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
12 static unsigned __read_mostly afs_volume_record_life = 60 * 60;
13
14 /*
15 * Insert a volume into a cell. If there's an existing volume record, that is
16 * returned instead with a ref held.
17 */
afs_insert_volume_into_cell(struct afs_cell * cell,struct afs_volume * volume)18 static struct afs_volume *afs_insert_volume_into_cell(struct afs_cell *cell,
19 struct afs_volume *volume)
20 {
21 struct afs_volume *p;
22 struct rb_node *parent = NULL, **pp;
23
24 write_seqlock(&cell->volume_lock);
25
26 pp = &cell->volumes.rb_node;
27 while (*pp) {
28 parent = *pp;
29 p = rb_entry(parent, struct afs_volume, cell_node);
30 if (p->vid < volume->vid) {
31 pp = &(*pp)->rb_left;
32 } else if (p->vid > volume->vid) {
33 pp = &(*pp)->rb_right;
34 } else {
35 if (afs_try_get_volume(p, afs_volume_trace_get_cell_insert)) {
36 volume = p;
37 goto found;
38 }
39
40 set_bit(AFS_VOLUME_RM_TREE, &volume->flags);
41 rb_replace_node_rcu(&p->cell_node, &volume->cell_node, &cell->volumes);
42 }
43 }
44
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);
48
49 found:
50 write_sequnlock(&cell->volume_lock);
51 return volume;
52
53 }
54
afs_remove_volume_from_cell(struct afs_volume * volume)55 static void afs_remove_volume_from_cell(struct afs_volume *volume)
56 {
57 struct afs_cell *cell = volume->cell;
58
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);
67 }
68 }
69
70 /*
71 * Allocate a volume record and load it up from a vldb record.
72 */
afs_alloc_volume(struct afs_fs_context * params,struct afs_vldb_entry * vldb,unsigned long type_mask)73 static struct afs_volume *afs_alloc_volume(struct afs_fs_context *params,
74 struct afs_vldb_entry *vldb,
75 unsigned long type_mask)
76 {
77 struct afs_server_list *slist;
78 struct afs_volume *volume;
79 int ret = -ENOMEM;
80
81 volume = kzalloc(sizeof(struct afs_volume), GFP_KERNEL);
82 if (!volume)
83 goto error_0;
84
85 volume->vid = vldb->vid[params->type];
86 volume->update_at = ktime_get_real_seconds() + afs_volume_record_life;
87 volume->cell = afs_get_cell(params->cell, afs_cell_trace_get_vol);
88 volume->type = params->type;
89 volume->type_force = params->force;
90 volume->name_len = vldb->name_len;
91
92 refcount_set(&volume->ref, 1);
93 INIT_HLIST_NODE(&volume->proc_link);
94 rwlock_init(&volume->servers_lock);
95 rwlock_init(&volume->cb_v_break_lock);
96 memcpy(volume->name, vldb->name, vldb->name_len + 1);
97
98 slist = afs_alloc_server_list(params->cell, params->key, vldb, type_mask);
99 if (IS_ERR(slist)) {
100 ret = PTR_ERR(slist);
101 goto error_1;
102 }
103
104 refcount_set(&slist->usage, 1);
105 rcu_assign_pointer(volume->servers, slist);
106 trace_afs_volume(volume->vid, 1, afs_volume_trace_alloc);
107 return volume;
108
109 error_1:
110 afs_put_cell(volume->cell, afs_cell_trace_put_vol);
111 kfree(volume);
112 error_0:
113 return ERR_PTR(ret);
114 }
115
116 /*
117 * Look up or allocate a volume record.
