1/* AFS volume 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/kernel.h>
13#include <linux/module.h>
14#include <linux/init.h>
15#include <linux/slab.h>
16#include <linux/fs.h>
17#include <linux/pagemap.h>
18#include <linux/sched.h>
19#include "internal.h"
20
21static const char *afs_voltypes[] = { "R/W", "R/O", "BAK" };
22
23/*
24 * lookup a volume by name
25 * - this can be one of the following:
26 *	"%[cell:]volume[.]"		R/W volume
27 *	"#[cell:]volume[.]"		R/O or R/W volume (rwparent=0),
28 *					 or R/W (rwparent=1) volume
29 *	"%[cell:]volume.readonly"	R/O volume
30 *	"#[cell:]volume.readonly"	R/O volume
31 *	"%[cell:]volume.backup"		Backup volume
32 *	"#[cell:]volume.backup"		Backup volume
33 *
34 * The cell name is optional, and defaults to the current cell.
35 *
36 * See "The Rules of Mount Point Traversal" in Chapter 5 of the AFS SysAdmin
37 * Guide
38 * - Rule 1: Explicit type suffix forces access of that type or nothing
39 *           (no suffix, then use Rule 2 & 3)
40 * - Rule 2: If parent volume is R/O, then mount R/O volume by preference, R/W
41 *           if not available
42 * - Rule 3: If parent volume is R/W, then only mount R/W volume unless
43 *           explicitly told otherwise
44 */
45struct afs_volume *afs_volume_lookup(struct afs_mount_params *params)
46{
47	struct afs_vlocation *vlocation = NULL;
48	struct afs_volume *volume = NULL;
49	struct afs_server *server = NULL;
50	char srvtmask;
51	int ret, loop;
52
53	_enter("{%*.*s,%d}",
54	       params->volnamesz, params->volnamesz, params->volname, params->rwpath);
55
56	/* lookup the volume location record */
57	vlocation = afs_vlocation_lookup(params->cell, params->key,
58					 params->volname, params->volnamesz);
59	if (IS_ERR(vlocation)) {
60		ret = PTR_ERR(vlocation);
61		vlocation = NULL;
62		goto error;
63	}
64
65	/* make the final decision on the type we want */
66	ret = -ENOMEDIUM;
67	if (params->force && !(vlocation->vldb.vidmask & (1 << params->type)))
68		goto error;
69
70	srvtmask = 0;
71	for (loop = 0; loop < vlocation->vldb.nservers; loop++)
72		srvtmask |= vlocation->vldb.srvtmask[loop];
73
74	if (params->force) {
75		if (!(srvtmask & (1 << params->type)))
76			goto error;
77	} else if (srvtmask & AFS_VOL_VTM_RO) {
78		params->type = AFSVL_ROVOL;
79	} else if (srvtmask & AFS_VOL_VTM_RW) {
80		params->type = AFSVL_RWVOL;
81	} else {
82		goto error;
83	}
84
85	down_write(&params->cell->vl_sem);
86
87	/* is the volume already active? */
88	if (vlocation->vols[params->type]) {
89		/* yes - re-use it */
90		volume = vlocation->vols[params->type];
91		afs_get_volume(volume);
92		goto success;
93	}
94
95	/* create a new volume record */
96	_debug("creating new volume record");
97
98	ret = -ENOMEM;
99	volume = kzalloc(sizeof(struct afs_volume), GFP_KERNEL);
100	if (!volume)
101		goto error_up;
102
103	atomic_set(&volume->usage, 1);
104	volume->type		= params->type;
105	volume->type_force	= params->force;
106	volume->cell		= params->cell;
107	volume->vid		= vlocation->vldb.vid[params->type];
108
109	init_rwsem(&volume->server_sem);
110
111	/* look up all the applicable server records */
112	for (loop = 0; loop < 8; loop++) {
113		if (vlocation->vldb.srvtmask[loop] & (1 << volume->type)) {
114			server = afs_lookup_server(
115			       volume->cell, &vlocation->vldb.servers[loop]);
116			if (IS_ERR(server)) {
117				ret = PTR_ERR(server);
118				goto error_discard;
119			}
120
121			volume->servers[volume->nservers] = server;
122			volume->nservers++;
123		}
124	}
125
126	/* attach the cache and volume location */
