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