threads.c 6.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224
  1. /*
  2. * Copyright (c) 2006 Oracle. All rights reserved.
  3. *
  4. * This software is available to you under a choice of one of two
  5. * licenses. You may choose to be licensed under the terms of the GNU
  6. * General Public License (GPL) Version 2, available from the file
  7. * COPYING in the main directory of this source tree, or the
  8. * OpenIB.org BSD license below:
  9. *
  10. * Redistribution and use in source and binary forms, with or
  11. * without modification, are permitted provided that the following
  12. * conditions are met:
  13. *
  14. * - Redistributions of source code must retain the above
  15. * copyright notice, this list of conditions and the following
  16. * disclaimer.
  17. *
  18. * - Redistributions in binary form must reproduce the above
  19. * copyright notice, this list of conditions and the following
  20. * disclaimer in the documentation and/or other materials
  21. * provided with the distribution.
  22. *
  23. * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24. * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25. * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26. * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27. * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28. * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29. * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30. * SOFTWARE.
  31. *
  32. */
  33. #include <linux/kernel.h>
  34. #include <linux/random.h>
  35. #include <linux/export.h>
  36. #include "rds.h"
  37. /*
  38. * All of connection management is simplified by serializing it through
  39. * work queues that execute in a connection managing thread.
  40. *
  41. * TCP wants to send acks through sendpage() in response to data_ready(),
  42. * but it needs a process context to do so.
  43. *
  44. * The receive paths need to allocate but can't drop packets (!) so we have
  45. * a thread around to block allocating if the receive fast path sees an
  46. * allocation failure.
  47. */
  48. /* Grand Unified Theory of connection life cycle:
  49. * At any point in time, the connection can be in one of these states:
  50. * DOWN, CONNECTING, UP, DISCONNECTING, ERROR
  51. *
  52. * The following transitions are possible:
  53. * ANY -> ERROR
  54. * UP -> DISCONNECTING
  55. * ERROR -> DISCONNECTING
  56. * DISCONNECTING -> DOWN
  57. * DOWN -> CONNECTING
  58. * CONNECTING -> UP
  59. *
  60. * Transition to state DISCONNECTING/DOWN:
  61. * - Inside the shutdown worker; synchronizes with xmit path
  62. * through RDS_IN_XMIT, and with connection management callbacks
  63. * via c_cm_lock.
  64. *
  65. * For receive callbacks, we rely on the underlying transport
  66. * (TCP, IB/RDMA) to provide the necessary synchronisation.
  67. */
  68. struct workqueue_struct *rds_wq;
  69. EXPORT_SYMBOL_GPL(rds_wq);
  70. void rds_connect_complete(struct rds_connection *conn)
  71. {
  72. if (!rds_conn_transition(conn, RDS_CONN_CONNECTING, RDS_CONN_UP)) {
  73. printk(KERN_WARNING "%s: Cannot transition to state UP, "
  74. "current state is %d\n",
  75. __func__,
  76. atomic_read(&conn->c_state));
  77. rds_conn_drop(conn);
  78. return;
  79. }
  80. rdsdebug("conn %p for %pI4 to %pI4 complete\n",
  81. conn, &conn->c_laddr, &conn->c_faddr);
  82. conn->c_reconnect_jiffies = 0;
  83. set_bit(0, &conn->c_map_queued);
  84. queue_delayed_work(rds_wq, &conn->c_send_w, 0);
  85. queue_delayed_work(rds_wq, &conn->c_recv_w, 0);
  86. }
  87. EXPORT_SYMBOL_GPL(rds_connect_complete);
  88. /*
  89. * This random exponential backoff is relied on to eventually resolve racing
  90. * connects.
  91. *
  92. * If connect attempts race then both parties drop both connections and come
  93. * here to wait for a random amount of time before trying again. Eventually
  94. * the backoff range will be so much greater than the time it takes to
  95. * establish a connection that one of the pair will establish the connection
  96. * before the other's random delay fires.
  97. *
  98. * Connection attempts that arrive while a connection is already established
  99. * are also considered to be racing connects. This lets a connection from
  100. * a rebooted machine replace an existing stale connection before the transport
  101. * notices that the connection has failed.
  102. *
  103. * We should *always* start with a random backoff; otherwise a broken connection
  104. * will always take several iterations to be re-established.
