blob: 0ff4b86476e39288313fe3f92bd7ed1aaaedd3d3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/* volume.c: AFS volume management
2 *
3 * Copyright (C) 2002 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 "volume.h"
19#include "vnode.h"
20#include "cell.h"
21#include "cache.h"
22#include "cmservice.h"
23#include "fsclient.h"
24#include "vlclient.h"
25#include "internal.h"
26
27#ifdef __KDEBUG
28static const char *afs_voltypes[] = { "R/W", "R/O", "BAK" };
29#endif
30
31#ifdef AFS_CACHING_SUPPORT
32static cachefs_match_val_t afs_volume_cache_match(void *target,
33 const void *entry);
34static void afs_volume_cache_update(void *source, void *entry);
35
36struct cachefs_index_def afs_volume_cache_index_def = {
37 .name = "volume",
38 .data_size = sizeof(struct afs_cache_vhash),
39 .keys[0] = { CACHEFS_INDEX_KEYS_BIN, 1 },
40 .keys[1] = { CACHEFS_INDEX_KEYS_BIN, 1 },
41 .match = afs_volume_cache_match,
42 .update = afs_volume_cache_update,
43};
44#endif
45
46/*****************************************************************************/
47/*
48 * lookup a volume by name
49 * - this can be one of the following:
50 * "%[cell:]volume[.]" R/W volume
51 * "#[cell:]volume[.]" R/O or R/W volume (rwparent=0),
52 * or R/W (rwparent=1) volume
53 * "%[cell:]volume.readonly" R/O volume
54 * "#[cell:]volume.readonly" R/O volume
55 * "%[cell:]volume.backup" Backup volume
56 * "#[cell:]volume.backup" Backup volume
57 *
58 * The cell name is optional, and defaults to the current cell.
59 *
60 * See "The Rules of Mount Point Traversal" in Chapter 5 of the AFS SysAdmin
61 * Guide
62 * - Rule 1: Explicit type suffix forces access of that type or nothing
63 * (no suffix, then use Rule 2 & 3)
64 * - Rule 2: If parent volume is R/O, then mount R/O volume by preference, R/W
65 * if not available
66 * - Rule 3: If parent volume is R/W, then only mount R/W volume unless
67 * explicitly told otherwise
68 */
69int afs_volume_lookup(const char *name, struct afs_cell *cell, int rwpath,
70 struct afs_volume **_volume)
71{
72 struct afs_vlocation *vlocation = NULL;
73 struct afs_volume *volume = NULL;
74 afs_voltype_t type;
75 const char *cellname, *volname, *suffix;
76 char srvtmask;
77 int force, ret, loop, cellnamesz, volnamesz;
78
79 _enter("%s,,%d,", name, rwpath);
80
81 if (!name || (name[0] != '%' && name[0] != '#') || !name[1]) {
82 printk("kAFS: unparsable volume name\n");
83 return -EINVAL;
84 }
85
86 /* determine the type of volume we're looking for */
87 force = 0;
88 type = AFSVL_ROVOL;
89
90 if (rwpath || name[0] == '%') {
91 type = AFSVL_RWVOL;
92 force = 1;
93 }
94
95 suffix = strrchr(name, '.');
96 if (suffix) {
97 if (strcmp(suffix, ".readonly") == 0) {
98 type = AFSVL_ROVOL;
99 force = 1;
100 }
101 else if (strcmp(suffix, ".backup") == 0) {
102 type = AFSVL_BACKVOL;
103 force = 1;
104 }
105 else if (suffix[1] == 0) {
106 }
107 else {
108 suffix = NULL;
109 }
110 }
111
112 /* split the cell and volume names */
113 name++;
114 volname = strchr(name, ':');
115 if (volname) {
116 cellname = name;
117 cellnamesz = volname - name;
118 volname++;
119 }
120 else {
121 volname = name;
122 cellname = NULL;
123 cellnamesz = 0;
124 }
125
126 volnamesz = suffix ? suffix - volname : strlen(volname);
127
128 _debug("CELL:%*.*s [%p] VOLUME:%*.*s SUFFIX:%s TYPE:%d%s",
129 cellnamesz, cellnamesz, cellname ?: "", cell,
