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 17 : static void destroy_super_work(struct work_struct *work)
157 : {
158 17 : struct super_block *s = container_of(work, struct super_block,
159 : destroy_work);
160 : int i;
161 :
162 68 : for (i = 0; i < SB_FREEZE_LEVELS; i++)
163 51 : percpu_free_rwsem(&s->s_writers.rw_sem[i]);
164 17 : kfree(s);
165 17 : }
166 :
167 17 : static void destroy_super_rcu(struct rcu_head *head)
168 : {
169 17 : struct super_block *s = container_of(head, struct super_block, rcu);
170 34 : INIT_WORK(&s->destroy_work, destroy_super_work);
171 34 : schedule_work(&s->destroy_work);
172 17 : }
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 27 : static struct super_block *alloc_super(struct file_system_type *type, int flags,
200 : struct user_namespace *user_ns)
201 : {
202 27 : struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
203 : static const struct super_operations default_op;
204 : int i;
205 :
206 27 : if (!s)
207 : return NULL;
208 :
209 54 : INIT_LIST_HEAD(&s->s_mounts);
210 27 : s->s_user_ns = get_user_ns(user_ns);
211 27 : 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 27 : down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
229 :
230 27 : if (security_sb_alloc(s))
231 : goto fail;
232 :
233 108 : for (i = 0; i < SB_FREEZE_LEVELS; i++) {
234 162 : if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
235 81 : sb_writers_name[i],
236 : &type->s_writers_key[i]))
237 : goto fail;
238 : }
239 27 : init_waitqueue_head(&s->s_writers.wait_unfrozen);
240 27 : s->s_bdi = &noop_backing_dev_info;
241 27 : s->s_flags = flags;
242 27 : if (s->s_user_ns != &init_user_ns)
243 0 : s->s_iflags |= SB_I_NODEV;
244 54 : INIT_HLIST_NODE(&s->s_instances);
245 27 : INIT_HLIST_BL_HEAD(&s->s_roots);
246 27 : mutex_init(&s->s_sync_lock);
247 54 : INIT_LIST_HEAD(&s->s_inodes);
248 27 : spin_lock_init(&s->s_inode_list_lock);
249 54 : INIT_LIST_HEAD(&s->s_inodes_wb);
250 27 : spin_lock_init(&s->s_inode_wblist_lock);
251 :
252 27 : s->s_count = 1;
253 54 : atomic_set(&s->s_active, 1);
254 27 : mutex_init(&s->s_vfs_rename_mutex);
255 : lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
256 27 : init_rwsem(&s->s_dquot.dqio_sem);
257 27 : s->s_maxbytes = MAX_NON_LFS;
258 27 : s->s_op = &default_op;
259 27 : s->s_time_gran = 1000000000;
260 27 : s->s_time_min = TIME64_MIN;
261 27 : s->s_time_max = TIME64_MAX;
262 :
263 27 : s->s_shrink.seeks = DEFAULT_SEEKS;
264 27 : s->s_shrink.scan_objects = super_cache_scan;
265 27 : s->s_shrink.count_objects = super_cache_count;
266 27 : s->s_shrink.batch = 1024;
267 27 : s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
268 27 : if (prealloc_shrinker(&s->s_shrink, "sb-%s", type->name))
269 : goto fail;
270 27 : if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
271 : goto fail;
272 27 : if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
273 : goto fail;
274 : return s;
275 :
276 : fail:
277 0 : 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 : */
286 17 : static void __put_super(struct super_block *s)
287 : {
288 17 : if (!--s->s_count) {
289 34 : list_del_init(&s->s_list);
290 17 : WARN_ON(s->s_dentry_lru.node);
291 17 : WARN_ON(s->s_inode_lru.node);
292 17 : WARN_ON(!list_empty(&s->s_mounts));
293 17 : security_sb_free(s);
294 17 : put_user_ns(s->s_user_ns);
295 17 : kfree(s->s_subtype);
296 17 : call_rcu(&s->rcu, destroy_super_rcu);
297 : }
298 17 : }
299 :
300 : /**
301 : * put_super - drop a temporary reference to superblock
302 : * @sb: superblock in question
303 : *
304 : * Drops a temporary reference, frees superblock if there's no
305 : * references left.
306 : */
307 0 : void put_super(struct super_block *sb)
308 : {
309 17 : spin_lock(&sb_lock);
310 17 : __put_super(sb);
311 17 : spin_unlock(&sb_lock);
312 0 : }
313 :
314 :
315 : /**
316 : * deactivate_locked_super - drop an active reference to superblock
317 : * @s: superblock to deactivate
318 : *
319 : * Drops an active reference to superblock, converting it into a temporary
320 : * one if there is no other active references left. In that case we
321 : * tell fs driver to shut it down and drop the temporary reference we
322 : * had just acquired.
323 : *
324 : * Caller holds exclusive lock on superblock; that lock is released.
325 : */
326 17 : void deactivate_locked_super(struct super_block *s)
327 : {
328 17 : struct file_system_type *fs = s->s_type;
329 34 : if (atomic_dec_and_test(&s->s_active)) {
330 17 : unregister_shrinker(&s->s_shrink);
331 17 : fs->kill_sb(s);
332 :
333 : /*
334 : * Since list_lru_destroy() may sleep, we cannot call it from
335 : * put_super(), where we hold the sb_lock. Therefore we destroy
336 : * the lru lists right now.
337 : */
338 17 : list_lru_destroy(&s->s_dentry_lru);
339 17 : list_lru_destroy(&s->s_inode_lru);
340 :
341 17 : put_filesystem(fs);
342 : put_super(s);
343 : } else {
344 0 : up_write(&s->s_umount);
345 : }
346 17 : }
347 :
348 : EXPORT_SYMBOL(deactivate_locked_super);
349 :
350 : /**
351 : * deactivate_super - drop an active reference to superblock
352 : * @s: superblock to deactivate
353 : *
354 : * Variant of deactivate_locked_super(), except that superblock is *not*
355 : * locked by caller. If we are going to drop the final active reference,
356 : * lock will be acquired prior to that.
357 : */
358 44 : void deactivate_super(struct super_block *s)
359 : {
360 88 : if (!atomic_add_unless(&s->s_active, -1, 1)) {
361 17 : down_write(&s->s_umount);
362 17 : deactivate_locked_super(s);
363 : }
364 44 : }
365 :
366 : EXPORT_SYMBOL(deactivate_super);
367 :
368 : /**
369 : * grab_super - acquire an active reference
370 : * @s: reference we are trying to make active
371 : *
372 : * Tries to acquire an active reference. grab_super() is used when we
373 : * had just found a superblock in super_blocks or fs_type->fs_supers
374 : * and want to turn it into a full-blown active reference. grab_super()
375 : * is called with sb_lock held and drops it. Returns 1 in case of
376 : * success, 0 if we had failed (superblock contents was already dead or
377 : * dying when grab_super() had been called). Note that this is only
378 : * called for superblocks not in rundown mode (== ones still on ->fs_supers
379 : * of their type), so increment of ->s_count is OK here.
380 : */
381 0 : static int grab_super(struct super_block *s) __releases(sb_lock)
382 : {
383 0 : s->s_count++;
384 0 : spin_unlock(&sb_lock);
385 0 : down_write(&s->s_umount);
386 0 : if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
387 0 : put_super(s);
388 0 : return 1;
389 : }
390 0 : up_write(&s->s_umount);
391 0 : put_super(s);
392 0 : return 0;
393 : }
394 :
395 : /*
396 : * trylock_super - try to grab ->s_umount shared
397 : * @sb: reference we are trying to grab
398 : *
399 : * Try to prevent fs shutdown. This is used in places where we
400 : * cannot take an active reference but we need to ensure that the
401 : * filesystem is not shut down while we are working on it. It returns
402 : * false if we cannot acquire s_umount or if we lose the race and
403 : * filesystem already got into shutdown, and returns true with the s_umount
404 : * lock held in read mode in case of success. On successful return,
405 : * the caller must drop the s_umount lock when done.
