xref: /openbmc/linux/fs/afs/volume.c (revision c4e78957)
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  */
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 			volume = afs_get_volume(p, afs_volume_trace_get_cell_insert);
36 			goto found;
37 		}
38 	}
39 
40 	rb_link_node_rcu(&volume->cell_node, parent, pp);
41 	rb_insert_color(&volume->cell_node, &cell->volumes);
42 	hlist_add_head_rcu(&volume->proc_link, &cell->proc_volumes);
43 
44 found:
45 	write_sequnlock(&cell->volume_lock);
46 	return volume;
47 
48 }
49 
50 static void afs_remove_volume_from_cell(struct afs_volume *volume)
51 {
52 	struct afs_cell *cell = volume->cell;
53 
54 	if (!hlist_unhashed(&volume->proc_link)) {
55 		trace_afs_volume(volume->vid, atomic_read(&volume->usage),
56 				 afs_volume_trace_remove);
57 		write_seqlock(&cell->volume_lock);
58 		hlist_del_rcu(&volume->proc_link);
59 		rb_erase(&volume->cell_node, &cell->volumes);
60 		write_sequnlock(&cell->volume_lock);
61 	}
62 }
63 
64 /*
65  * Allocate a volume record and load it up from a vldb record.
66  */
67 static struct afs_volume *afs_alloc_volume(struct afs_fs_context *params,
68 					   struct afs_vldb_entry *vldb,
69 					   unsigned long type_mask)
70 {
71 	struct afs_server_list *slist;
72 	struct afs_volume *volume;
73 	int ret = -ENOMEM, nr_servers = 0, i;
74 
75 	for (i = 0; i < vldb->nr_servers; i++)
76 		if (vldb->fs_mask[i] & type_mask)
77 			nr_servers++;
78 
79 	volume = kzalloc(sizeof(struct afs_volume), GFP_KERNEL);
80 	if (!volume)
81 		goto error_0;
82 
83 	volume->vid		= vldb->vid[params->type];
84 	volume->update_at	= ktime_get_real_seconds() + afs_volume_record_life;
85 	volume->cell		= afs_get_cell(params->cell, afs_cell_trace_get_vol);
86 	volume->type		= params->type;
87 	volume->type_force	= params->force;
88 	volume->name_len	= vldb->name_len;
89 
90 	atomic_set(&volume->usage, 1);
91 	INIT_HLIST_NODE(&volume->proc_link);
92 	rwlock_init(&volume->servers_lock);
93 	rwlock_init(&volume->cb_v_break_lock);
94 	memcpy(volume->name, vldb->name, vldb->name_len + 1);
95 
96 	slist = afs_alloc_server_list(params->cell, params->key, vldb, type_mask);
97 	if (IS_ERR(slist)) {
98 		ret = PTR_ERR(slist);
99 		goto error_1;
100 	}
101 
102 	refcount_set(&slist->usage, 1);
103 	rcu_assign_pointer(volume->servers, slist);
104 	trace_afs_volume(volume->vid, 1, afs_volume_trace_alloc);
105 	return volume;
106 
107 error_1:
108 	afs_put_cell(volume->cell, afs_cell_trace_put_vol);
109 	kfree(volume);
110 error_0:
111 	return ERR_PTR(ret);
112 }
113 
114 /*
115  * Look up or allocate a volume record.
116  */
117 static struct afs_volume *afs_lookup_volume(struct afs_fs_context *params,
118 					    struct afs_vldb_entry *vldb,
119 					    unsigned long type_mask)
120 {
121 	struct afs_volume *candidate, *volume;
122 
123 	candidate = afs_alloc_volume(params, vldb, type_mask);
124 	if (IS_ERR(candidate))
125 		return candidate;
126 
127 	volume = afs_insert_volume_into_cell(params->cell, candidate);
128 	if (volume != candidate)
129 		afs_put_volume(params->net, candidate, afs_volume_trace_put_cell_dup);
130 	return volume;
131 }
132 
133 /*
134  * Look up a VLDB record for a volume.
135  */
136 static struct afs_vldb_entry *afs_vl_lookup_vldb(struct afs_cell *cell,
137 						 struct key *key,
138 						 const char *volname,
139 						 size_t volnamesz)
140 {
141 	struct afs_vldb_entry *vldb = ERR_PTR(-EDESTADDRREQ);
142 	struct afs_vl_cursor vc;
143 	int ret;
144 
145 	if (!afs_begin_vlserver_operation(&vc, cell, key))
146 		return ERR_PTR(-ERESTARTSYS);
147 
148 	while (afs_select_vlserver(&vc)) {
149 		vldb = afs_vl_get_entry_by_name_u(&vc, volname, volnamesz);
150 	}
151 
152 	ret = afs_end_vlserver_operation(&vc);
153 	return ret < 0 ? ERR_PTR(ret) : vldb;
154 }
155 
156 /*
157  * Look up a volume in the VL server and create a candidate volume record for
158  * it.
