~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/super.c

Version: ~ [ linux-5.18-rc6 ] ~ [ linux-5.17.6 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.38 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.114 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.192 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.241 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.277 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.312 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0
  2 /*
  3  *  linux/fs/super.c
  4  *
  5  *  Copyright (C) 1991, 1992  Linus Torvalds
  6  *
  7  *  super.c contains code to handle: - mount structures
  8  *                                   - super-block tables
  9  *                                   - filesystem drivers list
 10  *                                   - mount system call
 11  *                                   - umount system call
 12  *                                   - ustat system call
 13  *
 14  * GK 2/5/95  -  Changed to support mounting the root fs via NFS
 15  *
 16  *  Added kerneld support: Jacques Gelinas and Bjorn Ekwall
 17  *  Added change_root: Werner Almesberger & Hans Lermen, Feb '96
 18  *  Added options to /proc/mounts:
 19  *    Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
 20  *  Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
 21  *  Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
 22  */
 23 
 24 #include <linux/export.h>
 25 #include <linux/slab.h>
 26 #include <linux/blkdev.h>
 27 #include <linux/mount.h>
 28 #include <linux/security.h>
 29 #include <linux/writeback.h>            /* for the emergency remount stuff */
 30 #include <linux/idr.h>
 31 #include <linux/mutex.h>
 32 #include <linux/backing-dev.h>
 33 #include <linux/rculist_bl.h>
 34 #include <linux/cleancache.h>
 35 #include <linux/fscrypt.h>
 36 #include <linux/fsnotify.h>
 37 #include <linux/lockdep.h>
 38 #include <linux/user_namespace.h>
 39 #include <linux/fs_context.h>
 40 #include <uapi/linux/mount.h>
 41 #include "internal.h"
 42 
 43 static int thaw_super_locked(struct super_block *sb);
 44 
 45 static LIST_HEAD(super_blocks);
 46 static DEFINE_SPINLOCK(sb_lock);
 47 
 48 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
 49         "sb_writers",
 50         "sb_pagefaults",
 51         "sb_internal",
 52 };
 53 
 54 /*
 55  * One thing we have to be careful of with a per-sb shrinker is that we don't
 56  * drop the last active reference to the superblock from within the shrinker.
 57  * If that happens we could trigger unregistering the shrinker from within the
 58  * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
 59  * take a passive reference to the superblock to avoid this from occurring.
 60  */
 61 static unsigned long super_cache_scan(struct shrinker *shrink,
 62                                       struct shrink_control *sc)
 63 {
 64         struct super_block *sb;
 65         long    fs_objects = 0;
 66         long    total_objects;
 67         long    freed = 0;
 68         long    dentries;
 69         long    inodes;
 70 
 71         sb = container_of(shrink, struct super_block, s_shrink);
 72 
 73         /*
 74          * Deadlock avoidance.  We may hold various FS locks, and we don't want
 75          * to recurse into the FS that called us in clear_inode() and friends..
 76          */
 77         if (!(sc->gfp_mask & __GFP_FS))
 78                 return SHRINK_STOP;
 79 
 80         if (!trylock_super(sb))
 81                 return SHRINK_STOP;
 82 
 83         if (sb->s_op->nr_cached_objects)
 84                 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
 85 
 86         inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
 87         dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
 88         total_objects = dentries + inodes + fs_objects + 1;
 89         if (!total_objects)
 90                 total_objects = 1;
 91 
 92         /* proportion the scan between the caches */
 93         dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
 94         inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
 95         fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
 96 
 97         /*
 98          * prune the dcache first as the icache is pinned by it, then
 99          * prune the icache, followed by the filesystem specific caches
100          *
101          * Ensure that we always scan at least one object - memcg kmem
102          * accounting uses this to fully empty the caches.
103          */
104         sc->nr_to_scan = dentries + 1;
105         freed = prune_dcache_sb(sb, sc);
106         sc->nr_to_scan = inodes + 1;
107         freed += prune_icache_sb(sb, sc);
108 
109         if (fs_objects) {
110                 sc->nr_to_scan = fs_objects + 1;
111                 freed += sb->s_op->free_cached_objects(sb, sc);
112         }
113 
114         up_read(&sb->s_umount);
115         return freed;
116 }
117 
118 static unsigned long super_cache_count(struct shrinker *shrink,
119                                        struct shrink_control *sc)
120 {
121         struct super_block *sb;
122         long    total_objects = 0;
123 
124         sb = container_of(shrink, struct super_block, s_shrink);
125 
126         /*
127          * We don't call trylock_super() here as it is a scalability bottleneck,
128          * so we're exposed to partial setup state. The shrinker rwsem does not
129          * protect filesystem operations backing list_lru_shrink_count() or
130          * s_op->nr_cached_objects(). Counts can change between
131          * super_cache_count and super_cache_scan, so we really don't need locks
132          * here.
133          *
134          * However, if we are currently mounting the superblock, the underlying
135          * filesystem might be in a state of partial construction and hence it
136          * is dangerous to access it.  trylock_super() uses a SB_BORN check to
137          * avoid this situation, so do the same here. The memory barrier is
138          * matched with the one in mount_fs() as we don't hold locks here.
139          */
140         if (!(sb->s_flags & SB_BORN))
141                 return 0;
142         smp_rmb();
143 
144         if (sb->s_op && sb->s_op->nr_cached_objects)
145                 total_objects = sb->s_op->nr_cached_objects(sb, sc);
146 
147         total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
148         total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
149 
150         if (!total_objects)
151                 return SHRINK_EMPTY;
152 
153         total_objects = vfs_pressure_ratio(total_objects);
154         return total_objects;
155 }
156 
157 static void destroy_super_work(struct work_struct *work)
158 {
159         struct super_block *s = container_of(work, struct super_block,
160                                                         destroy_work);
161         int i;
162 
163         for (i = 0; i < SB_FREEZE_LEVELS; i++)
164                 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
165         kfree(s);
166 }
167 
168 static void destroy_super_rcu(struct rcu_head *head)
169 {
170         struct super_block *s = container_of(head, struct super_block, rcu);
171         INIT_WORK(&s->destroy_work, destroy_super_work);
172         schedule_work(&s->destroy_work);
173 }
174 
175 /* Free a superblock that has never been seen by anyone */
176 static void destroy_unused_super(struct super_block *s)
177 {
178         if (!s)
179                 return;
180         up_write(&s->s_umount);
181         list_lru_destroy(&s->s_dentry_lru);
182         list_lru_destroy(&s->s_inode_lru);
183         security_sb_free(s);
184         put_user_ns(s->s_user_ns);
185         kfree(s->s_subtype);
186         free_prealloced_shrinker(&s->s_shrink);
187         /* no delays needed */
188         destroy_super_work(&s->destroy_work);
189 }
190 
191 /**
192  *      alloc_super     -       create new superblock
193  *      @type:  filesystem type superblock should belong to
194  *      @flags: the mount flags
195  *      @user_ns: User namespace for the super_block
196  *
197  *      Allocates and initializes a new &struct super_block.  alloc_super()
198  *      returns a pointer new superblock or %NULL if allocation had failed.
199  */
200 static struct super_block *alloc_super(struct file_system_type *type, int flags,
201                                        struct user_namespace *user_ns)
202 {
203         struct super_block *s = kzalloc(sizeof(struct super_block),  GFP_USER);
204         static const struct super_operations default_op;
205         int i;
206 
207         if (!s)
208                 return NULL;
209 
210         INIT_LIST_HEAD(&s->s_mounts);
211         s->s_user_ns = get_user_ns(user_ns);
212         init_rwsem(&s->s_umount);
213         lockdep_set_class(&s->s_umount, &type->s_umount_key);
214         /*
215          * sget() can have s_umount recursion.
216          *
217          * When it cannot find a suitable sb, it allocates a new
218          * one (this one), and tries again to find a suitable old
219          * one.
