Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * linux/fs/super.c
4 : *
5 : * Copyright (C) 1991, 1992 Linus Torvalds
6 : *
7 : * super.c contains code to handle: - mount structures
8 : * - super-block tables
9 : * - filesystem drivers list
10 : * - mount system call
11 : * - umount system call
12 : * - ustat system call
13 : *
14 : * GK 2/5/95 - Changed to support mounting the root fs via NFS
15 : *
16 : * Added kerneld support: Jacques Gelinas and Bjorn Ekwall
17 : * Added change_root: Werner Almesberger & Hans Lermen, Feb '96
18 : * Added options to /proc/mounts:
19 : * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
20 : * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
21 : * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
22 : */
23 :
24 : #include <linux/export.h>
25 : #include <linux/slab.h>
26 : #include <linux/blkdev.h>
27 : #include <linux/mount.h>
28 : #include <linux/security.h>
29 : #include <linux/writeback.h> /* for the emergency remount stuff */
30 : #include <linux/idr.h>
31 : #include <linux/mutex.h>
32 : #include <linux/backing-dev.h>
33 : #include <linux/rculist_bl.h>
34 : #include <linux/fscrypt.h>
35 : #include <linux/fsnotify.h>
36 : #include <linux/lockdep.h>
37 : #include <linux/user_namespace.h>
38 : #include <linux/fs_context.h>
39 : #include <uapi/linux/mount.h>
40 : #include "internal.h"
41 :
42 : static int thaw_super_locked(struct super_block *sb, enum freeze_holder who);
43 :
44 : static LIST_HEAD(super_blocks);
45 : static DEFINE_SPINLOCK(sb_lock);
46 :
47 : static char *sb_writers_name[SB_FREEZE_LEVELS] = {
48 : "sb_writers",
49 : "sb_pagefaults",
50 : "sb_internal",
51 : };
52 :
53 : /*
54 : * One thing we have to be careful of with a per-sb shrinker is that we don't
55 : * drop the last active reference to the superblock from within the shrinker.
56 : * If that happens we could trigger unregistering the shrinker from within the
57 : * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
58 : * take a passive reference to the superblock to avoid this from occurring.
59 : */
60 383430 : static unsigned long super_cache_scan(struct shrinker *shrink,
61 : struct shrink_control *sc)
62 : {
63 383430 : struct super_block *sb;
64 383430 : long fs_objects = 0;
65 383430 : long total_objects;
66 383430 : long freed = 0;
67 383430 : long dentries;
68 383430 : long inodes;
69 :
70 383430 : sb = container_of(shrink, struct super_block, s_shrink);
71 :
72 : /*
73 : * Deadlock avoidance. We may hold various FS locks, and we don't want
74 : * to recurse into the FS that called us in clear_inode() and friends..
75 : */
76 383430 : if (!(sc->gfp_mask & __GFP_FS))
77 : return SHRINK_STOP;
78 :
79 383429 : if (!trylock_super(sb))
80 : return SHRINK_STOP;
81 :
82 383258 : if (sb->s_op->nr_cached_objects)
83 338052 : fs_objects = sb->s_op->nr_cached_objects(sb, sc);
84 :
85 383258 : inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
86 383258 : dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
87 383258 : total_objects = dentries + inodes + fs_objects + 1;
88 383258 : if (!total_objects)
89 0 : total_objects = 1;
90 :
91 : /* proportion the scan between the caches */
92 383258 : dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
93 383258 : inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
94 383258 : fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
95 :
96 : /*
97 : * prune the dcache first as the icache is pinned by it, then
98 : * prune the icache, followed by the filesystem specific caches
99 : *
100 : * Ensure that we always scan at least one object - memcg kmem
101 : * accounting uses this to fully empty the caches.
102 : */
103 383258 : sc->nr_to_scan = dentries + 1;
104 383258 : freed = prune_dcache_sb(sb, sc);
105 383258 : sc->nr_to_scan = inodes + 1;
106 383258 : freed += prune_icache_sb(sb, sc);
107 :
108 383258 : if (fs_objects) {
109 40085 : sc->nr_to_scan = fs_objects + 1;
110 40085 : freed += sb->s_op->free_cached_objects(sb, sc);
111 : }
112 :
113 383258 : up_read(&sb->s_umount);
114 383258 : return freed;
115 : }
116 :
117 601415 : static unsigned long super_cache_count(struct shrinker *shrink,
118 : struct shrink_control *sc)
119 : {
120 601415 : struct super_block *sb;
121 601415 : long total_objects = 0;
122 :
123 601415 : sb = container_of(shrink, struct super_block, s_shrink);
124 :
125 : /*
126 : * We don't call trylock_super() here as it is a scalability bottleneck,
127 : * so we're exposed to partial setup state. The shrinker rwsem does not
128 : * protect filesystem operations backing list_lru_shrink_count() or
129 : * s_op->nr_cached_objects(). Counts can change between
130 : * super_cache_count and super_cache_scan, so we really don't need locks
131 : * here.
132 : *
133 : * However, if we are currently mounting the superblock, the underlying
134 : * filesystem might be in a state of partial construction and hence it
135 : * is dangerous to access it. trylock_super() uses a SB_BORN check to
136 : * avoid this situation, so do the same here. The memory barrier is
137 : * matched with the one in mount_fs() as we don't hold locks here.
138 : */
139 601415 : if (!(sb->s_flags & SB_BORN))
140 : return 0;
141 601415 : smp_rmb();
142 :
143 601415 : if (sb->s_op && sb->s_op->nr_cached_objects)
144 108094 : total_objects = sb->s_op->nr_cached_objects(sb, sc);
145 :
146 601415 : total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
147 601415 : total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
148 :
149 601415 : if (!total_objects)
150 : return SHRINK_EMPTY;
151 :
152 237169 : total_objects = vfs_pressure_ratio(total_objects);
153 237169 : return total_objects;
154 : }
155 :
156 125808 : static void destroy_super_work(struct work_struct *work)
157 : {
158 125808 : struct super_block *s = container_of(work, struct super_block,
159 : destroy_work);
160 125808 : int i;
161 :
162 503265 : for (i = 0; i < SB_FREEZE_LEVELS; i++)
163 377444 : percpu_free_rwsem(&s->s_writers.rw_sem[i]);
164 125821 : kfree(s);
165 125819 : }
166 :
167 125810 : static void destroy_super_rcu(struct rcu_head *head)
168 : {
169 125810 : struct super_block *s = container_of(head, struct super_block, rcu);
170 125810 : INIT_WORK(&s->destroy_work, destroy_super_work);
171 125810 : schedule_work(&s->destroy_work);
172 125812 : }
173 :
174 : /* Free a superblock that has never been seen by anyone */
175 2977 : static void destroy_unused_super(struct super_block *s)
176 : {
177 2977 : if (!s)
178 : return;
179 0 : up_write(&s->s_umount);
180 0 : list_lru_destroy(&s->s_dentry_lru);
181 0 : list_lru_destroy(&s->s_inode_lru);
182 0 : security_sb_free(s);
183 0 : put_user_ns(s->s_user_ns);
184 0 : kfree(s->s_subtype);
185 0 : free_prealloced_shrinker(&s->s_shrink);
186 : /* no delays needed */
187 0 : destroy_super_work(&s->destroy_work);
188 : }
189 :
190 : /**
191 : * alloc_super - create new superblock
192 : * @type: filesystem type superblock should belong to
193 : * @flags: the mount flags
194 : * @user_ns: User namespace for the super_block
195 : *
196 : * Allocates and initializes a new &struct super_block. alloc_super()
197 : * returns a pointer new superblock or %NULL if allocation had failed.
198 : */
199 125768 : static struct super_block *alloc_super(struct file_system_type *type, int flags,
200 : struct user_namespace *user_ns)
201 : {
202 125768 : struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
203 125801 : static const struct super_operations default_op;
204 125801 : int i;
205 :
206 125801 : if (!s)
207 : return NULL;
208 :
209 125801 : INIT_LIST_HEAD(&s->s_mounts);
210 125801 : s->s_user_ns = get_user_ns(user_ns);
211 125791 : init_rwsem(&s->s_umount);
212 125758 : lockdep_set_class(&s->s_umount, &type->s_umount_key);
213 : /*
214 : * sget() can have s_umount recursion.
215 : *
216 : * When it cannot find a suitable sb, it allocates a new
217 : * one (this one), and tries again to find a suitable old
218 : * one.
219 : *
220 : * In case that succeeds, it will acquire the s_umount
221 : * lock of the old one. Since these are clearly distrinct
222 : * locks, and this object isn't exposed yet, there's no
223 : * risk of deadlocks.
224 : *
225 : * Annotate this by putting this lock in a different
226 : * subclass.
