output.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709
  1. // SPDX-License-Identifier: GPL-2.0-or-later
  2. /*
  3. * net/dccp/output.c
  4. *
  5. * An implementation of the DCCP protocol
  6. * Arnaldo Carvalho de Melo <[email protected]>
  7. */
  8. #include <linux/dccp.h>
  9. #include <linux/kernel.h>
  10. #include <linux/skbuff.h>
  11. #include <linux/slab.h>
  12. #include <linux/sched/signal.h>
  13. #include <net/inet_sock.h>
  14. #include <net/sock.h>
  15. #include "ackvec.h"
  16. #include "ccid.h"
  17. #include "dccp.h"
  18. static inline void dccp_event_ack_sent(struct sock *sk)
  19. {
  20. inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
  21. }
  22. /* enqueue @skb on sk_send_head for retransmission, return clone to send now */
  23. static struct sk_buff *dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
  24. {
  25. skb_set_owner_w(skb, sk);
  26. WARN_ON(sk->sk_send_head);
  27. sk->sk_send_head = skb;
  28. return skb_clone(sk->sk_send_head, gfp_any());
  29. }
  30. /*
  31. * All SKB's seen here are completely headerless. It is our
  32. * job to build the DCCP header, and pass the packet down to
  33. * IP so it can do the same plus pass the packet off to the
  34. * device.
  35. */
  36. static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
  37. {
  38. if (likely(skb != NULL)) {
  39. struct inet_sock *inet = inet_sk(sk);
  40. const struct inet_connection_sock *icsk = inet_csk(sk);
  41. struct dccp_sock *dp = dccp_sk(sk);
  42. struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
  43. struct dccp_hdr *dh;
  44. /* XXX For now we're using only 48 bits sequence numbers */
  45. const u32 dccp_header_size = sizeof(*dh) +
  46. sizeof(struct dccp_hdr_ext) +
  47. dccp_packet_hdr_len(dcb->dccpd_type);
  48. int err, set_ack = 1;
  49. u64 ackno = dp->dccps_gsr;
  50. /*
  51. * Increment GSS here already in case the option code needs it.
  52. * Update GSS for real only if option processing below succeeds.
  53. */
  54. dcb->dccpd_seq = ADD48(dp->dccps_gss, 1);
  55. switch (dcb->dccpd_type) {
  56. case DCCP_PKT_DATA:
  57. set_ack = 0;
  58. fallthrough;
  59. case DCCP_PKT_DATAACK:
  60. case DCCP_PKT_RESET:
  61. break;
  62. case DCCP_PKT_REQUEST:
  63. set_ack = 0;
  64. /* Use ISS on the first (non-retransmitted) Request. */
  65. if (icsk->icsk_retransmits == 0)
  66. dcb->dccpd_seq = dp->dccps_iss;
  67. fallthrough;
  68. case DCCP_PKT_SYNC:
  69. case DCCP_PKT_SYNCACK:
  70. ackno = dcb->dccpd_ack_seq;
  71. fallthrough;
  72. default:
  73. /*
  74. * Set owner/destructor: some skbs are allocated via
  75. * alloc_skb (e.g. when retransmission may happen).
  76. * Only Data, DataAck, and Reset packets should come
  77. * through here with skb->sk set.
  78. */
  79. WARN_ON(skb->sk);
  80. skb_set_owner_w(skb, sk);
  81. break;
  82. }
  83. if (dccp_insert_options(sk, skb)) {
  84. kfree_skb(skb);
  85. return -EPROTO;
  86. }
  87. /* Build DCCP header and checksum it. */
  88. dh = dccp_zeroed_hdr(skb, dccp_header_size);
  89. dh->dccph_type = dcb->dccpd_type;
  90. dh->dccph_sport = inet->inet_sport;
  91. dh->dccph_dport = inet->inet_dport;
  92. dh->dccph_doff = (dccp_header_size + dcb->dccpd_opt_len) / 4;
  93. dh->dccph_ccval = dcb->dccpd_ccval;
  94. dh->dccph_cscov = dp->dccps_pcslen;
  95. /* XXX For now we're using only 48 bits sequence numbers */
  96. dh->dccph_x = 1;
  97. dccp_update_gss(sk, dcb->dccpd_seq);
  98. dccp_hdr_set_seq(dh, dp->dccps_gss);
  99. if (set_ack)
  100. dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
  101. switch (dcb->dccpd_type) {
  102. case DCCP_PKT_REQUEST:
  103. dccp_hdr_request(skb)->dccph_req_service =
  104. dp->dccps_service;
  105. /*
  106. * Limit Ack window to ISS <= P.ackno <= GSS, so that
  107. * only Responses to Requests we sent are considered.
