]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/super.c
Merge branch 'for-linus' of git://ftp.arm.linux.org.uk/~rmk/linux-arm
[mirror_ubuntu-bionic-kernel.git] / fs / super.c
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/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 <linux/fsnotify.h>
36 #include <linux/lockdep.h>
37 #include "internal.h"
38
39
40 LIST_HEAD(super_blocks);
41 DEFINE_SPINLOCK(sb_lock);
42
43 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
44 "sb_writers",
45 "sb_pagefaults",
46 "sb_internal",
47 };
48
49 /*
50 * One thing we have to be careful of with a per-sb shrinker is that we don't
51 * drop the last active reference to the superblock from within the shrinker.
52 * If that happens we could trigger unregistering the shrinker from within the
53 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
54 * take a passive reference to the superblock to avoid this from occurring.
55 */
56 static unsigned long super_cache_scan(struct shrinker *shrink,
57 struct shrink_control *sc)
58 {
59 struct super_block *sb;
60 long fs_objects = 0;
61 long total_objects;
62 long freed = 0;
63 long dentries;
64 long inodes;
65
66 sb = container_of(shrink, struct super_block, s_shrink);
67
68 /*
69 * Deadlock avoidance. We may hold various FS locks, and we don't want
70 * to recurse into the FS that called us in clear_inode() and friends..
71 */
72 if (!(sc->gfp_mask & __GFP_FS))
73 return SHRINK_STOP;
74
75 if (!grab_super_passive(sb))
76 return SHRINK_STOP;
77
78 if (sb->s_op->nr_cached_objects)
79 fs_objects = sb->s_op->nr_cached_objects(sb, sc->nid);
80
81 inodes = list_lru_count_node(&sb->s_inode_lru, sc->nid);
82 dentries = list_lru_count_node(&sb->s_dentry_lru, sc->nid);
83 total_objects = dentries + inodes + fs_objects + 1;
84
85 /* proportion the scan between the caches */
86 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
87 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
88
89 /*
90 * prune the dcache first as the icache is pinned by it, then
91 * prune the icache, followed by the filesystem specific caches
92 */
93 freed = prune_dcache_sb(sb, dentries, sc->nid);
94 freed += prune_icache_sb(sb, inodes, sc->nid);
95
96 if (fs_objects) {
97 fs_objects = mult_frac(sc->nr_to_scan, fs_objects,
98 total_objects);
99 freed += sb->s_op->free_cached_objects(sb, fs_objects,
100 sc->nid);
101 }
102
103 drop_super(sb);
104 return freed;
105 }
106
107 static unsigned long super_cache_count(struct shrinker *shrink,
108 struct shrink_control *sc)
109 {
110 struct super_block *sb;
111 long total_objects = 0;
112
113 sb = container_of(shrink, struct super_block, s_shrink);
114
115 if (!grab_super_passive(sb))
116 return 0;
117
118 if (sb->s_op && sb->s_op->nr_cached_objects)
119 total_objects = sb->s_op->nr_cached_objects(sb,
120 sc->nid);
121
122 total_objects += list_lru_count_node(&sb->s_dentry_lru,
123 sc->nid);
124 total_objects += list_lru_count_node(&sb->s_inode_lru,
125 sc->nid);
126
127 total_objects = vfs_pressure_ratio(total_objects);
128 drop_super(sb);
129 return total_objects;
130 }
131
132 /**
133 * destroy_super - frees a superblock
134 * @s: superblock to free
135 *
136 * Frees a superblock.
137 */
138 static void destroy_super(struct super_block *s)
139 {
140 int i;
141 list_lru_destroy(&s->s_dentry_lru);
142 list_lru_destroy(&s->s_inode_lru);
143 for (i = 0; i < SB_FREEZE_LEVELS; i++)
144 percpu_counter_destroy(&s->s_writers.counter[i]);
145 security_sb_free(s);
146 WARN_ON(!list_empty(&s->s_mounts));
147 kfree(s->s_subtype);
148 kfree(s->s_options);
149 kfree_rcu(s, rcu);
150 }
151
152 /**
153 * alloc_super - create new superblock
154 * @type: filesystem type superblock should belong to
155 * @flags: the mount flags
156 *
157 * Allocates and initializes a new &struct super_block. alloc_super()
158 * returns a pointer new superblock or %NULL if allocation had failed.
159 */
160 static struct super_block *alloc_super(struct file_system_type *type, int flags)
161 {
162 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
163 static const struct super_operations default_op;
164 int i;
165
166 if (!s)
167 return NULL;
168
169 INIT_LIST_HEAD(&s->s_mounts);
170
171 if (security_sb_alloc(s))
172 goto fail;
173
174 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
175 if (percpu_counter_init(&s->s_writers.counter[i], 0) < 0)
176 goto fail;
177 lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i],
178 &type->s_writers_key[i], 0);
179 }
180 init_waitqueue_head(&s->s_writers.wait);
181 init_waitqueue_head(&s->s_writers.wait_unfrozen);
182 s->s_flags = flags;
183 s->s_bdi = &default_backing_dev_info;
184 INIT_HLIST_NODE(&s->s_instances);
185 INIT_HLIST_BL_HEAD(&s->s_anon);
186 INIT_LIST_HEAD(&s->s_inodes);
187
188 if (list_lru_init(&s->s_dentry_lru))
189 goto fail;
190 if (list_lru_init(&s->s_inode_lru))
191 goto fail;
192
193 init_rwsem(&s->s_umount);
194 lockdep_set_class(&s->s_umount, &type->s_umount_key);
195 /*
196 * sget() can have s_umount recursion.
