1/* AFS server record management
2 *
3 * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12#include <linux/sched.h>
13#include <linux/slab.h>
14#include "internal.h"
15
16static unsigned afs_server_timeout = 10;	/* server timeout in seconds */
17
18static void afs_reap_server(struct work_struct *);
19
20/* tree of all the servers, indexed by IP address */
21static struct rb_root afs_servers = RB_ROOT;
22static DEFINE_RWLOCK(afs_servers_lock);
23
24/* LRU list of all the servers not currently in use */
25static LIST_HEAD(afs_server_graveyard);
26static DEFINE_SPINLOCK(afs_server_graveyard_lock);
27static DECLARE_DELAYED_WORK(afs_server_reaper, afs_reap_server);
28
29/*
30 * install a server record in the master tree
31 */
32static int afs_install_server(struct afs_server *server)
33{
34	struct afs_server *xserver;
35	struct rb_node **pp, *p;
36	int ret;
37
38	_enter("%p", server);
39
40	write_lock(&afs_servers_lock);
41
42	ret = -EEXIST;
43	pp = &afs_servers.rb_node;
44	p = NULL;
45	while (*pp) {
46		p = *pp;
47		_debug("- consider %p", p);
48		xserver = rb_entry(p, struct afs_server, master_rb);
49		if (server->addr.s_addr < xserver->addr.s_addr)
50			pp = &(*pp)->rb_left;
51		else if (server->addr.s_addr > xserver->addr.s_addr)
52			pp = &(*pp)->rb_right;
53		else
54			goto error;
55	}
56
57	rb_link_node(&server->master_rb, p, pp);
58	rb_insert_color(&server->master_rb, &afs_servers);
59	ret = 0;
60
61error:
62	write_unlock(&afs_servers_lock);
63	return ret;
64}
65
66/*
67 * allocate a new server record
68 */
69static struct afs_server *afs_alloc_server(struct afs_cell *cell,
70					   const struct in_addr *addr)
71{
72	struct afs_server *server;
73
74	_enter("");
75
76	server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
77	if (server) {
78		atomic_set(&server->usage, 1);
79		server->cell = cell;
80
81		INIT_LIST_HEAD(&server->link);
82		INIT_LIST_HEAD(&server->grave);
83		init_rwsem(&server->sem);
84		spin_lock_init(&server->fs_lock);
85		server->fs_vnodes = RB_ROOT;
86		server->cb_promises = RB_ROOT;
87		spin_lock_init(&server->cb_lock);
88		init_waitqueue_head(&server->cb_break_waitq);
89		INIT_DELAYED_WORK(&server->cb_break_work,
90				  afs_dispatch_give_up_callbacks);
91
92		memcpy(&server->addr, addr, sizeof(struct in_addr));
93		server->addr.s_addr = addr->s_addr;
94		_leave(" = %p{%d}", server, atomic_read(&server->usage));
95	} else {
96		_leave(" = NULL [nomem]");
97	}
98	return server;
99}
100
101/*
102 * get an FS-server record for a cell
103 */
104struct afs_server *afs_lookup_server(struct afs_cell *cell,
105				     const struct in_addr *addr)
106{
107	struct afs_server *server, *candidate;
108
109	_enter("%p,%pI4", cell, &addr->s_addr);
110
111	/* quick scan of the list to see if we already have the server */
112	read_lock(&cell->servers_lock);
113
114	list_for_each_entry(server, &cell->servers, link) {
115		if (server->addr.s_addr == addr->s_addr)
116			goto found_server_quickly;
117	}
118	read_unlock(&cell->servers_lock);
119
120	candidate = afs_alloc_server(cell, addr);
121	if (!candidate) {
122		_leave(" = -ENOMEM");
123		return ERR_PTR(-ENOMEM);
124	}
125
126	write_lock(&cell->servers_lock);
127
128	/* check the cell's server list again */
129	list_for_each_entry(server, &cell->servers, link) {
130		if (server->addr.s_addr == addr->s_addr)
131			goto found_server;
132	}
133
134	_debug("new");
135	server = candidate;
136	if (afs_install_server(server) < 0)
137		goto server_in_two_cells;
138
139	afs_get_cell(cell);
140	list_add_tail(&server->link, &cell->servers);
141
142	write_unlock(&cell->servers_lock);
143	_leave(" = %p{%d}", server, atomic_read(&server->usage));
144	return server;
145
146	/* found a matching server quickly */
147found_server_quickly:
148	_debug("found quickly");
149	afs_get_server(server);
150	read_unlock(&cell->servers_lock);
151no_longer_unused:
152	if (!list_empty(&server->grave)) {
153		spin_lock(&afs_server_graveyard_lock);
154		list_del_init(&server->grave);
155		spin_unlock(&afs_server_graveyard_lock);
156	}
157	_leave(" = %p{%d}", server, atomic_read(&server->usage));
158	return server;
159
160	/* found a matching server on the second pass */
161found_server:
162	_debug("found");