118 */
afs_lookup_volume(struct afs_fs_context * params,struct afs_vldb_entry * vldb,unsigned long type_mask)119 static struct afs_volume *afs_lookup_volume(struct afs_fs_context *params,
120 struct afs_vldb_entry *vldb,
121 unsigned long type_mask)
122 {
123 struct afs_volume *candidate, *volume;
124
125 candidate = afs_alloc_volume(params, vldb, type_mask);
126 if (IS_ERR(candidate))
127 return candidate;
128
129 volume = afs_insert_volume_into_cell(params->cell, candidate);
130 if (volume != candidate)
131 afs_put_volume(params->net, candidate, afs_volume_trace_put_cell_dup);
132 return volume;
133 }
134
135 /*
136 * Look up a VLDB record for a volume.
137 */
afs_vl_lookup_vldb(struct afs_cell * cell,struct key * key,const char * volname,size_t volnamesz)138 static struct afs_vldb_entry *afs_vl_lookup_vldb(struct afs_cell *cell,
139 struct key *key,
140 const char *volname,
141 size_t volnamesz)
142 {
143 struct afs_vldb_entry *vldb = ERR_PTR(-EDESTADDRREQ);
144 struct afs_vl_cursor vc;
145 int ret;
146
147 if (!afs_begin_vlserver_operation(&vc, cell, key))
148 return ERR_PTR(-ERESTARTSYS);
149
150 while (afs_select_vlserver(&vc)) {
151 vldb = afs_vl_get_entry_by_name_u(&vc, volname, volnamesz);
152 }
153
154 ret = afs_end_vlserver_operation(&vc);
155 return ret < 0 ? ERR_PTR(ret) : vldb;
156 }
157
158 /*
159 * Look up a volume in the VL server and create a candidate volume record for
160 * it.
161 *
162 * The volume name can be one of the following:
163 * "%[cell:]volume[.]" R/W volume
164 * "#[cell:]volume[.]" R/O or R/W volume (rwparent=0),
165 * or R/W (rwparent=1) volume
166 * "%[cell:]volume.readonly" R/O volume
167 * "#[cell:]volume.readonly" R/O volume
168 * "%[cell:]volume.backup" Backup volume
169 * "#[cell:]volume.backup" Backup volume
170 *
171 * The cell name is optional, and defaults to the current cell.
172 *
173 * See "The Rules of Mount Point Traversal" in Chapter 5 of the AFS SysAdmin
174 * Guide
175 * - Rule 1: Explicit type suffix forces access of that type or nothing
176 * (no suffix, then use Rule 2 & 3)
177 * - Rule 2: If parent volume is R/O, then mount R/O volume by preference, R/W
178 * if not available
179 * - Rule 3: If parent volume is R/W, then only mount R/W volume unless
180 * explicitly told otherwise
181 */
afs_create_volume(struct afs_fs_context * params)182 struct afs_volume *afs_create_volume(struct afs_fs_context *params)
183 {
184 struct afs_vldb_entry *vldb;
185 struct afs_volume *volume;
186 unsigned long type_mask = 1UL << params->type;
187
188 vldb = afs_vl_lookup_vldb(params->cell, params->key,
189 params->volname, params->volnamesz);
190 if (IS_ERR(vldb))
191 return ERR_CAST(vldb);
192
193 if (test_bit(AFS_VLDB_QUERY_ERROR, &vldb->flags)) {
194 volume = ERR_PTR(vldb->error);
195 goto error;
196 }
197
198 /* Make the final decision on the type we want */
199 volume = ERR_PTR(-ENOMEDIUM);
200 if (params->force) {
201 if (!(vldb->flags & type_mask))
202 goto error;
203 } else if (test_bit(AFS_VLDB_HAS_RO, &vldb->flags)) {
204 params->type = AFSVL_ROVOL;
205 } else if (test_bit(AFS_VLDB_HAS_RW, &vldb->flags)) {
206 params->type = AFSVL_RWVOL;
207 } else {
208 goto error;
209 }
210
211 type_mask = 1UL << params->type;
212 volume = afs_lookup_volume(params, vldb, type_mask);
213
214 error:
215 kfree(vldb);
216 return volume;
217 }
218
219 /*
220 * Destroy a volume record
221 */
afs_destroy_volume(struct afs_net * net,struct afs_volume * volume)222 static void afs_destroy_volume(struct afs_net *net, struct afs_volume *volume)
223 {
224 _enter("%p", volume);
225
226 #ifdef CONFIG_AFS_FSCACHE
227 ASSERTCMP(volume->cache, ==, NULL);
228 #endif
229
230 afs_remove_volume_from_cell(volume);
231 afs_put_serverlist(net, rcu_access_pointer(volume->servers));
232 afs_put_cell(volume->cell, afs_cell_trace_put_vol);
233 trace_afs_volume(volume->vid, refcount_read(&volume->ref),
234 afs_volume_trace_free);
235 kfree_rcu(volume, rcu);
236
237 _leave(" [destroyed]");
238 }
239
240 /*
241 * Try to get a reference on a volume record.
