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

TOMOYO Linux Cross Reference
Linux/fs/super.c

Version: ~ [ linux-5.0-rc6 ] ~ [ linux-4.20.7 ] ~ [ linux-4.19.20 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.98 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.155 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.174 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.134 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.62 ] ~ [ 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-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ 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 <uapi/linux/mount.h>
 39 #include "internal.h"
 40 
 41 static int thaw_super_locked(struct super_block *sb);
 42 
 43 static LIST_HEAD(super_blocks);
 44 static DEFINE_SPINLOCK(sb_lock);
 45 
 46 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
 47         "sb_writers",
 48         "sb_pagefaults",
 49         "sb_internal",
 50 };
 51 
 52 /*
 53  * One thing we have to be careful of with a per-sb shrinker is that we don't
 54  * drop the last active reference to the superblock from within the shrinker.
 55  * If that happens we could trigger unregistering the shrinker from within the
 56  * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
 57  * take a passive reference to the superblock to avoid this from occurring.
 58  */
 59 static unsigned long super_cache_scan(struct shrinker *shrink,
 60                                       struct shrink_control *sc)
 61 {
 62         struct super_block *sb;
 63         long    fs_objects = 0;
 64         long    total_objects;
 65         long    freed = 0;
 66         long    dentries;
 67         long    inodes;
 68 
 69         sb = container_of(shrink, struct super_block, s_shrink);
 70 
 71         /*
 72          * Deadlock avoidance.  We may hold various FS locks, and we don't want
 73          * to recurse into the FS that called us in clear_inode() and friends..
 74          */
 75         if (!(sc->gfp_mask & __GFP_FS))
 76                 return SHRINK_STOP;
 77 
 78         if (!trylock_super(sb))
 79                 return SHRINK_STOP;
 80 
 81         if (sb->s_op->nr_cached_objects)
 82                 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
 83 
 84         inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
 85         dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
 86         total_objects = dentries + inodes + fs_objects + 1;
 87         if (!total_objects)
 88                 total_objects = 1;
 89 
 90         /* proportion the scan between the caches */
 91         dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
 92         inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
 93         fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
 94 
 95         /*
 96          * prune the dcache first as the icache is pinned by it, then
 97          * prune the icache, followed by the filesystem specific caches
 98          *
 99          * Ensure that we always scan at least one object - memcg kmem
100          * accounting uses this to fully empty the caches.
101          */
102         sc->nr_to_scan = dentries + 1;
103         freed = prune_dcache_sb(sb, sc);
104         sc->nr_to_scan = inodes + 1;
105         freed += prune_icache_sb(sb, sc);
106 
107         if (fs_objects) {
108                 sc->nr_to_scan = fs_objects + 1;
109                 freed += sb->s_op->free_cached_objects(sb, sc);
110         }
111 
112         up_read(&sb->s_umount);
113         return freed;
114 }
115 
116 static unsigned long super_cache_count(struct shrinker *shrink,
117                                        struct shrink_control *sc)
118 {
119         struct super_block *sb;
120         long    total_objects = 0;
121 
122         sb = container_of(shrink, struct super_block, s_shrink);
123 
124         /*
125          * We don't call trylock_super() here as it is a scalability bottleneck,
126          * so we're exposed to partial setup state. The shrinker rwsem does not
127          * protect filesystem operations backing list_lru_shrink_count() or
128          * s_op->nr_cached_objects(). Counts can change between
129          * super_cache_count and super_cache_scan, so we really don't need locks
130          * here.
131          *
132          * However, if we are currently mounting the superblock, the underlying
133          * filesystem might be in a state of partial construction and hence it
134          * is dangerous to access it.  trylock_super() uses a SB_BORN check to
135          * avoid this situation, so do the same here. The memory barrier is
136          * matched with the one in mount_fs() as we don't hold locks here.
