output.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695
  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 <net/inet_sock.h>
  16. #include <net/sock.h>
  17. #include "ackvec.h"
  18. #include "ccid.h"
  19. #include "dccp.h"
  20. static inline void dccp_event_ack_sent(struct sock *sk)
  21. {
  22. inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
  23. }
  24. /* enqueue @skb on sk_send_head for retransmission, return clone to send now */
  25. static struct sk_buff *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. return skb_clone(sk->sk_send_head, gfp_any());
  31. }
  32. /*
  33. * All SKB's seen here are completely headerless. It is our
  34. * job to build the DCCP header, and pass the packet down to
  35. * IP so it can do the same plus pass the packet off to the
  36. * device.
  37. */
  38. static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
  39. {
  40. if (likely(skb != NULL)) {
  41. const struct inet_sock *inet = inet_sk(sk);
  42. const struct inet_connection_sock *icsk = inet_csk(sk);
  43. struct dccp_sock *dp = dccp_sk(sk);
  44. struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
  45. struct dccp_hdr *dh;
  46. /* XXX For now we're using only 48 bits sequence numbers */
  47. const u32 dccp_header_size = sizeof(*dh) +
  48. sizeof(struct dccp_hdr_ext) +
  49. dccp_packet_hdr_len(dcb->dccpd_type);
  50. int err, set_ack = 1;
  51. u64 ackno = dp->dccps_gsr;
  52. /*
  53. * Increment GSS here already in case the option code needs it.
  54. * Update GSS for real only if option processing below succeeds.
  55. */
  56. dcb->dccpd_seq = ADD48(dp->dccps_gss, 1);
  57. switch (dcb->dccpd_type) {
  58. case DCCP_PKT_DATA:
  59. set_ack = 0;
  60. /* fall through */
  61. case DCCP_PKT_DATAACK:
  62. case DCCP_PKT_RESET:
  63. break;
  64. case DCCP_PKT_REQUEST:
  65. set_ack = 0;
  66. /* Use ISS on the first (non-retransmitted) Request. */
  67. if (icsk->icsk_retransmits == 0)
  68. dcb->dccpd_seq = dp->dccps_iss;
  69. /* fall through */
  70. case DCCP_PKT_SYNC:
  71. case DCCP_PKT_SYNCACK:
  72. ackno = dcb->dccpd_ack_seq;
  73. /* fall through */
  74. default:
  75. /*
  76. * Set owner/destructor: some skbs are allocated via
  77. * alloc_skb (e.g. when retransmission may happen).
  78. * Only Data, DataAck, and Reset packets should come
  79. * through here with skb->sk set.
  80. */
  81. WARN_ON(skb->sk);
  82. skb_set_owner_w(skb, sk);
  83. break;
  84. }
  85. if (dccp_insert_options(sk, skb)) {
  86. kfree_skb(skb);
  87. return -EPROTO;
  88. }
  89. /* Build DCCP header and checksum it. */
  90. dh = dccp_zeroed_hdr(skb, dccp_header_size);
  91. dh->dccph_type = dcb->dccpd_type;
  92. dh->dccph_sport = inet->sport;
  93. dh->dccph_dport = inet->dport;
  94. dh->dccph_doff = (dccp_header_size + dcb->dccpd_opt_len) / 4;
  95. dh->dccph_ccval = dcb->dccpd_ccval;
  96. dh->dccph_cscov = dp->dccps_pcslen;
  97. /* XXX For now we're using only 48 bits sequence numbers */
  98. dh->dccph_x = 1;
  99. dccp_update_gss(sk, dcb->dccpd_seq);
  100. dccp_hdr_set_seq(dh, dp->dccps_gss);
  101. if (set_ack)
  102. dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
  103. switch (dcb->dccpd_type) {
  104. case DCCP_PKT_REQUEST:
  105. dccp_hdr_request(skb)->dccph_req_service =
  106. dp->dccps_service;
  107. /*
  108. * Limit Ack window to ISS <= P.ackno <= GSS, so that
  109. * only Responses to Requests we sent are considered.