197 *
198 * When it cannot find a suitable sb, it allocates a new
199 * one (this one), and tries again to find a suitable old
200 * one.
201 *
202 * In case that succeeds, it will acquire the s_umount
203 * lock of the old one. Since these are clearly distrinct
204 * locks, and this object isn't exposed yet, there's no
205 * risk of deadlocks.
206 *
207 * Annotate this by putting this lock in a different
208 * subclass.
209 */
210 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
211 s->s_count = 1;
212 atomic_set(&s->s_active, 1);
213 mutex_init(&s->s_vfs_rename_mutex);
214 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
215 mutex_init(&s->s_dquot.dqio_mutex);
216 mutex_init(&s->s_dquot.dqonoff_mutex);
217 init_rwsem(&s->s_dquot.dqptr_sem);
218 s->s_maxbytes = MAX_NON_LFS;
219 s->s_op = &default_op;
220 s->s_time_gran = 1000000000;
221 s->cleancache_poolid = -1;
222
223 s->s_shrink.seeks = DEFAULT_SEEKS;
224 s->s_shrink.scan_objects = super_cache_scan;
225 s->s_shrink.count_objects = super_cache_count;
226 s->s_shrink.batch = 1024;
227 s->s_shrink.flags = SHRINKER_NUMA_AWARE;
228 return s;
229
230 fail:
231 destroy_super(s);
232 return NULL;
233 }
234
235 /* Superblock refcounting */
236
237 /*
238 * Drop a superblock's refcount. The caller must hold sb_lock.
239 */
240 static void __put_super(struct super_block *sb)
241 {
242 if (!--sb->s_count) {
243 list_del_init(&sb->s_list);
244 destroy_super(sb);
245 }
246 }
247
248 /**
249 * put_super - drop a temporary reference to superblock
250 * @sb: superblock in question
251 *
252 * Drops a temporary reference, frees superblock if there's no
253 * references left.
254 */
255 static void put_super(struct super_block *sb)
256 {
257 spin_lock(&sb_lock);
258 __put_super(sb);
259 spin_unlock(&sb_lock);
260 }
261
262
263 /**
264 * deactivate_locked_super - drop an active reference to superblock
265 * @s: superblock to deactivate
266 *
267 * Drops an active reference to superblock, converting it into a temprory
268 * one if there is no other active references left. In that case we
269 * tell fs driver to shut it down and drop the temporary reference we
270 * had just acquired.
271 *
272 * Caller holds exclusive lock on superblock; that lock is released.
273 */
274 void deactivate_locked_super(struct super_block *s)
275 {
276 struct file_system_type *fs = s->s_type;
277 if (atomic_dec_and_test(&s->s_active)) {
278 cleancache_invalidate_fs(s);
279 fs->kill_sb(s);
280
281 /* caches are now gone, we can safely kill the shrinker now */
282 unregister_shrinker(&s->s_shrink);
283
284 put_filesystem(fs);
285 put_super(s);
286 } else {
287 up_write(&s->s_umount);
288 }
289 }
290
291 EXPORT_SYMBOL(deactivate_locked_super);
292
293 /**
294 * deactivate_super - drop an active reference to superblock
295 * @s: superblock to deactivate
296 *
297 * Variant of deactivate_locked_super(), except that superblock is *not*
298 * locked by caller. If we are going to drop the final active reference,
299 * lock will be acquired prior to that.
300 */
301 void deactivate_super(struct super_block *s)
302 {
303 if (!atomic_add_unless(&s->s_active, -1, 1)) {
304 down_write(&s->s_umount);
305 deactivate_locked_super(s);
306 }
307 }
308
309 EXPORT_SYMBOL(deactivate_super);
310
311 /**
312 * grab_super - acquire an active reference
313 * @s: reference we are trying to make active
314 *
315 * Tries to acquire an active reference. grab_super() is used when we
316 * had just found a superblock in super_blocks or fs_type->fs_supers
317 * and want to turn it into a full-blown active reference. grab_super()
318 * is called with sb_lock held and drops it. Returns 1 in case of
319 * success, 0 if we had failed (superblock contents was already dead or
320 * dying when grab_super() had been called). Note that this is only
321 * called for superblocks not in rundown mode (== ones still on ->fs_supers
322 * of their type), so increment of ->s_count is OK here.
323 */
324 static int grab_super(struct super_block *s) __releases(sb_lock)
325 {
326 s->s_count++;
327 spin_unlock(&sb_lock);
328 down_write(&s->s_umount);
329 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
330 put_super(s);
331 return 1;
332 }
333 up_write(&s->s_umount);
334 put_super(s);
335 return 0;
336 }
337
338 /*
339 * grab_super_passive - acquire a passive reference
340 * @sb: reference we are trying to grab
341 *
342 * Tries to acquire a passive reference. This is used in places where we
343 * cannot take an active reference but we need to ensure that the
344 * superblock does not go away while we are working on it. It returns
345 * false if a reference was not gained, and returns true with the s_umount
346 * lock held in read mode if a reference is gained. On successful return,
347 * the caller must drop the s_umount lock and the passive reference when
348 * done.
