1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* AFS volume management
4 * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
8 #include <linux/kernel.h>
9 #include <linux/slab.h>
12 static unsigned __read_mostly afs_volume_record_life = 60 * 60;
15 * Insert a volume into a cell. If there's an existing volume record, that is
16 * returned instead with a ref held.
18 static struct afs_volume *afs_insert_volume_into_cell(struct afs_cell *cell,
19 struct afs_volume *volume)
22 struct rb_node *parent = NULL, **pp;
24 write_seqlock(&cell->volume_lock);
26 pp = &cell->volumes.rb_node;
29 p = rb_entry(parent, struct afs_volume, cell_node);
30 if (p->vid < volume->vid) {
32 } else if (p->vid > volume->vid) {
33 pp = &(*pp)->rb_right;
35 if (afs_try_get_volume(p, afs_volume_trace_get_cell_insert)) {
40 set_bit(AFS_VOLUME_RM_TREE, &volume->flags);
41 rb_replace_node_rcu(&p->cell_node, &volume->cell_node, &cell->volumes);
45 rb_link_node_rcu(&volume->cell_node, parent, pp);
46 rb_insert_color(&volume->cell_node, &cell->volumes);
47 hlist_add_head_rcu(&volume->proc_link, &cell->proc_volumes);
50 write_sequnlock(&cell->volume_lock);
55 static void afs_remove_volume_from_cell(struct afs_volume *volume)
57 struct afs_cell *cell = volume->cell;
59 if (!hlist_unhashed(&volume->proc_link)) {
60 trace_afs_volume(volume->vid, refcount_read(&cell->ref),
61 afs_volume_trace_remove);
62 write_seqlock(&cell->volume_lock);
63 hlist_del_rcu(&volume->proc_link);
64 if (!test_and_set_bit(AFS_VOLUME_RM_TREE, &volume->flags))
65 rb_erase(&volume->cell_node, &cell->volumes);
66 write_sequnlock(&cell->volume_lock);
71 * Allocate a volume record and load it up from a vldb record.
73 static struct afs_volume *afs_alloc_volume(struct afs_fs_context *params,
74 struct afs_vldb_entry *vldb,
75 unsigned long type_mask)
77 struct afs_server_list *slist;
78 struct afs_volume *volume;
79 int ret = -ENOMEM, nr_servers = 0, i;
81 for (i = 0; i < vldb->nr_servers; i++)
82 if (vldb->fs_mask[i] & type_mask)
85 volume = kzalloc(sizeof(struct afs_volume), GFP_KERNEL);
89 volume->vid = vldb->vid[params->type];
90 volume->update_at = ktime_get_real_seconds() + afs_volume_record_life;
91 volume->cell = afs_get_cell(params->cell, afs_cell_trace_get_vol);
92 volume->type = params->type;
93 volume->type_force = params->force;
94 volume->name_len = vldb->name_len;
96 refcount_set(&volume->ref, 1);
97 INIT_HLIST_NODE(&volume->proc_link);
98 rwlock_init(&volume->servers_lock);
99 rwlock_init(&volume->cb_v_break_lock);
100 memcpy(volume->name, vldb->name, vldb->name_len + 1);
102 slist = afs_alloc_server_list(params->cell, params->key, vldb, type_mask);
104 ret = PTR_ERR(slist);
108 refcount_set(&slist->usage, 1);
109 rcu_assign_pointer(volume->servers, slist);
110 trace_afs_volume(volume->vid, 1, afs_volume_trace_alloc);
114 afs_put_cell(volume->cell, afs_cell_trace_put_vol);
121 * Look up or allocate a volume record.
123 static struct afs_volume *afs_lookup_volume(struct afs_fs_context *params,
124 struct afs_vldb_entry *vldb,
125 unsigned long type_mask)
127 struct afs_volume *candidate, *volume;
129 candidate = afs_alloc_volume(params, vldb, type_mask);
130 if (IS_ERR(candidate))
133 volume = afs_insert_volume_into_cell(params->cell, candidate);
134 if (volume != candidate)
135 afs_put_volume(params->net, candidate, afs_volume_trace_put_cell_dup);
140 * Look up a VLDB record for a volume.