  110. */
  111. dp->dccps_awl = dp->dccps_iss;
  112. break;
  113. case DCCP_PKT_RESET:
  114. dccp_hdr_reset(skb)->dccph_reset_code =
  115. dcb->dccpd_reset_code;
  116. break;
  117. }
  118. icsk->icsk_af_ops->send_check(sk, 0, skb);
  119. if (set_ack)
  120. dccp_event_ack_sent(sk);
  121. DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
  122. err = icsk->icsk_af_ops->queue_xmit(skb, 0);
  123. return net_xmit_eval(err);
  124. }
  125. return -ENOBUFS;
  126. }
  127. /**
  128. * dccp_determine_ccmps - Find out about CCID-specfic packet-size limits
  129. * We only consider the HC-sender CCID for setting the CCMPS (RFC 4340, 14.),
  130. * since the RX CCID is restricted to feedback packets (Acks), which are small
  131. * in comparison with the data traffic. A value of 0 means "no current CCMPS".
  132. */
  133. static u32 dccp_determine_ccmps(const struct dccp_sock *dp)
  134. {
  135. const struct ccid *tx_ccid = dp->dccps_hc_tx_ccid;
  136. if (tx_ccid == NULL || tx_ccid->ccid_ops == NULL)
  137. return 0;
  138. return tx_ccid->ccid_ops->ccid_ccmps;
  139. }
  140. unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
  141. {
  142. struct inet_connection_sock *icsk = inet_csk(sk);
  143. struct dccp_sock *dp = dccp_sk(sk);
  144. u32 ccmps = dccp_determine_ccmps(dp);
  145. u32 cur_mps = ccmps ? min(pmtu, ccmps) : pmtu;
  146. /* Account for header lengths and IPv4/v6 option overhead */
  147. cur_mps -= (icsk->icsk_af_ops->net_header_len + icsk->icsk_ext_hdr_len +
  148. sizeof(struct dccp_hdr) + sizeof(struct dccp_hdr_ext));
  149. /*
  150. * Leave enough headroom for common DCCP header options.
  151. * This only considers options which may appear on DCCP-Data packets, as
  152. * per table 3 in RFC 4340, 5.8. When running out of space for other
  153. * options (eg. Ack Vector which can take up to 255 bytes), it is better
  154. * to schedule a separate Ack. Thus we leave headroom for the following:
  155. * - 1 byte for Slow Receiver (11.6)
  156. * - 6 bytes for Timestamp (13.1)
  157. * - 10 bytes for Timestamp Echo (13.3)
  158. * - 8 bytes for NDP count (7.7, when activated)
  159. * - 6 bytes for Data Checksum (9.3)
  160. * - %DCCPAV_MIN_OPTLEN bytes for Ack Vector size (11.4, when enabled)
  161. */
  162. cur_mps -= roundup(1 + 6 + 10 + dp->dccps_send_ndp_count * 8 + 6 +
  163. (dp->dccps_hc_rx_ackvec ? DCCPAV_MIN_OPTLEN : 0), 4);
  164. /* And store cached results */
  165. icsk->icsk_pmtu_cookie = pmtu;
  166. dp->dccps_mss_cache = cur_mps;
  167. return cur_mps;
  168. }
  169. EXPORT_SYMBOL_GPL(dccp_sync_mss);
  170. void dccp_write_space(struct sock *sk)
  171. {
  172. read_lock(&sk->sk_callback_lock);
  173. if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
  174. wake_up_interruptible(sk->sk_sleep);
  175. /* Should agree with poll, otherwise some programs break */
  176. if (sock_writeable(sk))
  177. sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
  178. read_unlock(&sk->sk_callback_lock);
  179. }
  180. /**
  181. * dccp_wait_for_ccid - Await CCID send permission
  182. * @sk: socket to wait for
  183. * @delay: timeout in jiffies
  184. * This is used by CCIDs which need to delay the send time in process context.
  185. */
  186. static int dccp_wait_for_ccid(struct sock *sk, unsigned long delay)
  187. {
  188. DEFINE_WAIT(wait);
  189. long remaining;
  190. prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
  191. sk->sk_write_pending++;
  192. release_sock(sk);
  193. remaining = schedule_timeout(delay);
  194. lock_sock(sk);
  195. sk->sk_write_pending--;
  196. finish_wait(sk->sk_sleep, &wait);
  197. if (signal_pending(current) || sk->sk_err)
  198. return -1;
  199. return remaining;
  200. }
  201. /**
  202. * dccp_xmit_packet - Send data packet under control of CCID
  203. * Transmits next-queued payload and informs CCID to account for the packet.
