Loading...
1/*
2 * linux/fs/super.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * super.c contains code to handle: - mount structures
7 * - super-block tables
8 * - filesystem drivers list
9 * - mount system call
10 * - umount system call
11 * - ustat system call
12 *
13 * GK 2/5/95 - Changed to support mounting the root fs via NFS
14 *
15 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall
16 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96
17 * Added options to /proc/mounts:
18 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
19 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
20 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
21 */
22
23#include <linux/export.h>
24#include <linux/slab.h>
25#include <linux/blkdev.h>
26#include <linux/mount.h>
27#include <linux/security.h>
28#include <linux/writeback.h> /* for the emergency remount stuff */
29#include <linux/idr.h>
30#include <linux/mutex.h>
31#include <linux/backing-dev.h>
32#include <linux/rculist_bl.h>
33#include <linux/cleancache.h>
34#include <linux/fsnotify.h>
35#include <linux/lockdep.h>
36#include "internal.h"
37
38
39static LIST_HEAD(super_blocks);
40static DEFINE_SPINLOCK(sb_lock);
41
42static char *sb_writers_name[SB_FREEZE_LEVELS] = {
43 "sb_writers",
44 "sb_pagefaults",
45 "sb_internal",
46};
47
48/*
49 * One thing we have to be careful of with a per-sb shrinker is that we don't
50 * drop the last active reference to the superblock from within the shrinker.
51 * If that happens we could trigger unregistering the shrinker from within the
52 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
53 * take a passive reference to the superblock to avoid this from occurring.
54 */
55static unsigned long super_cache_scan(struct shrinker *shrink,
56 struct shrink_control *sc)
57{
58 struct super_block *sb;
59 long fs_objects = 0;
60 long total_objects;
61 long freed = 0;
62 long dentries;
63 long inodes;
64
65 sb = container_of(shrink, struct super_block, s_shrink);
66
67 /*
68 * Deadlock avoidance. We may hold various FS locks, and we don't want
69 * to recurse into the FS that called us in clear_inode() and friends..
70 */
71 if (!(sc->gfp_mask & __GFP_FS))
72 return SHRINK_STOP;
73
74 if (!trylock_super(sb))
75 return SHRINK_STOP;
76
77 if (sb->s_op->nr_cached_objects)
78 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
79
80 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
81 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
82 total_objects = dentries + inodes + fs_objects + 1;
83 if (!total_objects)
84 total_objects = 1;
85
86 /* proportion the scan between the caches */
87 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
88 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
89 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
90
91 /*
92 * prune the dcache first as the icache is pinned by it, then
93 * prune the icache, followed by the filesystem specific caches
94 *
95 * Ensure that we always scan at least one object - memcg kmem
96 * accounting uses this to fully empty the caches.
97 */
98 sc->nr_to_scan = dentries + 1;
99 freed = prune_dcache_sb(sb, sc);
100 sc->nr_to_scan = inodes + 1;
101 freed += prune_icache_sb(sb, sc);
102
103 if (fs_objects) {
104 sc->nr_to_scan = fs_objects + 1;
105 freed += sb->s_op->free_cached_objects(sb, sc);
106 }
107
108 up_read(&sb->s_umount);
109 return freed;
110}
111
112static unsigned long super_cache_count(struct shrinker *shrink,
113 struct shrink_control *sc)
114{
115 struct super_block *sb;
116 long total_objects = 0;
117
118 sb = container_of(shrink, struct super_block, s_shrink);
119
120 /*
121 * Don't call trylock_super as it is a potential
122 * scalability bottleneck. The counts could get updated
123 * between super_cache_count and super_cache_scan anyway.
124 * Call to super_cache_count with shrinker_rwsem held
125 * ensures the safety of call to list_lru_shrink_count() and
126 * s_op->nr_cached_objects().
127 */
128 if (sb->s_op && sb->s_op->nr_cached_objects)
129 total_objects = sb->s_op->nr_cached_objects(sb, sc);
130
131 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
132 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
133
134 total_objects = vfs_pressure_ratio(total_objects);
135 return total_objects;
136}
137
138static void destroy_super_work(struct work_struct *work)
139{
140 struct super_block *s = container_of(work, struct super_block,
141 destroy_work);
142 int i;
143
144 for (i = 0; i < SB_FREEZE_LEVELS; i++)
145 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
146 kfree(s);
147}
148
149static void destroy_super_rcu(struct rcu_head *head)
150{
151 struct super_block *s = container_of(head, struct super_block, rcu);
152 INIT_WORK(&s->destroy_work, destroy_super_work);
153 schedule_work(&s->destroy_work);
154}
155
156/**
157 * destroy_super - frees a superblock
158 * @s: superblock to free
159 *
160 * Frees a superblock.
161 */
162static void destroy_super(struct super_block *s)
163{
164 list_lru_destroy(&s->s_dentry_lru);
165 list_lru_destroy(&s->s_inode_lru);
166 security_sb_free(s);
167 WARN_ON(!list_empty(&s->s_mounts));
168 kfree(s->s_subtype);
169 kfree(s->s_options);
170 call_rcu(&s->rcu, destroy_super_rcu);
171}
172
173/**
174 * alloc_super - create new superblock
175 * @type: filesystem type superblock should belong to
176 * @flags: the mount flags
177 *
178 * Allocates and initializes a new &struct super_block. alloc_super()
179 * returns a pointer new superblock or %NULL if allocation had failed.
180 */
181static struct super_block *alloc_super(struct file_system_type *type, int flags)
182{
183 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
184 static const struct super_operations default_op;
185 int i;
186
187 if (!s)
188 return NULL;
189
190 INIT_LIST_HEAD(&s->s_mounts);
191
192 if (security_sb_alloc(s))
193 goto fail;
194
195 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
196 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
197 sb_writers_name[i],
198 &type->s_writers_key[i]))
199 goto fail;
200 }
201 init_waitqueue_head(&s->s_writers.wait_unfrozen);
202 s->s_bdi = &noop_backing_dev_info;
203 s->s_flags = flags;
204 INIT_HLIST_NODE(&s->s_instances);
205 INIT_HLIST_BL_HEAD(&s->s_anon);
206 mutex_init(&s->s_sync_lock);
207 INIT_LIST_HEAD(&s->s_inodes);
208 spin_lock_init(&s->s_inode_list_lock);
209
210 if (list_lru_init_memcg(&s->s_dentry_lru))
211 goto fail;
212 if (list_lru_init_memcg(&s->s_inode_lru))
213 goto fail;
214
215 init_rwsem(&s->s_umount);
216 lockdep_set_class(&s->s_umount, &type->s_umount_key);
217 /*
218 * sget() can have s_umount recursion.
219 *
220 * When it cannot find a suitable sb, it allocates a new
221 * one (this one), and tries again to find a suitable old
222 * one.
223 *
224 * In case that succeeds, it will acquire the s_umount
225 * lock of the old one. Since these are clearly distrinct
226 * locks, and this object isn't exposed yet, there's no
227 * risk of deadlocks.
228 *
229 * Annotate this by putting this lock in a different
230 * subclass.
231 */
232 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
233 s->s_count = 1;
234 atomic_set(&s->s_active, 1);
235 mutex_init(&s->s_vfs_rename_mutex);
236 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
237 mutex_init(&s->s_dquot.dqio_mutex);
238 mutex_init(&s->s_dquot.dqonoff_mutex);
239 s->s_maxbytes = MAX_NON_LFS;
240 s->s_op = &default_op;
241 s->s_time_gran = 1000000000;
242 s->cleancache_poolid = CLEANCACHE_NO_POOL;
243
244 s->s_shrink.seeks = DEFAULT_SEEKS;
245 s->s_shrink.scan_objects = super_cache_scan;
246 s->s_shrink.count_objects = super_cache_count;
247 s->s_shrink.batch = 1024;
248 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
249 return s;
250
251fail:
252 destroy_super(s);
253 return NULL;
254}
255
256/* Superblock refcounting */
257
258/*
259 * Drop a superblock's refcount. The caller must hold sb_lock.
260 */
261static void __put_super(struct super_block *sb)
262{
263 if (!--sb->s_count) {
264 list_del_init(&sb->s_list);
265 destroy_super(sb);
266 }
267}
268
269/**
270 * put_super - drop a temporary reference to superblock
271 * @sb: superblock in question
272 *
273 * Drops a temporary reference, frees superblock if there's no
274 * references left.
275 */
276static void put_super(struct super_block *sb)
277{
278 spin_lock(&sb_lock);
279 __put_super(sb);
280 spin_unlock(&sb_lock);
281}
282
283
284/**
285 * deactivate_locked_super - drop an active reference to superblock
286 * @s: superblock to deactivate
287 *
288 * Drops an active reference to superblock, converting it into a temprory
289 * one if there is no other active references left. In that case we
290 * tell fs driver to shut it down and drop the temporary reference we
291 * had just acquired.
292 *
293 * Caller holds exclusive lock on superblock; that lock is released.
294 */
295void deactivate_locked_super(struct super_block *s)
296{
297 struct file_system_type *fs = s->s_type;
298 if (atomic_dec_and_test(&s->s_active)) {
299 cleancache_invalidate_fs(s);
300 unregister_shrinker(&s->s_shrink);
301 fs->kill_sb(s);
302
303 /*
304 * Since list_lru_destroy() may sleep, we cannot call it from
305 * put_super(), where we hold the sb_lock. Therefore we destroy
306 * the lru lists right now.