220          *
221          * In case that succeeds, it will acquire the s_umount
222          * lock of the old one. Since these are clearly distrinct
223          * locks, and this object isn't exposed yet, there's no
224          * risk of deadlocks.
225          *
226          * Annotate this by putting this lock in a different
227          * subclass.
228          */
229         down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
230 
231         if (security_sb_alloc(s))
232                 goto fail;
233 
234         for (i = 0; i < SB_FREEZE_LEVELS; i++) {
235                 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
236                                         sb_writers_name[i],
237                                         &type->s_writers_key[i]))
238                         goto fail;
239         }
240         init_waitqueue_head(&s->s_writers.wait_unfrozen);
241         s->s_bdi = &noop_backing_dev_info;
242         s->s_flags = flags;
243         if (s->s_user_ns != &init_user_ns)
244                 s->s_iflags |= SB_I_NODEV;
245         INIT_HLIST_NODE(&s->s_instances);
246         INIT_HLIST_BL_HEAD(&s->s_roots);
247         mutex_init(&s->s_sync_lock);
248         INIT_LIST_HEAD(&s->s_inodes);
249         spin_lock_init(&s->s_inode_list_lock);
250         INIT_LIST_HEAD(&s->s_inodes_wb);
251         spin_lock_init(&s->s_inode_wblist_lock);
252 
253         s->s_count = 1;
254         atomic_set(&s->s_active, 1);
255         mutex_init(&s->s_vfs_rename_mutex);
256         lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
257         init_rwsem(&s->s_dquot.dqio_sem);
258         s->s_maxbytes = MAX_NON_LFS;
259         s->s_op = &default_op;
260         s->s_time_gran = 1000000000;
261         s->s_time_min = TIME64_MIN;
262         s->s_time_max = TIME64_MAX;
263         s->cleancache_poolid = CLEANCACHE_NO_POOL;
264 
265         s->s_shrink.seeks = DEFAULT_SEEKS;
266         s->s_shrink.scan_objects = super_cache_scan;
267         s->s_shrink.count_objects = super_cache_count;
268         s->s_shrink.batch = 1024;
269         s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
270         if (prealloc_shrinker(&s->s_shrink))
271                 goto fail;
272         if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
273                 goto fail;
274         if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
275                 goto fail;
276         return s;
277 
278 fail:
279         destroy_unused_super(s);
280         return NULL;
281 }
282 
283 /* Superblock refcounting  */
284 
285 /*
286  * Drop a superblock's refcount.  The caller must hold sb_lock.
287  */
288 static void __put_super(struct super_block *s)
289 {
290         if (!--s->s_count) {
291                 list_del_init(&s->s_list);
292                 WARN_ON(s->s_dentry_lru.node);
293                 WARN_ON(s->s_inode_lru.node);
294                 WARN_ON(!list_empty(&s->s_mounts));
295                 security_sb_free(s);
296                 fscrypt_sb_free(s);
297                 put_user_ns(s->s_user_ns);
298                 kfree(s->s_subtype);
299                 call_rcu(&s->rcu, destroy_super_rcu);
300         }
301 }
302 
303 /**
304  *      put_super       -       drop a temporary reference to superblock
305  *      @sb: superblock in question
306  *
307  *      Drops a temporary reference, frees superblock if there's no
308  *      references left.
309  */
310 static void put_super(struct super_block *sb)
311 {
312         spin_lock(&sb_lock);
313         __put_super(sb);
314         spin_unlock(&sb_lock);
315 }
316 
317 
318 /**
319  *      deactivate_locked_super -       drop an active reference to superblock
320  *      @s: superblock to deactivate
321  *
322  *      Drops an active reference to superblock, converting it into a temporary
323  *      one if there is no other active references left.  In that case we
324  *      tell fs driver to shut it down and drop the temporary reference we
325  *      had just acquired.
326  *
327  *      Caller holds exclusive lock on superblock; that lock is released.
328  */
329 void deactivate_locked_super(struct super_block *s)
330 {
331         struct file_system_type *fs = s->s_type;
332         if (atomic_dec_and_test(&s->s_active)) {
333                 cleancache_invalidate_fs(s);
334                 unregister_shrinker(&s->s_shrink);
335                 fs->kill_sb(s);
336 
337                 /*
338                  * Since list_lru_destroy() may sleep, we cannot call it from
339                  * put_super(), where we hold the sb_lock. Therefore we destroy
340                  * the lru lists right now.
341                  */
342                 list_lru_destroy(&s->s_dentry_lru);
343                 list_lru_destroy(&s->s_inode_lru);
344 
345                 put_filesystem(fs);
346                 put_super(s);
347         } else {
348                 up_write(&s->s_umount);
349         }
350 }
351 
352 EXPORT_SYMBOL(deactivate_locked_super);
353 
354 /**
355  *      deactivate_super        -       drop an active reference to superblock
356  *      @s: superblock to deactivate
357  *
358  *      Variant of deactivate_locked_super(), except that superblock is *not*
359  *      locked by caller.  If we are going to drop the final active reference,
360  *      lock will be acquired prior to that.
361  */
362 void deactivate_super(struct super_block *s)
363 {
364         if (!atomic_add_unless(&s->s_active, -1, 1)) {
365                 down_write(&s->s_umount);
366                 deactivate_locked_super(s);
367         }
368 }
369 
370 EXPORT_SYMBOL(deactivate_super);
371 
372 /**
373  *      grab_super - acquire an active reference
374  *      @s: reference we are trying to make active
375  *
376  *      Tries to acquire an active reference.  grab_super() is used when we
377  *      had just found a superblock in super_blocks or fs_type->fs_supers
378  *      and want to turn it into a full-blown active reference.  grab_super()
379  *      is called with sb_lock held and drops it.  Returns 1 in case of
380  *      success, 0 if we had failed (superblock contents was already dead or
381  *      dying when grab_super() had been called).  Note that this is only
382  *      called for superblocks not in rundown mode (== ones still on ->fs_supers
383  *      of their type), so increment of ->s_count is OK here.
384  */
385 static int grab_super(struct super_block *s) __releases(sb_lock)
386 {
387         s->s_count++;
388         spin_unlock(&sb_lock);
389         down_write(&s->s_umount);
390         if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
391                 put_super(s);
392                 return 1;
393         }
394         up_write(&s->s_umount);
395         put_super(s);
396         return 0;
397 }
398 
399 /*
400  *      trylock_super - try to grab ->s_umount shared
401  *      @sb: reference we are trying to grab
402  *
403  *      Try to prevent fs shutdown.  This is used in places where we
404  *      cannot take an active reference but we need to ensure that the
405  *      filesystem is not shut down while we are working on it. It returns
406  *      false if we cannot acquire s_umount or if we lose the race and
407  *      filesystem already got into shutdown, and returns true with the s_umount
408  *      lock held in read mode in case of success. On successful return,
409  *      the caller must drop the s_umount lock when done.
410  *
411  *      Note that unlike get_super() et.al. this one does *not* bump ->s_count.
412  *      The reason why it's safe is that we are OK with doing trylock instead
413  *      of down_read().  There's a couple of places that are OK with that, but
414  *      it's very much not a general-purpose interface.
415  */
416 bool trylock_super(struct super_block *sb)
417 {
418         if (down_read_trylock(&sb->s_umount)) {
419                 if (!hlist_unhashed(&sb->s_instances) &&
420                     sb->s_root && (sb->s_flags & SB_BORN))
421                         return true;
422                 up_read(&sb->s_umount);
423         }
424 
425         return false;
426 }
427 
428 /**
429  *      generic_shutdown_super  -       common helper for ->kill_sb()
430  *      @sb: superblock to kill
431  *
432  *      generic_shutdown_super() does all fs-independent work on superblock
433  *      shutdown.  Typical ->kill_sb() should pick all fs-specific objects
434  *      that need destruction out of superblock, call generic_shutdown_super()
435  *      and release aforementioned objects.  Note: dentries and inodes _are_
436  *      taken care of and do not need specific handling.