406 : *
407 : * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
408 : * The reason why it's safe is that we are OK with doing trylock instead
409 : * of down_read(). There's a couple of places that are OK with that, but
410 : * it's very much not a general-purpose interface.
411 : */
412 0 : bool trylock_super(struct super_block *sb)
413 : {
414 0 : if (down_read_trylock(&sb->s_umount)) {
415 0 : if (!hlist_unhashed(&sb->s_instances) &&
416 0 : sb->s_root && (sb->s_flags & SB_BORN))
417 : return true;
418 0 : up_read(&sb->s_umount);
419 : }
420 :
421 : return false;
422 : }
423 :
424 : /**
425 : * retire_super - prevents superblock from being reused
426 : * @sb: superblock to retire
427 : *
428 : * The function marks superblock to be ignored in superblock test, which
429 : * prevents it from being reused for any new mounts. If the superblock has
430 : * a private bdi, it also unregisters it, but doesn't reduce the refcount
431 : * of the superblock to prevent potential races. The refcount is reduced
432 : * by generic_shutdown_super(). The function can not be called
433 : * concurrently with generic_shutdown_super(). It is safe to call the
434 : * function multiple times, subsequent calls have no effect.
435 : *
436 : * The marker will affect the re-use only for block-device-based
437 : * superblocks. Other superblocks will still get marked if this function
438 : * is used, but that will not affect their reusability.
439 : */
440 0 : void retire_super(struct super_block *sb)
441 : {
442 0 : WARN_ON(!sb->s_bdev);
443 0 : down_write(&sb->s_umount);
444 0 : if (sb->s_iflags & SB_I_PERSB_BDI) {
445 0 : bdi_unregister(sb->s_bdi);
446 0 : sb->s_iflags &= ~SB_I_PERSB_BDI;
447 : }
448 0 : sb->s_iflags |= SB_I_RETIRED;
449 0 : up_write(&sb->s_umount);
450 0 : }
451 : EXPORT_SYMBOL(retire_super);
452 :
453 : /**
454 : * generic_shutdown_super - common helper for ->kill_sb()
455 : * @sb: superblock to kill
456 : *
457 : * generic_shutdown_super() does all fs-independent work on superblock
458 : * shutdown. Typical ->kill_sb() should pick all fs-specific objects
459 : * that need destruction out of superblock, call generic_shutdown_super()
460 : * and release aforementioned objects. Note: dentries and inodes _are_
461 : * taken care of and do not need specific handling.
462 : *
463 : * Upon calling this function, the filesystem may no longer alter or
464 : * rearrange the set of dentries belonging to this super_block, nor may it
465 : * change the attachments of dentries to inodes.
466 : */
467 17 : void generic_shutdown_super(struct super_block *sb)
468 : {
469 17 : const struct super_operations *sop = sb->s_op;
470 :
471 17 : if (sb->s_root) {
472 17 : shrink_dcache_for_umount(sb);
473 17 : sync_filesystem(sb);
474 17 : sb->s_flags &= ~SB_ACTIVE;
475 :
476 : cgroup_writeback_umount();
477 :
478 : /* evict all inodes with zero refcount */
479 17 : evict_inodes(sb);
480 : /* only nonzero refcount inodes can have marks */
481 17 : fsnotify_sb_delete(sb);
482 17 : fscrypt_destroy_keyring(sb);
483 17 : security_sb_delete(sb);
484 :
485 17 : if (sb->s_dio_done_wq) {
486 0 : destroy_workqueue(sb->s_dio_done_wq);
487 0 : sb->s_dio_done_wq = NULL;
488 : }
489 :
490 17 : if (sop->put_super)
491 0 : sop->put_super(sb);
492 :
493 34 : if (CHECK_DATA_CORRUPTION(!list_empty(&sb->s_inodes),
494 : "VFS: Busy inodes after unmount of %s (%s)",
495 : sb->s_id, sb->s_type->name)) {
496 : /*
497 : * Adding a proper bailout path here would be hard, but
498 : * we can at least make it more likely that a later
499 : * iput_final() or such crashes cleanly.
500 : */
501 : struct inode *inode;
502 :
503 0 : spin_lock(&sb->s_inode_list_lock);
504 0 : list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
505 0 : inode->i_op = VFS_PTR_POISON;
506 0 : inode->i_sb = VFS_PTR_POISON;
507 0 : inode->i_mapping = VFS_PTR_POISON;
508 : }
509 0 : spin_unlock(&sb->s_inode_list_lock);
510 : }
511 : }
512 17 : spin_lock(&sb_lock);
513 : /* should be initialized for __put_super_and_need_restart() */
514 34 : hlist_del_init(&sb->s_instances);
515 17 : spin_unlock(&sb_lock);
516 17 : up_write(&sb->s_umount);
517 17 : if (sb->s_bdi != &noop_backing_dev_info) {
518 0 : if (sb->s_iflags & SB_I_PERSB_BDI)
519 0 : bdi_unregister(sb->s_bdi);
520 0 : bdi_put(sb->s_bdi);
521 0 : sb->s_bdi = &noop_backing_dev_info;
522 : }
523 17 : }
524 :
525 : EXPORT_SYMBOL(generic_shutdown_super);
526 :
527 0 : bool mount_capable(struct fs_context *fc)
528 : {
529 0 : if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT))
530 0 : return capable(CAP_SYS_ADMIN);
531 : else
532 0 : return ns_capable(fc->user_ns, CAP_SYS_ADMIN);
533 : }
534 :
535 : /**
536 : * sget_fc - Find or create a superblock
537 : * @fc: Filesystem context.
538 : * @test: Comparison callback
539 : * @set: Setup callback
540 : *
541 : * Find or create a superblock using the parameters stored in the filesystem
542 : * context and the two callback functions.
543 : *
544 : * If an extant superblock is matched, then that will be returned with an
545 : * elevated reference count that the caller must transfer or discard.
546 : *
547 : * If no match is made, a new superblock will be allocated and basic
548 : * initialisation will be performed (s_type, s_fs_info and s_id will be set and
549 : * the set() callback will be invoked), the superblock will be published and it
550 : * will be returned in a partially constructed state with SB_BORN and SB_ACTIVE
551 : * as yet unset.
