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

TOMOYO Linux Cross Reference
Linux/fs/super.c

Version: ~ [ linux-5.10-rc1 ] ~ [ linux-5.9.1 ] ~ [ linux-5.8.16 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.72 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.152 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.202 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.240 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.240 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ 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/fsnotify.h>
 36 #include <linux/lockdep.h>
 37 #include <linux/user_namespace.h>
 38 #include "internal.h"
 39 
 40 
 41 static LIST_HEAD(super_blocks);
 42 static DEFINE_SPINLOCK(sb_lock);
 43 
 44 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
 45         "sb_writers",
 46         "sb_pagefaults",
 47         "sb_internal",
 48 };
 49 
 50 /*
 51  * One thing we have to be careful of with a per-sb shrinker is that we don't
 52  * drop the last active reference to the superblock from within the shrinker.
 53  * If that happens we could trigger unregistering the shrinker from within the
 54  * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
 55  * take a passive reference to the superblock to avoid this from occurring.
 56  */
 57 static unsigned long super_cache_scan(struct shrinker *shrink,
 58                                       struct shrink_control *sc)
 59 {
 60         struct super_block *sb;
 61         long    fs_objects = 0;
 62         long    total_objects;
 63         long    freed = 0;
 64         long    dentries;
 65         long    inodes;
 66 
 67         sb = container_of(shrink, struct super_block, s_shrink);
 68 
 69         /*
 70          * Deadlock avoidance.  We may hold various FS locks, and we don't want
 71          * to recurse into the FS that called us in clear_inode() and friends..
 72          */
 73         if (!(sc->gfp_mask & __GFP_FS))
 74                 return SHRINK_STOP;
 75 
 76         if (!trylock_super(sb))
 77                 return SHRINK_STOP;
 78 
 79         if (sb->s_op->nr_cached_objects)
 80                 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
 81 
 82         inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
 83         dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
 84         total_objects = dentries + inodes + fs_objects + 1;
 85         if (!total_objects)
 86                 total_objects = 1;
 87 
 88         /* proportion the scan between the caches */
 89         dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
 90         inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
 91         fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
 92 
 93         /*
 94          * prune the dcache first as the icache is pinned by it, then
 95          * prune the icache, followed by the filesystem specific caches
 96          *
 97          * Ensure that we always scan at least one object - memcg kmem
 98          * accounting uses this to fully empty the caches.
 99          */
100         sc->nr_to_scan = dentries + 1;
101         freed = prune_dcache_sb(sb, sc);
102         sc->nr_to_scan = inodes + 1;
103         freed += prune_icache_sb(sb, sc);
104 
105         if (fs_objects) {
106                 sc->nr_to_scan = fs_objects + 1;
107                 freed += sb->s_op->free_cached_objects(sb, sc);
108         }
109 
110         up_read(&sb->s_umount);
111         return freed;
112 }
113 
114 static unsigned long super_cache_count(struct shrinker *shrink,
115                                        struct shrink_control *sc)
116 {
117         struct super_block *sb;
118         long    total_objects = 0;
119 
120         sb = container_of(shrink, struct super_block, s_shrink);
121 
122         /*
123          * Don't call trylock_super as it is a potential
124          * scalability bottleneck. The counts could get updated
125          * between super_cache_count and super_cache_scan anyway.
126          * Call to super_cache_count with shrinker_rwsem held
127          * ensures the safety of call to list_lru_shrink_count() and
128          * s_op->nr_cached_objects().
129          */
130         if (sb->s_op && sb->s_op->nr_cached_objects)
131                 total_objects = sb->s_op->nr_cached_objects(sb, sc);
132 
133         total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
134         total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
135 
136         total_objects = vfs_pressure_ratio(total_objects);
137         return total_objects;
138 }
139 
140 static void destroy_super_work(struct work_struct *work)
141 {
142         struct super_block *s = container_of(work, struct super_block,
143                                                         destroy_work);
144         int i;
145 
146         for (i = 0; i < SB_FREEZE_LEVELS; i++)
147                 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
148         kfree(s);
149 }
150 
151 static void destroy_super_rcu(struct rcu_head *head)
152 {
153         struct super_block *s = container_of(head, struct super_block, rcu);
154         INIT_WORK(&s->destroy_work, destroy_super_work);
155         schedule_work(&s->destroy_work);
156 }
157 
158 /* Free a superblock that has never been seen by anyone */
159 static void destroy_unused_super(struct super_block *s)
160 {
161         if (!s)
162                 return;
163         up_write(&s->s_umount);
164         list_lru_destroy(&s->s_dentry_lru);
165         list_lru_destroy(&s->s_inode_lru);
166         security_sb_free(s);
167         put_user_ns(s->s_user_ns);
168         kfree(s->s_subtype);
169         /* no delays needed */
170         destroy_super_work(&s->destroy_work);
171 }
172 
173 /**
174  *      alloc_super     -       create new superblock
175  *      @type:  filesystem type superblock should belong to
176  *      @flags: the mount flags
177  *      @user_ns: User namespace for the super_block
178  *
179  *      Allocates and initializes a new &struct super_block.  alloc_super()
180  *      returns a pointer new superblock or %NULL if allocation had failed.
181  */
182 static struct super_block *alloc_super(struct file_system_type *type, int flags,
183                                        struct user_namespace *user_ns)
184 {
185         struct super_block *s = kzalloc(sizeof(struct super_block),  GFP_USER);
186         static const struct super_operations default_op;
187         int i;
188 
189         if (!s)
190                 return NULL;
191 
192         INIT_LIST_HEAD(&s->s_mounts);
193         s->s_user_ns = get_user_ns(user_ns);
194         init_rwsem(&s->s_umount);
195         lockdep_set_class(&s->s_umount, &type->s_umount_key);
196         /*
197          * sget() can have s_umount recursion.
198          *
199          * When it cannot find a suitable sb, it allocates a new
200          * one (this one), and tries again to find a suitable old
201          * one.
202          *
203          * In case that succeeds, it will acquire the s_umount
204          * lock of the old one. Since these are clearly distrinct
205          * locks, and this object isn't exposed yet, there's no
206          * risk of deadlocks.
207          *
208          * Annotate this by putting this lock in a different
209          * subclass.
