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/fscrypt.h>
35#include <linux/fsnotify.h>
36#include <linux/lockdep.h>
37#include <linux/user_namespace.h>
38#include <linux/fs_context.h>
39#include <uapi/linux/mount.h>
40#include "internal.h"
41
42static int thaw_super_locked(struct super_block *sb);
43
44static LIST_HEAD(super_blocks);
45static DEFINE_SPINLOCK(sb_lock);
46
47static char *sb_writers_name[SB_FREEZE_LEVELS] = {
48 "sb_writers",
49 "sb_pagefaults",
50 "sb_internal",
51};
52
53/*
54 * One thing we have to be careful of with a per-sb shrinker is that we don't
55 * drop the last active reference to the superblock from within the shrinker.
56 * If that happens we could trigger unregistering the shrinker from within the
57 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
58 * take a passive reference to the superblock to avoid this from occurring.
59 */
60static unsigned long super_cache_scan(struct shrinker *shrink,
61 struct shrink_control *sc)
62{
63 struct super_block *sb;
64 long fs_objects = 0;
65 long total_objects;
66 long freed = 0;
67 long dentries;
68 long inodes;
69
70 sb = container_of(shrink, struct super_block, s_shrink);
71
72 /*
73 * Deadlock avoidance. We may hold various FS locks, and we don't want
74 * to recurse into the FS that called us in clear_inode() and friends..
75 */
76 if (!(sc->gfp_mask & __GFP_FS))
77 return SHRINK_STOP;
78
79 if (!trylock_super(sb))
80 return SHRINK_STOP;
81
82 if (sb->s_op->nr_cached_objects)
83 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
84
85 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
86 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
87 total_objects = dentries + inodes + fs_objects + 1;
88 if (!total_objects)
89 total_objects = 1;
90
91 /* proportion the scan between the caches */
92 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
93 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
94 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
95
96 /*
97 * prune the dcache first as the icache is pinned by it, then
98 * prune the icache, followed by the filesystem specific caches
99 *
100 * Ensure that we always scan at least one object - memcg kmem
101 * accounting uses this to fully empty the caches.
102 */
103 sc->nr_to_scan = dentries + 1;
104 freed = prune_dcache_sb(sb, sc);
105 sc->nr_to_scan = inodes + 1;
106 freed += prune_icache_sb(sb, sc);
107
108 if (fs_objects) {
109 sc->nr_to_scan = fs_objects + 1;
110 freed += sb->s_op->free_cached_objects(sb, sc);
111 }
112
113 up_read(&sb->s_umount);
114 return freed;
115}
116
117static unsigned long super_cache_count(struct shrinker *shrink,
118 struct shrink_control *sc)
119{
120 struct super_block *sb;
121 long total_objects = 0;
122
123 sb = container_of(shrink, struct super_block, s_shrink);
124
125 /*
126 * We don't call trylock_super() here as it is a scalability bottleneck,
127 * so we're exposed to partial setup state. The shrinker rwsem does not
128 * protect filesystem operations backing list_lru_shrink_count() or
129 * s_op->nr_cached_objects(). Counts can change between
130 * super_cache_count and super_cache_scan, so we really don't need locks
131 * here.
132 *
133 * However, if we are currently mounting the superblock, the underlying
134 * filesystem might be in a state of partial construction and hence it
135 * is dangerous to access it. trylock_super() uses a SB_BORN check to
136 * avoid this situation, so do the same here. The memory barrier is
137 * matched with the one in mount_fs() as we don't hold locks here.
138 */
139 if (!(sb->s_flags & SB_BORN))
140 return 0;
141 smp_rmb();
142
143 if (sb->s_op && sb->s_op->nr_cached_objects)
144 total_objects = sb->s_op->nr_cached_objects(sb, sc);
145
146 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
147 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
148
149 if (!total_objects)
150 return SHRINK_EMPTY;
151
152 total_objects = vfs_pressure_ratio(total_objects);
153 return total_objects;
154}
155
156static void destroy_super_work(struct work_struct *work)
157{
158 struct super_block *s = container_of(work, struct super_block,
159 destroy_work);
160 int i;
161
162 for (i = 0; i < SB_FREEZE_LEVELS; i++)
163 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
164 kfree(s);
165}
166
167static void destroy_super_rcu(struct rcu_head *head)
168{
169 struct super_block *s = container_of(head, struct super_block, rcu);
170 INIT_WORK(&s->destroy_work, destroy_super_work);
171 schedule_work(&s->destroy_work);
172}
173
174/* Free a superblock that has never been seen by anyone */
175static void destroy_unused_super(struct super_block *s)
176{
177 if (!s)
178 return;
179 up_write(&s->s_umount);
180 list_lru_destroy(&s->s_dentry_lru);
181 list_lru_destroy(&s->s_inode_lru);
182 security_sb_free(s);
183 put_user_ns(s->s_user_ns);
184 kfree(s->s_subtype);
185 free_prealloced_shrinker(&s->s_shrink);
186 /* no delays needed */
187 destroy_super_work(&s->destroy_work);
188}
189
190/**
191 * alloc_super - create new superblock
192 * @type: filesystem type superblock should belong to
193 * @flags: the mount flags
194 * @user_ns: User namespace for the super_block
195 *
196 * Allocates and initializes a new &struct super_block. alloc_super()
197 * returns a pointer new superblock or %NULL if allocation had failed.
198 */
199static struct super_block *alloc_super(struct file_system_type *type, int flags,
200 struct user_namespace *user_ns)
201{
202 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
203 static const struct super_operations default_op;
204 int i;
205
206 if (!s)
207 return NULL;
208
209 INIT_LIST_HEAD(&s->s_mounts);
210 s->s_user_ns = get_user_ns(user_ns);
211 init_rwsem(&s->s_umount);
212 lockdep_set_class(&s->s_umount, &type->s_umount_key);
213 /*
214 * sget() can have s_umount recursion.
215 *
216 * When it cannot find a suitable sb, it allocates a new
217 * one (this one), and tries again to find a suitable old
218 * one.
219 *
220 * In case that succeeds, it will acquire the s_umount
221 * lock of the old one. Since these are clearly distrinct
222 * locks, and this object isn't exposed yet, there's no
223 * risk of deadlocks.
224 *
225 * Annotate this by putting this lock in a different
226 * subclass.
227 */
228 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
229
230 if (security_sb_alloc(s))
231 goto fail;
232
233 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
234 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
235 sb_writers_name[i],
236 &type->s_writers_key[i]))
237 goto fail;
238 }
239 init_waitqueue_head(&s->s_writers.wait_unfrozen);
240 s->s_bdi = &noop_backing_dev_info;
241 s->s_flags = flags;
242 if (s->s_user_ns != &init_user_ns)
243 s->s_iflags |= SB_I_NODEV;
244 INIT_HLIST_NODE(&s->s_instances);
245 INIT_HLIST_BL_HEAD(&s->s_roots);
246 mutex_init(&s->s_sync_lock);
247 INIT_LIST_HEAD(&s->s_inodes);
248 spin_lock_init(&s->s_inode_list_lock);
249 INIT_LIST_HEAD(&s->s_inodes_wb);
250 spin_lock_init(&s->s_inode_wblist_lock);
251
252 s->s_count = 1;
253 atomic_set(&s->s_active, 1);
254 mutex_init(&s->s_vfs_rename_mutex);
255 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
256 init_rwsem(&s->s_dquot.dqio_sem);
257 s->s_maxbytes = MAX_NON_LFS;
258 s->s_op = &default_op;
259 s->s_time_gran = 1000000000;
260 s->s_time_min = TIME64_MIN;
261 s->s_time_max = TIME64_MAX;
262
263 s->s_shrink.seeks = DEFAULT_SEEKS;
264 s->s_shrink.scan_objects = super_cache_scan;
265 s->s_shrink.count_objects = super_cache_count;
266 s->s_shrink.batch = 1024;
267 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
268 if (prealloc_shrinker(&s->s_shrink, "sb-%s", type->name))
269 goto fail;
270 if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
271 goto fail;
272 if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
273 goto fail;
274 return s;
275
276fail:
277 destroy_unused_super(s);
278 return NULL;
279}
280
281/* Superblock refcounting */
282
283/*
284 * Drop a superblock's refcount. The caller must hold sb_lock.
285 */
286static void __put_super(struct super_block *s)
287{
288 if (!--s->s_count) {
289 list_del_init(&s->s_list);
290 WARN_ON(s->s_dentry_lru.node);
291 WARN_ON(s->s_inode_lru.node);
292 WARN_ON(!list_empty(&s->s_mounts));
293 security_sb_free(s);
294 fscrypt_destroy_keyring(s);
295 put_user_ns(s->s_user_ns);
296 kfree(s->s_subtype);
297 call_rcu(&s->rcu, destroy_super_rcu);
298 }
299}
300
301/**
302 * put_super - drop a temporary reference to superblock
303 * @sb: superblock in question
304 *
305 * Drops a temporary reference, frees superblock if there's no
306 * references left.
307 */
308void put_super(struct super_block *sb)
309{
310 spin_lock(&sb_lock);
311 __put_super(sb);
312 spin_unlock(&sb_lock);
313}
314
315
316/**
317 * deactivate_locked_super - drop an active reference to superblock
318 * @s: superblock to deactivate
319 *
320 * Drops an active reference to superblock, converting it into a temporary
321 * one if there is no other active references left. In that case we
322 * tell fs driver to shut it down and drop the temporary reference we
323 * had just acquired.
324 *
325 * Caller holds exclusive lock on superblock; that lock is released.
326 */
327void deactivate_locked_super(struct super_block *s)
328{
329 struct file_system_type *fs = s->s_type;
330 if (atomic_dec_and_test(&s->s_active)) {
331 unregister_shrinker(&s->s_shrink);
332 fs->kill_sb(s);
333
334 /*
335 * Since list_lru_destroy() may sleep, we cannot call it from
336 * put_super(), where we hold the sb_lock. Therefore we destroy
337 * the lru lists right now.
338 */
339 list_lru_destroy(&s->s_dentry_lru);
340 list_lru_destroy(&s->s_inode_lru);
341
342 put_filesystem(fs);
343 put_super(s);
344 } else {
345 up_write(&s->s_umount);
346 }
347}
348
349EXPORT_SYMBOL(deactivate_locked_super);
350
351/**
352 * deactivate_super - drop an active reference to superblock
353 * @s: superblock to deactivate
354 *
355 * Variant of deactivate_locked_super(), except that superblock is *not*
356 * locked by caller. If we are going to drop the final active reference,
357 * lock will be acquired prior to that.
358 */
359void deactivate_super(struct super_block *s)
360{
361 if (!atomic_add_unless(&s->s_active, -1, 1)) {
362 down_write(&s->s_umount);
363 deactivate_locked_super(s);
364 }
365}
366
367EXPORT_SYMBOL(deactivate_super);
368
369/**
370 * grab_super - acquire an active reference
371 * @s: reference we are trying to make active
372 *
373 * Tries to acquire an active reference. grab_super() is used when we
374 * had just found a superblock in super_blocks or fs_type->fs_supers
375 * and want to turn it into a full-blown active reference. grab_super()
376 * is called with sb_lock held and drops it. Returns 1 in case of
377 * success, 0 if we had failed (superblock contents was already dead or
378 * dying when grab_super() had been called). Note that this is only
379 * called for superblocks not in rundown mode (== ones still on ->fs_supers
380 * of their type), so increment of ->s_count is OK here.
381 */
382static int grab_super(struct super_block *s) __releases(sb_lock)
383{
384 s->s_count++;
385 spin_unlock(&sb_lock);
386 down_write(&s->s_umount);
387 if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
388 put_super(s);
389 return 1;
390 }
391 up_write(&s->s_umount);
392 put_super(s);
393 return 0;
394}
395
396/*
397 * trylock_super - try to grab ->s_umount shared
398 * @sb: reference we are trying to grab
399 *
400 * Try to prevent fs shutdown. This is used in places where we
401 * cannot take an active reference but we need to ensure that the
402 * filesystem is not shut down while we are working on it. It returns
403 * false if we cannot acquire s_umount or if we lose the race and
404 * filesystem already got into shutdown, and returns true with the s_umount
405 * lock held in read mode in case of success. On successful return,
406 * the caller must drop the s_umount lock when done.
407 *
408 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
409 * The reason why it's safe is that we are OK with doing trylock instead
410 * of down_read(). There's a couple of places that are OK with that, but
411 * it's very much not a general-purpose interface.
412 */
413bool trylock_super(struct super_block *sb)
414{
415 if (down_read_trylock(&sb->s_umount)) {
416 if (!hlist_unhashed(&sb->s_instances) &&
417 sb->s_root && (sb->s_flags & SB_BORN))
418 return true;
419 up_read(&sb->s_umount);
420 }
421
422 return false;
423}
424
425/**
426 * retire_super - prevents superblock from being reused
427 * @sb: superblock to retire
428 *
429 * The function marks superblock to be ignored in superblock test, which
430 * prevents it from being reused for any new mounts. If the superblock has
431 * a private bdi, it also unregisters it, but doesn't reduce the refcount
432 * of the superblock to prevent potential races. The refcount is reduced
433 * by generic_shutdown_super(). The function can not be called
434 * concurrently with generic_shutdown_super(). It is safe to call the
435 * function multiple times, subsequent calls have no effect.
436 *
437 * The marker will affect the re-use only for block-device-based
438 * superblocks. Other superblocks will still get marked if this function
439 * is used, but that will not affect their reusability.
440 */
441void retire_super(struct super_block *sb)
442{
443 WARN_ON(!sb->s_bdev);
444 down_write(&sb->s_umount);
445 if (sb->s_iflags & SB_I_PERSB_BDI) {
446 bdi_unregister(sb->s_bdi);
447 sb->s_iflags &= ~SB_I_PERSB_BDI;
448 }
449 sb->s_iflags |= SB_I_RETIRED;
450 up_write(&sb->s_umount);
451}
452EXPORT_SYMBOL(retire_super);
453
454/**
455 * generic_shutdown_super - common helper for ->kill_sb()
456 * @sb: superblock to kill
457 *
458 * generic_shutdown_super() does all fs-independent work on superblock
459 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
460 * that need destruction out of superblock, call generic_shutdown_super()
461 * and release aforementioned objects. Note: dentries and inodes _are_
462 * taken care of and do not need specific handling.
463 *
464 * Upon calling this function, the filesystem may no longer alter or
465 * rearrange the set of dentries belonging to this super_block, nor may it
466 * change the attachments of dentries to inodes.
467 */
468void generic_shutdown_super(struct super_block *sb)
469{
470 const struct super_operations *sop = sb->s_op;
471
472 if (sb->s_root) {
473 shrink_dcache_for_umount(sb);
474 sync_filesystem(sb);
475 sb->s_flags &= ~SB_ACTIVE;
476
477 cgroup_writeback_umount();
478
479 /* evict all inodes with zero refcount */
480 evict_inodes(sb);
481 /* only nonzero refcount inodes can have marks */
482 fsnotify_sb_delete(sb);
483 fscrypt_destroy_keyring(sb);
484 security_sb_delete(sb);
485
486 if (sb->s_dio_done_wq) {
487 destroy_workqueue(sb->s_dio_done_wq);
488 sb->s_dio_done_wq = NULL;
489 }
490
491 if (sop->put_super)
492 sop->put_super(sb);
493
494 if (!list_empty(&sb->s_inodes)) {
495 printk("VFS: Busy inodes after unmount of %s. "
496 "Self-destruct in 5 seconds. Have a nice day...\n",
497 sb->s_id);
498 }
499 }
500 spin_lock(&sb_lock);
501 /* should be initialized for __put_super_and_need_restart() */
502 hlist_del_init(&sb->s_instances);
503 spin_unlock(&sb_lock);
504 up_write(&sb->s_umount);
505 if (sb->s_bdi != &noop_backing_dev_info) {
506 if (sb->s_iflags & SB_I_PERSB_BDI)
507 bdi_unregister(sb->s_bdi);
508 bdi_put(sb->s_bdi);
509 sb->s_bdi = &noop_backing_dev_info;
510 }
511}
512
513EXPORT_SYMBOL(generic_shutdown_super);
514
515bool mount_capable(struct fs_context *fc)
516{
517 if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT))
518 return capable(CAP_SYS_ADMIN);
519 else
520 return ns_capable(fc->user_ns, CAP_SYS_ADMIN);
521}
522
523/**
524 * sget_fc - Find or create a superblock
525 * @fc: Filesystem context.
