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