227 : */
228 125758 : down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
229 :
230 125758 : if (security_sb_alloc(s))
231 : goto fail;
232 :
233 628918 : for (i = 0; i < SB_FREEZE_LEVELS; i++) {
234 377352 : if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
235 377351 : sb_writers_name[i],
236 377351 : &type->s_writers_key[i]))
237 0 : goto fail;
238 : }
239 125809 : s->s_bdi = &noop_backing_dev_info;
240 125809 : s->s_flags = flags;
241 125809 : if (s->s_user_ns != &init_user_ns)
242 13 : s->s_iflags |= SB_I_NODEV;
243 125809 : INIT_HLIST_NODE(&s->s_instances);
244 125809 : INIT_HLIST_BL_HEAD(&s->s_roots);
245 125809 : mutex_init(&s->s_sync_lock);
246 125807 : INIT_LIST_HEAD(&s->s_inodes);
247 125807 : spin_lock_init(&s->s_inode_list_lock);
248 125809 : INIT_LIST_HEAD(&s->s_inodes_wb);
249 125809 : spin_lock_init(&s->s_inode_wblist_lock);
250 :
251 125808 : s->s_count = 1;
252 125808 : atomic_set(&s->s_active, 1);
253 125808 : mutex_init(&s->s_vfs_rename_mutex);
254 125809 : lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
255 125809 : init_rwsem(&s->s_dquot.dqio_sem);
256 125807 : s->s_maxbytes = MAX_NON_LFS;
257 125807 : s->s_op = &default_op;
258 125807 : s->s_time_gran = 1000000000;
259 125807 : s->s_time_min = TIME64_MIN;
260 125807 : s->s_time_max = TIME64_MAX;
261 :
262 125807 : s->s_shrink.seeks = DEFAULT_SEEKS;
263 125807 : s->s_shrink.scan_objects = super_cache_scan;
264 125807 : s->s_shrink.count_objects = super_cache_count;
265 125807 : s->s_shrink.batch = 1024;
266 125807 : s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
267 125807 : if (prealloc_shrinker(&s->s_shrink, "sb-%s", type->name))
268 0 : goto fail;
269 125810 : if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
270 0 : goto fail;
271 125809 : if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
272 0 : goto fail;
273 : return s;
274 :
275 0 : fail:
276 0 : destroy_unused_super(s);
277 0 : return NULL;
278 : }
279 :
280 : /* Superblock refcounting */
281 :
282 : /*
283 : * Drop a superblock's refcount. The caller must hold sb_lock.
284 : */
285 347779518 : static void __put_super(struct super_block *s)
286 : {
287 347779518 : if (!--s->s_count) {
288 125821 : list_del_init(&s->s_list);
289 125821 : WARN_ON(s->s_dentry_lru.node);
290 125821 : WARN_ON(s->s_inode_lru.node);
291 125821 : WARN_ON(!list_empty(&s->s_mounts));
292 125821 : security_sb_free(s);
293 125821 : put_user_ns(s->s_user_ns);
294 125821 : kfree(s->s_subtype);
295 125821 : call_rcu(&s->rcu, destroy_super_rcu);
296 : }
297 347779518 : }
298 :
299 : /**
300 : * put_super - drop a temporary reference to superblock
301 : * @sb: superblock in question
302 : *
303 : * Drops a temporary reference, frees superblock if there's no
304 : * references left.
305 : */
306 1860517 : void put_super(struct super_block *sb)
307 : {
308 1860517 : spin_lock(&sb_lock);
309 1861847 : __put_super(sb);
310 1861847 : spin_unlock(&sb_lock);
311 1861767 : }
312 :
313 :
314 : /**
315 : * deactivate_locked_super - drop an active reference to superblock
316 : * @s: superblock to deactivate
317 : *
318 : * Drops an active reference to superblock, converting it into a temporary
319 : * one if there is no other active references left. In that case we
320 : * tell fs driver to shut it down and drop the temporary reference we
321 : * had just acquired.
322 : *
323 : * Caller holds exclusive lock on superblock; that lock is released.
324 : */
325 196459 : void deactivate_locked_super(struct super_block *s)
326 : {
327 196459 : struct file_system_type *fs = s->s_type;
328 196459 : if (atomic_dec_and_test(&s->s_active)) {
329 125821 : unregister_shrinker(&s->s_shrink);
330 125821 : fs->kill_sb(s);
331 :
332 : /*
333 : * Since list_lru_destroy() may sleep, we cannot call it from
334 : * put_super(), where we hold the sb_lock. Therefore we destroy
335 : * the lru lists right now.
336 : */
337 125821 : list_lru_destroy(&s->s_dentry_lru);
338 125813 : list_lru_destroy(&s->s_inode_lru);
339 :
340 125816 : put_filesystem(fs);
341 125819 : put_super(s);
342 : } else {
343 70638 : up_write(&s->s_umount);
344 : }
345 196458 : }
346 :
347 : EXPORT_SYMBOL(deactivate_locked_super);
348 :
349 : /**
350 : * deactivate_super - drop an active reference to superblock
351 : * @s: superblock to deactivate
352 : *
353 : * Variant of deactivate_locked_super(), except that superblock is *not*
354 : * locked by caller. If we are going to drop the final active reference,
355 : * lock will be acquired prior to that.
356 : */
357 15427270 : void deactivate_super(struct super_block *s)
358 : {
359 30856143 : if (!atomic_add_unless(&s->s_active, -1, 1)) {
360 124366 : down_write(&s->s_umount);
361 124366 : deactivate_locked_super(s);
362 : }
363 15428873 : }
364 :
365 : EXPORT_SYMBOL(deactivate_super);
366 :
367 : /**
368 : * grab_super - acquire an active reference
369 : * @s: reference we are trying to make active
370 : *
371 : * Tries to acquire an active reference. grab_super() is used when we
372 : * had just found a superblock in super_blocks or fs_type->fs_supers
373 : * and want to turn it into a full-blown active reference. grab_super()
374 : * is called with sb_lock held and drops it. Returns 1 in case of
375 : * success, 0 if we had failed (superblock contents was already dead or
376 : * dying when grab_super() had been called). Note that this is only
377 : * called for superblocks not in rundown mode (== ones still on ->fs_supers
378 : * of their type), so increment of ->s_count is OK here.
379 : */
380 4005 : static int grab_super(struct super_block *s) __releases(sb_lock)
381 : {
382 4005 : s->s_count++;
383 4005 : spin_unlock(&sb_lock);
384 4005 : down_write(&s->s_umount);
385 7713 : if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
386 3694 : put_super(s);
387 3694 : return 1;
388 : }
389 311 : up_write(&s->s_umount);
390 311 : put_super(s);
391 311 : return 0;
392 : }
393 :
394 : /*
395 : * trylock_super - try to grab ->s_umount shared
396 : * @sb: reference we are trying to grab
397 : *
398 : * Try to prevent fs shutdown. This is used in places where we
399 : * cannot take an active reference but we need to ensure that the
400 : * filesystem is not shut down while we are working on it. It returns
401 : * false if we cannot acquire s_umount or if we lose the race and
402 : * filesystem already got into shutdown, and returns true with the s_umount
403 : * lock held in read mode in case of success. On successful return,
404 : * the caller must drop the s_umount lock when done.
405 : *
406 : * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
407 : * The reason why it's safe is that we are OK with doing trylock instead
408 : * of down_read(). There's a couple of places that are OK with that, but
409 : * it's very much not a general-purpose interface.
410 : */
411 4136658 : bool trylock_super(struct super_block *sb)
412 : {
413 4136658 : if (down_read_trylock(&sb->s_umount)) {
414 2903143 : if (!hlist_unhashed(&sb->s_instances) &&
415 2903143 : sb->s_root && (sb->s_flags & SB_BORN))
416 : return true;
417 3 : up_read(&sb->s_umount);
418 : }
419 :
420 : return false;
421 : }
422 :
423 : /**
424 : * retire_super - prevents superblock from being reused
425 : * @sb: superblock to retire
426 : *
427 : * The function marks superblock to be ignored in superblock test, which
428 : * prevents it from being reused for any new mounts. If the superblock has
429 : * a private bdi, it also unregisters it, but doesn't reduce the refcount
430 : * of the superblock to prevent potential races. The refcount is reduced
431 : * by generic_shutdown_super(). The function can not be called
432 : * concurrently with generic_shutdown_super(). It is safe to call the
433 : * function multiple times, subsequent calls have no effect.
434 : *
435 : * The marker will affect the re-use only for block-device-based
436 : * superblocks. Other superblocks will still get marked if this function
437 : * is used, but that will not affect their reusability.
438 : */
439 0 : void retire_super(struct super_block *sb)
440 : {
441 0 : WARN_ON(!sb->s_bdev);
442 0 : down_write(&sb->s_umount);
443 0 : if (sb->s_iflags & SB_I_PERSB_BDI) {
444 0 : bdi_unregister(sb->s_bdi);
445 0 : sb->s_iflags &= ~SB_I_PERSB_BDI;
446 : }
447 0 : sb->s_iflags |= SB_I_RETIRED;
448 0 : up_write(&sb->s_umount);
449 0 : }
450 : EXPORT_SYMBOL(retire_super);
451 :
452 : /**
453 : * generic_shutdown_super - common helper for ->kill_sb()
454 : * @sb: superblock to kill
455 : *
456 : * generic_shutdown_super() does all fs-independent work on superblock
457 : * shutdown. Typical ->kill_sb() should pick all fs-specific objects
458 : * that need destruction out of superblock, call generic_shutdown_super()
459 : * and release aforementioned objects. Note: dentries and inodes _are_
460 : * taken care of and do not need specific handling.
461 : *
462 : * Upon calling this function, the filesystem may no longer alter or
463 : * rearrange the set of dentries belonging to this super_block, nor may it
464 : * change the attachments of dentries to inodes.