349 */
350 bool grab_super_passive(struct super_block *sb)
351 {
352 spin_lock(&sb_lock);
353 if (hlist_unhashed(&sb->s_instances)) {
354 spin_unlock(&sb_lock);
355 return false;
356 }
357
358 sb->s_count++;
359 spin_unlock(&sb_lock);
360
361 if (down_read_trylock(&sb->s_umount)) {
362 if (sb->s_root && (sb->s_flags & MS_BORN))
363 return true;
364 up_read(&sb->s_umount);
365 }
366
367 put_super(sb);
368 return false;
369 }
370
371 /**
372 * generic_shutdown_super - common helper for ->kill_sb()
373 * @sb: superblock to kill
374 *
375 * generic_shutdown_super() does all fs-independent work on superblock
376 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
377 * that need destruction out of superblock, call generic_shutdown_super()
378 * and release aforementioned objects. Note: dentries and inodes _are_
379 * taken care of and do not need specific handling.
380 *
381 * Upon calling this function, the filesystem may no longer alter or
382 * rearrange the set of dentries belonging to this super_block, nor may it
383 * change the attachments of dentries to inodes.
384 */
385 void generic_shutdown_super(struct super_block *sb)
386 {
387 const struct super_operations *sop = sb->s_op;
388
389 if (sb->s_root) {
390 shrink_dcache_for_umount(sb);
391 sync_filesystem(sb);
392 sb->s_flags &= ~MS_ACTIVE;
393
394 fsnotify_unmount_inodes(&sb->s_inodes);
395
396 evict_inodes(sb);
397
398 if (sb->s_dio_done_wq) {
399 destroy_workqueue(sb->s_dio_done_wq);
400 sb->s_dio_done_wq = NULL;
401 }
402
403 if (sop->put_super)
404 sop->put_super(sb);
405
406 if (!list_empty(&sb->s_inodes)) {
407 printk("VFS: Busy inodes after unmount of %s. "
408 "Self-destruct in 5 seconds. Have a nice day...\n",
409 sb->s_id);
410 }
411 }
412 spin_lock(&sb_lock);
413 /* should be initialized for __put_super_and_need_restart() */
414 hlist_del_init(&sb->s_instances);
415 spin_unlock(&sb_lock);
416 up_write(&sb->s_umount);
417 }
418
419 EXPORT_SYMBOL(generic_shutdown_super);
420
421 /**
422 * sget - find or create a superblock
423 * @type: filesystem type superblock should belong to
424 * @test: comparison callback
425 * @set: setup callback
426 * @flags: mount flags
427 * @data: argument to each of them
428 */
429 struct super_block *sget(struct file_system_type *type,
430 int (*test)(struct super_block *,void *),
431 int (*set)(struct super_block *,void *),
432 int flags,
433 void *data)
434 {
435 struct super_block *s = NULL;
436 struct super_block *old;
437 int err;
438
439 retry:
440 spin_lock(&sb_lock);
441 if (test) {
442 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
443 if (!test(old, data))
444 continue;
445 if (!grab_super(old))
446 goto retry;
447 if (s) {
448 up_write(&s->s_umount);
449 destroy_super(s);
450 s = NULL;
451 }
452 return old;
453 }
454 }
455 if (!s) {
456 spin_unlock(&sb_lock);
457 s = alloc_super(type, flags);
458 if (!s)
459 return ERR_PTR(-ENOMEM);
460 goto retry;
461 }
462
463 err = set(s, data);
464 if (err) {
465 spin_unlock(&sb_lock);
466 up_write(&s->s_umount);
467 destroy_super(s);
468 return ERR_PTR(err);
469 }
470 s->s_type = type;
471 strlcpy(s->s_id, type->name, sizeof(s->s_id));
472 list_add_tail(&s->s_list, &super_blocks);
473 hlist_add_head(&s->s_instances, &type->fs_supers);
474 spin_unlock(&sb_lock);
475 get_filesystem(type);
476 register_shrinker(&s->s_shrink);
477 return s;
478 }
479
480 EXPORT_SYMBOL(sget);
481
482 void drop_super(struct super_block *sb)
483 {
484 up_read(&sb->s_umount);
485 put_super(sb);
486 }
487
488 EXPORT_SYMBOL(drop_super);
489
490 /**
491 * iterate_supers - call function for all active superblocks
492 * @f: function to call
493 * @arg: argument to pass to it
494 *
495 * Scans the superblock list and calls given function, passing it
496 * locked superblock and given argument.
497 */
498 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
499 {
500 struct super_block *sb, *p = NULL;
501
502 spin_lock(&sb_lock);
503 list_for_each_entry(sb, &super_blocks, s_list) {
504 if (hlist_unhashed(&sb->s_instances))
505 continue;
506 sb->s_count++;
507 spin_unlock(&sb_lock);
508
509 down_read(&sb->s_umount);
510 if (sb->s_root && (sb->s_flags & MS_BORN))
511 f(sb, arg);
512 up_read(&sb->s_umount);
513
514 spin_lock(&sb_lock);
515 if (p)
516 __put_super(p);
517 p = sb;
518 }
519 if (p)
520 __put_super(p);
521 spin_unlock(&sb_lock);
522 }
523
524 /**
525 * iterate_supers_type - call function for superblocks of given type
526 * @type: fs type
527 * @f: function to call
528 * @arg: argument to pass to it
529 *
530 * Scans the superblock list and calls given function, passing it
531 * locked superblock and given argument.