552 : */
553 27 : struct super_block *sget_fc(struct fs_context *fc,
554 : int (*test)(struct super_block *, struct fs_context *),
555 : int (*set)(struct super_block *, struct fs_context *))
556 : {
557 27 : struct super_block *s = NULL;
558 : struct super_block *old;
559 27 : struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
560 : int err;
561 :
562 : retry:
563 54 : spin_lock(&sb_lock);
564 54 : if (test) {
565 0 : hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
566 0 : if (test(old, fc))
567 : goto share_extant_sb;
568 : }
569 : }
570 54 : if (!s) {
571 27 : spin_unlock(&sb_lock);
572 27 : s = alloc_super(fc->fs_type, fc->sb_flags, user_ns);
573 27 : if (!s)
574 : return ERR_PTR(-ENOMEM);
575 : goto retry;
576 : }
577 :
578 27 : s->s_fs_info = fc->s_fs_info;
579 27 : err = set(s, fc);
580 27 : if (err) {
581 0 : s->s_fs_info = NULL;
582 0 : spin_unlock(&sb_lock);
583 0 : destroy_unused_super(s);
584 0 : return ERR_PTR(err);
585 : }
586 27 : fc->s_fs_info = NULL;
587 27 : s->s_type = fc->fs_type;
588 27 : s->s_iflags |= fc->s_iflags;
589 27 : strlcpy(s->s_id, s->s_type->name, sizeof(s->s_id));
590 54 : list_add_tail(&s->s_list, &super_blocks);
591 54 : hlist_add_head(&s->s_instances, &s->s_type->fs_supers);
592 27 : spin_unlock(&sb_lock);
593 27 : get_filesystem(s->s_type);
594 27 : register_shrinker_prepared(&s->s_shrink);
595 27 : return s;
596 :
597 : share_extant_sb:
598 0 : if (user_ns != old->s_user_ns) {
599 0 : spin_unlock(&sb_lock);
600 0 : destroy_unused_super(s);
601 0 : return ERR_PTR(-EBUSY);
602 : }
603 0 : if (!grab_super(old))
604 : goto retry;
605 0 : destroy_unused_super(s);
606 0 : return old;
607 : }
608 : EXPORT_SYMBOL(sget_fc);
609 :
610 : /**
611 : * sget - find or create a superblock
612 : * @type: filesystem type superblock should belong to
613 : * @test: comparison callback
614 : * @set: setup callback
615 : * @flags: mount flags
616 : * @data: argument to each of them
617 : */
618 0 : struct super_block *sget(struct file_system_type *type,
619 : int (*test)(struct super_block *,void *),
620 : int (*set)(struct super_block *,void *),
621 : int flags,
622 : void *data)
623 : {
624 0 : struct user_namespace *user_ns = current_user_ns();
625 0 : struct super_block *s = NULL;
626 : struct super_block *old;
627 : int err;
628 :
629 : /* We don't yet pass the user namespace of the parent
630 : * mount through to here so always use &init_user_ns
631 : * until that changes.
632 : */
633 : if (flags & SB_SUBMOUNT)
634 : user_ns = &init_user_ns;
635 :
636 : retry:
637 0 : spin_lock(&sb_lock);
638 0 : if (test) {
639 0 : hlist_for_each_entry(old, &type->fs_supers, s_instances) {
640 0 : if (!test(old, data))
641 0 : continue;
642 0 : if (user_ns != old->s_user_ns) {
643 0 : spin_unlock(&sb_lock);
644 0 : destroy_unused_super(s);
645 0 : return ERR_PTR(-EBUSY);
646 : }
647 0 : if (!grab_super(old))
648 : goto retry;
649 0 : destroy_unused_super(s);
650 0 : return old;
651 : }
652 : }
653 0 : if (!s) {
654 0 : spin_unlock(&sb_lock);
655 0 : s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
656 0 : if (!s)
657 : return ERR_PTR(-ENOMEM);
658 : goto retry;
659 : }
660 :
661 0 : err = set(s, data);
662 0 : if (err) {
663 0 : spin_unlock(&sb_lock);
664 0 : destroy_unused_super(s);
665 0 : return ERR_PTR(err);
666 : }
667 0 : s->s_type = type;
668 0 : strlcpy(s->s_id, type->name, sizeof(s->s_id));
669 0 : list_add_tail(&s->s_list, &super_blocks);
670 0 : hlist_add_head(&s->s_instances, &type->fs_supers);
671 0 : spin_unlock(&sb_lock);
672 0 : get_filesystem(type);
673 0 : register_shrinker_prepared(&s->s_shrink);
674 0 : return s;
675 : }
676 : EXPORT_SYMBOL(sget);
677 :
678 0 : void drop_super(struct super_block *sb)
679 : {
680 0 : up_read(&sb->s_umount);
681 0 : put_super(sb);
682 0 : }
683 :
684 : EXPORT_SYMBOL(drop_super);
685 :
686 0 : void drop_super_exclusive(struct super_block *sb)
687 : {
688 0 : up_write(&sb->s_umount);
689 0 : put_super(sb);
690 0 : }
691 : EXPORT_SYMBOL(drop_super_exclusive);
692 :
693 0 : static void __iterate_supers(void (*f)(struct super_block *))
694 : {
695 0 : struct super_block *sb, *p = NULL;
696 :
697 0 : spin_lock(&sb_lock);
698 0 : list_for_each_entry(sb, &super_blocks, s_list) {
699 0 : if (hlist_unhashed(&sb->s_instances))
700 0 : continue;
701 0 : sb->s_count++;
702 0 : spin_unlock(&sb_lock);
703 :
704 0 : f(sb);
705 :
706 0 : spin_lock(&sb_lock);
707 0 : if (p)
708 0 : __put_super(p);
709 : p = sb;
710 : }
711 0 : if (p)
712 0 : __put_super(p);
713 0 : spin_unlock(&sb_lock);
714 0 : }
715 : /**
716 : * iterate_supers - call function for all active superblocks
717 : * @f: function to call
718 : * @arg: argument to pass to it
719 : *
720 : * Scans the superblock list and calls given function, passing it
721 : * locked superblock and given argument.
722 : */
723 0 : void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
724 : {
725 0 : struct super_block *sb, *p = NULL;
726 :
727 0 : spin_lock(&sb_lock);
728 0 : list_for_each_entry(sb, &super_blocks, s_list) {
729 0 : if (hlist_unhashed(&sb->s_instances))
730 0 : continue;
731 0 : sb->s_count++;
732 0 : spin_unlock(&sb_lock);
733 :
734 0 : down_read(&sb->s_umount);
735 0 : if (sb->s_root && (sb->s_flags & SB_BORN))
736 0 : f(sb, arg);
737 0 : up_read(&sb->s_umount);
738 :
739 0 : spin_lock(&sb_lock);
740 0 : if (p)
741 0 : __put_super(p);
742 : p = sb;
743 : }
744 0 : if (p)
745 0 : __put_super(p);
746 0 : spin_unlock(&sb_lock);
747 0 : }
748 :
749 : /**
750 : * iterate_supers_type - call function for superblocks of given type
751 : * @type: fs type
752 : * @f: function to call
753 : * @arg: argument to pass to it
754 : *
755 : * Scans the superblock list and calls given function, passing it
756 : * locked superblock and given argument.
757 : */
758 0 : void iterate_supers_type(struct file_system_type *type,
759 : void (*f)(struct super_block *, void *), void *arg)
760 : {
761 0 : struct super_block *sb, *p = NULL;
762 :
763 0 : spin_lock(&sb_lock);
764 0 : hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
765 0 : sb->s_count++;
766 0 : spin_unlock(&sb_lock);
767 :
768 0 : down_read(&sb->s_umount);
769 0 : if (sb->s_root && (sb->s_flags & SB_BORN))
770 0 : f(sb, arg);
771 0 : up_read(&sb->s_umount);
772 :
773 0 : spin_lock(&sb_lock);
774 0 : if (p)
775 0 : __put_super(p);
776 0 : p = sb;
777 : }
778 0 : if (p)
779 0 : __put_super(p);
780 0 : spin_unlock(&sb_lock);
781 0 : }
782 :
783 : EXPORT_SYMBOL(iterate_supers_type);
784 :
785 : /**
786 : * get_super - get the superblock of a device
787 : * @bdev: device to get the superblock for
788 : *
789 : * Scans the superblock list and finds the superblock of the file system
790 : * mounted on the device given. %NULL is returned if no match is found.