  108. */
  109. dp->dccps_awl = dp->dccps_iss;
  110. break;
  111. case DCCP_PKT_RESET:
  112. dccp_hdr_reset(skb)->dccph_reset_code =
  113. dcb->dccpd_reset_code;
  114. break;
  115. }
  116. icsk->icsk_af_ops->send_check(sk, skb);
  117. if (set_ack)
  118. dccp_event_ack_sent(sk);
  119. DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
  120. err = icsk->icsk_af_ops->queue_xmit(sk, skb, &inet->cork.fl);
  121. return net_xmit_eval(err);
  122. }
  123. return -ENOBUFS;
  124. }
  125. /**
  126. * dccp_determine_ccmps - Find out about CCID-specific packet-size limits
  127. * @dp: socket to find packet size limits of
  128. *
  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. WRITE_ONCE(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. struct socket_wq *wq;
  173. rcu_read_lock();
  174. wq = rcu_dereference(sk->sk_wq);
  175. if (skwq_has_sleeper(wq))
  176. wake_up_interruptible(&wq->wait);
  177. /* Should agree with poll, otherwise some programs break */
  178. if (sock_writeable(sk))
  179. sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
  180. rcu_read_unlock();
  181. }
  182. /**
  183. * dccp_wait_for_ccid - Await CCID send permission
  184. * @sk: socket to wait for
  185. * @delay: timeout in jiffies
  186. *
  187. * This is used by CCIDs which need to delay the send time in process context.
  188. */
  189. static int dccp_wait_for_ccid(struct sock *sk, unsigned long delay)
  190. {
  191. DEFINE_WAIT(wait);
  192. long remaining;
  193. prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
  194. sk->sk_write_pending++;
  195. release_sock(sk);
  196. remaining = schedule_timeout(delay);
  197. lock_sock(sk);
  198. sk->sk_write_pending--;
  199. finish_wait(sk_sleep(sk), &wait);
  200. if (signal_pending(current) || sk->sk_err)
  201. return -1;
  202. return remaining;
  203. }
  204. /**
  205. * dccp_xmit_packet - Send data packet under control of CCID
  206. * @sk: socket to send data packet on
  207. *
  208. * Transmits next-queued payload and informs CCID to account for the packet.
  209. */
  210. static void dccp_xmit_packet(struct sock *sk)
  211. {
  212. int err, len;
  213. struct dccp_sock *dp = dccp_sk(sk);
  214. struct sk_buff *skb = dccp_qpolicy_pop(sk);
  215. if (unlikely(skb == NULL))
  216. return;
  217. len = skb->len;
  218. if (sk->sk_state == DCCP_PARTOPEN) {
  219. const u32 cur_mps = dp->dccps_mss_cache - DCCP_FEATNEG_OVERHEAD;
  220. /*
  221. * See 8.1.5 - Handshake Completion.
  222. *
  223. * For robustness we resend Confirm options until the client has
  224. * entered OPEN. During the initial feature negotiation, the MPS
  225. * is smaller than usual, reduced by the Change/Confirm options.
  226. */
  227. if (!list_empty(&dp->dccps_featneg) && len > cur_mps) {
  228. DCCP_WARN("Payload too large (%d) for featneg.\n", len);
  229. dccp_send_ack(sk);
  230. dccp_feat_list_purge(&dp->dccps_featneg);
  231. }
  232. inet_csk_schedule_ack(sk);
  233. inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
  234. inet_csk(sk)->icsk_rto,
  235. DCCP_RTO_MAX);
  236. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
  237. } else if (dccp_ack_pending(sk)) {
  238. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
  239. } else {
  240. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATA;
  241. }
  242. err = dccp_transmit_skb(sk, skb);
  243. if (err)
  244. dccp_pr_debug("transmit_skb() returned err=%d\n", err);
  245. /*
  246. * Register this one as sent even if an error occurred. To the remote
  247. * end a local packet drop is indistinguishable from network loss, i.e.
