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/slab.h>
14 #include "internal.h"
15 
16 unsigned __read_mostly afs_volume_gc_delay = 10;
17 unsigned __read_mostly afs_volume_record_life = 60 * 60;
18 
19 static const char *const afs_voltypes[] = { "R/W", "R/O", "BAK" };
20 
21 /*
22  * Allocate a volume record and load it up from a vldb record.
23  */
afs_alloc_volume(struct afs_mount_params * params,struct afs_vldb_entry * vldb,unsigned long type_mask)24 static struct afs_volume *afs_alloc_volume(struct afs_mount_params *params,
25 					   struct afs_vldb_entry *vldb,
26 					   unsigned long type_mask)
27 {
28 	struct afs_server_list *slist;
29 	struct afs_volume *volume;
30 	int ret = -ENOMEM, nr_servers = 0, i;
31 
32 	for (i = 0; i < vldb->nr_servers; i++)
33 		if (vldb->fs_mask[i] & type_mask)
34 			nr_servers++;
35 
36 	volume = kzalloc(sizeof(struct afs_volume), GFP_KERNEL);
37 	if (!volume)
38 		goto error_0;
39 
40 	volume->vid		= vldb->vid[params->type];
41 	volume->update_at	= ktime_get_real_seconds() + afs_volume_record_life;
42 	volume->cell		= afs_get_cell(params->cell);
43 	volume->type		= params->type;
44 	volume->type_force	= params->force;
45 	volume->name_len	= vldb->name_len;
46 
47 	atomic_set(&volume->usage, 1);
48 	INIT_LIST_HEAD(&volume->proc_link);
49 	rwlock_init(&volume->servers_lock);
50 	memcpy(volume->name, vldb->name, vldb->name_len + 1);
51 
52 	slist = afs_alloc_server_list(params->cell, params->key, vldb, type_mask);
53 	if (IS_ERR(slist)) {
54 		ret = PTR_ERR(slist);
55 		goto error_1;
56 	}
57 
58 	refcount_set(&slist->usage, 1);
59 	volume->servers = slist;
60 	return volume;
61 
62 error_1:
63 	afs_put_cell(params->net, volume->cell);
64 	kfree(volume);
65 error_0:
66 	return ERR_PTR(ret);
67 }
68 
69 /*
70  * Look up a VLDB record for a volume.
71  */
afs_vl_lookup_vldb(struct afs_cell * cell,struct key * key,const char * volname,size_t volnamesz)72 static struct afs_vldb_entry *afs_vl_lookup_vldb(struct afs_cell *cell,
73 						 struct key *key,
74 						 const char *volname,
75 						 size_t volnamesz)
76 {
77 	struct afs_addr_cursor ac;
78 	struct afs_vldb_entry *vldb;
79 	int ret;
80 
81 	ret = afs_set_vl_cursor(&ac, cell);
82 	if (ret < 0)
83 		return ERR_PTR(ret);
84 
85 	while (afs_iterate_addresses(&ac)) {
86 		if (!test_bit(ac.index, &ac.alist->probed)) {
87 			ret = afs_vl_get_capabilities(cell->net, &ac, key);
88 			switch (ret) {
89 			case VL_SERVICE:
90 				clear_bit(ac.index, &ac.alist->yfs);
91 				set_bit(ac.index, &ac.alist->probed);
92 				ac.addr->srx_service = ret;
93 				break;
94 			case YFS_VL_SERVICE:
95 				set_bit(ac.index, &ac.alist->yfs);
96 				set_bit(ac.index, &ac.alist->probed);
97 				ac.addr->srx_service = ret;
98 				break;
99 			}
100 		}
101 
102 		vldb = afs_vl_get_entry_by_name_u(cell->net, &ac, key,
103 						  volname, volnamesz);
104 		switch (ac.error) {
105 		case 0:
106 			afs_end_cursor(&ac);
107 			return vldb;
108 		case -ECONNABORTED:
109 			ac.error = afs_abort_to_error(ac.abort_code);
110 			goto error;
111 		case -ENOMEM:
112 		case -ENONET:
113 			goto error;
114 		case -ENETUNREACH:
115 		case -EHOSTUNREACH:
116 		case -ECONNREFUSED:
117 			break;
118 		default:
119 			ac.error = -EIO;
120 			goto error;
121 		}
122 	}
123 
124 error:
125 	return ERR_PTR(afs_end_cursor(&ac));
126 }
127 
128 /*
129  * Look up a volume in the VL server and create a candidate volume record for
130  * it.