791 : */
792 0 : struct super_block *get_super(struct block_device *bdev)
793 : {
794 : struct super_block *sb;
795 :
796 0 : if (!bdev)
797 : return NULL;
798 :
799 : spin_lock(&sb_lock);
800 : rescan:
801 0 : list_for_each_entry(sb, &super_blocks, s_list) {
802 0 : if (hlist_unhashed(&sb->s_instances))
803 0 : continue;
804 0 : if (sb->s_bdev == bdev) {
805 0 : sb->s_count++;
806 0 : spin_unlock(&sb_lock);
807 0 : down_read(&sb->s_umount);
808 : /* still alive? */
809 0 : if (sb->s_root && (sb->s_flags & SB_BORN))
810 : return sb;
811 0 : up_read(&sb->s_umount);
812 : /* nope, got unmounted */
813 0 : spin_lock(&sb_lock);
814 0 : __put_super(sb);
815 0 : goto rescan;
816 : }
817 : }
818 0 : spin_unlock(&sb_lock);
819 0 : return NULL;
820 : }
821 :
822 : /**
823 : * get_active_super - get an active reference to the superblock of a device
824 : * @bdev: device to get the superblock for
825 : *
826 : * Scans the superblock list and finds the superblock of the file system
827 : * mounted on the device given. Returns the superblock with an active
828 : * reference or %NULL if none was found.
829 : */
830 0 : struct super_block *get_active_super(struct block_device *bdev)
831 : {
832 : struct super_block *sb;
833 :
834 0 : if (!bdev)
835 : return NULL;
836 :
837 : restart:
838 0 : spin_lock(&sb_lock);
839 0 : list_for_each_entry(sb, &super_blocks, s_list) {
840 0 : if (hlist_unhashed(&sb->s_instances))
841 0 : continue;
842 0 : if (sb->s_bdev == bdev) {
843 0 : if (!grab_super(sb))
844 : goto restart;
845 0 : up_write(&sb->s_umount);
846 0 : return sb;
847 : }
848 : }
849 0 : spin_unlock(&sb_lock);
850 0 : return NULL;
851 : }
852 :
853 0 : struct super_block *user_get_super(dev_t dev, bool excl)
854 : {
855 : struct super_block *sb;
856 :
857 : spin_lock(&sb_lock);
858 : rescan:
859 0 : list_for_each_entry(sb, &super_blocks, s_list) {
860 0 : if (hlist_unhashed(&sb->s_instances))
861 0 : continue;
862 0 : if (sb->s_dev == dev) {
863 0 : sb->s_count++;
864 0 : spin_unlock(&sb_lock);
865 0 : if (excl)
866 0 : down_write(&sb->s_umount);
867 : else
868 0 : down_read(&sb->s_umount);
869 : /* still alive? */
870 0 : if (sb->s_root && (sb->s_flags & SB_BORN))
871 : return sb;
872 0 : if (excl)
873 0 : up_write(&sb->s_umount);
874 : else
875 0 : up_read(&sb->s_umount);
876 : /* nope, got unmounted */
877 0 : spin_lock(&sb_lock);
878 0 : __put_super(sb);
879 0 : goto rescan;
880 : }
881 : }
882 0 : spin_unlock(&sb_lock);
883 0 : return NULL;
884 : }
885 :
886 : /**
887 : * reconfigure_super - asks filesystem to change superblock parameters
888 : * @fc: The superblock and configuration
889 : *
890 : * Alters the configuration parameters of a live superblock.
891 : */
892 0 : int reconfigure_super(struct fs_context *fc)
893 : {
894 0 : struct super_block *sb = fc->root->d_sb;
895 : int retval;
896 0 : bool remount_ro = false;
897 0 : bool force = fc->sb_flags & SB_FORCE;
898 :
899 0 : if (fc->sb_flags_mask & ~MS_RMT_MASK)
900 : return -EINVAL;
901 0 : if (sb->s_writers.frozen != SB_UNFROZEN)
902 : return -EBUSY;
903 :
904 0 : retval = security_sb_remount(sb, fc->security);
905 : if (retval)
906 : return retval;
907 :
908 0 : if (fc->sb_flags_mask & SB_RDONLY) {
909 : #ifdef CONFIG_BLOCK
910 0 : if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev &&
911 0 : bdev_read_only(sb->s_bdev))
912 : return -EACCES;
913 : #endif
914 :
915 0 : remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
916 : }
917 :
918 0 : if (remount_ro) {
919 0 : if (!hlist_empty(&sb->s_pins)) {
920 0 : up_write(&sb->s_umount);
921 0 : group_pin_kill(&sb->s_pins);
922 0 : down_write(&sb->s_umount);
923 0 : if (!sb->s_root)
924 : return 0;
925 0 : if (sb->s_writers.frozen != SB_UNFROZEN)
926 : return -EBUSY;
927 0 : remount_ro = !sb_rdonly(sb);
928 : }
929 : }
930 0 : shrink_dcache_sb(sb);
931 :
932 : /* If we are reconfiguring to RDONLY and current sb is read/write,
933 : * make sure there are no files open for writing.
934 : */
935 0 : if (remount_ro) {
936 0 : if (force) {
937 0 : sb->s_readonly_remount = 1;
938 0 : smp_wmb();
939 : } else {
940 0 : retval = sb_prepare_remount_readonly(sb);
941 0 : if (retval)
942 : return retval;
943 : }
944 : }
945 :
946 0 : if (fc->ops->reconfigure) {
947 0 : retval = fc->ops->reconfigure(fc);
948 0 : if (retval) {
949 0 : if (!force)
950 : goto cancel_readonly;
951 : /* If forced remount, go ahead despite any errors */
952 0 : WARN(1, "forced remount of a %s fs returned %i\n",
953 : sb->s_type->name, retval);
954 : }
955 : }
956 :
957 0 : WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
958 : (fc->sb_flags & fc->sb_flags_mask)));
959 : /* Needs to be ordered wrt mnt_is_readonly() */
960 0 : smp_wmb();
961 0 : sb->s_readonly_remount = 0;
962 :
963 : /*
964 : * Some filesystems modify their metadata via some other path than the
965 : * bdev buffer cache (eg. use a private mapping, or directories in
966 : * pagecache, etc). Also file data modifications go via their own
967 : * mappings. So If we try to mount readonly then copy the filesystem
968 : * from bdev, we could get stale data, so invalidate it to give a best
969 : * effort at coherency.