142 static struct afs_vldb_entry *afs_vl_lookup_vldb(struct afs_cell *cell,
147 struct afs_vldb_entry *vldb = ERR_PTR(-EDESTADDRREQ);
148 struct afs_vl_cursor vc;
151 if (!afs_begin_vlserver_operation(&vc, cell, key))
152 return ERR_PTR(-ERESTARTSYS);
154 while (afs_select_vlserver(&vc)) {
155 vldb = afs_vl_get_entry_by_name_u(&vc, volname, volnamesz);
158 ret = afs_end_vlserver_operation(&vc);
159 return ret < 0 ? ERR_PTR(ret) : vldb;
163 * Look up a volume in the VL server and create a candidate volume record for
166 * The volume name can be one of the following:
167 * "%[cell:]volume[.]" R/W volume
168 * "#[cell:]volume[.]" R/O or R/W volume (rwparent=0),
169 * or R/W (rwparent=1) volume
170 * "%[cell:]volume.readonly" R/O volume
171 * "#[cell:]volume.readonly" R/O volume
172 * "%[cell:]volume.backup" Backup volume
173 * "#[cell:]volume.backup" Backup volume
175 * The cell name is optional, and defaults to the current cell.
177 * See "The Rules of Mount Point Traversal" in Chapter 5 of the AFS SysAdmin
179 * - Rule 1: Explicit type suffix forces access of that type or nothing
180 * (no suffix, then use Rule 2 & 3)
181 * - Rule 2: If parent volume is R/O, then mount R/O volume by preference, R/W
183 * - Rule 3: If parent volume is R/W, then only mount R/W volume unless
184 * explicitly told otherwise
186 struct afs_volume *afs_create_volume(struct afs_fs_context *params)
188 struct afs_vldb_entry *vldb;
189 struct afs_volume *volume;
190 unsigned long type_mask = 1UL << params->type;
192 vldb = afs_vl_lookup_vldb(params->cell, params->key,
193 params->volname, params->volnamesz);
195 return ERR_CAST(vldb);
197 if (test_bit(AFS_VLDB_QUERY_ERROR, &vldb->flags)) {
198 volume = ERR_PTR(vldb->error);
202 /* Make the final decision on the type we want */
203 volume = ERR_PTR(-ENOMEDIUM);
205 if (!(vldb->flags & type_mask))
207 } else if (test_bit(AFS_VLDB_HAS_RO, &vldb->flags)) {
208 params->type = AFSVL_ROVOL;
209 } else if (test_bit(AFS_VLDB_HAS_RW, &vldb->flags)) {
210 params->type = AFSVL_RWVOL;
215 type_mask = 1UL << params->type;
216 volume = afs_lookup_volume(params, vldb, type_mask);
224 * Destroy a volume record
226 static void afs_destroy_volume(struct afs_net *net, struct afs_volume *volume)
228 _enter("%p", volume);
230 #ifdef CONFIG_AFS_FSCACHE
231 ASSERTCMP(volume->cache, ==, NULL);
234 afs_remove_volume_from_cell(volume);
235 afs_put_serverlist(net, rcu_access_pointer(volume->servers));
236 afs_put_cell(volume->cell, afs_cell_trace_put_vol);
237 trace_afs_volume(volume->vid, refcount_read(&volume->ref),
238 afs_volume_trace_free);
239 kfree_rcu(volume, rcu);
241 _leave(" [destroyed]");
245 * Try to get a reference on a volume record.
247 bool afs_try_get_volume(struct afs_volume *volume, enum afs_volume_trace reason)
251 if (__refcount_inc_not_zero(&volume->ref, &r)) {
252 trace_afs_volume(volume->vid, r + 1, reason);
259 * Get a reference on a volume record.
261 struct afs_volume *afs_get_volume(struct afs_volume *volume,
262 enum afs_volume_trace reason)
267 __refcount_inc(&volume->ref, &r);
268 trace_afs_volume(volume->vid, r + 1, reason);
275 * Drop a reference on a volume record.