437  *
438  *      Upon calling this function, the filesystem may no longer alter or
439  *      rearrange the set of dentries belonging to this super_block, nor may it
440  *      change the attachments of dentries to inodes.
441  */
442 void generic_shutdown_super(struct super_block *sb)
443 {
444         const struct super_operations *sop = sb->s_op;
445 
446         if (sb->s_root) {
447                 shrink_dcache_for_umount(sb);
448                 sync_filesystem(sb);
449                 sb->s_flags &= ~SB_ACTIVE;
450 
451                 cgroup_writeback_umount();
452 
453                 /* evict all inodes with zero refcount */
454                 evict_inodes(sb);
455                 /* only nonzero refcount inodes can have marks */
456                 fsnotify_sb_delete(sb);
457 
458                 if (sb->s_dio_done_wq) {
459                         destroy_workqueue(sb->s_dio_done_wq);
460                         sb->s_dio_done_wq = NULL;
461                 }
462 
463                 if (sop->put_super)
464                         sop->put_super(sb);
465 
466                 if (!list_empty(&sb->s_inodes)) {
467                         printk("VFS: Busy inodes after unmount of %s. "
468                            "Self-destruct in 5 seconds.  Have a nice day...\n",
469                            sb->s_id);
470                 }
471         }
472         spin_lock(&sb_lock);
473         /* should be initialized for __put_super_and_need_restart() */
474         hlist_del_init(&sb->s_instances);
475         spin_unlock(&sb_lock);
476         up_write(&sb->s_umount);
477         if (sb->s_bdi != &noop_backing_dev_info) {
478                 bdi_put(sb->s_bdi);
479                 sb->s_bdi = &noop_backing_dev_info;
480         }
481 }
482 
483 EXPORT_SYMBOL(generic_shutdown_super);
484 
485 bool mount_capable(struct fs_context *fc)
486 {
487         if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT))
488                 return capable(CAP_SYS_ADMIN);
489         else
490                 return ns_capable(fc->user_ns, CAP_SYS_ADMIN);
491 }
492 
493 /**
494  * sget_fc - Find or create a superblock
495  * @fc: Filesystem context.
496  * @test: Comparison callback
497  * @set: Setup callback
498  *
499  * Find or create a superblock using the parameters stored in the filesystem
500  * context and the two callback functions.
501  *
502  * If an extant superblock is matched, then that will be returned with an
503  * elevated reference count that the caller must transfer or discard.
504  *
505  * If no match is made, a new superblock will be allocated and basic
506  * initialisation will be performed (s_type, s_fs_info and s_id will be set and
507  * the set() callback will be invoked), the superblock will be published and it
508  * will be returned in a partially constructed state with SB_BORN and SB_ACTIVE
509  * as yet unset.
510  */
511 struct super_block *sget_fc(struct fs_context *fc,
512                             int (*test)(struct super_block *, struct fs_context *),
513                             int (*set)(struct super_block *, struct fs_context *))
514 {
515         struct super_block *s = NULL;
516         struct super_block *old;
517         struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
518         int err;
519 
520 retry:
521         spin_lock(&sb_lock);
522         if (test) {
523                 hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
524                         if (test(old, fc))
525                                 goto share_extant_sb;
526                 }
527         }
528         if (!s) {
529                 spin_unlock(&sb_lock);
530                 s = alloc_super(fc->fs_type, fc->sb_flags, user_ns);
531                 if (!s)
532                         return ERR_PTR(-ENOMEM);
533                 goto retry;
534         }
535 
536         s->s_fs_info = fc->s_fs_info;
537         err = set(s, fc);
538         if (err) {
539                 s->s_fs_info = NULL;
540                 spin_unlock(&sb_lock);
541                 destroy_unused_super(s);
542                 return ERR_PTR(err);
543         }
544         fc->s_fs_info = NULL;
545         s->s_type = fc->fs_type;
546         s->s_iflags |= fc->s_iflags;
547         strlcpy(s->s_id, s->s_type->name, sizeof(s->s_id));
548         list_add_tail(&s->s_list, &super_blocks);
549         hlist_add_head(&s->s_instances, &s->s_type->fs_supers);
550         spin_unlock(&sb_lock);
551         get_filesystem(s->s_type);
552         register_shrinker_prepared(&s->s_shrink);
553         return s;
554 
555 share_extant_sb:
556         if (user_ns != old->s_user_ns) {
557                 spin_unlock(&sb_lock);
558                 destroy_unused_super(s);
559                 return ERR_PTR(-EBUSY);
560         }
561         if (!grab_super(old))
562                 goto retry;
563         destroy_unused_super(s);
564         return old;
565 }
566 EXPORT_SYMBOL(sget_fc);
567 
568 /**
569  *      sget    -       find or create a superblock
570  *      @type:    filesystem type superblock should belong to
571  *      @test:    comparison callback
572  *      @set:     setup callback
573  *      @flags:   mount flags
574  *      @data:    argument to each of them
575  */
576 struct super_block *sget(struct file_system_type *type,
577                         int (*test)(struct super_block *,void *),
578                         int (*set)(struct super_block *,void *),
579                         int flags,
580                         void *data)
581 {
582         struct user_namespace *user_ns = current_user_ns();
583         struct super_block *s = NULL;
584         struct super_block *old;
585         int err;
586 
587         /* We don't yet pass the user namespace of the parent
588          * mount through to here so always use &init_user_ns
589          * until that changes.
590          */
591         if (flags & SB_SUBMOUNT)
592                 user_ns = &init_user_ns;
593 
594 retry:
595         spin_lock(&sb_lock);
596         if (test) {
597                 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
598                         if (!test(old, data))
599                                 continue;
600                         if (user_ns != old->s_user_ns) {
601                                 spin_unlock(&sb_lock);
602                                 destroy_unused_super(s);
603                                 return ERR_PTR(-EBUSY);
604                         }
605                         if (!grab_super(old))
606                                 goto retry;
607                         destroy_unused_super(s);
608                         return old;
609                 }
610         }
611         if (!s) {
612                 spin_unlock(&sb_lock);
613                 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
614                 if (!s)
615                         return ERR_PTR(-ENOMEM);
616                 goto retry;
617         }
618 
619         err = set(s, data);
620         if (err) {
621                 spin_unlock(&sb_lock);
622                 destroy_unused_super(s);
623                 return ERR_PTR(err);
624         }
625         s->s_type = type;
626         strlcpy(s->s_id, type->name, sizeof(s->s_id));
627         list_add_tail(&s->s_list, &super_blocks);
628         hlist_add_head(&s->s_instances, &type->fs_supers);
629         spin_unlock(&sb_lock);
630         get_filesystem(type);
631         register_shrinker_prepared(&s->s_shrink);
632         return s;
633 }
634 EXPORT_SYMBOL(sget);
635 
636 void drop_super(struct super_block *sb)
637 {
638         up_read(&sb->s_umount);
639         put_super(sb);
640 }
641 
642 EXPORT_SYMBOL(drop_super);
643 
644 void drop_super_exclusive(struct super_block *sb)
645 {
646         up_write(&sb->s_umount);
647         put_super(sb);
648 }
649 EXPORT_SYMBOL(drop_super_exclusive);
650 
651 static void __iterate_supers(void (*f)(struct super_block *))
652 {
653         struct super_block *sb, *p = NULL;
654 
655         spin_lock(&sb_lock);
656         list_for_each_entry(sb, &super_blocks, s_list) {
657                 if (hlist_unhashed(&sb->s_instances))
658                         continue;
659                 sb->s_count++;
660                 spin_unlock(&sb_lock);
661 
662                 f(sb);
663 
664                 spin_lock(&sb_lock);
665                 if (p)
666                         __put_super(p);
667                 p = sb;
668         }
669         if (p)
670                 __put_super(p);
671         spin_unlock(&sb_lock);
672 }
673 /**
674  *      iterate_supers - call function for all active superblocks
675  *      @f: function to call
676  *      @arg: argument to pass to it
677  *
678  *      Scans the superblock list and calls given function, passing it
679  *      locked superblock and given argument.
