1/* -*- 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/* 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
32 * on their number */
33#define O2NET_QUORUM_DELAY_MS	((o2hb_dead_threshold + 2) * O2HB_REGION_TIMEOUT_MS)
34
35/*
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.
40 *
41 * New in version 8:
42 * 	- Replace delete inode votes with a cluster lock
43 *
44 * New in version 7:
45 * 	- DLM join domain includes the live nodemap
46 *
47 * New in version 6:
48 * 	- DLM lockres remote refcount fixes.
49 *
50 * New in version 5:
51 * 	- Network timeout checking protocol
52 *
53 * New in version 4:
54 * 	- Remove i_generation from lock names for better stat performance.
55 *
56 * New in version 3:
57 * 	- Replace dentry votes with a cluster lock
58 *
59 * New in version 2:
60 * 	- full 64 bit i_size in the metadata lock lvbs
61 * 	- introduction of "rw" lock and pushing meta/data locking down
62 */
63#define O2NET_PROTOCOL_VERSION 8ULL
64struct o2net_handshake {
65	__be64	protocol_version;
66	__be64	connector_id;
67	__be32  o2hb_heartbeat_timeout_ms;
68	__be32  o2net_idle_timeout_ms;
69	__be32  o2net_keepalive_delay_ms;
70	__be32  o2net_reconnect_delay_ms;
71};
72
73struct o2net_node {
74	/* this is never called from int/bh */
75	spinlock_t			nn_lock;
76
77	/* set the moment an sc is allocated and a connect is started */
78	struct o2net_sock_container	*nn_sc;
79	/* _valid is only set after the handshake passes and tx can happen */
80	unsigned			nn_sc_valid:1;
81	/* if this is set tx just returns it */
82	int				nn_persistent_error;
83
84	/* threads waiting for an sc to arrive wait on the wq for generation
85	 * to increase.  it is increased when a connecting socket succeeds
86	 * or fails or when an accepted socket is attached. */
87	wait_queue_head_t		nn_sc_wq;
88
89	struct idr			nn_status_idr;
90	struct list_head		nn_status_list;
91
92	/* connects are attempted from when heartbeat comes up until either hb
93	 * goes down, the node is unconfigured, no connect attempts succeed
94	 * before O2NET_CONN_IDLE_DELAY, or a connect succeeds.  connect_work
95	 * is queued from set_nn_state both from hb up and from itself if a
96	 * connect attempt fails and so can be self-arming.  shutdown is
97	 * careful to first mark the nn such that no connects will be attempted
98	 * before canceling delayed connect work and flushing the queue. */
99	struct delayed_work		nn_connect_work;
100	unsigned long			nn_last_connect_attempt;
101
102	/* this is queued as nodes come up and is canceled when a connection is
103	 * established.  this expiring gives up on the node and errors out
104	 * transmits */
105	struct delayed_work		nn_connect_expired;
106
107	/* after we give up on a socket we wait a while before deciding
108	 * that it is still heartbeating and that we should do some
109	 * quorum work */
110	struct delayed_work		nn_still_up;
111};
112
113struct o2net_sock_container {
114	struct kref		sc_kref;
115	/* the next two are vaild for the life time of the sc */
116	struct socket		*sc_sock;
117	struct o2nm_node	*sc_node;
118
119	/* all of these sc work structs hold refs on the sc while they are
120	 * queued.  they should not be able to ref a freed sc.  the teardown
121	 * race is with o2net_wq destruction in o2net_stop_listening() */
122
123	/* rx and connect work are generated from socket callbacks.  sc
124	 * shutdown removes the callbacks and then flushes the work queue */
125	struct work_struct	sc_rx_work;
126	struct work_struct	sc_connect_work;
127	/* shutdown work is triggered in two ways.  the simple way is
128	 * for a code path calls ensure_shutdown which gets a lock, removes
129	 * the sc from the nn, and queues the work.  in this case the
130	 * work is single-shot.  the work is also queued from a sock
131	 * callback, though, and in this case the work will find the sc
132	 * still on the nn and will call ensure_shutdown itself.. this
133	 * ends up triggering the shutdown work again, though nothing
134	 * will be done in that second iteration.  so work queue teardown
135	 * has to be careful to remove the sc from the nn before waiting
136	 * on the work queue so that the shutdown work doesn't remove the
137	 * sc and rearm itself.
138	 */
139	struct work_struct	sc_shutdown_work;
140
141	struct timer_list	sc_idle_timeout;
142	struct delayed_work	sc_keepalive_work;
143
144	unsigned		sc_handshake_ok:1;
145
146	struct page 		*sc_page;
147	size_t			sc_page_off;
148
149	/* original handlers for the sockets */
150	void			(*sc_state_change)(struct sock *sk);
151	void			(*sc_data_ready)(struct sock *sk, int bytes);
152
153	struct timeval 		sc_tv_timer;
154	struct timeval 		sc_tv_data_ready;
155	struct timeval 		sc_tv_advance_start;
156	struct timeval 		sc_tv_advance_stop;
157	struct timeval 		sc_tv_func_start;
158	struct timeval 		sc_tv_func_stop;
159	u32			sc_msg_key;
160	u16			sc_msg_type;
161
162	struct mutex		sc_send_lock;
163};
164
165struct o2net_msg_handler {
166	struct rb_node		nh_node;
167	u32			nh_max_len;
168	u32			nh_msg_type;
169	u32			nh_key;
170	o2net_msg_handler_func	*nh_func;
171	o2net_msg_handler_func	*nh_func_data;
172	o2net_post_msg_handler_func
173				*nh_post_func;
174	struct kref		nh_kref;
175	struct list_head	nh_unregister_item;
176};
177
178enum o2net_system_error {
179	O2NET_ERR_NONE = 0,
180	O2NET_ERR_NO_HNDLR,
181	O2NET_ERR_OVERFLOW,
182	O2NET_ERR_DIED,
183	O2NET_ERR_MAX
184};
185
186struct o2net_status_wait {
187	enum o2net_system_error	ns_sys_status;
188	s32			ns_status;
189	int			ns_id;
190	wait_queue_head_t	ns_wq;
191	struct list_head	ns_node_item;
192};
193
194#endif /* O2CLUSTER_TCP_INTERNAL_H */
195