tcp_timer.c 18 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651
  1. /*
  2. * INET An implementation of the TCP/IP protocol suite for the LINUX
  3. * operating system. INET is implemented using the BSD Socket
  4. * interface as the means of communication with the user level.
  5. *
  6. * Implementation of the Transmission Control Protocol(TCP).
  7. *
  8. * Authors: Ross Biro
  9. * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10. * Mark Evans, <evansmp@uhura.aston.ac.uk>
  11. * Corey Minyard <wf-rch!minyard@relay.EU.net>
  12. * Florian La Roche, <flla@stud.uni-sb.de>
  13. * Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
  14. * Linus Torvalds, <torvalds@cs.helsinki.fi>
  15. * Alan Cox, <gw4pts@gw4pts.ampr.org>
  16. * Matthew Dillon, <dillon@apollo.west.oic.com>
  17. * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  18. * Jorge Cwik, <jorge@laser.satlink.net>
  19. */
  20. #include <linux/module.h>
  21. #include <linux/gfp.h>
  22. #include <net/tcp.h>
  23. int sysctl_tcp_syn_retries __read_mostly = TCP_SYN_RETRIES;
  24. int sysctl_tcp_synack_retries __read_mostly = TCP_SYNACK_RETRIES;
  25. int sysctl_tcp_keepalive_time __read_mostly = TCP_KEEPALIVE_TIME;
  26. int sysctl_tcp_keepalive_probes __read_mostly = TCP_KEEPALIVE_PROBES;
  27. int sysctl_tcp_keepalive_intvl __read_mostly = TCP_KEEPALIVE_INTVL;
  28. int sysctl_tcp_retries1 __read_mostly = TCP_RETR1;
  29. int sysctl_tcp_retries2 __read_mostly = TCP_RETR2;
  30. int sysctl_tcp_orphan_retries __read_mostly;
  31. int sysctl_tcp_thin_linear_timeouts __read_mostly;
  32. static void tcp_write_err(struct sock *sk)
  33. {
  34. sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT;
  35. sk->sk_error_report(sk);
  36. tcp_done(sk);
  37. NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT);
  38. }
  39. /* Do not allow orphaned sockets to eat all our resources.
  40. * This is direct violation of TCP specs, but it is required
  41. * to prevent DoS attacks. It is called when a retransmission timeout
  42. * or zero probe timeout occurs on orphaned socket.
  43. *
  44. * Criteria is still not confirmed experimentally and may change.
  45. * We kill the socket, if:
  46. * 1. If number of orphaned sockets exceeds an administratively configured
  47. * limit.
  48. * 2. If we have strong memory pressure.
  49. */
  50. static int tcp_out_of_resources(struct sock *sk, int do_reset)
  51. {
  52. struct tcp_sock *tp = tcp_sk(sk);
  53. int shift = 0;
  54. /* If peer does not open window for long time, or did not transmit
  55. * anything for long time, penalize it. */
  56. if ((s32)(tcp_time_stamp - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset)
  57. shift++;
  58. /* If some dubious ICMP arrived, penalize even more. */
  59. if (sk->sk_err_soft)
  60. shift++;
  61. if (tcp_check_oom(sk, shift)) {
  62. /* Catch exceptional cases, when connection requires reset.
