Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* AFS volume management
3 *
4 * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 */
7
8#include <linux/kernel.h>
9#include <linux/slab.h>
10#include "internal.h"
11
12static unsigned __read_mostly afs_volume_record_life = 60 * 60;
13
14static void afs_destroy_volume(struct work_struct *work);
15
16/*
17 * Insert a volume into a cell. If there's an existing volume record, that is
18 * returned instead with a ref held.
19 */
20static struct afs_volume *afs_insert_volume_into_cell(struct afs_cell *cell,
21 struct afs_volume *volume)
22{
23 struct afs_volume *p;
24 struct rb_node *parent = NULL, **pp;
25
26 write_seqlock(&cell->volume_lock);
27
28 pp = &cell->volumes.rb_node;
29 while (*pp) {
30 parent = *pp;
31 p = rb_entry(parent, struct afs_volume, cell_node);
32 if (p->vid < volume->vid) {
33 pp = &(*pp)->rb_left;
34 } else if (p->vid > volume->vid) {
35 pp = &(*pp)->rb_right;
36 } else {
37 if (afs_try_get_volume(p, afs_volume_trace_get_cell_insert)) {
38 volume = p;
39 goto found;
40 }
41
42 set_bit(AFS_VOLUME_RM_TREE, &volume->flags);
43 rb_replace_node_rcu(&p->cell_node, &volume->cell_node, &cell->volumes);
44 }
45 }
46
47 rb_link_node_rcu(&volume->cell_node, parent, pp);
48 rb_insert_color(&volume->cell_node, &cell->volumes);
49 hlist_add_head_rcu(&volume->proc_link, &cell->proc_volumes);
50
51found:
52 write_sequnlock(&cell->volume_lock);
53 return volume;
54
55}
56
57static void afs_remove_volume_from_cell(struct afs_volume *volume)
58{
59 struct afs_cell *cell = volume->cell;
60
61 if (!hlist_unhashed(&volume->proc_link)) {
62 trace_afs_volume(volume->vid, refcount_read(&cell->ref),
63 afs_volume_trace_remove);
64 write_seqlock(&cell->volume_lock);
65 hlist_del_rcu(&volume->proc_link);
66 if (!test_and_set_bit(AFS_VOLUME_RM_TREE, &volume->flags))
67 rb_erase(&volume->cell_node, &cell->volumes);
68 write_sequnlock(&cell->volume_lock);
69 }
70}
71
72/*
73 * Allocate a volume record and load it up from a vldb record.
74 */
75static struct afs_volume *afs_alloc_volume(struct afs_fs_context *params,
76 struct afs_vldb_entry *vldb,
77 struct afs_server_list **_slist)
78{
79 struct afs_server_list *slist;
80 struct afs_volume *volume;
81 int ret = -ENOMEM, i;
82
83 volume = kzalloc(sizeof(struct afs_volume), GFP_KERNEL);
84 if (!volume)
85 goto error_0;
86
87 volume->vid = vldb->vid[params->type];
88 volume->update_at = ktime_get_real_seconds() + afs_volume_record_life;
89 volume->cell = afs_get_cell(params->cell, afs_cell_trace_get_vol);
90 volume->type = params->type;
91 volume->type_force = params->force;
92 volume->name_len = vldb->name_len;
93 volume->creation_time = TIME64_MIN;
94 volume->update_time = TIME64_MIN;
95
96 refcount_set(&volume->ref, 1);
97 INIT_HLIST_NODE(&volume->proc_link);
98 INIT_WORK(&volume->destructor, afs_destroy_volume);
99 rwlock_init(&volume->servers_lock);
100 mutex_init(&volume->volsync_lock);
101 mutex_init(&volume->cb_check_lock);
102 rwlock_init(&volume->cb_v_break_lock);
103 INIT_LIST_HEAD(&volume->open_mmaps);
104 init_rwsem(&volume->open_mmaps_lock);
105 memcpy(volume->name, vldb->name, vldb->name_len + 1);
106
107 for (i = 0; i < AFS_MAXTYPES; i++)
108 volume->vids[i] = vldb->vid[i];
109
110 slist = afs_alloc_server_list(volume, params->key, vldb);
111 if (IS_ERR(slist)) {
112 ret = PTR_ERR(slist);
113 goto error_1;
114 }
115
116 *_slist = slist;
117 rcu_assign_pointer(volume->servers, slist);
118 trace_afs_volume(volume->vid, 1, afs_volume_trace_alloc);
119 return volume;
120
121error_1:
122 afs_put_cell(volume->cell, afs_cell_trace_put_vol);
123 kfree(volume);
124error_0:
125 return ERR_PTR(ret);
126}
127
128/*
129 * Look up or allocate a volume record.
