xref: /linux/fs/afs/volume.c (revision 2dbc0838bcf24ca59cabc3130cf3b1d6809cdcd4)
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 unsigned __read_mostly afs_volume_gc_delay = 10;
13 unsigned __read_mostly afs_volume_record_life = 60 * 60;
14 
15 static const char *const afs_voltypes[] = { "R/W", "R/O", "BAK" };
16 
17 /*
18  * Allocate a volume record and load it up from a vldb record.
19  */
20 static struct afs_volume *afs_alloc_volume(struct afs_fs_context *params,
21 					   struct afs_vldb_entry *vldb,
22 					   unsigned long type_mask)
23 {
24 	struct afs_server_list *slist;
25 	struct afs_volume *volume;
26 	int ret = -ENOMEM, nr_servers = 0, i;
27 
28 	for (i = 0; i < vldb->nr_servers; i++)
29 		if (vldb->fs_mask[i] & type_mask)
30 			nr_servers++;
31 
32 	volume = kzalloc(sizeof(struct afs_volume), GFP_KERNEL);
33 	if (!volume)
34 		goto error_0;
35 
36 	volume->vid		= vldb->vid[params->type];
37 	volume->update_at	= ktime_get_real_seconds() + afs_volume_record_life;
38 	volume->cell		= afs_get_cell(params->cell);
39 	volume->type		= params->type;
40 	volume->type_force	= params->force;
41 	volume->name_len	= vldb->name_len;
42 
43 	atomic_set(&volume->usage, 1);
44 	INIT_LIST_HEAD(&volume->proc_link);
45 	rwlock_init(&volume->servers_lock);
46 	rwlock_init(&volume->cb_v_break_lock);
47 	memcpy(volume->name, vldb->name, vldb->name_len + 1);
48 
49 	slist = afs_alloc_server_list(params->cell, params->key, vldb, type_mask);
50 	if (IS_ERR(slist)) {
51 		ret = PTR_ERR(slist);
52 		goto error_1;
53 	}
54 
55 	refcount_set(&slist->usage, 1);
56 	volume->servers = slist;
57 	return volume;
58 
59 error_1:
60 	afs_put_cell(params->net, volume->cell);
61 	kfree(volume);
62 error_0:
63 	return ERR_PTR(ret);
64 }
65 
66 /*
67  * Look up a VLDB record for a volume.
68  */
69 static struct afs_vldb_entry *afs_vl_lookup_vldb(struct afs_cell *cell,
70 						 struct key *key,
71 						 const char *volname,
72 						 size_t volnamesz)
73 {
74 	struct afs_vldb_entry *vldb = ERR_PTR(-EDESTADDRREQ);
75 	struct afs_vl_cursor vc;
76 	int ret;
77 
78 	if (!afs_begin_vlserver_operation(&vc, cell, key))
79 		return ERR_PTR(-ERESTARTSYS);
80 
81 	while (afs_select_vlserver(&vc)) {
82 		vldb = afs_vl_get_entry_by_name_u(&vc, volname, volnamesz);
83 	}
84 
85 	ret = afs_end_vlserver_operation(&vc);
86 	return ret < 0 ? ERR_PTR(ret) : vldb;
87 }
88 
89 /*
90  * Look up a volume in the VL server and create a candidate volume record for
91  * it.
92  *
93  * The volume name can be one of the following:
94  *	"%[cell:]volume[.]"		R/W volume
95  *	"#[cell:]volume[.]"		R/O or R/W volume (rwparent=0),
96  *					 or R/W (rwparent=1) volume
97  *	"%[cell:]volume.readonly"	R/O volume
98  *	"#[cell:]volume.readonly"	R/O volume
99  *	"%[cell:]volume.backup"		Backup volume
100  *	"#[cell:]volume.backup"		Backup volume
101  *
102  * The cell name is optional, and defaults to the current cell.