307 */
308 list_lru_destroy(&s->s_dentry_lru);
309 list_lru_destroy(&s->s_inode_lru);
310
311 put_filesystem(fs);
312 put_super(s);
313 } else {
314 up_write(&s->s_umount);
315 }
316}
317
318EXPORT_SYMBOL(deactivate_locked_super);
319
320/**
321 * deactivate_super - drop an active reference to superblock
322 * @s: superblock to deactivate
323 *
324 * Variant of deactivate_locked_super(), except that superblock is *not*
325 * locked by caller. If we are going to drop the final active reference,
326 * lock will be acquired prior to that.
327 */
328void deactivate_super(struct super_block *s)
329{
330 if (!atomic_add_unless(&s->s_active, -1, 1)) {
331 down_write(&s->s_umount);
332 deactivate_locked_super(s);
333 }
334}
335
336EXPORT_SYMBOL(deactivate_super);
337
338/**
339 * grab_super - acquire an active reference
340 * @s: reference we are trying to make active
341 *
342 * Tries to acquire an active reference. grab_super() is used when we
343 * had just found a superblock in super_blocks or fs_type->fs_supers
344 * and want to turn it into a full-blown active reference. grab_super()
345 * is called with sb_lock held and drops it. Returns 1 in case of
346 * success, 0 if we had failed (superblock contents was already dead or
347 * dying when grab_super() had been called). Note that this is only
348 * called for superblocks not in rundown mode (== ones still on ->fs_supers
349 * of their type), so increment of ->s_count is OK here.
350 */
351static int grab_super(struct super_block *s) __releases(sb_lock)
352{
353 s->s_count++;
354 spin_unlock(&sb_lock);
355 down_write(&s->s_umount);
356 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
357 put_super(s);
358 return 1;
359 }
360 up_write(&s->s_umount);
361 put_super(s);
362 return 0;
363}
364
365/*
366 * trylock_super - try to grab ->s_umount shared
367 * @sb: reference we are trying to grab
368 *
369 * Try to prevent fs shutdown. This is used in places where we
370 * cannot take an active reference but we need to ensure that the
371 * filesystem is not shut down while we are working on it. It returns
372 * false if we cannot acquire s_umount or if we lose the race and
373 * filesystem already got into shutdown, and returns true with the s_umount
374 * lock held in read mode in case of success. On successful return,
375 * the caller must drop the s_umount lock when done.
376 *
377 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
378 * The reason why it's safe is that we are OK with doing trylock instead
379 * of down_read(). There's a couple of places that are OK with that, but
380 * it's very much not a general-purpose interface.
381 */
382bool trylock_super(struct super_block *sb)
383{
384 if (down_read_trylock(&sb->s_umount)) {
385 if (!hlist_unhashed(&sb->s_instances) &&
386 sb->s_root && (sb->s_flags & MS_BORN))
387 return true;
388 up_read(&sb->s_umount);
389 }
390
391 return false;
392}
393
394/**
395 * generic_shutdown_super - common helper for ->kill_sb()
396 * @sb: superblock to kill
397 *
398 * generic_shutdown_super() does all fs-independent work on superblock
399 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
400 * that need destruction out of superblock, call generic_shutdown_super()
401 * and release aforementioned objects. Note: dentries and inodes _are_
402 * taken care of and do not need specific handling.
403 *
404 * Upon calling this function, the filesystem may no longer alter or
405 * rearrange the set of dentries belonging to this super_block, nor may it
406 * change the attachments of dentries to inodes.
407 */
408void generic_shutdown_super(struct super_block *sb)
409{
410 const struct super_operations *sop = sb->s_op;
411
412 if (sb->s_root) {
413 shrink_dcache_for_umount(sb);
414 sync_filesystem(sb);
415 sb->s_flags &= ~MS_ACTIVE;
416
417 fsnotify_unmount_inodes(sb);
418 cgroup_writeback_umount();
419
420 evict_inodes(sb);
421
422 if (sb->s_dio_done_wq) {
423 destroy_workqueue(sb->s_dio_done_wq);
424 sb->s_dio_done_wq = NULL;
425 }
426
427 if (sop->put_super)
428 sop->put_super(sb);
429
430 if (!list_empty(&sb->s_inodes)) {
431 printk("VFS: Busy inodes after unmount of %s. "
432 "Self-destruct in 5 seconds. Have a nice day...\n",
433 sb->s_id);
434 }
435 }
436 spin_lock(&sb_lock);
437 /* should be initialized for __put_super_and_need_restart() */
438 hlist_del_init(&sb->s_instances);
439 spin_unlock(&sb_lock);
440 up_write(&sb->s_umount);
441}
442
443EXPORT_SYMBOL(generic_shutdown_super);
444
445/**
446 * sget - find or create a superblock
447 * @type: filesystem type superblock should belong to
448 * @test: comparison callback
449 * @set: setup callback
450 * @flags: mount flags
451 * @data: argument to each of them
452 */
453struct super_block *sget(struct file_system_type *type,
454 int (*test)(struct super_block *,void *),
455 int (*set)(struct super_block *,void *),
456 int flags,
457 void *data)
458{
459 struct super_block *s = NULL;
460 struct super_block *old;
461 int err;
462
463retry:
464 spin_lock(&sb_lock);
465 if (test) {
466 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
467 if (!test(old, data))
468 continue;
469 if (!grab_super(old))
470 goto retry;
471 if (s) {
472 up_write(&s->s_umount);
473 destroy_super(s);
474 s = NULL;
475 }
476 return old;
477 }
478 }
479 if (!s) {
480 spin_unlock(&sb_lock);
481 s = alloc_super(type, flags);
482 if (!s)
483 return ERR_PTR(-ENOMEM);
484 goto retry;
485 }
486
487 err = set(s, data);
488 if (err) {
489 spin_unlock(&sb_lock);
490 up_write(&s->s_umount);
491 destroy_super(s);
492 return ERR_PTR(err);
493 }
494 s->s_type = type;
495 strlcpy(s->s_id, type->name, sizeof(s->s_id));
496 list_add_tail(&s->s_list, &super_blocks);
497 hlist_add_head(&s->s_instances, &type->fs_supers);
498 spin_unlock(&sb_lock);
499 get_filesystem(type);
500 register_shrinker(&s->s_shrink);
501 return s;
502}
503
504EXPORT_SYMBOL(sget);
505
506void drop_super(struct super_block *sb)
507{
508 up_read(&sb->s_umount);
509 put_super(sb);
510}
511
512EXPORT_SYMBOL(drop_super);
513
514/**
515 * iterate_supers - call function for all active superblocks
516 * @f: function to call
517 * @arg: argument to pass to it
518 *
519 * Scans the superblock list and calls given function, passing it
520 * locked superblock and given argument.
521 */
522void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
523{
524 struct super_block *sb, *p = NULL;
525
526 spin_lock(&sb_lock);
527 list_for_each_entry(sb, &super_blocks, s_list) {
528 if (hlist_unhashed(&sb->s_instances))
529 continue;
530 sb->s_count++;
531 spin_unlock(&sb_lock);
532
533 down_read(&sb->s_umount);
534 if (sb->s_root && (sb->s_flags & MS_BORN))
535 f(sb, arg);
536 up_read(&sb->s_umount);
537
538 spin_lock(&sb_lock);
539 if (p)
540 __put_super(p);
541 p = sb;
542 }
543 if (p)
544 __put_super(p);
545 spin_unlock(&sb_lock);
546}
547
548/**
549 * iterate_supers_type - call function for superblocks of given type
550 * @type: fs type
551 * @f: function to call
552 * @arg: argument to pass to it
553 *
554 * Scans the superblock list and calls given function, passing it
555 * locked superblock and given argument.
556 */
557void iterate_supers_type(struct file_system_type *type,
558 void (*f)(struct super_block *, void *), void *arg)
559{
560 struct super_block *sb, *p = NULL;
561
562 spin_lock(&sb_lock);
563 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
564 sb->s_count++;
565 spin_unlock(&sb_lock);
566
567 down_read(&sb->s_umount);
568 if (sb->s_root && (sb->s_flags & MS_BORN))
569 f(sb, arg);
570 up_read(&sb->s_umount);
571
572 spin_lock(&sb_lock);
573 if (p)
574 __put_super(p);
575 p = sb;
576 }
577 if (p)
578 __put_super(p);
579 spin_unlock(&sb_lock);
580}
581
582EXPORT_SYMBOL(iterate_supers_type);
583
584/**
585 * get_super - get the superblock of a device
586 * @bdev: device to get the superblock for
587 *
588 * Scans the superblock list and finds the superblock of the file system
589 * mounted on the device given. %NULL is returned if no match is found.
590 */
591
592struct super_block *get_super(struct block_device *bdev)
593{
594 struct super_block *sb;
595
596 if (!bdev)
597 return NULL;
598
599 spin_lock(&sb_lock);
600rescan:
601 list_for_each_entry(sb, &super_blocks, s_list) {
602 if (hlist_unhashed(&sb->s_instances))
603 continue;
604 if (sb->s_bdev == bdev) {
605 sb->s_count++;
606 spin_unlock(&sb_lock);
607 down_read(&sb->s_umount);
608 /* still alive? */
609 if (sb->s_root && (sb->s_flags & MS_BORN))
610 return sb;
611 up_read(&sb->s_umount);
612 /* nope, got unmounted */
613 spin_lock(&sb_lock);
614 __put_super(sb);
615 goto rescan;
616 }
617 }
618 spin_unlock(&sb_lock);
619 return NULL;
620}
621
622EXPORT_SYMBOL(get_super);
623
624/**
625 * get_super_thawed - get thawed superblock of a device
626 * @bdev: device to get the superblock for
627 *
628 * Scans the superblock list and finds the superblock of the file system
629 * mounted on the device. The superblock is returned once it is thawed
630 * (or immediately if it was not frozen). %NULL is returned if no match
631 * is found.