  248. * any local drop will eventually be reported via receiver feedback.
  249. */
  250. ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, len);
  251. /*
  252. * If the CCID needs to transfer additional header options out-of-band
  253. * (e.g. Ack Vectors or feature-negotiation options), it activates this
  254. * flag to schedule a Sync. The Sync will automatically incorporate all
  255. * currently pending header options, thus clearing the backlog.
  256. */
  257. if (dp->dccps_sync_scheduled)
  258. dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
  259. }
  260. /**
  261. * dccp_flush_write_queue - Drain queue at end of connection
  262. * @sk: socket to be drained
  263. * @time_budget: time allowed to drain the queue
  264. *
  265. * Since dccp_sendmsg queues packets without waiting for them to be sent, it may
  266. * happen that the TX queue is not empty at the end of a connection. We give the
  267. * HC-sender CCID a grace period of up to @time_budget jiffies. If this function
  268. * returns with a non-empty write queue, it will be purged later.
  269. */
  270. void dccp_flush_write_queue(struct sock *sk, long *time_budget)
  271. {
  272. struct dccp_sock *dp = dccp_sk(sk);
  273. struct sk_buff *skb;
  274. long delay, rc;
  275. while (*time_budget > 0 && (skb = skb_peek(&sk->sk_write_queue))) {
  276. rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
  277. switch (ccid_packet_dequeue_eval(rc)) {
  278. case CCID_PACKET_WILL_DEQUEUE_LATER:
  279. /*
  280. * If the CCID determines when to send, the next sending
  281. * time is unknown or the CCID may not even send again
  282. * (e.g. remote host crashes or lost Ack packets).
  283. */
  284. DCCP_WARN("CCID did not manage to send all packets\n");
  285. return;
  286. case CCID_PACKET_DELAY:
  287. delay = msecs_to_jiffies(rc);
  288. if (delay > *time_budget)
  289. return;
  290. rc = dccp_wait_for_ccid(sk, delay);
  291. if (rc < 0)
  292. return;
  293. *time_budget -= (delay - rc);
  294. /* check again if we can send now */
  295. break;
  296. case CCID_PACKET_SEND_AT_ONCE:
  297. dccp_xmit_packet(sk);
  298. break;
  299. case CCID_PACKET_ERR:
  300. skb_dequeue(&sk->sk_write_queue);
  301. kfree_skb(skb);
  302. dccp_pr_debug("packet discarded due to err=%ld\n", rc);
  303. }
  304. }
  305. }
  306. void dccp_write_xmit(struct sock *sk)
  307. {
  308. struct dccp_sock *dp = dccp_sk(sk);
  309. struct sk_buff *skb;
  310. while ((skb = dccp_qpolicy_top(sk))) {
  311. int rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
  312. switch (ccid_packet_dequeue_eval(rc)) {
  313. case CCID_PACKET_WILL_DEQUEUE_LATER:
  314. return;
  315. case CCID_PACKET_DELAY:
  316. sk_reset_timer(sk, &dp->dccps_xmit_timer,
  317. jiffies + msecs_to_jiffies(rc));
  318. return;
  319. case CCID_PACKET_SEND_AT_ONCE:
  320. dccp_xmit_packet(sk);
  321. break;
  322. case CCID_PACKET_ERR:
  323. dccp_qpolicy_drop(sk, skb);
  324. dccp_pr_debug("packet discarded due to err=%d\n", rc);
  325. }
  326. }
  327. }
  328. /**
  329. * dccp_retransmit_skb - Retransmit Request, Close, or CloseReq packets
  330. * @sk: socket to perform retransmit on
  331. *
  332. * There are only four retransmittable packet types in DCCP:
  333. * - Request in client-REQUEST state (sec. 8.1.1),
  334. * - CloseReq in server-CLOSEREQ state (sec. 8.3),
  335. * - Close in node-CLOSING state (sec. 8.3),
  336. * - Acks in client-PARTOPEN state (sec. 8.1.5, handled by dccp_delack_timer()).