277 void afs_put_volume(struct afs_net *net, struct afs_volume *volume,
278 enum afs_volume_trace reason)
281 afs_volid_t vid = volume->vid;
285 zero = __refcount_dec_and_test(&volume->ref, &r);
286 trace_afs_volume(vid, r - 1, reason);
288 afs_destroy_volume(net, volume);
295 int afs_activate_volume(struct afs_volume *volume)
297 #ifdef CONFIG_AFS_FSCACHE
298 struct fscache_volume *vcookie;
301 name = kasprintf(GFP_KERNEL, "afs,%s,%llx",
302 volume->cell->name, volume->vid);
306 vcookie = fscache_acquire_volume(name, NULL, NULL, 0);
307 if (IS_ERR(vcookie)) {
308 if (vcookie != ERR_PTR(-EBUSY)) {
310 return PTR_ERR(vcookie);
312 pr_err("AFS: Cache volume key already in use (%s)\n", name);
315 volume->cache = vcookie;
322 * Deactivate a volume.
324 void afs_deactivate_volume(struct afs_volume *volume)
326 _enter("%s", volume->name);
328 #ifdef CONFIG_AFS_FSCACHE
329 fscache_relinquish_volume(volume->cache, NULL,
330 test_bit(AFS_VOLUME_DELETED, &volume->flags));
331 volume->cache = NULL;
338 * Query the VL service to update the volume status.
340 static int afs_update_volume_status(struct afs_volume *volume, struct key *key)
342 struct afs_server_list *new, *old, *discard;
343 struct afs_vldb_entry *vldb;
349 /* We look up an ID by passing it as a decimal string in the
350 * operation's name parameter.
352 idsz = snprintf(idbuf, sizeof(idbuf), "%llu", volume->vid);
354 vldb = afs_vl_lookup_vldb(volume->cell, key, idbuf, idsz);
360 /* See if the volume got renamed. */
361 if (vldb->name_len != volume->name_len ||
362 memcmp(vldb->name, volume->name, vldb->name_len) != 0) {
363 /* TODO: Use RCU'd string. */
364 memcpy(volume->name, vldb->name, AFS_MAXVOLNAME);
365 volume->name_len = vldb->name_len;
368 /* See if the volume's server list got updated. */
369 new = afs_alloc_server_list(volume->cell, key,
370 vldb, (1 << volume->type));
376 write_lock(&volume->servers_lock);
379 old = rcu_dereference_protected(volume->servers,
380 lockdep_is_held(&volume->servers_lock));
381 if (afs_annotate_server_list(new, old)) {
382 new->seq = volume->servers_seq + 1;
383 rcu_assign_pointer(volume->servers, new);
385 volume->servers_seq++;
389 volume->update_at = ktime_get_real_seconds() + afs_volume_record_life;
390 write_unlock(&volume->servers_lock);
393 afs_put_serverlist(volume->cell->net, discard);
397 _leave(" = %d", ret);
402 * Make sure the volume record is up to date.
404 int afs_check_volume_status(struct afs_volume *volume, struct afs_operation *op)
406 int ret, retries = 0;
411 if (test_bit(AFS_VOLUME_WAIT, &volume->flags))
413 if (volume->update_at <= ktime_get_real_seconds() ||
414 test_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags))
420 if (!test_and_set_bit_lock(AFS_VOLUME_UPDATING, &volume->flags)) {
421 clear_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
422 ret = afs_update_volume_status(volume, op->key);
424 set_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
425 clear_bit_unlock(AFS_VOLUME_WAIT, &volume->flags);
426 clear_bit_unlock(AFS_VOLUME_UPDATING, &volume->flags);
427 wake_up_bit(&volume->flags, AFS_VOLUME_WAIT);
428 _leave(" = %d", ret);
433 if (!test_bit(AFS_VOLUME_WAIT, &volume->flags)) {
434 _leave(" = 0 [no wait]");
438 ret = wait_on_bit(&volume->flags, AFS_VOLUME_WAIT,
439 (op->flags & AFS_OPERATION_UNINTR) ?
440 TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
441 if (ret == -ERESTARTSYS) {
442 _leave(" = %d", ret);
448 _leave(" = -ESTALE");