632 */
633struct super_block *get_super_thawed(struct block_device *bdev)
634{
635 while (1) {
636 struct super_block *s = get_super(bdev);
637 if (!s || s->s_writers.frozen == SB_UNFROZEN)
638 return s;
639 up_read(&s->s_umount);
640 wait_event(s->s_writers.wait_unfrozen,
641 s->s_writers.frozen == SB_UNFROZEN);
642 put_super(s);
643 }
644}
645EXPORT_SYMBOL(get_super_thawed);
646
647/**
648 * get_active_super - get an active reference to the superblock of a device
649 * @bdev: device to get the superblock for
650 *
651 * Scans the superblock list and finds the superblock of the file system
652 * mounted on the device given. Returns the superblock with an active
653 * reference or %NULL if none was found.
654 */
655struct super_block *get_active_super(struct block_device *bdev)
656{
657 struct super_block *sb;
658
659 if (!bdev)
660 return NULL;
661
662restart:
663 spin_lock(&sb_lock);
664 list_for_each_entry(sb, &super_blocks, s_list) {
665 if (hlist_unhashed(&sb->s_instances))
666 continue;
667 if (sb->s_bdev == bdev) {
668 if (!grab_super(sb))
669 goto restart;
670 up_write(&sb->s_umount);
671 return sb;
672 }
673 }
674 spin_unlock(&sb_lock);
675 return NULL;
676}
677
678struct super_block *user_get_super(dev_t dev)
679{
680 struct super_block *sb;
681
682 spin_lock(&sb_lock);
683rescan:
684 list_for_each_entry(sb, &super_blocks, s_list) {
685 if (hlist_unhashed(&sb->s_instances))
686 continue;
687 if (sb->s_dev == dev) {
688 sb->s_count++;
689 spin_unlock(&sb_lock);
690 down_read(&sb->s_umount);
691 /* still alive? */
692 if (sb->s_root && (sb->s_flags & MS_BORN))
693 return sb;
694 up_read(&sb->s_umount);
695 /* nope, got unmounted */
696 spin_lock(&sb_lock);
697 __put_super(sb);
698 goto rescan;
699 }
700 }
701 spin_unlock(&sb_lock);
702 return NULL;
703}
704
705/**
706 * do_remount_sb - asks filesystem to change mount options.
707 * @sb: superblock in question
708 * @flags: numeric part of options
709 * @data: the rest of options
710 * @force: whether or not to force the change
711 *
712 * Alters the mount options of a mounted file system.
713 */
714int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
715{
716 int retval;
717 int remount_ro;
718
719 if (sb->s_writers.frozen != SB_UNFROZEN)
720 return -EBUSY;
721
722#ifdef CONFIG_BLOCK
723 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
724 return -EACCES;
725#endif
726
727 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
728
729 if (remount_ro) {
730 if (!hlist_empty(&sb->s_pins)) {
731 up_write(&sb->s_umount);
732 group_pin_kill(&sb->s_pins);
733 down_write(&sb->s_umount);
734 if (!sb->s_root)
735 return 0;
736 if (sb->s_writers.frozen != SB_UNFROZEN)
737 return -EBUSY;
738 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
739 }
740 }
741 shrink_dcache_sb(sb);
742
743 /* If we are remounting RDONLY and current sb is read/write,
744 make sure there are no rw files opened */
745 if (remount_ro) {
746 if (force) {
747 sb->s_readonly_remount = 1;
748 smp_wmb();
749 } else {
750 retval = sb_prepare_remount_readonly(sb);
751 if (retval)
752 return retval;
753 }
754 }
755
756 if (sb->s_op->remount_fs) {
757 retval = sb->s_op->remount_fs(sb, &flags, data);
758 if (retval) {
759 if (!force)
760 goto cancel_readonly;
761 /* If forced remount, go ahead despite any errors */
762 WARN(1, "forced remount of a %s fs returned %i\n",
763 sb->s_type->name, retval);
764 }
765 }
766 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
767 /* Needs to be ordered wrt mnt_is_readonly() */
768 smp_wmb();
769 sb->s_readonly_remount = 0;
770
771 /*
772 * Some filesystems modify their metadata via some other path than the
773 * bdev buffer cache (eg. use a private mapping, or directories in
774 * pagecache, etc). Also file data modifications go via their own
775 * mappings. So If we try to mount readonly then copy the filesystem
776 * from bdev, we could get stale data, so invalidate it to give a best
777 * effort at coherency.
778 */
779 if (remount_ro && sb->s_bdev)
780 invalidate_bdev(sb->s_bdev);
781 return 0;
782
783cancel_readonly:
784 sb->s_readonly_remount = 0;
785 return retval;
786}
787
788static void do_emergency_remount(struct work_struct *work)
789{
790 struct super_block *sb, *p = NULL;
791
792 spin_lock(&sb_lock);
793 list_for_each_entry(sb, &super_blocks, s_list) {
794 if (hlist_unhashed(&sb->s_instances))
795 continue;
796 sb->s_count++;
797 spin_unlock(&sb_lock);
798 down_write(&sb->s_umount);
799 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
800 !(sb->s_flags & MS_RDONLY)) {
801 /*
802 * What lock protects sb->s_flags??
803 */
804 do_remount_sb(sb, MS_RDONLY, NULL, 1);
805 }
806 up_write(&sb->s_umount);
807 spin_lock(&sb_lock);
808 if (p)
809 __put_super(p);
810 p = sb;
811 }
812 if (p)
813 __put_super(p);
814 spin_unlock(&sb_lock);
815 kfree(work);
816 printk("Emergency Remount complete\n");
817}
818
819void emergency_remount(void)
820{
821 struct work_struct *work;
822
823 work = kmalloc(sizeof(*work), GFP_ATOMIC);
824 if (work) {
825 INIT_WORK(work, do_emergency_remount);
826 schedule_work(work);
827 }
828}
829
830/*
831 * Unnamed block devices are dummy devices used by virtual
832 * filesystems which don't use real block-devices. -- jrs
833 */
834
835static DEFINE_IDA(unnamed_dev_ida);
836static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
837/* Many userspace utilities consider an FSID of 0 invalid.
838 * Always return at least 1 from get_anon_bdev.
839 */
840static int unnamed_dev_start = 1;
841
842int get_anon_bdev(dev_t *p)
843{
844 int dev;
845 int error;
846
847 retry:
848 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
849 return -ENOMEM;
850 spin_lock(&unnamed_dev_lock);
851 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
852 if (!error)
853 unnamed_dev_start = dev + 1;
854 spin_unlock(&unnamed_dev_lock);
855 if (error == -EAGAIN)
856 /* We raced and lost with another CPU. */
857 goto retry;
858 else if (error)
859 return -EAGAIN;
860
861 if (dev >= (1 << MINORBITS)) {
862 spin_lock(&unnamed_dev_lock);
863 ida_remove(&unnamed_dev_ida, dev);
864 if (unnamed_dev_start > dev)
865 unnamed_dev_start = dev;
866 spin_unlock(&unnamed_dev_lock);
867 return -EMFILE;
868 }
869 *p = MKDEV(0, dev & MINORMASK);
870 return 0;
871}
872EXPORT_SYMBOL(get_anon_bdev);
873
874void free_anon_bdev(dev_t dev)
875{
876 int slot = MINOR(dev);
877 spin_lock(&unnamed_dev_lock);
878 ida_remove(&unnamed_dev_ida, slot);
879 if (slot < unnamed_dev_start)
880 unnamed_dev_start = slot;
881 spin_unlock(&unnamed_dev_lock);
882}
883EXPORT_SYMBOL(free_anon_bdev);
884
885int set_anon_super(struct super_block *s, void *data)
886{
887 return get_anon_bdev(&s->s_dev);
888}
889
890EXPORT_SYMBOL(set_anon_super);
891
892void kill_anon_super(struct super_block *sb)
893{
894 dev_t dev = sb->s_dev;
895 generic_shutdown_super(sb);
896 free_anon_bdev(dev);
897}
898
899EXPORT_SYMBOL(kill_anon_super);
900
901void kill_litter_super(struct super_block *sb)
902{
903 if (sb->s_root)
904 d_genocide(sb->s_root);
905 kill_anon_super(sb);
906}
907
908EXPORT_SYMBOL(kill_litter_super);
909
910static int ns_test_super(struct super_block *sb, void *data)
911{
912 return sb->s_fs_info == data;
913}
914
915static int ns_set_super(struct super_block *sb, void *data)
916{
917 sb->s_fs_info = data;
918 return set_anon_super(sb, NULL);
919}
920
921struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
922 void *data, int (*fill_super)(struct super_block *, void *, int))
923{
924 struct super_block *sb;
925
926 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
927 if (IS_ERR(sb))
928 return ERR_CAST(sb);
929
930 if (!sb->s_root) {
931 int err;
932 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
933 if (err) {
934 deactivate_locked_super(sb);
935 return ERR_PTR(err);
936 }
937
938 sb->s_flags |= MS_ACTIVE;
939 }
940
941 return dget(sb->s_root);
942}
943
944EXPORT_SYMBOL(mount_ns);
945
946#ifdef CONFIG_BLOCK
947static int set_bdev_super(struct super_block *s, void *data)
948{
949 s->s_bdev = data;
950 s->s_dev = s->s_bdev->bd_dev;
951
952 /*
953 * We set the bdi here to the queue backing, file systems can
954 * overwrite this in ->fill_super()
955 */
956 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
957 return 0;
958}
959
960static int test_bdev_super(struct super_block *s, void *data)
961{
962 return (void *)s->s_bdev == data;
963}
964
965struct dentry *mount_bdev(struct file_system_type *fs_type,
966 int flags, const char *dev_name, void *data,
967 int (*fill_super)(struct super_block *, void *, int))
968{
969 struct block_device *bdev;
970 struct super_block *s;
971 fmode_t mode = FMODE_READ | FMODE_EXCL;
972 int error = 0;
973
974 if (!(flags & MS_RDONLY))
975 mode |= FMODE_WRITE;
976
977 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
978 if (IS_ERR(bdev))
979 return ERR_CAST(bdev);
980
981 /*
982 * once the super is inserted into the list by sget, s_umount
983 * will protect the lockfs code from trying to start a snapshot
984 * while we are mounting
985 */
986 mutex_lock(&bdev->bd_fsfreeze_mutex);
987 if (bdev->bd_fsfreeze_count > 0) {
988 mutex_unlock(&bdev->bd_fsfreeze_mutex);
989 error = -EBUSY;
990 goto error_bdev;
991 }
992 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
993 bdev);
994 mutex_unlock(&bdev->bd_fsfreeze_mutex);
995 if (IS_ERR(s))
996 goto error_s;
997
998 if (s->s_root) {
999 if ((flags ^ s->s_flags) & MS_RDONLY) {
1000 deactivate_locked_super(s);
1001 error = -EBUSY;
1002 goto error_bdev;
1003 }
1004
1005 /*
1006 * s_umount nests inside bd_mutex during
1007 * __invalidate_device(). blkdev_put() acquires
1008 * bd_mutex and can't be called under s_umount. Drop
1009 * s_umount temporarily. This is safe as we're
1010 * holding an active reference.
