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 	rwlock_init(&volume->cb_v_break_lock);
51 	memcpy(volume->name, vldb->name, vldb->name_len + 1);
52 
53 	slist = afs_alloc_server_list(params->cell, params->key, vldb, type_mask);
54 	if (IS_ERR(slist)) {
55 		ret = PTR_ERR(slist);
56 		goto error_1;
57 	}
58 
59 	refcount_set(&slist->usage, 1);
60 	volume->servers = slist;
61 	return volume;
62 
63 error_1:
64 	afs_put_cell(params->net, volume->cell);
65 	kfree(volume);
66 error_0:
67 	return ERR_PTR(ret);
68 }
69 
70 /*
71  * Look up a VLDB record for a volume.
72  */
afs_vl_lookup_vldb(struct afs_cell * cell,struct key * key,const char * volname,size_t volnamesz)73 static struct afs_vldb_entry *afs_vl_lookup_vldb(struct afs_cell *cell,
74 						 struct key *key,
75 						 const char *volname,
76 						 size_t volnamesz)
77 {
78 	struct afs_addr_cursor ac;
79 	struct afs_vldb_entry *vldb;
80 	int ret;
81 
82 	ret = afs_set_vl_cursor(&ac, cell);
83 	if (ret < 0)
84 		return ERR_PTR(ret);
85 
86 	while (afs_iterate_addresses(&ac)) {
87 		if (!test_bit(ac.index, &ac.alist->probed)) {
88 			ret = afs_vl_get_capabilities(cell->net, &ac, key);
89 			switch (ret) {
90 			case VL_SERVICE:
91 				clear_bit(ac.index, &ac.alist->yfs);
92 				set_bit(ac.index, &ac.alist->probed);
93 				ac.addr->srx_service = ret;
94 				break;
95 			case YFS_VL_SERVICE:
96 				set_bit(ac.index, &ac.alist->yfs);
97 				set_bit(ac.index, &ac.alist->probed);
98 				ac.addr->srx_service = ret;
99 				break;
100 			}
101 		}
102 
103 		vldb = afs_vl_get_entry_by_name_u(cell->net, &ac, key,
104 						  volname, volnamesz);
105 		switch (ac.error) {
106 		case 0:
107 			afs_end_cursor(&ac);
108 			return vldb;
109 		case -ECONNABORTED:
110 			ac.error = afs_abort_to_error(ac.abort_code);
111 			goto error;
112 		case -ENOMEM:
113 		case -ENONET:
114 			goto error;
115 		case -ENETUNREACH:
116 		case -EHOSTUNREACH:
117 		case -ECONNREFUSED:
118 			break;
119 		default:
120 			ac.error = -EIO;
121 			goto error;
122 		}
123 	}
124 
125 error:
126 	return ERR_PTR(afs_end_cursor(&ac));
127 }
128 
129 /*
130  * Look up a volume in the VL server and create a candidate volume record for
131  * it.
132  *
133  * The volume name can be one of the following:
134  *	"%[cell:]volume[.]"		R/W volume
135  *	"#[cell:]volume[.]"		R/O or R/W volume (rwparent=0),
136  *					 or R/W (rwparent=1) volume
137  *	"%[cell:]volume.readonly"	R/O volume
138  *	"#[cell:]volume.readonly"	R/O volume
139  *	"%[cell:]volume.backup"		Backup volume
140  *	"#[cell:]volume.backup"		Backup volume
141  *
142  * The cell name is optional, and defaults to the current cell.
