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 R2CLUSTER_TCP_INTERNAL_H
23 #define R2CLUSTER_TCP_INTERNAL_H
25 #define R2NET_MSG_MAGIC ((u16)0xfa55)
26 #define R2NET_MSG_STATUS_MAGIC ((u16)0xfa56)
27 #define R2NET_MSG_KEEP_REQ_MAGIC ((u16)0xfa57)
28 #define R2NET_MSG_KEEP_RESP_MAGIC ((u16)0xfa58)
30 * "data magic" is a long version of "status magic" where the message
31 * payload actually contains data to be passed in reply to certain messages
33 #define R2NET_MSG_DATA_MAGIC ((u16)0xfa59)
35 /* we're delaying our quorum decision so that heartbeat will have timed
36 * out truly dead nodes by the time we come around to making decisions
38 #define R2NET_QUORUM_DELAY_MS \
39 ((r2hb_dead_threshold + 2) * R2HB_REGION_TIMEOUT_MS)
42 * This version number represents quite a lot, unfortunately. It not
43 * only represents the raw network message protocol on the wire but also
44 * locking semantics of the file system using the protocol. It should
45 * be somewhere else, I'm sure, but right now it isn't.
47 * With version 11, we separate out the filesystem locking portion. The
48 * filesystem now has a major.minor version it negotiates. Version 11
49 * introduces this negotiation to the r2dlm protocol, and as such the
50 * version here in tcp_internal.h should not need to be bumped for
51 * filesystem locking changes.
54 * - Negotiation of filesystem locking in the dlm join.
57 * - Meta/data locks combined
63 * - Replace delete inode votes with a cluster lock
66 * - DLM join domain includes the live nodemap
69 * - DLM lockres remote refcount fixes.
72 * - Network timeout checking protocol
75 * - Remove i_generation from lock names for better stat performance.
78 * - Replace dentry votes with a cluster lock
81 * - full 64 bit i_size in the metadata lock lvbs
82 * - introduction of "rw" lock and pushing meta/data locking down
84 #define R2NET_PROTOCOL_VERSION 11ULL
85 struct r2net_handshake {
86 __be64 protocol_version;
88 __be32 r2hb_heartbeat_timeout_ms;
89 __be32 r2net_idle_timeout_ms;
90 __be32 r2net_keepalive_delay_ms;
91 __be32 r2net_reconnect_delay_ms;
95 /* this is never called from int/bh */
98 /* set the moment an sc is allocated and a connect is started */
99 struct r2net_sock_container *nn_sc;
100 /* _valid is only set after the handshake passes and tx can happen */
101 unsigned nn_sc_valid:1;
102 /* if this is set tx just returns it */
103 int nn_persistent_error;
104 /* It is only set to 1 after the idle time out. */
107 /* threads waiting for an sc to arrive wait on the wq for generation
108 * to increase. it is increased when a connecting socket succeeds
109 * or fails or when an accepted socket is attached. */
110 wait_queue_head_t nn_sc_wq;
112 struct idr nn_status_idr;
113 struct list_head nn_status_list;
115 /* connects are attempted from when heartbeat comes up until either hb
116 * goes down, the node is unconfigured, no connect attempts succeed
117 * before R2NET_CONN_IDLE_DELAY, or a connect succeeds. connect_work
118 * is queued from set_nn_state both from hb up and from itself if a
119 * connect attempt fails and so can be self-arming. shutdown is
120 * careful to first mark the nn such that no connects will be attempted
121 * before canceling delayed connect work and flushing the queue. */
122 struct delayed_work nn_connect_work;
123 unsigned long nn_last_connect_attempt;
125 /* this is queued as nodes come up and is canceled when a connection is
126 * established. this expiring gives up on the node and errors out
128 struct delayed_work nn_connect_expired;
130 /* after we give up on a socket we wait a while before deciding
131 * that it is still heartbeating and that we should do some
133 struct delayed_work nn_still_up;
136 struct r2net_sock_container {
138 /* the next two are valid for the life time of the sc */
139 struct socket *sc_sock;
140 struct r2nm_node *sc_node;
142 /* all of these sc work structs hold refs on the sc while they are
143 * queued. they should not be able to ref a freed sc. the teardown
144 * race is with r2net_wq destruction in r2net_stop_listening() */
146 /* rx and connect work are generated from socket callbacks. sc
147 * shutdown removes the callbacks and then flushes the work queue */
148 struct work_struct sc_rx_work;
149 struct work_struct sc_connect_work;
150 /* shutdown work is triggered in two ways. the simple way is
151 * for a code path calls ensure_shutdown which gets a lock, removes
152 * the sc from the nn, and queues the work. in this case the
153 * work is single-shot. the work is also queued from a sock
154 * callback, though, and in this case the work will find the sc
155 * still on the nn and will call ensure_shutdown itself.. this
156 * ends up triggering the shutdown work again, though nothing
157 * will be done in that second iteration. so work queue teardown
158 * has to be careful to remove the sc from the nn before waiting
159 * on the work queue so that the shutdown work doesn't remove the
160 * sc and rearm itself.
162 struct work_struct sc_shutdown_work;
164 struct timer_list sc_idle_timeout;
165 struct delayed_work sc_keepalive_work;
167 unsigned sc_handshake_ok:1;
169 struct page *sc_page;
172 /* original handlers for the sockets */
173 void (*sc_state_change)(struct sock *sk);
174 void (*sc_data_ready)(struct sock *sk, int bytes);
179 #ifdef CONFIG_DEBUG_FS
180 struct list_head sc_net_debug_item;
182 ktime_t sc_tv_data_ready;
183 ktime_t sc_tv_advance_start;
184 ktime_t sc_tv_advance_stop;
185 ktime_t sc_tv_func_start;
186 ktime_t sc_tv_func_stop;
188 #ifdef CONFIG_RAMSTER_FS_STATS
189 ktime_t sc_tv_acquiry_total;
190 ktime_t sc_tv_send_total;
191 ktime_t sc_tv_status_total;
194 ktime_t sc_tv_process_total;
196 struct mutex sc_send_lock;
199 struct r2net_msg_handler {
200 struct rb_node nh_node;
204 r2net_msg_handler_func *nh_func;
205 r2net_msg_handler_func *nh_func_data;
206 r2net_post_msg_handler_func
209 struct list_head nh_unregister_item;
212 enum r2net_system_error {
220 struct r2net_status_wait {
221 enum r2net_system_error ns_sys_status;
224 wait_queue_head_t ns_wq;
225 struct list_head ns_node_item;
228 #ifdef CONFIG_DEBUG_FS
229 /* just for state dumps */
230 struct r2net_send_tracking {
231 struct list_head st_net_debug_item;
232 struct task_struct *st_task;
233 struct r2net_sock_container *st_sc;
238 ktime_t st_sock_time;
239 ktime_t st_send_time;
240 ktime_t st_status_time;
243 struct r2net_send_tracking {
246 #endif /* CONFIG_DEBUG_FS */
248 #endif /* R2CLUSTER_TCP_INTERNAL_H */