680  */
681 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
682 {
683         struct super_block *sb, *p = NULL;
684 
685         spin_lock(&sb_lock);
686         list_for_each_entry(sb, &super_blocks, s_list) {
687                 if (hlist_unhashed(&sb->s_instances))
688                         continue;
689                 sb->s_count++;
690                 spin_unlock(&sb_lock);
691 
692                 down_read(&sb->s_umount);
693                 if (sb->s_root && (sb->s_flags & SB_BORN))
694                         f(sb, arg);
695                 up_read(&sb->s_umount);
696 
697                 spin_lock(&sb_lock);
698                 if (p)
699                         __put_super(p);
700                 p = sb;
701         }
702         if (p)
703                 __put_super(p);
704         spin_unlock(&sb_lock);
705 }
706 
707 /**
708  *      iterate_supers_type - call function for superblocks of given type
709  *      @type: fs type
710  *      @f: function to call
711  *      @arg: argument to pass to it
712  *
713  *      Scans the superblock list and calls given function, passing it
714  *      locked superblock and given argument.
715  */
716 void iterate_supers_type(struct file_system_type *type,
717         void (*f)(struct super_block *, void *), void *arg)
718 {
719         struct super_block *sb, *p = NULL;
720 
721         spin_lock(&sb_lock);
722         hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
723                 sb->s_count++;
724                 spin_unlock(&sb_lock);
725 
726                 down_read(&sb->s_umount);
727                 if (sb->s_root && (sb->s_flags & SB_BORN))
728                         f(sb, arg);
729                 up_read(&sb->s_umount);
730 
731                 spin_lock(&sb_lock);
732                 if (p)
733                         __put_super(p);
734                 p = sb;
735         }
736         if (p)
737                 __put_super(p);
738         spin_unlock(&sb_lock);
739 }
740 
741 EXPORT_SYMBOL(iterate_supers_type);
742 
743 static struct super_block *__get_super(struct block_device *bdev, bool excl)
744 {
745         struct super_block *sb;
746 
747         if (!bdev)
748                 return NULL;
749 
750         spin_lock(&sb_lock);
751 rescan:
752         list_for_each_entry(sb, &super_blocks, s_list) {
753                 if (hlist_unhashed(&sb->s_instances))
754                         continue;
755                 if (sb->s_bdev == bdev) {
756                         sb->s_count++;
757                         spin_unlock(&sb_lock);
758                         if (!excl)
759                                 down_read(&sb->s_umount);
760                         else
761                                 down_write(&sb->s_umount);
762                         /* still alive? */
763                         if (sb->s_root && (sb->s_flags & SB_BORN))
764                                 return sb;
765                         if (!excl)
766                                 up_read(&sb->s_umount);
767                         else
768                                 up_write(&sb->s_umount);
769                         /* nope, got unmounted */
770                         spin_lock(&sb_lock);
771                         __put_super(sb);
772                         goto rescan;
773                 }
774         }
775         spin_unlock(&sb_lock);
776         return NULL;
777 }
778 
779 /**
780  *      get_super - get the superblock of a device
781  *      @bdev: device to get the superblock for
782  *
783  *      Scans the superblock list and finds the superblock of the file system
784  *      mounted on the device given. %NULL is returned if no match is found.
785  */
786 struct super_block *get_super(struct block_device *bdev)
787 {
788         return __get_super(bdev, false);
789 }
790 EXPORT_SYMBOL(get_super);
791 
792 static struct super_block *__get_super_thawed(struct block_device *bdev,
793                                               bool excl)
794 {
795         while (1) {
796                 struct super_block *s = __get_super(bdev, excl);
797                 if (!s || s->s_writers.frozen == SB_UNFROZEN)
798                         return s;
799                 if (!excl)
800                         up_read(&s->s_umount);
801                 else
802                         up_write(&s->s_umount);
803                 wait_event(s->s_writers.wait_unfrozen,
804                            s->s_writers.frozen == SB_UNFROZEN);
805                 put_super(s);
806         }
807 }
808 
809 /**
810  *      get_super_thawed - get thawed superblock of a device
811  *      @bdev: device to get the superblock for
812  *
813  *      Scans the superblock list and finds the superblock of the file system
814  *      mounted on the device. The superblock is returned once it is thawed
815  *      (or immediately if it was not frozen). %NULL is returned if no match
816  *      is found.
817  */
818 struct super_block *get_super_thawed(struct block_device *bdev)
819 {
820         return __get_super_thawed(bdev, false);
821 }
822 EXPORT_SYMBOL(get_super_thawed);
823 
824 /**
825  *      get_super_exclusive_thawed - get thawed superblock of a device
826  *      @bdev: device to get the superblock for
827  *
828  *      Scans the superblock list and finds the superblock of the file system
829  *      mounted on the device. The superblock is returned once it is thawed
830  *      (or immediately if it was not frozen) and s_umount semaphore is held
831  *      in exclusive mode. %NULL is returned if no match is found.
832  */
833 struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
834 {
835         return __get_super_thawed(bdev, true);
836 }
837 EXPORT_SYMBOL(get_super_exclusive_thawed);
838 
839 /**
840  * get_active_super - get an active reference to the superblock of a device
841  * @bdev: device to get the superblock for
842  *
843  * Scans the superblock list and finds the superblock of the file system
844  * mounted on the device given.  Returns the superblock with an active
845  * reference or %NULL if none was found.
846  */
847 struct super_block *get_active_super(struct block_device *bdev)
848 {
849         struct super_block *sb;
850 
851         if (!bdev)
852                 return NULL;
853 
854 restart:
855         spin_lock(&sb_lock);
856         list_for_each_entry(sb, &super_blocks, s_list) {
857                 if (hlist_unhashed(&sb->s_instances))
858                         continue;
859                 if (sb->s_bdev == bdev) {
860                         if (!grab_super(sb))
861                                 goto restart;
862                         up_write(&sb->s_umount);
863                         return sb;
864                 }
865         }
866         spin_unlock(&sb_lock);
867         return NULL;
868 }
869 
870 struct super_block *user_get_super(dev_t dev)
871 {
872         struct super_block *sb;
873 
874         spin_lock(&sb_lock);
875 rescan:
876         list_for_each_entry(sb, &super_blocks, s_list) {
877                 if (hlist_unhashed(&sb->s_instances))
878                         continue;
879                 if (sb->s_dev ==  dev) {
880                         sb->s_count++;
881                         spin_unlock(&sb_lock);
882                         down_read(&sb->s_umount);
883                         /* still alive? */
884                         if (sb->s_root && (sb->s_flags & SB_BORN))
885                                 return sb;
886                         up_read(&sb->s_umount);
887                         /* nope, got unmounted */
888                         spin_lock(&sb_lock);
889                         __put_super(sb);
890                         goto rescan;
891                 }
892         }
893         spin_unlock(&sb_lock);
894         return NULL;
895 }
896 
897 /**
898  * reconfigure_super - asks filesystem to change superblock parameters
899  * @fc: The superblock and configuration
900  *
901  * Alters the configuration parameters of a live superblock.
