4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * super.c contains code to handle: - mount structures
8 * - filesystem drivers list
10 * - umount system call
13 * GK 2/5/95 - Changed to support mounting the root fs via NFS
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
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>
39 static LIST_HEAD(super_blocks);
40 static DEFINE_SPINLOCK(sb_lock);
42 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
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.
55 static unsigned long super_cache_scan(struct shrinker *shrink,
56 struct shrink_control *sc)
58 struct super_block *sb;
65 sb = container_of(shrink, struct super_block, s_shrink);
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..
71 if (!(sc->gfp_mask & __GFP_FS))
74 if (!grab_super_passive(sb))
77 if (sb->s_op->nr_cached_objects)
78 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
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;
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);
92 * prune the dcache first as the icache is pinned by it, then
93 * prune the icache, followed by the filesystem specific caches
95 * Ensure that we always scan at least one object - memcg kmem
96 * accounting uses this to fully empty the caches.
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);
104 sc->nr_to_scan = fs_objects + 1;
105 freed += sb->s_op->free_cached_objects(sb, sc);
112 static unsigned long super_cache_count(struct shrinker *shrink,
113 struct shrink_control *sc)
115 struct super_block *sb;
116 long total_objects = 0;
118 sb = container_of(shrink, struct super_block, s_shrink);
121 * Don't call grab_super_passive 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().
128 if (sb->s_op && sb->s_op->nr_cached_objects)
129 total_objects = sb->s_op->nr_cached_objects(sb, sc);
131 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
132 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
134 total_objects = vfs_pressure_ratio(total_objects);
135 return total_objects;
139 * destroy_super - frees a superblock
140 * @s: superblock to free
142 * Frees a superblock.
144 static void destroy_super(struct super_block *s)
147 list_lru_destroy(&s->s_dentry_lru);
148 list_lru_destroy(&s->s_inode_lru);
149 for (i = 0; i < SB_FREEZE_LEVELS; i++)
150 percpu_counter_destroy(&s->s_writers.counter[i]);
152 WARN_ON(!list_empty(&s->s_mounts));
159 * alloc_super - create new superblock
160 * @type: filesystem type superblock should belong to
161 * @flags: the mount flags
163 * Allocates and initializes a new &struct super_block. alloc_super()
164 * returns a pointer new superblock or %NULL if allocation had failed.
166 static struct super_block *alloc_super(struct file_system_type *type, int flags)
168 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
169 static const struct super_operations default_op;
175 INIT_LIST_HEAD(&s->s_mounts);
177 if (security_sb_alloc(s))
180 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
181 if (percpu_counter_init(&s->s_writers.counter[i], 0,
184 lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i],
185 &type->s_writers_key[i], 0);
187 init_waitqueue_head(&s->s_writers.wait);
188 init_waitqueue_head(&s->s_writers.wait_unfrozen);
189 s->s_bdi = &noop_backing_dev_info;
191 INIT_HLIST_NODE(&s->s_instances);
192 INIT_HLIST_BL_HEAD(&s->s_anon);
193 INIT_LIST_HEAD(&s->s_inodes);
195 if (list_lru_init_memcg(&s->s_dentry_lru))
197 if (list_lru_init_memcg(&s->s_inode_lru))
200 init_rwsem(&s->s_umount);
201 lockdep_set_class(&s->s_umount, &type->s_umount_key);
203 * sget() can have s_umount recursion.
205 * When it cannot find a suitable sb, it allocates a new
206 * one (this one), and tries again to find a suitable old
209 * In case that succeeds, it will acquire the s_umount
210 * lock of the old one. Since these are clearly distrinct
211 * locks, and this object isn't exposed yet, there's no
214 * Annotate this by putting this lock in a different
217 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
219 atomic_set(&s->s_active, 1);
220 mutex_init(&s->s_vfs_rename_mutex);
221 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
222 mutex_init(&s->s_dquot.dqio_mutex);
223 mutex_init(&s->s_dquot.dqonoff_mutex);
224 s->s_maxbytes = MAX_NON_LFS;
225 s->s_op = &default_op;
226 s->s_time_gran = 1000000000;
227 s->cleancache_poolid = -1;
229 s->s_shrink.seeks = DEFAULT_SEEKS;
230 s->s_shrink.scan_objects = super_cache_scan;
231 s->s_shrink.count_objects = super_cache_count;
232 s->s_shrink.batch = 1024;
233 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
241 /* Superblock refcounting */
244 * Drop a superblock's refcount. The caller must hold sb_lock.
