blob: 601c99bd26114874d42cb10ae6f1524ee5d494f2 [file] [log] [blame]
Thomas Gleixner328970d2019-05-24 12:04:05 +02001/* SPDX-License-Identifier: GPL-2.0-or-later */
Masahiro Yamadafa60ce22021-05-06 18:06:44 -07002/*
Zach Brown982114892005-12-15 14:31:23 -08003 * Copyright (C) 2005 Oracle. All rights reserved.
Zach Brown982114892005-12-15 14:31:23 -08004 */
5
6#ifndef O2CLUSTER_TCP_INTERNAL_H
7#define O2CLUSTER_TCP_INTERNAL_H
8
9#define O2NET_MSG_MAGIC ((u16)0xfa55)
10#define O2NET_MSG_STATUS_MAGIC ((u16)0xfa56)
11#define O2NET_MSG_KEEP_REQ_MAGIC ((u16)0xfa57)
12#define O2NET_MSG_KEEP_RESP_MAGIC ((u16)0xfa58)
13
Zach Brown982114892005-12-15 14:31:23 -080014/* we're delaying our quorum decision so that heartbeat will have timed
15 * out truly dead nodes by the time we come around to making decisions
16 * on their number */
17#define O2NET_QUORUM_DELAY_MS ((o2hb_dead_threshold + 2) * O2HB_REGION_TIMEOUT_MS)
18
Sunil Mushran2bd63212010-01-25 16:57:38 -080019/*
Zach Brown982114892005-12-15 14:31:23 -080020 * This version number represents quite a lot, unfortunately. It not
21 * only represents the raw network message protocol on the wire but also
Sunil Mushran2bd63212010-01-25 16:57:38 -080022 * locking semantics of the file system using the protocol. It should
Zach Brown982114892005-12-15 14:31:23 -080023 * be somewhere else, I'm sure, but right now it isn't.
24 *
Joel Beckerd24fbcd2008-01-25 17:02:21 -080025 * With version 11, we separate out the filesystem locking portion. The
26 * filesystem now has a major.minor version it negotiates. Version 11
27 * introduces this negotiation to the o2dlm protocol, and as such the
28 * version here in tcp_internal.h should not need to be bumped for
29 * filesystem locking changes.
30 *
31 * New in version 11
32 * - Negotiation of filesystem locking in the dlm join.
33 *
Mark Fashehc934a922007-10-18 15:23:46 -070034 * New in version 10:
35 * - Meta/data locks combined
36 *
Mark Fasheh34d024f2007-09-24 15:56:19 -070037 * New in version 9:
38 * - All votes removed
39 *
Tiger Yang50008632007-03-20 16:01:38 -070040 * New in version 8:
41 * - Replace delete inode votes with a cluster lock
42 *
Srinivas Eeda1faf2892007-01-29 15:31:35 -080043 * New in version 7:
44 * - DLM join domain includes the live nodemap
45 *
Kurt Hackelba2bf212006-12-01 14:47:20 -080046 * New in version 6:
47 * - DLM lockres remote refcount fixes.
48 *
Andrew Beekhof828ae6a2006-12-04 14:04:55 +010049 * New in version 5:
50 * - Network timeout checking protocol
51 *
Mark Fasheh24c19ef2006-09-22 17:28:19 -070052 * New in version 4:
53 * - Remove i_generation from lock names for better stat performance.