210          */
211         down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
212 
213         if (security_sb_alloc(s))
214                 goto fail;
215 
216         for (i = 0; i < SB_FREEZE_LEVELS; i++) {
217                 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
218                                         sb_writers_name[i],
219                                         &type->s_writers_key[i]))
220                         goto fail;
221         }
222         init_waitqueue_head(&s->s_writers.wait_unfrozen);
223         s->s_bdi = &noop_backing_dev_info;
224         s->s_flags = flags;
225         if (s->s_user_ns != &init_user_ns)
226                 s->s_iflags |= SB_I_NODEV;
227         INIT_HLIST_NODE(&s->s_instances);
228         INIT_HLIST_BL_HEAD(&s->s_anon);
229         mutex_init(&s->s_sync_lock);
230         INIT_LIST_HEAD(&s->s_inodes);
231         spin_lock_init(&s->s_inode_list_lock);
232         INIT_LIST_HEAD(&s->s_inodes_wb);
233         spin_lock_init(&s->s_inode_wblist_lock);
234 
235         if (list_lru_init_memcg(&s->s_dentry_lru))
236                 goto fail;
237         if (list_lru_init_memcg(&s->s_inode_lru))
238                 goto fail;
239         s->s_count = 1;
240         atomic_set(&s->s_active, 1);
241         mutex_init(&s->s_vfs_rename_mutex);
242         lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
243         init_rwsem(&s->s_dquot.dqio_sem);
244         s->s_maxbytes = MAX_NON_LFS;
245         s->s_op = &default_op;
246         s->s_time_gran = 1000000000;
247         s->cleancache_poolid = CLEANCACHE_NO_POOL;
248 
249         s->s_shrink.seeks = DEFAULT_SEEKS;
250         s->s_shrink.scan_objects = super_cache_scan;
251         s->s_shrink.count_objects = super_cache_count;
252         s->s_shrink.batch = 1024;
253         s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
254         return s;
255 
256 fail:
257         destroy_unused_super(s);
258         return NULL;
259 }
260 
261 /* Superblock refcounting  */
262 
263 /*
264  * Drop a superblock's refcount.  The caller must hold sb_lock.
265  */
266 static void __put_super(struct super_block *s)
267 {
268         if (!--s->s_count) {
269                 list_del_init(&s->s_list);
270                 WARN_ON(s->s_dentry_lru.node);
271                 WARN_ON(s->s_inode_lru.node);
272                 WARN_ON(!list_empty(&s->s_mounts));
273                 security_sb_free(s);
274                 put_user_ns(s->s_user_ns);
275                 kfree(s->s_subtype);
276                 call_rcu(&s->rcu, destroy_super_rcu);
277         }
278 }
279 
280 /**
281  *      put_super       -       drop a temporary reference to superblock
282  *      @sb: superblock in question
283  *
284  *      Drops a temporary reference, frees superblock if there's no
285  *      references left.
286  */
287 static void put_super(struct super_block *sb)
288 {
289         spin_lock(&sb_lock);
290         __put_super(sb);
291         spin_unlock(&sb_lock);
292 }
293 
294 
295 /**
296  *      deactivate_locked_super -       drop an active reference to superblock
297  *      @s: superblock to deactivate
298  *
299  *      Drops an active reference to superblock, converting it into a temporary
300  *      one if there is no other active references left.  In that case we
301  *      tell fs driver to shut it down and drop the temporary reference we
302  *      had just acquired.
303  *
304  *      Caller holds exclusive lock on superblock; that lock is released.
305  */
306 void deactivate_locked_super(struct super_block *s)
307 {
308         struct file_system_type *fs = s->s_type;
309         if (atomic_dec_and_test(&s->s_active)) {
310                 cleancache_invalidate_fs(s);
311                 unregister_shrinker(&s->s_shrink);
312                 fs->kill_sb(s);
313 
314                 /*
315                  * Since list_lru_destroy() may sleep, we cannot call it from
316                  * put_super(), where we hold the sb_lock. Therefore we destroy
317                  * the lru lists right now.
318                  */
319                 list_lru_destroy(&s->s_dentry_lru);
320                 list_lru_destroy(&s->s_inode_lru);
321 
322                 put_filesystem(fs);
323                 put_super(s);
324         } else {
325                 up_write(&s->s_umount);
326         }
327 }
328 
329 EXPORT_SYMBOL(deactivate_locked_super);
330 
331 /**
332  *      deactivate_super        -       drop an active reference to superblock
333  *      @s: superblock to deactivate
334  *
335  *      Variant of deactivate_locked_super(), except that superblock is *not*
336  *      locked by caller.  If we are going to drop the final active reference,
337  *      lock will be acquired prior to that.
338  */
339 void deactivate_super(struct super_block *s)
340 {
341         if (!atomic_add_unless(&s->s_active, -1, 1)) {
342                 down_write(&s->s_umount);
343                 deactivate_locked_super(s);
344         }
345 }
346 
347 EXPORT_SYMBOL(deactivate_super);
348 
349 /**
350  *      grab_super - acquire an active reference
351  *      @s: reference we are trying to make active
352  *
353  *      Tries to acquire an active reference.  grab_super() is used when we
354  *      had just found a superblock in super_blocks or fs_type->fs_supers
355  *      and want to turn it into a full-blown active reference.  grab_super()
356  *      is called with sb_lock held and drops it.  Returns 1 in case of
357  *      success, 0 if we had failed (superblock contents was already dead or
358  *      dying when grab_super() had been called).  Note that this is only
359  *      called for superblocks not in rundown mode (== ones still on ->fs_supers
360  *      of their type), so increment of ->s_count is OK here.
361  */
362 static int grab_super(struct super_block *s) __releases(sb_lock)
363 {
364         s->s_count++;
365         spin_unlock(&sb_lock);
366         down_write(&s->s_umount);
367         if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
368                 put_super(s);
369                 return 1;
370         }
371         up_write(&s->s_umount);
372         put_super(s);
373         return 0;
374 }
375 
376 /*
377  *      trylock_super - try to grab ->s_umount shared
378  *      @sb: reference we are trying to grab
379  *
380  *      Try to prevent fs shutdown.  This is used in places where we
381  *      cannot take an active reference but we need to ensure that the
382  *      filesystem is not shut down while we are working on it. It returns
383  *      false if we cannot acquire s_umount or if we lose the race and
384  *      filesystem already got into shutdown, and returns true with the s_umount
385  *      lock held in read mode in case of success. On successful return,
386  *      the caller must drop the s_umount lock when done.