246 static void __put_super(struct super_block *sb)
248 if (!--sb->s_count) {
249 list_del_init(&sb->s_list);
255 * put_super - drop a temporary reference to superblock
256 * @sb: superblock in question
258 * Drops a temporary reference, frees superblock if there's no
261 static void put_super(struct super_block *sb)
265 spin_unlock(&sb_lock);
270 * deactivate_locked_super - drop an active reference to superblock
271 * @s: superblock to deactivate
273 * Drops an active reference to superblock, converting it into a temprory
274 * one if there is no other active references left. In that case we
275 * tell fs driver to shut it down and drop the temporary reference we
278 * Caller holds exclusive lock on superblock; that lock is released.
280 void deactivate_locked_super(struct super_block *s)
282 struct file_system_type *fs = s->s_type;
283 if (atomic_dec_and_test(&s->s_active)) {
284 cleancache_invalidate_fs(s);
285 unregister_shrinker(&s->s_shrink);
289 * Since list_lru_destroy() may sleep, we cannot call it from
290 * put_super(), where we hold the sb_lock. Therefore we destroy
291 * the lru lists right now.
293 list_lru_destroy(&s->s_dentry_lru);
294 list_lru_destroy(&s->s_inode_lru);
299 up_write(&s->s_umount);
303 EXPORT_SYMBOL(deactivate_locked_super);
306 * deactivate_super - drop an active reference to superblock
307 * @s: superblock to deactivate
309 * Variant of deactivate_locked_super(), except that superblock is *not*
310 * locked by caller. If we are going to drop the final active reference,
311 * lock will be acquired prior to that.
313 void deactivate_super(struct super_block *s)
315 if (!atomic_add_unless(&s->s_active, -1, 1)) {
316 down_write(&s->s_umount);
317 deactivate_locked_super(s);
321 EXPORT_SYMBOL(deactivate_super);
324 * grab_super - acquire an active reference
325 * @s: reference we are trying to make active
327 * Tries to acquire an active reference. grab_super() is used when we
328 * had just found a superblock in super_blocks or fs_type->fs_supers
329 * and want to turn it into a full-blown active reference. grab_super()
330 * is called with sb_lock held and drops it. Returns 1 in case of
331 * success, 0 if we had failed (superblock contents was already dead or
332 * dying when grab_super() had been called). Note that this is only
333 * called for superblocks not in rundown mode (== ones still on ->fs_supers
334 * of their type), so increment of ->s_count is OK here.
336 static int grab_super(struct super_block *s) __releases(sb_lock)
339 spin_unlock(&sb_lock);
340 down_write(&s->s_umount);
341 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
345 up_write(&s->s_umount);
351 * grab_super_passive - acquire a passive reference
352 * @sb: reference we are trying to grab
354 * Tries to acquire a passive reference. This is used in places where we
355 * cannot take an active reference but we need to ensure that the
356 * superblock does not go away while we are working on it. It returns
357 * false if a reference was not gained, and returns true with the s_umount
358 * lock held in read mode if a reference is gained. On successful return,
359 * the caller must drop the s_umount lock and the passive reference when
362 bool grab_super_passive(struct super_block *sb)
365 if (hlist_unhashed(&sb->s_instances)) {
366 spin_unlock(&sb_lock);
371 spin_unlock(&sb_lock);
373 if (down_read_trylock(&sb->s_umount)) {
374 if (sb->s_root && (sb->s_flags & MS_BORN))
376 up_read(&sb->s_umount);
384 * generic_shutdown_super - common helper for ->kill_sb()
385 * @sb: superblock to kill
387 * generic_shutdown_super() does all fs-independent work on superblock
388 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
389 * that need destruction out of superblock, call generic_shutdown_super()
390 * and release aforementioned objects. Note: dentries and inodes _are_
391 * taken care of and do not need specific handling.
393 * Upon calling this function, the filesystem may no longer alter or
394 * rearrange the set of dentries belonging to this super_block, nor may it
395 * change the attachments of dentries to inodes.