465 : */
466 125820 : void generic_shutdown_super(struct super_block *sb)
467 : {
468 125820 : const struct super_operations *sop = sb->s_op;
469 :
470 125820 : if (sb->s_root) {
471 124609 : shrink_dcache_for_umount(sb);
472 124610 : sync_filesystem(sb);
473 124568 : sb->s_flags &= ~SB_ACTIVE;
474 :
475 124568 : cgroup_writeback_umount();
476 :
477 : /* Evict all inodes with zero refcount. */
478 124570 : evict_inodes(sb);
479 :
480 : /*
481 : * Clean up and evict any inodes that still have references due
482 : * to fsnotify or the security policy.
483 : */
484 124532 : fsnotify_sb_delete(sb);
485 124560 : security_sb_delete(sb);
486 :
487 : /*
488 : * Now that all potentially-encrypted inodes have been evicted,
489 : * the fscrypt keyring can be destroyed.
490 : */
491 124560 : fscrypt_destroy_keyring(sb);
492 :
493 124560 : if (sb->s_dio_done_wq) {
494 8133 : destroy_workqueue(sb->s_dio_done_wq);
495 8133 : sb->s_dio_done_wq = NULL;
496 : }
497 :
498 124560 : if (sop->put_super)
499 124478 : sop->put_super(sb);
500 :
501 124608 : if (CHECK_DATA_CORRUPTION(!list_empty(&sb->s_inodes),
502 : "VFS: Busy inodes after unmount of %s (%s)",
503 : sb->s_id, sb->s_type->name)) {
504 : /*
505 : * Adding a proper bailout path here would be hard, but
506 : * we can at least make it more likely that a later
507 : * iput_final() or such crashes cleanly.
508 : */
509 : struct inode *inode;
510 :
511 : spin_lock(&sb->s_inode_list_lock);
512 : list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
513 : inode->i_op = VFS_PTR_POISON;
514 : inode->i_sb = VFS_PTR_POISON;
515 : inode->i_mapping = VFS_PTR_POISON;
516 : }
517 : spin_unlock(&sb->s_inode_list_lock);
518 : }
519 : }
520 125819 : spin_lock(&sb_lock);
521 : /* should be initialized for __put_super_and_need_restart() */
522 125821 : hlist_del_init(&sb->s_instances);
523 125821 : spin_unlock(&sb_lock);
524 125821 : up_write(&sb->s_umount);
525 125821 : if (sb->s_bdi != &noop_backing_dev_info) {
526 73622 : if (sb->s_iflags & SB_I_PERSB_BDI)
527 3244 : bdi_unregister(sb->s_bdi);
528 73622 : bdi_put(sb->s_bdi);
529 73622 : sb->s_bdi = &noop_backing_dev_info;
530 : }
531 125821 : }
532 :
533 : EXPORT_SYMBOL(generic_shutdown_super);
534 :
535 129090 : bool mount_capable(struct fs_context *fc)
536 : {
537 129090 : if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT))
538 76951 : return capable(CAP_SYS_ADMIN);
539 : else
540 52139 : return ns_capable(fc->user_ns, CAP_SYS_ADMIN);
541 : }
542 :
543 : /**
544 : * sget_fc - Find or create a superblock
545 : * @fc: Filesystem context.
546 : * @test: Comparison callback
547 : * @set: Setup callback
548 : *
549 : * Find or create a superblock using the parameters stored in the filesystem
550 : * context and the two callback functions.
551 : *
552 : * If an extant superblock is matched, then that will be returned with an
553 : * elevated reference count that the caller must transfer or discard.
554 : *
555 : * If no match is made, a new superblock will be allocated and basic
556 : * initialisation will be performed (s_type, s_fs_info and s_id will be set and
557 : * the set() callback will be invoked), the superblock will be published and it
558 : * will be returned in a partially constructed state with SB_BORN and SB_ACTIVE
559 : * as yet unset.
560 : */
561 125182 : struct super_block *sget_fc(struct fs_context *fc,
562 : int (*test)(struct super_block *, struct fs_context *),
563 : int (*set)(struct super_block *, struct fs_context *))
564 : {
565 125182 : struct super_block *s = NULL;
566 125182 : struct super_block *old;
567 125182 : struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
568 247742 : int err;
569 :
570 : retry:
571 247742 : spin_lock(&sb_lock);
572 247861 : if (test) {
573 389474 : hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
574 105299 : if (test(old, fc))
575 2751 : goto share_extant_sb;
576 : }
577 : }
578 245110 : if (!s) {
579 122555 : spin_unlock(&sb_lock);
580 122551 : s = alloc_super(fc->fs_type, fc->sb_flags, user_ns);
581 122548 : if (!s)
582 : return ERR_PTR(-ENOMEM);
583 122548 : goto retry;
584 : }
585 :
586 122555 : s->s_fs_info = fc->s_fs_info;
587 122555 : err = set(s, fc);
588 122555 : if (err) {
589 0 : s->s_fs_info = NULL;
590 0 : spin_unlock(&sb_lock);
591 0 : destroy_unused_super(s);
592 0 : return ERR_PTR(err);
593 : }
594 122555 : fc->s_fs_info = NULL;
595 122555 : s->s_type = fc->fs_type;
596 122555 : s->s_iflags |= fc->s_iflags;
597 122555 : strscpy(s->s_id, s->s_type->name, sizeof(s->s_id));
598 122555 : list_add_tail(&s->s_list, &super_blocks);
599 122555 : hlist_add_head(&s->s_instances, &s->s_type->fs_supers);
600 122555 : spin_unlock(&sb_lock);
601 122555 : get_filesystem(s->s_type);
602 122555 : register_shrinker_prepared(&s->s_shrink);
603 122555 : return s;
604 :
605 : share_extant_sb:
606 2751 : if (user_ns != old->s_user_ns) {
607 0 : spin_unlock(&sb_lock);
608 0 : destroy_unused_super(s);
609 0 : return ERR_PTR(-EBUSY);
610 : }
611 2751 : if (!grab_super(old))
612 12 : goto retry;
613 2739 : destroy_unused_super(s);
614 2739 : return old;
615 : }
616 : EXPORT_SYMBOL(sget_fc);
617 :
618 : /**
619 : * sget - find or create a superblock
620 : * @type: filesystem type superblock should belong to
621 : * @test: comparison callback
622 : * @set: setup callback
623 : * @flags: mount flags
624 : * @data: argument to each of them
625 : */
626 3493 : struct super_block *sget(struct file_system_type *type,
627 : int (*test)(struct super_block *,void *),
628 : int (*set)(struct super_block *,void *),
629 : int flags,
630 : void *data)
631 : {
632 3493 : struct user_namespace *user_ns = current_user_ns();
633 3493 : struct super_block *s = NULL;
634 3493 : struct super_block *old;
635 3493 : int err;
636 :
637 : /* We don't yet pass the user namespace of the parent
638 : * mount through to here so always use &init_user_ns
639 : * until that changes.
640 : */
641 3493 : if (flags & SB_SUBMOUNT)
642 11 : user_ns = &init_user_ns;
643 :
644 3493 : retry:
645 6749 : spin_lock(&sb_lock);
646 6749 : if (test) {
647 13502 : hlist_for_each_entry(old, &type->fs_supers, s_instances) {
648 4553 : if (!test(old, data))
649 4314 : continue;
650 239 : if (user_ns != old->s_user_ns) {
651 0 : spin_unlock(&sb_lock);
652 0 : destroy_unused_super(s);
653 0 : return ERR_PTR(-EBUSY);
654 : }
655 239 : if (!grab_super(old))
656 1 : goto retry;
657 238 : destroy_unused_super(s);
658 238 : return old;
659 : }
660 : }
661 6510 : if (!s) {
662 3255 : spin_unlock(&sb_lock);
663 3255 : s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
664 3255 : if (!s)
665 : return ERR_PTR(-ENOMEM);
666 3255 : goto retry;
667 : }
668 :
669 3255 : err = set(s, data);
670 3255 : if (err) {
671 0 : spin_unlock(&sb_lock);
672 0 : destroy_unused_super(s);
673 0 : return ERR_PTR(err);
674 : }
675 3255 : s->s_type = type;
676 3255 : strscpy(s->s_id, type->name, sizeof(s->s_id));
677 3255 : list_add_tail(&s->s_list, &super_blocks);
678 3255 : hlist_add_head(&s->s_instances, &type->fs_supers);
679 3255 : spin_unlock(&sb_lock);
680 3255 : get_filesystem(type);
681 3255 : register_shrinker_prepared(&s->s_shrink);
682 3255 : return s;
683 : }
684 : EXPORT_SYMBOL(sget);
685 :
686 1729717 : void drop_super(struct super_block *sb)
687 : {
688 1729717 : up_read(&sb->s_umount);
689 1730522 : put_super(sb);
690 0 : }
691 :
692 : EXPORT_SYMBOL(drop_super);
693 :
694 535 : void drop_super_exclusive(struct super_block *sb)
695 : {
696 535 : up_write(&sb->s_umount);
697 535 : put_super(sb);
698 535 : }
699 : EXPORT_SYMBOL(drop_super_exclusive);
700 :
701 0 : static void __iterate_supers(void (*f)(struct super_block *))
702 : {
703 0 : struct super_block *sb, *p = NULL;
704 :
705 0 : spin_lock(&sb_lock);
706 0 : list_for_each_entry(sb, &super_blocks, s_list) {
707 0 : if (hlist_unhashed(&sb->s_instances))
708 0 : continue;
709 0 : sb->s_count++;
710 0 : spin_unlock(&sb_lock);
711 :
712 0 : f(sb);
713 :
714 0 : spin_lock(&sb_lock);
715 0 : if (p)
716 0 : __put_super(p);
717 : p = sb;
718 : }
719 0 : if (p)
720 0 : __put_super(p);
721 0 : spin_unlock(&sb_lock);
722 0 : }
723 : /**
724 : * iterate_supers - call function for all active superblocks
725 : * @f: function to call
726 : * @arg: argument to pass to it
727 : *
728 : * Scans the superblock list and calls given function, passing it
729 : * locked superblock and given argument.