137          */
138         if (!(sb->s_flags & SB_BORN))
139                 return 0;
140         smp_rmb();
141 
142         if (sb->s_op && sb->s_op->nr_cached_objects)
143                 total_objects = sb->s_op->nr_cached_objects(sb, sc);
144 
145         total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
146         total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
147 
148         if (!total_objects)
149                 return SHRINK_EMPTY;
150 
151         total_objects = vfs_pressure_ratio(total_objects);
152         return total_objects;
153 }
154 
155 static void destroy_super_work(struct work_struct *work)
156 {
157         struct super_block *s = container_of(work, struct super_block,
158                                                         destroy_work);
159         int i;
160 
161         for (i = 0; i < SB_FREEZE_LEVELS; i++)
162                 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
163         kfree(s);
164 }
165 
166 static void destroy_super_rcu(struct rcu_head *head)
167 {
168         struct super_block *s = container_of(head, struct super_block, rcu);
169         INIT_WORK(&s->destroy_work, destroy_super_work);
170         schedule_work(&s->destroy_work);
171 }
172 
173 /* Free a superblock that has never been seen by anyone */
174 static void destroy_unused_super(struct super_block *s)
175 {
176         if (!s)
177                 return;
178         up_write(&s->s_umount);
179         list_lru_destroy(&s->s_dentry_lru);
180         list_lru_destroy(&s->s_inode_lru);
181         security_sb_free(s);
182         put_user_ns(s->s_user_ns);
183         kfree(s->s_subtype);
184         free_prealloced_shrinker(&s->s_shrink);
185         /* no delays needed */
186         destroy_super_work(&s->destroy_work);
187 }
188 
189 /**
190  *      alloc_super     -       create new superblock
191  *      @type:  filesystem type superblock should belong to
192  *      @flags: the mount flags
193  *      @user_ns: User namespace for the super_block
194  *
195  *      Allocates and initializes a new &struct super_block.  alloc_super()
196  *      returns a pointer new superblock or %NULL if allocation had failed.
197  */
198 static struct super_block *alloc_super(struct file_system_type *type, int flags,
199                                        struct user_namespace *user_ns)
200 {
201         struct super_block *s = kzalloc(sizeof(struct super_block),  GFP_USER);
202         static const struct super_operations default_op;
203         int i;
204 
205         if (!s)
206                 return NULL;
207 
208         INIT_LIST_HEAD(&s->s_mounts);
209         s->s_user_ns = get_user_ns(user_ns);
210         init_rwsem(&s->s_umount);
211         lockdep_set_class(&s->s_umount, &type->s_umount_key);
212         /*
213          * sget() can have s_umount recursion.
214          *
215          * When it cannot find a suitable sb, it allocates a new
216          * one (this one), and tries again to find a suitable old
217          * one.
218          *
219          * In case that succeeds, it will acquire the s_umount
220          * lock of the old one. Since these are clearly distrinct
221          * locks, and this object isn't exposed yet, there's no
222          * risk of deadlocks.
223          *
224          * Annotate this by putting this lock in a different
225          * subclass.
226          */
227         down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
228 
229         if (security_sb_alloc(s))
230                 goto fail;
231 
232         for (i = 0; i < SB_FREEZE_LEVELS; i++) {
233                 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
234                                         sb_writers_name[i],
235                                         &type->s_writers_key[i]))
236                         goto fail;
237         }
238         init_waitqueue_head(&s->s_writers.wait_unfrozen);
239         s->s_bdi = &noop_backing_dev_info;
240         s->s_flags = flags;
241         if (s->s_user_ns != &init_user_ns)
242                 s->s_iflags |= SB_I_NODEV;
243         INIT_HLIST_NODE(&s->s_instances);
244         INIT_HLIST_BL_HEAD(&s->s_roots);
245         mutex_init(&s->s_sync_lock);
246         INIT_LIST_HEAD(&s->s_inodes);
247         spin_lock_init(&s->s_inode_list_lock);
248         INIT_LIST_HEAD(&s->s_inodes_wb);
249         spin_lock_init(&s->s_inode_wblist_lock);
250 
251         s->s_count = 1;
252         atomic_set(&s->s_active, 1);
253         mutex_init(&s->s_vfs_rename_mutex);
254         lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
255         init_rwsem(&s->s_dquot.dqio_sem);
256         s->s_maxbytes = MAX_NON_LFS;
257         s->s_op = &default_op;
258         s->s_time_gran = 1000000000;
259         s->cleancache_poolid = CLEANCACHE_NO_POOL;
260 
261         s->s_shrink.seeks = DEFAULT_SEEKS;
262         s->s_shrink.scan_objects = super_cache_scan;
263         s->s_shrink.count_objects = super_cache_count;
264         s->s_shrink.batch = 1024;
265         s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
266         if (prealloc_shrinker(&s->s_shrink))
267                 goto fail;
268         if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
269                 goto fail;
270         if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
271                 goto fail;
272         return s;
273 
274 fail:
275         destroy_unused_super(s);
276         return NULL;
277 }
278 
279 /* Superblock refcounting  */
280 
281 /*
282  * Drop a superblock's refcount.  The caller must hold sb_lock.
283  */
284 static void __put_super(struct super_block *s)
285 {
286         if (!--s->s_count) {
287                 list_del_init(&s->s_list);
288                 WARN_ON(s->s_dentry_lru.node);
289                 WARN_ON(s->s_inode_lru.node);
290                 WARN_ON(!list_empty(&s->s_mounts));
291                 security_sb_free(s);
292                 put_user_ns(s->s_user_ns);
293                 kfree(s->s_subtype);
294                 call_rcu(&s->rcu, destroy_super_rcu);
295         }
296 }
297 
298 /**
299  *      put_super       -       drop a temporary reference to superblock
300  *      @sb: superblock in question
301  *
302  *      Drops a temporary reference, frees superblock if there's no
303  *      references left.