397 void generic_shutdown_super(struct super_block *sb)
399 const struct super_operations *sop = sb->s_op;
402 shrink_dcache_for_umount(sb);
404 sb->s_flags &= ~MS_ACTIVE;
406 fsnotify_unmount_inodes(&sb->s_inodes);
410 if (sb->s_dio_done_wq) {
411 destroy_workqueue(sb->s_dio_done_wq);
412 sb->s_dio_done_wq = NULL;
418 if (!list_empty(&sb->s_inodes)) {
419 printk("VFS: Busy inodes after unmount of %s. "
420 "Self-destruct in 5 seconds. Have a nice day...\n",
425 /* should be initialized for __put_super_and_need_restart() */
426 hlist_del_init(&sb->s_instances);
427 spin_unlock(&sb_lock);
428 up_write(&sb->s_umount);
431 EXPORT_SYMBOL(generic_shutdown_super);
434 * sget - find or create a superblock
435 * @type: filesystem type superblock should belong to
436 * @test: comparison callback
437 * @set: setup callback
438 * @flags: mount flags
439 * @data: argument to each of them
441 struct super_block *sget(struct file_system_type *type,
442 int (*test)(struct super_block *,void *),
443 int (*set)(struct super_block *,void *),
447 struct super_block *s = NULL;
448 struct super_block *old;
454 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
455 if (!test(old, data))
457 if (!grab_super(old))
460 up_write(&s->s_umount);
468 spin_unlock(&sb_lock);
469 s = alloc_super(type, flags);
471 return ERR_PTR(-ENOMEM);
477 spin_unlock(&sb_lock);
478 up_write(&s->s_umount);
483 strlcpy(s->s_id, type->name, sizeof(s->s_id));
484 list_add_tail(&s->s_list, &super_blocks);
485 hlist_add_head(&s->s_instances, &type->fs_supers);
486 spin_unlock(&sb_lock);
487 get_filesystem(type);
488 register_shrinker(&s->s_shrink);
494 void drop_super(struct super_block *sb)
496 up_read(&sb->s_umount);
500 EXPORT_SYMBOL(drop_super);
503 * iterate_supers - call function for all active superblocks
504 * @f: function to call
505 * @arg: argument to pass to it
507 * Scans the superblock list and calls given function, passing it
508 * locked superblock and given argument.
510 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
512 struct super_block *sb, *p = NULL;
515 list_for_each_entry(sb, &super_blocks, s_list) {
516 if (hlist_unhashed(&sb->s_instances))
519 spin_unlock(&sb_lock);
521 down_read(&sb->s_umount);
522 if (sb->s_root && (sb->s_flags & MS_BORN))
524 up_read(&sb->s_umount);
533 spin_unlock(&sb_lock);
537 * iterate_supers_type - call function for superblocks of given type
539 * @f: function to call
540 * @arg: argument to pass to it
542 * Scans the superblock list and calls given function, passing it
543 * locked superblock and given argument.
545 void iterate_supers_type(struct file_system_type *type,
546 void (*f)(struct super_block *, void *), void *arg)
548 struct super_block *sb, *p = NULL;
551 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
553 spin_unlock(&sb_lock);
555 down_read(&sb->s_umount);
556 if (sb->s_root && (sb->s_flags & MS_BORN))
558 up_read(&sb->s_umount);
567 spin_unlock(&sb_lock);
570 EXPORT_SYMBOL(iterate_supers_type);
573 * get_super - get the superblock of a device
574 * @bdev: device to get the superblock for
576 * Scans the superblock list and finds the superblock of the file system
577 * mounted on the device given. %NULL is returned if no match is found.
580 struct super_block *get_super(struct block_device *bdev)
582 struct super_block *sb;
589 list_for_each_entry(sb, &super_blocks, s_list) {
590 if (hlist_unhashed(&sb->s_instances))
592 if (sb->s_bdev == bdev) {
594 spin_unlock(&sb_lock);
595 down_read(&sb->s_umount);
597 if (sb->s_root && (sb->s_flags & MS_BORN))
599 up_read(&sb->s_umount);
600 /* nope, got unmounted */
606 spin_unlock(&sb_lock);
610 EXPORT_SYMBOL(get_super);
613 * get_super_thawed - get thawed superblock of a device
614 * @bdev: device to get the superblock for
616 * Scans the superblock list and finds the superblock of the file system
617 * mounted on the device. The superblock is returned once it is thawed
618 * (or immediately if it was not frozen). %NULL is returned if no match
621 struct super_block *get_super_thawed(struct block_device *bdev)
624 struct super_block *s = get_super(bdev);
625 if (!s || s->s_writers.frozen == SB_UNFROZEN)
627 up_read(&s->s_umount);
628 wait_event(s->s_writers.wait_unfrozen,
629 s->s_writers.frozen == SB_UNFROZEN);
633 EXPORT_SYMBOL(get_super_thawed);
636 * get_active_super - get an active reference to the superblock of a device
637 * @bdev: device to get the superblock for
639 * Scans the superblock list and finds the superblock of the file system
640 * mounted on the device given. Returns the superblock with an active
641 * reference or %NULL if none was found.