103  *
104  * See "The Rules of Mount Point Traversal" in Chapter 5 of the AFS SysAdmin
105  * Guide
106  * - Rule 1: Explicit type suffix forces access of that type or nothing
107  *           (no suffix, then use Rule 2 & 3)
108  * - Rule 2: If parent volume is R/O, then mount R/O volume by preference, R/W
109  *           if not available
110  * - Rule 3: If parent volume is R/W, then only mount R/W volume unless
111  *           explicitly told otherwise
112  */
113 struct afs_volume *afs_create_volume(struct afs_fs_context *params)
114 {
115 	struct afs_vldb_entry *vldb;
116 	struct afs_volume *volume;
117 	unsigned long type_mask = 1UL << params->type;
118 
119 	vldb = afs_vl_lookup_vldb(params->cell, params->key,
120 				  params->volname, params->volnamesz);
121 	if (IS_ERR(vldb))
122 		return ERR_CAST(vldb);
123 
124 	if (test_bit(AFS_VLDB_QUERY_ERROR, &vldb->flags)) {
125 		volume = ERR_PTR(vldb->error);
126 		goto error;
127 	}
128 
129 	/* Make the final decision on the type we want */
130 	volume = ERR_PTR(-ENOMEDIUM);
131 	if (params->force) {
132 		if (!(vldb->flags & type_mask))
133 			goto error;
134 	} else if (test_bit(AFS_VLDB_HAS_RO, &vldb->flags)) {
135 		params->type = AFSVL_ROVOL;
136 	} else if (test_bit(AFS_VLDB_HAS_RW, &vldb->flags)) {
137 		params->type = AFSVL_RWVOL;
138 	} else {
139 		goto error;
140 	}
141 
142 	type_mask = 1UL << params->type;
143 	volume = afs_alloc_volume(params, vldb, type_mask);
144 
145 error:
146 	kfree(vldb);
147 	return volume;
148 }
149 
150 /*
151  * Destroy a volume record
152  */
153 static void afs_destroy_volume(struct afs_net *net, struct afs_volume *volume)
154 {
155 	_enter("%p", volume);
156 
157 #ifdef CONFIG_AFS_FSCACHE
158 	ASSERTCMP(volume->cache, ==, NULL);
159 #endif
160 
161 	afs_put_serverlist(net, volume->servers);
162 	afs_put_cell(net, volume->cell);
163 	kfree(volume);
164 
165 	_leave(" [destroyed]");
166 }
167 
168 /*
169  * Drop a reference on a volume record.
170  */
171 void afs_put_volume(struct afs_cell *cell, struct afs_volume *volume)
172 {
173 	if (volume) {
174 		_enter("%s", volume->name);
175 
176 		if (atomic_dec_and_test(&volume->usage))
177 			afs_destroy_volume(cell->net, volume);
178 	}
179 }
180 
181 /*
182  * Activate a volume.
183  */
184 void afs_activate_volume(struct afs_volume *volume)
185 {
186 #ifdef CONFIG_AFS_FSCACHE
187 	volume->cache = fscache_acquire_cookie(volume->cell->cache,
188 					       &afs_volume_cache_index_def,
189 					       &volume->vid, sizeof(volume->vid),
190 					       NULL, 0,
191 					       volume, 0, true);
192 #endif
193 
194 	write_lock(&volume->cell->proc_lock);
195 	list_add_tail(&volume->proc_link, &volume->cell->proc_volumes);
196 	write_unlock(&volume->cell->proc_lock);
197 }
198 
199 /*
200  * Deactivate a volume.
201  */
202 void afs_deactivate_volume(struct afs_volume *volume)
203 {
204 	_enter("%s", volume->name);
205 
206 	write_lock(&volume->cell->proc_lock);
207 	list_del_init(&volume->proc_link);
208 	write_unlock(&volume->cell->proc_lock);
209 
210 #ifdef CONFIG_AFS_FSCACHE
211 	fscache_relinquish_cookie(volume->cache, NULL,
212 				  test_bit(AFS_VOLUME_DELETED, &volume->flags));
213 	volume->cache = NULL;
214 #endif
215 
216 	_leave("");
217 }
218 
219 /*
220  * Query the VL service to update the volume status.