387  *
388  *      Note that unlike get_super() et.al. this one does *not* bump ->s_count.
389  *      The reason why it's safe is that we are OK with doing trylock instead
390  *      of down_read().  There's a couple of places that are OK with that, but
391  *      it's very much not a general-purpose interface.
392  */
393 bool trylock_super(struct super_block *sb)
394 {
395         if (down_read_trylock(&sb->s_umount)) {
396                 if (!hlist_unhashed(&sb->s_instances) &&
397                     sb->s_root && (sb->s_flags & SB_BORN))
398                         return true;
399                 up_read(&sb->s_umount);
400         }
401 
402         return false;
403 }
404 
405 /**
406  *      generic_shutdown_super  -       common helper for ->kill_sb()
407  *      @sb: superblock to kill
408  *
409  *      generic_shutdown_super() does all fs-independent work on superblock
410  *      shutdown.  Typical ->kill_sb() should pick all fs-specific objects
411  *      that need destruction out of superblock, call generic_shutdown_super()
412  *      and release aforementioned objects.  Note: dentries and inodes _are_
413  *      taken care of and do not need specific handling.
414  *
415  *      Upon calling this function, the filesystem may no longer alter or
416  *      rearrange the set of dentries belonging to this super_block, nor may it
417  *      change the attachments of dentries to inodes.
418  */
419 void generic_shutdown_super(struct super_block *sb)
420 {
421         const struct super_operations *sop = sb->s_op;
422 
423         if (sb->s_root) {
424                 shrink_dcache_for_umount(sb);
425                 sync_filesystem(sb);
426                 sb->s_flags &= ~SB_ACTIVE;
427 
428                 fsnotify_unmount_inodes(sb);
429                 cgroup_writeback_umount();
430 
431                 evict_inodes(sb);
432 
433                 if (sb->s_dio_done_wq) {
434                         destroy_workqueue(sb->s_dio_done_wq);
435                         sb->s_dio_done_wq = NULL;
436                 }
437 
438                 if (sop->put_super)
439                         sop->put_super(sb);
440 
441                 if (!list_empty(&sb->s_inodes)) {
442                         printk("VFS: Busy inodes after unmount of %s. "
443                            "Self-destruct in 5 seconds.  Have a nice day...\n",
444                            sb->s_id);
445                 }
446         }
447         spin_lock(&sb_lock);
448         /* should be initialized for __put_super_and_need_restart() */
449         hlist_del_init(&sb->s_instances);
450         spin_unlock(&sb_lock);
451         up_write(&sb->s_umount);
452         if (sb->s_bdi != &noop_backing_dev_info) {
453                 bdi_put(sb->s_bdi);
454                 sb->s_bdi = &noop_backing_dev_info;
455         }
456 }
457 
458 EXPORT_SYMBOL(generic_shutdown_super);
459 
460 /**
461  *      sget_userns -   find or create a superblock
462  *      @type:  filesystem type superblock should belong to
463  *      @test:  comparison callback
464  *      @set:   setup callback
465  *      @flags: mount flags
466  *      @user_ns: User namespace for the super_block
467  *      @data:  argument to each of them
468  */
469 struct super_block *sget_userns(struct file_system_type *type,
470                         int (*test)(struct super_block *,void *),
471                         int (*set)(struct super_block *,void *),
472                         int flags, struct user_namespace *user_ns,
473                         void *data)
474 {
475         struct super_block *s = NULL;
476         struct super_block *old;
477         int err;
478 
479         if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) &&
480             !(type->fs_flags & FS_USERNS_MOUNT) &&
481             !capable(CAP_SYS_ADMIN))
482                 return ERR_PTR(-EPERM);
483 retry:
484         spin_lock(&sb_lock);
485         if (test) {
486                 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
487                         if (!test(old, data))
488                                 continue;
489                         if (user_ns != old->s_user_ns) {
490                                 spin_unlock(&sb_lock);
491                                 destroy_unused_super(s);
492                                 return ERR_PTR(-EBUSY);
493                         }
494                         if (!grab_super(old))
495                                 goto retry;
496                         destroy_unused_super(s);
497                         return old;
498                 }
499         }
500         if (!s) {
501                 spin_unlock(&sb_lock);
502                 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
503                 if (!s)
504                         return ERR_PTR(-ENOMEM);
505                 goto retry;
506         }
507 
508         err = set(s, data);
509         if (err) {
510                 spin_unlock(&sb_lock);
511                 destroy_unused_super(s);
512                 return ERR_PTR(err);
513         }
514         s->s_type = type;
515         strlcpy(s->s_id, type->name, sizeof(s->s_id));
516         list_add_tail(&s->s_list, &super_blocks);
517         hlist_add_head(&s->s_instances, &type->fs_supers);
518         spin_unlock(&sb_lock);
519         get_filesystem(type);
520         err = register_shrinker(&s->s_shrink);
521         if (err) {
522                 deactivate_locked_super(s);
523                 s = ERR_PTR(err);
524         }
525         return s;
526 }
527 
528 EXPORT_SYMBOL(sget_userns);
529 
530 /**
531  *      sget    -       find or create a superblock
532  *      @type:    filesystem type superblock should belong to
533  *      @test:    comparison callback
534  *      @set:     setup callback
535  *      @flags:   mount flags
536  *      @data:    argument to each of them
537  */
538 struct super_block *sget(struct file_system_type *type,
539                         int (*test)(struct super_block *,void *),
540                         int (*set)(struct super_block *,void *),
541                         int flags,
542                         void *data)
543 {
544         struct user_namespace *user_ns = current_user_ns();
545 
546         /* We don't yet pass the user namespace of the parent
547          * mount through to here so always use &init_user_ns
548          * until that changes.