526 * @test: Comparison callback
527 * @set: Setup callback
528 *
529 * Find or create a superblock using the parameters stored in the filesystem
530 * context and the two callback functions.
531 *
532 * If an extant superblock is matched, then that will be returned with an
533 * elevated reference count that the caller must transfer or discard.
534 *
535 * If no match is made, a new superblock will be allocated and basic
536 * initialisation will be performed (s_type, s_fs_info and s_id will be set and
537 * the set() callback will be invoked), the superblock will be published and it
538 * will be returned in a partially constructed state with SB_BORN and SB_ACTIVE
539 * as yet unset.
540 */
541struct super_block *sget_fc(struct fs_context *fc,
542 int (*test)(struct super_block *, struct fs_context *),
543 int (*set)(struct super_block *, struct fs_context *))
544{
545 struct super_block *s = NULL;
546 struct super_block *old;
547 struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
548 int err;
549
550retry:
551 spin_lock(&sb_lock);
552 if (test) {
553 hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
554 if (test(old, fc))
555 goto share_extant_sb;
556 }
557 }
558 if (!s) {
559 spin_unlock(&sb_lock);
560 s = alloc_super(fc->fs_type, fc->sb_flags, user_ns);
561 if (!s)
562 return ERR_PTR(-ENOMEM);
563 goto retry;
564 }
565
566 s->s_fs_info = fc->s_fs_info;
567 err = set(s, fc);
568 if (err) {
569 s->s_fs_info = NULL;
570 spin_unlock(&sb_lock);
571 destroy_unused_super(s);
572 return ERR_PTR(err);
573 }
574 fc->s_fs_info = NULL;
575 s->s_type = fc->fs_type;
576 s->s_iflags |= fc->s_iflags;
577 strlcpy(s->s_id, s->s_type->name, sizeof(s->s_id));
578 list_add_tail(&s->s_list, &super_blocks);
579 hlist_add_head(&s->s_instances, &s->s_type->fs_supers);
580 spin_unlock(&sb_lock);
581 get_filesystem(s->s_type);
582 register_shrinker_prepared(&s->s_shrink);
583 return s;
584
585share_extant_sb:
586 if (user_ns != old->s_user_ns) {
587 spin_unlock(&sb_lock);
588 destroy_unused_super(s);
589 return ERR_PTR(-EBUSY);
590 }
591 if (!grab_super(old))
592 goto retry;
593 destroy_unused_super(s);
594 return old;
595}
596EXPORT_SYMBOL(sget_fc);
597
598/**
599 * sget - find or create a superblock
600 * @type: filesystem type superblock should belong to
601 * @test: comparison callback
602 * @set: setup callback
603 * @flags: mount flags
604 * @data: argument to each of them
605 */
606struct super_block *sget(struct file_system_type *type,
607 int (*test)(struct super_block *,void *),
608 int (*set)(struct super_block *,void *),
609 int flags,
610 void *data)
611{
612 struct user_namespace *user_ns = current_user_ns();
613 struct super_block *s = NULL;
614 struct super_block *old;
615 int err;
616
617 /* We don't yet pass the user namespace of the parent
618 * mount through to here so always use &init_user_ns
619 * until that changes.
620 */
621 if (flags & SB_SUBMOUNT)
622 user_ns = &init_user_ns;
623
624retry:
625 spin_lock(&sb_lock);
626 if (test) {
627 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
628 if (!test(old, data))
629 continue;
630 if (user_ns != old->s_user_ns) {
631 spin_unlock(&sb_lock);
632 destroy_unused_super(s);
633 return ERR_PTR(-EBUSY);
634 }
635 if (!grab_super(old))
636 goto retry;
637 destroy_unused_super(s);
638 return old;
639 }
640 }
641 if (!s) {
642 spin_unlock(&sb_lock);
643 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
644 if (!s)
645 return ERR_PTR(-ENOMEM);
646 goto retry;
647 }
648
649 err = set(s, data);
650 if (err) {
651 spin_unlock(&sb_lock);
652 destroy_unused_super(s);
653 return ERR_PTR(err);
654 }
655 s->s_type = type;
656 strlcpy(s->s_id, type->name, sizeof(s->s_id));
657 list_add_tail(&s->s_list, &super_blocks);
658 hlist_add_head(&s->s_instances, &type->fs_supers);
659 spin_unlock(&sb_lock);
660 get_filesystem(type);
661 register_shrinker_prepared(&s->s_shrink);
662 return s;
663}
664EXPORT_SYMBOL(sget);
665
666void drop_super(struct super_block *sb)
667{
668 up_read(&sb->s_umount);
669 put_super(sb);
670}
671
672EXPORT_SYMBOL(drop_super);
673
674void drop_super_exclusive(struct super_block *sb)
675{
676 up_write(&sb->s_umount);
677 put_super(sb);
678}
679EXPORT_SYMBOL(drop_super_exclusive);
680
681static void __iterate_supers(void (*f)(struct super_block *))
682{
683 struct super_block *sb, *p = NULL;
684
685 spin_lock(&sb_lock);
686 list_for_each_entry(sb, &super_blocks, s_list) {
687 if (hlist_unhashed(&sb->s_instances))
688 continue;
689 sb->s_count++;
690 spin_unlock(&sb_lock);
691
692 f(sb);
693
694 spin_lock(&sb_lock);
695 if (p)
696 __put_super(p);
697 p = sb;
698 }
699 if (p)
700 __put_super(p);
701 spin_unlock(&sb_lock);
702}
703/**
704 * iterate_supers - call function for all active superblocks
705 * @f: function to call
706 * @arg: argument to pass to it
707 *
708 * Scans the superblock list and calls given function, passing it
709 * locked superblock and given argument.
710 */
711void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
712{
713 struct super_block *sb, *p = NULL;
714
715 spin_lock(&sb_lock);
716 list_for_each_entry(sb, &super_blocks, s_list) {
717 if (hlist_unhashed(&sb->s_instances))
718 continue;
719 sb->s_count++;
720 spin_unlock(&sb_lock);
721
722 down_read(&sb->s_umount);
723 if (sb->s_root && (sb->s_flags & SB_BORN))
724 f(sb, arg);
725 up_read(&sb->s_umount);
726
727 spin_lock(&sb_lock);
728 if (p)
729 __put_super(p);
730 p = sb;
731 }
732 if (p)
733 __put_super(p);
734 spin_unlock(&sb_lock);
735}
736
737/**
738 * iterate_supers_type - call function for superblocks of given type
739 * @type: fs type
740 * @f: function to call
741 * @arg: argument to pass to it
742 *
743 * Scans the superblock list and calls given function, passing it
744 * locked superblock and given argument.
745 */
746void iterate_supers_type(struct file_system_type *type,
747 void (*f)(struct super_block *, void *), void *arg)
748{
749 struct super_block *sb, *p = NULL;
750
751 spin_lock(&sb_lock);
752 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
753 sb->s_count++;
754 spin_unlock(&sb_lock);
755
756 down_read(&sb->s_umount);
757 if (sb->s_root && (sb->s_flags & SB_BORN))
758 f(sb, arg);
759 up_read(&sb->s_umount);
760
761 spin_lock(&sb_lock);
762 if (p)
763 __put_super(p);
764 p = sb;
765 }
766 if (p)
767 __put_super(p);
768 spin_unlock(&sb_lock);
769}
770
771EXPORT_SYMBOL(iterate_supers_type);
772
773/**
774 * get_super - get the superblock of a device
775 * @bdev: device to get the superblock for
776 *
777 * Scans the superblock list and finds the superblock of the file system
778 * mounted on the device given. %NULL is returned if no match is found.
779 */
780struct super_block *get_super(struct block_device *bdev)
781{
782 struct super_block *sb;
783
784 if (!bdev)
785 return NULL;
786
787 spin_lock(&sb_lock);
788rescan:
789 list_for_each_entry(sb, &super_blocks, s_list) {
790 if (hlist_unhashed(&sb->s_instances))
791 continue;
792 if (sb->s_bdev == bdev) {
793 sb->s_count++;
794 spin_unlock(&sb_lock);
795 down_read(&sb->s_umount);
796 /* still alive? */
797 if (sb->s_root && (sb->s_flags & SB_BORN))
798 return sb;
799 up_read(&sb->s_umount);
800 /* nope, got unmounted */
801 spin_lock(&sb_lock);
802 __put_super(sb);
803 goto rescan;
804 }
805 }
806 spin_unlock(&sb_lock);
807 return NULL;
808}
809
810/**
811 * get_active_super - get an active reference to the superblock of a device
812 * @bdev: device to get the superblock for
813 *
814 * Scans the superblock list and finds the superblock of the file system
815 * mounted on the device given. Returns the superblock with an active
816 * reference or %NULL if none was found.
817 */
818struct super_block *get_active_super(struct block_device *bdev)
819{
820 struct super_block *sb;
821
822 if (!bdev)
823 return NULL;
824
825restart:
826 spin_lock(&sb_lock);
827 list_for_each_entry(sb, &super_blocks, s_list) {
828 if (hlist_unhashed(&sb->s_instances))
829 continue;
830 if (sb->s_bdev == bdev) {
831 if (!grab_super(sb))
832 goto restart;
833 up_write(&sb->s_umount);
834 return sb;
835 }
836 }
837 spin_unlock(&sb_lock);
838 return NULL;
839}
840
841struct super_block *user_get_super(dev_t dev, bool excl)
842{
843 struct super_block *sb;
844
845 spin_lock(&sb_lock);
846rescan:
847 list_for_each_entry(sb, &super_blocks, s_list) {
848 if (hlist_unhashed(&sb->s_instances))
849 continue;
850 if (sb->s_dev == dev) {
851 sb->s_count++;
852 spin_unlock(&sb_lock);
853 if (excl)
854 down_write(&sb->s_umount);
855 else
856 down_read(&sb->s_umount);
857 /* still alive? */
858 if (sb->s_root && (sb->s_flags & SB_BORN))
859 return sb;
860 if (excl)
861 up_write(&sb->s_umount);
862 else
863 up_read(&sb->s_umount);
864 /* nope, got unmounted */
865 spin_lock(&sb_lock);
866 __put_super(sb);
867 goto rescan;
868 }
869 }
870 spin_unlock(&sb_lock);
871 return NULL;
872}
873
874/**
875 * reconfigure_super - asks filesystem to change superblock parameters
876 * @fc: The superblock and configuration
877 *
878 * Alters the configuration parameters of a live superblock.
879 */
880int reconfigure_super(struct fs_context *fc)
881{
882 struct super_block *sb = fc->root->d_sb;
883 int retval;
884 bool remount_ro = false;
885 bool force = fc->sb_flags & SB_FORCE;
886
887 if (fc->sb_flags_mask & ~MS_RMT_MASK)
888 return -EINVAL;
889 if (sb->s_writers.frozen != SB_UNFROZEN)
890 return -EBUSY;
891
892 retval = security_sb_remount(sb, fc->security);
893 if (retval)
894 return retval;
895
896 if (fc->sb_flags_mask & SB_RDONLY) {
897#ifdef CONFIG_BLOCK
898 if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev &&
899 bdev_read_only(sb->s_bdev))
900 return -EACCES;
901#endif
902
903 remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
904 }
905
906 if (remount_ro) {
907 if (!hlist_empty(&sb->s_pins)) {
908 up_write(&sb->s_umount);
909 group_pin_kill(&sb->s_pins);
910 down_write(&sb->s_umount);
911 if (!sb->s_root)
912 return 0;
913 if (sb->s_writers.frozen != SB_UNFROZEN)
914 return -EBUSY;
915 remount_ro = !sb_rdonly(sb);
916 }
917 }
918 shrink_dcache_sb(sb);
919
920 /* If we are reconfiguring to RDONLY and current sb is read/write,
921 * make sure there are no files open for writing.
922 */
923 if (remount_ro) {
924 if (force) {
925 sb->s_readonly_remount = 1;
926 smp_wmb();
927 } else {
928 retval = sb_prepare_remount_readonly(sb);
929 if (retval)
930 return retval;
931 }
932 }
933
934 if (fc->ops->reconfigure) {
935 retval = fc->ops->reconfigure(fc);
936 if (retval) {
937 if (!force)
938 goto cancel_readonly;
939 /* If forced remount, go ahead despite any errors */
940 WARN(1, "forced remount of a %s fs returned %i\n",
941 sb->s_type->name, retval);
942 }
943 }
944
945 WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
946 (fc->sb_flags & fc->sb_flags_mask)));
947 /* Needs to be ordered wrt mnt_is_readonly() */
948 smp_wmb();
949 sb->s_readonly_remount = 0;
950
951 /*
952 * Some filesystems modify their metadata via some other path than the
953 * bdev buffer cache (eg. use a private mapping, or directories in
954 * pagecache, etc). Also file data modifications go via their own
955 * mappings. So If we try to mount readonly then copy the filesystem
956 * from bdev, we could get stale data, so invalidate it to give a best
957 * effort at coherency.
958 */
959 if (remount_ro && sb->s_bdev)
960 invalidate_bdev(sb->s_bdev);
961 return 0;
962
963cancel_readonly:
964 sb->s_readonly_remount = 0;
965 return retval;
966}
967
968static void do_emergency_remount_callback(struct super_block *sb)
969{
970 down_write(&sb->s_umount);
971 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
972 !sb_rdonly(sb)) {
973 struct fs_context *fc;
974
975 fc = fs_context_for_reconfigure(sb->s_root,
976 SB_RDONLY | SB_FORCE, SB_RDONLY);
977 if (!IS_ERR(fc)) {
978 if (parse_monolithic_mount_data(fc, NULL) == 0)
979 (void)reconfigure_super(fc);
980 put_fs_context(fc);
981 }
982 }
983 up_write(&sb->s_umount);
984}
985
986static void do_emergency_remount(struct work_struct *work)
987{
988 __iterate_supers(do_emergency_remount_callback);
989 kfree(work);
990 printk("Emergency Remount complete\n");
991}
992
993void emergency_remount(void)
994{
995 struct work_struct *work;
996
997 work = kmalloc(sizeof(*work), GFP_ATOMIC);
998 if (work) {
999 INIT_WORK(work, do_emergency_remount);
1000 schedule_work(work);
1001 }
1002}
1003
1004static void do_thaw_all_callback(struct super_block *sb)
1005{
1006 down_write(&sb->s_umount);
1007 if (sb->s_root && sb->s_flags & SB_BORN) {
1008 emergency_thaw_bdev(sb);
1009 thaw_super_locked(sb);
1010 } else {
1011 up_write(&sb->s_umount);
1012 }
1013}
1014
1015static void do_thaw_all(struct work_struct *work)
1016{
1017 __iterate_supers(do_thaw_all_callback);
1018 kfree(work);
1019 printk(KERN_WARNING "Emergency Thaw complete\n");
1020}
1021
1022/**
1023 * emergency_thaw_all -- forcibly thaw every frozen filesystem
1024 *
1025 * Used for emergency unfreeze of all filesystems via SysRq
1026 */
1027void emergency_thaw_all(void)
1028{
1029 struct work_struct *work;
1030
1031 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1032 if (work) {
1033 INIT_WORK(work, do_thaw_all);
1034 schedule_work(work);
1035 }
1036}
1037
1038static DEFINE_IDA(unnamed_dev_ida);
1039
1040/**
1041 * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1042 * @p: Pointer to a dev_t.
1043 *
1044 * Filesystems which don't use real block devices can call this function
1045 * to allocate a virtual block device.
1046 *
1047 * Context: Any context. Frequently called while holding sb_lock.
1048 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1049 * or -ENOMEM if memory allocation failed.