730 : */
731 6068843 : void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
732 : {
733 6068843 : struct super_block *sb, *p = NULL;
734 :
735 6068843 : spin_lock(&sb_lock);
736 351986872 : list_for_each_entry(sb, &super_blocks, s_list) {
737 345917957 : if (hlist_unhashed(&sb->s_instances))
738 286 : continue;
739 345917671 : sb->s_count++;
740 345917671 : spin_unlock(&sb_lock);
741 :
742 345779303 : down_read(&sb->s_umount);
743 345781191 : if (sb->s_root && (sb->s_flags & SB_BORN))
744 345776225 : f(sb, arg);
745 345643349 : up_read(&sb->s_umount);
746 :
747 345460686 : spin_lock(&sb_lock);
748 345917671 : if (p)
749 339848756 : __put_super(p);
750 : p = sb;
751 : }
752 6068915 : if (p)
753 6068915 : __put_super(p);
754 6068915 : spin_unlock(&sb_lock);
755 6068887 : }
756 :
757 : /**
758 : * iterate_supers_type - call function for superblocks of given type
759 : * @type: fs type
760 : * @f: function to call
761 : * @arg: argument to pass to it
762 : *
763 : * Scans the superblock list and calls given function, passing it
764 : * locked superblock and given argument.
765 : */
766 0 : void iterate_supers_type(struct file_system_type *type,
767 : void (*f)(struct super_block *, void *), void *arg)
768 : {
769 0 : struct super_block *sb, *p = NULL;
770 :
771 0 : spin_lock(&sb_lock);
772 0 : hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
773 0 : sb->s_count++;
774 0 : spin_unlock(&sb_lock);
775 :
776 0 : down_read(&sb->s_umount);
777 0 : if (sb->s_root && (sb->s_flags & SB_BORN))
778 0 : f(sb, arg);
779 0 : up_read(&sb->s_umount);
780 :
781 0 : spin_lock(&sb_lock);
782 0 : if (p)
783 0 : __put_super(p);
784 0 : p = sb;
785 : }
786 0 : if (p)
787 0 : __put_super(p);
788 0 : spin_unlock(&sb_lock);
789 0 : }
790 :
791 : EXPORT_SYMBOL(iterate_supers_type);
792 :
793 : /**
794 : * get_super - get the superblock of a device
795 : * @bdev: device to get the superblock for
796 : *
797 : * Scans the superblock list and finds the superblock of the file system
798 : * mounted on the device given. %NULL is returned if no match is found.
799 : */
800 337795 : struct super_block *get_super(struct block_device *bdev)
801 : {
802 337795 : struct super_block *sb;
803 :
804 337795 : if (!bdev)
805 : return NULL;
806 :
807 337795 : spin_lock(&sb_lock);
808 337795 : rescan:
809 19123478 : list_for_each_entry(sb, &super_blocks, s_list) {
810 18787457 : if (hlist_unhashed(&sb->s_instances))
811 1425 : continue;
812 18786032 : if (sb->s_bdev == bdev) {
813 1774 : sb->s_count++;
814 1774 : spin_unlock(&sb_lock);
815 1774 : down_read(&sb->s_umount);
816 : /* still alive? */
817 1774 : if (sb->s_root && (sb->s_flags & SB_BORN))
818 1774 : return sb;
819 0 : up_read(&sb->s_umount);
820 : /* nope, got unmounted */
821 0 : spin_lock(&sb_lock);
822 0 : __put_super(sb);
823 0 : goto rescan;
824 : }
825 : }
826 336021 : spin_unlock(&sb_lock);
827 336021 : return NULL;
828 : }
829 :
830 : /**
831 : * get_active_super - get an active reference to the superblock of a device
832 : * @bdev: device to get the superblock for
833 : *
834 : * Scans the superblock list and finds the superblock of the file system
835 : * mounted on the device given. Returns the superblock with an active
836 : * reference or %NULL if none was found.
837 : */
838 2333 : struct super_block *get_active_super(struct block_device *bdev)
839 : {
840 2333 : struct super_block *sb;
841 :
842 2333 : if (!bdev)
843 : return NULL;
844 :
845 2333 : restart:
846 2631 : spin_lock(&sb_lock);
847 150588 : list_for_each_entry(sb, &super_blocks, s_list) {
848 148972 : if (hlist_unhashed(&sb->s_instances))
849 5 : continue;
850 148967 : if (sb->s_bdev == bdev) {
851 1015 : if (!grab_super(sb))
852 298 : goto restart;
853 717 : up_write(&sb->s_umount);
854 717 : return sb;
855 : }
856 : }
857 1616 : spin_unlock(&sb_lock);
858 1616 : return NULL;
859 : }
860 :
861 1727479 : struct super_block *user_get_super(dev_t dev, bool excl)
862 : {
863 1727479 : struct super_block *sb;
864 :
865 1727479 : spin_lock(&sb_lock);
866 1730247 : rescan:
867 98378380 : list_for_each_entry(sb, &super_blocks, s_list) {
868 98378380 : if (hlist_unhashed(&sb->s_instances))
869 0 : continue;
870 98378380 : if (sb->s_dev == dev) {
871 1730247 : sb->s_count++;
872 1730247 : spin_unlock(&sb_lock);
873 1730247 : if (excl)
874 535 : down_write(&sb->s_umount);
875 : else
876 1729712 : down_read(&sb->s_umount);
877 : /* still alive? */
878 1730247 : if (sb->s_root && (sb->s_flags & SB_BORN))
879 1730247 : return sb;
880 0 : if (excl)
881 0 : up_write(&sb->s_umount);
882 : else
883 0 : up_read(&sb->s_umount);
884 : /* nope, got unmounted */
885 0 : spin_lock(&sb_lock);
886 0 : __put_super(sb);
887 0 : goto rescan;
888 : }
889 : }
890 0 : spin_unlock(&sb_lock);
891 0 : return NULL;
892 : }
893 :
894 : /**
895 : * reconfigure_super - asks filesystem to change superblock parameters
896 : * @fc: The superblock and configuration
897 : *
898 : * Alters the configuration parameters of a live superblock.
899 : */
900 13293 : int reconfigure_super(struct fs_context *fc)
901 : {
902 13293 : struct super_block *sb = fc->root->d_sb;
903 13293 : int retval;
904 13293 : bool remount_ro = false;
905 13293 : bool remount_rw = false;
906 13293 : bool force = fc->sb_flags & SB_FORCE;
907 :
908 13293 : if (fc->sb_flags_mask & ~MS_RMT_MASK)
909 : return -EINVAL;
910 13293 : if (sb->s_writers.frozen != SB_UNFROZEN)
911 : return -EBUSY;
912 :
913 13293 : retval = security_sb_remount(sb, fc->security);
914 13293 : if (retval)
915 : return retval;
916 :
917 13293 : if (fc->sb_flags_mask & SB_RDONLY) {
918 : #ifdef CONFIG_BLOCK
919 14496 : if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev &&
920 1203 : bdev_read_only(sb->s_bdev))
921 : return -EACCES;
922 : #endif
923 13293 : remount_rw = !(fc->sb_flags & SB_RDONLY) && sb_rdonly(sb);
924 13293 : remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
925 : }
926 :
927 13293 : if (remount_ro) {
928 11964 : if (!hlist_empty(&sb->s_pins)) {
929 0 : up_write(&sb->s_umount);
930 0 : group_pin_kill(&sb->s_pins);
931 0 : down_write(&sb->s_umount);
932 0 : if (!sb->s_root)
933 : return 0;
934 0 : if (sb->s_writers.frozen != SB_UNFROZEN)
935 : return -EBUSY;
936 0 : remount_ro = !sb_rdonly(sb);
937 : }
938 : }
939 13293 : shrink_dcache_sb(sb);
940 :
941 : /* If we are reconfiguring to RDONLY and current sb is read/write,
942 : * make sure there are no files open for writing.
943 : */
944 13293 : if (remount_ro) {
945 11964 : if (force) {
946 0 : sb_start_ro_state_change(sb);
947 : } else {
948 11964 : retval = sb_prepare_remount_readonly(sb);
949 11964 : if (retval)
950 : return retval;
951 : }
952 1329 : } else if (remount_rw) {
953 : /*
954 : * Protect filesystem's reconfigure code from writes from
955 : * userspace until reconfigure finishes.