  337. * This function expects sk->sk_send_head to contain the original skb.
  338. */
  339. int dccp_retransmit_skb(struct sock *sk)
  340. {
  341. WARN_ON(sk->sk_send_head == NULL);
  342. if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
  343. return -EHOSTUNREACH; /* Routing failure or similar. */
  344. /* this count is used to distinguish original and retransmitted skb */
  345. inet_csk(sk)->icsk_retransmits++;
  346. return dccp_transmit_skb(sk, skb_clone(sk->sk_send_head, GFP_ATOMIC));
  347. }
  348. struct sk_buff *dccp_make_response(const struct sock *sk, struct dst_entry *dst,
  349. struct request_sock *req)
  350. {
  351. struct dccp_hdr *dh;
  352. struct dccp_request_sock *dreq;
  353. const u32 dccp_header_size = sizeof(struct dccp_hdr) +
  354. sizeof(struct dccp_hdr_ext) +
  355. sizeof(struct dccp_hdr_response);
  356. struct sk_buff *skb;
  357. /* sk is marked const to clearly express we dont hold socket lock.
  358. * sock_wmalloc() will atomically change sk->sk_wmem_alloc,
  359. * it is safe to promote sk to non const.
  360. */
  361. skb = sock_wmalloc((struct sock *)sk, MAX_DCCP_HEADER, 1,
  362. GFP_ATOMIC);
  363. if (!skb)
  364. return NULL;
  365. skb_reserve(skb, MAX_DCCP_HEADER);
  366. skb_dst_set(skb, dst_clone(dst));
  367. dreq = dccp_rsk(req);
  368. if (inet_rsk(req)->acked) /* increase GSS upon retransmission */
  369. dccp_inc_seqno(&dreq->dreq_gss);
  370. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
  371. DCCP_SKB_CB(skb)->dccpd_seq = dreq->dreq_gss;
  372. /* Resolve feature dependencies resulting from choice of CCID */
  373. if (dccp_feat_server_ccid_dependencies(dreq))
  374. goto response_failed;
  375. if (dccp_insert_options_rsk(dreq, skb))
  376. goto response_failed;
  377. /* Build and checksum header */
  378. dh = dccp_zeroed_hdr(skb, dccp_header_size);
  379. dh->dccph_sport = htons(inet_rsk(req)->ir_num);
  380. dh->dccph_dport = inet_rsk(req)->ir_rmt_port;
  381. dh->dccph_doff = (dccp_header_size +
  382. DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
  383. dh->dccph_type = DCCP_PKT_RESPONSE;
  384. dh->dccph_x = 1;
  385. dccp_hdr_set_seq(dh, dreq->dreq_gss);
  386. dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_gsr);
  387. dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
  388. dccp_csum_outgoing(skb);
  389. /* We use `acked' to remember that a Response was already sent. */
  390. inet_rsk(req)->acked = 1;
  391. DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
  392. return skb;
  393. response_failed:
  394. kfree_skb(skb);
  395. return NULL;
  396. }
  397. EXPORT_SYMBOL_GPL(dccp_make_response);
  398. /* answer offending packet in @rcv_skb with Reset from control socket @ctl */
  399. struct sk_buff *dccp_ctl_make_reset(struct sock *sk, struct sk_buff *rcv_skb)
  400. {
  401. struct dccp_hdr *rxdh = dccp_hdr(rcv_skb), *dh;
  402. struct dccp_skb_cb *dcb = DCCP_SKB_CB(rcv_skb);
  403. const u32 dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
  404. sizeof(struct dccp_hdr_ext) +
  405. sizeof(struct dccp_hdr_reset);
  406. struct dccp_hdr_reset *dhr;
  407. struct sk_buff *skb;
  408. skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
  409. if (skb == NULL)
  410. return NULL;
  411. skb_reserve(skb, sk->sk_prot->max_header);
  412. /* Swap the send and the receive. */
  413. dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
  414. dh->dccph_type = DCCP_PKT_RESET;
  415. dh->dccph_sport = rxdh->dccph_dport;
  416. dh->dccph_dport = rxdh->dccph_sport;
  417. dh->dccph_doff = dccp_hdr_reset_len / 4;
  418. dh->dccph_x = 1;
  419. dhr = dccp_hdr_reset(skb);
  420. dhr->dccph_reset_code = dcb->dccpd_reset_code;
  421. switch (dcb->dccpd_reset_code) {
  422. case DCCP_RESET_CODE_PACKET_ERROR:
  423. dhr->dccph_reset_data[0] = rxdh->dccph_type;
  424. break;
  425. case DCCP_RESET_CODE_OPTION_ERROR:
  426. case DCCP_RESET_CODE_MANDATORY_ERROR:
  427. memcpy(dhr->dccph_reset_data, dcb->dccpd_reset_data, 3);
  428. break;
  429. }
  430. /*
  431. * From RFC 4340, 8.3.1:
  432. * If P.ackno exists, set R.seqno := P.ackno + 1.