242 */
afs_try_get_volume(struct afs_volume * volume,enum afs_volume_trace reason)243 bool afs_try_get_volume(struct afs_volume *volume, enum afs_volume_trace reason)
244 {
245 int r;
246
247 if (__refcount_inc_not_zero(&volume->ref, &r)) {
248 trace_afs_volume(volume->vid, r + 1, reason);
249 return true;
250 }
251 return false;
252 }
253
254 /*
255 * Get a reference on a volume record.
256 */
afs_get_volume(struct afs_volume * volume,enum afs_volume_trace reason)257 struct afs_volume *afs_get_volume(struct afs_volume *volume,
258 enum afs_volume_trace reason)
259 {
260 if (volume) {
261 int r;
262
263 __refcount_inc(&volume->ref, &r);
264 trace_afs_volume(volume->vid, r + 1, reason);
265 }
266 return volume;
267 }
268
269
270 /*
271 * Drop a reference on a volume record.
272 */
afs_put_volume(struct afs_net * net,struct afs_volume * volume,enum afs_volume_trace reason)273 void afs_put_volume(struct afs_net *net, struct afs_volume *volume,
274 enum afs_volume_trace reason)
275 {
276 if (volume) {
277 afs_volid_t vid = volume->vid;
278 bool zero;
279 int r;
280
281 zero = __refcount_dec_and_test(&volume->ref, &r);
282 trace_afs_volume(vid, r - 1, reason);
283 if (zero)
284 afs_destroy_volume(net, volume);
285 }
286 }
287
288 /*
289 * Activate a volume.
290 */
afs_activate_volume(struct afs_volume * volume)291 int afs_activate_volume(struct afs_volume *volume)
292 {
293 #ifdef CONFIG_AFS_FSCACHE
294 struct fscache_volume *vcookie;
295 char *name;
296
297 name = kasprintf(GFP_KERNEL, "afs,%s,%llx",
298 volume->cell->name, volume->vid);
299 if (!name)
300 return -ENOMEM;
301
302 vcookie = fscache_acquire_volume(name, NULL, NULL, 0);
303 if (IS_ERR(vcookie)) {
304 if (vcookie != ERR_PTR(-EBUSY)) {
305 kfree(name);
306 return PTR_ERR(vcookie);
307 }
308 pr_err("AFS: Cache volume key already in use (%s)\n", name);
309 vcookie = NULL;
310 }
311 volume->cache = vcookie;
312 kfree(name);
313 #endif
314 return 0;
315 }
316
317 /*
318 * Deactivate a volume.
319 */
afs_deactivate_volume(struct afs_volume * volume)320 void afs_deactivate_volume(struct afs_volume *volume)
321 {
322 _enter("%s", volume->name);
323
324 #ifdef CONFIG_AFS_FSCACHE
325 fscache_relinquish_volume(volume->cache, NULL,
326 test_bit(AFS_VOLUME_DELETED, &volume->flags));
327 volume->cache = NULL;
328 #endif
329
330 _leave("");
331 }
332
333 /*
334 * Query the VL service to update the volume status.