902  */
903 int reconfigure_super(struct fs_context *fc)
904 {
905         struct super_block *sb = fc->root->d_sb;
906         int retval;
907         bool remount_ro = false;
908         bool force = fc->sb_flags & SB_FORCE;
909 
910         if (fc->sb_flags_mask & ~MS_RMT_MASK)
911                 return -EINVAL;
912         if (sb->s_writers.frozen != SB_UNFROZEN)
913                 return -EBUSY;
914 
915         retval = security_sb_remount(sb, fc->security);
916         if (retval)
917                 return retval;
918 
919         if (fc->sb_flags_mask & SB_RDONLY) {
920 #ifdef CONFIG_BLOCK
921                 if (!(fc->sb_flags & SB_RDONLY) && bdev_read_only(sb->s_bdev))
922                         return -EACCES;
923 #endif
924 
925                 remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
926         }
927 
928         if (remount_ro) {
929                 if (!hlist_empty(&sb->s_pins)) {
930                         up_write(&sb->s_umount);
931                         group_pin_kill(&sb->s_pins);
932                         down_write(&sb->s_umount);
933                         if (!sb->s_root)
934                                 return 0;
935                         if (sb->s_writers.frozen != SB_UNFROZEN)
936                                 return -EBUSY;
937                         remount_ro = !sb_rdonly(sb);
938                 }
939         }
940         shrink_dcache_sb(sb);
941 
942         /* If we are reconfiguring to RDONLY and current sb is read/write,
943          * make sure there are no files open for writing.
944          */
945         if (remount_ro) {
946                 if (force) {
947                         sb->s_readonly_remount = 1;
948                         smp_wmb();
949                 } else {
950                         retval = sb_prepare_remount_readonly(sb);
951                         if (retval)
952                                 return retval;
953                 }
954         }
955 
956         if (fc->ops->reconfigure) {
957                 retval = fc->ops->reconfigure(fc);
958                 if (retval) {
959                         if (!force)
960                                 goto cancel_readonly;
961                         /* If forced remount, go ahead despite any errors */
962                         WARN(1, "forced remount of a %s fs returned %i\n",
963                              sb->s_type->name, retval);
964                 }
965         }
966 
967         WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
968                                  (fc->sb_flags & fc->sb_flags_mask)));
969         /* Needs to be ordered wrt mnt_is_readonly() */
970         smp_wmb();
971         sb->s_readonly_remount = 0;
972 
973         /*
974          * Some filesystems modify their metadata via some other path than the
975          * bdev buffer cache (eg. use a private mapping, or directories in
976          * pagecache, etc). Also file data modifications go via their own
977          * mappings. So If we try to mount readonly then copy the filesystem
978          * from bdev, we could get stale data, so invalidate it to give a best
979          * effort at coherency.
980          */
981         if (remount_ro && sb->s_bdev)
982                 invalidate_bdev(sb->s_bdev);
983         return 0;
984 
985 cancel_readonly:
986         sb->s_readonly_remount = 0;
987         return retval;
988 }
989 
990 static void do_emergency_remount_callback(struct super_block *sb)
991 {
992         down_write(&sb->s_umount);
993         if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
994             !sb_rdonly(sb)) {
995                 struct fs_context *fc;
996 
997                 fc = fs_context_for_reconfigure(sb->s_root,
998                                         SB_RDONLY | SB_FORCE, SB_RDONLY);
999                 if (!IS_ERR(fc)) {
1000                         if (parse_monolithic_mount_data(fc, NULL) == 0)
1001                                 (void)reconfigure_super(fc);
1002                         put_fs_context(fc);
1003                 }
1004         }
1005         up_write(&sb->s_umount);
1006 }
1007 
1008 static void do_emergency_remount(struct work_struct *work)
1009 {
1010         __iterate_supers(do_emergency_remount_callback);
1011         kfree(work);
1012         printk("Emergency Remount complete\n");
1013 }
1014 
1015 void emergency_remount(void)
1016 {
1017         struct work_struct *work;
1018 
1019         work = kmalloc(sizeof(*work), GFP_ATOMIC);
1020         if (work) {
1021                 INIT_WORK(work, do_emergency_remount);
1022                 schedule_work(work);
1023         }
1024 }
1025 
1026 static void do_thaw_all_callback(struct super_block *sb)
1027 {
1028         down_write(&sb->s_umount);
1029         if (sb->s_root && sb->s_flags & SB_BORN) {
1030                 emergency_thaw_bdev(sb);
1031                 thaw_super_locked(sb);
1032         } else {
1033                 up_write(&sb->s_umount);
1034         }
1035 }
1036 
1037 static void do_thaw_all(struct work_struct *work)
1038 {
1039         __iterate_supers(do_thaw_all_callback);
1040         kfree(work);
1041         printk(KERN_WARNING "Emergency Thaw complete\n");
1042 }
1043 
1044 /**
1045  * emergency_thaw_all -- forcibly thaw every frozen filesystem
1046  *
1047  * Used for emergency unfreeze of all filesystems via SysRq
1048  */
1049 void emergency_thaw_all(void)
1050 {
1051         struct work_struct *work;
1052 
1053         work = kmalloc(sizeof(*work), GFP_ATOMIC);
1054         if (work) {
1055                 INIT_WORK(work, do_thaw_all);
1056                 schedule_work(work);
1057         }
1058 }
1059 
1060 static DEFINE_IDA(unnamed_dev_ida);
1061 
1062 /**
1063  * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1064  * @p: Pointer to a dev_t.
1065  *
1066  * Filesystems which don't use real block devices can call this function
1067  * to allocate a virtual block device.
1068  *
1069  * Context: Any context.  Frequently called while holding sb_lock.
1070  * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1071  * or -ENOMEM if memory allocation failed.
1072  */
1073 int get_anon_bdev(dev_t *p)
1074 {
1075         int dev;
1076 
1077         /*
1078          * Many userspace utilities consider an FSID of 0 invalid.
1079          * Always return at least 1 from get_anon_bdev.
1080          */
1081         dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1082                         GFP_ATOMIC);
1083         if (dev == -ENOSPC)
1084                 dev = -EMFILE;
1085         if (dev < 0)
1086                 return dev;
1087 
1088         *p = MKDEV(0, dev);
1089         return 0;
1090 }
1091 EXPORT_SYMBOL(get_anon_bdev);
1092 
1093 void free_anon_bdev(dev_t dev)
1094 {
1095         ida_free(&unnamed_dev_ida, MINOR(dev));
1096 }
1097 EXPORT_SYMBOL(free_anon_bdev);
1098 
1099 int set_anon_super(struct super_block *s, void *data)
1100 {
1101         return get_anon_bdev(&s->s_dev);
1102 }
1103 EXPORT_SYMBOL(set_anon_super);
1104 
1105 void kill_anon_super(struct super_block *sb)
1106 {
1107         dev_t dev = sb->s_dev;
1108         generic_shutdown_super(sb);
1109         free_anon_bdev(dev);
1110 }
1111 EXPORT_SYMBOL(kill_anon_super);
1112 
1113 void kill_litter_super(struct super_block *sb)
1114 {
1115         if (sb->s_root)
1116                 d_genocide(sb->s_root);
1117         kill_anon_super(sb);
1118 }
1119 EXPORT_SYMBOL(kill_litter_super);
1120 
1121 int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1122 {
1123         return set_anon_super(sb, NULL);
1124 }
1125 EXPORT_SYMBOL(set_anon_super_fc);
1126 
1127 static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1128 {
1129         return sb->s_fs_info == fc->s_fs_info;
1130 }
1131 
1132 static int test_single_super(struct super_block *s, struct fs_context *fc)
1133 {
1134         return 1;
1135 }
1136 
1137 /**
1138  * vfs_get_super - Get a superblock with a search key set in s_fs_info.
1139  * @fc: The filesystem context holding the parameters
1140  * @keying: How to distinguish superblocks
1141  * @fill_super: Helper to initialise a new superblock
1142  *
1143  * Search for a superblock and create a new one if not found.  The search
1144  * criterion is controlled by @keying.  If the search fails, a new superblock
1145  * is created and @fill_super() is called to initialise it.
1146  *
1147  * @keying can take one of a number of values:
1148  *
1149  * (1) vfs_get_single_super - Only one superblock of this type may exist on the
1150  *     system.  This is typically used for special system filesystems.