532 */
533 void iterate_supers_type(struct file_system_type *type,
534 void (*f)(struct super_block *, void *), void *arg)
535 {
536 struct super_block *sb, *p = NULL;
537
538 spin_lock(&sb_lock);
539 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
540 sb->s_count++;
541 spin_unlock(&sb_lock);
542
543 down_read(&sb->s_umount);
544 if (sb->s_root && (sb->s_flags & MS_BORN))
545 f(sb, arg);
546 up_read(&sb->s_umount);
547
548 spin_lock(&sb_lock);
549 if (p)
550 __put_super(p);
551 p = sb;
552 }
553 if (p)
554 __put_super(p);
555 spin_unlock(&sb_lock);
556 }
557
558 EXPORT_SYMBOL(iterate_supers_type);
559
560 /**
561 * get_super - get the superblock of a device
562 * @bdev: device to get the superblock for
563 *
564 * Scans the superblock list and finds the superblock of the file system
565 * mounted on the device given. %NULL is returned if no match is found.
566 */
567
568 struct super_block *get_super(struct block_device *bdev)
569 {
570 struct super_block *sb;
571
572 if (!bdev)
573 return NULL;
574
575 spin_lock(&sb_lock);
576 rescan:
577 list_for_each_entry(sb, &super_blocks, s_list) {
578 if (hlist_unhashed(&sb->s_instances))
579 continue;
580 if (sb->s_bdev == bdev) {
581 sb->s_count++;
582 spin_unlock(&sb_lock);
583 down_read(&sb->s_umount);
584 /* still alive? */
585 if (sb->s_root && (sb->s_flags & MS_BORN))
586 return sb;
587 up_read(&sb->s_umount);
588 /* nope, got unmounted */
589 spin_lock(&sb_lock);
590 __put_super(sb);
591 goto rescan;
592 }
593 }
594 spin_unlock(&sb_lock);
595 return NULL;
596 }
597
598 EXPORT_SYMBOL(get_super);
599
600 /**
601 * get_super_thawed - get thawed superblock of a device
602 * @bdev: device to get the superblock for
603 *
604 * Scans the superblock list and finds the superblock of the file system
605 * mounted on the device. The superblock is returned once it is thawed
606 * (or immediately if it was not frozen). %NULL is returned if no match
607 * is found.
608 */
609 struct super_block *get_super_thawed(struct block_device *bdev)
610 {
611 while (1) {
612 struct super_block *s = get_super(bdev);
613 if (!s || s->s_writers.frozen == SB_UNFROZEN)
614 return s;
615 up_read(&s->s_umount);
616 wait_event(s->s_writers.wait_unfrozen,
617 s->s_writers.frozen == SB_UNFROZEN);
618 put_super(s);
619 }
620 }
621 EXPORT_SYMBOL(get_super_thawed);
622
623 /**
624 * get_active_super - get an active reference to the superblock of a device
625 * @bdev: device to get the superblock for
626 *
627 * Scans the superblock list and finds the superblock of the file system
628 * mounted on the device given. Returns the superblock with an active
629 * reference or %NULL if none was found.
630 */
631 struct super_block *get_active_super(struct block_device *bdev)
632 {
633 struct super_block *sb;
634
635 if (!bdev)
636 return NULL;
637
638 restart:
639 spin_lock(&sb_lock);
640 list_for_each_entry(sb, &super_blocks, s_list) {
641 if (hlist_unhashed(&sb->s_instances))
642 continue;
643 if (sb->s_bdev == bdev) {
644 if (!grab_super(sb))
645 goto restart;
646 up_write(&sb->s_umount);
647 return sb;
648 }
649 }
650 spin_unlock(&sb_lock);
651 return NULL;
652 }
653
654 struct super_block *user_get_super(dev_t dev)
655 {
656 struct super_block *sb;
657
658 spin_lock(&sb_lock);
659 rescan:
660 list_for_each_entry(sb, &super_blocks, s_list) {
661 if (hlist_unhashed(&sb->s_instances))
662 continue;
663 if (sb->s_dev == dev) {
664 sb->s_count++;
665 spin_unlock(&sb_lock);
666 down_read(&sb->s_umount);
667 /* still alive? */
668 if (sb->s_root && (sb->s_flags & MS_BORN))
669 return sb;
670 up_read(&sb->s_umount);
671 /* nope, got unmounted */
672 spin_lock(&sb_lock);
673 __put_super(sb);
674 goto rescan;
675 }
676 }
677 spin_unlock(&sb_lock);
678 return NULL;
679 }
680
681 /**
682 * do_remount_sb - asks filesystem to change mount options.
683 * @sb: superblock in question
684 * @flags: numeric part of options
685 * @data: the rest of options
686 * @force: whether or not to force the change
687 *
688 * Alters the mount options of a mounted file system.