  63. * 1. Last segment was sent recently. */
  64. if ((s32)(tcp_time_stamp - tp->lsndtime) <= TCP_TIMEWAIT_LEN ||
  65. /* 2. Window is closed. */
  66. (!tp->snd_wnd && !tp->packets_out))
  67. do_reset = 1;
  68. if (do_reset)
  69. tcp_send_active_reset(sk, GFP_ATOMIC);
  70. tcp_done(sk);
  71. NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY);
  72. return 1;
  73. }
  74. return 0;
  75. }
  76. /* Calculate maximal number or retries on an orphaned socket. */
  77. static int tcp_orphan_retries(struct sock *sk, int alive)
  78. {
  79. int retries = sysctl_tcp_orphan_retries; /* May be zero. */
  80. /* We know from an ICMP that something is wrong. */
  81. if (sk->sk_err_soft && !alive)
  82. retries = 0;
  83. /* However, if socket sent something recently, select some safe
  84. * number of retries. 8 corresponds to >100 seconds with minimal
  85. * RTO of 200msec. */
  86. if (retries == 0 && alive)
  87. retries = 8;
  88. return retries;
  89. }
  90. static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
  91. {
  92. /* Black hole detection */
  93. if (sysctl_tcp_mtu_probing) {
  94. if (!icsk->icsk_mtup.enabled) {
  95. icsk->icsk_mtup.enabled = 1;
  96. tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
  97. } else {
  98. struct tcp_sock *tp = tcp_sk(sk);
  99. int mss;
  100. mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1;
  101. mss = min(sysctl_tcp_base_mss, mss);
  102. mss = max(mss, 68 - tp->tcp_header_len);
  103. icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
  104. tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
  105. }
  106. }
  107. }
  108. /* This function calculates a "timeout" which is equivalent to the timeout of a
  109. * TCP connection after "boundary" unsuccessful, exponentially backed-off
  110. * retransmissions with an initial RTO of TCP_RTO_MIN or TCP_TIMEOUT_INIT if
  111. * syn_set flag is set.
  112. */
  113. static bool retransmits_timed_out(struct sock *sk,
  114. unsigned int boundary,
  115. unsigned int timeout,
  116. bool syn_set)
  117. {
  118. unsigned int linear_backoff_thresh, start_ts;
  119. unsigned int rto_base = syn_set ? TCP_TIMEOUT_INIT : TCP_RTO_MIN;
  120. if (!inet_csk(sk)->icsk_retransmits)
  121. return false;
  122. if (unlikely(!tcp_sk(sk)->retrans_stamp))
  123. start_ts = TCP_SKB_CB(tcp_write_queue_head(sk))->when;
  124. else
  125. start_ts = tcp_sk(sk)->retrans_stamp;
  126. if (likely(timeout == 0)) {
  127. linear_backoff_thresh = ilog2(TCP_RTO_MAX/rto_base);
  128. if (boundary <= linear_backoff_thresh)
  129. timeout = ((2 << boundary) - 1) * rto_base;
  130. else
  131. timeout = ((2 << linear_backoff_thresh) - 1) * rto_base +
  132. (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
  133. }
  134. return (tcp_time_stamp - start_ts) >= timeout;
  135. }
  136. /* A write timeout has occurred. Process the after effects. */
  137. static int tcp_write_timeout(struct sock *sk)
  138. {
  139. struct inet_connection_sock *icsk = inet_csk(sk);
  140. struct tcp_sock *tp = tcp_sk(sk);
  141. int retry_until;
  142. bool do_reset, syn_set = false;
  143. if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
  144. if (icsk->icsk_retransmits) {
  145. dst_negative_advice(sk);
  146. if (tp->syn_fastopen || tp->syn_data)
  147. tcp_fastopen_cache_set(sk, 0, NULL, true);
  148. }
  149. retry_until = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
  150. syn_set = true;
  151. } else {
  152. if (retransmits_timed_out(sk, sysctl_tcp_retries1, 0, 0)) {
  153. /* Black hole detection */
  154. tcp_mtu_probing(icsk, sk);
  155. dst_negative_advice(sk);
  156. }
  157. retry_until = sysctl_tcp_retries2;
  158. if (sock_flag(sk, SOCK_DEAD)) {
  159. const int alive = (icsk->icsk_rto < TCP_RTO_MAX);
  160. retry_until = tcp_orphan_retries(sk, alive);
  161. do_reset = alive ||
  162. !retransmits_timed_out(sk, retry_until, 0, 0);