1011 */
1012 up_write(&s->s_umount);
1013 blkdev_put(bdev, mode);
1014 down_write(&s->s_umount);
1015 } else {
1016 s->s_mode = mode;
1017 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1018 sb_set_blocksize(s, block_size(bdev));
1019 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1020 if (error) {
1021 deactivate_locked_super(s);
1022 goto error;
1023 }
1024
1025 s->s_flags |= MS_ACTIVE;
1026 bdev->bd_super = s;
1027 }
1028
1029 return dget(s->s_root);
1030
1031error_s:
1032 error = PTR_ERR(s);
1033error_bdev:
1034 blkdev_put(bdev, mode);
1035error:
1036 return ERR_PTR(error);
1037}
1038EXPORT_SYMBOL(mount_bdev);
1039
1040void kill_block_super(struct super_block *sb)
1041{
1042 struct block_device *bdev = sb->s_bdev;
1043 fmode_t mode = sb->s_mode;
1044
1045 bdev->bd_super = NULL;
1046 generic_shutdown_super(sb);
1047 sync_blockdev(bdev);
1048 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1049 blkdev_put(bdev, mode | FMODE_EXCL);
1050}
1051
1052EXPORT_SYMBOL(kill_block_super);
1053#endif
1054
1055struct dentry *mount_nodev(struct file_system_type *fs_type,
1056 int flags, void *data,
1057 int (*fill_super)(struct super_block *, void *, int))
1058{
1059 int error;
1060 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1061
1062 if (IS_ERR(s))
1063 return ERR_CAST(s);
1064
1065 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1066 if (error) {
1067 deactivate_locked_super(s);
1068 return ERR_PTR(error);
1069 }
1070 s->s_flags |= MS_ACTIVE;
1071 return dget(s->s_root);
1072}
1073EXPORT_SYMBOL(mount_nodev);
1074
1075static int compare_single(struct super_block *s, void *p)
1076{
1077 return 1;
1078}
1079
1080struct dentry *mount_single(struct file_system_type *fs_type,
1081 int flags, void *data,
1082 int (*fill_super)(struct super_block *, void *, int))
1083{
1084 struct super_block *s;
1085 int error;
1086
1087 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1088 if (IS_ERR(s))
1089 return ERR_CAST(s);
1090 if (!s->s_root) {
1091 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1092 if (error) {
1093 deactivate_locked_super(s);
1094 return ERR_PTR(error);
1095 }
1096 s->s_flags |= MS_ACTIVE;
1097 } else {
1098 do_remount_sb(s, flags, data, 0);
1099 }
1100 return dget(s->s_root);
1101}
1102EXPORT_SYMBOL(mount_single);
1103
1104struct dentry *
1105mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1106{
1107 struct dentry *root;
1108 struct super_block *sb;
1109 char *secdata = NULL;
1110 int error = -ENOMEM;
1111
1112 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1113 secdata = alloc_secdata();
1114 if (!secdata)
1115 goto out;
1116
1117 error = security_sb_copy_data(data, secdata);
1118 if (error)
1119 goto out_free_secdata;
1120 }
1121
1122 root = type->mount(type, flags, name, data);
1123 if (IS_ERR(root)) {
1124 error = PTR_ERR(root);
1125 goto out_free_secdata;
1126 }
1127 sb = root->d_sb;
1128 BUG_ON(!sb);
1129 WARN_ON(!sb->s_bdi);
1130 sb->s_flags |= MS_BORN;
1131
1132 error = security_sb_kern_mount(sb, flags, secdata);
1133 if (error)
1134 goto out_sb;
1135
1136 /*
1137 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1138 * but s_maxbytes was an unsigned long long for many releases. Throw
1139 * this warning for a little while to try and catch filesystems that
1140 * violate this rule.
1141 */
1142 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1143 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1144
1145 up_write(&sb->s_umount);
1146 free_secdata(secdata);
1147 return root;
1148out_sb:
1149 dput(root);
1150 deactivate_locked_super(sb);
1151out_free_secdata:
1152 free_secdata(secdata);
1153out:
1154 return ERR_PTR(error);
1155}
1156
1157/*
1158 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1159 * instead.
1160 */
1161void __sb_end_write(struct super_block *sb, int level)
1162{
1163 percpu_up_read(sb->s_writers.rw_sem + level-1);
1164}
1165EXPORT_SYMBOL(__sb_end_write);
1166
1167/*
1168 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1169 * instead.
1170 */
1171int __sb_start_write(struct super_block *sb, int level, bool wait)
1172{
1173 bool force_trylock = false;
1174 int ret = 1;
1175
1176#ifdef CONFIG_LOCKDEP
1177 /*
1178 * We want lockdep to tell us about possible deadlocks with freezing
1179 * but it's it bit tricky to properly instrument it. Getting a freeze
1180 * protection works as getting a read lock but there are subtle
1181 * problems. XFS for example gets freeze protection on internal level
1182 * twice in some cases, which is OK only because we already hold a
1183 * freeze protection also on higher level. Due to these cases we have
1184 * to use wait == F (trylock mode) which must not fail.
1185 */
1186 if (wait) {
1187 int i;
1188
1189 for (i = 0; i < level - 1; i++)
1190 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
1191 force_trylock = true;
1192 break;
1193 }
1194 }
1195#endif
1196 if (wait && !force_trylock)
1197 percpu_down_read(sb->s_writers.rw_sem + level-1);
1198 else
1199 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1200
1201 WARN_ON(force_trylock && !ret);
1202 return ret;
1203}
1204EXPORT_SYMBOL(__sb_start_write);
1205
1206/**
1207 * sb_wait_write - wait until all writers to given file system finish
1208 * @sb: the super for which we wait
1209 * @level: type of writers we wait for (normal vs page fault)
1210 *
1211 * This function waits until there are no writers of given type to given file
1212 * system.
1213 */
1214static void sb_wait_write(struct super_block *sb, int level)
1215{
1216 percpu_down_write(sb->s_writers.rw_sem + level-1);
1217 /*
1218 * We are going to return to userspace and forget about this lock, the
1219 * ownership goes to the caller of thaw_super() which does unlock.
1220 *
1221 * FIXME: we should do this before return from freeze_super() after we
1222 * called sync_filesystem(sb) and s_op->freeze_fs(sb), and thaw_super()
1223 * should re-acquire these locks before s_op->unfreeze_fs(sb). However
1224 * this leads to lockdep false-positives, so currently we do the early
1225 * release right after acquire.
1226 */
1227 percpu_rwsem_release(sb->s_writers.rw_sem + level-1, 0, _THIS_IP_);
1228}
1229
1230static void sb_freeze_unlock(struct super_block *sb)
1231{
1232 int level;
1233
1234 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1235 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1236
1237 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1238 percpu_up_write(sb->s_writers.rw_sem + level);
1239}
1240
1241/**
1242 * freeze_super - lock the filesystem and force it into a consistent state
1243 * @sb: the super to lock
1244 *
1245 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1246 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1247 * -EBUSY.
1248 *
1249 * During this function, sb->s_writers.frozen goes through these values:
1250 *
1251 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1252 *
1253 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1254 * writes should be blocked, though page faults are still allowed. We wait for
1255 * all writes to complete and then proceed to the next stage.
1256 *
1257 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1258 * but internal fs threads can still modify the filesystem (although they
1259 * should not dirty new pages or inodes), writeback can run etc. After waiting
1260 * for all running page faults we sync the filesystem which will clean all
1261 * dirty pages and inodes (no new dirty pages or inodes can be created when
1262 * sync is running).
1263 *
1264 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1265 * modification are blocked (e.g. XFS preallocation truncation on inode
1266 * reclaim). This is usually implemented by blocking new transactions for
1267 * filesystems that have them and need this additional guard. After all
1268 * internal writers are finished we call ->freeze_fs() to finish filesystem
1269 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1270 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1271 *
1272 * sb->s_writers.frozen is protected by sb->s_umount.