54 *
Mark Fasheh379dfe92006-09-08 14:21:03 -070055 * New in version 3:
56 * - Replace dentry votes with a cluster lock
57 *
Zach Brown982114892005-12-15 14:31:23 -080058 * New in version 2:
59 * - full 64 bit i_size in the metadata lock lvbs
60 * - introduction of "rw" lock and pushing meta/data locking down
61 */
Junxiao Bi63b52c42016-09-19 14:44:44 -070062#define O2NET_PROTOCOL_VERSION 11ULL
Zach Brown982114892005-12-15 14:31:23 -080063struct o2net_handshake {
64 __be64 protocol_version;
65 __be64 connector_id;
Andrew Beekhof828ae6a2006-12-04 14:04:55 +010066 __be32 o2hb_heartbeat_timeout_ms;
67 __be32 o2net_idle_timeout_ms;
68 __be32 o2net_keepalive_delay_ms;
69 __be32 o2net_reconnect_delay_ms;
Zach Brown982114892005-12-15 14:31:23 -080070};
71
72struct o2net_node {
73 /* this is never called from int/bh */
74 spinlock_t nn_lock;
75
76 /* set the moment an sc is allocated and a connect is started */
77 struct o2net_sock_container *nn_sc;
78 /* _valid is only set after the handshake passes and tx can happen */
79 unsigned nn_sc_valid:1;
80 /* if this is set tx just returns it */
81 int nn_persistent_error;
Tao Ma5cc3bf22008-03-05 15:50:12 +080082 /* It is only set to 1 after the idle time out. */
83 atomic_t nn_timeout;
Zach Brown982114892005-12-15 14:31:23 -080084
85 /* threads waiting for an sc to arrive wait on the wq for generation
86 * to increase. it is increased when a connecting socket succeeds
87 * or fails or when an accepted socket is attached. */
88 wait_queue_head_t nn_sc_wq;
89
90 struct idr nn_status_idr;
91 struct list_head nn_status_list;
92
93 /* connects are attempted from when heartbeat comes up until either hb
Junxiao Bi41d62472015-02-10 14:08:48 -080094 * goes down, the node is unconfigured, or a connect succeeds.
95 * connect_work is queued from set_nn_state both from hb up and from
96 * itself if a connect attempt fails and so can be self-arming.
97 * shutdown is careful to first mark the nn such that no connects will
98 * be attempted before canceling delayed connect work and flushing the
99 * queue. */
David Howellsc4028952006-11-22 14:57:56 +0000100 struct delayed_work nn_connect_work;
Zach Brown982114892005-12-15 14:31:23 -0800101 unsigned long nn_last_connect_attempt;
102
103 /* this is queued as nodes come up and is canceled when a connection is
104 * established. this expiring gives up on the node and errors out
105 * transmits */
David Howellsc4028952006-11-22 14:57:56 +0000106 struct delayed_work nn_connect_expired;
Zach Brown982114892005-12-15 14:31:23 -0800107
108 /* after we give up on a socket we wait a while before deciding
109 * that it is still heartbeating and that we should do some
110 * quorum work */
David Howellsc4028952006-11-22 14:57:56 +0000111 struct delayed_work nn_still_up;
Zach Brown982114892005-12-15 14:31:23 -0800112};
113
114struct o2net_sock_container {
115 struct kref sc_kref;
Nikanth Karthikesan817f2c82010-09-20 11:44:00 +0530116 /* the next two are valid for the life time of the sc */
Zach Brown982114892005-12-15 14:31:23 -0800117 struct socket *sc_sock;
118 struct o2nm_node *sc_node;
119
120 /* all of these sc work structs hold refs on the sc while they are
121 * queued. they should not be able to ref a freed sc. the teardown
122 * race is with o2net_wq destruction in o2net_stop_listening() */
123
124 /* rx and connect work are generated from socket callbacks. sc
125 * shutdown removes the callbacks and then flushes the work queue */
126 struct work_struct sc_rx_work;
127 struct work_struct sc_connect_work;
128 /* shutdown work is triggered in two ways. the simple way is
129 * for a code path calls ensure_shutdown which gets a lock, removes
130 * the sc from the nn, and queues the work. in this case the
131 * work is single-shot. the work is also queued from a sock
132 * callback, though, and in this case the work will find the sc
133 * still on the nn and will call ensure_shutdown itself.. this
134 * ends up triggering the shutdown work again, though nothing
135 * will be done in that second iteration. so work queue teardown
136 * has to be careful to remove the sc from the nn before waiting
137 * on the work queue so that the shutdown work doesn't remove the
138 * sc and rearm itself.