689 */
690 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
691 {
692 int retval;
693 int remount_ro;
694
695 if (sb->s_writers.frozen != SB_UNFROZEN)
696 return -EBUSY;
697
698 #ifdef CONFIG_BLOCK
699 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
700 return -EACCES;
701 #endif
702
703 if (flags & MS_RDONLY)
704 acct_auto_close(sb);
705 shrink_dcache_sb(sb);
706
707 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
708
709 /* If we are remounting RDONLY and current sb is read/write,
710 make sure there are no rw files opened */
711 if (remount_ro) {
712 if (force) {
713 sb->s_readonly_remount = 1;
714 smp_wmb();
715 } else {
716 retval = sb_prepare_remount_readonly(sb);
717 if (retval)
718 return retval;
719 }
720 }
721
722 if (sb->s_op->remount_fs) {
723 retval = sb->s_op->remount_fs(sb, &flags, data);
724 if (retval) {
725 if (!force)
726 goto cancel_readonly;
727 /* If forced remount, go ahead despite any errors */
728 WARN(1, "forced remount of a %s fs returned %i\n",
729 sb->s_type->name, retval);
730 }
731 }
732 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
733 /* Needs to be ordered wrt mnt_is_readonly() */
734 smp_wmb();
735 sb->s_readonly_remount = 0;
736
737 /*
738 * Some filesystems modify their metadata via some other path than the
739 * bdev buffer cache (eg. use a private mapping, or directories in
740 * pagecache, etc). Also file data modifications go via their own
741 * mappings. So If we try to mount readonly then copy the filesystem
742 * from bdev, we could get stale data, so invalidate it to give a best
743 * effort at coherency.
744 */
745 if (remount_ro && sb->s_bdev)
746 invalidate_bdev(sb->s_bdev);
747 return 0;
748
749 cancel_readonly:
750 sb->s_readonly_remount = 0;
751 return retval;
752 }
753
754 static void do_emergency_remount(struct work_struct *work)
755 {
756 struct super_block *sb, *p = NULL;
757
758 spin_lock(&sb_lock);
759 list_for_each_entry(sb, &super_blocks, s_list) {
760 if (hlist_unhashed(&sb->s_instances))
761 continue;
762 sb->s_count++;
763 spin_unlock(&sb_lock);
764 down_write(&sb->s_umount);
765 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
766 !(sb->s_flags & MS_RDONLY)) {
767 /*
768 * What lock protects sb->s_flags??
769 */
770 do_remount_sb(sb, MS_RDONLY, NULL, 1);
771 }
772 up_write(&sb->s_umount);
773 spin_lock(&sb_lock);
774 if (p)
775 __put_super(p);
776 p = sb;
777 }
778 if (p)
779 __put_super(p);
780 spin_unlock(&sb_lock);
781 kfree(work);
782 printk("Emergency Remount complete\n");
783 }
784
785 void emergency_remount(void)
786 {
787 struct work_struct *work;
788
789 work = kmalloc(sizeof(*work), GFP_ATOMIC);
790 if (work) {
791 INIT_WORK(work, do_emergency_remount);
792 schedule_work(work);
793 }
794 }
795
796 /*
797 * Unnamed block devices are dummy devices used by virtual
798 * filesystems which don't use real block-devices. -- jrs
799 */
800
801 static DEFINE_IDA(unnamed_dev_ida);
802 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
803 static int unnamed_dev_start = 0; /* don't bother trying below it */
804
805 int get_anon_bdev(dev_t *p)
806 {
807 int dev;
808 int error;
809
810 retry:
811 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
812 return -ENOMEM;
813 spin_lock(&unnamed_dev_lock);
814 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
815 if (!error)
816 unnamed_dev_start = dev + 1;
817 spin_unlock(&unnamed_dev_lock);
818 if (error == -EAGAIN)
819 /* We raced and lost with another CPU. */
820 goto retry;
821 else if (error)
822 return -EAGAIN;
823
824 if (dev == (1 << MINORBITS)) {
825 spin_lock(&unnamed_dev_lock);
826 ida_remove(&unnamed_dev_ida, dev);
827 if (unnamed_dev_start > dev)
828 unnamed_dev_start = dev;
829 spin_unlock(&unnamed_dev_lock);
830 return -EMFILE;
831 }
832 *p = MKDEV(0, dev & MINORMASK);
833 return 0;
834 }
835 EXPORT_SYMBOL(get_anon_bdev);
836
837 void free_anon_bdev(dev_t dev)
838 {
839 int slot = MINOR(dev);
840 spin_lock(&unnamed_dev_lock);
841 ida_remove(&unnamed_dev_ida, slot);
842 if (slot < unnamed_dev_start)
843 unnamed_dev_start = slot;
844 spin_unlock(&unnamed_dev_lock);
845 }
846 EXPORT_SYMBOL(free_anon_bdev);
847
848 int set_anon_super(struct super_block *s, void *data)
849 {
850 int error = get_anon_bdev(&s->s_dev);
851 if (!error)
852 s->s_bdi = &noop_backing_dev_info;
853 return error;
854 }
855
856 EXPORT_SYMBOL(set_anon_super);
857
858 void kill_anon_super(struct super_block *sb)
859 {
860 dev_t dev = sb->s_dev;
861 generic_shutdown_super(sb);
862 free_anon_bdev(dev);
863 }
864
865 EXPORT_SYMBOL(kill_anon_super);
866
867 void kill_litter_super(struct super_block *sb)
868 {
869 if (sb->s_root)