127#ifdef AFS_CACHING_SUPPORT
128	cachefs_acquire_cookie(vlocation->cache,
129			       &afs_vnode_cache_index_def,
130			       volume,
131			       &volume->cache);
132#endif
133
134	afs_get_vlocation(vlocation);
135	volume->vlocation = vlocation;
136
137	vlocation->vols[volume->type] = volume;
138
139success:
140	_debug("kAFS selected %s volume %08x",
141	       afs_voltypes[volume->type], volume->vid);
142	up_write(&params->cell->vl_sem);
143	afs_put_vlocation(vlocation);
144	_leave(" = %p", volume);
145	return volume;
146
147	/* clean up */
148error_up:
149	up_write(&params->cell->vl_sem);
150error:
151	afs_put_vlocation(vlocation);
152	_leave(" = %d", ret);
153	return ERR_PTR(ret);
154
155error_discard:
156	up_write(&params->cell->vl_sem);
157
158	for (loop = volume->nservers - 1; loop >= 0; loop--)
159		afs_put_server(volume->servers[loop]);
160
161	kfree(volume);
162	goto error;
163}
164
165/*
166 * destroy a volume record
167 */
168void afs_put_volume(struct afs_volume *volume)
169{
170	struct afs_vlocation *vlocation;
171	int loop;
172
173	if (!volume)
174		return;
175
176	_enter("%p", volume);
177
178	ASSERTCMP(atomic_read(&volume->usage), >, 0);
179
180	vlocation = volume->vlocation;
181
182	/* to prevent a race, the decrement and the dequeue must be effectively
183	 * atomic */
184	down_write(&vlocation->cell->vl_sem);
185
186	if (likely(!atomic_dec_and_test(&volume->usage))) {
187		up_write(&vlocation->cell->vl_sem);
188		_leave("");
189		return;
190	}
191
192	vlocation->vols[volume->type] = NULL;
193
194	up_write(&vlocation->cell->vl_sem);
195
196	/* finish cleaning up the volume */
197#ifdef AFS_CACHING_SUPPORT
198	cachefs_relinquish_cookie(volume->cache, 0);
199#endif
200	afs_put_vlocation(vlocation);
201
202	for (loop = volume->nservers - 1; loop >= 0; loop--)
203		afs_put_server(volume->servers[loop]);
204
205	kfree(volume);
206
207	_leave(" [destroyed]");
208}
209
210/*
211 * pick a server to use to try accessing this volume
212 * - returns with an elevated usage count on the server chosen
213 */
214struct afs_server *afs_volume_pick_fileserver(struct afs_vnode *vnode)
215{
216	struct afs_volume *volume = vnode->volume;
217	struct afs_server *server;
218	int ret, state, loop;
219
220	_enter("%s", volume->vlocation->vldb.name);
221
222	/* stick with the server we're already using if we can */
223	if (vnode->server && vnode->server->fs_state == 0) {
224		afs_get_server(vnode->server);
225		_leave(" = %p [current]", vnode->server);
226		return vnode->server;
227	}
228
229	down_read(&volume->server_sem);
230
231	/* handle the no-server case */
232	if (volume->nservers == 0) {
233		ret = volume->rjservers ? -ENOMEDIUM : -ESTALE;
234		up_read(&volume->server_sem);
235		_leave(" = %d [no servers]", ret);
236		return ERR_PTR(ret);
237	}
238
239	/* basically, just search the list for the first live server and use
240	 * that */
241	ret = 0;
242	for (loop = 0; loop < volume->nservers; loop++) {
243		server = volume->servers[loop];
244		state = server->fs_state;
245
246		_debug("consider %d [%d]", loop, state);
247
248		switch (state) {
249			/* found an apparently healthy server */
250		case 0:
251			afs_get_server(server);
252			up_read(&volume->server_sem);
253			_leave(" = %p (picked %08x)",
254			       server, ntohl(server->addr.s_addr));
255			return server;
256
257		case -ENETUNREACH:
258			if (ret == 0)
259				ret = state;
260			break;
261
262		case -EHOSTUNREACH:
263			if (ret == 0 ||
264			    ret == -ENETUNREACH)
265				ret = state;
266			break;
267
268		case -ECONNREFUSED:
