xref: /openbmc/linux/fs/afs/volume.c (revision e15a5365)
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 /*
16  * Insert a volume into a cell.  If there's an existing volume record, that is
17  * returned instead with a ref held.
18  */
19 static struct afs_volume *afs_insert_volume_into_cell(struct afs_cell *cell,
20 						      struct afs_volume *volume)
21 {
22 	struct afs_volume *p;
23 	struct rb_node *parent = NULL, **pp;
24 
25 	write_seqlock(&cell->volume_lock);
26 
27 	pp = &cell->volumes.rb_node;
28 	while (*pp) {
29 		parent = *pp;
30 		p = rb_entry(parent, struct afs_volume, cell_node);
31 		if (p->vid < volume->vid) {
32 			pp = &(*pp)->rb_left;
33 		} else if (p->vid > volume->vid) {
34 			pp = &(*pp)->rb_right;
35 		} else {
36 			volume = afs_get_volume(p, afs_volume_trace_get_cell_insert);
37 			goto found;
38 		}
39 	}
40 
41 	rb_link_node_rcu(&volume->cell_node, parent, pp);
42 	rb_insert_color(&volume->cell_node, &cell->volumes);
43 	hlist_add_head_rcu(&volume->proc_link, &cell->proc_volumes);
44 
45 found:
46 	write_sequnlock(&cell->volume_lock);
47 	return volume;
48 
49 }
50 
51 static void afs_remove_volume_from_cell(struct afs_volume *volume)
52 {
53 	struct afs_cell *cell = volume->cell;
54 
55 	if (!hlist_unhashed(&volume->proc_link)) {
56 		trace_afs_volume(volume->vid, atomic_read(&volume->usage),
57 				 afs_volume_trace_remove);
58 		write_seqlock(&cell->volume_lock);
59 		hlist_del_rcu(&volume->proc_link);
60 		rb_erase(&volume->cell_node, &cell->volumes);
61 		write_sequnlock(&cell->volume_lock);
62 	}
63 }
64 
65 /*
66  * Allocate a volume record and load it up from a vldb record.
67  */
68 static struct afs_volume *afs_alloc_volume(struct afs_fs_context *params,
69 					   struct afs_vldb_entry *vldb,
70 					   unsigned long type_mask)
71 {
72 	struct afs_server_list *slist;
73 	struct afs_volume *volume;
74 	int ret = -ENOMEM, nr_servers = 0, i;
75 
76 	for (i = 0; i < vldb->nr_servers; i++)
77 		if (vldb->fs_mask[i] & type_mask)
78 			nr_servers++;
79 
80 	volume = kzalloc(sizeof(struct afs_volume), GFP_KERNEL);
81 	if (!volume)
82 		goto error_0;
83 
84 	volume->vid		= vldb->vid[params->type];
85 	volume->update_at	= ktime_get_real_seconds() + afs_volume_record_life;
86 	volume->cell		= afs_get_cell(params->cell, afs_cell_trace_get_vol);
87 	volume->type		= params->type;
88 	volume->type_force	= params->force;
89 	volume->name_len	= vldb->name_len;
90 
91 	atomic_set(&volume->usage, 1);
92 	INIT_HLIST_NODE(&volume->proc_link);
93 	rwlock_init(&volume->servers_lock);
94 	rwlock_init(&volume->cb_v_break_lock);
95 	memcpy(volume->name, vldb->name, vldb->name_len + 1);
96 
97 	slist = afs_alloc_server_list(params->cell, params->key, vldb, type_mask);
98 	if (IS_ERR(slist)) {
99 		ret = PTR_ERR(slist);
100 		goto error_1;
101 	}
102 
103 	refcount_set(&slist->usage, 1);
104 	rcu_assign_pointer(volume->servers, slist);
105 	trace_afs_volume(volume->vid, 1, afs_volume_trace_alloc);
106 	return volume;
107 
108 error_1:
109 	afs_put_cell(volume->cell, afs_cell_trace_put_vol);
110 	kfree(volume);
111 error_0:
112 	return ERR_PTR(ret);
113 }
114 
115 /*
116  * Look up or allocate a volume record.
117  */
118 static struct afs_volume *afs_lookup_volume(struct afs_fs_context *params,
119 					    struct afs_vldb_entry *vldb,
120 					    unsigned long type_mask)
121 {
122 	struct afs_volume *candidate, *volume;
123 
124 	candidate = afs_alloc_volume(params, vldb, type_mask);
125 	if (IS_ERR(candidate))
126 		return candidate;
127 
128 	volume = afs_insert_volume_into_cell(params->cell, candidate);
129 	if (volume != candidate)
130 		afs_put_volume(params->net, candidate, afs_volume_trace_put_cell_dup);
131 	return volume;
132 }
133 
134 /*
135  * Look up a VLDB record for a volume.