1151  *
1152  * (2) vfs_get_keyed_super - Multiple superblocks may exist, but they must have
1153  *     distinct keys (where the key is in s_fs_info).  Searching for the same
1154  *     key again will turn up the superblock for that key.
1155  *
1156  * (3) vfs_get_independent_super - Multiple superblocks may exist and are
1157  *     unkeyed.  Each call will get a new superblock.
1158  *
1159  * A permissions check is made by sget_fc() unless we're getting a superblock
1160  * for a kernel-internal mount or a submount.
1161  */
1162 int vfs_get_super(struct fs_context *fc,
1163                   enum vfs_get_super_keying keying,
1164                   int (*fill_super)(struct super_block *sb,
1165                                     struct fs_context *fc))
1166 {
1167         int (*test)(struct super_block *, struct fs_context *);
1168         struct super_block *sb;
1169         int err;
1170 
1171         switch (keying) {
1172         case vfs_get_single_super:
1173         case vfs_get_single_reconf_super:
1174                 test = test_single_super;
1175                 break;
1176         case vfs_get_keyed_super:
1177                 test = test_keyed_super;
1178                 break;
1179         case vfs_get_independent_super:
1180                 test = NULL;
1181                 break;
1182         default:
1183                 BUG();
1184         }
1185 
1186         sb = sget_fc(fc, test, set_anon_super_fc);
1187         if (IS_ERR(sb))
1188                 return PTR_ERR(sb);
1189 
1190         if (!sb->s_root) {
1191                 err = fill_super(sb, fc);
1192                 if (err)
1193                         goto error;
1194 
1195                 sb->s_flags |= SB_ACTIVE;
1196                 fc->root = dget(sb->s_root);
1197         } else {
1198                 fc->root = dget(sb->s_root);
1199                 if (keying == vfs_get_single_reconf_super) {
1200                         err = reconfigure_super(fc);
1201                         if (err < 0) {
1202                                 dput(fc->root);
1203                                 fc->root = NULL;
1204                                 goto error;
1205                         }
1206                 }
1207         }
1208 
1209         return 0;
1210 
1211 error:
1212         deactivate_locked_super(sb);
1213         return err;
1214 }
1215 EXPORT_SYMBOL(vfs_get_super);
1216 
1217 int get_tree_nodev(struct fs_context *fc,
1218                   int (*fill_super)(struct super_block *sb,
1219                                     struct fs_context *fc))
1220 {
1221         return vfs_get_super(fc, vfs_get_independent_super, fill_super);
1222 }
1223 EXPORT_SYMBOL(get_tree_nodev);
1224 
1225 int get_tree_single(struct fs_context *fc,
1226                   int (*fill_super)(struct super_block *sb,
1227                                     struct fs_context *fc))
1228 {
1229         return vfs_get_super(fc, vfs_get_single_super, fill_super);
1230 }
1231 EXPORT_SYMBOL(get_tree_single);
1232 
1233 int get_tree_single_reconf(struct fs_context *fc,
1234                   int (*fill_super)(struct super_block *sb,
1235                                     struct fs_context *fc))
1236 {
1237         return vfs_get_super(fc, vfs_get_single_reconf_super, fill_super);
1238 }
1239 EXPORT_SYMBOL(get_tree_single_reconf);
1240 
1241 int get_tree_keyed(struct fs_context *fc,
1242                   int (*fill_super)(struct super_block *sb,
1243                                     struct fs_context *fc),
1244                 void *key)
1245 {
1246         fc->s_fs_info = key;
1247         return vfs_get_super(fc, vfs_get_keyed_super, fill_super);
1248 }
1249 EXPORT_SYMBOL(get_tree_keyed);
1250 
1251 #ifdef CONFIG_BLOCK
1252 
1253 static int set_bdev_super(struct super_block *s, void *data)
1254 {
1255         s->s_bdev = data;
1256         s->s_dev = s->s_bdev->bd_dev;
1257         s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1258 
1259         return 0;
1260 }
1261 
1262 static int set_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1263 {
1264         return set_bdev_super(s, fc->sget_key);
1265 }
1266 
1267 static int test_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1268 {
1269         return s->s_bdev == fc->sget_key;
1270 }
1271 
1272 /**
1273  * get_tree_bdev - Get a superblock based on a single block device
1274  * @fc: The filesystem context holding the parameters
1275  * @fill_super: Helper to initialise a new superblock
1276  */
1277 int get_tree_bdev(struct fs_context *fc,
1278                 int (*fill_super)(struct super_block *,
1279                                   struct fs_context *))
1280 {
1281         struct block_device *bdev;
1282         struct super_block *s;
1283         fmode_t mode = FMODE_READ | FMODE_EXCL;
1284         int error = 0;
1285 
1286         if (!(fc->sb_flags & SB_RDONLY))
1287                 mode |= FMODE_WRITE;
1288 
1289         if (!fc->source)
1290                 return invalf(fc, "No source specified");
1291 
1292         bdev = blkdev_get_by_path(fc->source, mode, fc->fs_type);
1293         if (IS_ERR(bdev)) {
1294                 errorf(fc, "%s: Can't open blockdev", fc->source);
1295                 return PTR_ERR(bdev);
1296         }
1297 
1298         /* Once the superblock is inserted into the list by sget_fc(), s_umount
1299          * will protect the lockfs code from trying to start a snapshot while
1300          * we are mounting
1301          */
1302         mutex_lock(&bdev->bd_fsfreeze_mutex);
1303         if (bdev->bd_fsfreeze_count > 0) {
1304                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1305                 warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
1306                 blkdev_put(bdev, mode);
1307                 return -EBUSY;
1308         }
1309 
1310         fc->sb_flags |= SB_NOSEC;
1311         fc->sget_key = bdev;
1312         s = sget_fc(fc, test_bdev_super_fc, set_bdev_super_fc);
1313         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1314         if (IS_ERR(s)) {
1315                 blkdev_put(bdev, mode);
1316                 return PTR_ERR(s);
1317         }
1318 
1319         if (s->s_root) {
1320                 /* Don't summarily change the RO/RW state. */
1321                 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1322                         warnf(fc, "%pg: Can't mount, would change RO state", bdev);
1323                         deactivate_locked_super(s);
1324                         blkdev_put(bdev, mode);
1325                         return -EBUSY;
1326                 }
1327 
1328                 /*
1329                  * s_umount nests inside bd_mutex during
1330                  * __invalidate_device().  blkdev_put() acquires
1331                  * bd_mutex and can't be called under s_umount.  Drop
1332                  * s_umount temporarily.  This is safe as we're
1333                  * holding an active reference.
1334                  */
1335                 up_write(&s->s_umount);
1336                 blkdev_put(bdev, mode);
1337                 down_write(&s->s_umount);
1338         } else {
1339                 s->s_mode = mode;
1340                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1341                 sb_set_blocksize(s, block_size(bdev));
1342                 error = fill_super(s, fc);
1343                 if (error) {
1344                         deactivate_locked_super(s);
1345                         return error;
1346                 }
1347 
1348                 s->s_flags |= SB_ACTIVE;
1349                 bdev->bd_super = s;
1350         }
1351 
1352         BUG_ON(fc->root);
1353         fc->root = dget(s->s_root);
1354         return 0;
1355 }
1356 EXPORT_SYMBOL(get_tree_bdev);
1357 
1358 static int test_bdev_super(struct super_block *s, void *data)
1359 {
1360         return (void *)s->s_bdev == data;
1361 }
1362 
1363 struct dentry *mount_bdev(struct file_system_type *fs_type,
1364         int flags, const char *dev_name, void *data,
1365         int (*fill_super)(struct super_block *, void *, int))
1366 {
1367         struct block_device *bdev;
1368         struct super_block *s;
1369         fmode_t mode = FMODE_READ | FMODE_EXCL;
1370         int error = 0;
1371 
1372         if (!(flags & SB_RDONLY))
1373                 mode |= FMODE_WRITE;
1374 
1375         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1376         if (IS_ERR(bdev))
1377                 return ERR_CAST(bdev);
1378 
1379         /*
1380          * once the super is inserted into the list by sget, s_umount
1381          * will protect the lockfs code from trying to start a snapshot
1382          * while we are mounting
1383          */
1384         mutex_lock(&bdev->bd_fsfreeze_mutex);
1385         if (bdev->bd_fsfreeze_count > 0) {
1386                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1387                 error = -EBUSY;
1388                 goto error_bdev;
1389         }
1390         s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
1391                  bdev);
1392         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1393         if (IS_ERR(s))
1394                 goto error_s;
1395 
1396         if (s->s_root) {
1397                 if ((flags ^ s->s_flags) & SB_RDONLY) {
1398                         deactivate_locked_super(s);
1399                         error = -EBUSY;
1400                         goto error_bdev;
1401                 }
1402 
1403                 /*
1404                  * s_umount nests inside bd_mutex during
1405                  * __invalidate_device().  blkdev_put() acquires
1406                  * bd_mutex and can't be called under s_umount.  Drop
1407                  * s_umount temporarily.  This is safe as we're
1408                  * holding an active reference.