159  *
160  * The volume name can be one of the following:
161  *	"%[cell:]volume[.]"		R/W volume
162  *	"#[cell:]volume[.]"		R/O or R/W volume (rwparent=0),
163  *					 or R/W (rwparent=1) volume
164  *	"%[cell:]volume.readonly"	R/O volume
165  *	"#[cell:]volume.readonly"	R/O volume
166  *	"%[cell:]volume.backup"		Backup volume
167  *	"#[cell:]volume.backup"		Backup volume
168  *
169  * The cell name is optional, and defaults to the current cell.
170  *
171  * See "The Rules of Mount Point Traversal" in Chapter 5 of the AFS SysAdmin
172  * Guide
173  * - Rule 1: Explicit type suffix forces access of that type or nothing
174  *           (no suffix, then use Rule 2 & 3)
175  * - Rule 2: If parent volume is R/O, then mount R/O volume by preference, R/W
176  *           if not available
177  * - Rule 3: If parent volume is R/W, then only mount R/W volume unless
178  *           explicitly told otherwise
179  */
180 struct afs_volume *afs_create_volume(struct afs_fs_context *params)
181 {
182 	struct afs_vldb_entry *vldb;
183 	struct afs_volume *volume;
184 	unsigned long type_mask = 1UL << params->type;
185 
186 	vldb = afs_vl_lookup_vldb(params->cell, params->key,
187 				  params->volname, params->volnamesz);
188 	if (IS_ERR(vldb))
189 		return ERR_CAST(vldb);
190 
191 	if (test_bit(AFS_VLDB_QUERY_ERROR, &vldb->flags)) {
192 		volume = ERR_PTR(vldb->error);
193 		goto error;
194 	}
195 
196 	/* Make the final decision on the type we want */
197 	volume = ERR_PTR(-ENOMEDIUM);
198 	if (params->force) {
199 		if (!(vldb->flags & type_mask))
200 			goto error;
201 	} else if (test_bit(AFS_VLDB_HAS_RO, &vldb->flags)) {
202 		params->type = AFSVL_ROVOL;
203 	} else if (test_bit(AFS_VLDB_HAS_RW, &vldb->flags)) {
204 		params->type = AFSVL_RWVOL;
205 	} else {
206 		goto error;
207 	}
208 
209 	type_mask = 1UL << params->type;
210 	volume = afs_lookup_volume(params, vldb, type_mask);
211 
212 error:
213 	kfree(vldb);
214 	return volume;
215 }
216 
217 /*
218  * Destroy a volume record
219  */
220 static void afs_destroy_volume(struct afs_net *net, struct afs_volume *volume)
221 {
222 	_enter("%p", volume);
223 
224 #ifdef CONFIG_AFS_FSCACHE
225 	ASSERTCMP(volume->cache, ==, NULL);
226 #endif
227 
228 	afs_remove_volume_from_cell(volume);
229 	afs_put_serverlist(net, rcu_access_pointer(volume->servers));
230 	afs_put_cell(volume->cell, afs_cell_trace_put_vol);
231 	trace_afs_volume(volume->vid, atomic_read(&volume->usage),
232 			 afs_volume_trace_free);
233 	kfree_rcu(volume, rcu);
234 
235 	_leave(" [destroyed]");
236 }
237 
238 /*
239  * Get a reference on a volume record.
240  */
241 struct afs_volume *afs_get_volume(struct afs_volume *volume,
242 				  enum afs_volume_trace reason)
243 {
244 	if (volume) {
245 		int u = atomic_inc_return(&volume->usage);
246 		trace_afs_volume(volume->vid, u, reason);
247 	}
248 	return volume;
249 }
250 
251 
252 /*
253  * Drop a reference on a volume record.
254  */
255 void afs_put_volume(struct afs_net *net, struct afs_volume *volume,
256 		    enum afs_volume_trace reason)
257 {
258 	if (volume) {
259 		afs_volid_t vid = volume->vid;
260 		int u = atomic_dec_return(&volume->usage);
261 		trace_afs_volume(vid, u, reason);
262 		if (u == 0)
263 			afs_destroy_volume(net, volume);
264 	}
265 }
266 
267 /*
268  * Activate a volume.
269  */
270 int afs_activate_volume(struct afs_volume *volume)
271 {
272 #ifdef CONFIG_AFS_FSCACHE
273 	struct fscache_volume *vcookie;
274 	char *name;
275 
276 	name = kasprintf(GFP_KERNEL, "afs,%s,%llx",
277 			 volume->cell->name, volume->vid);
278 	if (!name)
279 		return -ENOMEM;
280 
281 	vcookie = fscache_acquire_volume(name, NULL, NULL, 0);
282 	if (IS_ERR(vcookie)) {
283 		if (vcookie != ERR_PTR(-EBUSY)) {
284 			kfree(name);
285 			return PTR_ERR(vcookie);
286 		}
287 		pr_err("AFS: Cache volume key already in use (%s)\n", name);
288 		vcookie = NULL;
289 	}
290 	volume->cache = vcookie;
291 	kfree(name);
292 #endif
293 	return 0;
294 }
295 
296 /*
297  * Deactivate a volume.