139 */
140 struct work_struct sc_shutdown_work;
141
142 struct timer_list sc_idle_timeout;
David Howellsc4028952006-11-22 14:57:56 +0000143 struct delayed_work sc_keepalive_work;
Zach Brown982114892005-12-15 14:31:23 -0800144
145 unsigned sc_handshake_ok:1;
146
147 struct page *sc_page;
148 size_t sc_page_off;
149
150 /* original handlers for the sockets */
151 void (*sc_state_change)(struct sock *sk);
David S. Miller676d2362014-04-11 16:15:36 -0400152 void (*sc_data_ready)(struct sock *sk);
Sunil Mushranff1becb2010-12-22 12:39:39 -0800153
Zach Brown982114892005-12-15 14:31:23 -0800154 u32 sc_msg_key;
155 u16 sc_msg_type;
Zhen Wei925037b2007-01-23 17:19:59 -0800156
Sunil Mushranff1becb2010-12-22 12:39:39 -0800157#ifdef CONFIG_DEBUG_FS
158 struct list_head sc_net_debug_item;
159 ktime_t sc_tv_timer;
160 ktime_t sc_tv_data_ready;
161 ktime_t sc_tv_advance_start;
162 ktime_t sc_tv_advance_stop;
163 ktime_t sc_tv_func_start;
164 ktime_t sc_tv_func_stop;
165#endif
Sunil Mushran3c193b382010-12-22 12:39:40 -0800166#ifdef CONFIG_OCFS2_FS_STATS
167 ktime_t sc_tv_acquiry_total;
168 ktime_t sc_tv_send_total;
169 ktime_t sc_tv_status_total;
170 u32 sc_send_count;
Sunil Mushrane4530392010-12-22 12:39:41 -0800171 u32 sc_recv_count;
172 ktime_t sc_tv_process_total;
Sunil Mushran3c193b382010-12-22 12:39:40 -0800173#endif
Zhen Wei925037b2007-01-23 17:19:59 -0800174 struct mutex sc_send_lock;
Zach Brown982114892005-12-15 14:31:23 -0800175};
176
177struct o2net_msg_handler {
178 struct rb_node nh_node;
179 u32 nh_max_len;
180 u32 nh_msg_type;
181 u32 nh_key;
182 o2net_msg_handler_func *nh_func;
Changwei Gecfdce252018-01-31 16:14:25 -0800183 void *nh_func_data;
Kurt Hackeld74c9802007-01-17 17:04:25 -0800184 o2net_post_msg_handler_func
185 *nh_post_func;
Zach Brown982114892005-12-15 14:31:23 -0800186 struct kref nh_kref;
187 struct list_head nh_unregister_item;
188};
189
190enum o2net_system_error {
191 O2NET_ERR_NONE = 0,
192 O2NET_ERR_NO_HNDLR,
193 O2NET_ERR_OVERFLOW,
194 O2NET_ERR_DIED,
195 O2NET_ERR_MAX
196};
197
198struct o2net_status_wait {
199 enum o2net_system_error ns_sys_status;
200 s32 ns_status;
201 int ns_id;
202 wait_queue_head_t ns_wq;
203 struct list_head ns_node_item;
204};
205
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700206#ifdef CONFIG_DEBUG_FS
207/* just for state dumps */
208struct o2net_send_tracking {
209 struct list_head st_net_debug_item;
210 struct task_struct *st_task;
211 struct o2net_sock_container *st_sc;
212 u32 st_id;
213 u32 st_msg_type;
214 u32 st_msg_key;
215 u8 st_node;
Sunil Mushran3f9c14f2010-12-22 12:39:38 -0800216 ktime_t st_sock_time;
217 ktime_t st_send_time;
218 ktime_t st_status_time;
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700219};
220#else
221struct o2net_send_tracking {
222 u32 dummy;
223};
224#endif /* CONFIG_DEBUG_FS */
225
Zach Brown982114892005-12-15 14:31:23 -0800226#endif /* O2CLUSTER_TCP_INTERNAL_H */