tcp_internal.h 7.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245
  1. /* -*- mode: c; c-basic-offset: 8; -*-
  2. * vim: noexpandtab sw=8 ts=8 sts=0:
  3. *
  4. * Copyright (C) 2005 Oracle. All rights reserved.
  5. *
  6. * This program is free software; you can redistribute it and/or
  7. * modify it under the terms of the GNU General Public
  8. * License as published by the Free Software Foundation; either
  9. * version 2 of the License, or (at your option) any later version.
  10. *
  11. * This program is distributed in the hope that it will be useful,
  12. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14. * General Public License for more details.
  15. *
  16. * You should have received a copy of the GNU General Public
  17. * License along with this program; if not, write to the
  18. * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19. * Boston, MA 021110-1307, USA.
  20. */
  21. #ifndef O2CLUSTER_TCP_INTERNAL_H
  22. #define O2CLUSTER_TCP_INTERNAL_H
  23. #define O2NET_MSG_MAGIC ((u16)0xfa55)
  24. #define O2NET_MSG_STATUS_MAGIC ((u16)0xfa56)
  25. #define O2NET_MSG_KEEP_REQ_MAGIC ((u16)0xfa57)
  26. #define O2NET_MSG_KEEP_RESP_MAGIC ((u16)0xfa58)
  27. /* we're delaying our quorum decision so that heartbeat will have timed
  28. * out truly dead nodes by the time we come around to making decisions
  29. * on their number */
  30. #define O2NET_QUORUM_DELAY_MS ((o2hb_dead_threshold + 2) * O2HB_REGION_TIMEOUT_MS)
  31. /*
  32. * This version number represents quite a lot, unfortunately. It not
  33. * only represents the raw network message protocol on the wire but also
  34. * locking semantics of the file system using the protocol. It should
  35. * be somewhere else, I'm sure, but right now it isn't.
  36. *
  37. * With version 11, we separate out the filesystem locking portion. The
  38. * filesystem now has a major.minor version it negotiates. Version 11
  39. * introduces this negotiation to the o2dlm protocol, and as such the
  40. * version here in tcp_internal.h should not need to be bumped for
  41. * filesystem locking changes.
  42. *
  43. * New in version 12
  44. * - Negotiate hb timeout when storage is down.
  45. *
  46. * New in version 11
  47. * - Negotiation of filesystem locking in the dlm join.
  48. *
  49. * New in version 10:
  50. * - Meta/data locks combined
  51. *
  52. * New in version 9:
  53. * - All votes removed
  54. *
  55. * New in version 8:
  56. * - Replace delete inode votes with a cluster lock
  57. *
  58. * New in version 7:
  59. * - DLM join domain includes the live nodemap
  60. *
  61. * New in version 6:
  62. * - DLM lockres remote refcount fixes.
  63. *
  64. * New in version 5:
  65. * - Network timeout checking protocol
  66. *
  67. * New in version 4:
  68. * - Remove i_generation from lock names for better stat performance.
  69. *
  70. * New in version 3:
  71. * - Replace dentry votes with a cluster lock
  72. *
  73. * New in version 2:
  74. * - full 64 bit i_size in the metadata lock lvbs
  75. * - introduction of "rw" lock and pushing meta/data locking down
  76. */
  77. #define O2NET_PROTOCOL_VERSION 12ULL
  78. struct o2net_handshake {
  79. __be64 protocol_version;
  80. __be64 connector_id;
  81. __be32 o2hb_heartbeat_timeout_ms;
  82. __be32 o2net_idle_timeout_ms;
  83. __be32 o2net_keepalive_delay_ms;
  84. __be32 o2net_reconnect_delay_ms;
  85. };
  86. struct o2net_node {
  87. /* this is never called from int/bh */
  88. spinlock_t nn_lock;
  89. /* set the moment an sc is allocated and a connect is started */
  90. struct o2net_sock_container *nn_sc;
  91. /* _valid is only set after the handshake passes and tx can happen */
  92. unsigned nn_sc_valid:1;
  93. /* if this is set tx just returns it */
  94. int nn_persistent_error;
  95. /* It is only set to 1 after the idle time out. */
  96. atomic_t nn_timeout;
  97. /* threads waiting for an sc to arrive wait on the wq for generation
  98. * to increase. it is increased when a connecting socket succeeds
  99. * or fails or when an accepted socket is attached. */
  100. wait_queue_head_t nn_sc_wq;
  101. struct idr nn_status_idr;
  102. struct list_head nn_status_list;
  103. /* connects are attempted from when heartbeat comes up until either hb
  104. * goes down, the node is unconfigured, or a connect succeeds.
  105. * connect_work is queued from set_nn_state both from hb up and from
  106. * itself if a connect attempt fails and so can be self-arming.