  163. if (tcp_out_of_resources(sk, do_reset))
  164. return 1;
  165. }
  166. }
  167. if (retransmits_timed_out(sk, retry_until,
  168. syn_set ? 0 : icsk->icsk_user_timeout, syn_set)) {
  169. /* Has it gone just too far? */
  170. tcp_write_err(sk);
  171. return 1;
  172. }
  173. return 0;
  174. }
  175. void tcp_delack_timer_handler(struct sock *sk)
  176. {
  177. struct tcp_sock *tp = tcp_sk(sk);
  178. struct inet_connection_sock *icsk = inet_csk(sk);
  179. sk_mem_reclaim_partial(sk);
  180. if (sk->sk_state == TCP_CLOSE || !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
  181. goto out;
  182. if (time_after(icsk->icsk_ack.timeout, jiffies)) {
  183. sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
  184. goto out;
  185. }
  186. icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
  187. if (!skb_queue_empty(&tp->ucopy.prequeue)) {
  188. struct sk_buff *skb;
  189. NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSCHEDULERFAILED);
  190. while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
  191. sk_backlog_rcv(sk, skb);
  192. tp->ucopy.memory = 0;
  193. }
  194. if (inet_csk_ack_scheduled(sk)) {
  195. if (!icsk->icsk_ack.pingpong) {
  196. /* Delayed ACK missed: inflate ATO. */
  197. icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
  198. } else {
  199. /* Delayed ACK missed: leave pingpong mode and
  200. * deflate ATO.
  201. */
  202. icsk->icsk_ack.pingpong = 0;
  203. icsk->icsk_ack.ato = TCP_ATO_MIN;
  204. }
  205. tcp_send_ack(sk);
  206. NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKS);
  207. }
  208. out:
  209. if (sk_under_memory_pressure(sk))
  210. sk_mem_reclaim(sk);
  211. }
  212. static void tcp_delack_timer(unsigned long data)
  213. {
  214. struct sock *sk = (struct sock *)data;
  215. bh_lock_sock(sk);
  216. if (!sock_owned_by_user(sk)) {
  217. tcp_delack_timer_handler(sk);
  218. } else {
  219. inet_csk(sk)->icsk_ack.blocked = 1;
  220. NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
  221. /* deleguate our work to tcp_release_cb() */
  222. if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags))
  223. sock_hold(sk);
  224. }
  225. bh_unlock_sock(sk);
  226. sock_put(sk);
  227. }
  228. static void tcp_probe_timer(struct sock *sk)
  229. {
  230. struct inet_connection_sock *icsk = inet_csk(sk);
  231. struct tcp_sock *tp = tcp_sk(sk);
  232. int max_probes;
  233. if (tp->packets_out || !tcp_send_head(sk)) {
  234. icsk->icsk_probes_out = 0;
  235. return;
  236. }
  237. /* *WARNING* RFC 1122 forbids this
  238. *
  239. * It doesn't AFAIK, because we kill the retransmit timer -AK
  240. *
  241. * FIXME: We ought not to do it, Solaris 2.5 actually has fixing
  242. * this behaviour in Solaris down as a bug fix. [AC]
  243. *
  244. * Let me to explain. icsk_probes_out is zeroed by incoming ACKs
  245. * even if they advertise zero window. Hence, connection is killed only
  246. * if we received no ACKs for normal connection timeout. It is not killed
  247. * only because window stays zero for some time, window may be zero
  248. * until armageddon and even later. We are in full accordance
  249. * with RFCs, only probe timer combines both retransmission timeout
  250. * and probe timeout in one bottle. --ANK
  251. */
  252. max_probes = sysctl_tcp_retries2;
  253. if (sock_flag(sk, SOCK_DEAD)) {
  254. const int alive = ((icsk->icsk_rto << icsk->icsk_backoff) < TCP_RTO_MAX);
  255. max_probes = tcp_orphan_retries(sk, alive);
  256. if (tcp_out_of_resources(sk, alive || icsk->icsk_probes_out <= max_probes))
  257. return;
  258. }
  259. if (icsk->icsk_probes_out > max_probes) {
  260. tcp_write_err(sk);
  261. } else {
  262. /* Only send another probe if we didn't close things up. */
  263. tcp_send_probe0(sk);
  264. }
  265. }
  266. /*
  267. * Timer for Fast Open socket to retransmit SYNACK. Note that the
  268. * sk here is the child socket, not the parent (listener) socket.