304  */
305 static void put_super(struct super_block *sb)
306 {
307         spin_lock(&sb_lock);
308         __put_super(sb);
309         spin_unlock(&sb_lock);
310 }
311 
312 
313 /**
314  *      deactivate_locked_super -       drop an active reference to superblock
315  *      @s: superblock to deactivate
316  *
317  *      Drops an active reference to superblock, converting it into a temporary
318  *      one if there is no other active references left.  In that case we
319  *      tell fs driver to shut it down and drop the temporary reference we
320  *      had just acquired.
321  *
322  *      Caller holds exclusive lock on superblock; that lock is released.
323  */
324 void deactivate_locked_super(struct super_block *s)
325 {
326         struct file_system_type *fs = s->s_type;
327         if (atomic_dec_and_test(&s->s_active)) {
328                 cleancache_invalidate_fs(s);
329                 unregister_shrinker(&s->s_shrink);
330                 fs->kill_sb(s);
331 
332                 /*
333                  * Since list_lru_destroy() may sleep, we cannot call it from
334                  * put_super(), where we hold the sb_lock. Therefore we destroy
335                  * the lru lists right now.
336                  */
337                 list_lru_destroy(&s->s_dentry_lru);
338                 list_lru_destroy(&s->s_inode_lru);
339 
340                 put_filesystem(fs);
341                 put_super(s);
342         } else {
343                 up_write(&s->s_umount);
344         }
345 }
346 
347 EXPORT_SYMBOL(deactivate_locked_super);
348 
349 /**
350  *      deactivate_super        -       drop an active reference to superblock
351  *      @s: superblock to deactivate
352  *
353  *      Variant of deactivate_locked_super(), except that superblock is *not*
354  *      locked by caller.  If we are going to drop the final active reference,
355  *      lock will be acquired prior to that.
356  */
357 void deactivate_super(struct super_block *s)
358 {
359         if (!atomic_add_unless(&s->s_active, -1, 1)) {
360                 down_write(&s->s_umount);
361                 deactivate_locked_super(s);
362         }
363 }
364 
365 EXPORT_SYMBOL(deactivate_super);
366 
367 /**
368  *      grab_super - acquire an active reference
369  *      @s: reference we are trying to make active
370  *
371  *      Tries to acquire an active reference.  grab_super() is used when we
372  *      had just found a superblock in super_blocks or fs_type->fs_supers
373  *      and want to turn it into a full-blown active reference.  grab_super()
374  *      is called with sb_lock held and drops it.  Returns 1 in case of
375  *      success, 0 if we had failed (superblock contents was already dead or
376  *      dying when grab_super() had been called).  Note that this is only
377  *      called for superblocks not in rundown mode (== ones still on ->fs_supers
378  *      of their type), so increment of ->s_count is OK here.
379  */
380 static int grab_super(struct super_block *s) __releases(sb_lock)
381 {
382         s->s_count++;
383         spin_unlock(&sb_lock);
384         down_write(&s->s_umount);
385         if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
386                 put_super(s);
387                 return 1;
388         }
389         up_write(&s->s_umount);
390         put_super(s);
391         return 0;
392 }
393 
394 /*
395  *      trylock_super - try to grab ->s_umount shared
396  *      @sb: reference we are trying to grab
397  *
398  *      Try to prevent fs shutdown.  This is used in places where we
399  *      cannot take an active reference but we need to ensure that the
400  *      filesystem is not shut down while we are working on it. It returns
401  *      false if we cannot acquire s_umount or if we lose the race and
402  *      filesystem already got into shutdown, and returns true with the s_umount
403  *      lock held in read mode in case of success. On successful return,
404  *      the caller must drop the s_umount lock when done.
405  *
406  *      Note that unlike get_super() et.al. this one does *not* bump ->s_count.
407  *      The reason why it's safe is that we are OK with doing trylock instead
408  *      of down_read().  There's a couple of places that are OK with that, but
409  *      it's very much not a general-purpose interface.
410  */
411 bool trylock_super(struct super_block *sb)
412 {
413         if (down_read_trylock(&sb->s_umount)) {
414                 if (!hlist_unhashed(&sb->s_instances) &&
415                     sb->s_root && (sb->s_flags & SB_BORN))
416                         return true;
417                 up_read(&sb->s_umount);
418         }
419 
420         return false;
421 }
422 
423 /**
424  *      generic_shutdown_super  -       common helper for ->kill_sb()
425  *      @sb: superblock to kill
426  *
427  *      generic_shutdown_super() does all fs-independent work on superblock
428  *      shutdown.  Typical ->kill_sb() should pick all fs-specific objects
429  *      that need destruction out of superblock, call generic_shutdown_super()
430  *      and release aforementioned objects.  Note: dentries and inodes _are_
431  *      taken care of and do not need specific handling.
