1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
4 *
5 * Copyright (C) 2005 Oracle.  All rights reserved.
6 */
7
8#ifndef O2CLUSTER_TCP_INTERNAL_H
9#define O2CLUSTER_TCP_INTERNAL_H
10
11#define O2NET_MSG_MAGIC           ((u16)0xfa55)
12#define O2NET_MSG_STATUS_MAGIC    ((u16)0xfa56)
13#define O2NET_MSG_KEEP_REQ_MAGIC  ((u16)0xfa57)
14#define O2NET_MSG_KEEP_RESP_MAGIC ((u16)0xfa58)
15
16/* we're delaying our quorum decision so that heartbeat will have timed
17 * out truly dead nodes by the time we come around to making decisions
18 * on their number */
19#define O2NET_QUORUM_DELAY_MS	((o2hb_dead_threshold + 2) * O2HB_REGION_TIMEOUT_MS)
20
21/*
22 * This version number represents quite a lot, unfortunately.  It not
23 * only represents the raw network message protocol on the wire but also
24 * locking semantics of the file system using the protocol.  It should
25 * be somewhere else, I'm sure, but right now it isn't.
26 *
27 * With version 11, we separate out the filesystem locking portion.  The
28 * filesystem now has a major.minor version it negotiates.  Version 11
29 * introduces this negotiation to the o2dlm protocol, and as such the
30 * version here in tcp_internal.h should not need to be bumped for
31 * filesystem locking changes.
32 *
33 * New in version 11
34 * 	- Negotiation of filesystem locking in the dlm join.
35 *
36 * New in version 10:
37 * 	- Meta/data locks combined
38 *
39 * New in version 9:
40 * 	- All votes removed
41 *
42 * New in version 8:
43 * 	- Replace delete inode votes with a cluster lock
44 *
45 * New in version 7:
46 * 	- DLM join domain includes the live nodemap
47 *
48 * New in version 6:
49 * 	- DLM lockres remote refcount fixes.
50 *
51 * New in version 5:
52 * 	- Network timeout checking protocol
53 *
54 * New in version 4:
55 * 	- Remove i_generation from lock names for better stat performance.
56 *
57 * New in version 3:
58 * 	- Replace dentry votes with a cluster lock
59 *
60 * New in version 2:
61 * 	- full 64 bit i_size in the metadata lock lvbs
62 * 	- introduction of "rw" lock and pushing meta/data locking down
63 */
64#define O2NET_PROTOCOL_VERSION 11ULL
65struct o2net_handshake {
66	__be64	protocol_version;
67	__be64	connector_id;
68	__be32  o2hb_heartbeat_timeout_ms;
69	__be32  o2net_idle_timeout_ms;
70	__be32  o2net_keepalive_delay_ms;
71	__be32  o2net_reconnect_delay_ms;
72};
73
74struct o2net_node {
75	/* this is never called from int/bh */
76	spinlock_t			nn_lock;
77
78	/* set the moment an sc is allocated and a connect is started */
79	struct o2net_sock_container	*nn_sc;
80	/* _valid is only set after the handshake passes and tx can happen */
81	unsigned			nn_sc_valid:1;
82	/* if this is set tx just returns it */
83	int				nn_persistent_error;
84	/* It is only set to 1 after the idle time out. */
85	atomic_t			nn_timeout;
86
87	/* threads waiting for an sc to arrive wait on the wq for generation
88	 * to increase.  it is increased when a connecting socket succeeds
89	 * or fails or when an accepted socket is attached. */
90	wait_queue_head_t		nn_sc_wq;
91
92	struct idr			nn_status_idr;
93	struct list_head		nn_status_list;
94
95	/* connects are attempted from when heartbeat comes up until either hb
96	 * goes down, the node is unconfigured, or a connect succeeds.
97	 * connect_work is queued from set_nn_state both from hb up and from
98	 * itself if a connect attempt fails and so can be self-arming.