  107. * shutdown is careful to first mark the nn such that no connects will
  108. * be attempted before canceling delayed connect work and flushing the
  109. * queue. */
  110. struct delayed_work nn_connect_work;
  111. unsigned long nn_last_connect_attempt;
  112. /* this is queued as nodes come up and is canceled when a connection is
  113. * established. this expiring gives up on the node and errors out
  114. * transmits */
  115. struct delayed_work nn_connect_expired;
  116. /* after we give up on a socket we wait a while before deciding
  117. * that it is still heartbeating and that we should do some
  118. * quorum work */
  119. struct delayed_work nn_still_up;
  120. };
  121. struct o2net_sock_container {
  122. struct kref sc_kref;
  123. /* the next two are valid for the life time of the sc */
  124. struct socket *sc_sock;
  125. struct o2nm_node *sc_node;
  126. /* all of these sc work structs hold refs on the sc while they are
  127. * queued. they should not be able to ref a freed sc. the teardown
  128. * race is with o2net_wq destruction in o2net_stop_listening() */
  129. /* rx and connect work are generated from socket callbacks. sc
  130. * shutdown removes the callbacks and then flushes the work queue */
  131. struct work_struct sc_rx_work;
  132. struct work_struct sc_connect_work;
  133. /* shutdown work is triggered in two ways. the simple way is
  134. * for a code path calls ensure_shutdown which gets a lock, removes
  135. * the sc from the nn, and queues the work. in this case the
  136. * work is single-shot. the work is also queued from a sock
  137. * callback, though, and in this case the work will find the sc
  138. * still on the nn and will call ensure_shutdown itself.. this
  139. * ends up triggering the shutdown work again, though nothing
  140. * will be done in that second iteration. so work queue teardown
  141. * has to be careful to remove the sc from the nn before waiting
  142. * on the work queue so that the shutdown work doesn't remove the
  143. * sc and rearm itself.
  144. */
  145. struct work_struct sc_shutdown_work;
  146. struct timer_list sc_idle_timeout;
  147. struct delayed_work sc_keepalive_work;
  148. unsigned sc_handshake_ok:1;
  149. struct page *sc_page;
  150. size_t sc_page_off;
  151. /* original handlers for the sockets */
  152. void (*sc_state_change)(struct sock *sk);
  153. void (*sc_data_ready)(struct sock *sk);
  154. u32 sc_msg_key;
  155. u16 sc_msg_type;
  156. #ifdef CONFIG_DEBUG_FS
  157. struct list_head sc_net_debug_item;
  158. ktime_t sc_tv_timer;
  159. ktime_t sc_tv_data_ready;
  160. ktime_t sc_tv_advance_start;
  161. ktime_t sc_tv_advance_stop;
  162. ktime_t sc_tv_func_start;
  163. ktime_t sc_tv_func_stop;
  164. #endif
  165. #ifdef CONFIG_OCFS2_FS_STATS
  166. ktime_t sc_tv_acquiry_total;
  167. ktime_t sc_tv_send_total;
  168. ktime_t sc_tv_status_total;
  169. u32 sc_send_count;
  170. u32 sc_recv_count;
  171. ktime_t sc_tv_process_total;
  172. #endif
  173. struct mutex sc_send_lock;
  174. };
  175. struct o2net_msg_handler {
  176. struct rb_node nh_node;
  177. u32 nh_max_len;
  178. u32 nh_msg_type;
  179. u32 nh_key;
  180. o2net_msg_handler_func *nh_func;
  181. o2net_msg_handler_func *nh_func_data;
  182. o2net_post_msg_handler_func
  183. *nh_post_func;
  184. struct kref nh_kref;
  185. struct list_head nh_unregister_item;
  186. };
  187. enum o2net_system_error {
  188. O2NET_ERR_NONE = 0,
  189. O2NET_ERR_NO_HNDLR,
  190. O2NET_ERR_OVERFLOW,
  191. O2NET_ERR_DIED,
  192. O2NET_ERR_MAX
  193. };
  194. struct o2net_status_wait {
  195. enum o2net_system_error ns_sys_status;
  196. s32 ns_status;
  197. int ns_id;
  198. wait_queue_head_t ns_wq;
  199. struct list_head ns_node_item;
  200. };
  201. #ifdef CONFIG_DEBUG_FS
  202. /* just for state dumps */
  203. struct o2net_send_tracking {
  204. struct list_head st_net_debug_item;
  205. struct task_struct *st_task;
  206. struct o2net_sock_container *st_sc;
  207. u32 st_id;
  208. u32 st_msg_type;
  209. u32 st_msg_key;
  210. u8 st_node;
  211. ktime_t st_sock_time;
  212. ktime_t st_send_time;
  213. ktime_t st_status_time;
  214. };
  215. #else
  216. struct o2net_send_tracking {
  217. u32 dummy;
  218. };
  219. #endif /* CONFIG_DEBUG_FS */
  220. #endif /* O2CLUSTER_TCP_INTERNAL_H */