![]() |
|
|||
0001 /* SPDX-License-Identifier: GPL-2.0-or-later */ 0002 /* 0003 * Copyright (C) 2005 Oracle. All rights reserved. 0004 */ 0005 0006 #ifndef O2CLUSTER_TCP_INTERNAL_H 0007 #define O2CLUSTER_TCP_INTERNAL_H 0008 0009 #define O2NET_MSG_MAGIC ((u16)0xfa55) 0010 #define O2NET_MSG_STATUS_MAGIC ((u16)0xfa56) 0011 #define O2NET_MSG_KEEP_REQ_MAGIC ((u16)0xfa57) 0012 #define O2NET_MSG_KEEP_RESP_MAGIC ((u16)0xfa58) 0013 0014 /* we're delaying our quorum decision so that heartbeat will have timed 0015 * out truly dead nodes by the time we come around to making decisions 0016 * on their number */ 0017 #define O2NET_QUORUM_DELAY_MS ((o2hb_dead_threshold + 2) * O2HB_REGION_TIMEOUT_MS) 0018 0019 /* 0020 * This version number represents quite a lot, unfortunately. It not 0021 * only represents the raw network message protocol on the wire but also 0022 * locking semantics of the file system using the protocol. It should 0023 * be somewhere else, I'm sure, but right now it isn't. 0024 * 0025 * With version 11, we separate out the filesystem locking portion. The 0026 * filesystem now has a major.minor version it negotiates. Version 11 0027 * introduces this negotiation to the o2dlm protocol, and as such the 0028 * version here in tcp_internal.h should not need to be bumped for 0029 * filesystem locking changes. 0030 * 0031 * New in version 11 0032 * - Negotiation of filesystem locking in the dlm join. 0033 * 0034 * New in version 10: 0035 * - Meta/data locks combined 0036 * 0037 * New in version 9: 0038 * - All votes removed 0039 * 0040 * New in version 8: 0041 * - Replace delete inode votes with a cluster lock 0042 * 0043 * New in version 7: 0044 * - DLM join domain includes the live nodemap 0045 * 0046 * New in version 6: 0047 * - DLM lockres remote refcount fixes. 0048 * 0049 * New in version 5: 0050 * - Network timeout checking protocol 0051 * 0052 * New in version 4: 0053 * - Remove i_generation from lock names for better stat performance. 0054 * 0055 * New in version 3: 0056 * - Replace dentry votes with a cluster lock 0057 * 0058 * New in version 2: 0059 * - full 64 bit i_size in the metadata lock lvbs 0060 * - introduction of "rw" lock and pushing meta/data locking down 0061 */ 0062 #define O2NET_PROTOCOL_VERSION 11ULL 0063 struct o2net_handshake { 0064 __be64 protocol_version; 0065 __be64 connector_id; 0066 __be32 o2hb_heartbeat_timeout_ms; 0067 __be32 o2net_idle_timeout_ms; 0068 __be32 o2net_keepalive_delay_ms; 0069 __be32 o2net_reconnect_delay_ms; 0070 }; 0071 0072 struct o2net_node { 0073 /* this is never called from int/bh */ 0074 spinlock_t nn_lock; 0075 0076 /* set the moment an sc is allocated and a connect is started */ 0077 struct o2net_sock_container *nn_sc; 0078 /* _valid is only set after the handshake passes and tx can happen */ 0079 unsigned nn_sc_valid:1; 0080 /* if this is set tx just returns it */ 0081 int nn_persistent_error; 0082 /* It is only set to 1 after the idle time out. */ 0083 atomic_t nn_timeout; 0084 0085 /* threads waiting for an sc to arrive wait on the wq for generation 0086 * to increase. it is increased when a connecting socket succeeds 0087 * or fails or when an accepted socket is attached. */ 0088 wait_queue_head_t nn_sc_wq; 0089 0090 struct idr nn_status_idr; 0091 struct list_head nn_status_list; 0092 0093 /* connects are attempted from when heartbeat comes up until either hb 0094 * goes down, the node is unconfigured, or a connect succeeds. 0095 * connect_work is queued from set_nn_state both from hb up and from 0096 * itself if a connect attempt fails and so can be self-arming. 