130 */
131static struct afs_volume *afs_lookup_volume(struct afs_fs_context *params,
132 struct afs_vldb_entry *vldb)
133{
134 struct afs_server_list *slist;
135 struct afs_volume *candidate, *volume;
136
137 candidate = afs_alloc_volume(params, vldb, &slist);
138 if (IS_ERR(candidate))
139 return candidate;
140
141 volume = afs_insert_volume_into_cell(params->cell, candidate);
142 if (volume == candidate)
143 afs_attach_volume_to_servers(volume, slist);
144 else
145 afs_put_volume(candidate, afs_volume_trace_put_cell_dup);
146 return volume;
147}
148
149/*
150 * Look up a VLDB record for a volume.
151 */
152static struct afs_vldb_entry *afs_vl_lookup_vldb(struct afs_cell *cell,
153 struct key *key,
154 const char *volname,
155 size_t volnamesz)
156{
157 struct afs_vldb_entry *vldb = ERR_PTR(-EDESTADDRREQ);
158 struct afs_vl_cursor vc;
159 int ret;
160
161 if (!afs_begin_vlserver_operation(&vc, cell, key))
162 return ERR_PTR(-ERESTARTSYS);
163
164 while (afs_select_vlserver(&vc)) {
165 vldb = afs_vl_get_entry_by_name_u(&vc, volname, volnamesz);
166 }
167
168 ret = afs_end_vlserver_operation(&vc);
169 return ret < 0 ? ERR_PTR(ret) : vldb;
170}
171
172/*
173 * Look up a volume in the VL server and create a candidate volume record for
174 * it.
175 *
176 * The volume name can be one of the following:
177 * "%[cell:]volume[.]" R/W volume
178 * "#[cell:]volume[.]" R/O or R/W volume (rwparent=0),
179 * or R/W (rwparent=1) volume
180 * "%[cell:]volume.readonly" R/O volume
181 * "#[cell:]volume.readonly" R/O volume
182 * "%[cell:]volume.backup" Backup volume
183 * "#[cell:]volume.backup" Backup volume
184 *
185 * The cell name is optional, and defaults to the current cell.
186 *
187 * See "The Rules of Mount Point Traversal" in Chapter 5 of the AFS SysAdmin
188 * Guide
189 * - Rule 1: Explicit type suffix forces access of that type or nothing
190 * (no suffix, then use Rule 2 & 3)
191 * - Rule 2: If parent volume is R/O, then mount R/O volume by preference, R/W
192 * if not available
193 * - Rule 3: If parent volume is R/W, then only mount R/W volume unless
194 * explicitly told otherwise
195 */
196struct afs_volume *afs_create_volume(struct afs_fs_context *params)
197{
198 struct afs_vldb_entry *vldb;
199 struct afs_volume *volume;
200 unsigned long type_mask = 1UL << params->type;
201
202 vldb = afs_vl_lookup_vldb(params->cell, params->key,
203 params->volname, params->volnamesz);
204 if (IS_ERR(vldb))
205 return ERR_CAST(vldb);
206
207 if (test_bit(AFS_VLDB_QUERY_ERROR, &vldb->flags)) {
208 volume = ERR_PTR(vldb->error);
209 goto error;
210 }
211
212 /* Make the final decision on the type we want */
213 volume = ERR_PTR(-ENOMEDIUM);
214 if (params->force) {
215 if (!(vldb->flags & type_mask))
216 goto error;
217 } else if (test_bit(AFS_VLDB_HAS_RO, &vldb->flags)) {
218 params->type = AFSVL_ROVOL;
219 } else if (test_bit(AFS_VLDB_HAS_RW, &vldb->flags)) {
220 params->type = AFSVL_RWVOL;
221 } else {
222 goto error;
223 }
224
225 volume = afs_lookup_volume(params, vldb);
226
227error:
228 kfree(vldb);
229 return volume;
230}
231
232/*
233 * Destroy a volume record
234 */
235static void afs_destroy_volume(struct work_struct *work)
236{
237 struct afs_volume *volume = container_of(work, struct afs_volume, destructor);
238 struct afs_server_list *slist = rcu_access_pointer(volume->servers);
239
240 _enter("%p", volume);
241
242#ifdef CONFIG_AFS_FSCACHE
243 ASSERTCMP(volume->cache, ==, NULL);
244#endif
245
246 afs_detach_volume_from_servers(volume, slist);
247 afs_remove_volume_from_cell(volume);
248 afs_put_serverlist(volume->cell->net, slist);
249 afs_put_cell(volume->cell, afs_cell_trace_put_vol);
250 trace_afs_volume(volume->vid, refcount_read(&volume->ref),
251 afs_volume_trace_free);
252 kfree_rcu(volume, rcu);
253
254 _leave(" [destroyed]");
255}
256
257/*
258 * Try to get a reference on a volume record.