870 d_genocide(sb->s_root);
871 kill_anon_super(sb);
872 }
873
874 EXPORT_SYMBOL(kill_litter_super);
875
876 static int ns_test_super(struct super_block *sb, void *data)
877 {
878 return sb->s_fs_info == data;
879 }
880
881 static int ns_set_super(struct super_block *sb, void *data)
882 {
883 sb->s_fs_info = data;
884 return set_anon_super(sb, NULL);
885 }
886
887 struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
888 void *data, int (*fill_super)(struct super_block *, void *, int))
889 {
890 struct super_block *sb;
891
892 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
893 if (IS_ERR(sb))
894 return ERR_CAST(sb);
895
896 if (!sb->s_root) {
897 int err;
898 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
899 if (err) {
900 deactivate_locked_super(sb);
901 return ERR_PTR(err);
902 }
903
904 sb->s_flags |= MS_ACTIVE;
905 }
906
907 return dget(sb->s_root);
908 }
909
910 EXPORT_SYMBOL(mount_ns);
911
912 #ifdef CONFIG_BLOCK
913 static int set_bdev_super(struct super_block *s, void *data)
914 {
915 s->s_bdev = data;
916 s->s_dev = s->s_bdev->bd_dev;
917
918 /*
919 * We set the bdi here to the queue backing, file systems can
920 * overwrite this in ->fill_super()
921 */
922 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
923 return 0;
924 }
925
926 static int test_bdev_super(struct super_block *s, void *data)
927 {
928 return (void *)s->s_bdev == data;
929 }
930
931 struct dentry *mount_bdev(struct file_system_type *fs_type,
932 int flags, const char *dev_name, void *data,
933 int (*fill_super)(struct super_block *, void *, int))
934 {
935 struct block_device *bdev;
936 struct super_block *s;
937 fmode_t mode = FMODE_READ | FMODE_EXCL;
938 int error = 0;
939
940 if (!(flags & MS_RDONLY))
941 mode |= FMODE_WRITE;
942
943 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
944 if (IS_ERR(bdev))
945 return ERR_CAST(bdev);
946
947 /*
948 * once the super is inserted into the list by sget, s_umount
949 * will protect the lockfs code from trying to start a snapshot
950 * while we are mounting
951 */
952 mutex_lock(&bdev->bd_fsfreeze_mutex);
953 if (bdev->bd_fsfreeze_count > 0) {
954 mutex_unlock(&bdev->bd_fsfreeze_mutex);
955 error = -EBUSY;
956 goto error_bdev;
957 }
958 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
959 bdev);
960 mutex_unlock(&bdev->bd_fsfreeze_mutex);
961 if (IS_ERR(s))
962 goto error_s;
963
964 if (s->s_root) {
965 if ((flags ^ s->s_flags) & MS_RDONLY) {
966 deactivate_locked_super(s);
967 error = -EBUSY;
968 goto error_bdev;
969 }
970
971 /*
972 * s_umount nests inside bd_mutex during
973 * __invalidate_device(). blkdev_put() acquires
974 * bd_mutex and can't be called under s_umount. Drop
975 * s_umount temporarily. This is safe as we're
976 * holding an active reference.
977 */
978 up_write(&s->s_umount);
979 blkdev_put(bdev, mode);
980 down_write(&s->s_umount);
981 } else {
982 char b[BDEVNAME_SIZE];
983
984 s->s_mode = mode;
985 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
986 sb_set_blocksize(s, block_size(bdev));
987 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
988 if (error) {
989 deactivate_locked_super(s);
990 goto error;
991 }
992
993 s->s_flags |= MS_ACTIVE;
994 bdev->bd_super = s;
995 }
996
997 return dget(s->s_root);
998
999 error_s:
1000 error = PTR_ERR(s);
1001 error_bdev:
1002 blkdev_put(bdev, mode);
1003 error:
1004 return ERR_PTR(error);
1005 }
1006 EXPORT_SYMBOL(mount_bdev);
1007
1008 void kill_block_super(struct super_block *sb)
1009 {
1010 struct block_device *bdev = sb->s_bdev;
1011 fmode_t mode = sb->s_mode;
1012
1013 bdev->bd_super = NULL;
1014 generic_shutdown_super(sb);
1015 sync_blockdev(bdev);
1016 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1017 blkdev_put(bdev, mode | FMODE_EXCL);
1018 }
1019
1020 EXPORT_SYMBOL(kill_block_super);
1021 #endif
1022
1023 struct dentry *mount_nodev(struct file_system_type *fs_type,
1024 int flags, void *data,
1025 int (*fill_super)(struct super_block *, void *, int))
1026 {
1027 int error;
1028 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1029
1030 if (IS_ERR(s))
1031 return ERR_CAST(s);
1032
1033 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1034 if (error) {
1035 deactivate_locked_super(s);
1036 return ERR_PTR(error);
1037 }
1038 s->s_flags |= MS_ACTIVE;
1039 return dget(s->s_root);
1040 }
1041 EXPORT_SYMBOL(mount_nodev);
1042
1043 static int compare_single(struct super_block *s, void *p)
1044 {
1045 return 1;
1046 }
1047
1048 struct dentry *mount_single(struct file_system_type *fs_type,
1049 int flags, void *data,
1050 int (*fill_super)(struct super_block *, void *, int))
1051 {
1052 struct super_block *s;
1053 int error;
1054