549          */
550         if (flags & SB_SUBMOUNT)
551                 user_ns = &init_user_ns;
552 
553         /* Ensure the requestor has permissions over the target filesystem */
554         if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
555                 return ERR_PTR(-EPERM);
556 
557         return sget_userns(type, test, set, flags, user_ns, data);
558 }
559 
560 EXPORT_SYMBOL(sget);
561 
562 void drop_super(struct super_block *sb)
563 {
564         up_read(&sb->s_umount);
565         put_super(sb);
566 }
567 
568 EXPORT_SYMBOL(drop_super);
569 
570 void drop_super_exclusive(struct super_block *sb)
571 {
572         up_write(&sb->s_umount);
573         put_super(sb);
574 }
575 EXPORT_SYMBOL(drop_super_exclusive);
576 
577 /**
578  *      iterate_supers - call function for all active superblocks
579  *      @f: function to call
580  *      @arg: argument to pass to it
581  *
582  *      Scans the superblock list and calls given function, passing it
583  *      locked superblock and given argument.
584  */
585 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
586 {
587         struct super_block *sb, *p = NULL;
588 
589         spin_lock(&sb_lock);
590         list_for_each_entry(sb, &super_blocks, s_list) {
591                 if (hlist_unhashed(&sb->s_instances))
592                         continue;
593                 sb->s_count++;
594                 spin_unlock(&sb_lock);
595 
596                 down_read(&sb->s_umount);
597                 if (sb->s_root && (sb->s_flags & SB_BORN))
598                         f(sb, arg);
599                 up_read(&sb->s_umount);
600 
601                 spin_lock(&sb_lock);
602                 if (p)
603                         __put_super(p);
604                 p = sb;
605         }
606         if (p)
607                 __put_super(p);
608         spin_unlock(&sb_lock);
609 }
610 
611 /**
612  *      iterate_supers_type - call function for superblocks of given type
613  *      @type: fs type
614  *      @f: function to call
615  *      @arg: argument to pass to it
616  *
617  *      Scans the superblock list and calls given function, passing it
618  *      locked superblock and given argument.
619  */
620 void iterate_supers_type(struct file_system_type *type,
621         void (*f)(struct super_block *, void *), void *arg)
622 {
623         struct super_block *sb, *p = NULL;
624 
625         spin_lock(&sb_lock);
626         hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
627                 sb->s_count++;
628                 spin_unlock(&sb_lock);
629 
630                 down_read(&sb->s_umount);
631                 if (sb->s_root && (sb->s_flags & SB_BORN))
632                         f(sb, arg);
633                 up_read(&sb->s_umount);
634 
635                 spin_lock(&sb_lock);
636                 if (p)
637                         __put_super(p);
638                 p = sb;
639         }
640         if (p)
641                 __put_super(p);
642         spin_unlock(&sb_lock);
643 }
644 
645 EXPORT_SYMBOL(iterate_supers_type);
646 
647 static struct super_block *__get_super(struct block_device *bdev, bool excl)
648 {
649         struct super_block *sb;
650 
651         if (!bdev)
652                 return NULL;
653 
654         spin_lock(&sb_lock);
655 rescan:
656         list_for_each_entry(sb, &super_blocks, s_list) {
657                 if (hlist_unhashed(&sb->s_instances))
658                         continue;
659                 if (sb->s_bdev == bdev) {
660                         sb->s_count++;
661                         spin_unlock(&sb_lock);
662                         if (!excl)
663                                 down_read(&sb->s_umount);
664                         else
665                                 down_write(&sb->s_umount);
666                         /* still alive? */
667                         if (sb->s_root && (sb->s_flags & SB_BORN))
668                                 return sb;
669                         if (!excl)
670                                 up_read(&sb->s_umount);
671                         else
672                                 up_write(&sb->s_umount);
673                         /* nope, got unmounted */
674                         spin_lock(&sb_lock);
675                         __put_super(sb);
676                         goto rescan;
677                 }
678         }
679         spin_unlock(&sb_lock);
680         return NULL;
681 }
682 
683 /**
684  *      get_super - get the superblock of a device
685  *      @bdev: device to get the superblock for
686  *
687  *      Scans the superblock list and finds the superblock of the file system
688  *      mounted on the device given. %NULL is returned if no match is found.
689  */
690 struct super_block *get_super(struct block_device *bdev)
691 {
692         return __get_super(bdev, false);
693 }
694 EXPORT_SYMBOL(get_super);
695 
696 static struct super_block *__get_super_thawed(struct block_device *bdev,
697                                               bool excl)
698 {
699         while (1) {
700                 struct super_block *s = __get_super(bdev, excl);
701                 if (!s || s->s_writers.frozen == SB_UNFROZEN)
702                         return s;
703                 if (!excl)
704                         up_read(&s->s_umount);
705                 else
706                         up_write(&s->s_umount);
707                 wait_event(s->s_writers.wait_unfrozen,
708                            s->s_writers.frozen == SB_UNFROZEN);
709                 put_super(s);
710         }
711 }
712 
713 /**
714  *      get_super_thawed - get thawed superblock of a device
715  *      @bdev: device to get the superblock for
716  *
717  *      Scans the superblock list and finds the superblock of the file system
718  *      mounted on the device. The superblock is returned once it is thawed
719  *      (or immediately if it was not frozen). %NULL is returned if no match
720  *      is found.
721  */
722 struct super_block *get_super_thawed(struct block_device *bdev)
723 {
724         return __get_super_thawed(bdev, false);
725 }
726 EXPORT_SYMBOL(get_super_thawed);
727 
728 /**
729  *      get_super_exclusive_thawed - get thawed superblock of a device
730  *      @bdev: device to get the superblock for
731  *
732  *      Scans the superblock list and finds the superblock of the file system
733  *      mounted on the device. The superblock is returned once it is thawed
734  *      (or immediately if it was not frozen) and s_umount semaphore is held
735  *      in exclusive mode. %NULL is returned if no match is found.
736  */
737 struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
738 {
739         return __get_super_thawed(bdev, true);
740 }
741 EXPORT_SYMBOL(get_super_exclusive_thawed);
742 
743 /**
744  * get_active_super - get an active reference to the superblock of a device
745  * @bdev: device to get the superblock for
746  *
747  * Scans the superblock list and finds the superblock of the file system
748  * mounted on the device given.  Returns the superblock with an active
749  * reference or %NULL if none was found.