643 struct super_block *get_active_super(struct block_device *bdev)
645 struct super_block *sb;
652 list_for_each_entry(sb, &super_blocks, s_list) {
653 if (hlist_unhashed(&sb->s_instances))
655 if (sb->s_bdev == bdev) {
658 up_write(&sb->s_umount);
662 spin_unlock(&sb_lock);
666 struct super_block *user_get_super(dev_t dev)
668 struct super_block *sb;
672 list_for_each_entry(sb, &super_blocks, s_list) {
673 if (hlist_unhashed(&sb->s_instances))
675 if (sb->s_dev == dev) {
677 spin_unlock(&sb_lock);
678 down_read(&sb->s_umount);
680 if (sb->s_root && (sb->s_flags & MS_BORN))
682 up_read(&sb->s_umount);
683 /* nope, got unmounted */
689 spin_unlock(&sb_lock);
694 * do_remount_sb - asks filesystem to change mount options.
695 * @sb: superblock in question
696 * @flags: numeric part of options
697 * @data: the rest of options
698 * @force: whether or not to force the change
700 * Alters the mount options of a mounted file system.
702 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
707 if (sb->s_writers.frozen != SB_UNFROZEN)
711 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
715 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
718 if (sb->s_pins.first) {
719 up_write(&sb->s_umount);
721 down_write(&sb->s_umount);
724 if (sb->s_writers.frozen != SB_UNFROZEN)
726 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
729 shrink_dcache_sb(sb);
731 /* If we are remounting RDONLY and current sb is read/write,
732 make sure there are no rw files opened */
735 sb->s_readonly_remount = 1;
738 retval = sb_prepare_remount_readonly(sb);
744 if (sb->s_op->remount_fs) {
745 retval = sb->s_op->remount_fs(sb, &flags, data);
748 goto cancel_readonly;
749 /* If forced remount, go ahead despite any errors */
750 WARN(1, "forced remount of a %s fs returned %i\n",
751 sb->s_type->name, retval);
754 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
755 /* Needs to be ordered wrt mnt_is_readonly() */
757 sb->s_readonly_remount = 0;
760 * Some filesystems modify their metadata via some other path than the
761 * bdev buffer cache (eg. use a private mapping, or directories in
762 * pagecache, etc). Also file data modifications go via their own
763 * mappings. So If we try to mount readonly then copy the filesystem
764 * from bdev, we could get stale data, so invalidate it to give a best
765 * effort at coherency.
767 if (remount_ro && sb->s_bdev)
768 invalidate_bdev(sb->s_bdev);
772 sb->s_readonly_remount = 0;
776 static void do_emergency_remount(struct work_struct *work)
778 struct super_block *sb, *p = NULL;
781 list_for_each_entry(sb, &super_blocks, s_list) {
782 if (hlist_unhashed(&sb->s_instances))
785 spin_unlock(&sb_lock);
786 down_write(&sb->s_umount);
787 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
788 !(sb->s_flags & MS_RDONLY)) {
790 * What lock protects sb->s_flags??
792 do_remount_sb(sb, MS_RDONLY, NULL, 1);
794 up_write(&sb->s_umount);
802 spin_unlock(&sb_lock);
804 printk("Emergency Remount complete\n");
807 void emergency_remount(void)
809 struct work_struct *work;
811 work = kmalloc(sizeof(*work), GFP_ATOMIC);
813 INIT_WORK(work, do_emergency_remount);
819 * Unnamed block devices are dummy devices used by virtual
820 * filesystems which don't use real block-devices. -- jrs
823 static DEFINE_IDA(unnamed_dev_ida);
824 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
825 /* Many userspace utilities consider an FSID of 0 invalid.
826 * Always return at least 1 from get_anon_bdev.