  204. */
  205. static void dccp_xmit_packet(struct sock *sk)
  206. {
  207. int err, len;
  208. struct dccp_sock *dp = dccp_sk(sk);
  209. struct sk_buff *skb = skb_dequeue(&sk->sk_write_queue);
  210. if (unlikely(skb == NULL))
  211. return;
  212. len = skb->len;
  213. if (sk->sk_state == DCCP_PARTOPEN) {
  214. const u32 cur_mps = dp->dccps_mss_cache - DCCP_FEATNEG_OVERHEAD;
  215. /*
  216. * See 8.1.5 - Handshake Completion.
  217. *
  218. * For robustness we resend Confirm options until the client has
  219. * entered OPEN. During the initial feature negotiation, the MPS
  220. * is smaller than usual, reduced by the Change/Confirm options.
  221. */
  222. if (!list_empty(&dp->dccps_featneg) && len > cur_mps) {
  223. DCCP_WARN("Payload too large (%d) for featneg.\n", len);
  224. dccp_send_ack(sk);
  225. dccp_feat_list_purge(&dp->dccps_featneg);
  226. }
  227. inet_csk_schedule_ack(sk);
  228. inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
  229. inet_csk(sk)->icsk_rto,
  230. DCCP_RTO_MAX);
  231. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
  232. } else if (dccp_ack_pending(sk)) {
  233. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
  234. } else {
  235. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATA;
  236. }
  237. err = dccp_transmit_skb(sk, skb);
  238. if (err)
  239. dccp_pr_debug("transmit_skb() returned err=%d\n", err);
  240. /*
  241. * Register this one as sent even if an error occurred. To the remote
  242. * end a local packet drop is indistinguishable from network loss, i.e.
  243. * any local drop will eventually be reported via receiver feedback.
  244. */
  245. ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, len);
  246. /*
  247. * If the CCID needs to transfer additional header options out-of-band
  248. * (e.g. Ack Vectors or feature-negotiation options), it activates this
  249. * flag to schedule a Sync. The Sync will automatically incorporate all
  250. * currently pending header options, thus clearing the backlog.
  251. */
  252. if (dp->dccps_sync_scheduled)
  253. dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
  254. }
  255. /**
  256. * dccp_flush_write_queue - Drain queue at end of connection
  257. * Since dccp_sendmsg queues packets without waiting for them to be sent, it may
  258. * happen that the TX queue is not empty at the end of a connection. We give the
  259. * HC-sender CCID a grace period of up to @time_budget jiffies. If this function
  260. * returns with a non-empty write queue, it will be purged later.
  261. */
  262. void dccp_flush_write_queue(struct sock *sk, long *time_budget)
  263. {
  264. struct dccp_sock *dp = dccp_sk(sk);
  265. struct sk_buff *skb;
  266. long delay, rc;
  267. while (*time_budget > 0 && (skb = skb_peek(&sk->sk_write_queue))) {
  268. rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
  269. switch (ccid_packet_dequeue_eval(rc)) {
  270. case CCID_PACKET_WILL_DEQUEUE_LATER:
  271. /*
  272. * If the CCID determines when to send, the next sending
  273. * time is unknown or the CCID may not even send again
  274. * (e.g. remote host crashes or lost Ack packets).