1409                  */
1410                 up_write(&s->s_umount);
1411                 blkdev_put(bdev, mode);
1412                 down_write(&s->s_umount);
1413         } else {
1414                 s->s_mode = mode;
1415                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1416                 sb_set_blocksize(s, block_size(bdev));
1417                 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1418                 if (error) {
1419                         deactivate_locked_super(s);
1420                         goto error;
1421                 }
1422 
1423                 s->s_flags |= SB_ACTIVE;
1424                 bdev->bd_super = s;
1425         }
1426 
1427         return dget(s->s_root);
1428 
1429 error_s:
1430         error = PTR_ERR(s);
1431 error_bdev:
1432         blkdev_put(bdev, mode);
1433 error:
1434         return ERR_PTR(error);
1435 }
1436 EXPORT_SYMBOL(mount_bdev);
1437 
1438 void kill_block_super(struct super_block *sb)
1439 {
1440         struct block_device *bdev = sb->s_bdev;
1441         fmode_t mode = sb->s_mode;
1442 
1443         bdev->bd_super = NULL;
1444         generic_shutdown_super(sb);
1445         sync_blockdev(bdev);
1446         WARN_ON_ONCE(!(mode & FMODE_EXCL));
1447         blkdev_put(bdev, mode | FMODE_EXCL);
1448 }
1449 
1450 EXPORT_SYMBOL(kill_block_super);
1451 #endif
1452 
1453 struct dentry *mount_nodev(struct file_system_type *fs_type,
1454         int flags, void *data,
1455         int (*fill_super)(struct super_block *, void *, int))
1456 {
1457         int error;
1458         struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1459 
1460         if (IS_ERR(s))
1461                 return ERR_CAST(s);
1462 
1463         error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1464         if (error) {
1465                 deactivate_locked_super(s);
1466                 return ERR_PTR(error);
1467         }
1468         s->s_flags |= SB_ACTIVE;
1469         return dget(s->s_root);
1470 }
1471 EXPORT_SYMBOL(mount_nodev);
1472 
1473 static int reconfigure_single(struct super_block *s,
1474                               int flags, void *data)
1475 {
1476         struct fs_context *fc;
1477         int ret;
1478 
1479         /* The caller really need to be passing fc down into mount_single(),
1480          * then a chunk of this can be removed.  [Bollocks -- AV]
1481          * Better yet, reconfiguration shouldn't happen, but rather the second
1482          * mount should be rejected if the parameters are not compatible.
1483          */
1484         fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1485         if (IS_ERR(fc))
1486                 return PTR_ERR(fc);
1487 
1488         ret = parse_monolithic_mount_data(fc, data);
1489         if (ret < 0)
1490                 goto out;
1491 
1492         ret = reconfigure_super(fc);
1493 out:
1494         put_fs_context(fc);
1495         return ret;
1496 }
1497 
1498 static int compare_single(struct super_block *s, void *p)
1499 {
1500         return 1;
1501 }
1502 
1503 struct dentry *mount_single(struct file_system_type *fs_type,
1504         int flags, void *data,
1505         int (*fill_super)(struct super_block *, void *, int))
1506 {
1507         struct super_block *s;
1508         int error;
1509 
1510         s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1511         if (IS_ERR(s))
1512                 return ERR_CAST(s);
1513         if (!s->s_root) {
1514                 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1515                 if (!error)
1516                         s->s_flags |= SB_ACTIVE;
1517         } else {
1518                 error = reconfigure_single(s, flags, data);
1519         }
1520         if (unlikely(error)) {
1521                 deactivate_locked_super(s);
1522                 return ERR_PTR(error);
1523         }
1524         return dget(s->s_root);
1525 }
1526 EXPORT_SYMBOL(mount_single);
1527 
1528 /**
1529  * vfs_get_tree - Get the mountable root
1530  * @fc: The superblock configuration context.
1531  *
1532  * The filesystem is invoked to get or create a superblock which can then later
1533  * be used for mounting.  The filesystem places a pointer to the root to be
1534  * used for mounting in @fc->root.
1535  */
1536 int vfs_get_tree(struct fs_context *fc)
1537 {
1538         struct super_block *sb;
1539         int error;
1540 
1541         if (fc->root)
1542                 return -EBUSY;
1543 
1544         /* Get the mountable root in fc->root, with a ref on the root and a ref
1545          * on the superblock.
1546          */
1547         error = fc->ops->get_tree(fc);
1548         if (error < 0)
1549                 return error;
1550 
1551         if (!fc->root) {
1552                 pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1553                        fc->fs_type->name);
1554                 /* We don't know what the locking state of the superblock is -
1555                  * if there is a superblock.
1556                  */
1557                 BUG();
1558         }
1559 
1560         sb = fc->root->d_sb;
1561         WARN_ON(!sb->s_bdi);
1562 
1563         /*
1564          * Write barrier is for super_cache_count(). We place it before setting
1565          * SB_BORN as the data dependency between the two functions is the
1566          * superblock structure contents that we just set up, not the SB_BORN
1567          * flag.
1568          */
1569         smp_wmb();
1570         sb->s_flags |= SB_BORN;
1571 
1572         error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
1573         if (unlikely(error)) {
1574                 fc_drop_locked(fc);
1575                 return error;
1576         }
1577 
1578         /*
1579          * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1580          * but s_maxbytes was an unsigned long long for many releases. Throw
1581          * this warning for a little while to try and catch filesystems that
1582          * violate this rule.
1583          */
1584         WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1585                 "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
1586 
1587         return 0;
1588 }
1589 EXPORT_SYMBOL(vfs_get_tree);
1590 
1591 /*
1592  * Setup private BDI for given superblock. It gets automatically cleaned up
1593  * in generic_shutdown_super().
1594  */
1595 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1596 {
1597         struct backing_dev_info *bdi;
1598         int err;
1599         va_list args;
1600 
1601         bdi = bdi_alloc(NUMA_NO_NODE);
1602         if (!bdi)
1603                 return -ENOMEM;
1604 
1605         va_start(args, fmt);
1606         err = bdi_register_va(bdi, fmt, args);
1607         va_end(args);
1608         if (err) {
1609                 bdi_put(bdi);
1610                 return err;
1611         }
1612         WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1613         sb->s_bdi = bdi;
1614 
1615         return 0;
1616 }
1617 EXPORT_SYMBOL(super_setup_bdi_name);
1618 
1619 /*
1620  * Setup private BDI for given superblock. I gets automatically cleaned up
1621  * in generic_shutdown_super().
1622  */
1623 int super_setup_bdi(struct super_block *sb)
1624 {
1625         static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1626 
1627         return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1628                                     atomic_long_inc_return(&bdi_seq));
1629 }
1630 EXPORT_SYMBOL(super_setup_bdi);
1631 
1632 /*
1633  * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1634  * instead.