  433. * Else set R.seqno := 0.
  434. */
  435. if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
  436. dccp_hdr_set_seq(dh, ADD48(dcb->dccpd_ack_seq, 1));
  437. dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dcb->dccpd_seq);
  438. dccp_csum_outgoing(skb);
  439. return skb;
  440. }
  441. EXPORT_SYMBOL_GPL(dccp_ctl_make_reset);
  442. /* send Reset on established socket, to close or abort the connection */
  443. int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
  444. {
  445. struct sk_buff *skb;
  446. /*
  447. * FIXME: what if rebuild_header fails?
  448. * Should we be doing a rebuild_header here?
  449. */
  450. int err = inet_csk(sk)->icsk_af_ops->rebuild_header(sk);
  451. if (err != 0)
  452. return err;
  453. skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
  454. if (skb == NULL)
  455. return -ENOBUFS;
  456. /* Reserve space for headers and prepare control bits. */
  457. skb_reserve(skb, sk->sk_prot->max_header);
  458. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESET;
  459. DCCP_SKB_CB(skb)->dccpd_reset_code = code;
  460. return dccp_transmit_skb(sk, skb);
  461. }
  462. /*
  463. * Do all connect socket setups that can be done AF independent.
  464. */
  465. int dccp_connect(struct sock *sk)
  466. {
  467. struct sk_buff *skb;
  468. struct dccp_sock *dp = dccp_sk(sk);
  469. struct dst_entry *dst = __sk_dst_get(sk);
  470. struct inet_connection_sock *icsk = inet_csk(sk);
  471. sk->sk_err = 0;
  472. sock_reset_flag(sk, SOCK_DONE);
  473. dccp_sync_mss(sk, dst_mtu(dst));
  474. /* do not connect if feature negotiation setup fails */
  475. if (dccp_feat_finalise_settings(dccp_sk(sk)))
  476. return -EPROTO;
  477. /* Initialise GAR as per 8.5; AWL/AWH are set in dccp_transmit_skb() */
  478. dp->dccps_gar = dp->dccps_iss;
  479. skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
  480. if (unlikely(skb == NULL))
  481. return -ENOBUFS;
  482. /* Reserve space for headers. */
  483. skb_reserve(skb, sk->sk_prot->max_header);
  484. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
  485. dccp_transmit_skb(sk, dccp_skb_entail(sk, skb));
  486. DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
  487. /* Timer for repeating the REQUEST until an answer. */
  488. icsk->icsk_retransmits = 0;
  489. inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
  490. icsk->icsk_rto, DCCP_RTO_MAX);
  491. return 0;
  492. }
  493. EXPORT_SYMBOL_GPL(dccp_connect);
  494. void dccp_send_ack(struct sock *sk)
  495. {
  496. /* If we have been reset, we may not send again. */
  497. if (sk->sk_state != DCCP_CLOSED) {
  498. struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
  499. GFP_ATOMIC);
  500. if (skb == NULL) {
  501. inet_csk_schedule_ack(sk);
  502. inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
  503. inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
  504. TCP_DELACK_MAX,
  505. DCCP_RTO_MAX);
  506. return;
  507. }
  508. /* Reserve space for headers */
  509. skb_reserve(skb, sk->sk_prot->max_header);
  510. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
  511. dccp_transmit_skb(sk, skb);
  512. }
  513. }
  514. EXPORT_SYMBOL_GPL(dccp_send_ack);
  515. #if 0
  516. /* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
  517. void dccp_send_delayed_ack(struct sock *sk)
  518. {
  519. struct inet_connection_sock *icsk = inet_csk(sk);
  520. /*
  521. * FIXME: tune this timer. elapsed time fixes the skew, so no problem
  522. * with using 2s, and active senders also piggyback the ACK into a
  523. * DATAACK packet, so this is really for quiescent senders.