828 static int unnamed_dev_start = 1;
830 int get_anon_bdev(dev_t *p)
836 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
838 spin_lock(&unnamed_dev_lock);
839 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
841 unnamed_dev_start = dev + 1;
842 spin_unlock(&unnamed_dev_lock);
843 if (error == -EAGAIN)
844 /* We raced and lost with another CPU. */
849 if (dev == (1 << MINORBITS)) {
850 spin_lock(&unnamed_dev_lock);
851 ida_remove(&unnamed_dev_ida, dev);
852 if (unnamed_dev_start > dev)
853 unnamed_dev_start = dev;
854 spin_unlock(&unnamed_dev_lock);
857 *p = MKDEV(0, dev & MINORMASK);
860 EXPORT_SYMBOL(get_anon_bdev);
862 void free_anon_bdev(dev_t dev)
864 int slot = MINOR(dev);
865 spin_lock(&unnamed_dev_lock);
866 ida_remove(&unnamed_dev_ida, slot);
867 if (slot < unnamed_dev_start)
868 unnamed_dev_start = slot;
869 spin_unlock(&unnamed_dev_lock);
871 EXPORT_SYMBOL(free_anon_bdev);
873 int set_anon_super(struct super_block *s, void *data)
875 return get_anon_bdev(&s->s_dev);
878 EXPORT_SYMBOL(set_anon_super);
880 void kill_anon_super(struct super_block *sb)
882 dev_t dev = sb->s_dev;
883 generic_shutdown_super(sb);
887 EXPORT_SYMBOL(kill_anon_super);
889 void kill_litter_super(struct super_block *sb)
892 d_genocide(sb->s_root);
896 EXPORT_SYMBOL(kill_litter_super);
898 static int ns_test_super(struct super_block *sb, void *data)
900 return sb->s_fs_info == data;
903 static int ns_set_super(struct super_block *sb, void *data)
905 sb->s_fs_info = data;
906 return set_anon_super(sb, NULL);
909 struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
910 void *data, int (*fill_super)(struct super_block *, void *, int))
912 struct super_block *sb;
914 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
920 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
922 deactivate_locked_super(sb);
926 sb->s_flags |= MS_ACTIVE;
929 return dget(sb->s_root);
932 EXPORT_SYMBOL(mount_ns);
935 static int set_bdev_super(struct super_block *s, void *data)
938 s->s_dev = s->s_bdev->bd_dev;
941 * We set the bdi here to the queue backing, file systems can
942 * overwrite this in ->fill_super()
944 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
948 static int test_bdev_super(struct super_block *s, void *data)
950 return (void *)s->s_bdev == data;
953 struct dentry *mount_bdev(struct file_system_type *fs_type,
954 int flags, const char *dev_name, void *data,
955 int (*fill_super)(struct super_block *, void *, int))
957 struct block_device *bdev;
958 struct super_block *s;
959 fmode_t mode = FMODE_READ | FMODE_EXCL;
962 if (!(flags & MS_RDONLY))
965 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
967 return ERR_CAST(bdev);
970 * once the super is inserted into the list by sget, s_umount
971 * will protect the lockfs code from trying to start a snapshot
972 * while we are mounting
974 mutex_lock(&bdev->bd_fsfreeze_mutex);
975 if (bdev->bd_fsfreeze_count > 0) {
976 mutex_unlock(&bdev->bd_fsfreeze_mutex);
980 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
982 mutex_unlock(&bdev->bd_fsfreeze_mutex);
987 if ((flags ^ s->s_flags) & MS_RDONLY) {
988 deactivate_locked_super(s);
994 * s_umount nests inside bd_mutex during
995 * __invalidate_device(). blkdev_put() acquires
996 * bd_mutex and can't be called under s_umount. Drop
997 * s_umount temporarily. This is safe as we're
998 * holding an active reference.