750  */
751 struct super_block *get_active_super(struct block_device *bdev)
752 {
753         struct super_block *sb;
754 
755         if (!bdev)
756                 return NULL;
757 
758 restart:
759         spin_lock(&sb_lock);
760         list_for_each_entry(sb, &super_blocks, s_list) {
761                 if (hlist_unhashed(&sb->s_instances))
762                         continue;
763                 if (sb->s_bdev == bdev) {
764                         if (!grab_super(sb))
765                                 goto restart;
766                         up_write(&sb->s_umount);
767                         return sb;
768                 }
769         }
770         spin_unlock(&sb_lock);
771         return NULL;
772 }
773 
774 struct super_block *user_get_super(dev_t dev)
775 {
776         struct super_block *sb;
777 
778         spin_lock(&sb_lock);
779 rescan:
780         list_for_each_entry(sb, &super_blocks, s_list) {
781                 if (hlist_unhashed(&sb->s_instances))
782                         continue;
783                 if (sb->s_dev ==  dev) {
784                         sb->s_count++;
785                         spin_unlock(&sb_lock);
786                         down_read(&sb->s_umount);
787                         /* still alive? */
788                         if (sb->s_root && (sb->s_flags & SB_BORN))
789                                 return sb;
790                         up_read(&sb->s_umount);
791                         /* nope, got unmounted */
792                         spin_lock(&sb_lock);
793                         __put_super(sb);
794                         goto rescan;
795                 }
796         }
797         spin_unlock(&sb_lock);
798         return NULL;
799 }
800 
801 /**
802  *      do_remount_sb - asks filesystem to change mount options.
803  *      @sb:    superblock in question
804  *      @sb_flags: revised superblock flags
805  *      @data:  the rest of options
806  *      @force: whether or not to force the change
807  *
808  *      Alters the mount options of a mounted file system.
809  */
810 int do_remount_sb(struct super_block *sb, int sb_flags, void *data, int force)
811 {
812         int retval;
813         int remount_ro;
814 
815         if (sb->s_writers.frozen != SB_UNFROZEN)
816                 return -EBUSY;
817 
818 #ifdef CONFIG_BLOCK
819         if (!(sb_flags & SB_RDONLY) && bdev_read_only(sb->s_bdev))
820                 return -EACCES;
821 #endif
822 
823         remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
824 
825         if (remount_ro) {
826                 if (!hlist_empty(&sb->s_pins)) {
827                         up_write(&sb->s_umount);
828                         group_pin_kill(&sb->s_pins);
829                         down_write(&sb->s_umount);
830                         if (!sb->s_root)
831                                 return 0;
832                         if (sb->s_writers.frozen != SB_UNFROZEN)
833                                 return -EBUSY;
834                         remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
835                 }
836         }
837         shrink_dcache_sb(sb);
838 
839         /* If we are remounting RDONLY and current sb is read/write,
840            make sure there are no rw files opened */
841         if (remount_ro) {
842                 if (force) {
843                         sb->s_readonly_remount = 1;
844                         smp_wmb();
845                 } else {
846                         retval = sb_prepare_remount_readonly(sb);
847                         if (retval)
848                                 return retval;
849                 }
850         }
851 
852         if (sb->s_op->remount_fs) {
853                 retval = sb->s_op->remount_fs(sb, &sb_flags, data);
854                 if (retval) {
855                         if (!force)
856                                 goto cancel_readonly;
857                         /* If forced remount, go ahead despite any errors */
858                         WARN(1, "forced remount of a %s fs returned %i\n",
859                              sb->s_type->name, retval);
860                 }
861         }
862         sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (sb_flags & MS_RMT_MASK);
863         /* Needs to be ordered wrt mnt_is_readonly() */
864         smp_wmb();
865         sb->s_readonly_remount = 0;
866 
867         /*
868          * Some filesystems modify their metadata via some other path than the
869          * bdev buffer cache (eg. use a private mapping, or directories in
870          * pagecache, etc). Also file data modifications go via their own
871          * mappings. So If we try to mount readonly then copy the filesystem
872          * from bdev, we could get stale data, so invalidate it to give a best
873          * effort at coherency.
874          */
875         if (remount_ro && sb->s_bdev)
876                 invalidate_bdev(sb->s_bdev);
877         return 0;
878 
879 cancel_readonly:
880         sb->s_readonly_remount = 0;
881         return retval;
882 }
883 
884 static void do_emergency_remount(struct work_struct *work)
885 {
886         struct super_block *sb, *p = NULL;
887 
888         spin_lock(&sb_lock);
889         list_for_each_entry(sb, &super_blocks, s_list) {
890                 if (hlist_unhashed(&sb->s_instances))
891                         continue;
892                 sb->s_count++;
893                 spin_unlock(&sb_lock);
894                 down_write(&sb->s_umount);
895                 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
896                     !sb_rdonly(sb)) {
897                         /*
898                          * What lock protects sb->s_flags??
899                          */
900                         do_remount_sb(sb, SB_RDONLY, NULL, 1);
901                 }
902                 up_write(&sb->s_umount);
903                 spin_lock(&sb_lock);
904                 if (p)
905                         __put_super(p);
906                 p = sb;
907         }
908         if (p)
909                 __put_super(p);
910         spin_unlock(&sb_lock);
911         kfree(work);
912         printk("Emergency Remount complete\n");
913 }
914 
915 void emergency_remount(void)
916 {
917         struct work_struct *work;
918 
919         work = kmalloc(sizeof(*work), GFP_ATOMIC);
920         if (work) {
921                 INIT_WORK(work, do_emergency_remount);
922                 schedule_work(work);
923         }
924 }
925 
926 /*
927  * Unnamed block devices are dummy devices used by virtual
928  * filesystems which don't use real block-devices.  -- jrs
929  */
930 
931 static DEFINE_IDA(unnamed_dev_ida);
932 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
933 /* Many userspace utilities consider an FSID of 0 invalid.
934  * Always return at least 1 from get_anon_bdev.