432  *
433  *      Upon calling this function, the filesystem may no longer alter or
434  *      rearrange the set of dentries belonging to this super_block, nor may it
435  *      change the attachments of dentries to inodes.
436  */
437 void generic_shutdown_super(struct super_block *sb)
438 {
439         const struct super_operations *sop = sb->s_op;
440 
441         if (sb->s_root) {
442                 shrink_dcache_for_umount(sb);
443                 sync_filesystem(sb);
444                 sb->s_flags &= ~SB_ACTIVE;
445 
446                 fsnotify_sb_delete(sb);
447                 cgroup_writeback_umount();
448 
449                 evict_inodes(sb);
450 
451                 if (sb->s_dio_done_wq) {
452                         destroy_workqueue(sb->s_dio_done_wq);
453                         sb->s_dio_done_wq = NULL;
454                 }
455 
456                 if (sop->put_super)
457                         sop->put_super(sb);
458 
459                 if (!list_empty(&sb->s_inodes)) {
460                         printk("VFS: Busy inodes after unmount of %s. "
461                            "Self-destruct in 5 seconds.  Have a nice day...\n",
462                            sb->s_id);
463                 }
464         }
465         spin_lock(&sb_lock);
466         /* should be initialized for __put_super_and_need_restart() */
467         hlist_del_init(&sb->s_instances);
468         spin_unlock(&sb_lock);
469         up_write(&sb->s_umount);
470         if (sb->s_bdi != &noop_backing_dev_info) {
471                 bdi_put(sb->s_bdi);
472                 sb->s_bdi = &noop_backing_dev_info;
473         }
474 }
475 
476 EXPORT_SYMBOL(generic_shutdown_super);
477 
478 /**
479  *      sget_userns -   find or create a superblock
480  *      @type:  filesystem type superblock should belong to
481  *      @test:  comparison callback
482  *      @set:   setup callback
483  *      @flags: mount flags
484  *      @user_ns: User namespace for the super_block
485  *      @data:  argument to each of them
486  */
487 struct super_block *sget_userns(struct file_system_type *type,
488                         int (*test)(struct super_block *,void *),
489                         int (*set)(struct super_block *,void *),
490                         int flags, struct user_namespace *user_ns,
491                         void *data)
492 {
493         struct super_block *s = NULL;
494         struct super_block *old;
495         int err;
496 
497         if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) &&
498             !(type->fs_flags & FS_USERNS_MOUNT) &&
499             !capable(CAP_SYS_ADMIN))
500                 return ERR_PTR(-EPERM);
501 retry:
502         spin_lock(&sb_lock);
503         if (test) {
504                 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
505                         if (!test(old, data))
506                                 continue;
507                         if (user_ns != old->s_user_ns) {
508                                 spin_unlock(&sb_lock);
509                                 destroy_unused_super(s);
510                                 return ERR_PTR(-EBUSY);
511                         }
512                         if (!grab_super(old))
513                                 goto retry;
514                         destroy_unused_super(s);
515                         return old;
516                 }
517         }
518         if (!s) {
519                 spin_unlock(&sb_lock);
520                 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
521                 if (!s)
522                         return ERR_PTR(-ENOMEM);
523                 goto retry;
524         }
525 
526         err = set(s, data);
527         if (err) {
528                 spin_unlock(&sb_lock);
529                 destroy_unused_super(s);
530                 return ERR_PTR(err);
531         }
532         s->s_type = type;
533         strlcpy(s->s_id, type->name, sizeof(s->s_id));
534         list_add_tail(&s->s_list, &super_blocks);
535         hlist_add_head(&s->s_instances, &type->fs_supers);
536         spin_unlock(&sb_lock);
537         get_filesystem(type);
538         register_shrinker_prepared(&s->s_shrink);
539         return s;
540 }
541 
542 EXPORT_SYMBOL(sget_userns);
543 
544 /**
545  *      sget    -       find or create a superblock
546  *      @type:    filesystem type superblock should belong to
547  *      @test:    comparison callback
548  *      @set:     setup callback
549  *      @flags:   mount flags
550  *      @data:    argument to each of them
551  */
552 struct super_block *sget(struct file_system_type *type,
553                         int (*test)(struct super_block *,void *),
554                         int (*set)(struct super_block *,void *),
555                         int flags,
556                         void *data)
557 {
558         struct user_namespace *user_ns = current_user_ns();
559 
560         /* We don't yet pass the user namespace of the parent
561          * mount through to here so always use &init_user_ns
562          * until that changes.