259 */
260bool afs_try_get_volume(struct afs_volume *volume, enum afs_volume_trace reason)
261{
262 int r;
263
264 if (__refcount_inc_not_zero(&volume->ref, &r)) {
265 trace_afs_volume(volume->vid, r + 1, reason);
266 return true;
267 }
268 return false;
269}
270
271/*
272 * Get a reference on a volume record.
273 */
274struct afs_volume *afs_get_volume(struct afs_volume *volume,
275 enum afs_volume_trace reason)
276{
277 if (volume) {
278 int r;
279
280 __refcount_inc(&volume->ref, &r);
281 trace_afs_volume(volume->vid, r + 1, reason);
282 }
283 return volume;
284}
285
286
287/*
288 * Drop a reference on a volume record.
289 */
290void afs_put_volume(struct afs_volume *volume, enum afs_volume_trace reason)
291{
292 if (volume) {
293 afs_volid_t vid = volume->vid;
294 bool zero;
295 int r;
296
297 zero = __refcount_dec_and_test(&volume->ref, &r);
298 trace_afs_volume(vid, r - 1, reason);
299 if (zero)
300 schedule_work(&volume->destructor);
301 }
302}
303
304/*
305 * Activate a volume.
306 */
307int afs_activate_volume(struct afs_volume *volume)
308{
309#ifdef CONFIG_AFS_FSCACHE
310 struct fscache_volume *vcookie;
311 char *name;
312
313 name = kasprintf(GFP_KERNEL, "afs,%s,%llx",
314 volume->cell->name, volume->vid);
315 if (!name)
316 return -ENOMEM;
317
318 vcookie = fscache_acquire_volume(name, NULL, NULL, 0);
319 if (IS_ERR(vcookie)) {
320 if (vcookie != ERR_PTR(-EBUSY)) {
321 kfree(name);
322 return PTR_ERR(vcookie);
323 }
324 pr_err("AFS: Cache volume key already in use (%s)\n", name);
325 vcookie = NULL;
326 }
327 volume->cache = vcookie;
328 kfree(name);
329#endif
330 return 0;
331}
332
333/*
334 * Deactivate a volume.
335 */
336void afs_deactivate_volume(struct afs_volume *volume)
337{
338 _enter("%s", volume->name);
339
340#ifdef CONFIG_AFS_FSCACHE
341 fscache_relinquish_volume(volume->cache, NULL,
342 test_bit(AFS_VOLUME_DELETED, &volume->flags));
343 volume->cache = NULL;
344#endif
345
346 _leave("");
347}
348
349/*
350 * Query the VL service to update the volume status.
351 */
352static int afs_update_volume_status(struct afs_volume *volume, struct key *key)
353{
354 struct afs_server_list *new, *old, *discard;
355 struct afs_vldb_entry *vldb;
356 char idbuf[24];
357 int ret, idsz;
358
359 _enter("");
360
361 /* We look up an ID by passing it as a decimal string in the
362 * operation's name parameter.