1000 up_write(&s->s_umount);
1001 blkdev_put(bdev, mode);
1002 down_write(&s->s_umount);
1004 char b[BDEVNAME_SIZE];
1007 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1008 sb_set_blocksize(s, block_size(bdev));
1009 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1011 deactivate_locked_super(s);
1015 s->s_flags |= MS_ACTIVE;
1019 return dget(s->s_root);
1024 blkdev_put(bdev, mode);
1026 return ERR_PTR(error);
1028 EXPORT_SYMBOL(mount_bdev);
1030 void kill_block_super(struct super_block *sb)
1032 struct block_device *bdev = sb->s_bdev;
1033 fmode_t mode = sb->s_mode;
1035 bdev->bd_super = NULL;
1036 generic_shutdown_super(sb);
1037 sync_blockdev(bdev);
1038 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1039 blkdev_put(bdev, mode | FMODE_EXCL);
1042 EXPORT_SYMBOL(kill_block_super);
1045 struct dentry *mount_nodev(struct file_system_type *fs_type,
1046 int flags, void *data,
1047 int (*fill_super)(struct super_block *, void *, int))
1050 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1055 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1057 deactivate_locked_super(s);
1058 return ERR_PTR(error);
1060 s->s_flags |= MS_ACTIVE;
1061 return dget(s->s_root);
1063 EXPORT_SYMBOL(mount_nodev);
1065 static int compare_single(struct super_block *s, void *p)
1070 struct dentry *mount_single(struct file_system_type *fs_type,
1071 int flags, void *data,
1072 int (*fill_super)(struct super_block *, void *, int))
1074 struct super_block *s;
1077 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1081 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1083 deactivate_locked_super(s);
1084 return ERR_PTR(error);
1086 s->s_flags |= MS_ACTIVE;
1088 do_remount_sb(s, flags, data, 0);
1090 return dget(s->s_root);
1092 EXPORT_SYMBOL(mount_single);
1095 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1097 struct dentry *root;
1098 struct super_block *sb;
1099 char *secdata = NULL;
1100 int error = -ENOMEM;
1102 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1103 secdata = alloc_secdata();
1107 error = security_sb_copy_data(data, secdata);
1109 goto out_free_secdata;
1112 root = type->mount(type, flags, name, data);
1114 error = PTR_ERR(root);
1115 goto out_free_secdata;
1119 WARN_ON(!sb->s_bdi);
1120 sb->s_flags |= MS_BORN;
1122 error = security_sb_kern_mount(sb, flags, secdata);
1127 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1128 * but s_maxbytes was an unsigned long long for many releases. Throw
1129 * this warning for a little while to try and catch filesystems that
1130 * violate this rule.
1132 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1133 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1135 up_write(&sb->s_umount);
1136 free_secdata(secdata);
1140 deactivate_locked_super(sb);
1142 free_secdata(secdata);
1144 return ERR_PTR(error);
1148 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1151 void __sb_end_write(struct super_block *sb, int level)
1153 percpu_counter_dec(&sb->s_writers.counter[level-1]);
1155 * Make sure s_writers are updated before we wake up waiters in
1159 if (waitqueue_active(&sb->s_writers.wait))
1160 wake_up(&sb->s_writers.wait);
1161 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_);
1163 EXPORT_SYMBOL(__sb_end_write);
1165 #ifdef CONFIG_LOCKDEP
1167 * We want lockdep to tell us about possible deadlocks with freezing but
1168 * it's it bit tricky to properly instrument it. Getting a freeze protection
1169 * works as getting a read lock but there are subtle problems. XFS for example
1170 * gets freeze protection on internal level twice in some cases, which is OK
1171 * only because we already hold a freeze protection also on higher level. Due
1172 * to these cases we have to tell lockdep we are doing trylock when we
1173 * already hold a freeze protection for a higher freeze level.
1175 static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock,
1181 for (i = 0; i < level - 1; i++)
1182 if (lock_is_held(&sb->s_writers.lock_map[i])) {
1187 rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip);
1192 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1195 int __sb_start_write(struct super_block *sb, int level, bool wait)
1198 if (unlikely(sb->s_writers.frozen >= level)) {
1201 wait_event(sb->s_writers.wait_unfrozen,
1202 sb->s_writers.frozen < level);
1205 #ifdef CONFIG_LOCKDEP
1206 acquire_freeze_lock(sb, level, !wait, _RET_IP_);
1208 percpu_counter_inc(&sb->s_writers.counter[level-1]);
1210 * Make sure counter is updated before we check for frozen.
1211 * freeze_super() first sets frozen and then checks the counter.