970 : */
971 0 : if (remount_ro && sb->s_bdev)
972 0 : invalidate_bdev(sb->s_bdev);
973 : return 0;
974 :
975 : cancel_readonly:
976 0 : sb->s_readonly_remount = 0;
977 0 : return retval;
978 : }
979 :
980 0 : static void do_emergency_remount_callback(struct super_block *sb)
981 : {
982 0 : down_write(&sb->s_umount);
983 0 : if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
984 0 : !sb_rdonly(sb)) {
985 : struct fs_context *fc;
986 :
987 0 : fc = fs_context_for_reconfigure(sb->s_root,
988 : SB_RDONLY | SB_FORCE, SB_RDONLY);
989 0 : if (!IS_ERR(fc)) {
990 0 : if (parse_monolithic_mount_data(fc, NULL) == 0)
991 0 : (void)reconfigure_super(fc);
992 0 : put_fs_context(fc);
993 : }
994 : }
995 0 : up_write(&sb->s_umount);
996 0 : }
997 :
998 0 : static void do_emergency_remount(struct work_struct *work)
999 : {
1000 0 : __iterate_supers(do_emergency_remount_callback);
1001 0 : kfree(work);
1002 0 : printk("Emergency Remount complete\n");
1003 0 : }
1004 :
1005 0 : void emergency_remount(void)
1006 : {
1007 : struct work_struct *work;
1008 :
1009 0 : work = kmalloc(sizeof(*work), GFP_ATOMIC);
1010 0 : if (work) {
1011 0 : INIT_WORK(work, do_emergency_remount);
1012 : schedule_work(work);
1013 : }
1014 0 : }
1015 :
1016 0 : static void do_thaw_all_callback(struct super_block *sb)
1017 : {
1018 0 : down_write(&sb->s_umount);
1019 0 : if (sb->s_root && sb->s_flags & SB_BORN) {
1020 0 : emergency_thaw_bdev(sb);
1021 0 : thaw_super_locked(sb);
1022 : } else {
1023 0 : up_write(&sb->s_umount);
1024 : }
1025 0 : }
1026 :
1027 0 : static void do_thaw_all(struct work_struct *work)
1028 : {
1029 0 : __iterate_supers(do_thaw_all_callback);
1030 0 : kfree(work);
1031 0 : printk(KERN_WARNING "Emergency Thaw complete\n");
1032 0 : }
1033 :
1034 : /**
1035 : * emergency_thaw_all -- forcibly thaw every frozen filesystem
1036 : *
1037 : * Used for emergency unfreeze of all filesystems via SysRq
1038 : */
1039 0 : void emergency_thaw_all(void)
1040 : {
1041 : struct work_struct *work;
1042 :
1043 0 : work = kmalloc(sizeof(*work), GFP_ATOMIC);
1044 0 : if (work) {
1045 0 : INIT_WORK(work, do_thaw_all);
1046 : schedule_work(work);
1047 : }
1048 0 : }
1049 :
1050 : static DEFINE_IDA(unnamed_dev_ida);
1051 :
1052 : /**
1053 : * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1054 : * @p: Pointer to a dev_t.
1055 : *
1056 : * Filesystems which don't use real block devices can call this function
1057 : * to allocate a virtual block device.
1058 : *
1059 : * Context: Any context. Frequently called while holding sb_lock.
1060 : * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1061 : * or -ENOMEM if memory allocation failed.
1062 : */
1063 27 : int get_anon_bdev(dev_t *p)
1064 : {
1065 : int dev;
1066 :
1067 : /*
1068 : * Many userspace utilities consider an FSID of 0 invalid.
1069 : * Always return at least 1 from get_anon_bdev.
1070 : */
1071 27 : dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1072 : GFP_ATOMIC);
1073 27 : if (dev == -ENOSPC)
1074 0 : dev = -EMFILE;
1075 27 : if (dev < 0)
1076 : return dev;
1077 :
1078 27 : *p = MKDEV(0, dev);
1079 27 : return 0;
1080 : }
1081 : EXPORT_SYMBOL(get_anon_bdev);
1082 :
1083 0 : void free_anon_bdev(dev_t dev)
1084 : {
1085 17 : ida_free(&unnamed_dev_ida, MINOR(dev));
1086 0 : }
1087 : EXPORT_SYMBOL(free_anon_bdev);
1088 :
1089 0 : int set_anon_super(struct super_block *s, void *data)
1090 : {
1091 27 : return get_anon_bdev(&s->s_dev);
1092 : }
1093 : EXPORT_SYMBOL(set_anon_super);
1094 :
1095 17 : void kill_anon_super(struct super_block *sb)
1096 : {
1097 17 : dev_t dev = sb->s_dev;
1098 17 : generic_shutdown_super(sb);
1099 17 : free_anon_bdev(dev);
1100 17 : }
1101 : EXPORT_SYMBOL(kill_anon_super);
1102 :
1103 0 : void kill_litter_super(struct super_block *sb)
1104 : {
1105 0 : if (sb->s_root)
1106 0 : d_genocide(sb->s_root);
1107 0 : kill_anon_super(sb);
1108 0 : }
1109 : EXPORT_SYMBOL(kill_litter_super);
1110 :
1111 27 : int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1112 : {
1113 27 : return set_anon_super(sb, NULL);
1114 : }
1115 : EXPORT_SYMBOL(set_anon_super_fc);
1116 :
1117 0 : static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1118 : {
1119 0 : return sb->s_fs_info == fc->s_fs_info;
1120 : }
1121 :
1122 0 : static int test_single_super(struct super_block *s, struct fs_context *fc)
1123 : {
1124 0 : return 1;
1125 : }
1126 :
1127 27 : static int vfs_get_super(struct fs_context *fc, bool reconf,
1128 : int (*test)(struct super_block *, struct fs_context *),
1129 : int (*fill_super)(struct super_block *sb,
1130 : struct fs_context *fc))
1131 : {
1132 : struct super_block *sb;
1133 : int err;
1134 :
1135 27 : sb = sget_fc(fc, test, set_anon_super_fc);
1136 27 : if (IS_ERR(sb))
1137 0 : return PTR_ERR(sb);
1138 :
1139 27 : if (!sb->s_root) {
1140 27 : err = fill_super(sb, fc);
1141 27 : if (err)
1142 : goto error;
1143 :
1144 27 : sb->s_flags |= SB_ACTIVE;
1145 54 : fc->root = dget(sb->s_root);
1146 : } else {
1147 0 : fc->root = dget(sb->s_root);
1148 0 : if (reconf) {
1149 0 : err = reconfigure_super(fc);
1150 0 : if (err < 0) {
1151 0 : dput(fc->root);
1152 0 : fc->root = NULL;
1153 0 : goto error;
1154 : }
1155 : }
1156 : }
1157 :
1158 : return 0;
1159 :
1160 : error:
1161 0 : deactivate_locked_super(sb);
1162 0 : return err;
1163 : }
1164 :
1165 27 : int get_tree_nodev(struct fs_context *fc,
1166 : int (*fill_super)(struct super_block *sb,
1167 : struct fs_context *fc))
1168 : {
1169 27 : return vfs_get_super(fc, false, NULL, fill_super);
1170 : }
1171 : EXPORT_SYMBOL(get_tree_nodev);
1172 :
1173 0 : int get_tree_single(struct fs_context *fc,
1174 : int (*fill_super)(struct super_block *sb,
1175 : struct fs_context *fc))
1176 : {
1177 0 : return vfs_get_super(fc, false, test_single_super, fill_super);
1178 : }
1179 : EXPORT_SYMBOL(get_tree_single);
1180 :
1181 0 : int get_tree_single_reconf(struct fs_context *fc,