363 */
364 idsz = snprintf(idbuf, sizeof(idbuf), "%llu", volume->vid);
365
366 vldb = afs_vl_lookup_vldb(volume->cell, key, idbuf, idsz);
367 if (IS_ERR(vldb)) {
368 ret = PTR_ERR(vldb);
369 goto error;
370 }
371
372 /* See if the volume got renamed. */
373 if (vldb->name_len != volume->name_len ||
374 memcmp(vldb->name, volume->name, vldb->name_len) != 0) {
375 /* TODO: Use RCU'd string. */
376 memcpy(volume->name, vldb->name, AFS_MAXVOLNAME);
377 volume->name_len = vldb->name_len;
378 }
379
380 /* See if the volume's server list got updated. */
381 new = afs_alloc_server_list(volume, key, vldb);
382 if (IS_ERR(new)) {
383 ret = PTR_ERR(new);
384 goto error_vldb;
385 }
386
387 write_lock(&volume->servers_lock);
388
389 discard = new;
390 old = rcu_dereference_protected(volume->servers,
391 lockdep_is_held(&volume->servers_lock));
392 if (afs_annotate_server_list(new, old)) {
393 new->seq = volume->servers_seq + 1;
394 rcu_assign_pointer(volume->servers, new);
395 smp_wmb();
396 volume->servers_seq++;
397 discard = old;
398 }
399
400 /* Check more often if replication is ongoing. */
401 if (new->ro_replicating)
402 volume->update_at = ktime_get_real_seconds() + 10 * 60;
403 else
404 volume->update_at = ktime_get_real_seconds() + afs_volume_record_life;
405 write_unlock(&volume->servers_lock);
406
407 if (discard == old)
408 afs_reattach_volume_to_servers(volume, new, old);
409 afs_put_serverlist(volume->cell->net, discard);
410 ret = 0;
411error_vldb:
412 kfree(vldb);
413error:
414 _leave(" = %d", ret);
415 return ret;
416}
417
418/*
419 * Make sure the volume record is up to date.
420 */
421int afs_check_volume_status(struct afs_volume *volume, struct afs_operation *op)
422{
423 int ret, retries = 0;
424
425 _enter("");
426
427retry:
428 if (test_bit(AFS_VOLUME_WAIT, &volume->flags))
429 goto wait;
430 if (volume->update_at <= ktime_get_real_seconds() ||
431 test_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags))
432 goto update;
433 _leave(" = 0");
434 return 0;
435
436update:
437 if (!test_and_set_bit_lock(AFS_VOLUME_UPDATING, &volume->flags)) {
438 clear_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
439 ret = afs_update_volume_status(volume, op->key);
440 if (ret < 0)
441 set_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
442 clear_bit_unlock(AFS_VOLUME_WAIT, &volume->flags);
443 clear_bit_unlock(AFS_VOLUME_UPDATING, &volume->flags);
444 wake_up_bit(&volume->flags, AFS_VOLUME_WAIT);
445 _leave(" = %d", ret);
446 return ret;
447 }
448
449wait:
450 if (!test_bit(AFS_VOLUME_WAIT, &volume->flags)) {
451 _leave(" = 0 [no wait]");
452 return 0;
453 }
454
455 ret = wait_on_bit(&volume->flags, AFS_VOLUME_WAIT,
456 (op->flags & AFS_OPERATION_UNINTR) ?