956 : */
957 428 : sb_start_ro_state_change(sb);
958 : }
959 :
960 1790 : if (fc->ops->reconfigure) {
961 1790 : retval = fc->ops->reconfigure(fc);
962 1790 : if (retval) {
963 155 : if (!force)
964 155 : goto cancel_readonly;
965 : /* If forced remount, go ahead despite any errors */
966 0 : WARN(1, "forced remount of a %s fs returned %i\n",
967 : sb->s_type->name, retval);
968 : }
969 : }
970 :
971 1635 : WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
972 : (fc->sb_flags & fc->sb_flags_mask)));
973 1635 : sb_end_ro_state_change(sb);
974 :
975 : /*
976 : * Some filesystems modify their metadata via some other path than the
977 : * bdev buffer cache (eg. use a private mapping, or directories in
978 : * pagecache, etc). Also file data modifications go via their own
979 : * mappings. So If we try to mount readonly then copy the filesystem
980 : * from bdev, we could get stale data, so invalidate it to give a best
981 : * effort at coherency.
982 : */
983 1635 : if (remount_ro && sb->s_bdev)
984 455 : invalidate_bdev(sb->s_bdev);
985 : return 0;
986 :
987 : cancel_readonly:
988 155 : sb_end_ro_state_change(sb);
989 155 : return retval;
990 : }
991 :
992 0 : static void do_emergency_remount_callback(struct super_block *sb)
993 : {
994 0 : down_write(&sb->s_umount);
995 0 : if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
996 : !sb_rdonly(sb)) {
997 0 : struct fs_context *fc;
998 :
999 0 : fc = fs_context_for_reconfigure(sb->s_root,
1000 : SB_RDONLY | SB_FORCE, SB_RDONLY);
1001 0 : if (!IS_ERR(fc)) {
1002 0 : if (parse_monolithic_mount_data(fc, NULL) == 0)
1003 0 : (void)reconfigure_super(fc);
1004 0 : put_fs_context(fc);
1005 : }
1006 : }
1007 0 : up_write(&sb->s_umount);
1008 0 : }
1009 :
1010 0 : static void do_emergency_remount(struct work_struct *work)
1011 : {
1012 0 : __iterate_supers(do_emergency_remount_callback);
1013 0 : kfree(work);
1014 0 : printk("Emergency Remount complete\n");
1015 0 : }
1016 :
1017 0 : void emergency_remount(void)
1018 : {
1019 0 : struct work_struct *work;
1020 :
1021 0 : work = kmalloc(sizeof(*work), GFP_ATOMIC);
1022 0 : if (work) {
1023 0 : INIT_WORK(work, do_emergency_remount);
1024 0 : schedule_work(work);
1025 : }
1026 0 : }
1027 :
1028 0 : static void do_thaw_all_callback(struct super_block *sb)
1029 : {
1030 0 : down_write(&sb->s_umount);
1031 0 : if (sb->s_root && sb->s_flags & SB_BORN) {
1032 0 : emergency_thaw_bdev(sb);
1033 0 : thaw_super_locked(sb, FREEZE_HOLDER_USERSPACE);
1034 : } else {
1035 0 : up_write(&sb->s_umount);
1036 : }
1037 0 : }
1038 :
1039 0 : static void do_thaw_all(struct work_struct *work)
1040 : {
1041 0 : __iterate_supers(do_thaw_all_callback);
1042 0 : kfree(work);
1043 0 : printk(KERN_WARNING "Emergency Thaw complete\n");
1044 0 : }
1045 :
1046 : /**
1047 : * emergency_thaw_all -- forcibly thaw every frozen filesystem
1048 : *
1049 : * Used for emergency unfreeze of all filesystems via SysRq
1050 : */
1051 0 : void emergency_thaw_all(void)
1052 : {
1053 0 : struct work_struct *work;
1054 :
1055 0 : work = kmalloc(sizeof(*work), GFP_ATOMIC);
1056 0 : if (work) {
1057 0 : INIT_WORK(work, do_thaw_all);
1058 0 : schedule_work(work);
1059 : }
1060 0 : }
1061 :
1062 : static DEFINE_IDA(unnamed_dev_ida);
1063 :
1064 : /**
1065 : * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1066 : * @p: Pointer to a dev_t.
1067 : *
1068 : * Filesystems which don't use real block devices can call this function
1069 : * to allocate a virtual block device.
1070 : *
1071 : * Context: Any context. Frequently called while holding sb_lock.
1072 : * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1073 : * or -ENOMEM if memory allocation failed.
1074 : */
1075 115475 : int get_anon_bdev(dev_t *p)
1076 : {
1077 115475 : int dev;
1078 :
1079 : /*
1080 : * Many userspace utilities consider an FSID of 0 invalid.
1081 : * Always return at least 1 from get_anon_bdev.
1082 : */
1083 115475 : dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1084 : GFP_ATOMIC);
1085 115492 : if (dev == -ENOSPC)
1086 : dev = -EMFILE;
1087 115492 : if (dev < 0)
1088 0 : return dev;
1089 :
1090 115492 : *p = MKDEV(0, dev);
1091 115492 : return 0;
1092 : }
1093 : EXPORT_SYMBOL(get_anon_bdev);
1094 :
1095 60075 : void free_anon_bdev(dev_t dev)
1096 : {
1097 60075 : ida_free(&unnamed_dev_ida, MINOR(dev));
1098 60076 : }
1099 : EXPORT_SYMBOL(free_anon_bdev);
1100 :
1101 3244 : int set_anon_super(struct super_block *s, void *data)
1102 : {
1103 3244 : return get_anon_bdev(&s->s_dev);
1104 : }
1105 : EXPORT_SYMBOL(set_anon_super);
1106 :
1107 55442 : void kill_anon_super(struct super_block *sb)
1108 : {
1109 55442 : dev_t dev = sb->s_dev;
1110 55442 : generic_shutdown_super(sb);
1111 55443 : free_anon_bdev(dev);
1112 55443 : }
1113 : EXPORT_SYMBOL(kill_anon_super);
1114 :
1115 916 : void kill_litter_super(struct super_block *sb)
1116 : {
1117 916 : if (sb->s_root)
1118 916 : d_genocide(sb->s_root);
1119 916 : kill_anon_super(sb);
1120 916 : }
1121 : EXPORT_SYMBOL(kill_litter_super);
1122 :
1123 52199 : int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1124 : {
1125 52199 : return set_anon_super(sb, NULL);
1126 : }
1127 : EXPORT_SYMBOL(set_anon_super_fc);
1128 :
1129 0 : static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1130 : {
1131 0 : return sb->s_fs_info == fc->s_fs_info;
1132 : }
1133 :
1134 0 : static int test_single_super(struct super_block *s, struct fs_context *fc)
1135 : {
1136 0 : return 1;
1137 : }
1138 :
1139 52137 : static int vfs_get_super(struct fs_context *fc, bool reconf,
1140 : int (*test)(struct super_block *, struct fs_context *),
1141 : int (*fill_super)(struct super_block *sb,
1142 : struct fs_context *fc))
1143 : {
1144 52137 : struct super_block *sb;
1145 52137 : int err;
1146 :
1147 52137 : sb = sget_fc(fc, test, set_anon_super_fc);
1148 52199 : if (IS_ERR(sb))
1149 0 : return PTR_ERR(sb);
1150 :
1151 52199 : if (!sb->s_root) {
1152 52199 : err = fill_super(sb, fc);
1153 52075 : if (err)
1154 0 : goto error;
1155 :
1156 52075 : sb->s_flags |= SB_ACTIVE;
1157 104154 : fc->root = dget(sb->s_root);
1158 : } else {
1159 0 : fc->root = dget(sb->s_root);
1160 0 : if (reconf) {
1161 0 : err = reconfigure_super(fc);
1162 0 : if (err < 0) {
1163 0 : dput(fc->root);
1164 0 : fc->root = NULL;
1165 0 : goto error;
1166 : }
1167 : }
1168 : }
1169 :
1170 : return 0;
1171 :
1172 0 : error:
1173 0 : deactivate_locked_super(sb);
1174 0 : return err;
1175 : }
1176 :
1177 52139 : int get_tree_nodev(struct fs_context *fc,
1178 : int (*fill_super)(struct super_block *sb,
1179 : struct fs_context *fc))
1180 : {
1181 52139 : return vfs_get_super(fc, false, NULL, fill_super);
1182 : }
1183 : EXPORT_SYMBOL(get_tree_nodev);
1184 :
1185 0 : int get_tree_single(struct fs_context *fc,
1186 : int (*fill_super)(struct super_block *sb,
1187 : struct fs_context *fc))
1188 : {
1189 0 : return vfs_get_super(fc, false, test_single_super, fill_super);
1190 : }
1191 : EXPORT_SYMBOL(get_tree_single);
1192 :
1193 0 : int get_tree_single_reconf(struct fs_context *fc,
1194 : int (*fill_super)(struct super_block *sb,
1195 : struct fs_context *fc))
1196 : {
1197 0 : return vfs_get_super(fc, true, test_single_super, fill_super);
1198 : }
1199 : EXPORT_SYMBOL(get_tree_single_reconf);
1200 :
1201 0 : int get_tree_keyed(struct fs_context *fc,
1202 : int (*fill_super)(struct super_block *sb,
1203 : struct fs_context *fc),
1204 : void *key)
1205 : {
1206 0 : fc->s_fs_info = key;
1207 0 : return vfs_get_super(fc, false, test_keyed_super, fill_super);
1208 : }
1209 : EXPORT_SYMBOL(get_tree_keyed);
1210 :
1211 : #ifdef CONFIG_BLOCK
1212 0 : static void fs_mark_dead(struct block_device *bdev)
1213 : {
1214 0 : struct super_block *sb;
1215 :
1216 0 : sb = get_super(bdev);
1217 0 : if (!sb)
1218 : return;
1219 :