130 volnamesz, volnamesz, volname, suffix ?: "-",
131 type,
132 force ? " FORCE" : "");
133
134 /* lookup the cell record */
135 if (cellname || !cell) {
136 ret = afs_cell_lookup(cellname, cellnamesz, &cell);
137 if (ret<0) {
138 printk("kAFS: unable to lookup cell '%s'\n",
139 cellname ?: "");
140 goto error;
141 }
142 }
143 else {
144 afs_get_cell(cell);
145 }
146
147 /* lookup the volume location record */
148 ret = afs_vlocation_lookup(cell, volname, volnamesz, &vlocation);
149 if (ret < 0)
150 goto error;
151
152 /* make the final decision on the type we want */
153 ret = -ENOMEDIUM;
154 if (force && !(vlocation->vldb.vidmask & (1 << type)))
155 goto error;
156
157 srvtmask = 0;
158 for (loop = 0; loop < vlocation->vldb.nservers; loop++)
159 srvtmask |= vlocation->vldb.srvtmask[loop];
160
161 if (force) {
162 if (!(srvtmask & (1 << type)))
163 goto error;
164 }
165 else if (srvtmask & AFS_VOL_VTM_RO) {
166 type = AFSVL_ROVOL;
167 }
168 else if (srvtmask & AFS_VOL_VTM_RW) {
169 type = AFSVL_RWVOL;
170 }
171 else {
172 goto error;
173 }
174
175 down_write(&cell->vl_sem);
176
177 /* is the volume already active? */
178 if (vlocation->vols[type]) {
179 /* yes - re-use it */
180 volume = vlocation->vols[type];
181 afs_get_volume(volume);
182 goto success;
183 }
184
185 /* create a new volume record */
186 _debug("creating new volume record");
187
188 ret = -ENOMEM;
189 volume = kmalloc(sizeof(struct afs_volume), GFP_KERNEL);
190 if (!volume)
191 goto error_up;
192
193 memset(volume, 0, sizeof(struct afs_volume));
194 atomic_set(&volume->usage, 1);
195 volume->type = type;
196 volume->type_force = force;
197 volume->cell = cell;
198 volume->vid = vlocation->vldb.vid[type];
199
200 init_rwsem(&volume->server_sem);
201
202 /* look up all the applicable server records */
203 for (loop = 0; loop < 8; loop++) {
204 if (vlocation->vldb.srvtmask[loop] & (1 << volume->type)) {
205 ret = afs_server_lookup(
206 volume->cell,
207 &vlocation->vldb.servers[loop],
208 &volume->servers[volume->nservers]);
209 if (ret < 0)
210 goto error_discard;
211
212 volume->nservers++;
213 }
214 }
215
216 /* attach the cache and volume location */
217#ifdef AFS_CACHING_SUPPORT
218 cachefs_acquire_cookie(vlocation->cache,
219 &afs_vnode_cache_index_def,
220 volume,
221 &volume->cache);
222#endif
223
224 afs_get_vlocation(vlocation);
225 volume->vlocation = vlocation;
226
227 vlocation->vols[type] = volume;
228
229 success:
230 _debug("kAFS selected %s volume %08x",
231 afs_voltypes[volume->type], volume->vid);
232 *_volume = volume;
233 ret = 0;
234
235 /* clean up */
236 error_up:
237 up_write(&cell->vl_sem);
238 error:
239 afs_put_vlocation(vlocation);
240 afs_put_cell(cell);
241
242 _leave(" = %d (%p)", ret, volume);
243 return ret;
244
245 error_discard:
246 up_write(&cell->vl_sem);
247
248 for (loop = volume->nservers - 1; loop >= 0; loop--)
249 afs_put_server(volume->servers[loop]);
250
251 kfree(volume);
252 goto error;
253} /* end afs_volume_lookup() */
254
255/*****************************************************************************/
256/*
257 * destroy a volume record
258 */
259void afs_put_volume(struct afs_volume *volume)
260{
261 struct afs_vlocation *vlocation;
262 int loop;
263
264 if (!volume)
265 return;
266
267 _enter("%p", volume);
268
269 vlocation = volume->vlocation;