99	 * shutdown is careful to first mark the nn such that no connects will
100	 * be attempted before canceling delayed connect work and flushing the
101	 * queue. */
102	struct delayed_work		nn_connect_work;
103	unsigned long			nn_last_connect_attempt;
104
105	/* this is queued as nodes come up and is canceled when a connection is
106	 * established.  this expiring gives up on the node and errors out
107	 * transmits */
108	struct delayed_work		nn_connect_expired;
109
110	/* after we give up on a socket we wait a while before deciding
111	 * that it is still heartbeating and that we should do some
112	 * quorum work */
113	struct delayed_work		nn_still_up;
114};
115
116struct o2net_sock_container {
117	struct kref		sc_kref;
118	/* the next two are valid for the life time of the sc */
119	struct socket		*sc_sock;
120	struct o2nm_node	*sc_node;
121
122	/* all of these sc work structs hold refs on the sc while they are
123	 * queued.  they should not be able to ref a freed sc.  the teardown
124	 * race is with o2net_wq destruction in o2net_stop_listening() */
125
126	/* rx and connect work are generated from socket callbacks.  sc
127	 * shutdown removes the callbacks and then flushes the work queue */
128	struct work_struct	sc_rx_work;
129	struct work_struct	sc_connect_work;
130	/* shutdown work is triggered in two ways.  the simple way is
131	 * for a code path calls ensure_shutdown which gets a lock, removes
132	 * the sc from the nn, and queues the work.  in this case the
133	 * work is single-shot.  the work is also queued from a sock
134	 * callback, though, and in this case the work will find the sc
135	 * still on the nn and will call ensure_shutdown itself.. this
136	 * ends up triggering the shutdown work again, though nothing
137	 * will be done in that second iteration.  so work queue teardown
138	 * has to be careful to remove the sc from the nn before waiting
139	 * on the work queue so that the shutdown work doesn't remove the
140	 * sc and rearm itself.
141	 */
142	struct work_struct	sc_shutdown_work;
143
144	struct timer_list	sc_idle_timeout;
145	struct delayed_work	sc_keepalive_work;
146
147	unsigned		sc_handshake_ok:1;
148
149	struct page 		*sc_page;
150	size_t			sc_page_off;
151
152	/* original handlers for the sockets */
153	void			(*sc_state_change)(struct sock *sk);
154	void			(*sc_data_ready)(struct sock *sk);
155
156	u32			sc_msg_key;
157	u16			sc_msg_type;
158
159#ifdef CONFIG_DEBUG_FS
160	struct list_head        sc_net_debug_item;
161	ktime_t			sc_tv_timer;
162	ktime_t			sc_tv_data_ready;
163	ktime_t			sc_tv_advance_start;
164	ktime_t			sc_tv_advance_stop;
165	ktime_t			sc_tv_func_start;
166	ktime_t			sc_tv_func_stop;
167#endif
168#ifdef CONFIG_OCFS2_FS_STATS
169	ktime_t			sc_tv_acquiry_total;
170	ktime_t			sc_tv_send_total;
171	ktime_t			sc_tv_status_total;
172	u32			sc_send_count;
173	u32			sc_recv_count;
174	ktime_t			sc_tv_process_total;
175#endif
176	struct mutex		sc_send_lock;
177};
178
179struct o2net_msg_handler {
180	struct rb_node		nh_node;
181	u32			nh_max_len;
182	u32			nh_msg_type;
183	u32			nh_key;
184	o2net_msg_handler_func	*nh_func;
185	void			*nh_func_data;
186	o2net_post_msg_handler_func
187				*nh_post_func;
188	struct kref		nh_kref;
189	struct list_head	nh_unregister_item;
190};
191
192enum o2net_system_error {
193	O2NET_ERR_NONE = 0,
194	O2NET_ERR_NO_HNDLR,
195	O2NET_ERR_OVERFLOW,
196	O2NET_ERR_DIED,
197	O2NET_ERR_MAX
198};
199
200struct o2net_status_wait {
201	enum o2net_system_error	ns_sys_status;
202	s32			ns_status;
203	int			ns_id;
204	wait_queue_head_t	ns_wq;
205	struct list_head	ns_node_item;
206};
207
208#ifdef CONFIG_DEBUG_FS
209/* just for state dumps */
210struct o2net_send_tracking {
211	struct list_head		st_net_debug_item;
212	struct task_struct		*st_task;
213	struct o2net_sock_container	*st_sc;
214	u32				st_id;
215	u32				st_msg_type;
216	u32				st_msg_key;
217	u8				st_node;
218	ktime_t				st_sock_time;
219	ktime_t				st_send_time;
220	ktime_t				st_status_time;
221};
222#else
223struct o2net_send_tracking {
224	u32	dummy;
225};
226#endif	/* CONFIG_DEBUG_FS */
227
228#endif /* O2CLUSTER_TCP_INTERNAL_H */
229