563          */
564         if (flags & SB_SUBMOUNT)
565                 user_ns = &init_user_ns;
566 
567         /* Ensure the requestor has permissions over the target filesystem */
568         if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
569                 return ERR_PTR(-EPERM);
570 
571         return sget_userns(type, test, set, flags, user_ns, data);
572 }
573 
574 EXPORT_SYMBOL(sget);
575 
576 void drop_super(struct super_block *sb)
577 {
578         up_read(&sb->s_umount);
579         put_super(sb);
580 }
581 
582 EXPORT_SYMBOL(drop_super);
583 
584 void drop_super_exclusive(struct super_block *sb)
585 {
586         up_write(&sb->s_umount);
587         put_super(sb);
588 }
589 EXPORT_SYMBOL(drop_super_exclusive);
590 
591 static void __iterate_supers(void (*f)(struct super_block *))
592 {
593         struct super_block *sb, *p = NULL;
594 
595         spin_lock(&sb_lock);
596         list_for_each_entry(sb, &super_blocks, s_list) {
597                 if (hlist_unhashed(&sb->s_instances))
598                         continue;
599                 sb->s_count++;
600                 spin_unlock(&sb_lock);
601 
602                 f(sb);
603 
604                 spin_lock(&sb_lock);
605                 if (p)
606                         __put_super(p);
607                 p = sb;
608         }
609         if (p)
610                 __put_super(p);
611         spin_unlock(&sb_lock);
612 }
613 /**
614  *      iterate_supers - call function for all active superblocks
615  *      @f: function to call
616  *      @arg: argument to pass to it
617  *
618  *      Scans the superblock list and calls given function, passing it
619  *      locked superblock and given argument.
620  */
621 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
622 {
623         struct super_block *sb, *p = NULL;
624 
625         spin_lock(&sb_lock);
626         list_for_each_entry(sb, &super_blocks, s_list) {
627                 if (hlist_unhashed(&sb->s_instances))
628                         continue;
629                 sb->s_count++;
630                 spin_unlock(&sb_lock);
631 
632                 down_read(&sb->s_umount);
633                 if (sb->s_root && (sb->s_flags & SB_BORN))
634                         f(sb, arg);
635                 up_read(&sb->s_umount);
636 
637                 spin_lock(&sb_lock);
638                 if (p)
639                         __put_super(p);
640                 p = sb;
641         }
642         if (p)
643                 __put_super(p);
644         spin_unlock(&sb_lock);
645 }
646 
647 /**
648  *      iterate_supers_type - call function for superblocks of given type
649  *      @type: fs type
650  *      @f: function to call
651  *      @arg: argument to pass to it
652  *
653  *      Scans the superblock list and calls given function, passing it
654  *      locked superblock and given argument.
655  */
656 void iterate_supers_type(struct file_system_type *type,
657         void (*f)(struct super_block *, void *), void *arg)
658 {
659         struct super_block *sb, *p = NULL;
660 
661         spin_lock(&sb_lock);
662         hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
663                 sb->s_count++;
664                 spin_unlock(&sb_lock);
665 
666                 down_read(&sb->s_umount);
667                 if (sb->s_root && (sb->s_flags & SB_BORN))
668                         f(sb, arg);
669                 up_read(&sb->s_umount);
670 
671                 spin_lock(&sb_lock);
672                 if (p)
673                         __put_super(p);
674                 p = sb;
675         }
676         if (p)
677                 __put_super(p);
678         spin_unlock(&sb_lock);
679 }
680 
681 EXPORT_SYMBOL(iterate_supers_type);
682 
683 static struct super_block *__get_super(struct block_device *bdev, bool excl)
684 {
685         struct super_block *sb;
686 
687         if (!bdev)
688                 return NULL;
689 
690         spin_lock(&sb_lock);
691 rescan:
692         list_for_each_entry(sb, &super_blocks, s_list) {
693                 if (hlist_unhashed(&sb->s_instances))
694                         continue;
695                 if (sb->s_bdev == bdev) {
696                         sb->s_count++;
697                         spin_unlock(&sb_lock);
698                         if (!excl)
699                                 down_read(&sb->s_umount);
700                         else
701                                 down_write(&sb->s_umount);
702                         /* still alive? */
703                         if (sb->s_root && (sb->s_flags & SB_BORN))
704                                 return sb;
705                         if (!excl)
706                                 up_read(&sb->s_umount);
707                         else
708                                 up_write(&sb->s_umount);
709                         /* nope, got unmounted */
710                         spin_lock(&sb_lock);
711                         __put_super(sb);
712                         goto rescan;
713                 }
714         }
715         spin_unlock(&sb_lock);
716         return NULL;
717 }
718 
719 /**
720  *      get_super - get the superblock of a device
721  *      @bdev: device to get the superblock for
722  *
723  *      Scans the superblock list and finds the superblock of the file system
724  *      mounted on the device given. %NULL is returned if no match is found.