  275. */
  276. DCCP_WARN("CCID did not manage to send all packets\n");
  277. return;
  278. case CCID_PACKET_DELAY:
  279. delay = msecs_to_jiffies(rc);
  280. if (delay > *time_budget)
  281. return;
  282. rc = dccp_wait_for_ccid(sk, delay);
  283. if (rc < 0)
  284. return;
  285. *time_budget -= (delay - rc);
  286. /* check again if we can send now */
  287. break;
  288. case CCID_PACKET_SEND_AT_ONCE:
  289. dccp_xmit_packet(sk);
  290. break;
  291. case CCID_PACKET_ERR:
  292. skb_dequeue(&sk->sk_write_queue);
  293. kfree_skb(skb);
  294. dccp_pr_debug("packet discarded due to err=%ld\n", rc);
  295. }
  296. }
  297. }
  298. void dccp_write_xmit(struct sock *sk)
  299. {
  300. struct dccp_sock *dp = dccp_sk(sk);
  301. struct sk_buff *skb;
  302. while ((skb = skb_peek(&sk->sk_write_queue))) {
  303. int rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
  304. switch (ccid_packet_dequeue_eval(rc)) {
  305. case CCID_PACKET_WILL_DEQUEUE_LATER:
  306. return;
  307. case CCID_PACKET_DELAY:
  308. sk_reset_timer(sk, &dp->dccps_xmit_timer,
  309. jiffies + msecs_to_jiffies(rc));
  310. return;
  311. case CCID_PACKET_SEND_AT_ONCE:
  312. dccp_xmit_packet(sk);
  313. break;
  314. case CCID_PACKET_ERR:
  315. skb_dequeue(&sk->sk_write_queue);
  316. kfree_skb(skb);
  317. dccp_pr_debug("packet discarded due to err=%d\n", rc);
  318. }
  319. }
  320. }
  321. /**
  322. * dccp_retransmit_skb - Retransmit Request, Close, or CloseReq packets
  323. * There are only four retransmittable packet types in DCCP:
  324. * - Request in client-REQUEST state (sec. 8.1.1),
  325. * - CloseReq in server-CLOSEREQ state (sec. 8.3),
  326. * - Close in node-CLOSING state (sec. 8.3),
  327. * - Acks in client-PARTOPEN state (sec. 8.1.5, handled by dccp_delack_timer()).
  328. * This function expects sk->sk_send_head to contain the original skb.
  329. */
  330. int dccp_retransmit_skb(struct sock *sk)
  331. {
  332. WARN_ON(sk->sk_send_head == NULL);
  333. if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
  334. return -EHOSTUNREACH; /* Routing failure or similar. */
  335. /* this count is used to distinguish original and retransmitted skb */
  336. inet_csk(sk)->icsk_retransmits++;
  337. return dccp_transmit_skb(sk, skb_clone(sk->sk_send_head, GFP_ATOMIC));
  338. }
  339. struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
  340. struct request_sock *req)
  341. {
  342. struct dccp_hdr *dh;
  343. struct dccp_request_sock *dreq;
  344. const u32 dccp_header_size = sizeof(struct dccp_hdr) +
  345. sizeof(struct dccp_hdr_ext) +
  346. sizeof(struct dccp_hdr_response);
  347. struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
  348. GFP_ATOMIC);
  349. if (skb == NULL)
  350. return NULL;
  351. /* Reserve space for headers. */
  352. skb_reserve(skb, sk->sk_prot->max_header);
  353. skb->dst = dst_clone(dst);
  354. dreq = dccp_rsk(req);
  355. if (inet_rsk(req)->acked) /* increase ISS upon retransmission */
  356. dccp_inc_seqno(&dreq->dreq_iss);
  357. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
  358. DCCP_SKB_CB(skb)->dccpd_seq = dreq->dreq_iss;
  359. /* Resolve feature dependencies resulting from choice of CCID */
  360. if (dccp_feat_server_ccid_dependencies(dreq))
  361. goto response_failed;
  362. if (dccp_insert_options_rsk(dreq, skb))
  363. goto response_failed;
  364. /* Build and checksum header */
  365. dh = dccp_zeroed_hdr(skb, dccp_header_size);
  366. dh->dccph_sport = inet_sk(sk)->sport;
  367. dh->dccph_dport = inet_rsk(req)->rmt_port;
  368. dh->dccph_doff = (dccp_header_size +
  369. DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
  370. dh->dccph_type = DCCP_PKT_RESPONSE;
  371. dh->dccph_x = 1;
  372. dccp_hdr_set_seq(dh, dreq->dreq_iss);
  373. dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
  374. dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
  375. dccp_csum_outgoing(skb);
  376. /* We use `acked' to remember that a Response was already sent. */
  377. inet_rsk(req)->acked = 1;
  378. DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
  379. return skb;
  380. response_failed:
  381. kfree_skb(skb);
  382. return NULL;
  383. }
  384. EXPORT_SYMBOL_GPL(dccp_make_response);
  385. /* answer offending packet in @rcv_skb with Reset from control socket @ctl */
  386. struct sk_buff *dccp_ctl_make_reset(struct sock *sk, struct sk_buff *rcv_skb)
  387. {
  388. struct dccp_hdr *rxdh = dccp_hdr(rcv_skb), *dh;
  389. struct dccp_skb_cb *dcb = DCCP_SKB_CB(rcv_skb);
  390. const u32 dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
  391. sizeof(struct dccp_hdr_ext) +
  392. sizeof(struct dccp_hdr_reset);
