  165. /* And store cached results */
  166. icsk->icsk_pmtu_cookie = pmtu;
  167. dp->dccps_mss_cache = cur_mps;
  168. return cur_mps;
  169. }
  170. EXPORT_SYMBOL_GPL(dccp_sync_mss);
  171. void dccp_write_space(struct sock *sk)
  172. {
  173. struct socket_wq *wq;
  174. rcu_read_lock();
  175. wq = rcu_dereference(sk->sk_wq);
  176. if (wq_has_sleeper(wq))
  177. wake_up_interruptible(&wq->wait);
  178. /* Should agree with poll, otherwise some programs break */
  179. if (sock_writeable(sk))
  180. sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
  181. rcu_read_unlock();
  182. }
  183. /**
  184. * dccp_wait_for_ccid - Await CCID send permission
  185. * @sk: socket to wait for
  186. * @delay: timeout in jiffies
  187. *
  188. * This is used by CCIDs which need to delay the send time in process context.
  189. */
  190. static int dccp_wait_for_ccid(struct sock *sk, unsigned long delay)
  191. {
  192. DEFINE_WAIT(wait);
  193. long remaining;
  194. prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
  195. sk->sk_write_pending++;
  196. release_sock(sk);
  197. remaining = schedule_timeout(delay);
  198. lock_sock(sk);
  199. sk->sk_write_pending--;
  200. finish_wait(sk_sleep(sk), &wait);
  201. if (signal_pending(current) || sk->sk_err)
  202. return -1;
  203. return remaining;
  204. }
  205. /**
  206. * dccp_xmit_packet - Send data packet under control of CCID
  207. * Transmits next-queued payload and informs CCID to account for the packet.
  208. */
  209. static void dccp_xmit_packet(struct sock *sk)
  210. {
  211. int err, len;
  212. struct dccp_sock *dp = dccp_sk(sk);
  213. struct sk_buff *skb = dccp_qpolicy_pop(sk);
  214. if (unlikely(skb == NULL))
  215. return;
  216. len = skb->len;
  217. if (sk->sk_state == DCCP_PARTOPEN) {
  218. const u32 cur_mps = dp->dccps_mss_cache - DCCP_FEATNEG_OVERHEAD;
  219. /*
  220. * See 8.1.5 - Handshake Completion.
  221. *
  222. * For robustness we resend Confirm options until the client has
  223. * entered OPEN. During the initial feature negotiation, the MPS
  224. * is smaller than usual, reduced by the Change/Confirm options.
  225. */
  226. if (!list_empty(&dp->dccps_featneg) && len > cur_mps) {
  227. DCCP_WARN("Payload too large (%d) for featneg.\n", len);
  228. dccp_send_ack(sk);
  229. dccp_feat_list_purge(&dp->dccps_featneg);
  230. }
  231. inet_csk_schedule_ack(sk);
  232. inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
  233. inet_csk(sk)->icsk_rto,
  234. DCCP_RTO_MAX);
  235. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
  236. } else if (dccp_ack_pending(sk)) {
  237. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
  238. } else {
  239. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATA;
  240. }
  241. err = dccp_transmit_skb(sk, skb);
  242. if (err)
  243. dccp_pr_debug("transmit_skb() returned err=%d\n", err);
  244. /*
  245. * Register this one as sent even if an error occurred. To the remote
  246. * end a local packet drop is indistinguishable from network loss, i.e.
  247. * any local drop will eventually be reported via receiver feedback.
  248. */
  249. ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, len);
  250. /*
  251. * If the CCID needs to transfer additional header options out-of-band
  252. * (e.g. Ack Vectors or feature-negotiation options), it activates this
  253. * flag to schedule a Sync. The Sync will automatically incorporate all
  254. * currently pending header options, thus clearing the backlog.
  255. */
  256. if (dp->dccps_sync_scheduled)
  257. dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
  258. }
  259. /**
  260. * dccp_flush_write_queue - Drain queue at end of connection
  261. * Since dccp_sendmsg queues packets without waiting for them to be sent, it may
  262. * happen that the TX queue is not empty at the end of a connection. We give the
  263. * HC-sender CCID a grace period of up to @time_budget jiffies. If this function
  264. * returns with a non-empty write queue, it will be purged later.
  265. */
  266. void dccp_flush_write_queue(struct sock *sk, long *time_budget)
  267. {
  268. struct dccp_sock *dp = dccp_sk(sk);
  269. struct sk_buff *skb;
  270. long delay, rc;
  271. while (*time_budget > 0 && (skb = skb_peek(&sk->sk_write_queue))) {
  272. rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
  273. switch (ccid_packet_dequeue_eval(rc)) {
  274. case CCID_PACKET_WILL_DEQUEUE_LATER:
  275. /*
  276. * If the CCID determines when to send, the next sending
  277. * time is unknown or the CCID may not even send again
  278. * (e.g. remote host crashes or lost Ack packets).
