tcp_internal.h 6.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228
  1. /* SPDX-License-Identifier: GPL-2.0-or-later */
  2. /* -*- mode: c; c-basic-offset: 8; -*-
  3. * vim: noexpandtab sw=8 ts=8 sts=0:
  4. *
  5. * Copyright (C) 2005 Oracle. All rights reserved.
  6. */
  7. #ifndef O2CLUSTER_TCP_INTERNAL_H
  8. #define O2CLUSTER_TCP_INTERNAL_H
  9. #define O2NET_MSG_MAGIC ((u16)0xfa55)
  10. #define O2NET_MSG_STATUS_MAGIC ((u16)0xfa56)
  11. #define O2NET_MSG_KEEP_REQ_MAGIC ((u16)0xfa57)
  12. #define O2NET_MSG_KEEP_RESP_MAGIC ((u16)0xfa58)
  13. /* we're delaying our quorum decision so that heartbeat will have timed
  14. * out truly dead nodes by the time we come around to making decisions
  15. * on their number */
  16. #define O2NET_QUORUM_DELAY_MS ((o2hb_dead_threshold + 2) * O2HB_REGION_TIMEOUT_MS)
  17. /*
  18. * This version number represents quite a lot, unfortunately. It not
  19. * only represents the raw network message protocol on the wire but also
  20. * locking semantics of the file system using the protocol. It should
  21. * be somewhere else, I'm sure, but right now it isn't.
  22. *
  23. * With version 11, we separate out the filesystem locking portion. The
  24. * filesystem now has a major.minor version it negotiates. Version 11
  25. * introduces this negotiation to the o2dlm protocol, and as such the
  26. * version here in tcp_internal.h should not need to be bumped for
  27. * filesystem locking changes.
  28. *
  29. * New in version 11
  30. * - Negotiation of filesystem locking in the dlm join.
  31. *
  32. * New in version 10:
  33. * - Meta/data locks combined
  34. *
  35. * New in version 9:
  36. * - All votes removed
  37. *
  38. * New in version 8:
  39. * - Replace delete inode votes with a cluster lock
  40. *
  41. * New in version 7:
  42. * - DLM join domain includes the live nodemap
  43. *
  44. * New in version 6:
  45. * - DLM lockres remote refcount fixes.
  46. *
  47. * New in version 5:
  48. * - Network timeout checking protocol
  49. *
  50. * New in version 4:
  51. * - Remove i_generation from lock names for better stat performance.
  52. *
  53. * New in version 3:
  54. * - Replace dentry votes with a cluster lock
  55. *
  56. * New in version 2:
  57. * - full 64 bit i_size in the metadata lock lvbs
  58. * - introduction of "rw" lock and pushing meta/data locking down
  59. */
  60. #define O2NET_PROTOCOL_VERSION 11ULL
  61. struct o2net_handshake {
  62. __be64 protocol_version;
  63. __be64 connector_id;
  64. __be32 o2hb_heartbeat_timeout_ms;
  65. __be32 o2net_idle_timeout_ms;
  66. __be32 o2net_keepalive_delay_ms;
  67. __be32 o2net_reconnect_delay_ms;
  68. };
  69. struct o2net_node {
  70. /* this is never called from int/bh */
  71. spinlock_t nn_lock;
  72. /* set the moment an sc is allocated and a connect is started */
  73. struct o2net_sock_container *nn_sc;
  74. /* _valid is only set after the handshake passes and tx can happen */
  75. unsigned nn_sc_valid:1;
  76. /* if this is set tx just returns it */
  77. int nn_persistent_error;
  78. /* It is only set to 1 after the idle time out. */
  79. atomic_t nn_timeout;
  80. /* threads waiting for an sc to arrive wait on the wq for generation
  81. * to increase. it is increased when a connecting socket succeeds
  82. * or fails or when an accepted socket is attached. */
  83. wait_queue_head_t nn_sc_wq;
  84. struct idr nn_status_idr;
  85. struct list_head nn_status_list;
  86. /* connects are attempted from when heartbeat comes up until either hb
  87. * goes down, the node is unconfigured, or a connect succeeds.
  88. * connect_work is queued from set_nn_state both from hb up and from
  89. * itself if a connect attempt fails and so can be self-arming.