  269. */
  270. static void tcp_fastopen_synack_timer(struct sock *sk)
  271. {
  272. struct inet_connection_sock *icsk = inet_csk(sk);
  273. int max_retries = icsk->icsk_syn_retries ? :
  274. sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */
  275. struct request_sock *req;
  276. req = tcp_sk(sk)->fastopen_rsk;
  277. req->rsk_ops->syn_ack_timeout(sk, req);
  278. if (req->num_timeout >= max_retries) {
  279. tcp_write_err(sk);
  280. return;
  281. }
  282. /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error
  283. * returned from rtx_syn_ack() to make it more persistent like
  284. * regular retransmit because if the child socket has been accepted
  285. * it's not good to give up too easily.
  286. */
  287. inet_rtx_syn_ack(sk, req);
  288. req->num_timeout++;
  289. inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
  290. TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
  291. }
  292. /*
  293. * The TCP retransmit timer.
  294. */
  295. void tcp_retransmit_timer(struct sock *sk)
  296. {
  297. struct tcp_sock *tp = tcp_sk(sk);
  298. struct inet_connection_sock *icsk = inet_csk(sk);
  299. if (tp->fastopen_rsk) {
  300. WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV &&
  301. sk->sk_state != TCP_FIN_WAIT1);
  302. tcp_fastopen_synack_timer(sk);
  303. /* Before we receive ACK to our SYN-ACK don't retransmit
  304. * anything else (e.g., data or FIN segments).
  305. */
  306. return;
  307. }
  308. if (!tp->packets_out)
  309. goto out;
  310. WARN_ON(tcp_write_queue_empty(sk));
  311. tp->tlp_high_seq = 0;
  312. if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
  313. !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
  314. /* Receiver dastardly shrinks window. Our retransmits
  315. * become zero probes, but we should not timeout this
  316. * connection. If the socket is an orphan, time it out,
  317. * we cannot allow such beasts to hang infinitely.
  318. */
  319. struct inet_sock *inet = inet_sk(sk);
  320. if (sk->sk_family == AF_INET) {
  321. LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n"),
  322. &inet->inet_daddr,
  323. ntohs(inet->inet_dport), inet->inet_num,
  324. tp->snd_una, tp->snd_nxt);
  325. }
  326. #if IS_ENABLED(CONFIG_IPV6)
  327. else if (sk->sk_family == AF_INET6) {
  328. LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n"),
  329. &sk->sk_v6_daddr,
  330. ntohs(inet->inet_dport), inet->inet_num,
  331. tp->snd_una, tp->snd_nxt);
  332. }
  333. #endif
  334. if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) {
  335. tcp_write_err(sk);
  336. goto out;
  337. }
  338. tcp_enter_loss(sk, 0);
  339. tcp_retransmit_skb(sk, tcp_write_queue_head(sk));
  340. __sk_dst_reset(sk);
  341. goto out_reset_timer;
  342. }
  343. if (tcp_write_timeout(sk))
  344. goto out;
  345. if (icsk->icsk_retransmits == 0) {
  346. int mib_idx;
  347. if (icsk->icsk_ca_state == TCP_CA_Recovery) {
  348. if (tcp_is_sack(tp))
  349. mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL;
  350. else
  351. mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL;
  352. } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
  353. mib_idx = LINUX_MIB_TCPLOSSFAILURES;
  354. } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) ||
  355. tp->sacked_out) {
  356. if (tcp_is_sack(tp))
  357. mib_idx = LINUX_MIB_TCPSACKFAILURES;
  358. else
  359. mib_idx = LINUX_MIB_TCPRENOFAILURES;
  360. } else {
  361. mib_idx = LINUX_MIB_TCPTIMEOUTS;
  362. }
  363. NET_INC_STATS_BH(sock_net(sk), mib_idx);
  364. }
  365. tcp_enter_loss(sk, 0);
  366. if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk)) > 0) {
  367. /* Retransmission failed because of local congestion,
  368. * do not backoff.