  105. */
  106. void rds_queue_reconnect(struct rds_connection *conn)
  107. {
  108. unsigned long rand;
  109. rdsdebug("conn %p for %pI4 to %pI4 reconnect jiffies %lu\n",
  110. conn, &conn->c_laddr, &conn->c_faddr,
  111. conn->c_reconnect_jiffies);
  112. set_bit(RDS_RECONNECT_PENDING, &conn->c_flags);
  113. if (conn->c_reconnect_jiffies == 0) {
  114. conn->c_reconnect_jiffies = rds_sysctl_reconnect_min_jiffies;
  115. queue_delayed_work(rds_wq, &conn->c_conn_w, 0);
  116. return;
  117. }
  118. get_random_bytes(&rand, sizeof(rand));
  119. rdsdebug("%lu delay %lu ceil conn %p for %pI4 -> %pI4\n",
  120. rand % conn->c_reconnect_jiffies, conn->c_reconnect_jiffies,
  121. conn, &conn->c_laddr, &conn->c_faddr);
  122. queue_delayed_work(rds_wq, &conn->c_conn_w,
  123. rand % conn->c_reconnect_jiffies);
  124. conn->c_reconnect_jiffies = min(conn->c_reconnect_jiffies * 2,
  125. rds_sysctl_reconnect_max_jiffies);
  126. }
  127. void rds_connect_worker(struct work_struct *work)
  128. {
  129. struct rds_connection *conn = container_of(work, struct rds_connection, c_conn_w.work);
  130. int ret;
  131. clear_bit(RDS_RECONNECT_PENDING, &conn->c_flags);
  132. if (rds_conn_transition(conn, RDS_CONN_DOWN, RDS_CONN_CONNECTING)) {
  133. ret = conn->c_trans->conn_connect(conn);
  134. rdsdebug("conn %p for %pI4 to %pI4 dispatched, ret %d\n",
  135. conn, &conn->c_laddr, &conn->c_faddr, ret);
  136. if (ret) {
  137. if (rds_conn_transition(conn, RDS_CONN_CONNECTING, RDS_CONN_DOWN))
  138. rds_queue_reconnect(conn);
  139. else
  140. rds_conn_error(conn, "RDS: connect failed\n");
  141. }
  142. }
  143. }
  144. void rds_send_worker(struct work_struct *work)
  145. {
  146. struct rds_connection *conn = container_of(work, struct rds_connection, c_send_w.work);
  147. int ret;
  148. if (rds_conn_state(conn) == RDS_CONN_UP) {
  149. clear_bit(RDS_LL_SEND_FULL, &conn->c_flags);
  150. ret = rds_send_xmit(conn);
  151. cond_resched();
  152. rdsdebug("conn %p ret %d\n", conn, ret);
  153. switch (ret) {
  154. case -EAGAIN:
  155. rds_stats_inc(s_send_immediate_retry);
  156. queue_delayed_work(rds_wq, &conn->c_send_w, 0);
  157. break;
  158. case -ENOMEM:
  159. rds_stats_inc(s_send_delayed_retry);
  160. queue_delayed_work(rds_wq, &conn->c_send_w, 2);
  161. default:
  162. break;
  163. }
  164. }
  165. }
  166. void rds_recv_worker(struct work_struct *work)
  167. {
  168. struct rds_connection *conn = container_of(work, struct rds_connection, c_recv_w.work);
  169. int ret;
  170. if (rds_conn_state(conn) == RDS_CONN_UP) {
  171. ret = conn->c_trans->recv(conn);
  172. rdsdebug("conn %p ret %d\n", conn, ret);
  173. switch (ret) {
  174. case -EAGAIN:
  175. rds_stats_inc(s_recv_immediate_retry);
  176. queue_delayed_work(rds_wq, &conn->c_recv_w, 0);
  177. break;
  178. case -ENOMEM:
  179. rds_stats_inc(s_recv_delayed_retry);
  180. queue_delayed_work(rds_wq, &conn->c_recv_w, 2);
  181. default:
  182. break;
  183. }
  184. }
  185. }
  186. void rds_shutdown_worker(struct work_struct *work)
  187. {
  188. struct rds_connection *conn = container_of(work, struct rds_connection, c_down_w);
  189. rds_conn_shutdown(conn);
  190. }
  191. void rds_threads_exit(void)
  192. {
  193. destroy_workqueue(rds_wq);
  194. }
  195. int rds_threads_init(void)
  196. {
  197. rds_wq = create_singlethread_workqueue("krdsd");
  198. if (!rds_wq)
  199. return -ENOMEM;
  200. return 0;
  201. }