457 TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
458 if (ret == -ERESTARTSYS) {
459 _leave(" = %d", ret);
460 return ret;
461 }
462
463 retries++;
464 if (retries == 4) {
465 _leave(" = -ESTALE");
466 return -ESTALE;
467 }
468 goto retry;
469}
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(¶ms->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 ret = bdi_setup_and_register(&volume->bdi, "afs", BDI_CAP_MAP_COPY);
110 if (ret)
111 goto error_bdi;
112
113 init_rwsem(&volume->server_sem);
114
115 /* look up all the applicable server records */
116 for (loop = 0; loop < 8; loop++) {
117 if (vlocation->vldb.srvtmask[loop] & (1 << volume->type)) {
118 server = afs_lookup_server(
119 volume->cell, &vlocation->vldb.servers[loop]);
120 if (IS_ERR(server)) {
121 ret = PTR_ERR(server);
122 goto error_discard;
123 }
124
125 volume->servers[volume->nservers] = server;
126 volume->nservers++;
127 }
128 }
129
130 /* attach the cache and volume location */
131#ifdef CONFIG_AFS_FSCACHE
132 volume->cache = fscache_acquire_cookie(vlocation->cache,
133 &afs_volume_cache_index_def,
134 volume);
135#endif
136 afs_get_vlocation(vlocation);
137 volume->vlocation = vlocation;
138
139 vlocation->vols[volume->type] = volume;
140
141success:
142 _debug("kAFS selected %s volume %08x",
143 afs_voltypes[volume->type], volume->vid);
144 up_write(¶ms->cell->vl_sem);
145 afs_put_vlocation(vlocation);
146 _leave(" = %p", volume);
147 return volume;
148
149 /* clean up */
150error_up:
151 up_write(¶ms->cell->vl_sem);
152error:
153 afs_put_vlocation(vlocation);
154 _leave(" = %d", ret);
155 return ERR_PTR(ret);
156
157error_discard:
158 bdi_destroy(&volume->bdi);
159error_bdi:
160 up_write(¶ms->cell->vl_sem);
161
162 for (loop = volume->nservers - 1; loop >= 0; loop--)
163 afs_put_server(volume->servers[loop]);
164
165 kfree(volume);
166 goto error;
167}
168
169/*
170 * destroy a volume record
171 */
172void afs_put_volume(struct afs_volume *volume)
173{
174 struct afs_vlocation *vlocation;
175 int loop;
176
177 if (!volume)
178 return;
179
180 _enter("%p", volume);
181
182 ASSERTCMP(atomic_read(&volume->usage), >, 0);
183
184 vlocation = volume->vlocation;
185
186 /* to prevent a race, the decrement and the dequeue must be effectively
187 * atomic */
188 down_write(&vlocation->cell->vl_sem);
189
190 if (likely(!atomic_dec_and_test(&volume->usage))) {
191 up_write(&vlocation->cell->vl_sem);
192 _leave("");
193 return;
194 }
195
196 vlocation->vols[volume->type] = NULL;
197
198 up_write(&vlocation->cell->vl_sem);
199
200 /* finish cleaning up the volume */
201#ifdef CONFIG_AFS_FSCACHE
202 fscache_relinquish_cookie(volume->cache, 0);
203#endif
204 afs_put_vlocation(vlocation);
205
206 for (loop = volume->nservers - 1; loop >= 0; loop--)
207 afs_put_server(volume->servers[loop]);
208
209 bdi_destroy(&volume->bdi);
210 kfree(volume);
211
212 _leave(" [destroyed]");
213}
214
215/*
216 * pick a server to use to try accessing this volume
217 * - returns with an elevated usage count on the server chosen
218 */
219struct afs_server *afs_volume_pick_fileserver(struct afs_vnode *vnode)
220{
221 struct afs_volume *volume = vnode->volume;
222 struct afs_server *server;
223 int ret, state, loop;
224
225 _enter("%s", volume->vlocation->vldb.name);
226
227 /* stick with the server we're already using if we can */
228 if (vnode->server && vnode->server->fs_state == 0) {
229 afs_get_server(vnode->server);
230 _leave(" = %p [current]", vnode->server);
231 return vnode->server;
232 }
233
234 down_read(&volume->server_sem);
235
236 /* handle the no-server case */
237 if (volume->nservers == 0) {
238 ret = volume->rjservers ? -ENOMEDIUM : -ESTALE;
239 up_read(&volume->server_sem);
240 _leave(" = %d [no servers]", ret);
241 return ERR_PTR(ret);
242 }
243
244 /* basically, just search the list for the first live server and use
245 * that */
246 ret = 0;
247 for (loop = 0; loop < volume->nservers; loop++) {