0097 * shutdown is careful to first mark the nn such that no connects will 0098 * be attempted before canceling delayed connect work and flushing the 0099 * queue. */ 0100 struct delayed_work nn_connect_work; 0101 unsigned long nn_last_connect_attempt; 0102 0103 /* this is queued as nodes come up and is canceled when a connection is 0104 * established. this expiring gives up on the node and errors out 0105 * transmits */ 0106 struct delayed_work nn_connect_expired; 0107 0108 /* after we give up on a socket we wait a while before deciding 0109 * that it is still heartbeating and that we should do some 0110 * quorum work */ 0111 struct delayed_work nn_still_up; 0112 }; 0113 0114 struct o2net_sock_container { 0115 struct kref sc_kref; 0116 /* the next two are valid for the life time of the sc */ 0117 struct socket *sc_sock; 0118 struct o2nm_node *sc_node; 0119 0120 /* all of these sc work structs hold refs on the sc while they are 0121 * queued. they should not be able to ref a freed sc. the teardown 0122 * race is with o2net_wq destruction in o2net_stop_listening() */ 0123 0124 /* rx and connect work are generated from socket callbacks. sc 0125 * shutdown removes the callbacks and then flushes the work queue */ 0126 struct work_struct sc_rx_work; 0127 struct work_struct sc_connect_work; 0128 /* shutdown work is triggered in two ways. the simple way is 0129 * for a code path calls ensure_shutdown which gets a lock, removes 0130 * the sc from the nn, and queues the work. in this case the 0131 * work is single-shot. the work is also queued from a sock 0132 * callback, though, and in this case the work will find the sc 0133 * still on the nn and will call ensure_shutdown itself.. this 0134 * ends up triggering the shutdown work again, though nothing 0135 * will be done in that second iteration. so work queue teardown 0136 * has to be careful to remove the sc from the nn before waiting 0137 * on the work queue so that the shutdown work doesn't remove the 0138 * sc and rearm itself. 0139 */ 0140 struct work_struct sc_shutdown_work; 0141 0142 struct timer_list sc_idle_timeout; 0143 struct delayed_work sc_keepalive_work; 0144 0145 unsigned sc_handshake_ok:1; 0146 0147 struct page *sc_page; 0148 size_t sc_page_off; 0149 0150 /* original handlers for the sockets */ 0151 void (*sc_state_change)(struct sock *sk); 0152 void (*sc_data_ready)(struct sock *sk); 0153 0154 u32 sc_msg_key; 0155 u16 sc_msg_type; 0156 0157 #ifdef CONFIG_DEBUG_FS 0158 struct list_head sc_net_debug_item; 0159 ktime_t sc_tv_timer; 0160 ktime_t sc_tv_data_ready; 0161 ktime_t sc_tv_advance_start; 0162 ktime_t sc_tv_advance_stop; 0163 ktime_t sc_tv_func_start; 0164 ktime_t sc_tv_func_stop; 0165 #endif 0166 #ifdef CONFIG_OCFS2_FS_STATS 0167 ktime_t sc_tv_acquiry_total; 0168 ktime_t sc_tv_send_total; 0169 ktime_t sc_tv_status_total; 0170 u32 sc_send_count; 0171 u32 sc_recv_count; 0172 ktime_t sc_tv_process_total; 0173 #endif 0174 struct mutex sc_send_lock; 0175 }; 0176 0177 struct o2net_msg_handler { 0178 struct rb_node nh_node; 0179 u32 nh_max_len; 0180 u32 nh_msg_type; 0181 u32 nh_key; 0182 o2net_msg_handler_func *nh_func; 0183 void *nh_func_data; 0184 o2net_post_msg_handler_func 0185 *nh_post_func; 0186 struct kref nh_kref; 0187 struct list_head nh_unregister_item; 0188 }; 0189 0190 enum o2net_system_error { 0191 O2NET_ERR_NONE = 0, 0192 O2NET_ERR_NO_HNDLR, 0193 O2NET_ERR_OVERFLOW, 0194 O2NET_ERR_DIED, 0195 O2NET_ERR_MAX 0196 }; 0197 0198 struct o2net_status_wait { 0199 enum o2net_system_error ns_sys_status; 0200 s32 ns_status; 0201 int ns_id; 0202 wait_queue_head_t ns_wq; 0203 struct list_head ns_node_item; 0204 }; 0205 0206 #ifdef CONFIG_DEBUG_FS 0207 /* just for state dumps */ 0208 struct o2net_send_tracking { 0209 struct list_head st_net_debug_item; 0210 struct task_struct *st_task; 0211 struct o2net_sock_container *st_sc; 0212 u32 st_id; 0213 u32 st_msg_type; 0214 u32 st_msg_key; 0215 u8 st_node; 0216 ktime_t st_sock_time; 0217 ktime_t st_send_time; 0218 ktime_t st_status_time; 0219 }; 0220 #else 0221 struct o2net_send_tracking { 0222 u32 dummy; 0223 }; 0224 #endif /* CONFIG_DEBUG_FS */ 0225 0226 #endif /* O2CLUSTER_TCP_INTERNAL_H */
[ Source navigation ] | [ Diff markup ] | [ Identifier search ] | [ general search ] |
This page was automatically generated by the 2.1.0 LXR engine. The LXR team |
![]() ![]() |