1273 */
1274int freeze_super(struct super_block *sb)
1275{
1276 int ret;
1277
1278 atomic_inc(&sb->s_active);
1279 down_write(&sb->s_umount);
1280 if (sb->s_writers.frozen != SB_UNFROZEN) {
1281 deactivate_locked_super(sb);
1282 return -EBUSY;
1283 }
1284
1285 if (!(sb->s_flags & MS_BORN)) {
1286 up_write(&sb->s_umount);
1287 return 0; /* sic - it's "nothing to do" */
1288 }
1289
1290 if (sb->s_flags & MS_RDONLY) {
1291 /* Nothing to do really... */
1292 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1293 up_write(&sb->s_umount);
1294 return 0;
1295 }
1296
1297 sb->s_writers.frozen = SB_FREEZE_WRITE;
1298 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1299 up_write(&sb->s_umount);
1300 sb_wait_write(sb, SB_FREEZE_WRITE);
1301 down_write(&sb->s_umount);
1302
1303 /* Now we go and block page faults... */
1304 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1305 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1306
1307 /* All writers are done so after syncing there won't be dirty data */
1308 sync_filesystem(sb);
1309
1310 /* Now wait for internal filesystem counter */
1311 sb->s_writers.frozen = SB_FREEZE_FS;
1312 sb_wait_write(sb, SB_FREEZE_FS);
1313
1314 if (sb->s_op->freeze_fs) {
1315 ret = sb->s_op->freeze_fs(sb);
1316 if (ret) {
1317 printk(KERN_ERR
1318 "VFS:Filesystem freeze failed\n");
1319 sb->s_writers.frozen = SB_UNFROZEN;
1320 sb_freeze_unlock(sb);
1321 wake_up(&sb->s_writers.wait_unfrozen);
1322 deactivate_locked_super(sb);
1323 return ret;
1324 }
1325 }
1326 /*
1327 * This is just for debugging purposes so that fs can warn if it
1328 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1329 */
1330 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1331 up_write(&sb->s_umount);
1332 return 0;
1333}
1334EXPORT_SYMBOL(freeze_super);
1335
1336/**
1337 * thaw_super -- unlock filesystem
1338 * @sb: the super to thaw
1339 *
1340 * Unlocks the filesystem and marks it writeable again after freeze_super().
1341 */
1342int thaw_super(struct super_block *sb)
1343{
1344 int error;
1345
1346 down_write(&sb->s_umount);
1347 if (sb->s_writers.frozen == SB_UNFROZEN) {
1348 up_write(&sb->s_umount);
1349 return -EINVAL;
1350 }
1351
1352 if (sb->s_flags & MS_RDONLY) {
1353 sb->s_writers.frozen = SB_UNFROZEN;
1354 goto out;
1355 }
1356
1357 if (sb->s_op->unfreeze_fs) {
1358 error = sb->s_op->unfreeze_fs(sb);
1359 if (error) {
1360 printk(KERN_ERR
1361 "VFS:Filesystem thaw failed\n");
1362 up_write(&sb->s_umount);
1363 return error;
1364 }
1365 }
1366
1367 sb->s_writers.frozen = SB_UNFROZEN;
1368 sb_freeze_unlock(sb);
1369out:
1370 wake_up(&sb->s_writers.wait_unfrozen);
1371 deactivate_locked_super(sb);
1372 return 0;
1373}
1374EXPORT_SYMBOL(thaw_super);
1/*
2 * linux/fs/super.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * super.c contains code to handle: - mount structures
7 * - super-block tables
8 * - filesystem drivers list
9 * - mount system call
10 * - umount system call
11 * - ustat system call
12 *
13 * GK 2/5/95 - Changed to support mounting the root fs via NFS
14 *
15 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall
16 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96
17 * Added options to /proc/mounts:
18 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
19 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
20 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
21 */
22
23#include <linux/module.h>
24#include <linux/slab.h>
25#include <linux/acct.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/cleancache.h>
35#include "internal.h"
36
37
38LIST_HEAD(super_blocks);
39DEFINE_SPINLOCK(sb_lock);
40
41/*
42 * One thing we have to be careful of with a per-sb shrinker is that we don't
43 * drop the last active reference to the superblock from within the shrinker.
44 * If that happens we could trigger unregistering the shrinker from within the
45 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
46 * take a passive reference to the superblock to avoid this from occurring.
47 */
48static int prune_super(struct shrinker *shrink, struct shrink_control *sc)
49{
50 struct super_block *sb;
51 int fs_objects = 0;
52 int total_objects;
53
54 sb = container_of(shrink, struct super_block, s_shrink);
55
56 /*
57 * Deadlock avoidance. We may hold various FS locks, and we don't want
58 * to recurse into the FS that called us in clear_inode() and friends..
59 */
60 if (sc->nr_to_scan && !(sc->gfp_mask & __GFP_FS))
61 return -1;
62
63 if (!grab_super_passive(sb))
64 return -1;
65
66 if (sb->s_op && sb->s_op->nr_cached_objects)
67 fs_objects = sb->s_op->nr_cached_objects(sb);
68
69 total_objects = sb->s_nr_dentry_unused +
70 sb->s_nr_inodes_unused + fs_objects + 1;
71
72 if (sc->nr_to_scan) {
73 int dentries;
74 int inodes;
75
76 /* proportion the scan between the caches */
77 dentries = (sc->nr_to_scan * sb->s_nr_dentry_unused) /
78 total_objects;
79 inodes = (sc->nr_to_scan * sb->s_nr_inodes_unused) /
80 total_objects;
81 if (fs_objects)
82 fs_objects = (sc->nr_to_scan * fs_objects) /
83 total_objects;
84 /*
85 * prune the dcache first as the icache is pinned by it, then
86 * prune the icache, followed by the filesystem specific caches
87 */
88 prune_dcache_sb(sb, dentries);
89 prune_icache_sb(sb, inodes);
90
91 if (fs_objects && sb->s_op->free_cached_objects) {
92 sb->s_op->free_cached_objects(sb, fs_objects);
93 fs_objects = sb->s_op->nr_cached_objects(sb);
94 }
95 total_objects = sb->s_nr_dentry_unused +
96 sb->s_nr_inodes_unused + fs_objects;
97 }
98
99 total_objects = (total_objects / 100) * sysctl_vfs_cache_pressure;
100 drop_super(sb);
101 return total_objects;
102}
103
104/**
105 * alloc_super - create new superblock
106 * @type: filesystem type superblock should belong to
107 *
108 * Allocates and initializes a new &struct super_block. alloc_super()
109 * returns a pointer new superblock or %NULL if allocation had failed.
110 */
111static struct super_block *alloc_super(struct file_system_type *type)
112{
113 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
114 static const struct super_operations default_op;
115
116 if (s) {
117 if (security_sb_alloc(s)) {
118 kfree(s);
119 s = NULL;
120 goto out;
121 }
122#ifdef CONFIG_SMP
123 s->s_files = alloc_percpu(struct list_head);
124 if (!s->s_files) {
125 security_sb_free(s);
126 kfree(s);
127 s = NULL;
128 goto out;
129 } else {
130 int i;
131
132 for_each_possible_cpu(i)
133 INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i));
134 }
135#else
136 INIT_LIST_HEAD(&s->s_files);
137#endif
138 s->s_bdi = &default_backing_dev_info;
139 INIT_LIST_HEAD(&s->s_instances);
140 INIT_HLIST_BL_HEAD(&s->s_anon);
141 INIT_LIST_HEAD(&s->s_inodes);
142 INIT_LIST_HEAD(&s->s_dentry_lru);
143 INIT_LIST_HEAD(&s->s_inode_lru);
144 spin_lock_init(&s->s_inode_lru_lock);
145 init_rwsem(&s->s_umount);
146 mutex_init(&s->s_lock);
147 lockdep_set_class(&s->s_umount, &type->s_umount_key);
148 /*
149 * The locking rules for s_lock are up to the
150 * filesystem. For example ext3fs has different
151 * lock ordering than usbfs:
152 */
153 lockdep_set_class(&s->s_lock, &type->s_lock_key);
154 /*
155 * sget() can have s_umount recursion.
156 *
157 * When it cannot find a suitable sb, it allocates a new
158 * one (this one), and tries again to find a suitable old
159 * one.
160 *
161 * In case that succeeds, it will acquire the s_umount
162 * lock of the old one. Since these are clearly distrinct
163 * locks, and this object isn't exposed yet, there's no
164 * risk of deadlocks.
165 *
166 * Annotate this by putting this lock in a different
167 * subclass.
168 */
169 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
170 s->s_count = 1;
171 atomic_set(&s->s_active, 1);
172 mutex_init(&s->s_vfs_rename_mutex);
173 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
174 mutex_init(&s->s_dquot.dqio_mutex);
175 mutex_init(&s->s_dquot.dqonoff_mutex);
176 init_rwsem(&s->s_dquot.dqptr_sem);
177 init_waitqueue_head(&s->s_wait_unfrozen);
178 s->s_maxbytes = MAX_NON_LFS;
179 s->s_op = &default_op;
180 s->s_time_gran = 1000000000;
181 s->cleancache_poolid = -1;
182
183 s->s_shrink.seeks = DEFAULT_SEEKS;
184 s->s_shrink.shrink = prune_super;
185 s->s_shrink.batch = 1024;
186 }
187out:
188 return s;
189}
190
191/**
192 * destroy_super - frees a superblock
193 * @s: superblock to free
194 *
195 * Frees a superblock.
196 */
197static inline void destroy_super(struct super_block *s)
198{
199#ifdef CONFIG_SMP
200 free_percpu(s->s_files);
201#endif
202 security_sb_free(s);
203 kfree(s->s_subtype);
204 kfree(s->s_options);
205 kfree(s);
206}
207
208/* Superblock refcounting */
209
210/*
211 * Drop a superblock's refcount. The caller must hold sb_lock.