725  */
726 struct super_block *get_super(struct block_device *bdev)
727 {
728         return __get_super(bdev, false);
729 }
730 EXPORT_SYMBOL(get_super);
731 
732 static struct super_block *__get_super_thawed(struct block_device *bdev,
733                                               bool excl)
734 {
735         while (1) {
736                 struct super_block *s = __get_super(bdev, excl);
737                 if (!s || s->s_writers.frozen == SB_UNFROZEN)
738                         return s;
739                 if (!excl)
740                         up_read(&s->s_umount);
741                 else
742                         up_write(&s->s_umount);
743                 wait_event(s->s_writers.wait_unfrozen,
744                            s->s_writers.frozen == SB_UNFROZEN);
745                 put_super(s);
746         }
747 }
748 
749 /**
750  *      get_super_thawed - get thawed superblock of a device
751  *      @bdev: device to get the superblock for
752  *
753  *      Scans the superblock list and finds the superblock of the file system
754  *      mounted on the device. The superblock is returned once it is thawed
755  *      (or immediately if it was not frozen). %NULL is returned if no match
756  *      is found.
757  */
758 struct super_block *get_super_thawed(struct block_device *bdev)
759 {
760         return __get_super_thawed(bdev, false);
761 }
762 EXPORT_SYMBOL(get_super_thawed);
763 
764 /**
765  *      get_super_exclusive_thawed - get thawed superblock of a device
766  *      @bdev: device to get the superblock for
767  *
768  *      Scans the superblock list and finds the superblock of the file system
769  *      mounted on the device. The superblock is returned once it is thawed
770  *      (or immediately if it was not frozen) and s_umount semaphore is held
771  *      in exclusive mode. %NULL is returned if no match is found.
772  */
773 struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
774 {
775         return __get_super_thawed(bdev, true);
776 }
777 EXPORT_SYMBOL(get_super_exclusive_thawed);
778 
779 /**
780  * get_active_super - get an active reference to 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.  Returns the superblock with an active
785  * reference or %NULL if none was found.
786  */
787 struct super_block *get_active_super(struct block_device *bdev)
788 {
789         struct super_block *sb;
790 
791         if (!bdev)
792                 return NULL;
793 
794 restart:
795         spin_lock(&sb_lock);
796         list_for_each_entry(sb, &super_blocks, s_list) {
797                 if (hlist_unhashed(&sb->s_instances))
798                         continue;
799                 if (sb->s_bdev == bdev) {
800                         if (!grab_super(sb))
801                                 goto restart;
802                         up_write(&sb->s_umount);
803                         return sb;
804                 }
805         }
806         spin_unlock(&sb_lock);
807         return NULL;
808 }
809 
810 struct super_block *user_get_super(dev_t dev)
811 {
812         struct super_block *sb;
813 
814         spin_lock(&sb_lock);
815 rescan:
816         list_for_each_entry(sb, &super_blocks, s_list) {
817                 if (hlist_unhashed(&sb->s_instances))
818                         continue;
819                 if (sb->s_dev ==  dev) {
820                         sb->s_count++;
821                         spin_unlock(&sb_lock);
822                         down_read(&sb->s_umount);
823                         /* still alive? */
824                         if (sb->s_root && (sb->s_flags & SB_BORN))
825                                 return sb;
826                         up_read(&sb->s_umount);
827                         /* nope, got unmounted */
828                         spin_lock(&sb_lock);
829                         __put_super(sb);
830                         goto rescan;
831                 }
832         }
833         spin_unlock(&sb_lock);
834         return NULL;
835 }
836 
837 /**
838  *      do_remount_sb - asks filesystem to change mount options.
839  *      @sb:    superblock in question
840  *      @sb_flags: revised superblock flags
841  *      @data:  the rest of options
842  *      @force: whether or not to force the change
843  *
844  *      Alters the mount options of a mounted file system.