136  */
137 static struct afs_vldb_entry *afs_vl_lookup_vldb(struct afs_cell *cell,
138 						 struct key *key,
139 						 const char *volname,
140 						 size_t volnamesz)
141 {
142 	struct afs_vldb_entry *vldb = ERR_PTR(-EDESTADDRREQ);
143 	struct afs_vl_cursor vc;
144 	int ret;
145 
146 	if (!afs_begin_vlserver_operation(&vc, cell, key))
147 		return ERR_PTR(-ERESTARTSYS);
148 
149 	while (afs_select_vlserver(&vc)) {
150 		vldb = afs_vl_get_entry_by_name_u(&vc, volname, volnamesz);
151 	}
152 
153 	ret = afs_end_vlserver_operation(&vc);
154 	return ret < 0 ? ERR_PTR(ret) : vldb;
155 }
156 
157 /*
158  * Look up a volume in the VL server and create a candidate volume record for
159  * it.
160  *
161  * The volume name can be one of the following:
162  *	"%[cell:]volume[.]"		R/W volume
163  *	"#[cell:]volume[.]"		R/O or R/W volume (rwparent=0),
164  *					 or R/W (rwparent=1) volume
165  *	"%[cell:]volume.readonly"	R/O volume
166  *	"#[cell:]volume.readonly"	R/O volume
167  *	"%[cell:]volume.backup"		Backup volume
168  *	"#[cell:]volume.backup"		Backup volume
169  *
170  * The cell name is optional, and defaults to the current cell.
171  *
172  * See "The Rules of Mount Point Traversal" in Chapter 5 of the AFS SysAdmin
173  * Guide
174  * - Rule 1: Explicit type suffix forces access of that type or nothing
175  *           (no suffix, then use Rule 2 & 3)
176  * - Rule 2: If parent volume is R/O, then mount R/O volume by preference, R/W
177  *           if not available
178  * - Rule 3: If parent volume is R/W, then only mount R/W volume unless
179  *           explicitly told otherwise
180  */
181 struct afs_volume *afs_create_volume(struct afs_fs_context *params)
182 {
183 	struct afs_vldb_entry *vldb;
184 	struct afs_volume *volume;
185 	unsigned long type_mask = 1UL << params->type;
186 
187 	vldb = afs_vl_lookup_vldb(params->cell, params->key,
188 				  params->volname, params->volnamesz);
189 	if (IS_ERR(vldb))
190 		return ERR_CAST(vldb);
191 
192 	if (test_bit(AFS_VLDB_QUERY_ERROR, &vldb->flags)) {
193 		volume = ERR_PTR(vldb->error);
194 		goto error;
195 	}
196 
197 	/* Make the final decision on the type we want */
198 	volume = ERR_PTR(-ENOMEDIUM);
199 	if (params->force) {
200 		if (!(vldb->flags & type_mask))
201 			goto error;
202 	} else if (test_bit(AFS_VLDB_HAS_RO, &vldb->flags)) {
203 		params->type = AFSVL_ROVOL;
204 	} else if (test_bit(AFS_VLDB_HAS_RW, &vldb->flags)) {
205 		params->type = AFSVL_RWVOL;
206 	} else {
207 		goto error;
208 	}
209 
210 	type_mask = 1UL << params->type;
211 	volume = afs_lookup_volume(params, vldb, type_mask);
212 
213 error:
214 	kfree(vldb);
215 	return volume;
216 }
217 
218 /*
219  * Destroy a volume record
220  */
221 static void afs_destroy_volume(struct afs_net *net, struct afs_volume *volume)
222 {
223 	_enter("%p", volume);
224 
225 #ifdef CONFIG_AFS_FSCACHE
226 	ASSERTCMP(volume->cache, ==, NULL);
227 #endif
228 
229 	afs_remove_volume_from_cell(volume);
230 	afs_put_serverlist(net, rcu_access_pointer(volume->servers));
231 	afs_put_cell(volume->cell, afs_cell_trace_put_vol);
232 	trace_afs_volume(volume->vid, atomic_read(&volume->usage),
233 			 afs_volume_trace_free);
234 	kfree_rcu(volume, rcu);
235 
236 	_leave(" [destroyed]");
237 }
238 
239 /*
240  * Get a reference on a volume record.
241  */
242 struct afs_volume *afs_get_volume(struct afs_volume *volume,
243 				  enum afs_volume_trace reason)
244 {
245 	if (volume) {
246 		int u = atomic_inc_return(&volume->usage);
247 		trace_afs_volume(volume->vid, u, reason);
248 	}
249 	return volume;
250 }
251 
252 
253 /*
254  * Drop a reference on a volume record.
255  */
256 void afs_put_volume(struct afs_net *net, struct afs_volume *volume,
257 		    enum afs_volume_trace reason)
258 {
259 	if (volume) {
260 		afs_volid_t vid = volume->vid;
261 		int u = atomic_dec_return(&volume->usage);
262 		trace_afs_volume(vid, u, reason);
263 		if (u == 0)
264 			afs_destroy_volume(net, volume);
265 	}
266 }
267 
268 /*
269  * Activate a volume.