270
271 /* sanity check */
272 BUG_ON(atomic_read(&volume->usage) <= 0);
273
274 /* to prevent a race, the decrement and the dequeue must be effectively
275 * atomic */
276 down_write(&vlocation->cell->vl_sem);
277
278 if (likely(!atomic_dec_and_test(&volume->usage))) {
279 up_write(&vlocation->cell->vl_sem);
280 _leave("");
281 return;
282 }
283
284 vlocation->vols[volume->type] = NULL;
285
286 up_write(&vlocation->cell->vl_sem);
287
288 /* finish cleaning up the volume */
289#ifdef AFS_CACHING_SUPPORT
290 cachefs_relinquish_cookie(volume->cache, 0);
291#endif
292 afs_put_vlocation(vlocation);
293
294 for (loop = volume->nservers - 1; loop >= 0; loop--)
295 afs_put_server(volume->servers[loop]);
296
297 kfree(volume);
298
299 _leave(" [destroyed]");
300} /* end afs_put_volume() */
301
302/*****************************************************************************/
303/*
304 * pick a server to use to try accessing this volume
305 * - returns with an elevated usage count on the server chosen
306 */
307int afs_volume_pick_fileserver(struct afs_volume *volume,
308 struct afs_server **_server)
309{
310 struct afs_server *server;
311 int ret, state, loop;
312
313 _enter("%s", volume->vlocation->vldb.name);
314
315 down_read(&volume->server_sem);
316
317 /* handle the no-server case */
318 if (volume->nservers == 0) {
319 ret = volume->rjservers ? -ENOMEDIUM : -ESTALE;
320 up_read(&volume->server_sem);
321 _leave(" = %d [no servers]", ret);
322 return ret;
323 }
324
325 /* basically, just search the list for the first live server and use
326 * that */
327 ret = 0;
328 for (loop = 0; loop < volume->nservers; loop++) {
329 server = volume->servers[loop];
330 state = server->fs_state;
331
332 switch (state) {
333 /* found an apparently healthy server */
334 case 0:
335 afs_get_server(server);
336 up_read(&volume->server_sem);
337 *_server = server;
338 _leave(" = 0 (picked %08x)",
339 ntohl(server->addr.s_addr));
340 return 0;
341
342 case -ENETUNREACH:
343 if (ret == 0)
344 ret = state;
345 break;
346
347 case -EHOSTUNREACH:
348 if (ret == 0 ||
349 ret == -ENETUNREACH)
350 ret = state;
351 break;
352
353 case -ECONNREFUSED:
354 if (ret == 0 ||
355 ret == -ENETUNREACH ||
356 ret == -EHOSTUNREACH)
357 ret = state;
358 break;
359
360 default:
361 case -EREMOTEIO:
362 if (ret == 0 ||
363 ret == -ENETUNREACH ||
364 ret == -EHOSTUNREACH ||
365 ret == -ECONNREFUSED)
366 ret = state;
367 break;
368 }
369 }
370
371 /* no available servers
372 * - TODO: handle the no active servers case better
373 */
374 up_read(&volume->server_sem);
375 _leave(" = %d", ret);
376 return ret;
377} /* end afs_volume_pick_fileserver() */
378
379/*****************************************************************************/
380/*
381 * release a server after use
382 * - releases the ref on the server struct that was acquired by picking
383 * - records result of using a particular server to access a volume
384 * - return 0 to try again, 1 if okay or to issue error
385 */
386int afs_volume_release_fileserver(struct afs_volume *volume,
387 struct afs_server *server,
388 int result)
389{
390 unsigned loop;
391
392 _enter("%s,%08x,%d",
393 volume->vlocation->vldb.name, ntohl(server->addr.s_addr),
394 result);
395
396 switch (result) {
397 /* success */
398 case 0:
399 server->fs_act_jif = jiffies;