1214 if (unlikely(sb->s_writers.frozen >= level)) {
1215 __sb_end_write(sb, level);
1220 EXPORT_SYMBOL(__sb_start_write);
1223 * sb_wait_write - wait until all writers to given file system finish
1224 * @sb: the super for which we wait
1225 * @level: type of writers we wait for (normal vs page fault)
1227 * This function waits until there are no writers of given type to given file
1228 * system. Caller of this function should make sure there can be no new writers
1229 * of type @level before calling this function. Otherwise this function can
1232 static void sb_wait_write(struct super_block *sb, int level)
1237 * We just cycle-through lockdep here so that it does not complain
1238 * about returning with lock to userspace
1240 rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_);
1241 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_);
1247 * We use a barrier in prepare_to_wait() to separate setting
1248 * of frozen and checking of the counter
1250 prepare_to_wait(&sb->s_writers.wait, &wait,
1251 TASK_UNINTERRUPTIBLE);
1253 writers = percpu_counter_sum(&sb->s_writers.counter[level-1]);
1257 finish_wait(&sb->s_writers.wait, &wait);
1262 * freeze_super - lock the filesystem and force it into a consistent state
1263 * @sb: the super to lock
1265 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1266 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1269 * During this function, sb->s_writers.frozen goes through these values:
1271 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1273 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1274 * writes should be blocked, though page faults are still allowed. We wait for
1275 * all writes to complete and then proceed to the next stage.
1277 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1278 * but internal fs threads can still modify the filesystem (although they
1279 * should not dirty new pages or inodes), writeback can run etc. After waiting
1280 * for all running page faults we sync the filesystem which will clean all
1281 * dirty pages and inodes (no new dirty pages or inodes can be created when
1284 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1285 * modification are blocked (e.g. XFS preallocation truncation on inode
1286 * reclaim). This is usually implemented by blocking new transactions for
1287 * filesystems that have them and need this additional guard. After all
1288 * internal writers are finished we call ->freeze_fs() to finish filesystem
1289 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1290 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1292 * sb->s_writers.frozen is protected by sb->s_umount.
1294 int freeze_super(struct super_block *sb)
1298 atomic_inc(&sb->s_active);
1299 down_write(&sb->s_umount);
1300 if (sb->s_writers.frozen != SB_UNFROZEN) {
1301 deactivate_locked_super(sb);
1305 if (!(sb->s_flags & MS_BORN)) {
1306 up_write(&sb->s_umount);
1307 return 0; /* sic - it's "nothing to do" */
1310 if (sb->s_flags & MS_RDONLY) {
1311 /* Nothing to do really... */
1312 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1313 up_write(&sb->s_umount);
1317 /* From now on, no new normal writers can start */
1318 sb->s_writers.frozen = SB_FREEZE_WRITE;
1321 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1322 up_write(&sb->s_umount);
1324 sb_wait_write(sb, SB_FREEZE_WRITE);
1326 /* Now we go and block page faults... */
1327 down_write(&sb->s_umount);
1328 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1331 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1333 /* All writers are done so after syncing there won't be dirty data */
1334 sync_filesystem(sb);
1336 /* Now wait for internal filesystem counter */
1337 sb->s_writers.frozen = SB_FREEZE_FS;
1339 sb_wait_write(sb, SB_FREEZE_FS);
1341 if (sb->s_op->freeze_fs) {
1342 ret = sb->s_op->freeze_fs(sb);
1345 "VFS:Filesystem freeze failed\n");
1346 sb->s_writers.frozen = SB_UNFROZEN;
1348 wake_up(&sb->s_writers.wait_unfrozen);
1349 deactivate_locked_super(sb);
1354 * This is just for debugging purposes so that fs can warn if it
1355 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1357 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1358 up_write(&sb->s_umount);
1361 EXPORT_SYMBOL(freeze_super);
1364 * thaw_super -- unlock filesystem
1365 * @sb: the super to thaw
1367 * Unlocks the filesystem and marks it writeable again after freeze_super().
1369 int thaw_super(struct super_block *sb)
1373 down_write(&sb->s_umount);
1374 if (sb->s_writers.frozen == SB_UNFROZEN) {
1375 up_write(&sb->s_umount);
1379 if (sb->s_flags & MS_RDONLY)
1382 if (sb->s_op->unfreeze_fs) {
1383 error = sb->s_op->unfreeze_fs(sb);
1386 "VFS:Filesystem thaw failed\n");
1387 up_write(&sb->s_umount);
1393 sb->s_writers.frozen = SB_UNFROZEN;
1395 wake_up(&sb->s_writers.wait_unfrozen);
1396 deactivate_locked_super(sb);
1400 EXPORT_SYMBOL(thaw_super);