  90. * shutdown is careful to first mark the nn such that no connects will
  91. * be attempted before canceling delayed connect work and flushing the
  92. * queue. */
  93. struct delayed_work nn_connect_work;
  94. unsigned long nn_last_connect_attempt;
  95. /* this is queued as nodes come up and is canceled when a connection is
  96. * established. this expiring gives up on the node and errors out
  97. * transmits */
  98. struct delayed_work nn_connect_expired;
  99. /* after we give up on a socket we wait a while before deciding
  100. * that it is still heartbeating and that we should do some
  101. * quorum work */
  102. struct delayed_work nn_still_up;
  103. };
  104. struct o2net_sock_container {
  105. struct kref sc_kref;
  106. /* the next two are valid for the life time of the sc */
  107. struct socket *sc_sock;
  108. struct o2nm_node *sc_node;
  109. /* all of these sc work structs hold refs on the sc while they are
  110. * queued. they should not be able to ref a freed sc. the teardown
  111. * race is with o2net_wq destruction in o2net_stop_listening() */
  112. /* rx and connect work are generated from socket callbacks. sc
  113. * shutdown removes the callbacks and then flushes the work queue */
  114. struct work_struct sc_rx_work;
  115. struct work_struct sc_connect_work;
  116. /* shutdown work is triggered in two ways. the simple way is
  117. * for a code path calls ensure_shutdown which gets a lock, removes
  118. * the sc from the nn, and queues the work. in this case the
  119. * work is single-shot. the work is also queued from a sock
  120. * callback, though, and in this case the work will find the sc
  121. * still on the nn and will call ensure_shutdown itself.. this
  122. * ends up triggering the shutdown work again, though nothing
  123. * will be done in that second iteration. so work queue teardown
  124. * has to be careful to remove the sc from the nn before waiting
  125. * on the work queue so that the shutdown work doesn't remove the
  126. * sc and rearm itself.
  127. */
  128. struct work_struct sc_shutdown_work;
  129. struct timer_list sc_idle_timeout;
  130. struct delayed_work sc_keepalive_work;
  131. unsigned sc_handshake_ok:1;
  132. struct page *sc_page;
  133. size_t sc_page_off;
  134. /* original handlers for the sockets */
  135. void (*sc_state_change)(struct sock *sk);
  136. void (*sc_data_ready)(struct sock *sk);
  137. u32 sc_msg_key;
  138. u16 sc_msg_type;
  139. #ifdef CONFIG_DEBUG_FS
  140. struct list_head sc_net_debug_item;
  141. ktime_t sc_tv_timer;
  142. ktime_t sc_tv_data_ready;
  143. ktime_t sc_tv_advance_start;
  144. ktime_t sc_tv_advance_stop;
  145. ktime_t sc_tv_func_start;
  146. ktime_t sc_tv_func_stop;
  147. #endif
  148. #ifdef CONFIG_OCFS2_FS_STATS
  149. ktime_t sc_tv_acquiry_total;
  150. ktime_t sc_tv_send_total;
  151. ktime_t sc_tv_status_total;
  152. u32 sc_send_count;
  153. u32 sc_recv_count;
  154. ktime_t sc_tv_process_total;
  155. #endif
  156. struct mutex sc_send_lock;
  157. };
  158. struct o2net_msg_handler {
  159. struct rb_node nh_node;
  160. u32 nh_max_len;
  161. u32 nh_msg_type;
  162. u32 nh_key;
  163. o2net_msg_handler_func *nh_func;
  164. void *nh_func_data;
  165. o2net_post_msg_handler_func
  166. *nh_post_func;
  167. struct kref nh_kref;
  168. struct list_head nh_unregister_item;
  169. };
  170. enum o2net_system_error {
  171. O2NET_ERR_NONE = 0,
  172. O2NET_ERR_NO_HNDLR,
  173. O2NET_ERR_OVERFLOW,
  174. O2NET_ERR_DIED,
  175. O2NET_ERR_MAX
  176. };
  177. struct o2net_status_wait {
  178. enum o2net_system_error ns_sys_status;
  179. s32 ns_status;
  180. int ns_id;
  181. wait_queue_head_t ns_wq;
  182. struct list_head ns_node_item;
  183. };
  184. #ifdef CONFIG_DEBUG_FS
  185. /* just for state dumps */
  186. struct o2net_send_tracking {
  187. struct list_head st_net_debug_item;
  188. struct task_struct *st_task;
  189. struct o2net_sock_container *st_sc;
  190. u32 st_id;
  191. u32 st_msg_type;
  192. u32 st_msg_key;
  193. u8 st_node;
  194. ktime_t st_sock_time;
  195. ktime_t st_send_time;
  196. ktime_t st_status_time;
  197. };
  198. #else
  199. struct o2net_send_tracking {
  200. u32 dummy;
  201. };
  202. #endif /* CONFIG_DEBUG_FS */
  203. #endif /* O2CLUSTER_TCP_INTERNAL_H */