212 */
213void __put_super(struct super_block *sb)
214{
215 if (!--sb->s_count) {
216 list_del_init(&sb->s_list);
217 destroy_super(sb);
218 }
219}
220
221/**
222 * put_super - drop a temporary reference to superblock
223 * @sb: superblock in question
224 *
225 * Drops a temporary reference, frees superblock if there's no
226 * references left.
227 */
228void put_super(struct super_block *sb)
229{
230 spin_lock(&sb_lock);
231 __put_super(sb);
232 spin_unlock(&sb_lock);
233}
234
235
236/**
237 * deactivate_locked_super - drop an active reference to superblock
238 * @s: superblock to deactivate
239 *
240 * Drops an active reference to superblock, converting it into a temprory
241 * one if there is no other active references left. In that case we
242 * tell fs driver to shut it down and drop the temporary reference we
243 * had just acquired.
244 *
245 * Caller holds exclusive lock on superblock; that lock is released.
246 */
247void deactivate_locked_super(struct super_block *s)
248{
249 struct file_system_type *fs = s->s_type;
250 if (atomic_dec_and_test(&s->s_active)) {
251 cleancache_flush_fs(s);
252 fs->kill_sb(s);
253
254 /* caches are now gone, we can safely kill the shrinker now */
255 unregister_shrinker(&s->s_shrink);
256
257 /*
258 * We need to call rcu_barrier so all the delayed rcu free
259 * inodes are flushed before we release the fs module.
260 */
261 rcu_barrier();
262 put_filesystem(fs);
263 put_super(s);
264 } else {
265 up_write(&s->s_umount);
266 }
267}
268
269EXPORT_SYMBOL(deactivate_locked_super);
270
271/**
272 * deactivate_super - drop an active reference to superblock
273 * @s: superblock to deactivate
274 *
275 * Variant of deactivate_locked_super(), except that superblock is *not*
276 * locked by caller. If we are going to drop the final active reference,
277 * lock will be acquired prior to that.
278 */
279void deactivate_super(struct super_block *s)
280{
281 if (!atomic_add_unless(&s->s_active, -1, 1)) {
282 down_write(&s->s_umount);
283 deactivate_locked_super(s);
284 }
285}
286
287EXPORT_SYMBOL(deactivate_super);
288
289/**
290 * grab_super - acquire an active reference
291 * @s: reference we are trying to make active
292 *
293 * Tries to acquire an active reference. grab_super() is used when we
294 * had just found a superblock in super_blocks or fs_type->fs_supers
295 * and want to turn it into a full-blown active reference. grab_super()
296 * is called with sb_lock held and drops it. Returns 1 in case of
297 * success, 0 if we had failed (superblock contents was already dead or
298 * dying when grab_super() had been called).
299 */
300static int grab_super(struct super_block *s) __releases(sb_lock)
301{
302 if (atomic_inc_not_zero(&s->s_active)) {
303 spin_unlock(&sb_lock);
304 return 1;
305 }
306 /* it's going away */
307 s->s_count++;
308 spin_unlock(&sb_lock);
309 /* wait for it to die */
310 down_write(&s->s_umount);
311 up_write(&s->s_umount);
312 put_super(s);
313 return 0;
314}
315
316/*
317 * grab_super_passive - acquire a passive reference
318 * @s: reference we are trying to grab
319 *
320 * Tries to acquire a passive reference. This is used in places where we
321 * cannot take an active reference but we need to ensure that the
322 * superblock does not go away while we are working on it. It returns
323 * false if a reference was not gained, and returns true with the s_umount
324 * lock held in read mode if a reference is gained. On successful return,
325 * the caller must drop the s_umount lock and the passive reference when
326 * done.
327 */
328bool grab_super_passive(struct super_block *sb)
329{
330 spin_lock(&sb_lock);
331 if (list_empty(&sb->s_instances)) {
332 spin_unlock(&sb_lock);
333 return false;
334 }
335
336 sb->s_count++;
337 spin_unlock(&sb_lock);
338
339 if (down_read_trylock(&sb->s_umount)) {
340 if (sb->s_root)
341 return true;
342 up_read(&sb->s_umount);
343 }
344
345 put_super(sb);
346 return false;
347}
348
349/*
350 * Superblock locking. We really ought to get rid of these two.
351 */
352void lock_super(struct super_block * sb)
353{
354 mutex_lock(&sb->s_lock);
355}
356
357void unlock_super(struct super_block * sb)
358{
359 mutex_unlock(&sb->s_lock);
360}
361
362EXPORT_SYMBOL(lock_super);
363EXPORT_SYMBOL(unlock_super);
364
365/**
366 * generic_shutdown_super - common helper for ->kill_sb()
367 * @sb: superblock to kill
368 *
369 * generic_shutdown_super() does all fs-independent work on superblock
370 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
371 * that need destruction out of superblock, call generic_shutdown_super()
372 * and release aforementioned objects. Note: dentries and inodes _are_
373 * taken care of and do not need specific handling.
374 *
375 * Upon calling this function, the filesystem may no longer alter or
376 * rearrange the set of dentries belonging to this super_block, nor may it
377 * change the attachments of dentries to inodes.
378 */
379void generic_shutdown_super(struct super_block *sb)
380{
381 const struct super_operations *sop = sb->s_op;
382
383 if (sb->s_root) {
384 shrink_dcache_for_umount(sb);
385 sync_filesystem(sb);
386 sb->s_flags &= ~MS_ACTIVE;
387
388 fsnotify_unmount_inodes(&sb->s_inodes);
389
390 evict_inodes(sb);
391
392 if (sop->put_super)
393 sop->put_super(sb);
394
395 if (!list_empty(&sb->s_inodes)) {
396 printk("VFS: Busy inodes after unmount of %s. "
397 "Self-destruct in 5 seconds. Have a nice day...\n",
398 sb->s_id);
399 }
400 }
401 spin_lock(&sb_lock);
402 /* should be initialized for __put_super_and_need_restart() */
403 list_del_init(&sb->s_instances);
404 spin_unlock(&sb_lock);
405 up_write(&sb->s_umount);
406}
407
408EXPORT_SYMBOL(generic_shutdown_super);
409
410/**
411 * sget - find or create a superblock
412 * @type: filesystem type superblock should belong to
413 * @test: comparison callback
414 * @set: setup callback
415 * @data: argument to each of them
416 */
417struct super_block *sget(struct file_system_type *type,
418 int (*test)(struct super_block *,void *),
419 int (*set)(struct super_block *,void *),
420 void *data)
421{
422 struct super_block *s = NULL;
423 struct super_block *old;
424 int err;
425
426retry:
427 spin_lock(&sb_lock);
428 if (test) {
429 list_for_each_entry(old, &type->fs_supers, s_instances) {
430 if (!test(old, data))
431 continue;
432 if (!grab_super(old))
433 goto retry;
434 if (s) {
435 up_write(&s->s_umount);
436 destroy_super(s);
437 s = NULL;
438 }
439 down_write(&old->s_umount);
440 if (unlikely(!(old->s_flags & MS_BORN))) {
441 deactivate_locked_super(old);
442 goto retry;
443 }
444 return old;
445 }
446 }
447 if (!s) {
448 spin_unlock(&sb_lock);
449 s = alloc_super(type);
450 if (!s)
451 return ERR_PTR(-ENOMEM);
452 goto retry;
453 }
454
455 err = set(s, data);
456 if (err) {
457 spin_unlock(&sb_lock);
458 up_write(&s->s_umount);
459 destroy_super(s);
460 return ERR_PTR(err);
461 }
462 s->s_type = type;
463 strlcpy(s->s_id, type->name, sizeof(s->s_id));
464 list_add_tail(&s->s_list, &super_blocks);
465 list_add(&s->s_instances, &type->fs_supers);
466 spin_unlock(&sb_lock);
467 get_filesystem(type);
468 register_shrinker(&s->s_shrink);
469 return s;
470}
471
472EXPORT_SYMBOL(sget);
473
474void drop_super(struct super_block *sb)
475{
476 up_read(&sb->s_umount);
477 put_super(sb);
478}
479
480EXPORT_SYMBOL(drop_super);
481
482/**
483 * sync_supers - helper for periodic superblock writeback
484 *
485 * Call the write_super method if present on all dirty superblocks in
486 * the system. This is for the periodic writeback used by most older
487 * filesystems. For data integrity superblock writeback use
488 * sync_filesystems() instead.
489 *
490 * Note: check the dirty flag before waiting, so we don't
491 * hold up the sync while mounting a device. (The newly
492 * mounted device won't need syncing.)
493 */
494void sync_supers(void)
495{
496 struct super_block *sb, *p = NULL;
497
498 spin_lock(&sb_lock);
499 list_for_each_entry(sb, &super_blocks, s_list) {
500 if (list_empty(&sb->s_instances))
501 continue;
502 if (sb->s_op->write_super && sb->s_dirt) {
503 sb->s_count++;
504 spin_unlock(&sb_lock);
505
506 down_read(&sb->s_umount);
507 if (sb->s_root && sb->s_dirt)
508 sb->s_op->write_super(sb);
509 up_read(&sb->s_umount);
510
511 spin_lock(&sb_lock);
512 if (p)
513 __put_super(p);
514 p = sb;
515 }
516 }
517 if (p)
518 __put_super(p);
519 spin_unlock(&sb_lock);
520}
521
522/**
523 * iterate_supers - call function for all active superblocks
524 * @f: function to call
525 * @arg: argument to pass to it
526 *
527 * Scans the superblock list and calls given function, passing it
528 * locked superblock and given argument.