  393. struct dccp_hdr_reset *dhr;
  394. struct sk_buff *skb;
  395. skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
  396. if (skb == NULL)
  397. return NULL;
  398. skb_reserve(skb, sk->sk_prot->max_header);
  399. /* Swap the send and the receive. */
  400. dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
  401. dh->dccph_type = DCCP_PKT_RESET;
  402. dh->dccph_sport = rxdh->dccph_dport;
  403. dh->dccph_dport = rxdh->dccph_sport;
  404. dh->dccph_doff = dccp_hdr_reset_len / 4;
  405. dh->dccph_x = 1;
  406. dhr = dccp_hdr_reset(skb);
  407. dhr->dccph_reset_code = dcb->dccpd_reset_code;
  408. switch (dcb->dccpd_reset_code) {
  409. case DCCP_RESET_CODE_PACKET_ERROR:
  410. dhr->dccph_reset_data[0] = rxdh->dccph_type;
  411. break;
  412. case DCCP_RESET_CODE_OPTION_ERROR: /* fall through */
  413. case DCCP_RESET_CODE_MANDATORY_ERROR:
  414. memcpy(dhr->dccph_reset_data, dcb->dccpd_reset_data, 3);
  415. break;
  416. }
  417. /*
  418. * From RFC 4340, 8.3.1:
  419. * If P.ackno exists, set R.seqno := P.ackno + 1.
  420. * Else set R.seqno := 0.
  421. */
  422. if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
  423. dccp_hdr_set_seq(dh, ADD48(dcb->dccpd_ack_seq, 1));
  424. dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dcb->dccpd_seq);
  425. dccp_csum_outgoing(skb);
  426. return skb;
  427. }
  428. EXPORT_SYMBOL_GPL(dccp_ctl_make_reset);
  429. /* send Reset on established socket, to close or abort the connection */
  430. int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
  431. {
  432. struct sk_buff *skb;
  433. /*
  434. * FIXME: what if rebuild_header fails?
  435. * Should we be doing a rebuild_header here?
  436. */
  437. int err = inet_csk(sk)->icsk_af_ops->rebuild_header(sk);
  438. if (err != 0)
  439. return err;
  440. skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
  441. if (skb == NULL)
  442. return -ENOBUFS;
  443. /* Reserve space for headers and prepare control bits. */
  444. skb_reserve(skb, sk->sk_prot->max_header);
  445. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESET;
  446. DCCP_SKB_CB(skb)->dccpd_reset_code = code;
  447. return dccp_transmit_skb(sk, skb);
  448. }
  449. /*
  450. * Do all connect socket setups that can be done AF independent.
  451. */
  452. int dccp_connect(struct sock *sk)
  453. {
  454. struct sk_buff *skb;
  455. struct dccp_sock *dp = dccp_sk(sk);
  456. struct dst_entry *dst = __sk_dst_get(sk);
  457. struct inet_connection_sock *icsk = inet_csk(sk);
  458. sk->sk_err = 0;
  459. sock_reset_flag(sk, SOCK_DONE);
  460. dccp_sync_mss(sk, dst_mtu(dst));
  461. /* do not connect if feature negotiation setup fails */
  462. if (dccp_feat_finalise_settings(dccp_sk(sk)))
  463. return -EPROTO;
  464. /* Initialise GAR as per 8.5; AWL/AWH are set in dccp_transmit_skb() */
  465. dp->dccps_gar = dp->dccps_iss;
  466. skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
  467. if (unlikely(skb == NULL))
  468. return -ENOBUFS;
  469. /* Reserve space for headers. */
  470. skb_reserve(skb, sk->sk_prot->max_header);
  471. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
  472. dccp_transmit_skb(sk, dccp_skb_entail(sk, skb));
  473. DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
  474. /* Timer for repeating the REQUEST until an answer. */
  475. icsk->icsk_retransmits = 0;
  476. inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
  477. icsk->icsk_rto, DCCP_RTO_MAX);
  478. return 0;
  479. }
  480. EXPORT_SYMBOL_GPL(dccp_connect);
  481. void dccp_send_ack(struct sock *sk)
  482. {
  483. /* If we have been reset, we may not send again. */
  484. if (sk->sk_state != DCCP_CLOSED) {
  485. struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
  486. GFP_ATOMIC);
  487. if (skb == NULL) {
  488. inet_csk_schedule_ack(sk);
  489. inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
  490. inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
  491. TCP_DELACK_MAX,
  492. DCCP_RTO_MAX);
  493. return;
  494. }
  495. /* Reserve space for headers */
  496. skb_reserve(skb, sk->sk_prot->max_header);
  497. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
  498. dccp_transmit_skb(sk, skb);
  499. }
  500. }
  501. EXPORT_SYMBOL_GPL(dccp_send_ack);
  502. #if 0
  503. /* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
  504. void dccp_send_delayed_ack(struct sock *sk)
  505. {
  506. struct inet_connection_sock *icsk = inet_csk(sk);
  507. /*
  508. * FIXME: tune this timer. elapsed time fixes the skew, so no problem
  509. * with using 2s, and active senders also piggyback the ACK into a
  510. * DATAACK packet, so this is really for quiescent senders.