  369. */
  370. if (!icsk->icsk_retransmits)
  371. icsk->icsk_retransmits = 1;
  372. inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
  373. min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL),
  374. TCP_RTO_MAX);
  375. goto out;
  376. }
  377. /* Increase the timeout each time we retransmit. Note that
  378. * we do not increase the rtt estimate. rto is initialized
  379. * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
  380. * that doubling rto each time is the least we can get away with.
  381. * In KA9Q, Karn uses this for the first few times, and then
  382. * goes to quadratic. netBSD doubles, but only goes up to *64,
  383. * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
  384. * defined in the protocol as the maximum possible RTT. I guess
  385. * we'll have to use something other than TCP to talk to the
  386. * University of Mars.
  387. *
  388. * PAWS allows us longer timeouts and large windows, so once
  389. * implemented ftp to mars will work nicely. We will have to fix
  390. * the 120 second clamps though!
  391. */
  392. icsk->icsk_backoff++;
  393. icsk->icsk_retransmits++;
  394. out_reset_timer:
  395. /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is
  396. * used to reset timer, set to 0. Recalculate 'icsk_rto' as this
  397. * might be increased if the stream oscillates between thin and thick,
  398. * thus the old value might already be too high compared to the value
  399. * set by 'tcp_set_rto' in tcp_input.c which resets the rto without
  400. * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating
  401. * exponential backoff behaviour to avoid continue hammering
  402. * linear-timeout retransmissions into a black hole
  403. */
  404. if (sk->sk_state == TCP_ESTABLISHED &&
  405. (tp->thin_lto || sysctl_tcp_thin_linear_timeouts) &&
  406. tcp_stream_is_thin(tp) &&
  407. icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) {
  408. icsk->icsk_backoff = 0;
  409. icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX);
  410. } else {
  411. /* Use normal (exponential) backoff */
  412. icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
  413. }
  414. inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX);
  415. if (retransmits_timed_out(sk, sysctl_tcp_retries1 + 1, 0, 0))
  416. __sk_dst_reset(sk);
  417. out:;
  418. }
  419. void tcp_write_timer_handler(struct sock *sk)
  420. {
  421. struct inet_connection_sock *icsk = inet_csk(sk);
  422. int event;
  423. if (sk->sk_state == TCP_CLOSE || !icsk->icsk_pending)
  424. goto out;
  425. if (time_after(icsk->icsk_timeout, jiffies)) {
  426. sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
  427. goto out;
  428. }
  429. event = icsk->icsk_pending;
  430. switch (event) {
  431. case ICSK_TIME_EARLY_RETRANS:
  432. tcp_resume_early_retransmit(sk);
  433. break;
  434. case ICSK_TIME_LOSS_PROBE:
  435. tcp_send_loss_probe(sk);
  436. break;
  437. case ICSK_TIME_RETRANS:
  438. icsk->icsk_pending = 0;
  439. tcp_retransmit_timer(sk);
  440. break;
  441. case ICSK_TIME_PROBE0:
  442. icsk->icsk_pending = 0;
  443. tcp_probe_timer(sk);
  444. break;
  445. }
  446. out:
  447. sk_mem_reclaim(sk);
  448. }
  449. static void tcp_write_timer(unsigned long data)
  450. {
  451. struct sock *sk = (struct sock *)data;
  452. bh_lock_sock(sk);
  453. if (!sock_owned_by_user(sk)) {
  454. tcp_write_timer_handler(sk);
  455. } else {
  456. /* deleguate our work to tcp_release_cb() */