400 break;
401
402 /* the fileserver denied all knowledge of the volume */
403 case -ENOMEDIUM:
404 server->fs_act_jif = jiffies;
405 down_write(&volume->server_sem);
406
407 /* first, find where the server is in the active list (if it
408 * is) */
409 for (loop = 0; loop < volume->nservers; loop++)
410 if (volume->servers[loop] == server)
411 goto present;
412
413 /* no longer there - may have been discarded by another op */
414 goto try_next_server_upw;
415
416 present:
417 volume->nservers--;
418 memmove(&volume->servers[loop],
419 &volume->servers[loop + 1],
420 sizeof(volume->servers[loop]) *
421 (volume->nservers - loop));
422 volume->servers[volume->nservers] = NULL;
423 afs_put_server(server);
424 volume->rjservers++;
425
426 if (volume->nservers > 0)
427 /* another server might acknowledge its existence */
428 goto try_next_server_upw;
429
430 /* handle the case where all the fileservers have rejected the
431 * volume
432 * - TODO: try asking the fileservers for volume information
433 * - TODO: contact the VL server again to see if the volume is
434 * no longer registered
435 */
436 up_write(&volume->server_sem);
437 afs_put_server(server);
438 _leave(" [completely rejected]");
439 return 1;
440
441 /* problem reaching the server */
442 case -ENETUNREACH:
443 case -EHOSTUNREACH:
444 case -ECONNREFUSED:
445 case -ETIMEDOUT:
446 case -EREMOTEIO:
447 /* mark the server as dead
448 * TODO: vary dead timeout depending on error
449 */
450 spin_lock(&server->fs_lock);
451 if (!server->fs_state) {
452 server->fs_dead_jif = jiffies + HZ * 10;
453 server->fs_state = result;
454 printk("kAFS: SERVER DEAD state=%d\n", result);
455 }
456 spin_unlock(&server->fs_lock);
457 goto try_next_server;
458
459 /* miscellaneous error */
460 default:
461 server->fs_act_jif = jiffies;
462 case -ENOMEM:
463 case -ENONET:
464 break;
465 }
466
467 /* tell the caller to accept the result */
468 afs_put_server(server);
469 _leave("");
470 return 1;
471
472 /* tell the caller to loop around and try the next server */
473 try_next_server_upw:
474 up_write(&volume->server_sem);
475 try_next_server:
476 afs_put_server(server);
477 _leave(" [try next server]");
478 return 0;
479
480} /* end afs_volume_release_fileserver() */
481
482/*****************************************************************************/
483/*
484 * match a volume hash record stored in the cache
485 */
486#ifdef AFS_CACHING_SUPPORT
487static cachefs_match_val_t afs_volume_cache_match(void *target,
488 const void *entry)
489{
490 const struct afs_cache_vhash *vhash = entry;
491 struct afs_volume *volume = target;
492
493 _enter("{%u},{%u}", volume->type, vhash->vtype);
494
495 if (volume->type == vhash->vtype) {
496 _leave(" = SUCCESS");
497 return CACHEFS_MATCH_SUCCESS;
498 }
499
500 _leave(" = FAILED");
501 return CACHEFS_MATCH_FAILED;
502} /* end afs_volume_cache_match() */
503#endif
504
505/*****************************************************************************/
506/*
507 * update a volume hash record stored in the cache
508 */
509#ifdef AFS_CACHING_SUPPORT
510static void afs_volume_cache_update(void *source, void *entry)
511{
512 struct afs_cache_vhash *vhash = entry;
513 struct afs_volume *volume = source;
514
515 _enter("");
516
517 vhash->vtype = volume->type;
518
519} /* end afs_volume_cache_update() */
520#endif