1050 */
1051int get_anon_bdev(dev_t *p)
1052{
1053 int dev;
1054
1055 /*
1056 * Many userspace utilities consider an FSID of 0 invalid.
1057 * Always return at least 1 from get_anon_bdev.
1058 */
1059 dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1060 GFP_ATOMIC);
1061 if (dev == -ENOSPC)
1062 dev = -EMFILE;
1063 if (dev < 0)
1064 return dev;
1065
1066 *p = MKDEV(0, dev);
1067 return 0;
1068}
1069EXPORT_SYMBOL(get_anon_bdev);
1070
1071void free_anon_bdev(dev_t dev)
1072{
1073 ida_free(&unnamed_dev_ida, MINOR(dev));
1074}
1075EXPORT_SYMBOL(free_anon_bdev);
1076
1077int set_anon_super(struct super_block *s, void *data)
1078{
1079 return get_anon_bdev(&s->s_dev);
1080}
1081EXPORT_SYMBOL(set_anon_super);
1082
1083void kill_anon_super(struct super_block *sb)
1084{
1085 dev_t dev = sb->s_dev;
1086 generic_shutdown_super(sb);
1087 free_anon_bdev(dev);
1088}
1089EXPORT_SYMBOL(kill_anon_super);
1090
1091void kill_litter_super(struct super_block *sb)
1092{
1093 if (sb->s_root)
1094 d_genocide(sb->s_root);
1095 kill_anon_super(sb);
1096}
1097EXPORT_SYMBOL(kill_litter_super);
1098
1099int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1100{
1101 return set_anon_super(sb, NULL);
1102}
1103EXPORT_SYMBOL(set_anon_super_fc);
1104
1105static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1106{
1107 return sb->s_fs_info == fc->s_fs_info;
1108}
1109
1110static int test_single_super(struct super_block *s, struct fs_context *fc)
1111{
1112 return 1;
1113}
1114
1115static int vfs_get_super(struct fs_context *fc, bool reconf,
1116 int (*test)(struct super_block *, struct fs_context *),
1117 int (*fill_super)(struct super_block *sb,
1118 struct fs_context *fc))
1119{
1120 struct super_block *sb;
1121 int err;
1122
1123 sb = sget_fc(fc, test, set_anon_super_fc);
1124 if (IS_ERR(sb))
1125 return PTR_ERR(sb);
1126
1127 if (!sb->s_root) {
1128 err = fill_super(sb, fc);
1129 if (err)
1130 goto error;
1131
1132 sb->s_flags |= SB_ACTIVE;
1133 fc->root = dget(sb->s_root);
1134 } else {
1135 fc->root = dget(sb->s_root);
1136 if (reconf) {
1137 err = reconfigure_super(fc);
1138 if (err < 0) {
1139 dput(fc->root);
1140 fc->root = NULL;
1141 goto error;
1142 }
1143 }
1144 }
1145
1146 return 0;
1147
1148error:
1149 deactivate_locked_super(sb);
1150 return err;
1151}
1152
1153int get_tree_nodev(struct fs_context *fc,
1154 int (*fill_super)(struct super_block *sb,
1155 struct fs_context *fc))
1156{
1157 return vfs_get_super(fc, false, NULL, fill_super);
1158}
1159EXPORT_SYMBOL(get_tree_nodev);
1160
1161int get_tree_single(struct fs_context *fc,
1162 int (*fill_super)(struct super_block *sb,
1163 struct fs_context *fc))
1164{
1165 return vfs_get_super(fc, false, test_single_super, fill_super);
1166}
1167EXPORT_SYMBOL(get_tree_single);
1168
1169int get_tree_single_reconf(struct fs_context *fc,
1170 int (*fill_super)(struct super_block *sb,
1171 struct fs_context *fc))
1172{
1173 return vfs_get_super(fc, true, test_single_super, fill_super);
1174}
1175EXPORT_SYMBOL(get_tree_single_reconf);
1176
1177int get_tree_keyed(struct fs_context *fc,
1178 int (*fill_super)(struct super_block *sb,
1179 struct fs_context *fc),
1180 void *key)
1181{
1182 fc->s_fs_info = key;
1183 return vfs_get_super(fc, false, test_keyed_super, fill_super);
1184}
1185EXPORT_SYMBOL(get_tree_keyed);
1186
1187#ifdef CONFIG_BLOCK
1188
1189static int set_bdev_super(struct super_block *s, void *data)
1190{
1191 s->s_bdev = data;
1192 s->s_dev = s->s_bdev->bd_dev;
1193 s->s_bdi = bdi_get(s->s_bdev->bd_disk->bdi);
1194
1195 if (bdev_stable_writes(s->s_bdev))
1196 s->s_iflags |= SB_I_STABLE_WRITES;
1197 return 0;
1198}
1199
1200static int set_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1201{
1202 return set_bdev_super(s, fc->sget_key);
1203}
1204
1205static int test_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1206{
1207 return !(s->s_iflags & SB_I_RETIRED) && s->s_bdev == fc->sget_key;
1208}
1209
1210/**
1211 * get_tree_bdev - Get a superblock based on a single block device
1212 * @fc: The filesystem context holding the parameters
1213 * @fill_super: Helper to initialise a new superblock
1214 */
1215int get_tree_bdev(struct fs_context *fc,
1216 int (*fill_super)(struct super_block *,
1217 struct fs_context *))
1218{
1219 struct block_device *bdev;
1220 struct super_block *s;
1221 fmode_t mode = FMODE_READ | FMODE_EXCL;
1222 int error = 0;
1223
1224 if (!(fc->sb_flags & SB_RDONLY))
1225 mode |= FMODE_WRITE;
1226
1227 if (!fc->source)
1228 return invalf(fc, "No source specified");
1229
1230 bdev = blkdev_get_by_path(fc->source, mode, fc->fs_type);
1231 if (IS_ERR(bdev)) {
1232 errorf(fc, "%s: Can't open blockdev", fc->source);
1233 return PTR_ERR(bdev);
1234 }
1235
1236 /* Once the superblock is inserted into the list by sget_fc(), s_umount
1237 * will protect the lockfs code from trying to start a snapshot while
1238 * we are mounting
1239 */
1240 mutex_lock(&bdev->bd_fsfreeze_mutex);
1241 if (bdev->bd_fsfreeze_count > 0) {
1242 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1243 warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
1244 blkdev_put(bdev, mode);
1245 return -EBUSY;
1246 }
1247
1248 fc->sb_flags |= SB_NOSEC;
1249 fc->sget_key = bdev;
1250 s = sget_fc(fc, test_bdev_super_fc, set_bdev_super_fc);
1251 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1252 if (IS_ERR(s)) {
1253 blkdev_put(bdev, mode);
1254 return PTR_ERR(s);
1255 }
1256
1257 if (s->s_root) {
1258 /* Don't summarily change the RO/RW state. */
1259 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1260 warnf(fc, "%pg: Can't mount, would change RO state", bdev);
1261 deactivate_locked_super(s);
1262 blkdev_put(bdev, mode);
1263 return -EBUSY;
1264 }
1265
1266 /*
1267 * s_umount nests inside open_mutex during
1268 * __invalidate_device(). blkdev_put() acquires
1269 * open_mutex and can't be called under s_umount. Drop
1270 * s_umount temporarily. This is safe as we're
1271 * holding an active reference.
1272 */
1273 up_write(&s->s_umount);
1274 blkdev_put(bdev, mode);
1275 down_write(&s->s_umount);
1276 } else {
1277 s->s_mode = mode;
1278 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1279 shrinker_debugfs_rename(&s->s_shrink, "sb-%s:%s",
1280 fc->fs_type->name, s->s_id);
1281 sb_set_blocksize(s, block_size(bdev));
1282 error = fill_super(s, fc);
1283 if (error) {
1284 deactivate_locked_super(s);
1285 return error;
1286 }
1287
1288 s->s_flags |= SB_ACTIVE;
1289 bdev->bd_super = s;
1290 }
1291
1292 BUG_ON(fc->root);
1293 fc->root = dget(s->s_root);
1294 return 0;
1295}
1296EXPORT_SYMBOL(get_tree_bdev);
1297
1298static int test_bdev_super(struct super_block *s, void *data)
1299{
1300 return !(s->s_iflags & SB_I_RETIRED) && (void *)s->s_bdev == data;
1301}
1302
1303struct dentry *mount_bdev(struct file_system_type *fs_type,
1304 int flags, const char *dev_name, void *data,
1305 int (*fill_super)(struct super_block *, void *, int))
1306{
1307 struct block_device *bdev;
1308 struct super_block *s;
1309 fmode_t mode = FMODE_READ | FMODE_EXCL;
1310 int error = 0;
1311
1312 if (!(flags & SB_RDONLY))
1313 mode |= FMODE_WRITE;
1314
1315 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1316 if (IS_ERR(bdev))
1317 return ERR_CAST(bdev);
1318
1319 /*
1320 * once the super is inserted into the list by sget, s_umount
1321 * will protect the lockfs code from trying to start a snapshot
1322 * while we are mounting
1323 */
1324 mutex_lock(&bdev->bd_fsfreeze_mutex);
1325 if (bdev->bd_fsfreeze_count > 0) {
1326 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1327 error = -EBUSY;
1328 goto error_bdev;
1329 }
1330 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
1331 bdev);
1332 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1333 if (IS_ERR(s))
1334 goto error_s;
1335
1336 if (s->s_root) {
1337 if ((flags ^ s->s_flags) & SB_RDONLY) {
1338 deactivate_locked_super(s);
1339 error = -EBUSY;
1340 goto error_bdev;
1341 }
1342
1343 /*
1344 * s_umount nests inside open_mutex during
1345 * __invalidate_device(). blkdev_put() acquires
1346 * open_mutex and can't be called under s_umount. Drop
1347 * s_umount temporarily. This is safe as we're
1348 * holding an active reference.
1349 */
1350 up_write(&s->s_umount);
1351 blkdev_put(bdev, mode);
1352 down_write(&s->s_umount);
1353 } else {
1354 s->s_mode = mode;
1355 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1356 shrinker_debugfs_rename(&s->s_shrink, "sb-%s:%s",
1357 fs_type->name, s->s_id);
1358 sb_set_blocksize(s, block_size(bdev));
1359 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1360 if (error) {
1361 deactivate_locked_super(s);
1362 goto error;
1363 }
1364
1365 s->s_flags |= SB_ACTIVE;
1366 bdev->bd_super = s;
1367 }
1368
1369 return dget(s->s_root);
1370
1371error_s:
1372 error = PTR_ERR(s);
1373error_bdev:
1374 blkdev_put(bdev, mode);
1375error:
1376 return ERR_PTR(error);
1377}
1378EXPORT_SYMBOL(mount_bdev);
1379
1380void kill_block_super(struct super_block *sb)
1381{
1382 struct block_device *bdev = sb->s_bdev;
1383 fmode_t mode = sb->s_mode;
1384
1385 bdev->bd_super = NULL;
1386 generic_shutdown_super(sb);
1387 sync_blockdev(bdev);
1388 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1389 blkdev_put(bdev, mode | FMODE_EXCL);
1390}
1391
1392EXPORT_SYMBOL(kill_block_super);
1393#endif
1394
1395struct dentry *mount_nodev(struct file_system_type *fs_type,
1396 int flags, void *data,
1397 int (*fill_super)(struct super_block *, void *, int))
1398{
1399 int error;
1400 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1401
1402 if (IS_ERR(s))
1403 return ERR_CAST(s);
1404
1405 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1406 if (error) {
1407 deactivate_locked_super(s);
1408 return ERR_PTR(error);
1409 }
1410 s->s_flags |= SB_ACTIVE;
1411 return dget(s->s_root);
1412}
1413EXPORT_SYMBOL(mount_nodev);
1414
1415int reconfigure_single(struct super_block *s,
1416 int flags, void *data)
1417{
1418 struct fs_context *fc;
1419 int ret;
1420
1421 /* The caller really need to be passing fc down into mount_single(),
1422 * then a chunk of this can be removed. [Bollocks -- AV]
1423 * Better yet, reconfiguration shouldn't happen, but rather the second
1424 * mount should be rejected if the parameters are not compatible.
1425 */
1426 fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1427 if (IS_ERR(fc))
1428 return PTR_ERR(fc);
1429
1430 ret = parse_monolithic_mount_data(fc, data);
1431 if (ret < 0)
1432 goto out;
1433
1434 ret = reconfigure_super(fc);
1435out:
1436 put_fs_context(fc);
1437 return ret;
1438}
1439
1440static int compare_single(struct super_block *s, void *p)
1441{
1442 return 1;
1443}
1444
1445struct dentry *mount_single(struct file_system_type *fs_type,
1446 int flags, void *data,
1447 int (*fill_super)(struct super_block *, void *, int))
1448{
1449 struct super_block *s;
1450 int error;
1451
1452 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1453 if (IS_ERR(s))
1454 return ERR_CAST(s);
1455 if (!s->s_root) {
1456 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1457 if (!error)
1458 s->s_flags |= SB_ACTIVE;
1459 } else {
1460 error = reconfigure_single(s, flags, data);
1461 }
1462 if (unlikely(error)) {
1463 deactivate_locked_super(s);
1464 return ERR_PTR(error);
1465 }
1466 return dget(s->s_root);
1467}
1468EXPORT_SYMBOL(mount_single);
1469
1470/**
1471 * vfs_get_tree - Get the mountable root
1472 * @fc: The superblock configuration context.
1473 *
1474 * The filesystem is invoked to get or create a superblock which can then later
1475 * be used for mounting. The filesystem places a pointer to the root to be
1476 * used for mounting in @fc->root.
1477 */
1478int vfs_get_tree(struct fs_context *fc)
1479{
1480 struct super_block *sb;
1481 int error;
1482
1483 if (fc->root)
1484 return -EBUSY;
1485
1486 /* Get the mountable root in fc->root, with a ref on the root and a ref
1487 * on the superblock.
1488 */
1489 error = fc->ops->get_tree(fc);
1490 if (error < 0)
1491 return error;
1492
1493 if (!fc->root) {
1494 pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1495 fc->fs_type->name);
1496 /* We don't know what the locking state of the superblock is -
1497 * if there is a superblock.
1498 */
1499 BUG();
1500 }
1501
1502 sb = fc->root->d_sb;
1503 WARN_ON(!sb->s_bdi);
1504
1505 /*
1506 * Write barrier is for super_cache_count(). We place it before setting
1507 * SB_BORN as the data dependency between the two functions is the
1508 * superblock structure contents that we just set up, not the SB_BORN
1509 * flag.
1510 */
1511 smp_wmb();
1512 sb->s_flags |= SB_BORN;
1513
1514 error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
1515 if (unlikely(error)) {
1516 fc_drop_locked(fc);
1517 return error;
1518 }
1519
1520 /*
1521 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1522 * but s_maxbytes was an unsigned long long for many releases. Throw
1523 * this warning for a little while to try and catch filesystems that
1524 * violate this rule.
1525 */
1526 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1527 "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
1528
1529 return 0;
1530}
1531EXPORT_SYMBOL(vfs_get_tree);
1532
1533/*
1534 * Setup private BDI for given superblock. It gets automatically cleaned up
1535 * in generic_shutdown_super().
1536 */
1537int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1538{
1539 struct backing_dev_info *bdi;
1540 int err;
1541 va_list args;
1542
1543 bdi = bdi_alloc(NUMA_NO_NODE);
1544 if (!bdi)
1545 return -ENOMEM;
1546
1547 va_start(args, fmt);
1548 err = bdi_register_va(bdi, fmt, args);
1549 va_end(args);
1550 if (err) {
1551 bdi_put(bdi);
1552 return err;
1553 }
1554 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1555 sb->s_bdi = bdi;
1556 sb->s_iflags |= SB_I_PERSB_BDI;
1557
1558 return 0;
1559}
1560EXPORT_SYMBOL(super_setup_bdi_name);
1561
1562/*
1563 * Setup private BDI for given superblock. I gets automatically cleaned up
1564 * in generic_shutdown_super().
1565 */
1566int super_setup_bdi(struct super_block *sb)
1567{
1568 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1569
1570 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1571 atomic_long_inc_return(&bdi_seq));
1572}
1573EXPORT_SYMBOL(super_setup_bdi);
1574
1575/**
1576 * sb_wait_write - wait until all writers to given file system finish
1577 * @sb: the super for which we wait
1578 * @level: type of writers we wait for (normal vs page fault)
1579 *
1580 * This function waits until there are no writers of given type to given file
1581 * system.
1582 */
1583static void sb_wait_write(struct super_block *sb, int level)
1584{
1585 percpu_down_write(sb->s_writers.rw_sem + level-1);
1586}
1587
1588/*
1589 * We are going to return to userspace and forget about these locks, the
1590 * ownership goes to the caller of thaw_super() which does unlock().
1591 */
1592static void lockdep_sb_freeze_release(struct super_block *sb)
1593{
1594 int level;
1595
1596 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1597 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1598}
1599
1600/*
1601 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1602 */
1603static void lockdep_sb_freeze_acquire(struct super_block *sb)
1604{
1605 int level;
1606
1607 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1608 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1609}
1610
1611static void sb_freeze_unlock(struct super_block *sb, int level)
1612{
1613 for (level--; level >= 0; level--)
1614 percpu_up_write(sb->s_writers.rw_sem + level);
1615}
1616
1617/**
1618 * freeze_super - lock the filesystem and force it into a consistent state
1619 * @sb: the super to lock
1620 *
1621 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1622 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1623 * -EBUSY.
1624 *
1625 * During this function, sb->s_writers.frozen goes through these values:
1626 *
1627 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1628 *
1629 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1630 * writes should be blocked, though page faults are still allowed. We wait for
1631 * all writes to complete and then proceed to the next stage.
1632 *
1633 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1634 * but internal fs threads can still modify the filesystem (although they
1635 * should not dirty new pages or inodes), writeback can run etc. After waiting
1636 * for all running page faults we sync the filesystem which will clean all
1637 * dirty pages and inodes (no new dirty pages or inodes can be created when
1638 * sync is running).
1639 *
1640 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1641 * modification are blocked (e.g. XFS preallocation truncation on inode
1642 * reclaim). This is usually implemented by blocking new transactions for
1643 * filesystems that have them and need this additional guard. After all
1644 * internal writers are finished we call ->freeze_fs() to finish filesystem
1645 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1646 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1647 *
1648 * sb->s_writers.frozen is protected by sb->s_umount.
1649 */
1650int freeze_super(struct super_block *sb)
1651{
1652 int ret;
1653
1654 atomic_inc(&sb->s_active);
1655 down_write(&sb->s_umount);
1656 if (sb->s_writers.frozen != SB_UNFROZEN) {
1657 deactivate_locked_super(sb);
1658 return -EBUSY;
1659 }
1660
1661 if (!(sb->s_flags & SB_BORN)) {
1662 up_write(&sb->s_umount);
1663 return 0; /* sic - it's "nothing to do" */
1664 }
1665
1666 if (sb_rdonly(sb)) {
1667 /* Nothing to do really... */
1668 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1669 up_write(&sb->s_umount);
1670 return 0;
1671 }
1672
1673 sb->s_writers.frozen = SB_FREEZE_WRITE;
1674 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1675 up_write(&sb->s_umount);
1676 sb_wait_write(sb, SB_FREEZE_WRITE);
1677 down_write(&sb->s_umount);
1678
1679 /* Now we go and block page faults... */
1680 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1681 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1682
1683 /* All writers are done so after syncing there won't be dirty data */
1684 ret = sync_filesystem(sb);
1685 if (ret) {
1686 sb->s_writers.frozen = SB_UNFROZEN;
1687 sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT);
1688 wake_up(&sb->s_writers.wait_unfrozen);
1689 deactivate_locked_super(sb);
1690 return ret;
1691 }
1692
1693 /* Now wait for internal filesystem counter */
1694 sb->s_writers.frozen = SB_FREEZE_FS;
1695 sb_wait_write(sb, SB_FREEZE_FS);
1696
1697 if (sb->s_op->freeze_fs) {
1698 ret = sb->s_op->freeze_fs(sb);
1699 if (ret) {
1700 printk(KERN_ERR
1701 "VFS:Filesystem freeze failed\n");
1702 sb->s_writers.frozen = SB_UNFROZEN;
1703 sb_freeze_unlock(sb, SB_FREEZE_FS);
1704 wake_up(&sb->s_writers.wait_unfrozen);
1705 deactivate_locked_super(sb);
1706 return ret;
1707 }
1708 }
1709 /*
1710 * For debugging purposes so that fs can warn if it sees write activity
1711 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1712 */
1713 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1714 lockdep_sb_freeze_release(sb);
1715 up_write(&sb->s_umount);
1716 return 0;
1717}
1718EXPORT_SYMBOL(freeze_super);
1719
1720static int thaw_super_locked(struct super_block *sb)
1721{
1722 int error;
1723
1724 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1725 up_write(&sb->s_umount);
1726 return -EINVAL;
1727 }
1728
1729 if (sb_rdonly(sb)) {
1730 sb->s_writers.frozen = SB_UNFROZEN;
1731 goto out;
1732 }
1733
1734 lockdep_sb_freeze_acquire(sb);
1735
1736 if (sb->s_op->unfreeze_fs) {
1737 error = sb->s_op->unfreeze_fs(sb);
1738 if (error) {
1739 printk(KERN_ERR
1740 "VFS:Filesystem thaw failed\n");
1741 lockdep_sb_freeze_release(sb);
1742 up_write(&sb->s_umount);
1743 return error;
1744 }
1745 }
1746
1747 sb->s_writers.frozen = SB_UNFROZEN;
1748 sb_freeze_unlock(sb, SB_FREEZE_FS);
1749out:
1750 wake_up(&sb->s_writers.wait_unfrozen);
1751 deactivate_locked_super(sb);
1752 return 0;
1753}
1754
1755/**
1756 * thaw_super -- unlock filesystem
1757 * @sb: the super to thaw
1758 *
1759 * Unlocks the filesystem and marks it writeable again after freeze_super().
1760 */
1761int thaw_super(struct super_block *sb)
1762{
1763 down_write(&sb->s_umount);
1764 return thaw_super_locked(sb);
1765}
1766EXPORT_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/fscrypt.h>
35#include <linux/fsnotify.h>
36#include <linux/lockdep.h>
37#include <linux/user_namespace.h>
38#include <linux/fs_context.h>
39#include <uapi/linux/mount.h>
40#include "internal.h"
41
42static int thaw_super_locked(struct super_block *sb, enum freeze_holder who);
43
44static LIST_HEAD(super_blocks);
45static DEFINE_SPINLOCK(sb_lock);
46
47static char *sb_writers_name[SB_FREEZE_LEVELS] = {
48 "sb_writers",
49 "sb_pagefaults",
50 "sb_internal",
51};
52
53static inline void __super_lock(struct super_block *sb, bool excl)
54{
55 if (excl)
56 down_write(&sb->s_umount);
57 else
58 down_read(&sb->s_umount);
59}
60
61static inline void super_unlock(struct super_block *sb, bool excl)
62{
63 if (excl)
64 up_write(&sb->s_umount);
65 else
66 up_read(&sb->s_umount);
67}
68
69static inline void __super_lock_excl(struct super_block *sb)
70{
71 __super_lock(sb, true);
72}
73
74static inline void super_unlock_excl(struct super_block *sb)
75{
76 super_unlock(sb, true);
77}
78
79static inline void super_unlock_shared(struct super_block *sb)
80{
81 super_unlock(sb, false);
82}
83
84static bool super_flags(const struct super_block *sb, unsigned int flags)
85{
86 /*
87 * Pairs with smp_store_release() in super_wake() and ensures
88 * that we see @flags after we're woken.
89 */
90 return smp_load_acquire(&sb->s_flags) & flags;
91}
92
93/**
94 * super_lock - wait for superblock to become ready and lock it
95 * @sb: superblock to wait for
96 * @excl: whether exclusive access is required
97 *
98 * If the superblock has neither passed through vfs_get_tree() or
99 * generic_shutdown_super() yet wait for it to happen. Either superblock
100 * creation will succeed and SB_BORN is set by vfs_get_tree() or we're
101 * woken and we'll see SB_DYING.
102 *
103 * The caller must have acquired a temporary reference on @sb->s_count.
104 *
105 * Return: The function returns true if SB_BORN was set and with
106 * s_umount held. The function returns false if SB_DYING was
107 * set and without s_umount held.
108 */
109static __must_check bool super_lock(struct super_block *sb, bool excl)
110{
111 lockdep_assert_not_held(&sb->s_umount);
112
113 /* wait until the superblock is ready or dying */
114 wait_var_event(&sb->s_flags, super_flags(sb, SB_BORN | SB_DYING));
115
116 /* Don't pointlessly acquire s_umount. */
117 if (super_flags(sb, SB_DYING))
118 return false;
119
120 __super_lock(sb, excl);
121
122 /*
123 * Has gone through generic_shutdown_super() in the meantime.
124 * @sb->s_root is NULL and @sb->s_active is 0. No one needs to
125 * grab a reference to this. Tell them so.
126 */
127 if (sb->s_flags & SB_DYING) {
128 super_unlock(sb, excl);
129 return false;
130 }
131
132 WARN_ON_ONCE(!(sb->s_flags & SB_BORN));
133 return true;
134}
135
136/* wait and try to acquire read-side of @sb->s_umount */
137static inline bool super_lock_shared(struct super_block *sb)
138{
139 return super_lock(sb, false);
140}
141
142/* wait and try to acquire write-side of @sb->s_umount */
143static inline bool super_lock_excl(struct super_block *sb)
144{
145 return super_lock(sb, true);
146}
147
148/* wake waiters */
149#define SUPER_WAKE_FLAGS (SB_BORN | SB_DYING | SB_DEAD)
150static void super_wake(struct super_block *sb, unsigned int flag)
151{
152 WARN_ON_ONCE((flag & ~SUPER_WAKE_FLAGS));
153 WARN_ON_ONCE(hweight32(flag & SUPER_WAKE_FLAGS) > 1);
154
155 /*
156 * Pairs with smp_load_acquire() in super_lock() to make sure
157 * all initializations in the superblock are seen by the user
158 * seeing SB_BORN sent.
159 */
160 smp_store_release(&sb->s_flags, sb->s_flags | flag);
161 /*
162 * Pairs with the barrier in prepare_to_wait_event() to make sure
163 * ___wait_var_event() either sees SB_BORN set or
164 * waitqueue_active() check in wake_up_var() sees the waiter.
165 */
166 smp_mb();
167 wake_up_var(&sb->s_flags);
168}
169
170/*
171 * One thing we have to be careful of with a per-sb shrinker is that we don't
172 * drop the last active reference to the superblock from within the shrinker.
173 * If that happens we could trigger unregistering the shrinker from within the
174 * shrinker path and that leads to deadlock on the shrinker_mutex. Hence we
175 * take a passive reference to the superblock to avoid this from occurring.
176 */
177static unsigned long super_cache_scan(struct shrinker *shrink,
178 struct shrink_control *sc)
179{
180 struct super_block *sb;
181 long fs_objects = 0;
182 long total_objects;
183 long freed = 0;
184 long dentries;
185 long inodes;
186
187 sb = shrink->private_data;
188
189 /*
190 * Deadlock avoidance. We may hold various FS locks, and we don't want
191 * to recurse into the FS that called us in clear_inode() and friends..
192 */
193 if (!(sc->gfp_mask & __GFP_FS))
194 return SHRINK_STOP;
195
196 if (!super_trylock_shared(sb))
197 return SHRINK_STOP;
198
199 if (sb->s_op->nr_cached_objects)
200 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
201
202 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
203 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
204 total_objects = dentries + inodes + fs_objects + 1;
205 if (!total_objects)
206 total_objects = 1;
207
208 /* proportion the scan between the caches */
209 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
210 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
211 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
212
213 /*
214 * prune the dcache first as the icache is pinned by it, then
215 * prune the icache, followed by the filesystem specific caches
216 *
217 * Ensure that we always scan at least one object - memcg kmem
218 * accounting uses this to fully empty the caches.
219 */
220 sc->nr_to_scan = dentries + 1;
221 freed = prune_dcache_sb(sb, sc);
222 sc->nr_to_scan = inodes + 1;
223 freed += prune_icache_sb(sb, sc);
224
225 if (fs_objects) {
226 sc->nr_to_scan = fs_objects + 1;
227 freed += sb->s_op->free_cached_objects(sb, sc);
228 }
229
230 super_unlock_shared(sb);
231 return freed;
232}
233
234static unsigned long super_cache_count(struct shrinker *shrink,
235 struct shrink_control *sc)
236{
237 struct super_block *sb;
238 long total_objects = 0;
239
240 sb = shrink->private_data;
241
242 /*
243 * We don't call super_trylock_shared() here as it is a scalability
244 * bottleneck, so we're exposed to partial setup state. The shrinker
245 * rwsem does not protect filesystem operations backing
246 * list_lru_shrink_count() or s_op->nr_cached_objects(). Counts can
247 * change between super_cache_count and super_cache_scan, so we really
248 * don't need locks here.
249 *
250 * However, if we are currently mounting the superblock, the underlying
251 * filesystem might be in a state of partial construction and hence it
252 * is dangerous to access it. super_trylock_shared() uses a SB_BORN check
253 * to avoid this situation, so do the same here. The memory barrier is
254 * matched with the one in mount_fs() as we don't hold locks here.
255 */
256 if (!(sb->s_flags & SB_BORN))
257 return 0;
258 smp_rmb();
259
260 if (sb->s_op && sb->s_op->nr_cached_objects)
261 total_objects = sb->s_op->nr_cached_objects(sb, sc);
262
263 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
264 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
265
266 if (!total_objects)
267 return SHRINK_EMPTY;
268
269 total_objects = vfs_pressure_ratio(total_objects);
270 return total_objects;
271}
272
273static void destroy_super_work(struct work_struct *work)
274{
275 struct super_block *s = container_of(work, struct super_block,
276 destroy_work);
277 fsnotify_sb_free(s);
278 security_sb_free(s);
279 put_user_ns(s->s_user_ns);
280 kfree(s->s_subtype);
281 for (int i = 0; i < SB_FREEZE_LEVELS; i++)
282 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
283 kfree(s);
284}
285
286static void destroy_super_rcu(struct rcu_head *head)
287{
288 struct super_block *s = container_of(head, struct super_block, rcu);
289 INIT_WORK(&s->destroy_work, destroy_super_work);
290 schedule_work(&s->destroy_work);
291}
292
293/* Free a superblock that has never been seen by anyone */
294static void destroy_unused_super(struct super_block *s)
295{
296 if (!s)
297 return;
298 super_unlock_excl(s);
299 list_lru_destroy(&s->s_dentry_lru);
300 list_lru_destroy(&s->s_inode_lru);
301 shrinker_free(s->s_shrink);
302 /* no delays needed */
303 destroy_super_work(&s->destroy_work);
304}
305
306/**
307 * alloc_super - create new superblock
308 * @type: filesystem type superblock should belong to
309 * @flags: the mount flags
310 * @user_ns: User namespace for the super_block
311 *
312 * Allocates and initializes a new &struct super_block. alloc_super()
313 * returns a pointer new superblock or %NULL if allocation had failed.
314 */
315static struct super_block *alloc_super(struct file_system_type *type, int flags,
316 struct user_namespace *user_ns)
317{
318 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_KERNEL);
319 static const struct super_operations default_op;
320 int i;
321
322 if (!s)
323 return NULL;
324
325 INIT_LIST_HEAD(&s->s_mounts);
326 s->s_user_ns = get_user_ns(user_ns);
327 init_rwsem(&s->s_umount);
328 lockdep_set_class(&s->s_umount, &type->s_umount_key);
329 /*
330 * sget() can have s_umount recursion.
331 *
332 * When it cannot find a suitable sb, it allocates a new
333 * one (this one), and tries again to find a suitable old
334 * one.
335 *
336 * In case that succeeds, it will acquire the s_umount
337 * lock of the old one. Since these are clearly distrinct
338 * locks, and this object isn't exposed yet, there's no
339 * risk of deadlocks.
340 *
341 * Annotate this by putting this lock in a different
342 * subclass.
343 */
344 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
345
346 if (security_sb_alloc(s))
347 goto fail;
348
349 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
350 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
351 sb_writers_name[i],
352 &type->s_writers_key[i]))
353 goto fail;
354 }
355 s->s_bdi = &noop_backing_dev_info;
356 s->s_flags = flags;
357 if (s->s_user_ns != &init_user_ns)
358 s->s_iflags |= SB_I_NODEV;
359 INIT_HLIST_NODE(&s->s_instances);
360 INIT_HLIST_BL_HEAD(&s->s_roots);
361 mutex_init(&s->s_sync_lock);
362 INIT_LIST_HEAD(&s->s_inodes);
363 spin_lock_init(&s->s_inode_list_lock);
364 INIT_LIST_HEAD(&s->s_inodes_wb);
365 spin_lock_init(&s->s_inode_wblist_lock);
366
367 s->s_count = 1;
368 atomic_set(&s->s_active, 1);
369 mutex_init(&s->s_vfs_rename_mutex);
370 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
371 init_rwsem(&s->s_dquot.dqio_sem);
372 s->s_maxbytes = MAX_NON_LFS;
373 s->s_op = &default_op;
374 s->s_time_gran = 1000000000;
375 s->s_time_min = TIME64_MIN;
376 s->s_time_max = TIME64_MAX;
377
378 s->s_shrink = shrinker_alloc(SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE,
379 "sb-%s", type->name);
380 if (!s->s_shrink)
381 goto fail;
382
383 s->s_shrink->scan_objects = super_cache_scan;
384 s->s_shrink->count_objects = super_cache_count;
385 s->s_shrink->batch = 1024;
386 s->s_shrink->private_data = s;
387
388 if (list_lru_init_memcg(&s->s_dentry_lru, s->s_shrink))
389 goto fail;
390 if (list_lru_init_memcg(&s->s_inode_lru, s->s_shrink))
391 goto fail;
392 return s;
393
394fail:
395 destroy_unused_super(s);
396 return NULL;
397}
398
399/* Superblock refcounting */
400
401/*
402 * Drop a superblock's refcount. The caller must hold sb_lock.
403 */
404static void __put_super(struct super_block *s)
405{
406 if (!--s->s_count) {
407 list_del_init(&s->s_list);
408 WARN_ON(s->s_dentry_lru.node);
409 WARN_ON(s->s_inode_lru.node);
410 WARN_ON(!list_empty(&s->s_mounts));
411 call_rcu(&s->rcu, destroy_super_rcu);
412 }
413}
414
415/**
416 * put_super - drop a temporary reference to superblock
417 * @sb: superblock in question
418 *
419 * Drops a temporary reference, frees superblock if there's no
420 * references left.
421 */
422void put_super(struct super_block *sb)
423{
424 spin_lock(&sb_lock);
425 __put_super(sb);
426 spin_unlock(&sb_lock);
427}
428
429static void kill_super_notify(struct super_block *sb)
430{
431 lockdep_assert_not_held(&sb->s_umount);
432
433 /* already notified earlier */
434 if (sb->s_flags & SB_DEAD)
435 return;
436
437 /*
438 * Remove it from @fs_supers so it isn't found by new
439 * sget{_fc}() walkers anymore. Any concurrent mounter still
440 * managing to grab a temporary reference is guaranteed to
441 * already see SB_DYING and will wait until we notify them about
442 * SB_DEAD.
443 */
444 spin_lock(&sb_lock);
445 hlist_del_init(&sb->s_instances);
446 spin_unlock(&sb_lock);
447
448 /*
449 * Let concurrent mounts know that this thing is really dead.
450 * We don't need @sb->s_umount here as every concurrent caller
451 * will see SB_DYING and either discard the superblock or wait
452 * for SB_DEAD.
453 */
454 super_wake(sb, SB_DEAD);
455}
456
457/**
458 * deactivate_locked_super - drop an active reference to superblock
459 * @s: superblock to deactivate
460 *
461 * Drops an active reference to superblock, converting it into a temporary
462 * one if there is no other active references left. In that case we
463 * tell fs driver to shut it down and drop the temporary reference we
464 * had just acquired.
465 *
466 * Caller holds exclusive lock on superblock; that lock is released.
467 */
468void deactivate_locked_super(struct super_block *s)
469{
470 struct file_system_type *fs = s->s_type;
471 if (atomic_dec_and_test(&s->s_active)) {
472 shrinker_free(s->s_shrink);
473 fs->kill_sb(s);
474
475 kill_super_notify(s);
476
477 /*
478 * Since list_lru_destroy() may sleep, we cannot call it from
479 * put_super(), where we hold the sb_lock. Therefore we destroy
480 * the lru lists right now.
481 */
482 list_lru_destroy(&s->s_dentry_lru);
483 list_lru_destroy(&s->s_inode_lru);
484
485 put_filesystem(fs);
486 put_super(s);
487 } else {
488 super_unlock_excl(s);
489 }
490}
491
492EXPORT_SYMBOL(deactivate_locked_super);
493
494/**
495 * deactivate_super - drop an active reference to superblock
496 * @s: superblock to deactivate
497 *
498 * Variant of deactivate_locked_super(), except that superblock is *not*
499 * locked by caller. If we are going to drop the final active reference,
500 * lock will be acquired prior to that.
501 */
502void deactivate_super(struct super_block *s)
503{
504 if (!atomic_add_unless(&s->s_active, -1, 1)) {
505 __super_lock_excl(s);
506 deactivate_locked_super(s);
507 }
508}
509
510EXPORT_SYMBOL(deactivate_super);
511
512/**
513 * grab_super - acquire an active reference to a superblock
514 * @sb: superblock to acquire
515 *
516 * Acquire a temporary reference on a superblock and try to trade it for
517 * an active reference. This is used in sget{_fc}() to wait for a
518 * superblock to either become SB_BORN or for it to pass through
519 * sb->kill() and be marked as SB_DEAD.
520 *
521 * Return: This returns true if an active reference could be acquired,
522 * false if not.
523 */
524static bool grab_super(struct super_block *sb)
525{
526 bool locked;
527
528 sb->s_count++;
529 spin_unlock(&sb_lock);
530 locked = super_lock_excl(sb);
531 if (locked) {
532 if (atomic_inc_not_zero(&sb->s_active)) {
533 put_super(sb);
534 return true;
535 }
536 super_unlock_excl(sb);
537 }
538 wait_var_event(&sb->s_flags, super_flags(sb, SB_DEAD));
539 put_super(sb);
540 return false;
541}
542
543/*
544 * super_trylock_shared - try to grab ->s_umount shared
545 * @sb: reference we are trying to grab
546 *
547 * Try to prevent fs shutdown. This is used in places where we
548 * cannot take an active reference but we need to ensure that the
549 * filesystem is not shut down while we are working on it. It returns
550 * false if we cannot acquire s_umount or if we lose the race and
551 * filesystem already got into shutdown, and returns true with the s_umount
552 * lock held in read mode in case of success. On successful return,
553 * the caller must drop the s_umount lock when done.
554 *
555 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
556 * The reason why it's safe is that we are OK with doing trylock instead
557 * of down_read(). There's a couple of places that are OK with that, but
558 * it's very much not a general-purpose interface.
559 */
560bool super_trylock_shared(struct super_block *sb)
561{
562 if (down_read_trylock(&sb->s_umount)) {
563 if (!(sb->s_flags & SB_DYING) && sb->s_root &&
564 (sb->s_flags & SB_BORN))
565 return true;
566 super_unlock_shared(sb);
567 }
568
569 return false;
570}
571
572/**
573 * retire_super - prevents superblock from being reused
574 * @sb: superblock to retire
575 *
576 * The function marks superblock to be ignored in superblock test, which
577 * prevents it from being reused for any new mounts. If the superblock has
578 * a private bdi, it also unregisters it, but doesn't reduce the refcount
579 * of the superblock to prevent potential races. The refcount is reduced
580 * by generic_shutdown_super(). The function can not be called
581 * concurrently with generic_shutdown_super(). It is safe to call the
582 * function multiple times, subsequent calls have no effect.
583 *
584 * The marker will affect the re-use only for block-device-based
585 * superblocks. Other superblocks will still get marked if this function
586 * is used, but that will not affect their reusability.
587 */
588void retire_super(struct super_block *sb)
589{
590 WARN_ON(!sb->s_bdev);
591 __super_lock_excl(sb);
592 if (sb->s_iflags & SB_I_PERSB_BDI) {
593 bdi_unregister(sb->s_bdi);
594 sb->s_iflags &= ~SB_I_PERSB_BDI;
595 }
596 sb->s_iflags |= SB_I_RETIRED;
597 super_unlock_excl(sb);
598}
599EXPORT_SYMBOL(retire_super);
600
601/**
602 * generic_shutdown_super - common helper for ->kill_sb()
603 * @sb: superblock to kill
604 *
605 * generic_shutdown_super() does all fs-independent work on superblock
606 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
607 * that need destruction out of superblock, call generic_shutdown_super()
608 * and release aforementioned objects. Note: dentries and inodes _are_
609 * taken care of and do not need specific handling.
610 *
611 * Upon calling this function, the filesystem may no longer alter or
612 * rearrange the set of dentries belonging to this super_block, nor may it
613 * change the attachments of dentries to inodes.
614 */
615void generic_shutdown_super(struct super_block *sb)
616{
617 const struct super_operations *sop = sb->s_op;
618
619 if (sb->s_root) {
620 shrink_dcache_for_umount(sb);
621 sync_filesystem(sb);
622 sb->s_flags &= ~SB_ACTIVE;
623
624 cgroup_writeback_umount(sb);
625
626 /* Evict all inodes with zero refcount. */
627 evict_inodes(sb);
628
629 /*
630 * Clean up and evict any inodes that still have references due
631 * to fsnotify or the security policy.
632 */
633 fsnotify_sb_delete(sb);
634 security_sb_delete(sb);
635
636 if (sb->s_dio_done_wq) {
637 destroy_workqueue(sb->s_dio_done_wq);
638 sb->s_dio_done_wq = NULL;
639 }
640
641 if (sop->put_super)
642 sop->put_super(sb);
643
644 /*
645 * Now that all potentially-encrypted inodes have been evicted,
646 * the fscrypt keyring can be destroyed.
647 */
648 fscrypt_destroy_keyring(sb);
649
650 if (CHECK_DATA_CORRUPTION(!list_empty(&sb->s_inodes),
651 "VFS: Busy inodes after unmount of %s (%s)",
652 sb->s_id, sb->s_type->name)) {
653 /*
654 * Adding a proper bailout path here would be hard, but
655 * we can at least make it more likely that a later
656 * iput_final() or such crashes cleanly.
657 */
658 struct inode *inode;
659
660 spin_lock(&sb->s_inode_list_lock);
661 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
662 inode->i_op = VFS_PTR_POISON;
663 inode->i_sb = VFS_PTR_POISON;
664 inode->i_mapping = VFS_PTR_POISON;
665 }
666 spin_unlock(&sb->s_inode_list_lock);
667 }
668 }
669 /*
670 * Broadcast to everyone that grabbed a temporary reference to this
671 * superblock before we removed it from @fs_supers that the superblock
672 * is dying. Every walker of @fs_supers outside of sget{_fc}() will now
673 * discard this superblock and treat it as dead.
674 *
675 * We leave the superblock on @fs_supers so it can be found by
676 * sget{_fc}() until we passed sb->kill_sb().
677 */
678 super_wake(sb, SB_DYING);
679 super_unlock_excl(sb);
680 if (sb->s_bdi != &noop_backing_dev_info) {
681 if (sb->s_iflags & SB_I_PERSB_BDI)
682 bdi_unregister(sb->s_bdi);
683 bdi_put(sb->s_bdi);
684 sb->s_bdi = &noop_backing_dev_info;
685 }
686}
687
688EXPORT_SYMBOL(generic_shutdown_super);
689
690bool mount_capable(struct fs_context *fc)
691{
692 if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT))
693 return capable(CAP_SYS_ADMIN);
694 else
695 return ns_capable(fc->user_ns, CAP_SYS_ADMIN);
696}
697
698/**
699 * sget_fc - Find or create a superblock
700 * @fc: Filesystem context.
701 * @test: Comparison callback
702 * @set: Setup callback
703 *
704 * Create a new superblock or find an existing one.
705 *
706 * The @test callback is used to find a matching existing superblock.
707 * Whether or not the requested parameters in @fc are taken into account
708 * is specific to the @test callback that is used. They may even be
709 * completely ignored.
710 *
711 * If an extant superblock is matched, it will be returned unless:
712 *
713 * (1) the namespace the filesystem context @fc and the extant
714 * superblock's namespace differ
715 *
716 * (2) the filesystem context @fc has requested that reusing an extant
717 * superblock is not allowed
718 *
719 * In both cases EBUSY will be returned.
720 *
721 * If no match is made, a new superblock will be allocated and basic
722 * initialisation will be performed (s_type, s_fs_info and s_id will be
723 * set and the @set callback will be invoked), the superblock will be
724 * published and it will be returned in a partially constructed state
725 * with SB_BORN and SB_ACTIVE as yet unset.
726 *
727 * Return: On success, an extant or newly created superblock is
728 * returned. On failure an error pointer is returned.
729 */
730struct super_block *sget_fc(struct fs_context *fc,
731 int (*test)(struct super_block *, struct fs_context *),
732 int (*set)(struct super_block *, struct fs_context *))
733{
734 struct super_block *s = NULL;
735 struct super_block *old;
736 struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
737 int err;
738
739 /*
740 * Never allow s_user_ns != &init_user_ns when FS_USERNS_MOUNT is
741 * not set, as the filesystem is likely unprepared to handle it.
742 * This can happen when fsconfig() is called from init_user_ns with
743 * an fs_fd opened in another user namespace.
744 */
745 if (user_ns != &init_user_ns && !(fc->fs_type->fs_flags & FS_USERNS_MOUNT)) {
746 errorfc(fc, "VFS: Mounting from non-initial user namespace is not allowed");
747 return ERR_PTR(-EPERM);
748 }
749
750retry:
751 spin_lock(&sb_lock);
752 if (test) {
753 hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
754 if (test(old, fc))
755 goto share_extant_sb;
756 }
757 }
758 if (!s) {
759 spin_unlock(&sb_lock);
760 s = alloc_super(fc->fs_type, fc->sb_flags, user_ns);
761 if (!s)
762 return ERR_PTR(-ENOMEM);
763 goto retry;
764 }
765
766 s->s_fs_info = fc->s_fs_info;
767 err = set(s, fc);
768 if (err) {
769 s->s_fs_info = NULL;
770 spin_unlock(&sb_lock);
771 destroy_unused_super(s);
772 return ERR_PTR(err);
773 }
774 fc->s_fs_info = NULL;
775 s->s_type = fc->fs_type;
776 s->s_iflags |= fc->s_iflags;
777 strscpy(s->s_id, s->s_type->name, sizeof(s->s_id));
778 /*
779 * Make the superblock visible on @super_blocks and @fs_supers.
780 * It's in a nascent state and users should wait on SB_BORN or
781 * SB_DYING to be set.
782 */
783 list_add_tail(&s->s_list, &super_blocks);
784 hlist_add_head(&s->s_instances, &s->s_type->fs_supers);
785 spin_unlock(&sb_lock);
786 get_filesystem(s->s_type);
787 shrinker_register(s->s_shrink);
788 return s;
789
790share_extant_sb:
791 if (user_ns != old->s_user_ns || fc->exclusive) {
792 spin_unlock(&sb_lock);
793 destroy_unused_super(s);
794 if (fc->exclusive)
795 warnfc(fc, "reusing existing filesystem not allowed");
796 else
797 warnfc(fc, "reusing existing filesystem in another namespace not allowed");
798 return ERR_PTR(-EBUSY);
799 }
800 if (!grab_super(old))
801 goto retry;
802 destroy_unused_super(s);
803 return old;
804}
805EXPORT_SYMBOL(sget_fc);
806
807/**
808 * sget - find or create a superblock
809 * @type: filesystem type superblock should belong to
810 * @test: comparison callback
811 * @set: setup callback
812 * @flags: mount flags
813 * @data: argument to each of them
814 */
815struct super_block *sget(struct file_system_type *type,
816 int (*test)(struct super_block *,void *),
817 int (*set)(struct super_block *,void *),
818 int flags,
819 void *data)
820{
821 struct user_namespace *user_ns = current_user_ns();
822 struct super_block *s = NULL;
823 struct super_block *old;
824 int err;
825
826 /* We don't yet pass the user namespace of the parent
827 * mount through to here so always use &init_user_ns
828 * until that changes.
829 */
830 if (flags & SB_SUBMOUNT)
831 user_ns = &init_user_ns;
832
833retry:
834 spin_lock(&sb_lock);
835 if (test) {
836 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
837 if (!test(old, data))
838 continue;
839 if (user_ns != old->s_user_ns) {
840 spin_unlock(&sb_lock);
841 destroy_unused_super(s);
842 return ERR_PTR(-EBUSY);
843 }
844 if (!grab_super(old))
845 goto retry;
846 destroy_unused_super(s);
847 return old;
848 }
849 }
850 if (!s) {
851 spin_unlock(&sb_lock);
852 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
853 if (!s)
854 return ERR_PTR(-ENOMEM);
855 goto retry;
856 }
857
858 err = set(s, data);
859 if (err) {
860 spin_unlock(&sb_lock);
861 destroy_unused_super(s);
862 return ERR_PTR(err);
863 }
864 s->s_type = type;
865 strscpy(s->s_id, type->name, sizeof(s->s_id));
866 list_add_tail(&s->s_list, &super_blocks);
867 hlist_add_head(&s->s_instances, &type->fs_supers);
868 spin_unlock(&sb_lock);
869 get_filesystem(type);
870 shrinker_register(s->s_shrink);
871 return s;
872}
873EXPORT_SYMBOL(sget);
874
875void drop_super(struct super_block *sb)
876{
877 super_unlock_shared(sb);
878 put_super(sb);
879}
880
881EXPORT_SYMBOL(drop_super);
882
883void drop_super_exclusive(struct super_block *sb)
884{
885 super_unlock_excl(sb);
886 put_super(sb);
887}
888EXPORT_SYMBOL(drop_super_exclusive);
889
890static void __iterate_supers(void (*f)(struct super_block *))
891{
892 struct super_block *sb, *p = NULL;
893
894 spin_lock(&sb_lock);
895 list_for_each_entry(sb, &super_blocks, s_list) {
896 if (super_flags(sb, SB_DYING))
897 continue;
898 sb->s_count++;
899 spin_unlock(&sb_lock);
900
901 f(sb);
902
903 spin_lock(&sb_lock);
904 if (p)
905 __put_super(p);
906 p = sb;
907 }
908 if (p)
909 __put_super(p);
910 spin_unlock(&sb_lock);
911}
912/**
913 * iterate_supers - call function for all active superblocks
914 * @f: function to call
915 * @arg: argument to pass to it
916 *
917 * Scans the superblock list and calls given function, passing it
918 * locked superblock and given argument.
919 */
920void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
921{
922 struct super_block *sb, *p = NULL;
923
924 spin_lock(&sb_lock);
925 list_for_each_entry(sb, &super_blocks, s_list) {
926 bool locked;
927
928 sb->s_count++;
929 spin_unlock(&sb_lock);
930
931 locked = super_lock_shared(sb);
932 if (locked) {
933 if (sb->s_root)
934 f(sb, arg);
935 super_unlock_shared(sb);
936 }
937
938 spin_lock(&sb_lock);
939 if (p)
940 __put_super(p);
941 p = sb;
942 }
943 if (p)
944 __put_super(p);
945 spin_unlock(&sb_lock);
946}
947
948/**
949 * iterate_supers_type - call function for superblocks of given type
950 * @type: fs type
951 * @f: function to call
952 * @arg: argument to pass to it
953 *
954 * Scans the superblock list and calls given function, passing it
955 * locked superblock and given argument.
956 */
957void iterate_supers_type(struct file_system_type *type,
958 void (*f)(struct super_block *, void *), void *arg)
959{
960 struct super_block *sb, *p = NULL;
961
962 spin_lock(&sb_lock);
963 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
964 bool locked;
965
966 sb->s_count++;
967 spin_unlock(&sb_lock);
968
969 locked = super_lock_shared(sb);
970 if (locked) {
971 if (sb->s_root)
972 f(sb, arg);
973 super_unlock_shared(sb);
974 }
975
976 spin_lock(&sb_lock);
977 if (p)
978 __put_super(p);
979 p = sb;
980 }
981 if (p)
982 __put_super(p);
983 spin_unlock(&sb_lock);
984}
985
986EXPORT_SYMBOL(iterate_supers_type);
987
988struct super_block *user_get_super(dev_t dev, bool excl)
989{
990 struct super_block *sb;
991
992 spin_lock(&sb_lock);
993 list_for_each_entry(sb, &super_blocks, s_list) {
994 if (sb->s_dev == dev) {
995 bool locked;
996
997 sb->s_count++;
998 spin_unlock(&sb_lock);
999 /* still alive? */
1000 locked = super_lock(sb, excl);
1001 if (locked) {
1002 if (sb->s_root)
1003 return sb;
1004 super_unlock(sb, excl);
1005 }
1006 /* nope, got unmounted */
1007 spin_lock(&sb_lock);
1008 __put_super(sb);
1009 break;
1010 }
1011 }
1012 spin_unlock(&sb_lock);
1013 return NULL;
1014}
1015
1016/**
1017 * reconfigure_super - asks filesystem to change superblock parameters
1018 * @fc: The superblock and configuration
1019 *
1020 * Alters the configuration parameters of a live superblock.
1021 */
1022int reconfigure_super(struct fs_context *fc)
1023{
1024 struct super_block *sb = fc->root->d_sb;
1025 int retval;
1026 bool remount_ro = false;
1027 bool remount_rw = false;
1028 bool force = fc->sb_flags & SB_FORCE;
1029
1030 if (fc->sb_flags_mask & ~MS_RMT_MASK)
1031 return -EINVAL;
1032 if (sb->s_writers.frozen != SB_UNFROZEN)
1033 return -EBUSY;
1034
1035 retval = security_sb_remount(sb, fc->security);
1036 if (retval)
1037 return retval;
1038
1039 if (fc->sb_flags_mask & SB_RDONLY) {
1040#ifdef CONFIG_BLOCK
1041 if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev &&
1042 bdev_read_only(sb->s_bdev))
1043 return -EACCES;
1044#endif
1045 remount_rw = !(fc->sb_flags & SB_RDONLY) && sb_rdonly(sb);
1046 remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
1047 }
1048
1049 if (remount_ro) {
1050 if (!hlist_empty(&sb->s_pins)) {
1051 super_unlock_excl(sb);
1052 group_pin_kill(&sb->s_pins);
1053 __super_lock_excl(sb);
1054 if (!sb->s_root)
1055 return 0;
1056 if (sb->s_writers.frozen != SB_UNFROZEN)
1057 return -EBUSY;
1058 remount_ro = !sb_rdonly(sb);
1059 }
1060 }
1061 shrink_dcache_sb(sb);
1062
1063 /* If we are reconfiguring to RDONLY and current sb is read/write,
1064 * make sure there are no files open for writing.
1065 */
1066 if (remount_ro) {
1067 if (force) {
1068 sb_start_ro_state_change(sb);
1069 } else {
1070 retval = sb_prepare_remount_readonly(sb);
1071 if (retval)
1072 return retval;
1073 }
1074 } else if (remount_rw) {
1075 /*
1076 * Protect filesystem's reconfigure code from writes from
1077 * userspace until reconfigure finishes.
1078 */
1079 sb_start_ro_state_change(sb);
1080 }
1081
1082 if (fc->ops->reconfigure) {
1083 retval = fc->ops->reconfigure(fc);
1084 if (retval) {
1085 if (!force)
1086 goto cancel_readonly;
1087 /* If forced remount, go ahead despite any errors */
1088 WARN(1, "forced remount of a %s fs returned %i\n",
1089 sb->s_type->name, retval);
1090 }
1091 }
1092
1093 WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
1094 (fc->sb_flags & fc->sb_flags_mask)));
1095 sb_end_ro_state_change(sb);
1096
1097 /*
1098 * Some filesystems modify their metadata via some other path than the
1099 * bdev buffer cache (eg. use a private mapping, or directories in
1100 * pagecache, etc). Also file data modifications go via their own
1101 * mappings. So If we try to mount readonly then copy the filesystem
1102 * from bdev, we could get stale data, so invalidate it to give a best
1103 * effort at coherency.
1104 */
1105 if (remount_ro && sb->s_bdev)
1106 invalidate_bdev(sb->s_bdev);
1107 return 0;
1108
1109cancel_readonly:
1110 sb_end_ro_state_change(sb);
1111 return retval;
1112}
1113
1114static void do_emergency_remount_callback(struct super_block *sb)
1115{
1116 bool locked = super_lock_excl(sb);
1117
1118 if (locked && sb->s_root && sb->s_bdev && !sb_rdonly(sb)) {
1119 struct fs_context *fc;
1120
1121 fc = fs_context_for_reconfigure(sb->s_root,
1122 SB_RDONLY | SB_FORCE, SB_RDONLY);
1123 if (!IS_ERR(fc)) {
1124 if (parse_monolithic_mount_data(fc, NULL) == 0)
1125 (void)reconfigure_super(fc);
1126 put_fs_context(fc);
1127 }
1128 }
1129 if (locked)
1130 super_unlock_excl(sb);
1131}
1132
1133static void do_emergency_remount(struct work_struct *work)
1134{
1135 __iterate_supers(do_emergency_remount_callback);
1136 kfree(work);
1137 printk("Emergency Remount complete\n");
1138}
1139
1140void emergency_remount(void)
1141{
1142 struct work_struct *work;
1143
1144 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1145 if (work) {
1146 INIT_WORK(work, do_emergency_remount);
1147 schedule_work(work);
1148 }
1149}
1150
1151static void do_thaw_all_callback(struct super_block *sb)
1152{
1153 bool locked = super_lock_excl(sb);
1154
1155 if (locked && sb->s_root) {
1156 if (IS_ENABLED(CONFIG_BLOCK))
1157 while (sb->s_bdev && !bdev_thaw(sb->s_bdev))
1158 pr_warn("Emergency Thaw on %pg\n", sb->s_bdev);
1159 thaw_super_locked(sb, FREEZE_HOLDER_USERSPACE);
1160 return;
1161 }
1162 if (locked)
1163 super_unlock_excl(sb);
1164}
1165
1166static void do_thaw_all(struct work_struct *work)
1167{
1168 __iterate_supers(do_thaw_all_callback);
1169 kfree(work);
1170 printk(KERN_WARNING "Emergency Thaw complete\n");
1171}
1172
1173/**
1174 * emergency_thaw_all -- forcibly thaw every frozen filesystem
1175 *
1176 * Used for emergency unfreeze of all filesystems via SysRq
1177 */
1178void emergency_thaw_all(void)
1179{
1180 struct work_struct *work;
1181
1182 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1183 if (work) {
1184 INIT_WORK(work, do_thaw_all);
1185 schedule_work(work);
1186 }
1187}
1188
1189static DEFINE_IDA(unnamed_dev_ida);
1190
1191/**
1192 * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1193 * @p: Pointer to a dev_t.
1194 *
1195 * Filesystems which don't use real block devices can call this function
1196 * to allocate a virtual block device.
1197 *
1198 * Context: Any context. Frequently called while holding sb_lock.
1199 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1200 * or -ENOMEM if memory allocation failed.
1201 */
1202int get_anon_bdev(dev_t *p)
1203{
1204 int dev;
1205
1206 /*
1207 * Many userspace utilities consider an FSID of 0 invalid.
1208 * Always return at least 1 from get_anon_bdev.
1209 */
1210 dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1211 GFP_ATOMIC);
1212 if (dev == -ENOSPC)
1213 dev = -EMFILE;
1214 if (dev < 0)
1215 return dev;
1216
1217 *p = MKDEV(0, dev);
1218 return 0;
1219}
1220EXPORT_SYMBOL(get_anon_bdev);
1221
1222void free_anon_bdev(dev_t dev)
1223{
1224 ida_free(&unnamed_dev_ida, MINOR(dev));
1225}
1226EXPORT_SYMBOL(free_anon_bdev);
1227
1228int set_anon_super(struct super_block *s, void *data)
1229{
1230 return get_anon_bdev(&s->s_dev);
1231}
1232EXPORT_SYMBOL(set_anon_super);
1233
1234void kill_anon_super(struct super_block *sb)
1235{
1236 dev_t dev = sb->s_dev;
1237 generic_shutdown_super(sb);
1238 kill_super_notify(sb);
1239 free_anon_bdev(dev);
1240}
1241EXPORT_SYMBOL(kill_anon_super);
1242
1243void kill_litter_super(struct super_block *sb)
1244{
1245 if (sb->s_root)
1246 d_genocide(sb->s_root);
1247 kill_anon_super(sb);
1248}
1249EXPORT_SYMBOL(kill_litter_super);
1250
1251int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1252{
1253 return set_anon_super(sb, NULL);
1254}
1255EXPORT_SYMBOL(set_anon_super_fc);
1256
1257static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1258{
1259 return sb->s_fs_info == fc->s_fs_info;
1260}
1261
1262static int test_single_super(struct super_block *s, struct fs_context *fc)
1263{
1264 return 1;
1265}
1266
1267static int vfs_get_super(struct fs_context *fc,
1268 int (*test)(struct super_block *, struct fs_context *),
1269 int (*fill_super)(struct super_block *sb,
1270 struct fs_context *fc))
1271{
1272 struct super_block *sb;
1273 int err;
1274
1275 sb = sget_fc(fc, test, set_anon_super_fc);
1276 if (IS_ERR(sb))
1277 return PTR_ERR(sb);
1278
1279 if (!sb->s_root) {
1280 err = fill_super(sb, fc);
1281 if (err)
1282 goto error;
1283
1284 sb->s_flags |= SB_ACTIVE;
1285 }
1286
1287 fc->root = dget(sb->s_root);
1288 return 0;
1289
1290error:
1291 deactivate_locked_super(sb);
1292 return err;
1293}
1294
1295int get_tree_nodev(struct fs_context *fc,
1296 int (*fill_super)(struct super_block *sb,
1297 struct fs_context *fc))
1298{
1299 return vfs_get_super(fc, NULL, fill_super);
1300}
1301EXPORT_SYMBOL(get_tree_nodev);
1302
1303int get_tree_single(struct fs_context *fc,
1304 int (*fill_super)(struct super_block *sb,
1305 struct fs_context *fc))
1306{
1307 return vfs_get_super(fc, test_single_super, fill_super);
1308}
1309EXPORT_SYMBOL(get_tree_single);
1310
1311int get_tree_keyed(struct fs_context *fc,
1312 int (*fill_super)(struct super_block *sb,
1313 struct fs_context *fc),
1314 void *key)
1315{
1316 fc->s_fs_info = key;
1317 return vfs_get_super(fc, test_keyed_super, fill_super);
1318}
1319EXPORT_SYMBOL(get_tree_keyed);
1320
1321static int set_bdev_super(struct super_block *s, void *data)
1322{
1323 s->s_dev = *(dev_t *)data;
1324 return 0;
1325}
1326
1327static int super_s_dev_set(struct super_block *s, struct fs_context *fc)
1328{
1329 return set_bdev_super(s, fc->sget_key);
1330}
1331
1332static int super_s_dev_test(struct super_block *s, struct fs_context *fc)
1333{
1334 return !(s->s_iflags & SB_I_RETIRED) &&
1335 s->s_dev == *(dev_t *)fc->sget_key;
1336}
1337
1338/**
1339 * sget_dev - Find or create a superblock by device number
1340 * @fc: Filesystem context.
1341 * @dev: device number
1342 *
1343 * Find or create a superblock using the provided device number that
1344 * will be stored in fc->sget_key.
1345 *
1346 * If an extant superblock is matched, then that will be returned with
1347 * an elevated reference count that the caller must transfer or discard.
1348 *
1349 * If no match is made, a new superblock will be allocated and basic
1350 * initialisation will be performed (s_type, s_fs_info, s_id, s_dev will
1351 * be set). The superblock will be published and it will be returned in
1352 * a partially constructed state with SB_BORN and SB_ACTIVE as yet
1353 * unset.
1354 *
1355 * Return: an existing or newly created superblock on success, an error
1356 * pointer on failure.
1357 */
1358struct super_block *sget_dev(struct fs_context *fc, dev_t dev)
1359{
1360 fc->sget_key = &dev;
1361 return sget_fc(fc, super_s_dev_test, super_s_dev_set);
1362}
1363EXPORT_SYMBOL(sget_dev);
1364
1365#ifdef CONFIG_BLOCK
1366/*
1367 * Lock the superblock that is holder of the bdev. Returns the superblock
1368 * pointer if we successfully locked the superblock and it is alive. Otherwise
1369 * we return NULL and just unlock bdev->bd_holder_lock.
1370 *
1371 * The function must be called with bdev->bd_holder_lock and releases it.
1372 */
1373static struct super_block *bdev_super_lock(struct block_device *bdev, bool excl)
1374 __releases(&bdev->bd_holder_lock)
1375{
1376 struct super_block *sb = bdev->bd_holder;
1377 bool locked;
1378
1379 lockdep_assert_held(&bdev->bd_holder_lock);
1380 lockdep_assert_not_held(&sb->s_umount);
1381 lockdep_assert_not_held(&bdev->bd_disk->open_mutex);
1382
1383 /* Make sure sb doesn't go away from under us */
1384 spin_lock(&sb_lock);
1385 sb->s_count++;
1386 spin_unlock(&sb_lock);
1387
1388 mutex_unlock(&bdev->bd_holder_lock);
1389
1390 locked = super_lock(sb, excl);
1391
1392 /*
1393 * If the superblock wasn't already SB_DYING then we hold
1394 * s_umount and can safely drop our temporary reference.
1395 */
1396 put_super(sb);
1397
1398 if (!locked)
1399 return NULL;
1400
1401 if (!sb->s_root || !(sb->s_flags & SB_ACTIVE)) {
1402 super_unlock(sb, excl);
1403 return NULL;
1404 }
1405
1406 return sb;
1407}
1408
1409static void fs_bdev_mark_dead(struct block_device *bdev, bool surprise)
1410{
1411 struct super_block *sb;
1412
1413 sb = bdev_super_lock(bdev, false);
1414 if (!sb)
1415 return;
1416
1417 if (!surprise)
1418 sync_filesystem(sb);
1419 shrink_dcache_sb(sb);
1420 invalidate_inodes(sb);
1421 if (sb->s_op->shutdown)
1422 sb->s_op->shutdown(sb);
1423
1424 super_unlock_shared(sb);
1425}
1426
1427static void fs_bdev_sync(struct block_device *bdev)
1428{
1429 struct super_block *sb;
1430
1431 sb = bdev_super_lock(bdev, false);
1432 if (!sb)
1433 return;
1434
1435 sync_filesystem(sb);
1436 super_unlock_shared(sb);
1437}
1438
1439static struct super_block *get_bdev_super(struct block_device *bdev)
1440{
1441 bool active = false;
1442 struct super_block *sb;
1443
1444 sb = bdev_super_lock(bdev, true);
1445 if (sb) {
1446 active = atomic_inc_not_zero(&sb->s_active);
1447 super_unlock_excl(sb);
1448 }
1449 if (!active)
1450 return NULL;
1451 return sb;
1452}
1453
1454/**
1455 * fs_bdev_freeze - freeze owning filesystem of block device
1456 * @bdev: block device
1457 *
1458 * Freeze the filesystem that owns this block device if it is still
1459 * active.
1460 *
1461 * A filesystem that owns multiple block devices may be frozen from each
1462 * block device and won't be unfrozen until all block devices are
1463 * unfrozen. Each block device can only freeze the filesystem once as we
1464 * nest freezes for block devices in the block layer.
1465 *
1466 * Return: If the freeze was successful zero is returned. If the freeze
1467 * failed a negative error code is returned.
1468 */
1469static int fs_bdev_freeze(struct block_device *bdev)
1470{
1471 struct super_block *sb;
1472 int error = 0;
1473
1474 lockdep_assert_held(&bdev->bd_fsfreeze_mutex);
1475
1476 sb = get_bdev_super(bdev);
1477 if (!sb)
1478 return -EINVAL;
1479
1480 if (sb->s_op->freeze_super)
1481 error = sb->s_op->freeze_super(sb,
1482 FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE);
1483 else
1484 error = freeze_super(sb,
1485 FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE);
1486 if (!error)
1487 error = sync_blockdev(bdev);
1488 deactivate_super(sb);
1489 return error;
1490}
1491
1492/**
1493 * fs_bdev_thaw - thaw owning filesystem of block device
1494 * @bdev: block device
1495 *
1496 * Thaw the filesystem that owns this block device.
1497 *
1498 * A filesystem that owns multiple block devices may be frozen from each
1499 * block device and won't be unfrozen until all block devices are
1500 * unfrozen. Each block device can only freeze the filesystem once as we
1501 * nest freezes for block devices in the block layer.
1502 *
1503 * Return: If the thaw was successful zero is returned. If the thaw
1504 * failed a negative error code is returned. If this function
1505 * returns zero it doesn't mean that the filesystem is unfrozen
1506 * as it may have been frozen multiple times (kernel may hold a
1507 * freeze or might be frozen from other block devices).
1508 */
1509static int fs_bdev_thaw(struct block_device *bdev)
1510{
1511 struct super_block *sb;
1512 int error;
1513
1514 lockdep_assert_held(&bdev->bd_fsfreeze_mutex);
1515
1516 /*
1517 * The block device may have been frozen before it was claimed by a
1518 * filesystem. Concurrently another process might try to mount that
1519 * frozen block device and has temporarily claimed the block device for
1520 * that purpose causing a concurrent fs_bdev_thaw() to end up here. The
1521 * mounter is already about to abort mounting because they still saw an
1522 * elevanted bdev->bd_fsfreeze_count so get_bdev_super() will return
1523 * NULL in that case.
1524 */
1525 sb = get_bdev_super(bdev);
1526 if (!sb)
1527 return -EINVAL;
1528
1529 if (sb->s_op->thaw_super)
1530 error = sb->s_op->thaw_super(sb,
1531 FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE);
1532 else
1533 error = thaw_super(sb,
1534 FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE);
1535 deactivate_super(sb);
1536 return error;
1537}
1538
1539const struct blk_holder_ops fs_holder_ops = {
1540 .mark_dead = fs_bdev_mark_dead,
1541 .sync = fs_bdev_sync,
1542 .freeze = fs_bdev_freeze,
1543 .thaw = fs_bdev_thaw,
1544};
1545EXPORT_SYMBOL_GPL(fs_holder_ops);
1546
1547int setup_bdev_super(struct super_block *sb, int sb_flags,
1548 struct fs_context *fc)
1549{
1550 blk_mode_t mode = sb_open_mode(sb_flags);
1551 struct file *bdev_file;
1552 struct block_device *bdev;
1553
1554 bdev_file = bdev_file_open_by_dev(sb->s_dev, mode, sb, &fs_holder_ops);
1555 if (IS_ERR(bdev_file)) {
1556 if (fc)
1557 errorf(fc, "%s: Can't open blockdev", fc->source);
1558 return PTR_ERR(bdev_file);
1559 }
1560 bdev = file_bdev(bdev_file);
1561
1562 /*
1563 * This really should be in blkdev_get_by_dev, but right now can't due
1564 * to legacy issues that require us to allow opening a block device node
1565 * writable from userspace even for a read-only block device.
1566 */
1567 if ((mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) {
1568 bdev_fput(bdev_file);
1569 return -EACCES;
1570 }
1571
1572 /*
1573 * It is enough to check bdev was not frozen before we set
1574 * s_bdev as freezing will wait until SB_BORN is set.
1575 */
1576 if (atomic_read(&bdev->bd_fsfreeze_count) > 0) {
1577 if (fc)
1578 warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
1579 bdev_fput(bdev_file);
1580 return -EBUSY;
1581 }
1582 spin_lock(&sb_lock);
1583 sb->s_bdev_file = bdev_file;
1584 sb->s_bdev = bdev;
1585 sb->s_bdi = bdi_get(bdev->bd_disk->bdi);
1586 if (bdev_stable_writes(bdev))
1587 sb->s_iflags |= SB_I_STABLE_WRITES;
1588 spin_unlock(&sb_lock);
1589
1590 snprintf(sb->s_id, sizeof(sb->s_id), "%pg", bdev);
1591 shrinker_debugfs_rename(sb->s_shrink, "sb-%s:%s", sb->s_type->name,
1592 sb->s_id);
1593 sb_set_blocksize(sb, block_size(bdev));
1594 return 0;
1595}
1596EXPORT_SYMBOL_GPL(setup_bdev_super);
1597
1598/**
1599 * get_tree_bdev_flags - Get a superblock based on a single block device
1600 * @fc: The filesystem context holding the parameters
1601 * @fill_super: Helper to initialise a new superblock
1602 * @flags: GET_TREE_BDEV_* flags
1603 */
1604int get_tree_bdev_flags(struct fs_context *fc,
1605 int (*fill_super)(struct super_block *sb,
1606 struct fs_context *fc), unsigned int flags)
1607{
1608 struct super_block *s;
1609 int error = 0;
1610 dev_t dev;
1611
1612 if (!fc->source)
1613 return invalf(fc, "No source specified");
1614
1615 error = lookup_bdev(fc->source, &dev);
1616 if (error) {
1617 if (!(flags & GET_TREE_BDEV_QUIET_LOOKUP))
1618 errorf(fc, "%s: Can't lookup blockdev", fc->source);
1619 return error;
1620 }
1621 fc->sb_flags |= SB_NOSEC;
1622 s = sget_dev(fc, dev);
1623 if (IS_ERR(s))
1624 return PTR_ERR(s);
1625
1626 if (s->s_root) {
1627 /* Don't summarily change the RO/RW state. */
1628 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1629 warnf(fc, "%pg: Can't mount, would change RO state", s->s_bdev);
1630 deactivate_locked_super(s);
1631 return -EBUSY;
1632 }
1633 } else {
1634 error = setup_bdev_super(s, fc->sb_flags, fc);
1635 if (!error)
1636 error = fill_super(s, fc);
1637 if (error) {
1638 deactivate_locked_super(s);
1639 return error;
1640 }
1641 s->s_flags |= SB_ACTIVE;
1642 }
1643
1644 BUG_ON(fc->root);
1645 fc->root = dget(s->s_root);
1646 return 0;
1647}
1648EXPORT_SYMBOL_GPL(get_tree_bdev_flags);
1649
1650/**
1651 * get_tree_bdev - Get a superblock based on a single block device
1652 * @fc: The filesystem context holding the parameters
1653 * @fill_super: Helper to initialise a new superblock
1654 */
1655int get_tree_bdev(struct fs_context *fc,
1656 int (*fill_super)(struct super_block *,
1657 struct fs_context *))
1658{
1659 return get_tree_bdev_flags(fc, fill_super, 0);
1660}
1661EXPORT_SYMBOL(get_tree_bdev);
1662
1663static int test_bdev_super(struct super_block *s, void *data)
1664{
1665 return !(s->s_iflags & SB_I_RETIRED) && s->s_dev == *(dev_t *)data;
1666}
1667
1668struct dentry *mount_bdev(struct file_system_type *fs_type,
1669 int flags, const char *dev_name, void *data,
1670 int (*fill_super)(struct super_block *, void *, int))
1671{
1672 struct super_block *s;
1673 int error;
1674 dev_t dev;
1675
1676 error = lookup_bdev(dev_name, &dev);
1677 if (error)
1678 return ERR_PTR(error);
1679
1680 flags |= SB_NOSEC;
1681 s = sget(fs_type, test_bdev_super, set_bdev_super, flags, &dev);
1682 if (IS_ERR(s))
1683 return ERR_CAST(s);
1684
1685 if (s->s_root) {
1686 if ((flags ^ s->s_flags) & SB_RDONLY) {
1687 deactivate_locked_super(s);
1688 return ERR_PTR(-EBUSY);
1689 }
1690 } else {
1691 error = setup_bdev_super(s, flags, NULL);
1692 if (!error)
1693 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1694 if (error) {
1695 deactivate_locked_super(s);
1696 return ERR_PTR(error);
1697 }
1698
1699 s->s_flags |= SB_ACTIVE;
1700 }
1701
1702 return dget(s->s_root);
1703}
1704EXPORT_SYMBOL(mount_bdev);
1705
1706void kill_block_super(struct super_block *sb)
1707{
1708 struct block_device *bdev = sb->s_bdev;
1709
1710 generic_shutdown_super(sb);
1711 if (bdev) {
1712 sync_blockdev(bdev);
1713 bdev_fput(sb->s_bdev_file);
1714 }
1715}
1716
1717EXPORT_SYMBOL(kill_block_super);
1718#endif
1719
1720struct dentry *mount_nodev(struct file_system_type *fs_type,
1721 int flags, void *data,
1722 int (*fill_super)(struct super_block *, void *, int))
1723{
1724 int error;
1725 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1726
1727 if (IS_ERR(s))
1728 return ERR_CAST(s);
1729
1730 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1731 if (error) {
1732 deactivate_locked_super(s);
1733 return ERR_PTR(error);
1734 }
1735 s->s_flags |= SB_ACTIVE;
1736 return dget(s->s_root);
1737}
1738EXPORT_SYMBOL(mount_nodev);
1739
1740int reconfigure_single(struct super_block *s,
1741 int flags, void *data)
1742{
1743 struct fs_context *fc;
1744 int ret;
1745
1746 /* The caller really need to be passing fc down into mount_single(),
1747 * then a chunk of this can be removed. [Bollocks -- AV]
1748 * Better yet, reconfiguration shouldn't happen, but rather the second
1749 * mount should be rejected if the parameters are not compatible.
1750 */
1751 fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1752 if (IS_ERR(fc))
1753 return PTR_ERR(fc);
1754
1755 ret = parse_monolithic_mount_data(fc, data);
1756 if (ret < 0)
1757 goto out;
1758
1759 ret = reconfigure_super(fc);
1760out:
1761 put_fs_context(fc);
1762 return ret;
1763}
1764
1765static int compare_single(struct super_block *s, void *p)
1766{
1767 return 1;
1768}
1769
1770struct dentry *mount_single(struct file_system_type *fs_type,
1771 int flags, void *data,
1772 int (*fill_super)(struct super_block *, void *, int))
1773{
1774 struct super_block *s;
1775 int error;
1776
1777 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1778 if (IS_ERR(s))
1779 return ERR_CAST(s);
1780 if (!s->s_root) {
1781 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1782 if (!error)
1783 s->s_flags |= SB_ACTIVE;
1784 } else {
1785 error = reconfigure_single(s, flags, data);
1786 }
1787 if (unlikely(error)) {
1788 deactivate_locked_super(s);
1789 return ERR_PTR(error);
1790 }
1791 return dget(s->s_root);
1792}
1793EXPORT_SYMBOL(mount_single);
1794
1795/**
1796 * vfs_get_tree - Get the mountable root
1797 * @fc: The superblock configuration context.
1798 *
1799 * The filesystem is invoked to get or create a superblock which can then later
1800 * be used for mounting. The filesystem places a pointer to the root to be
1801 * used for mounting in @fc->root.
1802 */
1803int vfs_get_tree(struct fs_context *fc)
1804{
1805 struct super_block *sb;
1806 int error;
1807
1808 if (fc->root)
1809 return -EBUSY;
1810
1811 /* Get the mountable root in fc->root, with a ref on the root and a ref
1812 * on the superblock.
1813 */
1814 error = fc->ops->get_tree(fc);
1815 if (error < 0)
1816 return error;
1817
1818 if (!fc->root) {
1819 pr_err("Filesystem %s get_tree() didn't set fc->root, returned %i\n",
1820 fc->fs_type->name, error);
1821 /* We don't know what the locking state of the superblock is -
1822 * if there is a superblock.
1823 */
1824 BUG();
1825 }
1826
1827 sb = fc->root->d_sb;
1828 WARN_ON(!sb->s_bdi);
1829
1830 /*
1831 * super_wake() contains a memory barrier which also care of
1832 * ordering for super_cache_count(). We place it before setting
1833 * SB_BORN as the data dependency between the two functions is
1834 * the superblock structure contents that we just set up, not
1835 * the SB_BORN flag.
1836 */
1837 super_wake(sb, SB_BORN);
1838
1839 error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
1840 if (unlikely(error)) {
1841 fc_drop_locked(fc);
1842 return error;
1843 }
1844
1845 /*
1846 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1847 * but s_maxbytes was an unsigned long long for many releases. Throw
1848 * this warning for a little while to try and catch filesystems that
1849 * violate this rule.
1850 */
1851 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1852 "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
1853
1854 return 0;
1855}
1856EXPORT_SYMBOL(vfs_get_tree);
1857
1858/*
1859 * Setup private BDI for given superblock. It gets automatically cleaned up
1860 * in generic_shutdown_super().
1861 */
1862int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1863{
1864 struct backing_dev_info *bdi;
1865 int err;
1866 va_list args;
1867
1868 bdi = bdi_alloc(NUMA_NO_NODE);
1869 if (!bdi)
1870 return -ENOMEM;
1871
1872 va_start(args, fmt);
1873 err = bdi_register_va(bdi, fmt, args);
1874 va_end(args);
1875 if (err) {
1876 bdi_put(bdi);
1877 return err;
1878 }
1879 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1880 sb->s_bdi = bdi;
1881 sb->s_iflags |= SB_I_PERSB_BDI;
1882
1883 return 0;
1884}
1885EXPORT_SYMBOL(super_setup_bdi_name);
1886
1887/*
1888 * Setup private BDI for given superblock. I gets automatically cleaned up
1889 * in generic_shutdown_super().
1890 */
1891int super_setup_bdi(struct super_block *sb)
1892{
1893 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1894
1895 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1896 atomic_long_inc_return(&bdi_seq));
1897}
1898EXPORT_SYMBOL(super_setup_bdi);
1899
1900/**
1901 * sb_wait_write - wait until all writers to given file system finish
1902 * @sb: the super for which we wait
1903 * @level: type of writers we wait for (normal vs page fault)
1904 *
1905 * This function waits until there are no writers of given type to given file
1906 * system.
1907 */
1908static void sb_wait_write(struct super_block *sb, int level)
1909{
1910 percpu_down_write(sb->s_writers.rw_sem + level-1);
1911}
1912
1913/*
1914 * We are going to return to userspace and forget about these locks, the
1915 * ownership goes to the caller of thaw_super() which does unlock().
1916 */
1917static void lockdep_sb_freeze_release(struct super_block *sb)
1918{
1919 int level;
1920
1921 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1922 percpu_rwsem_release(sb->s_writers.rw_sem + level, _THIS_IP_);
1923}
1924
1925/*
1926 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1927 */
1928static void lockdep_sb_freeze_acquire(struct super_block *sb)
1929{
1930 int level;
1931
1932 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1933 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1934}
1935
1936static void sb_freeze_unlock(struct super_block *sb, int level)
1937{
1938 for (level--; level >= 0; level--)
1939 percpu_up_write(sb->s_writers.rw_sem + level);
1940}
1941
1942static int wait_for_partially_frozen(struct super_block *sb)
1943{
1944 int ret = 0;
1945
1946 do {
1947 unsigned short old = sb->s_writers.frozen;
1948
1949 up_write(&sb->s_umount);
1950 ret = wait_var_event_killable(&sb->s_writers.frozen,
1951 sb->s_writers.frozen != old);
1952 down_write(&sb->s_umount);
1953 } while (ret == 0 &&
1954 sb->s_writers.frozen != SB_UNFROZEN &&
1955 sb->s_writers.frozen != SB_FREEZE_COMPLETE);
1956
1957 return ret;
1958}
1959
1960#define FREEZE_HOLDERS (FREEZE_HOLDER_KERNEL | FREEZE_HOLDER_USERSPACE)
1961#define FREEZE_FLAGS (FREEZE_HOLDERS | FREEZE_MAY_NEST)
1962
1963static inline int freeze_inc(struct super_block *sb, enum freeze_holder who)
1964{
1965 WARN_ON_ONCE((who & ~FREEZE_FLAGS));
1966 WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1);
1967
1968 if (who & FREEZE_HOLDER_KERNEL)
1969 ++sb->s_writers.freeze_kcount;
1970 if (who & FREEZE_HOLDER_USERSPACE)
1971 ++sb->s_writers.freeze_ucount;
1972 return sb->s_writers.freeze_kcount + sb->s_writers.freeze_ucount;
1973}
1974
1975static inline int freeze_dec(struct super_block *sb, enum freeze_holder who)
1976{
1977 WARN_ON_ONCE((who & ~FREEZE_FLAGS));
1978 WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1);
1979
1980 if ((who & FREEZE_HOLDER_KERNEL) && sb->s_writers.freeze_kcount)
1981 --sb->s_writers.freeze_kcount;
1982 if ((who & FREEZE_HOLDER_USERSPACE) && sb->s_writers.freeze_ucount)
1983 --sb->s_writers.freeze_ucount;
1984 return sb->s_writers.freeze_kcount + sb->s_writers.freeze_ucount;
1985}
1986
1987static inline bool may_freeze(struct super_block *sb, enum freeze_holder who)
1988{
1989 WARN_ON_ONCE((who & ~FREEZE_FLAGS));
1990 WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1);
1991
1992 if (who & FREEZE_HOLDER_KERNEL)
1993 return (who & FREEZE_MAY_NEST) ||
1994 sb->s_writers.freeze_kcount == 0;
1995 if (who & FREEZE_HOLDER_USERSPACE)
1996 return (who & FREEZE_MAY_NEST) ||
1997 sb->s_writers.freeze_ucount == 0;
1998 return false;
1999}
2000
2001/**
2002 * freeze_super - lock the filesystem and force it into a consistent state
2003 * @sb: the super to lock
2004 * @who: context that wants to freeze
2005 *
2006 * Syncs the super to make sure the filesystem is consistent and calls the fs's
2007 * freeze_fs. Subsequent calls to this without first thawing the fs may return
2008 * -EBUSY.
2009 *
2010 * @who should be:
2011 * * %FREEZE_HOLDER_USERSPACE if userspace wants to freeze the fs;
2012 * * %FREEZE_HOLDER_KERNEL if the kernel wants to freeze the fs.
2013 * * %FREEZE_MAY_NEST whether nesting freeze and thaw requests is allowed.
2014 *
2015 * The @who argument distinguishes between the kernel and userspace trying to
2016 * freeze the filesystem. Although there cannot be multiple kernel freezes or
2017 * multiple userspace freezes in effect at any given time, the kernel and
2018 * userspace can both hold a filesystem frozen. The filesystem remains frozen
2019 * until there are no kernel or userspace freezes in effect.
2020 *
2021 * A filesystem may hold multiple devices and thus a filesystems may be
2022 * frozen through the block layer via multiple block devices. In this
2023 * case the request is marked as being allowed to nest by passing
2024 * FREEZE_MAY_NEST. The filesystem remains frozen until all block
2025 * devices are unfrozen. If multiple freezes are attempted without
2026 * FREEZE_MAY_NEST -EBUSY will be returned.
2027 *
2028 * During this function, sb->s_writers.frozen goes through these values:
2029 *
2030 * SB_UNFROZEN: File system is normal, all writes progress as usual.
2031 *
2032 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
2033 * writes should be blocked, though page faults are still allowed. We wait for
2034 * all writes to complete and then proceed to the next stage.
2035 *
2036 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
2037 * but internal fs threads can still modify the filesystem (although they
2038 * should not dirty new pages or inodes), writeback can run etc. After waiting
2039 * for all running page faults we sync the filesystem which will clean all
2040 * dirty pages and inodes (no new dirty pages or inodes can be created when
2041 * sync is running).
2042 *
2043 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
2044 * modification are blocked (e.g. XFS preallocation truncation on inode
2045 * reclaim). This is usually implemented by blocking new transactions for
2046 * filesystems that have them and need this additional guard. After all
2047 * internal writers are finished we call ->freeze_fs() to finish filesystem
2048 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
2049 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
2050 *
2051 * sb->s_writers.frozen is protected by sb->s_umount.
2052 *
2053 * Return: If the freeze was successful zero is returned. If the freeze
2054 * failed a negative error code is returned.
2055 */
2056int freeze_super(struct super_block *sb, enum freeze_holder who)
2057{
2058 int ret;
2059
2060 if (!super_lock_excl(sb)) {
2061 WARN_ON_ONCE("Dying superblock while freezing!");
2062 return -EINVAL;
2063 }
2064 atomic_inc(&sb->s_active);
2065
2066retry:
2067 if (sb->s_writers.frozen == SB_FREEZE_COMPLETE) {
2068 if (may_freeze(sb, who))
2069 ret = !!WARN_ON_ONCE(freeze_inc(sb, who) == 1);
2070 else
2071 ret = -EBUSY;
2072 /* All freezers share a single active reference. */
2073 deactivate_locked_super(sb);
2074 return ret;
2075 }
2076
2077 if (sb->s_writers.frozen != SB_UNFROZEN) {
2078 ret = wait_for_partially_frozen(sb);
2079 if (ret) {
2080 deactivate_locked_super(sb);
2081 return ret;
2082 }
2083
2084 goto retry;
2085 }
2086
2087 if (sb_rdonly(sb)) {
2088 /* Nothing to do really... */
2089 WARN_ON_ONCE(freeze_inc(sb, who) > 1);
2090 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
2091 wake_up_var(&sb->s_writers.frozen);
2092 super_unlock_excl(sb);
2093 return 0;
2094 }
2095
2096 sb->s_writers.frozen = SB_FREEZE_WRITE;
2097 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
2098 super_unlock_excl(sb);
2099 sb_wait_write(sb, SB_FREEZE_WRITE);
2100 __super_lock_excl(sb);
2101
2102 /* Now we go and block page faults... */
2103 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
2104 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
2105
2106 /* All writers are done so after syncing there won't be dirty data */
2107 ret = sync_filesystem(sb);
2108 if (ret) {
2109 sb->s_writers.frozen = SB_UNFROZEN;
2110 sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT);
2111 wake_up_var(&sb->s_writers.frozen);
2112 deactivate_locked_super(sb);
2113 return ret;
2114 }
2115
2116 /* Now wait for internal filesystem counter */
2117 sb->s_writers.frozen = SB_FREEZE_FS;
2118 sb_wait_write(sb, SB_FREEZE_FS);
2119
2120 if (sb->s_op->freeze_fs) {
2121 ret = sb->s_op->freeze_fs(sb);
2122 if (ret) {
2123 printk(KERN_ERR
2124 "VFS:Filesystem freeze failed\n");
2125 sb->s_writers.frozen = SB_UNFROZEN;
2126 sb_freeze_unlock(sb, SB_FREEZE_FS);
2127 wake_up_var(&sb->s_writers.frozen);
2128 deactivate_locked_super(sb);
2129 return ret;
2130 }
2131 }
2132 /*
2133 * For debugging purposes so that fs can warn if it sees write activity
2134 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
2135 */
2136 WARN_ON_ONCE(freeze_inc(sb, who) > 1);
2137 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
2138 wake_up_var(&sb->s_writers.frozen);
2139 lockdep_sb_freeze_release(sb);
2140 super_unlock_excl(sb);
2141 return 0;
2142}
2143EXPORT_SYMBOL(freeze_super);
2144
2145/*
2146 * Undoes the effect of a freeze_super_locked call. If the filesystem is
2147 * frozen both by userspace and the kernel, a thaw call from either source
2148 * removes that state without releasing the other state or unlocking the
2149 * filesystem.
2150 */
2151static int thaw_super_locked(struct super_block *sb, enum freeze_holder who)
2152{
2153 int error = -EINVAL;
2154
2155 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE)
2156 goto out_unlock;
2157
2158 /*
2159 * All freezers share a single active reference.
2160 * So just unlock in case there are any left.
2161 */
2162 if (freeze_dec(sb, who))
2163 goto out_unlock;
2164
2165 if (sb_rdonly(sb)) {
2166 sb->s_writers.frozen = SB_UNFROZEN;
2167 wake_up_var(&sb->s_writers.frozen);
2168 goto out_deactivate;
2169 }
2170
2171 lockdep_sb_freeze_acquire(sb);
2172
2173 if (sb->s_op->unfreeze_fs) {
2174 error = sb->s_op->unfreeze_fs(sb);
2175 if (error) {
2176 pr_err("VFS: Filesystem thaw failed\n");
2177 freeze_inc(sb, who);
2178 lockdep_sb_freeze_release(sb);
2179 goto out_unlock;
2180 }
2181 }
2182
2183 sb->s_writers.frozen = SB_UNFROZEN;
2184 wake_up_var(&sb->s_writers.frozen);
2185 sb_freeze_unlock(sb, SB_FREEZE_FS);
2186out_deactivate:
2187 deactivate_locked_super(sb);
2188 return 0;
2189
2190out_unlock:
2191 super_unlock_excl(sb);
2192 return error;
2193}
2194
2195/**
2196 * thaw_super -- unlock filesystem
2197 * @sb: the super to thaw
2198 * @who: context that wants to freeze
2199 *
2200 * Unlocks the filesystem and marks it writeable again after freeze_super()
2201 * if there are no remaining freezes on the filesystem.
2202 *
2203 * @who should be:
2204 * * %FREEZE_HOLDER_USERSPACE if userspace wants to thaw the fs;
2205 * * %FREEZE_HOLDER_KERNEL if the kernel wants to thaw the fs.
2206 * * %FREEZE_MAY_NEST whether nesting freeze and thaw requests is allowed
2207 *
2208 * A filesystem may hold multiple devices and thus a filesystems may
2209 * have been frozen through the block layer via multiple block devices.
2210 * The filesystem remains frozen until all block devices are unfrozen.
2211 */
2212int thaw_super(struct super_block *sb, enum freeze_holder who)
2213{
2214 if (!super_lock_excl(sb)) {
2215 WARN_ON_ONCE("Dying superblock while thawing!");
2216 return -EINVAL;
2217 }
2218 return thaw_super_locked(sb, who);
2219}
2220EXPORT_SYMBOL(thaw_super);
2221
2222/*
2223 * Create workqueue for deferred direct IO completions. We allocate the
2224 * workqueue when it's first needed. This avoids creating workqueue for
2225 * filesystems that don't need it and also allows us to create the workqueue
2226 * late enough so the we can include s_id in the name of the workqueue.
2227 */
2228int sb_init_dio_done_wq(struct super_block *sb)
2229{
2230 struct workqueue_struct *old;
2231 struct workqueue_struct *wq = alloc_workqueue("dio/%s",
2232 WQ_MEM_RECLAIM, 0,
2233 sb->s_id);
2234 if (!wq)
2235 return -ENOMEM;
2236 /*
2237 * This has to be atomic as more DIOs can race to create the workqueue
2238 */
2239 old = cmpxchg(&sb->s_dio_done_wq, NULL, wq);
2240 /* Someone created workqueue before us? Free ours... */
2241 if (old)
2242 destroy_workqueue(wq);
2243 return 0;
2244}
2245EXPORT_SYMBOL_GPL(sb_init_dio_done_wq);