935  */
936 static int unnamed_dev_start = 1;
937 
938 int get_anon_bdev(dev_t *p)
939 {
940         int dev;
941         int error;
942 
943  retry:
944         if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
945                 return -ENOMEM;
946         spin_lock(&unnamed_dev_lock);
947         error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
948         if (!error)
949                 unnamed_dev_start = dev + 1;
950         spin_unlock(&unnamed_dev_lock);
951         if (error == -EAGAIN)
952                 /* We raced and lost with another CPU. */
953                 goto retry;
954         else if (error)
955                 return -EAGAIN;
956 
957         if (dev >= (1 << MINORBITS)) {
958                 spin_lock(&unnamed_dev_lock);
959                 ida_remove(&unnamed_dev_ida, dev);
960                 if (unnamed_dev_start > dev)
961                         unnamed_dev_start = dev;
962                 spin_unlock(&unnamed_dev_lock);
963                 return -EMFILE;
964         }
965         *p = MKDEV(0, dev & MINORMASK);
966         return 0;
967 }
968 EXPORT_SYMBOL(get_anon_bdev);
969 
970 void free_anon_bdev(dev_t dev)
971 {
972         int slot = MINOR(dev);
973         spin_lock(&unnamed_dev_lock);
974         ida_remove(&unnamed_dev_ida, slot);
975         if (slot < unnamed_dev_start)
976                 unnamed_dev_start = slot;
977         spin_unlock(&unnamed_dev_lock);
978 }
979 EXPORT_SYMBOL(free_anon_bdev);
980 
981 int set_anon_super(struct super_block *s, void *data)
982 {
983         return get_anon_bdev(&s->s_dev);
984 }
985 
986 EXPORT_SYMBOL(set_anon_super);
987 
988 void kill_anon_super(struct super_block *sb)
989 {
990         dev_t dev = sb->s_dev;
991         generic_shutdown_super(sb);
992         free_anon_bdev(dev);
993 }
994 
995 EXPORT_SYMBOL(kill_anon_super);
996 
997 void kill_litter_super(struct super_block *sb)
998 {
999         if (sb->s_root)
1000                 d_genocide(sb->s_root);
1001         kill_anon_super(sb);
1002 }
1003 
1004 EXPORT_SYMBOL(kill_litter_super);
1005 
1006 static int ns_test_super(struct super_block *sb, void *data)
1007 {
1008         return sb->s_fs_info == data;
1009 }
1010 
1011 static int ns_set_super(struct super_block *sb, void *data)
1012 {
1013         sb->s_fs_info = data;
1014         return set_anon_super(sb, NULL);
1015 }
1016 
1017 struct dentry *mount_ns(struct file_system_type *fs_type,
1018         int flags, void *data, void *ns, struct user_namespace *user_ns,
1019         int (*fill_super)(struct super_block *, void *, int))
1020 {
1021         struct super_block *sb;
1022 
1023         /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1024          * over the namespace.
1025          */
1026         if (!(flags & SB_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
1027                 return ERR_PTR(-EPERM);
1028 
1029         sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1030                          user_ns, ns);
1031         if (IS_ERR(sb))
1032                 return ERR_CAST(sb);
1033 
1034         if (!sb->s_root) {
1035                 int err;
1036                 err = fill_super(sb, data, flags & SB_SILENT ? 1 : 0);
1037                 if (err) {
1038                         deactivate_locked_super(sb);
1039                         return ERR_PTR(err);
1040                 }
1041 
1042                 sb->s_flags |= SB_ACTIVE;
1043         }
1044 
1045         return dget(sb->s_root);
1046 }
1047 
1048 EXPORT_SYMBOL(mount_ns);
1049 
1050 #ifdef CONFIG_BLOCK
1051 static int set_bdev_super(struct super_block *s, void *data)
1052 {
1053         s->s_bdev = data;
1054         s->s_dev = s->s_bdev->bd_dev;
1055         s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1056 
1057         return 0;
1058 }
1059 
1060 static int test_bdev_super(struct super_block *s, void *data)
1061 {
1062         return (void *)s->s_bdev == data;
1063 }
1064 
1065 struct dentry *mount_bdev(struct file_system_type *fs_type,
1066         int flags, const char *dev_name, void *data,
1067         int (*fill_super)(struct super_block *, void *, int))
1068 {
1069         struct block_device *bdev;
1070         struct super_block *s;
1071         fmode_t mode = FMODE_READ | FMODE_EXCL;
1072         int error = 0;
1073 
1074         if (!(flags & SB_RDONLY))
1075                 mode |= FMODE_WRITE;
1076 
1077         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1078         if (IS_ERR(bdev))
1079                 return ERR_CAST(bdev);
1080 
1081         /*
1082          * once the super is inserted into the list by sget, s_umount
1083          * will protect the lockfs code from trying to start a snapshot
1084          * while we are mounting
1085          */
1086         mutex_lock(&bdev->bd_fsfreeze_mutex);
1087         if (bdev->bd_fsfreeze_count > 0) {
1088                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1089                 error = -EBUSY;
1090                 goto error_bdev;
1091         }
1092         s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
1093                  bdev);
1094         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1095         if (IS_ERR(s))
1096                 goto error_s;
1097 
1098         if (s->s_root) {
1099                 if ((flags ^ s->s_flags) & SB_RDONLY) {
1100                         deactivate_locked_super(s);
1101                         error = -EBUSY;
1102                         goto error_bdev;
1103                 }
1104 
1105                 /*
1106                  * s_umount nests inside bd_mutex during
1107                  * __invalidate_device().  blkdev_put() acquires
1108                  * bd_mutex and can't be called under s_umount.  Drop
1109                  * s_umount temporarily.  This is safe as we're
1110                  * holding an active reference.