335 */
afs_update_volume_status(struct afs_volume * volume,struct key * key)336 static int afs_update_volume_status(struct afs_volume *volume, struct key *key)
337 {
338 struct afs_server_list *new, *old, *discard;
339 struct afs_vldb_entry *vldb;
340 char idbuf[24];
341 int ret, idsz;
342
343 _enter("");
344
345 /* We look up an ID by passing it as a decimal string in the
346 * operation's name parameter.
347 */
348 idsz = snprintf(idbuf, sizeof(idbuf), "%llu", volume->vid);
349
350 vldb = afs_vl_lookup_vldb(volume->cell, key, idbuf, idsz);
351 if (IS_ERR(vldb)) {
352 ret = PTR_ERR(vldb);
353 goto error;
354 }
355
356 /* See if the volume got renamed. */
357 if (vldb->name_len != volume->name_len ||
358 memcmp(vldb->name, volume->name, vldb->name_len) != 0) {
359 /* TODO: Use RCU'd string. */
360 memcpy(volume->name, vldb->name, AFS_MAXVOLNAME);
361 volume->name_len = vldb->name_len;
362 }
363
364 /* See if the volume's server list got updated. */
365 new = afs_alloc_server_list(volume->cell, key,
366 vldb, (1 << volume->type));
367 if (IS_ERR(new)) {
368 ret = PTR_ERR(new);
369 goto error_vldb;
370 }
371
372 write_lock(&volume->servers_lock);
373
374 discard = new;
375 old = rcu_dereference_protected(volume->servers,
376 lockdep_is_held(&volume->servers_lock));
377 if (afs_annotate_server_list(new, old)) {
378 new->seq = volume->servers_seq + 1;
379 rcu_assign_pointer(volume->servers, new);
380 smp_wmb();
381 volume->servers_seq++;
382 discard = old;
383 }
384
385 volume->update_at = ktime_get_real_seconds() + afs_volume_record_life;
386 write_unlock(&volume->servers_lock);
387 ret = 0;
388
389 afs_put_serverlist(volume->cell->net, discard);
390 error_vldb:
391 kfree(vldb);
392 error:
393 _leave(" = %d", ret);
394 return ret;
395 }
396
397 /*
398 * Make sure the volume record is up to date.
399 */
afs_check_volume_status(struct afs_volume * volume,struct afs_operation * op)400 int afs_check_volume_status(struct afs_volume *volume, struct afs_operation *op)
401 {
402 int ret, retries = 0;
403
404 _enter("");
405
406 retry:
407 if (test_bit(AFS_VOLUME_WAIT, &volume->flags))
408 goto wait;
409 if (volume->update_at <= ktime_get_real_seconds() ||
410 test_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags))
411 goto update;
412 _leave(" = 0");
413 return 0;
414
415 update:
416 if (!test_and_set_bit_lock(AFS_VOLUME_UPDATING, &volume->flags)) {
417 clear_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
418 ret = afs_update_volume_status(volume, op->key);
419 if (ret < 0)
420 set_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
421 clear_bit_unlock(AFS_VOLUME_WAIT, &volume->flags);
422 clear_bit_unlock(AFS_VOLUME_UPDATING, &volume->flags);
423 wake_up_bit(&volume->flags, AFS_VOLUME_WAIT);
424 _leave(" = %d", ret);
425 return ret;
426 }
427
428 wait:
429 if (!test_bit(AFS_VOLUME_WAIT, &volume->flags)) {
430 _leave(" = 0 [no wait]");
431 return 0;
432 }
433
434 ret = wait_on_bit(&volume->flags, AFS_VOLUME_WAIT,
435 (op->flags & AFS_OPERATION_UNINTR) ?
436 TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
437 if (ret == -ERESTARTSYS) {
438 _leave(" = %d", ret);
439 return ret;
440 }
441
442 retries++;
443 if (retries == 4) {
444 _leave(" = -ESTALE");
445 return -ESTALE;
446 }
447 goto retry;
448 }
449