blob: 4b46aac7d243edd3d023935578d71ab457317bcc [file] [log] [blame]
Zach Brown98211482005-12-15 14:31:23 -08001/* -*- 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
22#ifndef O2CLUSTER_TCP_INTERNAL_H
23#define O2CLUSTER_TCP_INTERNAL_H
24
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)
29
30/* same as hb delay, we're waiting for another node to recognize our hb */
31#define O2NET_RECONNECT_DELAY_MS O2HB_REGION_TIMEOUT_MS
32
33/* we're delaying our quorum decision so that heartbeat will have timed
34 * out truly dead nodes by the time we come around to making decisions
35 * on their number */
36#define O2NET_QUORUM_DELAY_MS ((o2hb_dead_threshold + 2) * O2HB_REGION_TIMEOUT_MS)
37
38#define O2NET_KEEPALIVE_DELAY_SECS 5
39#define O2NET_IDLE_TIMEOUT_SECS 10
40
41/*
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.
46 *
Mark Fasheh24c19ef2006-09-22 17:28:19 -070047 * New in version 4:
48 * - Remove i_generation from lock names for better stat performance.
49 *
Mark Fasheh379dfe92006-09-08 14:21:03 -070050 * New in version 3:
51 * - Replace dentry votes with a cluster lock
52 *
Zach Brown98211482005-12-15 14:31:23 -080053 * New in version 2:
54 * - full 64 bit i_size in the metadata lock lvbs
55 * - introduction of "rw" lock and pushing meta/data locking down
56 */
Mark Fasheh24c19ef2006-09-22 17:28:19 -070057#define O2NET_PROTOCOL_VERSION 4ULL
Zach Brown98211482005-12-15 14:31:23 -080058struct o2net_handshake {
59 __be64 protocol_version;
60 __be64 connector_id;
61};
62
63struct o2net_node {
64 /* this is never called from int/bh */
65 spinlock_t nn_lock;
66
67 /* set the moment an sc is allocated and a connect is started */
68 struct o2net_sock_container *nn_sc;
69 /* _valid is only set after the handshake passes and tx can happen */
70 unsigned nn_sc_valid:1;
71 /* if this is set tx just returns it */
72 int nn_persistent_error;
73
74 /* threads waiting for an sc to arrive wait on the wq for generation
75 * to increase. it is increased when a connecting socket succeeds
76 * or fails or when an accepted socket is attached. */
77 wait_queue_head_t nn_sc_wq;
78
79 struct idr nn_status_idr;
80 struct list_head nn_status_list;
81
82 /* connects are attempted from when heartbeat comes up until either hb
83 * goes down, the node is unconfigured, no connect attempts succeed
84 * before O2NET_CONN_IDLE_DELAY, or a connect succeeds. connect_work
85 * is queued from set_nn_state both from hb up and from itself if a
86 * connect attempt fails and so can be self-arming. shutdown is
87 * careful to first mark the nn such that no connects will be attempted
88 * before canceling delayed connect work and flushing the queue. */
89 struct work_struct nn_connect_work;
90 unsigned long nn_last_connect_attempt;
91
92 /* this is queued as nodes come up and is canceled when a connection is
93 * established. this expiring gives up on the node and errors out
94 * transmits */
95 struct work_struct nn_connect_expired;
96
97 /* after we give up on a socket we wait a while before deciding
98 * that it is still heartbeating and that we should do some
99 * quorum work */
100 struct work_struct nn_still_up;
101};
102
103struct o2net_sock_container {
104 struct kref sc_kref;
105 /* the next two are vaild for the life time of the sc */
106 struct socket *sc_sock;
107 struct o2nm_node *sc_node;
108
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
113 /* rx and connect work are generated from socket callbacks. sc
114 * shutdown removes the callbacks and then flushes the work queue */
115 struct work_struct sc_rx_work;
116 struct work_struct sc_connect_work;
117 /* shutdown work is triggered in two ways. the simple way is
118 * for a code path calls ensure_shutdown which gets a lock, removes
119 * the sc from the nn, and queues the work. in this case the
120 * work is single-shot. the work is also queued from a sock
121 * callback, though, and in this case the work will find the sc
122 * still on the nn and will call ensure_shutdown itself.. this
123 * ends up triggering the shutdown work again, though nothing
124 * will be done in that second iteration. so work queue teardown
125 * has to be careful to remove the sc from the nn before waiting
126 * on the work queue so that the shutdown work doesn't remove the
127 * sc and rearm itself.
128 */
129 struct work_struct sc_shutdown_work;
130
131 struct timer_list sc_idle_timeout;
132 struct work_struct sc_keepalive_work;
133
134 unsigned sc_handshake_ok:1;
135
136 struct page *sc_page;
137 size_t sc_page_off;
138
139 /* original handlers for the sockets */
140 void (*sc_state_change)(struct sock *sk);
141 void (*sc_data_ready)(struct sock *sk, int bytes);
142
143 struct timeval sc_tv_timer;
144 struct timeval sc_tv_data_ready;
145 struct timeval sc_tv_advance_start;
146 struct timeval sc_tv_advance_stop;
147 struct timeval sc_tv_func_start;
148 struct timeval sc_tv_func_stop;
149 u32 sc_msg_key;
150 u16 sc_msg_type;
151};
152
153struct o2net_msg_handler {
154 struct rb_node nh_node;
155 u32 nh_max_len;
156 u32 nh_msg_type;
157 u32 nh_key;
158 o2net_msg_handler_func *nh_func;
159 o2net_msg_handler_func *nh_func_data;
160 struct kref nh_kref;
161 struct list_head nh_unregister_item;
162};
163
164enum o2net_system_error {
165 O2NET_ERR_NONE = 0,
166 O2NET_ERR_NO_HNDLR,
167 O2NET_ERR_OVERFLOW,
168 O2NET_ERR_DIED,
169 O2NET_ERR_MAX
170};
171
172struct o2net_status_wait {
173 enum o2net_system_error ns_sys_status;
174 s32 ns_status;
175 int ns_id;
176 wait_queue_head_t ns_wq;
177 struct list_head ns_node_item;
178};
179
180#endif /* O2CLUSTER_TCP_INTERNAL_H */