  279. */
  280. DCCP_WARN("CCID did not manage to send all packets\n");
  281. return;
  282. case CCID_PACKET_DELAY:
  283. delay = msecs_to_jiffies(rc);
  284. if (delay > *time_budget)
  285. return;
  286. rc = dccp_wait_for_ccid(sk, delay);
  287. if (rc < 0)
  288. return;
  289. *time_budget -= (delay - rc);
  290. /* check again if we can send now */
  291. break;
  292. case CCID_PACKET_SEND_AT_ONCE:
  293. dccp_xmit_packet(sk);
  294. break;
  295. case CCID_PACKET_ERR:
  296. skb_dequeue(&sk->sk_write_queue);
  297. kfree_skb(skb);
  298. dccp_pr_debug("packet discarded due to err=%ld\n", rc);
  299. }
  300. }
  301. }
  302. void dccp_write_xmit(struct sock *sk)
  303. {
  304. struct dccp_sock *dp = dccp_sk(sk);
  305. struct sk_buff *skb;
  306. while ((skb = dccp_qpolicy_top(sk))) {
  307. int rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
  308. switch (ccid_packet_dequeue_eval(rc)) {
  309. case CCID_PACKET_WILL_DEQUEUE_LATER:
  310. return;
  311. case CCID_PACKET_DELAY:
  312. sk_reset_timer(sk, &dp->dccps_xmit_timer,
  313. jiffies + msecs_to_jiffies(rc));
  314. return;
  315. case CCID_PACKET_SEND_AT_ONCE:
  316. dccp_xmit_packet(sk);
  317. break;
  318. case CCID_PACKET_ERR:
  319. dccp_qpolicy_drop(sk, skb);
  320. dccp_pr_debug("packet discarded due to err=%d\n", rc);
  321. }
  322. }
  323. }
  324. /**
  325. * dccp_retransmit_skb - Retransmit Request, Close, or CloseReq packets
  326. * There are only four retransmittable packet types in DCCP:
  327. * - Request in client-REQUEST state (sec. 8.1.1),
  328. * - CloseReq in server-CLOSEREQ state (sec. 8.3),
  329. * - Close in node-CLOSING state (sec. 8.3),
  330. * - Acks in client-PARTOPEN state (sec. 8.1.5, handled by dccp_delack_timer()).
  331. * This function expects sk->sk_send_head to contain the original skb.
  332. */
  333. int dccp_retransmit_skb(struct sock *sk)
  334. {
  335. WARN_ON(sk->sk_send_head == NULL);
  336. if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
  337. return -EHOSTUNREACH; /* Routing failure or similar. */
  338. /* this count is used to distinguish original and retransmitted skb */
  339. inet_csk(sk)->icsk_retransmits++;
  340. return dccp_transmit_skb(sk, skb_clone(sk->sk_send_head, GFP_ATOMIC));
  341. }
  342. struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
  343. struct request_sock *req)
  344. {
  345. struct dccp_hdr *dh;
  346. struct dccp_request_sock *dreq;
  347. const u32 dccp_header_size = sizeof(struct dccp_hdr) +
  348. sizeof(struct dccp_hdr_ext) +
  349. sizeof(struct dccp_hdr_response);
  350. struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
  351. GFP_ATOMIC);
  352. if (skb == NULL)
  353. return NULL;
  354. /* Reserve space for headers. */
  355. skb_reserve(skb, sk->sk_prot->max_header);
  356. skb_dst_set(skb, dst_clone(dst));
  357. dreq = dccp_rsk(req);
  358. if (inet_rsk(req)->acked) /* increase GSS upon retransmission */
  359. dccp_inc_seqno(&dreq->dreq_gss);
  360. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
  361. DCCP_SKB_CB(skb)->dccpd_seq = dreq->dreq_gss;
  362. /* Resolve feature dependencies resulting from choice of CCID */
  363. if (dccp_feat_server_ccid_dependencies(dreq))
  364. goto response_failed;
  365. if (dccp_insert_options_rsk(dreq, skb))
  366. goto response_failed;
  367. /* Build and checksum header */
  368. dh = dccp_zeroed_hdr(skb, dccp_header_size);
  369. dh->dccph_sport = inet_rsk(req)->loc_port;
  370. dh->dccph_dport = inet_rsk(req)->rmt_port;
  371. dh->dccph_doff = (dccp_header_size +
  372. DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
  373. dh->dccph_type = DCCP_PKT_RESPONSE;
  374. dh->dccph_x = 1;
  375. dccp_hdr_set_seq(dh, dreq->dreq_gss);
  376. dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_gsr);
  377. dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
  378. dccp_csum_outgoing(skb);
  379. /* We use `acked' to remember that a Response was already sent. */
  380. inet_rsk(req)->acked = 1;
  381. DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
  382. return skb;
  383. response_failed:
  384. kfree_skb(skb);
  385. return NULL;
  386. }
  387. EXPORT_SYMBOL_GPL(dccp_make_response);
  388. /* answer offending packet in @rcv_skb with Reset from control socket @ctl */
  389. struct sk_buff *dccp_ctl_make_reset(struct sock *sk, struct sk_buff *rcv_skb)
  390. {
  391. struct dccp_hdr *rxdh = dccp_hdr(rcv_skb), *dh;
  392. struct dccp_skb_cb *dcb = DCCP_SKB_CB(rcv_skb);
  393. const u32 dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
  394. sizeof(struct dccp_hdr_ext) +
  395. sizeof(struct dccp_hdr_reset);
  396. struct dccp_hdr_reset *dhr;
  397. struct sk_buff *skb;
  398. skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
  399. if (skb == NULL)
  400. return NULL;
  401. skb_reserve(skb, sk->sk_prot->max_header);
  402. /* Swap the send and the receive. */
  403. dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
  404. dh->dccph_type = DCCP_PKT_RESET;
  405. dh->dccph_sport = rxdh->dccph_dport;
  406. dh->dccph_dport = rxdh->dccph_sport;
  407. dh->dccph_doff = dccp_hdr_reset_len / 4;
  408. dh->dccph_x = 1;
  409. dhr = dccp_hdr_reset(skb);
  410. dhr->dccph_reset_code = dcb->dccpd_reset_code;
  411. switch (dcb->dccpd_reset_code) {
  412. case DCCP_RESET_CODE_PACKET_ERROR:
  413. dhr->dccph_reset_data[0] = rxdh->dccph_type;
  414. break;
  415. case DCCP_RESET_CODE_OPTION_ERROR: /* fall through */
  416. case DCCP_RESET_CODE_MANDATORY_ERROR:
  417. memcpy(dhr->dccph_reset_data, dcb->dccpd_reset_data, 3);
  418. break;
  419. }
  420. /*
  421. * From RFC 4340, 8.3.1:
  422. * If P.ackno exists, set R.seqno := P.ackno + 1.
  423. * Else set R.seqno := 0.
  424. */
  425. if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
  426. dccp_hdr_set_seq(dh, ADD48(dcb->dccpd_ack_seq, 1));
  427. dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dcb->dccpd_seq);
  428. dccp_csum_outgoing(skb);
  429. return skb;
  430. }
  431. EXPORT_SYMBOL_GPL(dccp_ctl_make_reset);
  432. /* send Reset on established socket, to close or abort the connection */
  433. int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
  434. {
  435. struct sk_buff *skb;
  436. /*
  437. * FIXME: what if rebuild_header fails?
  438. * Should we be doing a rebuild_header here?
  439. */
  440. int err = inet_csk(sk)->icsk_af_ops->rebuild_header(sk);
  441. if (err != 0)
  442. return err;
  443. skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
  444. if (skb == NULL)
  445. return -ENOBUFS;
  446. /* Reserve space for headers and prepare control bits. */
  447. skb_reserve(skb, sk->sk_prot->max_header);
  448. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESET;
  449. DCCP_SKB_CB(skb)->dccpd_reset_code = code;
  450. return dccp_transmit_skb(sk, skb);
  451. }
  452. /*
  453. * Do all connect socket setups that can be done AF independent.