529 */
530void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
531{
532 struct super_block *sb, *p = NULL;
533
534 spin_lock(&sb_lock);
535 list_for_each_entry(sb, &super_blocks, s_list) {
536 if (list_empty(&sb->s_instances))
537 continue;
538 sb->s_count++;
539 spin_unlock(&sb_lock);
540
541 down_read(&sb->s_umount);
542 if (sb->s_root)
543 f(sb, arg);
544 up_read(&sb->s_umount);
545
546 spin_lock(&sb_lock);
547 if (p)
548 __put_super(p);
549 p = sb;
550 }
551 if (p)
552 __put_super(p);
553 spin_unlock(&sb_lock);
554}
555
556/**
557 * iterate_supers_type - call function for superblocks of given type
558 * @type: fs type
559 * @f: function to call
560 * @arg: argument to pass to it
561 *
562 * Scans the superblock list and calls given function, passing it
563 * locked superblock and given argument.
564 */
565void iterate_supers_type(struct file_system_type *type,
566 void (*f)(struct super_block *, void *), void *arg)
567{
568 struct super_block *sb, *p = NULL;
569
570 spin_lock(&sb_lock);
571 list_for_each_entry(sb, &type->fs_supers, s_instances) {
572 sb->s_count++;
573 spin_unlock(&sb_lock);
574
575 down_read(&sb->s_umount);
576 if (sb->s_root)
577 f(sb, arg);
578 up_read(&sb->s_umount);
579
580 spin_lock(&sb_lock);
581 if (p)
582 __put_super(p);
583 p = sb;
584 }
585 if (p)
586 __put_super(p);
587 spin_unlock(&sb_lock);
588}
589
590EXPORT_SYMBOL(iterate_supers_type);
591
592/**
593 * get_super - get the superblock of a device
594 * @bdev: device to get the superblock for
595 *
596 * Scans the superblock list and finds the superblock of the file system
597 * mounted on the device given. %NULL is returned if no match is found.
598 */
599
600struct super_block *get_super(struct block_device *bdev)
601{
602 struct super_block *sb;
603
604 if (!bdev)
605 return NULL;
606
607 spin_lock(&sb_lock);
608rescan:
609 list_for_each_entry(sb, &super_blocks, s_list) {
610 if (list_empty(&sb->s_instances))
611 continue;
612 if (sb->s_bdev == bdev) {
613 sb->s_count++;
614 spin_unlock(&sb_lock);
615 down_read(&sb->s_umount);
616 /* still alive? */
617 if (sb->s_root)
618 return sb;
619 up_read(&sb->s_umount);
620 /* nope, got unmounted */
621 spin_lock(&sb_lock);
622 __put_super(sb);
623 goto rescan;
624 }
625 }
626 spin_unlock(&sb_lock);
627 return NULL;
628}
629
630EXPORT_SYMBOL(get_super);
631
632/**
633 * get_active_super - get an active reference to the superblock of a device
634 * @bdev: device to get the superblock for
635 *
636 * Scans the superblock list and finds the superblock of the file system
637 * mounted on the device given. Returns the superblock with an active
638 * reference or %NULL if none was found.
639 */
640struct super_block *get_active_super(struct block_device *bdev)
641{
642 struct super_block *sb;
643
644 if (!bdev)
645 return NULL;
646
647restart:
648 spin_lock(&sb_lock);
649 list_for_each_entry(sb, &super_blocks, s_list) {
650 if (list_empty(&sb->s_instances))
651 continue;
652 if (sb->s_bdev == bdev) {
653 if (grab_super(sb)) /* drops sb_lock */
654 return sb;
655 else
656 goto restart;
657 }
658 }
659 spin_unlock(&sb_lock);
660 return NULL;
661}
662
663struct super_block *user_get_super(dev_t dev)
664{
665 struct super_block *sb;
666
667 spin_lock(&sb_lock);
668rescan:
669 list_for_each_entry(sb, &super_blocks, s_list) {
670 if (list_empty(&sb->s_instances))
671 continue;
672 if (sb->s_dev == dev) {
673 sb->s_count++;
674 spin_unlock(&sb_lock);
675 down_read(&sb->s_umount);
676 /* still alive? */
677 if (sb->s_root)
678 return sb;
679 up_read(&sb->s_umount);
680 /* nope, got unmounted */
681 spin_lock(&sb_lock);
682 __put_super(sb);
683 goto rescan;
684 }
685 }
686 spin_unlock(&sb_lock);
687 return NULL;
688}
689
690/**
691 * do_remount_sb - asks filesystem to change mount options.
692 * @sb: superblock in question
693 * @flags: numeric part of options
694 * @data: the rest of options
695 * @force: whether or not to force the change
696 *
697 * Alters the mount options of a mounted file system.
698 */
699int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
700{
701 int retval;
702 int remount_ro;
703
704 if (sb->s_frozen != SB_UNFROZEN)
705 return -EBUSY;
706
707#ifdef CONFIG_BLOCK
708 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
709 return -EACCES;
710#endif
711
712 if (flags & MS_RDONLY)
713 acct_auto_close(sb);
714 shrink_dcache_sb(sb);
715 sync_filesystem(sb);
716
717 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
718
719 /* If we are remounting RDONLY and current sb is read/write,
720 make sure there are no rw files opened */
721 if (remount_ro) {
722 if (force)
723 mark_files_ro(sb);
724 else if (!fs_may_remount_ro(sb))
725 return -EBUSY;
726 }
727
728 if (sb->s_op->remount_fs) {
729 retval = sb->s_op->remount_fs(sb, &flags, data);
730 if (retval)
731 return retval;
732 }
733 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
734
735 /*
736 * Some filesystems modify their metadata via some other path than the
737 * bdev buffer cache (eg. use a private mapping, or directories in
738 * pagecache, etc). Also file data modifications go via their own
739 * mappings. So If we try to mount readonly then copy the filesystem
740 * from bdev, we could get stale data, so invalidate it to give a best
741 * effort at coherency.
742 */
743 if (remount_ro && sb->s_bdev)
744 invalidate_bdev(sb->s_bdev);
745 return 0;
746}
747
748static void do_emergency_remount(struct work_struct *work)
749{
750 struct super_block *sb, *p = NULL;
751
752 spin_lock(&sb_lock);
753 list_for_each_entry(sb, &super_blocks, s_list) {
754 if (list_empty(&sb->s_instances))
755 continue;
756 sb->s_count++;
757 spin_unlock(&sb_lock);
758 down_write(&sb->s_umount);
759 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
760 /*
761 * What lock protects sb->s_flags??
762 */
763 do_remount_sb(sb, MS_RDONLY, NULL, 1);
764 }
765 up_write(&sb->s_umount);
766 spin_lock(&sb_lock);
767 if (p)
768 __put_super(p);
769 p = sb;
770 }
771 if (p)
772 __put_super(p);
773 spin_unlock(&sb_lock);
774 kfree(work);
775 printk("Emergency Remount complete\n");
776}
777
778void emergency_remount(void)
779{
780 struct work_struct *work;
781
782 work = kmalloc(sizeof(*work), GFP_ATOMIC);
783 if (work) {
784 INIT_WORK(work, do_emergency_remount);
785 schedule_work(work);
786 }
787}
788
789/*
790 * Unnamed block devices are dummy devices used by virtual
791 * filesystems which don't use real block-devices. -- jrs
792 */
793
794static DEFINE_IDA(unnamed_dev_ida);
795static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
796static int unnamed_dev_start = 0; /* don't bother trying below it */
797
798int get_anon_bdev(dev_t *p)
799{
800 int dev;
801 int error;
802
803 retry:
804 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
805 return -ENOMEM;
806 spin_lock(&unnamed_dev_lock);
807 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
808 if (!error)
809 unnamed_dev_start = dev + 1;
810 spin_unlock(&unnamed_dev_lock);
811 if (error == -EAGAIN)
812 /* We raced and lost with another CPU. */
813 goto retry;
814 else if (error)
815 return -EAGAIN;
816
817 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
818 spin_lock(&unnamed_dev_lock);
819 ida_remove(&unnamed_dev_ida, dev);
820 if (unnamed_dev_start > dev)
821 unnamed_dev_start = dev;
822 spin_unlock(&unnamed_dev_lock);
823 return -EMFILE;
824 }
825 *p = MKDEV(0, dev & MINORMASK);
826 return 0;
827}
828EXPORT_SYMBOL(get_anon_bdev);
829
830void free_anon_bdev(dev_t dev)
831{
832 int slot = MINOR(dev);
833 spin_lock(&unnamed_dev_lock);
834 ida_remove(&unnamed_dev_ida, slot);
835 if (slot < unnamed_dev_start)
836 unnamed_dev_start = slot;
837 spin_unlock(&unnamed_dev_lock);
838}
839EXPORT_SYMBOL(free_anon_bdev);
840
841int set_anon_super(struct super_block *s, void *data)
842{
843 int error = get_anon_bdev(&s->s_dev);
844 if (!error)
845 s->s_bdi = &noop_backing_dev_info;
846 return error;
847}
848
849EXPORT_SYMBOL(set_anon_super);
850
851void kill_anon_super(struct super_block *sb)
852{
853 dev_t dev = sb->s_dev;
854 generic_shutdown_super(sb);
855 free_anon_bdev(dev);
856}
857
858EXPORT_SYMBOL(kill_anon_super);
859
860void kill_litter_super(struct super_block *sb)
861{
862 if (sb->s_root)
863 d_genocide(sb->s_root);
864 kill_anon_super(sb);
865}
866
867EXPORT_SYMBOL(kill_litter_super);
868
869static int ns_test_super(struct super_block *sb, void *data)
870{
871 return sb->s_fs_info == data;
872}
873
874static int ns_set_super(struct super_block *sb, void *data)
875{
876 sb->s_fs_info = data;
877 return set_anon_super(sb, NULL);
878}
879
880struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
881 void *data, int (*fill_super)(struct super_block *, void *, int))
882{
883 struct super_block *sb;
884
885 sb = sget(fs_type, ns_test_super, ns_set_super, data);
886 if (IS_ERR(sb))
887 return ERR_CAST(sb);
888
889 if (!sb->s_root) {
890 int err;
891 sb->s_flags = flags;
892 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
893 if (err) {
894 deactivate_locked_super(sb);
895 return ERR_PTR(err);
896 }
897
898 sb->s_flags |= MS_ACTIVE;
899 }
900
901 return dget(sb->s_root);
902}
903
904EXPORT_SYMBOL(mount_ns);
905
906#ifdef CONFIG_BLOCK
907static int set_bdev_super(struct super_block *s, void *data)
908{
909 s->s_bdev = data;
910 s->s_dev = s->s_bdev->bd_dev;
911
912 /*
913 * We set the bdi here to the queue backing, file systems can
914 * overwrite this in ->fill_super()
915 */
916 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
917 return 0;
918}
919
920static int test_bdev_super(struct super_block *s, void *data)
921{
922 return (void *)s->s_bdev == data;
923}
924
925struct dentry *mount_bdev(struct file_system_type *fs_type,
926 int flags, const char *dev_name, void *data,
927 int (*fill_super)(struct super_block *, void *, int))
928{
929 struct block_device *bdev;
930 struct super_block *s;
931 fmode_t mode = FMODE_READ | FMODE_EXCL;
932 int error = 0;
933
934 if (!(flags & MS_RDONLY))
935 mode |= FMODE_WRITE;
936
937 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
938 if (IS_ERR(bdev))
939 return ERR_CAST(bdev);
940
941 /*
942 * once the super is inserted into the list by sget, s_umount
943 * will protect the lockfs code from trying to start a snapshot
944 * while we are mounting
945 */
946 mutex_lock(&bdev->bd_fsfreeze_mutex);
947 if (bdev->bd_fsfreeze_count > 0) {
948 mutex_unlock(&bdev->bd_fsfreeze_mutex);
949 error = -EBUSY;
950 goto error_bdev;
951 }
952 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
953 mutex_unlock(&bdev->bd_fsfreeze_mutex);
954 if (IS_ERR(s))
955 goto error_s;
956
957 if (s->s_root) {
958 if ((flags ^ s->s_flags) & MS_RDONLY) {
959 deactivate_locked_super(s);
960 error = -EBUSY;
961 goto error_bdev;
962 }
963
964 /*
965 * s_umount nests inside bd_mutex during
966 * __invalidate_device(). blkdev_put() acquires
967 * bd_mutex and can't be called under s_umount. Drop
968 * s_umount temporarily. This is safe as we're
969 * holding an active reference.