1635  */
1636 void __sb_end_write(struct super_block *sb, int level)
1637 {
1638         percpu_up_read(sb->s_writers.rw_sem + level-1);
1639 }
1640 EXPORT_SYMBOL(__sb_end_write);
1641 
1642 /*
1643  * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1644  * instead.
1645  */
1646 int __sb_start_write(struct super_block *sb, int level, bool wait)
1647 {
1648         bool force_trylock = false;
1649         int ret = 1;
1650 
1651 #ifdef CONFIG_LOCKDEP
1652         /*
1653          * We want lockdep to tell us about possible deadlocks with freezing
1654          * but it's it bit tricky to properly instrument it. Getting a freeze
1655          * protection works as getting a read lock but there are subtle
1656          * problems. XFS for example gets freeze protection on internal level
1657          * twice in some cases, which is OK only because we already hold a
1658          * freeze protection also on higher level. Due to these cases we have
1659          * to use wait == F (trylock mode) which must not fail.
1660          */
1661         if (wait) {
1662                 int i;
1663 
1664                 for (i = 0; i < level - 1; i++)
1665                         if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
1666                                 force_trylock = true;
1667                                 break;
1668                         }
1669         }
1670 #endif
1671         if (wait && !force_trylock)
1672                 percpu_down_read(sb->s_writers.rw_sem + level-1);
1673         else
1674                 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1675 
1676         WARN_ON(force_trylock && !ret);
1677         return ret;
1678 }
1679 EXPORT_SYMBOL(__sb_start_write);
1680 
1681 /**
1682  * sb_wait_write - wait until all writers to given file system finish
1683  * @sb: the super for which we wait
1684  * @level: type of writers we wait for (normal vs page fault)
1685  *
1686  * This function waits until there are no writers of given type to given file
1687  * system.
1688  */
1689 static void sb_wait_write(struct super_block *sb, int level)
1690 {
1691         percpu_down_write(sb->s_writers.rw_sem + level-1);
1692 }
1693 
1694 /*
1695  * We are going to return to userspace and forget about these locks, the
1696  * ownership goes to the caller of thaw_super() which does unlock().
1697  */
1698 static void lockdep_sb_freeze_release(struct super_block *sb)
1699 {
1700         int level;
1701 
1702         for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1703                 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1704 }
1705 
1706 /*
1707  * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1708  */
1709 static void lockdep_sb_freeze_acquire(struct super_block *sb)
1710 {
1711         int level;
1712 
1713         for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1714                 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1715 }
1716 
1717 static void sb_freeze_unlock(struct super_block *sb)
1718 {
1719         int level;
1720 
1721         for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1722                 percpu_up_write(sb->s_writers.rw_sem + level);
1723 }
1724 
1725 /**
1726  * freeze_super - lock the filesystem and force it into a consistent state
1727  * @sb: the super to lock
1728  *
1729  * Syncs the super to make sure the filesystem is consistent and calls the fs's
1730  * freeze_fs.  Subsequent calls to this without first thawing the fs will return
1731  * -EBUSY.
1732  *
1733  * During this function, sb->s_writers.frozen goes through these values:
1734  *
1735  * SB_UNFROZEN: File system is normal, all writes progress as usual.
1736  *
1737  * SB_FREEZE_WRITE: The file system is in the process of being frozen.  New
1738  * writes should be blocked, though page faults are still allowed. We wait for
1739  * all writes to complete and then proceed to the next stage.
1740  *
1741  * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1742  * but internal fs threads can still modify the filesystem (although they
1743  * should not dirty new pages or inodes), writeback can run etc. After waiting
1744  * for all running page faults we sync the filesystem which will clean all
1745  * dirty pages and inodes (no new dirty pages or inodes can be created when
1746  * sync is running).
1747  *
1748  * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1749  * modification are blocked (e.g. XFS preallocation truncation on inode
1750  * reclaim). This is usually implemented by blocking new transactions for
1751  * filesystems that have them and need this additional guard. After all
1752  * internal writers are finished we call ->freeze_fs() to finish filesystem
1753  * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1754  * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1755  *
1756  * sb->s_writers.frozen is protected by sb->s_umount.
1757  */
1758 int freeze_super(struct super_block *sb)
1759 {
1760         int ret;
1761 
1762         atomic_inc(&sb->s_active);
1763         down_write(&sb->s_umount);
1764         if (sb->s_writers.frozen != SB_UNFROZEN) {
1765                 deactivate_locked_super(sb);
1766                 return -EBUSY;
1767         }
1768 
1769         if (!(sb->s_flags & SB_BORN)) {
1770                 up_write(&sb->s_umount);
1771                 return 0;       /* sic - it's "nothing to do" */
1772         }
1773 
1774         if (sb_rdonly(sb)) {
1775                 /* Nothing to do really... */
1776                 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1777                 up_write(&sb->s_umount);
1778                 return 0;
1779         }
1780 
1781         sb->s_writers.frozen = SB_FREEZE_WRITE;
1782         /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1783         up_write(&sb->s_umount);
1784         sb_wait_write(sb, SB_FREEZE_WRITE);
1785         down_write(&sb->s_umount);
1786 
1787         /* Now we go and block page faults... */
1788         sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1789         sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1790 
1791         /* All writers are done so after syncing there won't be dirty data */
1792         sync_filesystem(sb);
1793 
1794         /* Now wait for internal filesystem counter */
1795         sb->s_writers.frozen = SB_FREEZE_FS;
1796         sb_wait_write(sb, SB_FREEZE_FS);
1797 
1798         if (sb->s_op->freeze_fs) {
1799                 ret = sb->s_op->freeze_fs(sb);
1800                 if (ret) {
1801                         printk(KERN_ERR
1802                                 "VFS:Filesystem freeze failed\n");
1803                         sb->s_writers.frozen = SB_UNFROZEN;
1804                         sb_freeze_unlock(sb);
1805                         wake_up(&sb->s_writers.wait_unfrozen);
1806                         deactivate_locked_super(sb);
1807                         return ret;
1808                 }
1809         }
1810         /*
1811          * For debugging purposes so that fs can warn if it sees write activity
1812          * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1813          */
1814         sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1815         lockdep_sb_freeze_release(sb);
1816         up_write(&sb->s_umount);
1817         return 0;
1818 }
1819 EXPORT_SYMBOL(freeze_super);
1820 
1821 /**
1822  * thaw_super -- unlock filesystem
1823  * @sb: the super to thaw
1824  *
1825  * Unlocks the filesystem and marks it writeable again after freeze_super().
1826  */
1827 static int thaw_super_locked(struct super_block *sb)
1828 {
1829         int error;
1830 
1831         if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1832                 up_write(&sb->s_umount);
1833                 return -EINVAL;
1834         }
1835 
1836         if (sb_rdonly(sb)) {
1837                 sb->s_writers.frozen = SB_UNFROZEN;
1838                 goto out;
1839         }
1840 
1841         lockdep_sb_freeze_acquire(sb);
1842 
1843         if (sb->s_op->unfreeze_fs) {
1844                 error = sb->s_op->unfreeze_fs(sb);
1845                 if (error) {
1846                         printk(KERN_ERR
1847                                 "VFS:Filesystem thaw failed\n");
1848                         lockdep_sb_freeze_release(sb);
1849                         up_write(&sb->s_umount);
1850                         return error;
1851                 }
1852         }
1853 
1854         sb->s_writers.frozen = SB_UNFROZEN;
1855         sb_freeze_unlock(sb);
1856 out:
1857         wake_up(&sb->s_writers.wait_unfrozen);
1858         deactivate_locked_super(sb);
1859         return 0;
1860 }
1861 
1862 int thaw_super(struct super_block *sb)
1863 {
1864         down_write(&sb->s_umount);
1865         return thaw_super_locked(sb);
1866 }
1867 EXPORT_SYMBOL(thaw_super);
1868 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp