1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * Copyright (C) 2005 Oracle. All rights reserved.
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.
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.
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.
22 #ifndef O2CLUSTER_TCP_INTERNAL_H
23 #define O2CLUSTER_TCP_INTERNAL_H
25 #define O2NET_MSG_MAGIC ((u16)0xfa55)
26 #define O2NET_MSG_STATUS_MAGIC ((u16)0xfa56)
27 #define O2NET_MSG_KEEP_REQ_MAGIC ((u16)0xfa57)
28 #define O2NET_MSG_KEEP_RESP_MAGIC ((u16)0xfa58)
30 /* we're delaying our quorum decision so that heartbeat will have timed
31 * out truly dead nodes by the time we come around to making decisions
33 #define O2NET_QUORUM_DELAY_MS ((o2hb_dead_threshold + 2) * O2HB_REGION_TIMEOUT_MS)
36 * This version number represents quite a lot, unfortunately. It not
37 * only represents the raw network message protocol on the wire but also
38 * locking semantics of the file system using the protocol. It should
39 * be somewhere else, I'm sure, but right now it isn't.
41 * With version 11, we separate out the filesystem locking portion. The
42 * filesystem now has a major.minor version it negotiates. Version 11
43 * introduces this negotiation to the o2dlm protocol, and as such the
44 * version here in tcp_internal.h should not need to be bumped for
45 * filesystem locking changes.
48 * - Negotiation of filesystem locking in the dlm join.
51 * - Meta/data locks combined
57 * - Replace delete inode votes with a cluster lock
60 * - DLM join domain includes the live nodemap
63 * - DLM lockres remote refcount fixes.
66 * - Network timeout checking protocol
69 * - Remove i_generation from lock names for better stat performance.
72 * - Replace dentry votes with a cluster lock
75 * - full 64 bit i_size in the metadata lock lvbs
76 * - introduction of "rw" lock and pushing meta/data locking down
78 #define O2NET_PROTOCOL_VERSION 11ULL
79 struct o2net_handshake {
80 __be64 protocol_version;
82 __be32 o2hb_heartbeat_timeout_ms;
83 __be32 o2net_idle_timeout_ms;
84 __be32 o2net_keepalive_delay_ms;
85 __be32 o2net_reconnect_delay_ms;
89 /* this is never called from int/bh */
92 /* set the moment an sc is allocated and a connect is started */
93 struct o2net_sock_container *nn_sc;
94 /* _valid is only set after the handshake passes and tx can happen */
95 unsigned nn_sc_valid:1;
96 /* if this is set tx just returns it */
97 int nn_persistent_error;
98 /* It is only set to 1 after the idle time out. */
101 /* threads waiting for an sc to arrive wait on the wq for generation
102 * to increase. it is increased when a connecting socket succeeds
103 * or fails or when an accepted socket is attached. */
104 wait_queue_head_t nn_sc_wq;
106 struct idr nn_status_idr;
107 struct list_head nn_status_list;
109 /* connects are attempted from when heartbeat comes up until either hb
110 * goes down, the node is unconfigured, no connect attempts succeed
111 * before O2NET_CONN_IDLE_DELAY, or a connect succeeds. connect_work
112 * is queued from set_nn_state both from hb up and from itself if a
113 * connect attempt fails and so can be self-arming. shutdown is
114 * careful to first mark the nn such that no connects will be attempted
115 * before canceling delayed connect work and flushing the queue. */
116 struct delayed_work nn_connect_work;
117 unsigned long nn_last_connect_attempt;
119 /* this is queued as nodes come up and is canceled when a connection is
120 * established. this expiring gives up on the node and errors out
122 struct delayed_work nn_connect_expired;
124 /* after we give up on a socket we wait a while before deciding
125 * that it is still heartbeating and that we should do some
127 struct delayed_work nn_still_up;
130 struct o2net_sock_container {
132 /* the next two are vaild for the life time of the sc */
133 struct socket *sc_sock;
134 struct o2nm_node *sc_node;
136 /* all of these sc work structs hold refs on the sc while they are
137 * queued. they should not be able to ref a freed sc. the teardown
138 * race is with o2net_wq destruction in o2net_stop_listening() */
140 /* rx and connect work are generated from socket callbacks. sc
141 * shutdown removes the callbacks and then flushes the work queue */
142 struct work_struct sc_rx_work;
143 struct work_struct sc_connect_work;
144 /* shutdown work is triggered in two ways. the simple way is
145 * for a code path calls ensure_shutdown which gets a lock, removes
146 * the sc from the nn, and queues the work. in this case the
147 * work is single-shot. the work is also queued from a sock
148 * callback, though, and in this case the work will find the sc
149 * still on the nn and will call ensure_shutdown itself.. this
150 * ends up triggering the shutdown work again, though nothing
151 * will be done in that second iteration. so work queue teardown
152 * has to be careful to remove the sc from the nn before waiting
153 * on the work queue so that the shutdown work doesn't remove the
154 * sc and rearm itself.
156 struct work_struct sc_shutdown_work;
158 struct timer_list sc_idle_timeout;
159 struct delayed_work sc_keepalive_work;
161 unsigned sc_handshake_ok:1;
163 struct page *sc_page;
166 /* original handlers for the sockets */
167 void (*sc_state_change)(struct sock *sk);
168 void (*sc_data_ready)(struct sock *sk, int bytes);
169 #ifdef CONFIG_DEBUG_FS
170 struct list_head sc_net_debug_item;
172 struct timeval sc_tv_timer;
173 struct timeval sc_tv_data_ready;
174 struct timeval sc_tv_advance_start;
175 struct timeval sc_tv_advance_stop;
176 struct timeval sc_tv_func_start;
177 struct timeval sc_tv_func_stop;
181 struct mutex sc_send_lock;
184 struct o2net_msg_handler {
185 struct rb_node nh_node;
189 o2net_msg_handler_func *nh_func;
190 o2net_msg_handler_func *nh_func_data;
191 o2net_post_msg_handler_func
194 struct list_head nh_unregister_item;
197 enum o2net_system_error {
205 struct o2net_status_wait {
206 enum o2net_system_error ns_sys_status;
209 wait_queue_head_t ns_wq;
210 struct list_head ns_node_item;
213 #ifdef CONFIG_DEBUG_FS
214 /* just for state dumps */
215 struct o2net_send_tracking {
216 struct list_head st_net_debug_item;
217 struct task_struct *st_task;
218 struct o2net_sock_container *st_sc;
223 struct timeval st_sock_time;
224 struct timeval st_send_time;
225 struct timeval st_status_time;
228 void o2net_init_nst(struct o2net_send_tracking *nst, u32 msgtype,
229 u32 msgkey, struct task_struct *task, u8 node);
230 void o2net_set_nst_sock_time(struct o2net_send_tracking *nst);
231 void o2net_set_nst_send_time(struct o2net_send_tracking *nst);
232 void o2net_set_nst_status_time(struct o2net_send_tracking *nst);
233 void o2net_set_nst_sock_container(struct o2net_send_tracking *nst,
234 struct o2net_sock_container *sc);
235 void o2net_set_nst_msg_id(struct o2net_send_tracking *nst, u32 msg_id);
238 struct o2net_send_tracking {
242 static inline void o2net_init_nst(struct o2net_send_tracking *nst, u32 msgtype,
243 u32 msgkey, struct task_struct *task, u8 node)
246 static inline void o2net_set_nst_sock_time(struct o2net_send_tracking *nst)
249 static inline void o2net_set_nst_send_time(struct o2net_send_tracking *nst)
252 static inline void o2net_set_nst_status_time(struct o2net_send_tracking *nst)
255 static inline void o2net_set_nst_sock_container(struct o2net_send_tracking *nst,
256 struct o2net_sock_container *sc)
259 static inline void o2net_set_nst_msg_id(struct o2net_send_tracking *nst,
263 #endif /* CONFIG_DEBUG_FS */
265 #endif /* O2CLUSTER_TCP_INTERNAL_H */