131  *
132  * The volume name can be one of the following:
133  *	"%[cell:]volume[.]"		R/W volume
134  *	"#[cell:]volume[.]"		R/O or R/W volume (rwparent=0),
135  *					 or R/W (rwparent=1) volume
136  *	"%[cell:]volume.readonly"	R/O volume
137  *	"#[cell:]volume.readonly"	R/O volume
138  *	"%[cell:]volume.backup"		Backup volume
139  *	"#[cell:]volume.backup"		Backup volume
140  *
141  * The cell name is optional, and defaults to the current cell.
142  *
143  * See "The Rules of Mount Point Traversal" in Chapter 5 of the AFS SysAdmin
144  * Guide
145  * - Rule 1: Explicit type suffix forces access of that type or nothing
146  *           (no suffix, then use Rule 2 & 3)
147  * - Rule 2: If parent volume is R/O, then mount R/O volume by preference, R/W
148  *           if not available
149  * - Rule 3: If parent volume is R/W, then only mount R/W volume unless
150  *           explicitly told otherwise
151  */
afs_create_volume(struct afs_mount_params * params)152 struct afs_volume *afs_create_volume(struct afs_mount_params *params)
153 {
154 	struct afs_vldb_entry *vldb;
155 	struct afs_volume *volume;
156 	unsigned long type_mask = 1UL << params->type;
157 
158 	vldb = afs_vl_lookup_vldb(params->cell, params->key,
159 				  params->volname, params->volnamesz);
160 	if (IS_ERR(vldb))
161 		return ERR_CAST(vldb);
162 
163 	if (test_bit(AFS_VLDB_QUERY_ERROR, &vldb->flags)) {
164 		volume = ERR_PTR(vldb->error);
165 		goto error;
166 	}
167 
168 	/* Make the final decision on the type we want */
169 	volume = ERR_PTR(-ENOMEDIUM);
170 	if (params->force) {
171 		if (!(vldb->flags & type_mask))
172 			goto error;
173 	} else if (test_bit(AFS_VLDB_HAS_RO, &vldb->flags)) {
174 		params->type = AFSVL_ROVOL;
175 	} else if (test_bit(AFS_VLDB_HAS_RW, &vldb->flags)) {
176 		params->type = AFSVL_RWVOL;
177 	} else {
178 		goto error;
179 	}
180 
181 	type_mask = 1UL << params->type;
182 	volume = afs_alloc_volume(params, vldb, type_mask);
183 
184 error:
185 	kfree(vldb);
186 	return volume;
187 }
188 
189 /*
190  * Destroy a volume record
191  */
afs_destroy_volume(struct afs_net * net,struct afs_volume * volume)192 static void afs_destroy_volume(struct afs_net *net, struct afs_volume *volume)
193 {
194 	_enter("%p", volume);
195 
196 #ifdef CONFIG_AFS_FSCACHE
197 	ASSERTCMP(volume->cache, ==, NULL);
198 #endif
199 
200 	afs_put_serverlist(net, volume->servers);
201 	afs_put_cell(net, volume->cell);
202 	kfree(volume);
203 
204 	_leave(" [destroyed]");
205 }
206 
207 /*
208  * Drop a reference on a volume record.
209  */
afs_put_volume(struct afs_cell * cell,struct afs_volume * volume)210 void afs_put_volume(struct afs_cell *cell, struct afs_volume *volume)
211 {
212 	if (volume) {
213 		_enter("%s", volume->name);
214 
215 		if (atomic_dec_and_test(&volume->usage))
216 			afs_destroy_volume(cell->net, volume);
217 	}
218 }
219 
220 /*
221  * Activate a volume.
222  */
afs_activate_volume(struct afs_volume * volume)223 void afs_activate_volume(struct afs_volume *volume)
224 {
225 #ifdef CONFIG_AFS_FSCACHE
226 	volume->cache = fscache_acquire_cookie(volume->cell->cache,
227 					       &afs_volume_cache_index_def,
228 					       &volume->vid, sizeof(volume->vid),
229 					       NULL, 0,
230 					       volume, 0, true);
231 #endif
232 
233 	write_lock(&volume->cell->proc_lock);
234 	list_add_tail(&volume->proc_link, &volume->cell->proc_volumes);
235 	write_unlock(&volume->cell->proc_lock);
236 }
237 
238 /*
239  * Deactivate a volume.
240  */
afs_deactivate_volume(struct afs_volume * volume)241 void afs_deactivate_volume(struct afs_volume *volume)
242 {
243 	_enter("%s", volume->name);
244 
245 	write_lock(&volume->cell->proc_lock);
246 	list_del_init(&volume->proc_link);
247 	write_unlock(&volume->cell->proc_lock);
248 
249 #ifdef CONFIG_AFS_FSCACHE
250 	fscache_relinquish_cookie(volume->cache, NULL,
251 				  test_bit(AFS_VOLUME_DELETED, &volume->flags));
252 	volume->cache = NULL;
253 #endif
254 
255 	_leave("");
256 }
257 
258 /*
259  * Query the VL service to update the volume status.
260  */
afs_update_volume_status(struct afs_volume * volume,struct key * key)261 static int afs_update_volume_status(struct afs_volume *volume, struct key *key)
262 {
263 	struct afs_server_list *new, *old, *discard;
264 	struct afs_vldb_entry *vldb;
265 	char idbuf[16];
266 	int ret, idsz;
267 
268 	_enter("");
269 
270 	/* We look up an ID by passing it as a decimal string in the
271 	 * operation's name parameter.
272 	 */
273 	idsz = sprintf(idbuf, "%u", volume->vid);
274 
275 	vldb = afs_vl_lookup_vldb(volume->cell, key, idbuf, idsz);
276 	if (IS_ERR(vldb)) {
277 		ret = PTR_ERR(vldb);
278 		goto error;
279 	}
280 
281 	/* See if the volume got renamed. */
282 	if (vldb->name_len != volume->name_len ||
283 	    memcmp(vldb->name, volume->name, vldb->name_len) != 0) {
284 		/* TODO: Use RCU'd string. */
285 		memcpy(volume->name, vldb->name, AFS_MAXVOLNAME);
286 		volume->name_len = vldb->name_len;
287 	}
288 
289 	/* See if the volume's server list got updated. */
290 	new = afs_alloc_server_list(volume->cell, key,
291 				    vldb, (1 << volume->type));
292 	if (IS_ERR(new)) {
293 		ret = PTR_ERR(new);
294 		goto error_vldb;
295 	}
296 
297 	write_lock(&volume->servers_lock);
298 
299 	discard = new;
300 	old = volume->servers;
301 	if (afs_annotate_server_list(new, old)) {
302 		new->seq = volume->servers_seq + 1;
303 		volume->servers = new;
304 		smp_wmb();
305 		volume->servers_seq++;
306 		discard = old;
307 	}
308 
309 	volume->update_at = ktime_get_real_seconds() + afs_volume_record_life;
310 	clear_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
311 	write_unlock(&volume->servers_lock);
312 	ret = 0;
313 
314 	afs_put_serverlist(volume->cell->net, discard);
315 error_vldb:
316 	kfree(vldb);
317 error:
318 	_leave(" = %d", ret);
319 	return ret;
320 }
321 
322 /*
323  * Make sure the volume record is up to date.
324  */
afs_check_volume_status(struct afs_volume * volume,struct key * key)325 int afs_check_volume_status(struct afs_volume *volume, struct key *key)
326 {
327 	time64_t now = ktime_get_real_seconds();
328 	int ret, retries = 0;
329 
330 	_enter("");
331 
332 	if (volume->update_at <= now)
333 		set_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
334 
335 retry:
336 	if (!test_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags) &&
337 	    !test_bit(AFS_VOLUME_WAIT, &volume->flags)) {
338 		_leave(" = 0");
339 		return 0;
340 	}
341 
342 	if (!test_and_set_bit_lock(AFS_VOLUME_UPDATING, &volume->flags)) {
343 		ret = afs_update_volume_status(volume, key);
344 		clear_bit_unlock(AFS_VOLUME_WAIT, &volume->flags);
345 		clear_bit_unlock(AFS_VOLUME_UPDATING, &volume->flags);
346 		wake_up_bit(&volume->flags, AFS_VOLUME_WAIT);
347 		_leave(" = %d", ret);
348 		return ret;
349 	}
350 
351 	if (!test_bit(AFS_VOLUME_WAIT, &volume->flags)) {
352 		_leave(" = 0 [no wait]");
353 		return 0;
354 	}
355 
356 	ret = wait_on_bit(&volume->flags, AFS_VOLUME_WAIT, TASK_INTERRUPTIBLE);
357 	if (ret == -ERESTARTSYS) {
358 		_leave(" = %d", ret);
359 		return ret;
360 	}
361 
362 	retries++;
363 	if (retries == 4) {
364 		_leave(" = -ESTALE");
365 		return -ESTALE;
366 	}
367 	goto retry;
368 }
369