  511. */
  512. unsigned long timeout = jiffies + 2 * HZ;
  513. /* Use new timeout only if there wasn't a older one earlier. */
  514. if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
  515. /* If delack timer was blocked or is about to expire,
  516. * send ACK now.
  517. *
  518. * FIXME: check the "about to expire" part
  519. */
  520. if (icsk->icsk_ack.blocked) {
  521. dccp_send_ack(sk);
  522. return;
  523. }
  524. if (!time_before(timeout, icsk->icsk_ack.timeout))
  525. timeout = icsk->icsk_ack.timeout;
  526. }
  527. icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
  528. icsk->icsk_ack.timeout = timeout;
  529. sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
  530. }
  531. #endif
  532. void dccp_send_sync(struct sock *sk, const u64 ackno,
  533. const enum dccp_pkt_type pkt_type)
  534. {
  535. /*
  536. * We are not putting this on the write queue, so
  537. * dccp_transmit_skb() will set the ownership to this
  538. * sock.
  539. */
  540. struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
  541. if (skb == NULL) {
  542. /* FIXME: how to make sure the sync is sent? */
  543. DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
  544. return;
  545. }
  546. /* Reserve space for headers and prepare control bits. */
  547. skb_reserve(skb, sk->sk_prot->max_header);
  548. DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
  549. DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
  550. /*
  551. * Clear the flag in case the Sync was scheduled for out-of-band data,
  552. * such as carrying a long Ack Vector.
  553. */
  554. dccp_sk(sk)->dccps_sync_scheduled = 0;
  555. dccp_transmit_skb(sk, skb);
  556. }
  557. EXPORT_SYMBOL_GPL(dccp_send_sync);
  558. /*
  559. * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
  560. * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
  561. * any circumstances.
  562. */
  563. void dccp_send_close(struct sock *sk, const int active)
  564. {
  565. struct dccp_sock *dp = dccp_sk(sk);
  566. struct sk_buff *skb;
  567. const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
  568. skb = alloc_skb(sk->sk_prot->max_header, prio);
  569. if (skb == NULL)
  570. return;
  571. /* Reserve space for headers and prepare control bits. */
  572. skb_reserve(skb, sk->sk_prot->max_header);
  573. if (dp->dccps_role == DCCP_ROLE_SERVER && !dp->dccps_server_timewait)
  574. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSEREQ;
  575. else
  576. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSE;
  577. if (active) {
  578. skb = dccp_skb_entail(sk, skb);
  579. /*
  580. * Retransmission timer for active-close: RFC 4340, 8.3 requires
  581. * to retransmit the Close/CloseReq until the CLOSING/CLOSEREQ
  582. * state can be left. The initial timeout is 2 RTTs.
  583. * Since RTT measurement is done by the CCIDs, there is no easy
  584. * way to get an RTT sample. The fallback RTT from RFC 4340, 3.4
  585. * is too low (200ms); we use a high value to avoid unnecessary
  586. * retransmissions when the link RTT is > 0.2 seconds.
  587. * FIXME: Let main module sample RTTs and use that instead.
  588. */
  589. inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
  590. DCCP_TIMEOUT_INIT, DCCP_RTO_MAX);
  591. }
  592. dccp_transmit_skb(sk, skb);
  593. }