1182 : int (*fill_super)(struct super_block *sb,
1183 : struct fs_context *fc))
1184 : {
1185 0 : return vfs_get_super(fc, true, test_single_super, fill_super);
1186 : }
1187 : EXPORT_SYMBOL(get_tree_single_reconf);
1188 :
1189 0 : int get_tree_keyed(struct fs_context *fc,
1190 : int (*fill_super)(struct super_block *sb,
1191 : struct fs_context *fc),
1192 : void *key)
1193 : {
1194 0 : fc->s_fs_info = key;
1195 0 : return vfs_get_super(fc, false, test_keyed_super, fill_super);
1196 : }
1197 : EXPORT_SYMBOL(get_tree_keyed);
1198 :
1199 : #ifdef CONFIG_BLOCK
1200 :
1201 0 : static int set_bdev_super(struct super_block *s, void *data)
1202 : {
1203 0 : s->s_bdev = data;
1204 0 : s->s_dev = s->s_bdev->bd_dev;
1205 0 : s->s_bdi = bdi_get(s->s_bdev->bd_disk->bdi);
1206 :
1207 0 : if (bdev_stable_writes(s->s_bdev))
1208 0 : s->s_iflags |= SB_I_STABLE_WRITES;
1209 0 : return 0;
1210 : }
1211 :
1212 0 : static int set_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1213 : {
1214 0 : return set_bdev_super(s, fc->sget_key);
1215 : }
1216 :
1217 0 : static int test_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1218 : {
1219 0 : return !(s->s_iflags & SB_I_RETIRED) && s->s_bdev == fc->sget_key;
1220 : }
1221 :
1222 : /**
1223 : * get_tree_bdev - Get a superblock based on a single block device
1224 : * @fc: The filesystem context holding the parameters
1225 : * @fill_super: Helper to initialise a new superblock
1226 : */
1227 0 : int get_tree_bdev(struct fs_context *fc,
1228 : int (*fill_super)(struct super_block *,
1229 : struct fs_context *))
1230 : {
1231 : struct block_device *bdev;
1232 : struct super_block *s;
1233 0 : fmode_t mode = FMODE_READ | FMODE_EXCL;
1234 0 : int error = 0;
1235 :
1236 0 : if (!(fc->sb_flags & SB_RDONLY))
1237 0 : mode |= FMODE_WRITE;
1238 :
1239 0 : if (!fc->source)
1240 0 : return invalf(fc, "No source specified");
1241 :
1242 0 : bdev = blkdev_get_by_path(fc->source, mode, fc->fs_type);
1243 0 : if (IS_ERR(bdev)) {
1244 0 : errorf(fc, "%s: Can't open blockdev", fc->source);
1245 0 : return PTR_ERR(bdev);
1246 : }
1247 :
1248 : /* Once the superblock is inserted into the list by sget_fc(), s_umount
1249 : * will protect the lockfs code from trying to start a snapshot while
1250 : * we are mounting
1251 : */
1252 0 : mutex_lock(&bdev->bd_fsfreeze_mutex);
1253 0 : if (bdev->bd_fsfreeze_count > 0) {
1254 0 : mutex_unlock(&bdev->bd_fsfreeze_mutex);
1255 0 : warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
1256 0 : blkdev_put(bdev, mode);
1257 0 : return -EBUSY;
1258 : }
1259 :
1260 0 : fc->sb_flags |= SB_NOSEC;
1261 0 : fc->sget_key = bdev;
1262 0 : s = sget_fc(fc, test_bdev_super_fc, set_bdev_super_fc);
1263 0 : mutex_unlock(&bdev->bd_fsfreeze_mutex);
1264 0 : if (IS_ERR(s)) {
1265 0 : blkdev_put(bdev, mode);
1266 0 : return PTR_ERR(s);
1267 : }
1268 :
1269 0 : if (s->s_root) {
1270 : /* Don't summarily change the RO/RW state. */
1271 0 : if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1272 0 : warnf(fc, "%pg: Can't mount, would change RO state", bdev);
1273 0 : deactivate_locked_super(s);
1274 0 : blkdev_put(bdev, mode);
1275 0 : return -EBUSY;
1276 : }
1277 :
1278 : /*
1279 : * s_umount nests inside open_mutex during
1280 : * __invalidate_device(). blkdev_put() acquires
1281 : * open_mutex and can't be called under s_umount. Drop
1282 : * s_umount temporarily. This is safe as we're
1283 : * holding an active reference.
1284 : */
1285 0 : up_write(&s->s_umount);
1286 0 : blkdev_put(bdev, mode);
1287 0 : down_write(&s->s_umount);
1288 : } else {
1289 0 : s->s_mode = mode;
1290 0 : snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1291 0 : shrinker_debugfs_rename(&s->s_shrink, "sb-%s:%s",
1292 : fc->fs_type->name, s->s_id);
1293 0 : sb_set_blocksize(s, block_size(bdev));
1294 0 : error = fill_super(s, fc);
1295 0 : if (error) {
1296 0 : deactivate_locked_super(s);
1297 0 : return error;
1298 : }
1299 :
1300 0 : s->s_flags |= SB_ACTIVE;
1301 0 : bdev->bd_super = s;
1302 : }
1303 :
1304 0 : BUG_ON(fc->root);
1305 0 : fc->root = dget(s->s_root);
1306 0 : return 0;
1307 : }
1308 : EXPORT_SYMBOL(get_tree_bdev);
1309 :
1310 0 : static int test_bdev_super(struct super_block *s, void *data)
1311 : {
1312 0 : return !(s->s_iflags & SB_I_RETIRED) && (void *)s->s_bdev == data;
1313 : }
1314 :
1315 0 : struct dentry *mount_bdev(struct file_system_type *fs_type,
1316 : int flags, const char *dev_name, void *data,
1317 : int (*fill_super)(struct super_block *, void *, int))
1318 : {
1319 : struct block_device *bdev;
1320 : struct super_block *s;
1321 0 : fmode_t mode = FMODE_READ | FMODE_EXCL;
1322 0 : int error = 0;
1323 :
1324 0 : if (!(flags & SB_RDONLY))
1325 0 : mode |= FMODE_WRITE;
1326 :
1327 0 : bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1328 0 : if (IS_ERR(bdev))
1329 : return ERR_CAST(bdev);
1330 :
1331 : /*
1332 : * once the super is inserted into the list by sget, s_umount
1333 : * will protect the lockfs code from trying to start a snapshot
1334 : * while we are mounting
1335 : */
1336 0 : mutex_lock(&bdev->bd_fsfreeze_mutex);
1337 0 : if (bdev->bd_fsfreeze_count > 0) {
1338 0 : mutex_unlock(&bdev->bd_fsfreeze_mutex);
1339 0 : error = -EBUSY;
1340 0 : goto error_bdev;
1341 : }
1342 0 : s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
1343 : bdev);
1344 0 : mutex_unlock(&bdev->bd_fsfreeze_mutex);
1345 0 : if (IS_ERR(s))
1346 : goto error_s;
1347 :
1348 0 : if (s->s_root) {
1349 0 : if ((flags ^ s->s_flags) & SB_RDONLY) {
1350 0 : deactivate_locked_super(s);
1351 0 : error = -EBUSY;
1352 0 : goto error_bdev;
1353 : }
1354 :
1355 : /*
1356 : * s_umount nests inside open_mutex during
1357 : * __invalidate_device(). blkdev_put() acquires
1358 : * open_mutex and can't be called under s_umount. Drop
1359 : * s_umount temporarily. This is safe as we're
1360 : * holding an active reference.
1361 : */
1362 0 : up_write(&s->s_umount);
1363 0 : blkdev_put(bdev, mode);
1364 0 : down_write(&s->s_umount);
1365 : } else {
1366 0 : s->s_mode = mode;
1367 0 : snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1368 0 : shrinker_debugfs_rename(&s->s_shrink, "sb-%s:%s",
1369 : fs_type->name, s->s_id);
1370 0 : sb_set_blocksize(s, block_size(bdev));
1371 0 : error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1372 0 : if (error) {
1373 0 : deactivate_locked_super(s);
1374 0 : goto error;
1375 : }
1376 :
1377 0 : s->s_flags |= SB_ACTIVE;
1378 0 : bdev->bd_super = s;
1379 : }
1380 :
1381 0 : return dget(s->s_root);
1382 :
1383 : error_s:
1384 0 : error = PTR_ERR(s);
1385 : error_bdev:
1386 0 : blkdev_put(bdev, mode);
1387 : error:
1388 0 : return ERR_PTR(error);
1389 : }
1390 : EXPORT_SYMBOL(mount_bdev);
1391 :
1392 0 : void kill_block_super(struct super_block *sb)
1393 : {
1394 0 : struct block_device *bdev = sb->s_bdev;
1395 0 : fmode_t mode = sb->s_mode;
1396 :
1397 0 : bdev->bd_super = NULL;
1398 0 : generic_shutdown_super(sb);
1399 0 : sync_blockdev(bdev);
1400 0 : WARN_ON_ONCE(!(mode & FMODE_EXCL));
1401 0 : blkdev_put(bdev, mode | FMODE_EXCL);
1402 0 : }
1403 :
1404 : EXPORT_SYMBOL(kill_block_super);
1405 : #endif
1406 :
1407 0 : struct dentry *mount_nodev(struct file_system_type *fs_type,
1408 : int flags, void *data,
1409 : int (*fill_super)(struct super_block *, void *, int))
1410 : {
1411 : int error;
1412 0 : struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1413 :
1414 0 : if (IS_ERR(s))
1415 : return ERR_CAST(s);
1416 :
1417 0 : error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1418 0 : if (error) {
1419 0 : deactivate_locked_super(s);
1420 0 : return ERR_PTR(error);
1421 : }
1422 0 : s->s_flags |= SB_ACTIVE;
1423 0 : return dget(s->s_root);
1424 : }
1425 : EXPORT_SYMBOL(mount_nodev);
1426 :
1427 0 : int reconfigure_single(struct super_block *s,
1428 : int flags, void *data)
1429 : {
1430 : struct fs_context *fc;
1431 : int ret;
1432 :
1433 : /* The caller really need to be passing fc down into mount_single(),
1434 : * then a chunk of this can be removed. [Bollocks -- AV]
1435 : * Better yet, reconfiguration shouldn't happen, but rather the second
1436 : * mount should be rejected if the parameters are not compatible.
1437 : */
1438 0 : fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1439 0 : if (IS_ERR(fc))
1440 0 : return PTR_ERR(fc);
1441 :
1442 0 : ret = parse_monolithic_mount_data(fc, data);
1443 0 : if (ret < 0)
1444 : goto out;
1445 :
1446 0 : ret = reconfigure_super(fc);
1447 : out:
1448 0 : put_fs_context(fc);
1449 0 : return ret;
1450 : }
1451 :
1452 0 : static int compare_single(struct super_block *s, void *p)
1453 : {
1454 0 : return 1;
1455 : }
1456 :
1457 0 : struct dentry *mount_single(struct file_system_type *fs_type,
1458 : int flags, void *data,
1459 : int (*fill_super)(struct super_block *, void *, int))
1460 : {
1461 : struct super_block *s;
1462 : int error;
1463 :
1464 0 : s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1465 0 : if (IS_ERR(s))
1466 : return ERR_CAST(s);
1467 0 : if (!s->s_root) {
1468 0 : error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1469 0 : if (!error)
1470 0 : s->s_flags |= SB_ACTIVE;
1471 : } else {
1472 0 : error = reconfigure_single(s, flags, data);
1473 : }
1474 0 : if (unlikely(error)) {
1475 0 : deactivate_locked_super(s);
1476 0 : return ERR_PTR(error);
1477 : }
1478 0 : return dget(s->s_root);
1479 : }
1480 : EXPORT_SYMBOL(mount_single);
1481 :
1482 : /**
1483 : * vfs_get_tree - Get the mountable root
1484 : * @fc: The superblock configuration context.
1485 : *
1486 : * The filesystem is invoked to get or create a superblock which can then later
1487 : * be used for mounting. The filesystem places a pointer to the root to be
1488 : * used for mounting in @fc->root.
1489 : */
1490 27 : int vfs_get_tree(struct fs_context *fc)
1491 : {
1492 : struct super_block *sb;
1493 : int error;
1494 :
1495 27 : if (fc->root)
1496 : return -EBUSY;
1497 :
1498 : /* Get the mountable root in fc->root, with a ref on the root and a ref
1499 : * on the superblock.
1500 : */
1501 27 : error = fc->ops->get_tree(fc);
1502 27 : if (error < 0)
1503 : return error;
1504 :
1505 27 : if (!fc->root) {
1506 0 : pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1507 : fc->fs_type->name);
1508 : /* We don't know what the locking state of the superblock is -
1509 : * if there is a superblock.
1510 : */
1511 0 : BUG();
1512 : }
1513 :
1514 27 : sb = fc->root->d_sb;
1515 27 : WARN_ON(!sb->s_bdi);
1516 :
1517 : /*
1518 : * Write barrier is for super_cache_count(). We place it before setting
1519 : * SB_BORN as the data dependency between the two functions is the
1520 : * superblock structure contents that we just set up, not the SB_BORN
1521 : * flag.
1522 : */
1523 27 : smp_wmb();
1524 27 : sb->s_flags |= SB_BORN;
1525 :
1526 27 : error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
1527 : if (unlikely(error)) {
1528 : fc_drop_locked(fc);
1529 : return error;
1530 : }
1531 :
1532 : /*
1533 : * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1534 : * but s_maxbytes was an unsigned long long for many releases. Throw
1535 : * this warning for a little while to try and catch filesystems that
1536 : * violate this rule.
1537 : */
1538 27 : WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1539 : "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
1540 :
1541 : return 0;
1542 : }
1543 : EXPORT_SYMBOL(vfs_get_tree);
1544 :
1545 : /*
1546 : * Setup private BDI for given superblock. It gets automatically cleaned up
1547 : * in generic_shutdown_super().
1548 : */
1549 0 : int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1550 : {
1551 : struct backing_dev_info *bdi;
1552 : int err;
1553 : va_list args;
1554 :
1555 0 : bdi = bdi_alloc(NUMA_NO_NODE);
1556 0 : if (!bdi)
1557 : return -ENOMEM;
1558 :
1559 0 : va_start(args, fmt);
1560 0 : err = bdi_register_va(bdi, fmt, args);
1561 0 : va_end(args);
1562 0 : if (err) {
1563 0 : bdi_put(bdi);
1564 0 : return err;
1565 : }
1566 0 : WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1567 0 : sb->s_bdi = bdi;
1568 0 : sb->s_iflags |= SB_I_PERSB_BDI;
1569 :
1570 0 : return 0;
1571 : }
1572 : EXPORT_SYMBOL(super_setup_bdi_name);
1573 :
1574 : /*
1575 : * Setup private BDI for given superblock. I gets automatically cleaned up
1576 : * in generic_shutdown_super().
1577 : */
1578 0 : int super_setup_bdi(struct super_block *sb)
1579 : {
1580 : static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1581 :
1582 0 : return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1583 : atomic_long_inc_return(&bdi_seq));
1584 : }
1585 : EXPORT_SYMBOL(super_setup_bdi);
1586 :
1587 : /**
1588 : * sb_wait_write - wait until all writers to given file system finish
1589 : * @sb: the super for which we wait
1590 : * @level: type of writers we wait for (normal vs page fault)
1591 : *
1592 : * This function waits until there are no writers of given type to given file
1593 : * system.
1594 : */
1595 : static void sb_wait_write(struct super_block *sb, int level)
1596 : {
1597 0 : percpu_down_write(sb->s_writers.rw_sem + level-1);
1598 : }
1599 :
1600 : /*
1601 : * We are going to return to userspace and forget about these locks, the
1602 : * ownership goes to the caller of thaw_super() which does unlock().