  457. if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags))
  458. sock_hold(sk);
  459. }
  460. bh_unlock_sock(sk);
  461. sock_put(sk);
  462. }
  463. /*
  464. * Timer for listening sockets
  465. */
  466. static void tcp_synack_timer(struct sock *sk)
  467. {
  468. inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL,
  469. TCP_TIMEOUT_INIT, TCP_RTO_MAX);
  470. }
  471. void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req)
  472. {
  473. NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEOUTS);
  474. }
  475. EXPORT_SYMBOL(tcp_syn_ack_timeout);
  476. void tcp_set_keepalive(struct sock *sk, int val)
  477. {
  478. if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
  479. return;
  480. if (val && !sock_flag(sk, SOCK_KEEPOPEN))
  481. inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
  482. else if (!val)
  483. inet_csk_delete_keepalive_timer(sk);
  484. }
  485. static void tcp_keepalive_timer (unsigned long data)
  486. {
  487. struct sock *sk = (struct sock *) data;
  488. struct inet_connection_sock *icsk = inet_csk(sk);
  489. struct tcp_sock *tp = tcp_sk(sk);
  490. u32 elapsed;
  491. /* Only process if socket is not in use. */
  492. bh_lock_sock(sk);
  493. if (sock_owned_by_user(sk)) {
  494. /* Try again later. */
  495. inet_csk_reset_keepalive_timer (sk, HZ/20);
  496. goto out;
  497. }
  498. if (sk->sk_state == TCP_LISTEN) {
  499. tcp_synack_timer(sk);
  500. goto out;
  501. }
  502. if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
  503. if (tp->linger2 >= 0) {
  504. const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
  505. if (tmo > 0) {
  506. tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
  507. goto out;
  508. }
  509. }
  510. tcp_send_active_reset(sk, GFP_ATOMIC);
  511. goto death;
  512. }
  513. if (!sock_flag(sk, SOCK_KEEPOPEN) || sk->sk_state == TCP_CLOSE)
  514. goto out;
  515. elapsed = keepalive_time_when(tp);
  516. /* It is alive without keepalive 8) */
  517. if (tp->packets_out || tcp_send_head(sk))
  518. goto resched;
  519. elapsed = keepalive_time_elapsed(tp);
  520. if (elapsed >= keepalive_time_when(tp)) {
  521. /* If the TCP_USER_TIMEOUT option is enabled, use that
  522. * to determine when to timeout instead.
  523. */
  524. if ((icsk->icsk_user_timeout != 0 &&
  525. elapsed >= icsk->icsk_user_timeout &&
  526. icsk->icsk_probes_out > 0) ||
  527. (icsk->icsk_user_timeout == 0 &&
  528. icsk->icsk_probes_out >= keepalive_probes(tp))) {
  529. tcp_send_active_reset(sk, GFP_ATOMIC);
  530. tcp_write_err(sk);
  531. goto out;
  532. }
  533. if (tcp_write_wakeup(sk) <= 0) {
  534. icsk->icsk_probes_out++;
  535. elapsed = keepalive_intvl_when(tp);
  536. } else {
  537. /* If keepalive was lost due to local congestion,
  538. * try harder.
  539. */
  540. elapsed = TCP_RESOURCE_PROBE_INTERVAL;
  541. }
  542. } else {
  543. /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
  544. elapsed = keepalive_time_when(tp) - elapsed;
  545. }
  546. sk_mem_reclaim(sk);
  547. resched:
  548. inet_csk_reset_keepalive_timer (sk, elapsed);
  549. goto out;
  550. death:
  551. tcp_done(sk);
  552. out:
  553. bh_unlock_sock(sk);
  554. sock_put(sk);
  555. }
  556. void tcp_init_xmit_timers(struct sock *sk)
  557. {
  558. inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer,
  559. &tcp_keepalive_timer);
  560. }
  561. EXPORT_SYMBOL(tcp_init_xmit_timers);