221  */
222 static int afs_update_volume_status(struct afs_volume *volume, struct key *key)
223 {
224 	struct afs_server_list *new, *old, *discard;
225 	struct afs_vldb_entry *vldb;
226 	char idbuf[16];
227 	int ret, idsz;
228 
229 	_enter("");
230 
231 	/* We look up an ID by passing it as a decimal string in the
232 	 * operation's name parameter.
233 	 */
234 	idsz = sprintf(idbuf, "%llu", volume->vid);
235 
236 	vldb = afs_vl_lookup_vldb(volume->cell, key, idbuf, idsz);
237 	if (IS_ERR(vldb)) {
238 		ret = PTR_ERR(vldb);
239 		goto error;
240 	}
241 
242 	/* See if the volume got renamed. */
243 	if (vldb->name_len != volume->name_len ||
244 	    memcmp(vldb->name, volume->name, vldb->name_len) != 0) {
245 		/* TODO: Use RCU'd string. */
246 		memcpy(volume->name, vldb->name, AFS_MAXVOLNAME);
247 		volume->name_len = vldb->name_len;
248 	}
249 
250 	/* See if the volume's server list got updated. */
251 	new = afs_alloc_server_list(volume->cell, key,
252 				    vldb, (1 << volume->type));
253 	if (IS_ERR(new)) {
254 		ret = PTR_ERR(new);
255 		goto error_vldb;
256 	}
257 
258 	write_lock(&volume->servers_lock);
259 
260 	discard = new;
261 	old = volume->servers;
262 	if (afs_annotate_server_list(new, old)) {
263 		new->seq = volume->servers_seq + 1;
264 		volume->servers = new;
265 		smp_wmb();
266 		volume->servers_seq++;
267 		discard = old;
268 	}
269 
270 	volume->update_at = ktime_get_real_seconds() + afs_volume_record_life;
271 	clear_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
272 	write_unlock(&volume->servers_lock);
273 	ret = 0;
274 
275 	afs_put_serverlist(volume->cell->net, discard);
276 error_vldb:
277 	kfree(vldb);
278 error:
279 	_leave(" = %d", ret);
280 	return ret;
281 }
282 
283 /*
284  * Make sure the volume record is up to date.
285  */
286 int afs_check_volume_status(struct afs_volume *volume, struct key *key)
287 {
288 	time64_t now = ktime_get_real_seconds();
289 	int ret, retries = 0;
290 
291 	_enter("");
292 
293 	if (volume->update_at <= now)
294 		set_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
295 
296 retry:
297 	if (!test_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags) &&
298 	    !test_bit(AFS_VOLUME_WAIT, &volume->flags)) {
299 		_leave(" = 0");
300 		return 0;
301 	}
302 
303 	if (!test_and_set_bit_lock(AFS_VOLUME_UPDATING, &volume->flags)) {
304 		ret = afs_update_volume_status(volume, key);
305 		clear_bit_unlock(AFS_VOLUME_WAIT, &volume->flags);
306 		clear_bit_unlock(AFS_VOLUME_UPDATING, &volume->flags);
307 		wake_up_bit(&volume->flags, AFS_VOLUME_WAIT);
308 		_leave(" = %d", ret);
309 		return ret;
310 	}
311 
312 	if (!test_bit(AFS_VOLUME_WAIT, &volume->flags)) {
313 		_leave(" = 0 [no wait]");
314 		return 0;
315 	}
316 
317 	ret = wait_on_bit(&volume->flags, AFS_VOLUME_WAIT, TASK_INTERRUPTIBLE);
318 	if (ret == -ERESTARTSYS) {
319 		_leave(" = %d", ret);
320 		return ret;
321 	}
322 
323 	retries++;
324 	if (retries == 4) {
325 		_leave(" = -ESTALE");
326 		return -ESTALE;
327 	}
328 	goto retry;
329 }
330