1111                  */
1112                 up_write(&s->s_umount);
1113                 blkdev_put(bdev, mode);
1114                 down_write(&s->s_umount);
1115         } else {
1116                 s->s_mode = mode;
1117                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1118                 sb_set_blocksize(s, block_size(bdev));
1119                 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1120                 if (error) {
1121                         deactivate_locked_super(s);
1122                         goto error;
1123                 }
1124 
1125                 s->s_flags |= SB_ACTIVE;
1126                 bdev->bd_super = s;
1127         }
1128 
1129         return dget(s->s_root);
1130 
1131 error_s:
1132         error = PTR_ERR(s);
1133 error_bdev:
1134         blkdev_put(bdev, mode);
1135 error:
1136         return ERR_PTR(error);
1137 }
1138 EXPORT_SYMBOL(mount_bdev);
1139 
1140 void kill_block_super(struct super_block *sb)
1141 {
1142         struct block_device *bdev = sb->s_bdev;
1143         fmode_t mode = sb->s_mode;
1144 
1145         bdev->bd_super = NULL;
1146         generic_shutdown_super(sb);
1147         sync_blockdev(bdev);
1148         WARN_ON_ONCE(!(mode & FMODE_EXCL));
1149         blkdev_put(bdev, mode | FMODE_EXCL);
1150 }
1151 
1152 EXPORT_SYMBOL(kill_block_super);
1153 #endif
1154 
1155 struct dentry *mount_nodev(struct file_system_type *fs_type,
1156         int flags, void *data,
1157         int (*fill_super)(struct super_block *, void *, int))
1158 {
1159         int error;
1160         struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1161 
1162         if (IS_ERR(s))
1163                 return ERR_CAST(s);
1164 
1165         error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1166         if (error) {
1167                 deactivate_locked_super(s);
1168                 return ERR_PTR(error);
1169         }
1170         s->s_flags |= SB_ACTIVE;
1171         return dget(s->s_root);
1172 }
1173 EXPORT_SYMBOL(mount_nodev);
1174 
1175 static int compare_single(struct super_block *s, void *p)
1176 {
1177         return 1;
1178 }
1179 
1180 struct dentry *mount_single(struct file_system_type *fs_type,
1181         int flags, void *data,
1182         int (*fill_super)(struct super_block *, void *, int))
1183 {
1184         struct super_block *s;
1185         int error;
1186 
1187         s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1188         if (IS_ERR(s))
1189                 return ERR_CAST(s);
1190         if (!s->s_root) {
1191                 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1192                 if (error) {
1193                         deactivate_locked_super(s);
1194                         return ERR_PTR(error);
1195                 }
1196                 s->s_flags |= SB_ACTIVE;
1197         } else {
1198                 do_remount_sb(s, flags, data, 0);
1199         }
1200         return dget(s->s_root);
1201 }
1202 EXPORT_SYMBOL(mount_single);
1203 
1204 struct dentry *
1205 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1206 {
1207         struct dentry *root;
1208         struct super_block *sb;
1209         char *secdata = NULL;
1210         int error = -ENOMEM;
1211 
1212         if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1213                 secdata = alloc_secdata();
1214                 if (!secdata)
1215                         goto out;
1216 
1217                 error = security_sb_copy_data(data, secdata);
1218                 if (error)
1219                         goto out_free_secdata;
1220         }
1221 
1222         root = type->mount(type, flags, name, data);
1223         if (IS_ERR(root)) {
1224                 error = PTR_ERR(root);
1225                 goto out_free_secdata;
1226         }
1227         sb = root->d_sb;
1228         BUG_ON(!sb);
1229         WARN_ON(!sb->s_bdi);
1230         sb->s_flags |= SB_BORN;
1231 
1232         error = security_sb_kern_mount(sb, flags, secdata);
1233         if (error)
1234                 goto out_sb;
1235 
1236         /*
1237          * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1238          * but s_maxbytes was an unsigned long long for many releases. Throw
1239          * this warning for a little while to try and catch filesystems that
1240          * violate this rule.
1241          */
1242         WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1243                 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1244 
1245         up_write(&sb->s_umount);
1246         free_secdata(secdata);
1247         return root;
1248 out_sb:
1249         dput(root);
1250         deactivate_locked_super(sb);
1251 out_free_secdata:
1252         free_secdata(secdata);
1253 out:
1254         return ERR_PTR(error);
1255 }
1256 
1257 /*
1258  * Setup private BDI for given superblock. It gets automatically cleaned up
1259  * in generic_shutdown_super().
1260  */
1261 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1262 {
1263         struct backing_dev_info *bdi;
1264         int err;
1265         va_list args;
1266 
1267         bdi = bdi_alloc(GFP_KERNEL);
1268         if (!bdi)
1269                 return -ENOMEM;
1270 
1271         bdi->name = sb->s_type->name;
1272 
1273         va_start(args, fmt);
1274         err = bdi_register_va(bdi, fmt, args);
1275         va_end(args);
1276         if (err) {
1277                 bdi_put(bdi);
1278                 return err;
1279         }
1280         WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1281         sb->s_bdi = bdi;
1282 
1283         return 0;
1284 }
1285 EXPORT_SYMBOL(super_setup_bdi_name);
1286 
1287 /*
1288  * Setup private BDI for given superblock. I gets automatically cleaned up
1289  * in generic_shutdown_super().
1290  */
1291 int super_setup_bdi(struct super_block *sb)
1292 {
1293         static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1294 
1295         return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1296                                     atomic_long_inc_return(&bdi_seq));
1297 }
1298 EXPORT_SYMBOL(super_setup_bdi);
1299 
1300 /*
1301  * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1302  * instead.
1303  */
1304 void __sb_end_write(struct super_block *sb, int level)
1305 {
1306         percpu_up_read(sb->s_writers.rw_sem + level-1);
1307 }
1308 EXPORT_SYMBOL(__sb_end_write);
1309 
1310 /*
1311  * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1312  * instead.
1313  */
1314 int __sb_start_write(struct super_block *sb, int level, bool wait)
1315 {
1316         bool force_trylock = false;
1317         int ret = 1;
1318 
1319 #ifdef CONFIG_LOCKDEP
1320         /*
1321          * We want lockdep to tell us about possible deadlocks with freezing
1322          * but it's it bit tricky to properly instrument it. Getting a freeze
1323          * protection works as getting a read lock but there are subtle
1324          * problems. XFS for example gets freeze protection on internal level
1325          * twice in some cases, which is OK only because we already hold a
1326          * freeze protection also on higher level. Due to these cases we have
1327          * to use wait == F (trylock mode) which must not fail.
1328          */
1329         if (wait) {
1330                 int i;
1331 
1332                 for (i = 0; i < level - 1; i++)
1333                         if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
1334                                 force_trylock = true;
1335                                 break;
1336                         }
1337         }
1338 #endif
1339         if (wait && !force_trylock)
1340                 percpu_down_read(sb->s_writers.rw_sem + level-1);
1341         else
1342                 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1343 
1344         WARN_ON(force_trylock && !ret);
1345         return ret;
1346 }
1347 EXPORT_SYMBOL(__sb_start_write);
1348 
1349 /**
1350  * sb_wait_write - wait until all writers to given file system finish
1351  * @sb: the super for which we wait
1352  * @level: type of writers we wait for (normal vs page fault)
1353  *
1354  * This function waits until there are no writers of given type to given file
1355  * system.
1356  */
1357 static void sb_wait_write(struct super_block *sb, int level)
1358 {
1359         percpu_down_write(sb->s_writers.rw_sem + level-1);
1360 }
1361 
1362 /*
1363  * We are going to return to userspace and forget about these locks, the
1364  * ownership goes to the caller of thaw_super() which does unlock().
1365  */
1366 static void lockdep_sb_freeze_release(struct super_block *sb)
1367 {
1368         int level;
1369 
1370         for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1371                 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1372 }
1373 
1374 /*
1375  * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1376  */
1377 static void lockdep_sb_freeze_acquire(struct super_block *sb)
1378 {
1379         int level;
1380 
1381         for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1382                 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1383 }
1384 
1385 static void sb_freeze_unlock(struct super_block *sb)
1386 {
1387         int level;
1388 
1389         for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1390                 percpu_up_write(sb->s_writers.rw_sem + level);
1391 }
1392 
1393 /**
1394  * freeze_super - lock the filesystem and force it into a consistent state
1395  * @sb: the super to lock
1396  *
1397  * Syncs the super to make sure the filesystem is consistent and calls the fs's
1398  * freeze_fs.  Subsequent calls to this without first thawing the fs will return
1399  * -EBUSY.
1400  *
1401  * During this function, sb->s_writers.frozen goes through these values:
1402  *
1403  * SB_UNFROZEN: File system is normal, all writes progress as usual.
1404  *
1405  * SB_FREEZE_WRITE: The file system is in the process of being frozen.  New
1406  * writes should be blocked, though page faults are still allowed. We wait for
1407  * all writes to complete and then proceed to the next stage.
1408  *
1409  * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1410  * but internal fs threads can still modify the filesystem (although they
1411  * should not dirty new pages or inodes), writeback can run etc. After waiting
1412  * for all running page faults we sync the filesystem which will clean all
1413  * dirty pages and inodes (no new dirty pages or inodes can be created when
1414  * sync is running).
1415  *
1416  * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1417  * modification are blocked (e.g. XFS preallocation truncation on inode
1418  * reclaim). This is usually implemented by blocking new transactions for
1419  * filesystems that have them and need this additional guard. After all
1420  * internal writers are finished we call ->freeze_fs() to finish filesystem
1421  * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1422  * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1423  *
1424  * sb->s_writers.frozen is protected by sb->s_umount.
1425  */
1426 int freeze_super(struct super_block *sb)
1427 {
1428         int ret;
1429 
1430         atomic_inc(&sb->s_active);
1431         down_write(&sb->s_umount);
1432         if (sb->s_writers.frozen != SB_UNFROZEN) {
1433                 deactivate_locked_super(sb);
1434                 return -EBUSY;
1435         }
1436 
1437         if (!(sb->s_flags & SB_BORN)) {
1438                 up_write(&sb->s_umount);
1439                 return 0;       /* sic - it's "nothing to do" */
1440         }
1441 
1442         if (sb_rdonly(sb)) {
1443                 /* Nothing to do really... */
1444                 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1445                 up_write(&sb->s_umount);
1446                 return 0;
1447         }
1448 
1449         sb->s_writers.frozen = SB_FREEZE_WRITE;
1450         /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1451         up_write(&sb->s_umount);
1452         sb_wait_write(sb, SB_FREEZE_WRITE);
1453         down_write(&sb->s_umount);
1454 
1455         /* Now we go and block page faults... */
1456         sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1457         sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1458 
1459         /* All writers are done so after syncing there won't be dirty data */
1460         sync_filesystem(sb);
1461 
1462         /* Now wait for internal filesystem counter */
1463         sb->s_writers.frozen = SB_FREEZE_FS;
1464         sb_wait_write(sb, SB_FREEZE_FS);
1465 
1466         if (sb->s_op->freeze_fs) {
1467                 ret = sb->s_op->freeze_fs(sb);
1468                 if (ret) {
1469                         printk(KERN_ERR
1470                                 "VFS:Filesystem freeze failed\n");
1471                         sb->s_writers.frozen = SB_UNFROZEN;
1472                         sb_freeze_unlock(sb);
1473                         wake_up(&sb->s_writers.wait_unfrozen);
1474                         deactivate_locked_super(sb);
1475                         return ret;
1476                 }
1477         }
1478         /*
1479          * For debugging purposes so that fs can warn if it sees write activity
1480          * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1481          */
1482         sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1483         lockdep_sb_freeze_release(sb);
1484         up_write(&sb->s_umount);
1485         return 0;
1486 }
1487 EXPORT_SYMBOL(freeze_super);
1488 
1489 /**
1490  * thaw_super -- unlock filesystem
1491  * @sb: the super to thaw
1492  *
1493  * Unlocks the filesystem and marks it writeable again after freeze_super().
1494  */
1495 int thaw_super(struct super_block *sb)
1496 {
1497         int error;
1498 
1499         down_write(&sb->s_umount);
1500         if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1501                 up_write(&sb->s_umount);
1502                 return -EINVAL;
1503         }
1504 
1505         if (sb_rdonly(sb)) {
1506                 sb->s_writers.frozen = SB_UNFROZEN;
1507                 goto out;
1508         }
1509 
1510         lockdep_sb_freeze_acquire(sb);
1511 
1512         if (sb->s_op->unfreeze_fs) {
1513                 error = sb->s_op->unfreeze_fs(sb);
1514                 if (error) {
1515                         printk(KERN_ERR
1516                                 "VFS:Filesystem thaw failed\n");
1517                         lockdep_sb_freeze_release(sb);
1518                         up_write(&sb->s_umount);
1519                         return error;
1520                 }
1521         }
1522 
1523         sb->s_writers.frozen = SB_UNFROZEN;
1524         sb_freeze_unlock(sb);
1525 out:
1526         wake_up(&sb->s_writers.wait_unfrozen);
1527         deactivate_locked_super(sb);
1528         return 0;
1529 }
1530 EXPORT_SYMBOL(thaw_super);
1531 

~ [ 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