298  */
299 void afs_deactivate_volume(struct afs_volume *volume)
300 {
301 	_enter("%s", volume->name);
302 
303 #ifdef CONFIG_AFS_FSCACHE
304 	fscache_relinquish_volume(volume->cache, NULL,
305 				  test_bit(AFS_VOLUME_DELETED, &volume->flags));
306 	volume->cache = NULL;
307 #endif
308 
309 	_leave("");
310 }
311 
312 /*
313  * Query the VL service to update the volume status.
314  */
315 static int afs_update_volume_status(struct afs_volume *volume, struct key *key)
316 {
317 	struct afs_server_list *new, *old, *discard;
318 	struct afs_vldb_entry *vldb;
319 	char idbuf[16];
320 	int ret, idsz;
321 
322 	_enter("");
323 
324 	/* We look up an ID by passing it as a decimal string in the
325 	 * operation's name parameter.
326 	 */
327 	idsz = sprintf(idbuf, "%llu", volume->vid);
328 
329 	vldb = afs_vl_lookup_vldb(volume->cell, key, idbuf, idsz);
330 	if (IS_ERR(vldb)) {
331 		ret = PTR_ERR(vldb);
332 		goto error;
333 	}
334 
335 	/* See if the volume got renamed. */
336 	if (vldb->name_len != volume->name_len ||
337 	    memcmp(vldb->name, volume->name, vldb->name_len) != 0) {
338 		/* TODO: Use RCU'd string. */
339 		memcpy(volume->name, vldb->name, AFS_MAXVOLNAME);
340 		volume->name_len = vldb->name_len;
341 	}
342 
343 	/* See if the volume's server list got updated. */
344 	new = afs_alloc_server_list(volume->cell, key,
345 				    vldb, (1 << volume->type));
346 	if (IS_ERR(new)) {
347 		ret = PTR_ERR(new);
348 		goto error_vldb;
349 	}
350 
351 	write_lock(&volume->servers_lock);
352 
353 	discard = new;
354 	old = rcu_dereference_protected(volume->servers,
355 					lockdep_is_held(&volume->servers_lock));
356 	if (afs_annotate_server_list(new, old)) {
357 		new->seq = volume->servers_seq + 1;
358 		rcu_assign_pointer(volume->servers, new);
359 		smp_wmb();
360 		volume->servers_seq++;
361 		discard = old;
362 	}
363 
364 	volume->update_at = ktime_get_real_seconds() + afs_volume_record_life;
365 	write_unlock(&volume->servers_lock);
366 	ret = 0;
367 
368 	afs_put_serverlist(volume->cell->net, discard);
369 error_vldb:
370 	kfree(vldb);
371 error:
372 	_leave(" = %d", ret);
373 	return ret;
374 }
375 
376 /*
377  * Make sure the volume record is up to date.
378  */
379 int afs_check_volume_status(struct afs_volume *volume, struct afs_operation *op)
380 {
381 	int ret, retries = 0;
382 
383 	_enter("");
384 
385 retry:
386 	if (test_bit(AFS_VOLUME_WAIT, &volume->flags))
387 		goto wait;
388 	if (volume->update_at <= ktime_get_real_seconds() ||
389 	    test_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags))
390 		goto update;
391 	_leave(" = 0");
392 	return 0;
393 
394 update:
395 	if (!test_and_set_bit_lock(AFS_VOLUME_UPDATING, &volume->flags)) {
396 		clear_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
397 		ret = afs_update_volume_status(volume, op->key);
398 		if (ret < 0)
399 			set_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
400 		clear_bit_unlock(AFS_VOLUME_WAIT, &volume->flags);
401 		clear_bit_unlock(AFS_VOLUME_UPDATING, &volume->flags);
402 		wake_up_bit(&volume->flags, AFS_VOLUME_WAIT);
403 		_leave(" = %d", ret);
404 		return ret;
405 	}
406 
407 wait:
408 	if (!test_bit(AFS_VOLUME_WAIT, &volume->flags)) {
409 		_leave(" = 0 [no wait]");
410 		return 0;
411 	}
412 
413 	ret = wait_on_bit(&volume->flags, AFS_VOLUME_WAIT,
414 			  (op->flags & AFS_OPERATION_UNINTR) ?
415 			  TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
416 	if (ret == -ERESTARTSYS) {
417 		_leave(" = %d", ret);
418 		return ret;
419 	}
420 
421 	retries++;
422 	if (retries == 4) {
423 		_leave(" = -ESTALE");
424 		return -ESTALE;
425 	}
426 	goto retry;
427 }
428