248 server = volume->servers[loop];
249 state = server->fs_state;
250
251 _debug("consider %d [%d]", loop, state);
252
253 switch (state) {
254 /* found an apparently healthy server */
255 case 0:
256 afs_get_server(server);
257 up_read(&volume->server_sem);
258 _leave(" = %p (picked %08x)",
259 server, ntohl(server->addr.s_addr));
260 return server;
261
262 case -ENETUNREACH:
263 if (ret == 0)
264 ret = state;
265 break;
266
267 case -EHOSTUNREACH:
268 if (ret == 0 ||
269 ret == -ENETUNREACH)
270 ret = state;
271 break;
272
273 case -ECONNREFUSED:
274 if (ret == 0 ||
275 ret == -ENETUNREACH ||
276 ret == -EHOSTUNREACH)
277 ret = state;
278 break;
279
280 default:
281 case -EREMOTEIO:
282 if (ret == 0 ||
283 ret == -ENETUNREACH ||
284 ret == -EHOSTUNREACH ||
285 ret == -ECONNREFUSED)
286 ret = state;
287 break;
288 }
289 }
290
291 /* no available servers
292 * - TODO: handle the no active servers case better
293 */
294 up_read(&volume->server_sem);
295 _leave(" = %d", ret);
296 return ERR_PTR(ret);
297}
298
299/*
300 * release a server after use
301 * - releases the ref on the server struct that was acquired by picking
302 * - records result of using a particular server to access a volume
303 * - return 0 to try again, 1 if okay or to issue error
304 * - the caller must release the server struct if result was 0
305 */
306int afs_volume_release_fileserver(struct afs_vnode *vnode,
307 struct afs_server *server,
308 int result)
309{
310 struct afs_volume *volume = vnode->volume;
311 unsigned loop;
312
313 _enter("%s,%08x,%d",
314 volume->vlocation->vldb.name, ntohl(server->addr.s_addr),
315 result);
316
317 switch (result) {
318 /* success */
319 case 0:
320 server->fs_act_jif = jiffies;
321 server->fs_state = 0;
322 _leave("");
323 return 1;
324
325 /* the fileserver denied all knowledge of the volume */
326 case -ENOMEDIUM:
327 server->fs_act_jif = jiffies;
328 down_write(&volume->server_sem);
329
330 /* firstly, find where the server is in the active list (if it
331 * is) */
332 for (loop = 0; loop < volume->nservers; loop++)
333 if (volume->servers[loop] == server)
334 goto present;
335
336 /* no longer there - may have been discarded by another op */
337 goto try_next_server_upw;
338
339 present:
340 volume->nservers--;
341 memmove(&volume->servers[loop],
342 &volume->servers[loop + 1],
343 sizeof(volume->servers[loop]) *
344 (volume->nservers - loop));
345 volume->servers[volume->nservers] = NULL;
346 afs_put_server(server);
347 volume->rjservers++;
348
349 if (volume->nservers > 0)
350 /* another server might acknowledge its existence */
351 goto try_next_server_upw;
352
353 /* handle the case where all the fileservers have rejected the
354 * volume
355 * - TODO: try asking the fileservers for volume information
356 * - TODO: contact the VL server again to see if the volume is
357 * no longer registered
358 */
359 up_write(&volume->server_sem);
360 afs_put_server(server);
361 _leave(" [completely rejected]");
362 return 1;
363
364 /* problem reaching the server */
365 case -ENETUNREACH:
366 case -EHOSTUNREACH:
367 case -ECONNREFUSED:
368 case -ETIME:
369 case -ETIMEDOUT:
370 case -EREMOTEIO:
371 /* mark the server as dead
372 * TODO: vary dead timeout depending on error
373 */
374 spin_lock(&server->fs_lock);
375 if (!server->fs_state) {
376 server->fs_dead_jif = jiffies + HZ * 10;
377 server->fs_state = result;
378 printk("kAFS: SERVER DEAD state=%d\n", result);
379 }
380 spin_unlock(&server->fs_lock);
381 goto try_next_server;
382
383 /* miscellaneous error */
384 default:
385 server->fs_act_jif = jiffies;
386 case -ENOMEM:
387 case -ENONET:
388 /* tell the caller to accept the result */
389 afs_put_server(server);
390 _leave(" [local failure]");
391 return 1;
392 }
393
394 /* tell the caller to loop around and try the next server */
395try_next_server_upw:
396 up_write(&volume->server_sem);
397try_next_server:
398 afs_put_server(server);
399 _leave(" [try next server]");
400 return 0;
401}