1220 0 : if (sb->s_op->shutdown)
1221 0 : sb->s_op->shutdown(sb);
1222 0 : drop_super(sb);
1223 : }
1224 :
1225 : static const struct blk_holder_ops fs_holder_ops = {
1226 : .mark_dead = fs_mark_dead,
1227 : };
1228 :
1229 70367 : static int set_bdev_super(struct super_block *s, void *data)
1230 : {
1231 70367 : s->s_bdev = data;
1232 70367 : s->s_dev = s->s_bdev->bd_dev;
1233 70367 : s->s_bdi = bdi_get(s->s_bdev->bd_disk->bdi);
1234 :
1235 70367 : if (bdev_stable_writes(s->s_bdev))
1236 0 : s->s_iflags |= SB_I_STABLE_WRITES;
1237 70367 : return 0;
1238 : }
1239 :
1240 70356 : static int set_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1241 : {
1242 70356 : return set_bdev_super(s, fc->sget_key);
1243 : }
1244 :
1245 105299 : static int test_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1246 : {
1247 105299 : return !(s->s_iflags & SB_I_RETIRED) && s->s_bdev == fc->sget_key;
1248 : }
1249 :
1250 : /**
1251 : * get_tree_bdev - Get a superblock based on a single block device
1252 : * @fc: The filesystem context holding the parameters
1253 : * @fill_super: Helper to initialise a new superblock
1254 : */
1255 73467 : int get_tree_bdev(struct fs_context *fc,
1256 : int (*fill_super)(struct super_block *,
1257 : struct fs_context *))
1258 : {
1259 73467 : struct block_device *bdev;
1260 73467 : struct super_block *s;
1261 73467 : int error = 0;
1262 :
1263 73467 : if (!fc->source)
1264 0 : return invalf(fc, "No source specified");
1265 :
1266 73467 : bdev = blkdev_get_by_path(fc->source, sb_open_mode(fc->sb_flags),
1267 73467 : fc->fs_type, &fs_holder_ops);
1268 73467 : if (IS_ERR(bdev)) {
1269 36 : errorf(fc, "%s: Can't open blockdev", fc->source);
1270 36 : return PTR_ERR(bdev);
1271 : }
1272 :
1273 : /* Once the superblock is inserted into the list by sget_fc(), s_umount
1274 : * will protect the lockfs code from trying to start a snapshot while
1275 : * we are mounting
1276 : */
1277 73431 : mutex_lock(&bdev->bd_fsfreeze_mutex);
1278 73431 : if (bdev->bd_fsfreeze_count > 0) {
1279 336 : mutex_unlock(&bdev->bd_fsfreeze_mutex);
1280 336 : warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
1281 336 : blkdev_put(bdev, fc->fs_type);
1282 336 : return -EBUSY;
1283 : }
1284 :
1285 73095 : fc->sb_flags |= SB_NOSEC;
1286 73095 : fc->sget_key = bdev;
1287 73095 : s = sget_fc(fc, test_bdev_super_fc, set_bdev_super_fc);
1288 73095 : mutex_unlock(&bdev->bd_fsfreeze_mutex);
1289 73095 : if (IS_ERR(s)) {
1290 0 : blkdev_put(bdev, fc->fs_type);
1291 0 : return PTR_ERR(s);
1292 : }
1293 :
1294 73095 : if (s->s_root) {
1295 : /* Don't summarily change the RO/RW state. */
1296 2739 : if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1297 0 : warnf(fc, "%pg: Can't mount, would change RO state", bdev);
1298 0 : deactivate_locked_super(s);
1299 0 : blkdev_put(bdev, fc->fs_type);
1300 0 : return -EBUSY;
1301 : }
1302 :
1303 : /*
1304 : * s_umount nests inside open_mutex during
1305 : * __invalidate_device(). blkdev_put() acquires
1306 : * open_mutex and can't be called under s_umount. Drop
1307 : * s_umount temporarily. This is safe as we're
1308 : * holding an active reference.
1309 : */
1310 2739 : up_write(&s->s_umount);
1311 2739 : blkdev_put(bdev, fc->fs_type);
1312 2739 : down_write(&s->s_umount);
1313 : } else {
1314 70356 : snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1315 70356 : shrinker_debugfs_rename(&s->s_shrink, "sb-%s:%s",
1316 70356 : fc->fs_type->name, s->s_id);
1317 70356 : sb_set_blocksize(s, block_size(bdev));
1318 70356 : error = fill_super(s, fc);
1319 70356 : if (error) {
1320 1184 : deactivate_locked_super(s);
1321 1184 : return error;
1322 : }
1323 :
1324 69172 : s->s_flags |= SB_ACTIVE;
1325 69172 : bdev->bd_super = s;
1326 : }
1327 :
1328 71911 : BUG_ON(fc->root);
1329 71911 : fc->root = dget(s->s_root);
1330 71911 : return 0;
1331 : }
1332 : EXPORT_SYMBOL(get_tree_bdev);
1333 :
1334 10 : static int test_bdev_super(struct super_block *s, void *data)
1335 : {
1336 10 : return !(s->s_iflags & SB_I_RETIRED) && (void *)s->s_bdev == data;
1337 : }
1338 :
1339 11 : struct dentry *mount_bdev(struct file_system_type *fs_type,
1340 : int flags, const char *dev_name, void *data,
1341 : int (*fill_super)(struct super_block *, void *, int))
1342 : {
1343 11 : struct block_device *bdev;
1344 11 : struct super_block *s;
1345 11 : int error = 0;
1346 :
1347 22 : bdev = blkdev_get_by_path(dev_name, sb_open_mode(flags), fs_type,
1348 : &fs_holder_ops);
1349 11 : if (IS_ERR(bdev))
1350 : return ERR_CAST(bdev);
1351 :
1352 : /*
1353 : * once the super is inserted into the list by sget, s_umount
1354 : * will protect the lockfs code from trying to start a snapshot
1355 : * while we are mounting
1356 : */
1357 11 : mutex_lock(&bdev->bd_fsfreeze_mutex);
1358 11 : if (bdev->bd_fsfreeze_count > 0) {
1359 0 : mutex_unlock(&bdev->bd_fsfreeze_mutex);
1360 0 : error = -EBUSY;
1361 0 : goto error_bdev;
1362 : }
1363 11 : s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
1364 : bdev);
1365 11 : mutex_unlock(&bdev->bd_fsfreeze_mutex);
1366 11 : if (IS_ERR(s))
1367 0 : goto error_s;
1368 :
1369 11 : if (s->s_root) {
1370 0 : if ((flags ^ s->s_flags) & SB_RDONLY) {
1371 0 : deactivate_locked_super(s);
1372 0 : error = -EBUSY;
1373 0 : goto error_bdev;
1374 : }
1375 :
1376 : /*
1377 : * s_umount nests inside open_mutex during
1378 : * __invalidate_device(). blkdev_put() acquires
1379 : * open_mutex and can't be called under s_umount. Drop
1380 : * s_umount temporarily. This is safe as we're
1381 : * holding an active reference.
1382 : */
1383 0 : up_write(&s->s_umount);
1384 0 : blkdev_put(bdev, fs_type);
1385 0 : down_write(&s->s_umount);
1386 : } else {
1387 11 : snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1388 11 : shrinker_debugfs_rename(&s->s_shrink, "sb-%s:%s",
1389 : fs_type->name, s->s_id);
1390 11 : sb_set_blocksize(s, block_size(bdev));
1391 11 : error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1392 11 : if (error) {
1393 0 : deactivate_locked_super(s);
1394 0 : goto error;
1395 : }
1396 :
1397 11 : s->s_flags |= SB_ACTIVE;
1398 11 : bdev->bd_super = s;
1399 : }
1400 :
1401 11 : return dget(s->s_root);
1402 :
1403 : error_s:
1404 0 : error = PTR_ERR(s);
1405 0 : error_bdev:
1406 0 : blkdev_put(bdev, fs_type);
1407 0 : error:
1408 0 : return ERR_PTR(error);
1409 : }
1410 : EXPORT_SYMBOL(mount_bdev);
1411 :
1412 70378 : void kill_block_super(struct super_block *sb)
1413 : {
1414 70378 : struct block_device *bdev = sb->s_bdev;
1415 :
1416 70378 : bdev->bd_super = NULL;
1417 70378 : generic_shutdown_super(sb);
1418 70378 : sync_blockdev(bdev);
1419 70378 : blkdev_put(bdev, sb->s_type);
1420 70378 : }
1421 :
1422 : EXPORT_SYMBOL(kill_block_super);
1423 : #endif
1424 :
1425 0 : struct dentry *mount_nodev(struct file_system_type *fs_type,
1426 : int flags, void *data,
1427 : int (*fill_super)(struct super_block *, void *, int))
1428 : {
1429 0 : int error;
1430 0 : struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1431 :
1432 0 : if (IS_ERR(s))
1433 : return ERR_CAST(s);
1434 :
1435 0 : error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1436 0 : if (error) {
1437 0 : deactivate_locked_super(s);
1438 0 : return ERR_PTR(error);
1439 : }
1440 0 : s->s_flags |= SB_ACTIVE;
1441 0 : return dget(s->s_root);
1442 : }
1443 : EXPORT_SYMBOL(mount_nodev);
1444 :
1445 11 : int reconfigure_single(struct super_block *s,
1446 : int flags, void *data)
1447 : {
1448 11 : struct fs_context *fc;
1449 11 : int ret;
1450 :
1451 : /* The caller really need to be passing fc down into mount_single(),
1452 : * then a chunk of this can be removed. [Bollocks -- AV]
1453 : * Better yet, reconfiguration shouldn't happen, but rather the second
1454 : * mount should be rejected if the parameters are not compatible.