1055 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1056 if (IS_ERR(s))
1057 return ERR_CAST(s);
1058 if (!s->s_root) {
1059 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1060 if (error) {
1061 deactivate_locked_super(s);
1062 return ERR_PTR(error);
1063 }
1064 s->s_flags |= MS_ACTIVE;
1065 } else {
1066 do_remount_sb(s, flags, data, 0);
1067 }
1068 return dget(s->s_root);
1069 }
1070 EXPORT_SYMBOL(mount_single);
1071
1072 struct dentry *
1073 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1074 {
1075 struct dentry *root;
1076 struct super_block *sb;
1077 char *secdata = NULL;
1078 int error = -ENOMEM;
1079
1080 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1081 secdata = alloc_secdata();
1082 if (!secdata)
1083 goto out;
1084
1085 error = security_sb_copy_data(data, secdata);
1086 if (error)
1087 goto out_free_secdata;
1088 }
1089
1090 root = type->mount(type, flags, name, data);
1091 if (IS_ERR(root)) {
1092 error = PTR_ERR(root);
1093 goto out_free_secdata;
1094 }
1095 sb = root->d_sb;
1096 BUG_ON(!sb);
1097 WARN_ON(!sb->s_bdi);
1098 WARN_ON(sb->s_bdi == &default_backing_dev_info);
1099 sb->s_flags |= MS_BORN;
1100
1101 error = security_sb_kern_mount(sb, flags, secdata);
1102 if (error)
1103 goto out_sb;
1104
1105 /*
1106 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1107 * but s_maxbytes was an unsigned long long for many releases. Throw
1108 * this warning for a little while to try and catch filesystems that
1109 * violate this rule.
1110 */
1111 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1112 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1113
1114 up_write(&sb->s_umount);
1115 free_secdata(secdata);
1116 return root;
1117 out_sb:
1118 dput(root);
1119 deactivate_locked_super(sb);
1120 out_free_secdata:
1121 free_secdata(secdata);
1122 out:
1123 return ERR_PTR(error);
1124 }
1125
1126 /*
1127 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1128 * instead.
1129 */
1130 void __sb_end_write(struct super_block *sb, int level)
1131 {
1132 percpu_counter_dec(&sb->s_writers.counter[level-1]);
1133 /*
1134 * Make sure s_writers are updated before we wake up waiters in
1135 * freeze_super().
1136 */
1137 smp_mb();
1138 if (waitqueue_active(&sb->s_writers.wait))
1139 wake_up(&sb->s_writers.wait);
1140 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_);
1141 }
1142 EXPORT_SYMBOL(__sb_end_write);
1143
1144 #ifdef CONFIG_LOCKDEP
1145 /*
1146 * We want lockdep to tell us about possible deadlocks with freezing but
1147 * it's it bit tricky to properly instrument it. Getting a freeze protection
1148 * works as getting a read lock but there are subtle problems. XFS for example
1149 * gets freeze protection on internal level twice in some cases, which is OK
1150 * only because we already hold a freeze protection also on higher level. Due
1151 * to these cases we have to tell lockdep we are doing trylock when we
1152 * already hold a freeze protection for a higher freeze level.
1153 */
1154 static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock,
1155 unsigned long ip)
1156 {
1157 int i;
1158
1159 if (!trylock) {
1160 for (i = 0; i < level - 1; i++)
1161 if (lock_is_held(&sb->s_writers.lock_map[i])) {
1162 trylock = true;
1163 break;
1164 }
1165 }
1166 rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip);
1167 }
1168 #endif
1169
1170 /*
1171 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1172 * instead.
1173 */
1174 int __sb_start_write(struct super_block *sb, int level, bool wait)
1175 {
1176 retry:
1177 if (unlikely(sb->s_writers.frozen >= level)) {
1178 if (!wait)
1179 return 0;
1180 wait_event(sb->s_writers.wait_unfrozen,
1181 sb->s_writers.frozen < level);
1182 }
1183
1184 #ifdef CONFIG_LOCKDEP
1185 acquire_freeze_lock(sb, level, !wait, _RET_IP_);
1186 #endif
1187 percpu_counter_inc(&sb->s_writers.counter[level-1]);
1188 /*
1189 * Make sure counter is updated before we check for frozen.
1190 * freeze_super() first sets frozen and then checks the counter.
1191 */
1192 smp_mb();
1193 if (unlikely(sb->s_writers.frozen >= level)) {
1194 __sb_end_write(sb, level);
1195 goto retry;
1196 }
1197 return 1;
1198 }
1199 EXPORT_SYMBOL(__sb_start_write);
1200
1201 /**
1202 * sb_wait_write - wait until all writers to given file system finish
1203 * @sb: the super for which we wait
1204 * @level: type of writers we wait for (normal vs page fault)
1205 *
1206 * This function waits until there are no writers of given type to given file
1207 * system. Caller of this function should make sure there can be no new writers
1208 * of type @level before calling this function. Otherwise this function can
1209 * livelock.