270  */
271 void afs_activate_volume(struct afs_volume *volume)
272 {
273 #ifdef CONFIG_AFS_FSCACHE
274 	volume->cache = fscache_acquire_cookie(volume->cell->cache,
275 					       &afs_volume_cache_index_def,
276 					       &volume->vid, sizeof(volume->vid),
277 					       NULL, 0,
278 					       volume, 0, true);
279 #endif
280 }
281 
282 /*
283  * Deactivate a volume.
284  */
285 void afs_deactivate_volume(struct afs_volume *volume)
286 {
287 	_enter("%s", volume->name);
288 
289 #ifdef CONFIG_AFS_FSCACHE
290 	fscache_relinquish_cookie(volume->cache, NULL,
291 				  test_bit(AFS_VOLUME_DELETED, &volume->flags));
292 	volume->cache = NULL;
293 #endif
294 
295 	_leave("");
296 }
297 
298 /*
299  * Query the VL service to update the volume status.
300  */
301 static int afs_update_volume_status(struct afs_volume *volume, struct key *key)
302 {
303 	struct afs_server_list *new, *old, *discard;
304 	struct afs_vldb_entry *vldb;
305 	char idbuf[16];
306 	int ret, idsz;
307 
308 	_enter("");
309 
310 	/* We look up an ID by passing it as a decimal string in the
311 	 * operation's name parameter.
312 	 */
313 	idsz = sprintf(idbuf, "%llu", volume->vid);
314 
315 	vldb = afs_vl_lookup_vldb(volume->cell, key, idbuf, idsz);
316 	if (IS_ERR(vldb)) {
317 		ret = PTR_ERR(vldb);
318 		goto error;
319 	}
320 
321 	/* See if the volume got renamed. */
322 	if (vldb->name_len != volume->name_len ||
323 	    memcmp(vldb->name, volume->name, vldb->name_len) != 0) {
324 		/* TODO: Use RCU'd string. */
325 		memcpy(volume->name, vldb->name, AFS_MAXVOLNAME);
326 		volume->name_len = vldb->name_len;
327 	}
328 
329 	/* See if the volume's server list got updated. */
330 	new = afs_alloc_server_list(volume->cell, key,
331 				    vldb, (1 << volume->type));
332 	if (IS_ERR(new)) {
333 		ret = PTR_ERR(new);
334 		goto error_vldb;
335 	}
336 
337 	write_lock(&volume->servers_lock);
338 
339 	discard = new;
340 	old = rcu_dereference_protected(volume->servers,
341 					lockdep_is_held(&volume->servers_lock));
342 	if (afs_annotate_server_list(new, old)) {
343 		new->seq = volume->servers_seq + 1;
344 		rcu_assign_pointer(volume->servers, new);
345 		smp_wmb();
346 		volume->servers_seq++;
347 		discard = old;
348 	}
349 
350 	volume->update_at = ktime_get_real_seconds() + afs_volume_record_life;
351 	write_unlock(&volume->servers_lock);
352 	ret = 0;
353 
354 	afs_put_serverlist(volume->cell->net, discard);
355 error_vldb:
356 	kfree(vldb);
357 error:
358 	_leave(" = %d", ret);
359 	return ret;
360 }
361 
362 /*
363  * Make sure the volume record is up to date.
364  */
365 int afs_check_volume_status(struct afs_volume *volume, struct afs_operation *op)
366 {
367 	int ret, retries = 0;
368 
369 	_enter("");
370 
371 retry:
372 	if (test_bit(AFS_VOLUME_WAIT, &volume->flags))
373 		goto wait;
374 	if (volume->update_at <= ktime_get_real_seconds() ||
375 	    test_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags))
376 		goto update;
377 	_leave(" = 0");
378 	return 0;
379 
380 update:
381 	if (!test_and_set_bit_lock(AFS_VOLUME_UPDATING, &volume->flags)) {
382 		clear_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
383 		ret = afs_update_volume_status(volume, op->key);
384 		if (ret < 0)
385 			set_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
386 		clear_bit_unlock(AFS_VOLUME_WAIT, &volume->flags);
387 		clear_bit_unlock(AFS_VOLUME_UPDATING, &volume->flags);
388 		wake_up_bit(&volume->flags, AFS_VOLUME_WAIT);
389 		_leave(" = %d", ret);
390 		return ret;
391 	}
392 
393 wait:
394 	if (!test_bit(AFS_VOLUME_WAIT, &volume->flags)) {
395 		_leave(" = 0 [no wait]");
396 		return 0;
397 	}
398 
399 	ret = wait_on_bit(&volume->flags, AFS_VOLUME_WAIT,
400 			  (op->flags & AFS_OPERATION_UNINTR) ?
401 			  TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
402 	if (ret == -ERESTARTSYS) {
403 		_leave(" = %d", ret);
404 		return ret;
405 	}
406 
407 	retries++;
408 	if (retries == 4) {
409 		_leave(" = -ESTALE");
410 		return -ESTALE;
411 	}
412 	goto retry;
413 }
414