269			if (ret == 0 ||
270			    ret == -ENETUNREACH ||
271			    ret == -EHOSTUNREACH)
272				ret = state;
273			break;
274
275		default:
276		case -EREMOTEIO:
277			if (ret == 0 ||
278			    ret == -ENETUNREACH ||
279			    ret == -EHOSTUNREACH ||
280			    ret == -ECONNREFUSED)
281				ret = state;
282			break;
283		}
284	}
285
286	/* no available servers
287	 * - TODO: handle the no active servers case better
288	 */
289	up_read(&volume->server_sem);
290	_leave(" = %d", ret);
291	return ERR_PTR(ret);
292}
293
294/*
295 * release a server after use
296 * - releases the ref on the server struct that was acquired by picking
297 * - records result of using a particular server to access a volume
298 * - return 0 to try again, 1 if okay or to issue error
299 * - the caller must release the server struct if result was 0
300 */
301int afs_volume_release_fileserver(struct afs_vnode *vnode,
302				  struct afs_server *server,
303				  int result)
304{
305	struct afs_volume *volume = vnode->volume;
306	unsigned loop;
307
308	_enter("%s,%08x,%d",
309	       volume->vlocation->vldb.name, ntohl(server->addr.s_addr),
310	       result);
311
312	switch (result) {
313		/* success */
314	case 0:
315		server->fs_act_jif = jiffies;
316		server->fs_state = 0;
317		_leave("");
318		return 1;
319
320		/* the fileserver denied all knowledge of the volume */
321	case -ENOMEDIUM:
322		server->fs_act_jif = jiffies;
323		down_write(&volume->server_sem);
324
325		/* firstly, find where the server is in the active list (if it
326		 * is) */
327		for (loop = 0; loop < volume->nservers; loop++)
328			if (volume->servers[loop] == server)
329				goto present;
330
331		/* no longer there - may have been discarded by another op */
332		goto try_next_server_upw;
333
334	present:
335		volume->nservers--;
336		memmove(&volume->servers[loop],
337			&volume->servers[loop + 1],
338			sizeof(volume->servers[loop]) *
339			(volume->nservers - loop));
340		volume->servers[volume->nservers] = NULL;
341		afs_put_server(server);
342		volume->rjservers++;
343
344		if (volume->nservers > 0)
345			/* another server might acknowledge its existence */
346			goto try_next_server_upw;
347
348		/* handle the case where all the fileservers have rejected the
349		 * volume
350		 * - TODO: try asking the fileservers for volume information
351		 * - TODO: contact the VL server again to see if the volume is
352		 *         no longer registered
353		 */
354		up_write(&volume->server_sem);
355		afs_put_server(server);
356		_leave(" [completely rejected]");
357		return 1;
358
359		/* problem reaching the server */
360	case -ENETUNREACH:
361	case -EHOSTUNREACH:
362	case -ECONNREFUSED:
363	case -ETIME:
364	case -ETIMEDOUT:
365	case -EREMOTEIO:
366		/* mark the server as dead
367		 * TODO: vary dead timeout depending on error
368		 */
369		spin_lock(&server->fs_lock);
370		if (!server->fs_state) {
371			server->fs_dead_jif = jiffies + HZ * 10;
372			server->fs_state = result;
373			printk("kAFS: SERVER DEAD state=%d\n", result);
374		}
375		spin_unlock(&server->fs_lock);
376		goto try_next_server;
377
378		/* miscellaneous error */
379	default:
380		server->fs_act_jif = jiffies;
381	case -ENOMEM:
382	case -ENONET:
383		/* tell the caller to accept the result */
384		afs_put_server(server);
385		_leave(" [local failure]");
386		return 1;
387	}
388
389	/* tell the caller to loop around and try the next server */
390try_next_server_upw:
391	up_write(&volume->server_sem);
392try_next_server:
393	afs_put_server(server);
394	_leave(" [try next server]");
395	return 0;
396}
397