  524. */
  525. unsigned long timeout = jiffies + 2 * HZ;
  526. /* Use new timeout only if there wasn't a older one earlier. */
  527. if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
  528. /* If delack timer was blocked or is about to expire,
  529. * send ACK now.
  530. *
  531. * FIXME: check the "about to expire" part
  532. */
  533. if (icsk->icsk_ack.blocked) {
  534. dccp_send_ack(sk);
  535. return;
  536. }
  537. if (!time_before(timeout, icsk->icsk_ack.timeout))
  538. timeout = icsk->icsk_ack.timeout;
  539. }
  540. icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
  541. icsk->icsk_ack.timeout = timeout;
  542. sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
  543. }
  544. #endif
  545. void dccp_send_sync(struct sock *sk, const u64 ackno,
  546. const enum dccp_pkt_type pkt_type)
  547. {
  548. /*
  549. * We are not putting this on the write queue, so
  550. * dccp_transmit_skb() will set the ownership to this
  551. * sock.
  552. */
  553. struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
  554. if (skb == NULL) {
  555. /* FIXME: how to make sure the sync is sent? */
  556. DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
  557. return;
  558. }
  559. /* Reserve space for headers and prepare control bits. */
  560. skb_reserve(skb, sk->sk_prot->max_header);
  561. DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
  562. DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
  563. /*
  564. * Clear the flag in case the Sync was scheduled for out-of-band data,
  565. * such as carrying a long Ack Vector.
  566. */
  567. dccp_sk(sk)->dccps_sync_scheduled = 0;
  568. dccp_transmit_skb(sk, skb);
  569. }
  570. EXPORT_SYMBOL_GPL(dccp_send_sync);
  571. /*
  572. * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
  573. * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
  574. * any circumstances.
  575. */
  576. void dccp_send_close(struct sock *sk, const int active)
  577. {
  578. struct dccp_sock *dp = dccp_sk(sk);
  579. struct sk_buff *skb;
  580. const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
  581. skb = alloc_skb(sk->sk_prot->max_header, prio);
  582. if (skb == NULL)
  583. return;
  584. /* Reserve space for headers and prepare control bits. */
  585. skb_reserve(skb, sk->sk_prot->max_header);
  586. if (dp->dccps_role == DCCP_ROLE_SERVER && !dp->dccps_server_timewait)
  587. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSEREQ;
  588. else
  589. DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSE;
  590. if (active) {
  591. skb = dccp_skb_entail(sk, skb);
  592. /*
  593. * Retransmission timer for active-close: RFC 4340, 8.3 requires
  594. * to retransmit the Close/CloseReq until the CLOSING/CLOSEREQ
  595. * state can be left. The initial timeout is 2 RTTs.
  596. * Since RTT measurement is done by the CCIDs, there is no easy
  597. * way to get an RTT sample. The fallback RTT from RFC 4340, 3.4
  598. * is too low (200ms); we use a high value to avoid unnecessary
  599. * retransmissions when the link RTT is > 0.2 seconds.
  600. * FIXME: Let main module sample RTTs and use that instead.
  601. */
  602. inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
  603. DCCP_TIMEOUT_INIT, DCCP_RTO_MAX);
  604. }
  605. dccp_transmit_skb(sk, skb);
  606. }