143  *
144  * See "The Rules of Mount Point Traversal" in Chapter 5 of the AFS SysAdmin
145  * Guide
146  * - Rule 1: Explicit type suffix forces access of that type or nothing
147  *           (no suffix, then use Rule 2 & 3)
148  * - Rule 2: If parent volume is R/O, then mount R/O volume by preference, R/W
149  *           if not available
150  * - Rule 3: If parent volume is R/W, then only mount R/W volume unless
151  *           explicitly told otherwise
152  */
afs_create_volume(struct afs_mount_params * params)153 struct afs_volume *afs_create_volume(struct afs_mount_params *params)
154 {
155 	struct afs_vldb_entry *vldb;
156 	struct afs_volume *volume;
157 	unsigned long type_mask = 1UL << params->type;
158 
159 	vldb = afs_vl_lookup_vldb(params->cell, params->key,
160 				  params->volname, params->volnamesz);
161 	if (IS_ERR(vldb))
162 		return ERR_CAST(vldb);
163 
164 	if (test_bit(AFS_VLDB_QUERY_ERROR, &vldb->flags)) {
165 		volume = ERR_PTR(vldb->error);
166 		goto error;
167 	}
168 
169 	/* Make the final decision on the type we want */
170 	volume = ERR_PTR(-ENOMEDIUM);
171 	if (params->force) {
172 		if (!(vldb->flags & type_mask))
173 			goto error;
174 	} else if (test_bit(AFS_VLDB_HAS_RO, &vldb->flags)) {
175 		params->type = AFSVL_ROVOL;
176 	} else if (test_bit(AFS_VLDB_HAS_RW, &vldb->flags)) {
177 		params->type = AFSVL_RWVOL;
178 	} else {
179 		goto error;
180 	}
181 
182 	type_mask = 1UL << params->type;
183 	volume = afs_alloc_volume(params, vldb, type_mask);
184 
185 error:
186 	kfree(vldb);
187 	return volume;
188 }
189 
190 /*
191  * Destroy a volume record
192  */
afs_destroy_volume(struct afs_net * net,struct afs_volume * volume)193 static void afs_destroy_volume(struct afs_net *net, struct afs_volume *volume)
194 {
195 	_enter("%p", volume);
196 
197 #ifdef CONFIG_AFS_FSCACHE
198 	ASSERTCMP(volume->cache, ==, NULL);
199 #endif
200 
201 	afs_put_serverlist(net, volume->servers);
202 	afs_put_cell(net, volume->cell);
203 	kfree(volume);
204 
205 	_leave(" [destroyed]");
206 }
207 
208 /*
209  * Drop a reference on a volume record.
210  */
afs_put_volume(struct afs_cell * cell,struct afs_volume * volume)211 void afs_put_volume(struct afs_cell *cell, struct afs_volume *volume)
212 {
213 	if (volume) {
214 		_enter("%s", volume->name);
215 
216 		if (atomic_dec_and_test(&volume->usage))
217 			afs_destroy_volume(cell->net, volume);
218 	}
219 }
220 
221 /*
222  * Activate a volume.
223  */
afs_activate_volume(struct afs_volume * volume)224 void afs_activate_volume(struct afs_volume *volume)
225 {
226 #ifdef CONFIG_AFS_FSCACHE
227 	volume->cache = fscache_acquire_cookie(volume->cell->cache,
228 					       &afs_volume_cache_index_def,
229 					       &volume->vid, sizeof(volume->vid),
230 					       NULL, 0,
231 					       volume, 0, true);
232 #endif
233 
234 	write_lock(&volume->cell->proc_lock);
235 	list_add_tail(&volume->proc_link, &volume->cell->proc_volumes);
236 	write_unlock(&volume->cell->proc_lock);
237 }
238 
239 /*
240  * Deactivate a volume.
241  */
afs_deactivate_volume(struct afs_volume * volume)242 void afs_deactivate_volume(struct afs_volume *volume)
243 {
244 	_enter("%s", volume->name);
245 
246 	write_lock(&volume->cell->proc_lock);
247 	list_del_init(&volume->proc_link);
248 	write_unlock(&volume->cell->proc_lock);
249 
250 #ifdef CONFIG_AFS_FSCACHE
251 	fscache_relinquish_cookie(volume->cache, NULL,
252 				  test_bit(AFS_VOLUME_DELETED, &volume->flags));
253 	volume->cache = NULL;
254 #endif
255 
256 	_leave("");
257 }
258 
259 /*
260  * Query the VL service to update the volume status.
261  */
afs_update_volume_status(struct afs_volume * volume,struct key * key)262 static int afs_update_volume_status(struct afs_volume *volume, struct key *key)
263 {
264 	struct afs_server_list *new, *old, *discard;
265 	struct afs_vldb_entry *vldb;
266 	char idbuf[16];
267 	int ret, idsz;
268 
269 	_enter("");
270 
271 	/* We look up an ID by passing it as a decimal string in the
272 	 * operation's name parameter.
273 	 */
274 	idsz = sprintf(idbuf, "%u", volume->vid);
275 
276 	vldb = afs_vl_lookup_vldb(volume->cell, key, idbuf, idsz);
277 	if (IS_ERR(vldb)) {
278 		ret = PTR_ERR(vldb);
279 		goto error;
280 	}
281 
282 	/* See if the volume got renamed. */
283 	if (vldb->name_len != volume->name_len ||
284 	    memcmp(vldb->name, volume->name, vldb->name_len) != 0) {
285 		/* TODO: Use RCU'd string. */
286 		memcpy(volume->name, vldb->name, AFS_MAXVOLNAME);
287 		volume->name_len = vldb->name_len;
288 	}
289 
290 	/* See if the volume's server list got updated. */
291 	new = afs_alloc_server_list(volume->cell, key,
292 				    vldb, (1 << volume->type));
293 	if (IS_ERR(new)) {
294 		ret = PTR_ERR(new);
295 		goto error_vldb;
296 	}
297 
298 	write_lock(&volume->servers_lock);
299 
300 	discard = new;
301 	old = volume->servers;
302 	if (afs_annotate_server_list(new, old)) {
303 		new->seq = volume->servers_seq + 1;
304 		volume->servers = new;
305 		smp_wmb();
306 		volume->servers_seq++;
307 		discard = old;
308 	}
309 
310 	volume->update_at = ktime_get_real_seconds() + afs_volume_record_life;
311 	clear_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
312 	write_unlock(&volume->servers_lock);
313 	ret = 0;
314 
315 	afs_put_serverlist(volume->cell->net, discard);
316 error_vldb:
317 	kfree(vldb);
318 error:
319 	_leave(" = %d", ret);
320 	return ret;
321 }
322 
323 /*
324  * Make sure the volume record is up to date.
325  */
afs_check_volume_status(struct afs_volume * volume,struct key * key)326 int afs_check_volume_status(struct afs_volume *volume, struct key *key)
327 {
328 	time64_t now = ktime_get_real_seconds();
329 	int ret, retries = 0;
330 
331 	_enter("");
332 
333 	if (volume->update_at <= now)
334 		set_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
335 
336 retry:
337 	if (!test_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags) &&
338 	    !test_bit(AFS_VOLUME_WAIT, &volume->flags)) {
339 		_leave(" = 0");
340 		return 0;
341 	}
342 
343 	if (!test_and_set_bit_lock(AFS_VOLUME_UPDATING, &volume->flags)) {
344 		ret = afs_update_volume_status(volume, key);
345 		clear_bit_unlock(AFS_VOLUME_WAIT, &volume->flags);
346 		clear_bit_unlock(AFS_VOLUME_UPDATING, &volume->flags);
347 		wake_up_bit(&volume->flags, AFS_VOLUME_WAIT);
348 		_leave(" = %d", ret);
349 		return ret;
350 	}
351 
352 	if (!test_bit(AFS_VOLUME_WAIT, &volume->flags)) {
353 		_leave(" = 0 [no wait]");
354 		return 0;
355 	}
356 
357 	ret = wait_on_bit(&volume->flags, AFS_VOLUME_WAIT, TASK_INTERRUPTIBLE);
358 	if (ret == -ERESTARTSYS) {
359 		_leave(" = %d", ret);
360 		return ret;
361 	}
362 
363 	retries++;
364 	if (retries == 4) {
365 		_leave(" = -ESTALE");
366 		return -ESTALE;
367 	}
368 	goto retry;
369 }
370