163	afs_get_server(server);
164	write_unlock(&cell->servers_lock);
165	kfree(candidate);
166	goto no_longer_unused;
167
168	/* found a server that seems to be in two cells */
169server_in_two_cells:
170	write_unlock(&cell->servers_lock);
171	kfree(candidate);
172	printk(KERN_NOTICE "kAFS: Server %pI4 appears to be in two cells\n",
173	       addr);
174	_leave(" = -EEXIST");
175	return ERR_PTR(-EEXIST);
176}
177
178/*
179 * look up a server by its IP address
180 */
181struct afs_server *afs_find_server(const struct in_addr *_addr)
182{
183	struct afs_server *server = NULL;
184	struct rb_node *p;
185	struct in_addr addr = *_addr;
186
187	_enter("%pI4", &addr.s_addr);
188
189	read_lock(&afs_servers_lock);
190
191	p = afs_servers.rb_node;
192	while (p) {
193		server = rb_entry(p, struct afs_server, master_rb);
194
195		_debug("- consider %p", p);
196
197		if (addr.s_addr < server->addr.s_addr) {
198			p = p->rb_left;
199		} else if (addr.s_addr > server->addr.s_addr) {
200			p = p->rb_right;
201		} else {
202			afs_get_server(server);
203			goto found;
204		}
205	}
206
207	server = NULL;
208found:
209	read_unlock(&afs_servers_lock);
210	ASSERTIFCMP(server, server->addr.s_addr, ==, addr.s_addr);
211	_leave(" = %p", server);
212	return server;
213}
214
215/*
216 * destroy a server record
217 * - removes from the cell list
218 */
219void afs_put_server(struct afs_server *server)
220{
221	if (!server)
222		return;
223
224	_enter("%p{%d}", server, atomic_read(&server->usage));
225
226	_debug("PUT SERVER %d", atomic_read(&server->usage));
227
228	ASSERTCMP(atomic_read(&server->usage), >, 0);
229
230	if (likely(!atomic_dec_and_test(&server->usage))) {
231		_leave("");
232		return;
233	}
234
235	afs_flush_callback_breaks(server);
236
237	spin_lock(&afs_server_graveyard_lock);
238	if (atomic_read(&server->usage) == 0) {
239		list_move_tail(&server->grave, &afs_server_graveyard);
240		server->time_of_death = get_seconds();
241		queue_delayed_work(afs_wq, &afs_server_reaper,
242				   afs_server_timeout * HZ);
243	}
244	spin_unlock(&afs_server_graveyard_lock);
245	_leave(" [dead]");
246}
247
248/*
249 * destroy a dead server
250 */
251static void afs_destroy_server(struct afs_server *server)
252{
253	_enter("%p", server);
254
255	ASSERTIF(server->cb_break_head != server->cb_break_tail,
256		 delayed_work_pending(&server->cb_break_work));
257
258	ASSERTCMP(server->fs_vnodes.rb_node, ==, NULL);
259	ASSERTCMP(server->cb_promises.rb_node, ==, NULL);
260	ASSERTCMP(server->cb_break_head, ==, server->cb_break_tail);
261	ASSERTCMP(atomic_read(&server->cb_break_n), ==, 0);
262
263	afs_put_cell(server->cell);
264	kfree(server);
265}
266
267/*
268 * reap dead server records
269 */
270static void afs_reap_server(struct work_struct *work)
271{
272	LIST_HEAD(corpses);
273	struct afs_server *server;
274	unsigned long delay, expiry;
275	time_t now;
276
277	now = get_seconds();
278	spin_lock(&afs_server_graveyard_lock);
279
280	while (!list_empty(&afs_server_graveyard)) {
281		server = list_entry(afs_server_graveyard.next,
282				    struct afs_server, grave);
283
284		/* the queue is ordered most dead first */
285		expiry = server->time_of_death + afs_server_timeout;
286		if (expiry > now) {
287			delay = (expiry - now) * HZ;
288			mod_delayed_work(afs_wq, &afs_server_reaper, delay);
289			break;
290		}
291
292		write_lock(&server->cell->servers_lock);
293		write_lock(&afs_servers_lock);
294		if (atomic_read(&server->usage) > 0) {
295			list_del_init(&server->grave);
296		} else {
297			list_move_tail(&server->grave, &corpses);
298			list_del_init(&server->link);
299			rb_erase(&server->master_rb, &afs_servers);
300		}
301		write_unlock(&afs_servers_lock);
302		write_unlock(&server->cell->servers_lock);
303	}
304
305	spin_unlock(&afs_server_graveyard_lock);
306
307	/* now reap the corpses we've extracted */
308	while (!list_empty(&corpses)) {
309		server = list_entry(corpses.next, struct afs_server, grave);
310		list_del(&server->grave);
311		afs_destroy_server(server);
312	}
313}
314
315/*
316 * discard all the server records for rmmod
317 */
318void __exit afs_purge_servers(void)
319{
320	afs_server_timeout = 0;
321	mod_delayed_work(afs_wq, &afs_server_reaper, 0);
322}
323