1455 : */
1456 11 : fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1457 11 : if (IS_ERR(fc))
1458 0 : return PTR_ERR(fc);
1459 :
1460 11 : ret = parse_monolithic_mount_data(fc, data);
1461 11 : if (ret < 0)
1462 0 : goto out;
1463 :
1464 11 : ret = reconfigure_super(fc);
1465 11 : out:
1466 11 : put_fs_context(fc);
1467 11 : return ret;
1468 : }
1469 :
1470 11 : static int compare_single(struct super_block *s, void *p)
1471 : {
1472 11 : return 1;
1473 : }
1474 :
1475 11 : struct dentry *mount_single(struct file_system_type *fs_type,
1476 : int flags, void *data,
1477 : int (*fill_super)(struct super_block *, void *, int))
1478 : {
1479 11 : struct super_block *s;
1480 11 : int error;
1481 :
1482 11 : s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1483 11 : if (IS_ERR(s))
1484 : return ERR_CAST(s);
1485 11 : if (!s->s_root) {
1486 0 : error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1487 0 : if (!error)
1488 0 : s->s_flags |= SB_ACTIVE;
1489 : } else {
1490 11 : error = reconfigure_single(s, flags, data);
1491 : }
1492 11 : if (unlikely(error)) {
1493 0 : deactivate_locked_super(s);
1494 0 : return ERR_PTR(error);
1495 : }
1496 11 : return dget(s->s_root);
1497 : }
1498 : EXPORT_SYMBOL(mount_single);
1499 :
1500 : /**
1501 : * vfs_get_tree - Get the mountable root
1502 : * @fc: The superblock configuration context.
1503 : *
1504 : * The filesystem is invoked to get or create a superblock which can then later
1505 : * be used for mounting. The filesystem places a pointer to the root to be
1506 : * used for mounting in @fc->root.
1507 : */
1508 132587 : int vfs_get_tree(struct fs_context *fc)
1509 : {
1510 132587 : struct super_block *sb;
1511 132587 : int error;
1512 :
1513 132587 : if (fc->root)
1514 : return -EBUSY;
1515 :
1516 : /* Get the mountable root in fc->root, with a ref on the root and a ref
1517 : * on the superblock.
1518 : */
1519 132587 : error = fc->ops->get_tree(fc);
1520 132576 : if (error < 0)
1521 : return error;
1522 :
1523 130960 : if (!fc->root) {
1524 0 : pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1525 : fc->fs_type->name);
1526 : /* We don't know what the locking state of the superblock is -
1527 : * if there is a superblock.
1528 : */
1529 0 : BUG();
1530 : }
1531 :
1532 130960 : sb = fc->root->d_sb;
1533 130960 : WARN_ON(!sb->s_bdi);
1534 :
1535 : /*
1536 : * Write barrier is for super_cache_count(). We place it before setting
1537 : * SB_BORN as the data dependency between the two functions is the
1538 : * superblock structure contents that we just set up, not the SB_BORN
1539 : * flag.
1540 : */
1541 130960 : smp_wmb();
1542 130939 : sb->s_flags |= SB_BORN;
1543 :
1544 130939 : error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
1545 130939 : if (unlikely(error)) {
1546 : fc_drop_locked(fc);
1547 : return error;
1548 : }
1549 :
1550 : /*
1551 : * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1552 : * but s_maxbytes was an unsigned long long for many releases. Throw
1553 : * this warning for a little while to try and catch filesystems that
1554 : * violate this rule.
1555 : */
1556 130939 : WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1557 : "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
1558 :
1559 : return 0;
1560 : }
1561 : EXPORT_SYMBOL(vfs_get_tree);
1562 :
1563 : /*
1564 : * Setup private BDI for given superblock. It gets automatically cleaned up
1565 : * in generic_shutdown_super().
1566 : */
1567 3244 : int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1568 : {
1569 3244 : struct backing_dev_info *bdi;
1570 3244 : int err;
1571 3244 : va_list args;
1572 :
1573 3244 : bdi = bdi_alloc(NUMA_NO_NODE);
1574 3244 : if (!bdi)
1575 : return -ENOMEM;
1576 :
1577 3244 : va_start(args, fmt);
1578 3244 : err = bdi_register_va(bdi, fmt, args);
1579 3244 : va_end(args);
1580 3244 : if (err) {
1581 0 : bdi_put(bdi);
1582 0 : return err;
1583 : }
1584 3244 : WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1585 3244 : sb->s_bdi = bdi;
1586 3244 : sb->s_iflags |= SB_I_PERSB_BDI;
1587 :
1588 3244 : return 0;
1589 : }
1590 : EXPORT_SYMBOL(super_setup_bdi_name);
1591 :
1592 : /*
1593 : * Setup private BDI for given superblock. I gets automatically cleaned up
1594 : * in generic_shutdown_super().
1595 : */
1596 3244 : int super_setup_bdi(struct super_block *sb)
1597 : {
1598 3244 : static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1599 :
1600 3244 : return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1601 : atomic_long_inc_return(&bdi_seq));
1602 : }
1603 : EXPORT_SYMBOL(super_setup_bdi);
1604 :
1605 : /**
1606 : * sb_wait_write - wait until all writers to given file system finish
1607 : * @sb: the super for which we wait
1608 : * @level: type of writers we wait for (normal vs page fault)
1609 : *
1610 : * This function waits until there are no writers of given type to given file
1611 : * system.
1612 : */
1613 : static void sb_wait_write(struct super_block *sb, int level)
1614 : {
1615 209519 : percpu_down_write(sb->s_writers.rw_sem + level-1);
1616 : }
1617 :
1618 : /*
1619 : * We are going to return to userspace and forget about these locks, the
1620 : * ownership goes to the caller of thaw_super() which does unlock().
1621 : */
1622 : static void lockdep_sb_freeze_release(struct super_block *sb)
1623 : {
1624 69833 : int level;
1625 :
1626 69833 : for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1627 69833 : percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1628 : }
1629 :
1630 : /*
1631 : * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1632 : */
1633 : static void lockdep_sb_freeze_acquire(struct super_block *sb)
1634 : {
1635 69833 : int level;
1636 :
1637 69833 : for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1638 69833 : percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1639 : }
1640 :
1641 69843 : static void sb_freeze_unlock(struct super_block *sb, int level)
1642 : {
1643 279362 : for (level--; level >= 0; level--)
1644 209519 : percpu_up_write(sb->s_writers.rw_sem + level);
1645 69843 : }
1646 :
1647 691 : static int wait_for_partially_frozen(struct super_block *sb)
1648 : {
1649 691 : int ret = 0;
1650 :
1651 691 : do {
1652 691 : unsigned short old = sb->s_writers.frozen;
1653 :
1654 691 : up_write(&sb->s_umount);
1655 1372 : ret = wait_var_event_killable(&sb->s_writers.frozen,
1656 : sb->s_writers.frozen != old);
1657 691 : down_write(&sb->s_umount);
1658 691 : } while (ret == 0 &&
1659 691 : sb->s_writers.frozen != SB_UNFROZEN &&
1660 : sb->s_writers.frozen != SB_FREEZE_COMPLETE);
1661 :
1662 691 : return ret;
1663 : }
1664 :
1665 : /**
1666 : * freeze_super - lock the filesystem and force it into a consistent state
1667 : * @sb: the super to lock
1668 : * @who: context that wants to freeze
1669 : *
1670 : * Syncs the super to make sure the filesystem is consistent and calls the fs's
1671 : * freeze_fs. Subsequent calls to this without first thawing the fs may return
1672 : * -EBUSY.
1673 : *
1674 : * @who should be:
1675 : * * %FREEZE_HOLDER_USERSPACE if userspace wants to freeze the fs;
1676 : * * %FREEZE_HOLDER_KERNEL if the kernel wants to freeze the fs.
1677 : *
1678 : * The @who argument distinguishes between the kernel and userspace trying to
1679 : * freeze the filesystem. Although there cannot be multiple kernel freezes or
1680 : * multiple userspace freezes in effect at any given time, the kernel and
1681 : * userspace can both hold a filesystem frozen. The filesystem remains frozen
1682 : * until there are no kernel or userspace freezes in effect.
1683 : *
1684 : * During this function, sb->s_writers.frozen goes through these values:
1685 : *
1686 : * SB_UNFROZEN: File system is normal, all writes progress as usual.
1687 : *
1688 : * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1689 : * writes should be blocked, though page faults are still allowed. We wait for
1690 : * all writes to complete and then proceed to the next stage.