1210 */
1211 static void sb_wait_write(struct super_block *sb, int level)
1212 {
1213 s64 writers;
1214
1215 /*
1216 * We just cycle-through lockdep here so that it does not complain
1217 * about returning with lock to userspace
1218 */
1219 rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_);
1220 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_);
1221
1222 do {
1223 DEFINE_WAIT(wait);
1224
1225 /*
1226 * We use a barrier in prepare_to_wait() to separate setting
1227 * of frozen and checking of the counter
1228 */
1229 prepare_to_wait(&sb->s_writers.wait, &wait,
1230 TASK_UNINTERRUPTIBLE);
1231
1232 writers = percpu_counter_sum(&sb->s_writers.counter[level-1]);
1233 if (writers)
1234 schedule();
1235
1236 finish_wait(&sb->s_writers.wait, &wait);
1237 } while (writers);
1238 }
1239
1240 /**
1241 * freeze_super - lock the filesystem and force it into a consistent state
1242 * @sb: the super to lock
1243 *
1244 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1245 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1246 * -EBUSY.
1247 *
1248 * During this function, sb->s_writers.frozen goes through these values:
1249 *
1250 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1251 *
1252 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1253 * writes should be blocked, though page faults are still allowed. We wait for
1254 * all writes to complete and then proceed to the next stage.
1255 *
1256 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1257 * but internal fs threads can still modify the filesystem (although they
1258 * should not dirty new pages or inodes), writeback can run etc. After waiting
1259 * for all running page faults we sync the filesystem which will clean all
1260 * dirty pages and inodes (no new dirty pages or inodes can be created when
1261 * sync is running).
1262 *
1263 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1264 * modification are blocked (e.g. XFS preallocation truncation on inode
1265 * reclaim). This is usually implemented by blocking new transactions for
1266 * filesystems that have them and need this additional guard. After all
1267 * internal writers are finished we call ->freeze_fs() to finish filesystem
1268 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1269 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1270 *
1271 * sb->s_writers.frozen is protected by sb->s_umount.
1272 */
1273 int freeze_super(struct super_block *sb)
1274 {
1275 int ret;
1276
1277 atomic_inc(&sb->s_active);
1278 down_write(&sb->s_umount);
1279 if (sb->s_writers.frozen != SB_UNFROZEN) {
1280 deactivate_locked_super(sb);
1281 return -EBUSY;
1282 }
1283
1284 if (!(sb->s_flags & MS_BORN)) {
1285 up_write(&sb->s_umount);
1286 return 0; /* sic - it's "nothing to do" */
1287 }
1288
1289 if (sb->s_flags & MS_RDONLY) {
1290 /* Nothing to do really... */
1291 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1292 up_write(&sb->s_umount);
1293 return 0;
1294 }
1295
1296 /* From now on, no new normal writers can start */
1297 sb->s_writers.frozen = SB_FREEZE_WRITE;
1298 smp_wmb();
1299
1300 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1301 up_write(&sb->s_umount);
1302
1303 sb_wait_write(sb, SB_FREEZE_WRITE);
1304
1305 /* Now we go and block page faults... */
1306 down_write(&sb->s_umount);
1307 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1308 smp_wmb();
1309
1310 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1311
1312 /* All writers are done so after syncing there won't be dirty data */
1313 sync_filesystem(sb);
1314
1315 /* Now wait for internal filesystem counter */
1316 sb->s_writers.frozen = SB_FREEZE_FS;
1317 smp_wmb();
1318 sb_wait_write(sb, SB_FREEZE_FS);
1319
1320 if (sb->s_op->freeze_fs) {
1321 ret = sb->s_op->freeze_fs(sb);
1322 if (ret) {
1323 printk(KERN_ERR
1324 "VFS:Filesystem freeze failed\n");
1325 sb->s_writers.frozen = SB_UNFROZEN;
1326 smp_wmb();
1327 wake_up(&sb->s_writers.wait_unfrozen);
1328 deactivate_locked_super(sb);
1329 return ret;
1330 }
1331 }
1332 /*
1333 * This is just for debugging purposes so that fs can warn if it
1334 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1335 */
1336 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1337 up_write(&sb->s_umount);
1338 return 0;
1339 }
1340 EXPORT_SYMBOL(freeze_super);
1341
1342 /**
1343 * thaw_super -- unlock filesystem
1344 * @sb: the super to thaw
1345 *
1346 * Unlocks the filesystem and marks it writeable again after freeze_super().
1347 */
1348 int thaw_super(struct super_block *sb)
1349 {
1350 int error;
1351
1352 down_write(&sb->s_umount);
1353 if (sb->s_writers.frozen == SB_UNFROZEN) {
1354 up_write(&sb->s_umount);
1355 return -EINVAL;
1356 }
1357
1358 if (sb->s_flags & MS_RDONLY)
1359 goto out;
1360
1361 if (sb->s_op->unfreeze_fs) {
1362 error = sb->s_op->unfreeze_fs(sb);
1363 if (error) {
1364 printk(KERN_ERR
1365 "VFS:Filesystem thaw failed\n");
1366 up_write(&sb->s_umount);
1367 return error;
1368 }
1369 }
1370
1371 out:
1372 sb->s_writers.frozen = SB_UNFROZEN;
1373 smp_wmb();
1374 wake_up(&sb->s_writers.wait_unfrozen);
1375 deactivate_locked_super(sb);
1376
1377 return 0;
1378 }
1379 EXPORT_SYMBOL(thaw_super);