970 */
971 up_write(&s->s_umount);
972 blkdev_put(bdev, mode);
973 down_write(&s->s_umount);
974 } else {
975 char b[BDEVNAME_SIZE];
976
977 s->s_flags = flags | MS_NOSEC;
978 s->s_mode = mode;
979 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
980 sb_set_blocksize(s, block_size(bdev));
981 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
982 if (error) {
983 deactivate_locked_super(s);
984 goto error;
985 }
986
987 s->s_flags |= MS_ACTIVE;
988 bdev->bd_super = s;
989 }
990
991 return dget(s->s_root);
992
993error_s:
994 error = PTR_ERR(s);
995error_bdev:
996 blkdev_put(bdev, mode);
997error:
998 return ERR_PTR(error);
999}
1000EXPORT_SYMBOL(mount_bdev);
1001
1002void kill_block_super(struct super_block *sb)
1003{
1004 struct block_device *bdev = sb->s_bdev;
1005 fmode_t mode = sb->s_mode;
1006
1007 bdev->bd_super = NULL;
1008 generic_shutdown_super(sb);
1009 sync_blockdev(bdev);
1010 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1011 blkdev_put(bdev, mode | FMODE_EXCL);
1012}
1013
1014EXPORT_SYMBOL(kill_block_super);
1015#endif
1016
1017struct dentry *mount_nodev(struct file_system_type *fs_type,
1018 int flags, void *data,
1019 int (*fill_super)(struct super_block *, void *, int))
1020{
1021 int error;
1022 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
1023
1024 if (IS_ERR(s))
1025 return ERR_CAST(s);
1026
1027 s->s_flags = flags;
1028
1029 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1030 if (error) {
1031 deactivate_locked_super(s);
1032 return ERR_PTR(error);
1033 }
1034 s->s_flags |= MS_ACTIVE;
1035 return dget(s->s_root);
1036}
1037EXPORT_SYMBOL(mount_nodev);
1038
1039static int compare_single(struct super_block *s, void *p)
1040{
1041 return 1;
1042}
1043
1044struct dentry *mount_single(struct file_system_type *fs_type,
1045 int flags, void *data,
1046 int (*fill_super)(struct super_block *, void *, int))
1047{
1048 struct super_block *s;
1049 int error;
1050
1051 s = sget(fs_type, compare_single, set_anon_super, NULL);
1052 if (IS_ERR(s))
1053 return ERR_CAST(s);
1054 if (!s->s_root) {
1055 s->s_flags = flags;
1056 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1057 if (error) {
1058 deactivate_locked_super(s);
1059 return ERR_PTR(error);
1060 }
1061 s->s_flags |= MS_ACTIVE;
1062 } else {
1063 do_remount_sb(s, flags, data, 0);
1064 }
1065 return dget(s->s_root);
1066}
1067EXPORT_SYMBOL(mount_single);
1068
1069struct dentry *
1070mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1071{
1072 struct dentry *root;
1073 struct super_block *sb;
1074 char *secdata = NULL;
1075 int error = -ENOMEM;
1076
1077 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1078 secdata = alloc_secdata();
1079 if (!secdata)
1080 goto out;
1081
1082 error = security_sb_copy_data(data, secdata);
1083 if (error)
1084 goto out_free_secdata;
1085 }
1086
1087 root = type->mount(type, flags, name, data);
1088 if (IS_ERR(root)) {
1089 error = PTR_ERR(root);
1090 goto out_free_secdata;
1091 }
1092 sb = root->d_sb;
1093 BUG_ON(!sb);
1094 WARN_ON(!sb->s_bdi);
1095 WARN_ON(sb->s_bdi == &default_backing_dev_info);
1096 sb->s_flags |= MS_BORN;
1097
1098 error = security_sb_kern_mount(sb, flags, secdata);
1099 if (error)
1100 goto out_sb;
1101
1102 /*
1103 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1104 * but s_maxbytes was an unsigned long long for many releases. Throw
1105 * this warning for a little while to try and catch filesystems that
1106 * violate this rule.
1107 */
1108 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1109 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1110
1111 up_write(&sb->s_umount);
1112 free_secdata(secdata);
1113 return root;
1114out_sb:
1115 dput(root);
1116 deactivate_locked_super(sb);
1117out_free_secdata:
1118 free_secdata(secdata);
1119out:
1120 return ERR_PTR(error);
1121}
1122
1123/**
1124 * freeze_super - lock the filesystem and force it into a consistent state
1125 * @sb: the super to lock
1126 *
1127 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1128 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1129 * -EBUSY.
1130 */
1131int freeze_super(struct super_block *sb)
1132{
1133 int ret;
1134
1135 atomic_inc(&sb->s_active);
1136 down_write(&sb->s_umount);
1137 if (sb->s_frozen) {
1138 deactivate_locked_super(sb);
1139 return -EBUSY;
1140 }
1141
1142 if (sb->s_flags & MS_RDONLY) {
1143 sb->s_frozen = SB_FREEZE_TRANS;
1144 smp_wmb();
1145 up_write(&sb->s_umount);
1146 return 0;
1147 }
1148
1149 sb->s_frozen = SB_FREEZE_WRITE;
1150 smp_wmb();
1151
1152 sync_filesystem(sb);
1153
1154 sb->s_frozen = SB_FREEZE_TRANS;
1155 smp_wmb();
1156
1157 sync_blockdev(sb->s_bdev);
1158 if (sb->s_op->freeze_fs) {
1159 ret = sb->s_op->freeze_fs(sb);
1160 if (ret) {
1161 printk(KERN_ERR
1162 "VFS:Filesystem freeze failed\n");
1163 sb->s_frozen = SB_UNFROZEN;
1164 deactivate_locked_super(sb);
1165 return ret;
1166 }
1167 }
1168 up_write(&sb->s_umount);
1169 return 0;
1170}
1171EXPORT_SYMBOL(freeze_super);
1172
1173/**
1174 * thaw_super -- unlock filesystem
1175 * @sb: the super to thaw
1176 *
1177 * Unlocks the filesystem and marks it writeable again after freeze_super().
1178 */
1179int thaw_super(struct super_block *sb)
1180{
1181 int error;
1182
1183 down_write(&sb->s_umount);
1184 if (sb->s_frozen == SB_UNFROZEN) {
1185 up_write(&sb->s_umount);
1186 return -EINVAL;
1187 }
1188
1189 if (sb->s_flags & MS_RDONLY)
1190 goto out;
1191
1192 if (sb->s_op->unfreeze_fs) {
1193 error = sb->s_op->unfreeze_fs(sb);
1194 if (error) {
1195 printk(KERN_ERR
1196 "VFS:Filesystem thaw failed\n");
1197 sb->s_frozen = SB_FREEZE_TRANS;
1198 up_write(&sb->s_umount);
1199 return error;
1200 }
1201 }
1202
1203out:
1204 sb->s_frozen = SB_UNFROZEN;
1205 smp_wmb();
1206 wake_up(&sb->s_wait_unfrozen);
1207 deactivate_locked_super(sb);
1208
1209 return 0;
1210}
1211EXPORT_SYMBOL(thaw_super);