845  */
846 int do_remount_sb(struct super_block *sb, int sb_flags, void *data, int force)
847 {
848         int retval;
849         int remount_ro;
850 
851         if (sb->s_writers.frozen != SB_UNFROZEN)
852                 return -EBUSY;
853 
854 #ifdef CONFIG_BLOCK
855         if (!(sb_flags & SB_RDONLY) && bdev_read_only(sb->s_bdev))
856                 return -EACCES;
857 #endif
858 
859         remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
860 
861         if (remount_ro) {
862                 if (!hlist_empty(&sb->s_pins)) {
863                         up_write(&sb->s_umount);
864                         group_pin_kill(&sb->s_pins);
865                         down_write(&sb->s_umount);
866                         if (!sb->s_root)
867                                 return 0;
868                         if (sb->s_writers.frozen != SB_UNFROZEN)
869                                 return -EBUSY;
870                         remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
871                 }
872         }
873         shrink_dcache_sb(sb);
874 
875         /* If we are remounting RDONLY and current sb is read/write,
876            make sure there are no rw files opened */
877         if (remount_ro) {
878                 if (force) {
879                         sb->s_readonly_remount = 1;
880                         smp_wmb();
881                 } else {
882                         retval = sb_prepare_remount_readonly(sb);
883                         if (retval)
884                                 return retval;
885                 }
886         }
887 
888         if (sb->s_op->remount_fs) {
889                 retval = sb->s_op->remount_fs(sb, &sb_flags, data);
890                 if (retval) {
891                         if (!force)
892                                 goto cancel_readonly;
893                         /* If forced remount, go ahead despite any errors */
894                         WARN(1, "forced remount of a %s fs returned %i\n",
895                              sb->s_type->name, retval);
896                 }
897         }
898         sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (sb_flags & MS_RMT_MASK);
899         /* Needs to be ordered wrt mnt_is_readonly() */
900         smp_wmb();
901         sb->s_readonly_remount = 0;
902 
903         /*
904          * Some filesystems modify their metadata via some other path than the
905          * bdev buffer cache (eg. use a private mapping, or directories in
906          * pagecache, etc). Also file data modifications go via their own
907          * mappings. So If we try to mount readonly then copy the filesystem
908          * from bdev, we could get stale data, so invalidate it to give a best
909          * effort at coherency.
910          */
911         if (remount_ro && sb->s_bdev)
912                 invalidate_bdev(sb->s_bdev);
913         return 0;
914 
915 cancel_readonly:
916         sb->s_readonly_remount = 0;
917         return retval;
918 }
919 
920 static void do_emergency_remount_callback(struct super_block *sb)
921 {
922         down_write(&sb->s_umount);
923         if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
924             !sb_rdonly(sb)) {
925                 /*
926                  * What lock protects sb->s_flags??
927                  */
928                 do_remount_sb(sb, SB_RDONLY, NULL, 1);
929         }
930         up_write(&sb->s_umount);
931 }
932 
933 static void do_emergency_remount(struct work_struct *work)
934 {
935         __iterate_supers(do_emergency_remount_callback);
936         kfree(work);
937         printk("Emergency Remount complete\n");
938 }
939 
940 void emergency_remount(void)
941 {
942         struct work_struct *work;
943 
944         work = kmalloc(sizeof(*work), GFP_ATOMIC);
945         if (work) {
946                 INIT_WORK(work, do_emergency_remount);
947                 schedule_work(work);
948         }
949 }
950 
951 static void do_thaw_all_callback(struct super_block *sb)
952 {
953         down_write(&sb->s_umount);
954         if (sb->s_root && sb->s_flags & SB_BORN) {
955                 emergency_thaw_bdev(sb);
956                 thaw_super_locked(sb);
957         } else {
958                 up_write(&sb->s_umount);
959         }
960 }
961 
962 static void do_thaw_all(struct work_struct *work)
963 {
964         __iterate_supers(do_thaw_all_callback);
965         kfree(work);
966         printk(KERN_WARNING "Emergency Thaw complete\n");
967 }
968 
969 /**
970  * emergency_thaw_all -- forcibly thaw every frozen filesystem
971  *
972  * Used for emergency unfreeze of all filesystems via SysRq
973  */
974 void emergency_thaw_all(void)
975 {
976         struct work_struct *work;
977 
978         work = kmalloc(sizeof(*work), GFP_ATOMIC);
979         if (work) {
980                 INIT_WORK(work, do_thaw_all);
981                 schedule_work(work);
982         }
983 }
984 
985 static DEFINE_IDA(unnamed_dev_ida);
986 
987 /**
988  * get_anon_bdev - Allocate a block device for filesystems which don't have one.
989  * @p: Pointer to a dev_t.
990  *
991  * Filesystems which don't use real block devices can call this function
992  * to allocate a virtual block device.
993  *
994  * Context: Any context.  Frequently called while holding sb_lock.
995  * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
996  * or -ENOMEM if memory allocation failed.
997  */
998 int get_anon_bdev(dev_t *p)
999 {
1000         int dev;
1001 
1002         /*
1003          * Many userspace utilities consider an FSID of 0 invalid.
1004          * Always return at least 1 from get_anon_bdev.