  454. */
  455. int dccp_connect(struct sock *sk)
  456. {
  457. struct sk_buff *skb;
  458. struct dccp_sock *dp = dccp_sk(sk);
  459. struct dst_entry *dst = __sk_dst_get(sk);
  460. struct inet_connection_sock *icsk = inet_csk(sk);
  461. sk->sk_err = 0;
  462. sock_reset_flag(sk, SOCK_DONE);
  463. dccp_sync_mss(sk, dst_mtu(dst));
  464. /* do not connect if feature negotiation setup fails */
  465. if (dccp_feat_finalise_settings(dccp_sk(sk)))
  466. return -EPROTO;
  467. /* Initialise GAR as per 8.5; AWL/AWH are set in dccp_transmit_skb() */
  468. dp->dccps_gar = dp->dccps_iss;
  469. skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
  470. if (unlikely(skb == NULL))
  471. return -ENOBUFS;
  472. /* Reserve space for headers. */
  473. skb_reserve(skb, sk->sk_prot->max_header);
  474. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
  475. dccp_transmit_skb(sk, dccp_skb_entail(sk, skb));
  476. DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
  477. /* Timer for repeating the REQUEST until an answer. */
  478. icsk->icsk_retransmits = 0;
  479. inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
  480. icsk->icsk_rto, DCCP_RTO_MAX);
  481. return 0;
  482. }
  483. EXPORT_SYMBOL_GPL(dccp_connect);
  484. void dccp_send_ack(struct sock *sk)
  485. {
  486. /* If we have been reset, we may not send again. */
  487. if (sk->sk_state != DCCP_CLOSED) {
  488. struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
  489. GFP_ATOMIC);
  490. if (skb == NULL) {
  491. inet_csk_schedule_ack(sk);
  492. inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
  493. inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
  494. TCP_DELACK_MAX,
  495. DCCP_RTO_MAX);
  496. return;
  497. }
  498. /* Reserve space for headers */
  499. skb_reserve(skb, sk->sk_prot->max_header);
  500. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
  501. dccp_transmit_skb(sk, skb);
  502. }
  503. }
  504. EXPORT_SYMBOL_GPL(dccp_send_ack);
  505. #if 0
  506. /* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
  507. void dccp_send_delayed_ack(struct sock *sk)
  508. {
  509. struct inet_connection_sock *icsk = inet_csk(sk);
  510. /*
  511. * FIXME: tune this timer. elapsed time fixes the skew, so no problem
  512. * with using 2s, and active senders also piggyback the ACK into a
  513. * DATAACK packet, so this is really for quiescent senders.
  514. */
  515. unsigned long timeout = jiffies + 2 * HZ;
  516. /* Use new timeout only if there wasn't a older one earlier. */
  517. if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
  518. /* If delack timer was blocked or is about to expire,
  519. * send ACK now.
  520. *
  521. * FIXME: check the "about to expire" part
  522. */
  523. if (icsk->icsk_ack.blocked) {
  524. dccp_send_ack(sk);
  525. return;
  526. }
  527. if (!time_before(timeout, icsk->icsk_ack.timeout))
  528. timeout = icsk->icsk_ack.timeout;
  529. }
  530. icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
  531. icsk->icsk_ack.timeout = timeout;
  532. sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
  533. }
  534. #endif
  535. void dccp_send_sync(struct sock *sk, const u64 ackno,
  536. const enum dccp_pkt_type pkt_type)
  537. {
  538. /*
  539. * We are not putting this on the write queue, so
  540. * dccp_transmit_skb() will set the ownership to this
  541. * sock.
  542. */
  543. struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
  544. if (skb == NULL) {
  545. /* FIXME: how to make sure the sync is sent? */
  546. DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
  547. return;
  548. }
  549. /* Reserve space for headers and prepare control bits. */
  550. skb_reserve(skb, sk->sk_prot->max_header);
  551. DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
  552. DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
  553. /*
  554. * Clear the flag in case the Sync was scheduled for out-of-band data,
  555. * such as carrying a long Ack Vector.
  556. */
  557. dccp_sk(sk)->dccps_sync_scheduled = 0;
  558. dccp_transmit_skb(sk, skb);
  559. }
  560. EXPORT_SYMBOL_GPL(dccp_send_sync);
  561. /*
  562. * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
  563. * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
  564. * any circumstances.
  565. */
  566. void dccp_send_close(struct sock *sk, const int active)
  567. {
  568. struct dccp_sock *dp = dccp_sk(sk);
  569. struct sk_buff *skb;
  570. const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
  571. skb = alloc_skb(sk->sk_prot->max_header, prio);
  572. if (skb == NULL)
  573. return;
  574. /* Reserve space for headers and prepare control bits. */
  575. skb_reserve(skb, sk->sk_prot->max_header);
  576. if (dp->dccps_role == DCCP_ROLE_SERVER && !dp->dccps_server_timewait)
  577. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSEREQ;
  578. else
  579. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSE;
  580. if (active) {
  581. skb = dccp_skb_entail(sk, skb);
  582. /*
  583. * Retransmission timer for active-close: RFC 4340, 8.3 requires
  584. * to retransmit the Close/CloseReq until the CLOSING/CLOSEREQ
  585. * state can be left. The initial timeout is 2 RTTs.
  586. * Since RTT measurement is done by the CCIDs, there is no easy
  587. * way to get an RTT sample. The fallback RTT from RFC 4340, 3.4
  588. * is too low (200ms); we use a high value to avoid unnecessary
  589. * retransmissions when the link RTT is > 0.2 seconds.
  590. * FIXME: Let main module sample RTTs and use that instead.
  591. */
  592. inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
  593. DCCP_TIMEOUT_INIT, DCCP_RTO_MAX);
  594. }
  595. dccp_transmit_skb(sk, skb);
  596. }