1603 : */
1604 : static void lockdep_sb_freeze_release(struct super_block *sb)
1605 : {
1606 : int level;
1607 :
1608 0 : for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1609 : percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1610 : }
1611 :
1612 : /*
1613 : * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1614 : */
1615 : static void lockdep_sb_freeze_acquire(struct super_block *sb)
1616 : {
1617 : int level;
1618 :
1619 0 : for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1620 : percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1621 : }
1622 :
1623 : static void sb_freeze_unlock(struct super_block *sb, int level)
1624 : {
1625 0 : for (level--; level >= 0; level--)
1626 0 : percpu_up_write(sb->s_writers.rw_sem + level);
1627 : }
1628 :
1629 : /**
1630 : * freeze_super - lock the filesystem and force it into a consistent state
1631 : * @sb: the super to lock
1632 : *
1633 : * Syncs the super to make sure the filesystem is consistent and calls the fs's
1634 : * freeze_fs. Subsequent calls to this without first thawing the fs will return
1635 : * -EBUSY.
1636 : *
1637 : * During this function, sb->s_writers.frozen goes through these values:
1638 : *
1639 : * SB_UNFROZEN: File system is normal, all writes progress as usual.
1640 : *
1641 : * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1642 : * writes should be blocked, though page faults are still allowed. We wait for
1643 : * all writes to complete and then proceed to the next stage.
1644 : *
1645 : * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1646 : * but internal fs threads can still modify the filesystem (although they
1647 : * should not dirty new pages or inodes), writeback can run etc. After waiting
1648 : * for all running page faults we sync the filesystem which will clean all
1649 : * dirty pages and inodes (no new dirty pages or inodes can be created when
1650 : * sync is running).
1651 : *
1652 : * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1653 : * modification are blocked (e.g. XFS preallocation truncation on inode
1654 : * reclaim). This is usually implemented by blocking new transactions for
1655 : * filesystems that have them and need this additional guard. After all
1656 : * internal writers are finished we call ->freeze_fs() to finish filesystem
1657 : * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1658 : * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1659 : *
1660 : * sb->s_writers.frozen is protected by sb->s_umount.
1661 : */
1662 0 : int freeze_super(struct super_block *sb)
1663 : {
1664 : int ret;
1665 :
1666 0 : atomic_inc(&sb->s_active);
1667 0 : down_write(&sb->s_umount);
1668 0 : if (sb->s_writers.frozen != SB_UNFROZEN) {
1669 0 : deactivate_locked_super(sb);
1670 0 : return -EBUSY;
1671 : }
1672 :
1673 0 : if (!(sb->s_flags & SB_BORN)) {
1674 0 : up_write(&sb->s_umount);
1675 0 : return 0; /* sic - it's "nothing to do" */
1676 : }
1677 :
1678 0 : if (sb_rdonly(sb)) {
1679 : /* Nothing to do really... */
1680 0 : sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1681 0 : up_write(&sb->s_umount);
1682 0 : return 0;
1683 : }
1684 :
1685 0 : sb->s_writers.frozen = SB_FREEZE_WRITE;
1686 : /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1687 0 : up_write(&sb->s_umount);
1688 0 : sb_wait_write(sb, SB_FREEZE_WRITE);
1689 0 : down_write(&sb->s_umount);
1690 :
1691 : /* Now we go and block page faults... */
1692 0 : sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1693 0 : sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1694 :
1695 : /* All writers are done so after syncing there won't be dirty data */
1696 0 : ret = sync_filesystem(sb);
1697 0 : if (ret) {
1698 0 : sb->s_writers.frozen = SB_UNFROZEN;
1699 0 : sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT);
1700 0 : wake_up(&sb->s_writers.wait_unfrozen);
1701 0 : deactivate_locked_super(sb);
1702 0 : return ret;
1703 : }
1704 :
1705 : /* Now wait for internal filesystem counter */
1706 0 : sb->s_writers.frozen = SB_FREEZE_FS;
1707 0 : sb_wait_write(sb, SB_FREEZE_FS);
1708 :
1709 0 : if (sb->s_op->freeze_fs) {
1710 0 : ret = sb->s_op->freeze_fs(sb);
1711 0 : if (ret) {
1712 0 : printk(KERN_ERR
1713 : "VFS:Filesystem freeze failed\n");
1714 0 : sb->s_writers.frozen = SB_UNFROZEN;
1715 0 : sb_freeze_unlock(sb, SB_FREEZE_FS);
1716 0 : wake_up(&sb->s_writers.wait_unfrozen);
1717 0 : deactivate_locked_super(sb);
1718 0 : return ret;
1719 : }
1720 : }
1721 : /*
1722 : * For debugging purposes so that fs can warn if it sees write activity
1723 : * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1724 : */
1725 0 : sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1726 0 : lockdep_sb_freeze_release(sb);
1727 0 : up_write(&sb->s_umount);
1728 0 : return 0;
1729 : }
1730 : EXPORT_SYMBOL(freeze_super);
1731 :
1732 0 : static int thaw_super_locked(struct super_block *sb)
1733 : {
1734 : int error;
1735 :
1736 0 : if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1737 0 : up_write(&sb->s_umount);
1738 0 : return -EINVAL;
1739 : }
1740 :
1741 0 : if (sb_rdonly(sb)) {
1742 0 : sb->s_writers.frozen = SB_UNFROZEN;
1743 0 : goto out;
1744 : }
1745 :
1746 0 : lockdep_sb_freeze_acquire(sb);
1747 :
1748 0 : if (sb->s_op->unfreeze_fs) {
1749 0 : error = sb->s_op->unfreeze_fs(sb);
1750 0 : if (error) {
1751 0 : printk(KERN_ERR
1752 : "VFS:Filesystem thaw failed\n");
1753 0 : lockdep_sb_freeze_release(sb);
1754 0 : up_write(&sb->s_umount);
1755 0 : return error;
1756 : }
1757 : }
1758 :
1759 0 : sb->s_writers.frozen = SB_UNFROZEN;
1760 : sb_freeze_unlock(sb, SB_FREEZE_FS);
1761 : out:
1762 0 : wake_up(&sb->s_writers.wait_unfrozen);
1763 0 : deactivate_locked_super(sb);
1764 0 : return 0;
1765 : }
1766 :
1767 : /**
1768 : * thaw_super -- unlock filesystem
1769 : * @sb: the super to thaw
1770 : *
1771 : * Unlocks the filesystem and marks it writeable again after freeze_super().
1772 : */
1773 0 : int thaw_super(struct super_block *sb)
1774 : {
1775 0 : down_write(&sb->s_umount);
1776 0 : return thaw_super_locked(sb);
1777 : }
1778 : EXPORT_SYMBOL(thaw_super);
1779 :
1780 : /*
1781 : * Create workqueue for deferred direct IO completions. We allocate the
1782 : * workqueue when it's first needed. This avoids creating workqueue for
1783 : * filesystems that don't need it and also allows us to create the workqueue
1784 : * late enough so the we can include s_id in the name of the workqueue.
1785 : */
1786 0 : int sb_init_dio_done_wq(struct super_block *sb)
1787 : {
1788 : struct workqueue_struct *old;
1789 0 : struct workqueue_struct *wq = alloc_workqueue("dio/%s",
1790 : WQ_MEM_RECLAIM, 0,
1791 0 : sb->s_id);
1792 0 : if (!wq)
1793 : return -ENOMEM;
1794 : /*
1795 : * This has to be atomic as more DIOs can race to create the workqueue
1796 : */
1797 0 : old = cmpxchg(&sb->s_dio_done_wq, NULL, wq);
1798 : /* Someone created workqueue before us? Free ours... */
1799 0 : if (old)
1800 0 : destroy_workqueue(wq);
1801 : return 0;
1802 : }
|
