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