1691 : *
1692 : * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1693 : * but internal fs threads can still modify the filesystem (although they
1694 : * should not dirty new pages or inodes), writeback can run etc. After waiting
1695 : * for all running page faults we sync the filesystem which will clean all
1696 : * dirty pages and inodes (no new dirty pages or inodes can be created when
1697 : * sync is running).
1698 : *
1699 : * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1700 : * modification are blocked (e.g. XFS preallocation truncation on inode
1701 : * reclaim). This is usually implemented by blocking new transactions for
1702 : * filesystems that have them and need this additional guard. After all
1703 : * internal writers are finished we call ->freeze_fs() to finish filesystem
1704 : * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1705 : * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1706 : *
1707 : * sb->s_writers.frozen is protected by sb->s_umount.
1708 : */
1709 70882 : int freeze_super(struct super_block *sb, enum freeze_holder who)
1710 : {
1711 70882 : int ret;
1712 :
1713 70882 : atomic_inc(&sb->s_active);
1714 70881 : down_write(&sb->s_umount);
1715 :
1716 71573 : retry:
1717 71573 : if (sb->s_writers.frozen == SB_FREEZE_COMPLETE) {
1718 1039 : if (sb->s_writers.freeze_holders & who) {
1719 959 : deactivate_locked_super(sb);
1720 959 : return -EBUSY;
1721 : }
1722 :
1723 80 : WARN_ON(sb->s_writers.freeze_holders == 0);
1724 :
1725 : /*
1726 : * Someone else already holds this type of freeze; share the
1727 : * freeze and assign the active ref to the freeze.
1728 : */
1729 80 : sb->s_writers.freeze_holders |= who;
1730 80 : up_write(&sb->s_umount);
1731 80 : return 0;
1732 : }
1733 :
1734 70534 : if (sb->s_writers.frozen != SB_UNFROZEN) {
1735 691 : ret = wait_for_partially_frozen(sb);
1736 691 : if (ret) {
1737 0 : deactivate_locked_super(sb);
1738 0 : return ret;
1739 : }
1740 :
1741 691 : goto retry;
1742 : }
1743 :
1744 69843 : if (!(sb->s_flags & SB_BORN)) {
1745 0 : up_write(&sb->s_umount);
1746 0 : return 0; /* sic - it's "nothing to do" */
1747 : }
1748 :
1749 69843 : if (sb_rdonly(sb)) {
1750 : /* Nothing to do really... */
1751 0 : sb->s_writers.freeze_holders |= who;
1752 0 : sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1753 0 : wake_up_var(&sb->s_writers.frozen);
1754 0 : up_write(&sb->s_umount);
1755 0 : return 0;
1756 : }
1757 :
1758 69843 : sb->s_writers.frozen = SB_FREEZE_WRITE;
1759 : /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1760 69843 : up_write(&sb->s_umount);
1761 69843 : sb_wait_write(sb, SB_FREEZE_WRITE);
1762 69843 : down_write(&sb->s_umount);
1763 :
1764 : /* Now we go and block page faults... */
1765 69843 : sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1766 69843 : sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1767 :
1768 : /* All writers are done so after syncing there won't be dirty data */
1769 69843 : ret = sync_filesystem(sb);
1770 69843 : if (ret) {
1771 10 : sb->s_writers.frozen = SB_UNFROZEN;
1772 10 : sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT);
1773 10 : wake_up_var(&sb->s_writers.frozen);
1774 10 : deactivate_locked_super(sb);
1775 10 : return ret;
1776 : }
1777 :
1778 : /* Now wait for internal filesystem counter */
1779 69833 : sb->s_writers.frozen = SB_FREEZE_FS;
1780 69833 : sb_wait_write(sb, SB_FREEZE_FS);
1781 :
1782 69833 : if (sb->s_op->freeze_fs) {
1783 69833 : ret = sb->s_op->freeze_fs(sb);
1784 69833 : if (ret) {
1785 0 : printk(KERN_ERR
1786 : "VFS:Filesystem freeze failed\n");
1787 0 : sb->s_writers.frozen = SB_UNFROZEN;
1788 0 : sb_freeze_unlock(sb, SB_FREEZE_FS);
1789 0 : wake_up_var(&sb->s_writers.frozen);
1790 0 : deactivate_locked_super(sb);
1791 0 : return ret;
1792 : }
1793 : }
1794 : /*
1795 : * For debugging purposes so that fs can warn if it sees write activity
1796 : * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1797 : */
1798 69833 : sb->s_writers.freeze_holders |= who;
1799 69833 : sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1800 69833 : wake_up_var(&sb->s_writers.frozen);
1801 69833 : lockdep_sb_freeze_release(sb);
1802 69833 : up_write(&sb->s_umount);
1803 69833 : return 0;
1804 : }
1805 : EXPORT_SYMBOL(freeze_super);
1806 :
1807 : /*
1808 : * Undoes the effect of a freeze_super_locked call. If the filesystem is
1809 : * frozen both by userspace and the kernel, a thaw call from either source
1810 : * removes that state without releasing the other state or unlocking the
1811 : * filesystem.
1812 : */
1813 71045 : static int thaw_super_locked(struct super_block *sb, enum freeze_holder who)
1814 : {
1815 71045 : int error;
1816 :
1817 71045 : if (sb->s_writers.frozen == SB_FREEZE_COMPLETE) {
1818 69913 : if (!(sb->s_writers.freeze_holders & who)) {
1819 0 : up_write(&sb->s_umount);
1820 0 : return -EINVAL;
1821 : }
1822 :
1823 : /*
1824 : * Freeze is shared with someone else. Release our hold and
1825 : * drop the active ref that freeze_super assigned to the
1826 : * freezer.
1827 : */
1828 69913 : if (sb->s_writers.freeze_holders & ~who) {
1829 80 : sb->s_writers.freeze_holders &= ~who;
1830 80 : deactivate_locked_super(sb);
1831 80 : return 0;
1832 : }
1833 : } else {
1834 1132 : up_write(&sb->s_umount);
1835 1132 : return -EINVAL;
1836 : }
1837 :
1838 69833 : if (sb_rdonly(sb)) {
1839 0 : sb->s_writers.freeze_holders &= ~who;
1840 0 : sb->s_writers.frozen = SB_UNFROZEN;
1841 0 : wake_up_var(&sb->s_writers.frozen);
1842 0 : goto out;
1843 : }
1844 :
1845 69833 : lockdep_sb_freeze_acquire(sb);
1846 :
1847 69833 : if (sb->s_op->unfreeze_fs) {
1848 69833 : error = sb->s_op->unfreeze_fs(sb);
1849 69833 : if (error) {
1850 0 : printk(KERN_ERR
1851 : "VFS:Filesystem thaw failed\n");
1852 0 : lockdep_sb_freeze_release(sb);
1853 0 : up_write(&sb->s_umount);
1854 0 : return error;
1855 : }
1856 : }
1857 :
1858 69833 : sb->s_writers.freeze_holders &= ~who;
1859 69833 : sb->s_writers.frozen = SB_UNFROZEN;
1860 69833 : wake_up_var(&sb->s_writers.frozen);
1861 69833 : sb_freeze_unlock(sb, SB_FREEZE_FS);
1862 69833 : out:
1863 69833 : deactivate_locked_super(sb);
1864 69833 : return 0;
1865 : }
1866 :
1867 : /**
1868 : * thaw_super -- unlock filesystem
1869 : * @sb: the super to thaw
1870 : * @who: context that wants to freeze
1871 : *
1872 : * Unlocks the filesystem and marks it writeable again after freeze_super()
1873 : * if there are no remaining freezes on the filesystem.
1874 : *
1875 : * @who should be:
1876 : * * %FREEZE_HOLDER_USERSPACE if userspace wants to thaw the fs;
1877 : * * %FREEZE_HOLDER_KERNEL if the kernel wants to thaw the fs.
1878 : */
1879 71044 : int thaw_super(struct super_block *sb, enum freeze_holder who)
1880 : {
1881 71044 : down_write(&sb->s_umount);
1882 71045 : return thaw_super_locked(sb, who);
1883 : }
1884 : EXPORT_SYMBOL(thaw_super);
1885 :
1886 : /*
1887 : * Create workqueue for deferred direct IO completions. We allocate the
1888 : * workqueue when it's first needed. This avoids creating workqueue for
1889 : * filesystems that don't need it and also allows us to create the workqueue
1890 : * late enough so the we can include s_id in the name of the workqueue.
1891 : */
1892 8449 : int sb_init_dio_done_wq(struct super_block *sb)
1893 : {
1894 8449 : struct workqueue_struct *old;
1895 8449 : struct workqueue_struct *wq = alloc_workqueue("dio/%s",
1896 : WQ_MEM_RECLAIM, 0,
1897 8449 : sb->s_id);
1898 8461 : if (!wq)
1899 : return -ENOMEM;
1900 : /*
1901 : * This has to be atomic as more DIOs can race to create the workqueue
1902 : */
1903 8461 : old = cmpxchg(&sb->s_dio_done_wq, NULL, wq);
1904 : /* Someone created workqueue before us? Free ours... */
1905 8461 : if (old)
1906 328 : destroy_workqueue(wq);
1907 : return 0;
1908 : }
|