1005          */
1006         dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1007                         GFP_ATOMIC);
1008         if (dev == -ENOSPC)
1009                 dev = -EMFILE;
1010         if (dev < 0)
1011                 return dev;
1012 
1013         *p = MKDEV(0, dev);
1014         return 0;
1015 }
1016 EXPORT_SYMBOL(get_anon_bdev);
1017 
1018 void free_anon_bdev(dev_t dev)
1019 {
1020         ida_free(&unnamed_dev_ida, MINOR(dev));
1021 }
1022 EXPORT_SYMBOL(free_anon_bdev);
1023 
1024 int set_anon_super(struct super_block *s, void *data)
1025 {
1026         return get_anon_bdev(&s->s_dev);
1027 }
1028 EXPORT_SYMBOL(set_anon_super);
1029 
1030 void kill_anon_super(struct super_block *sb)
1031 {
1032         dev_t dev = sb->s_dev;
1033         generic_shutdown_super(sb);
1034         free_anon_bdev(dev);
1035 }
1036 EXPORT_SYMBOL(kill_anon_super);
1037 
1038 void kill_litter_super(struct super_block *sb)
1039 {
1040         if (sb->s_root)
1041                 d_genocide(sb->s_root);
1042         kill_anon_super(sb);
1043 }
1044 EXPORT_SYMBOL(kill_litter_super);
1045 
1046 static int ns_test_super(struct super_block *sb, void *data)
1047 {
1048         return sb->s_fs_info == data;
1049 }
1050 
1051 static int ns_set_super(struct super_block *sb, void *data)
1052 {
1053         sb->s_fs_info = data;
1054         return set_anon_super(sb, NULL);
1055 }
1056 
1057 struct dentry *mount_ns(struct file_system_type *fs_type,
1058         int flags, void *data, void *ns, struct user_namespace *user_ns,
1059         int (*fill_super)(struct super_block *, void *, int))
1060 {
1061         struct super_block *sb;
1062 
1063         /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1064          * over the namespace.
1065          */
1066         if (!(flags & SB_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
1067                 return ERR_PTR(-EPERM);
1068 
1069         sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1070                          user_ns, ns);
1071         if (IS_ERR(sb))
1072                 return ERR_CAST(sb);
1073 
1074         if (!sb->s_root) {
1075                 int err;
1076                 err = fill_super(sb, data, flags & SB_SILENT ? 1 : 0);
1077                 if (err) {
1078                         deactivate_locked_super(sb);
1079                         return ERR_PTR(err);
1080                 }
1081 
1082                 sb->s_flags |= SB_ACTIVE;
1083         }
1084 
1085         return dget(sb->s_root);
1086 }
1087 
1088 EXPORT_SYMBOL(mount_ns);
1089 
1090 #ifdef CONFIG_BLOCK
1091 static int set_bdev_super(struct super_block *s, void *data)
1092 {
1093         s->s_bdev = data;
1094         s->s_dev = s->s_bdev->bd_dev;
1095         s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1096 
1097         return 0;
1098 }
1099 
1100 static int test_bdev_super(struct super_block *s, void *data)
1101 {
1102         return (void *)s->s_bdev == data;
1103 }
1104 
1105 struct dentry *mount_bdev(struct file_system_type *fs_type,
1106         int flags, const char *dev_name, void *data,
1107         int (*fill_super)(struct super_block *, void *, int))
1108 {
1109         struct block_device *bdev;
1110         struct super_block *s;
1111         fmode_t mode = FMODE_READ | FMODE_EXCL;
1112         int error = 0;
1113 
1114         if (!(flags & SB_RDONLY))
1115                 mode |= FMODE_WRITE;
1116 
1117         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1118         if (IS_ERR(bdev))
1119                 return ERR_CAST(bdev);
1120 
1121         /*
1122          * once the super is inserted into the list by sget, s_umount
1123          * will protect the lockfs code from trying to start a snapshot
1124          * while we are mounting
1125          */
1126         mutex_lock(&bdev->bd_fsfreeze_mutex);
1127         if (bdev->bd_fsfreeze_count > 0) {
1128                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1129                 error = -EBUSY;
1130                 goto error_bdev;
1131         }
1132         s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
1133                  bdev);
1134         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1135         if (IS_ERR(s))
1136                 goto error_s;
1137 
1138         if (s->s_root) {
1139                 if ((flags ^ s->s_flags) & SB_RDONLY) {
1140                         deactivate_locked_super(s);
1141                         error = -EBUSY;
1142                         goto error_bdev;
1143                 }
1144 
1145                 /*
1146                  * s_umount nests inside bd_mutex during
1147                  * __invalidate_device().  blkdev_put() acquires
1148                  * bd_mutex and can't be called under s_umount.  Drop
1149                  * s_umount temporarily.  This is safe as we're
1150                  * holding an active reference.
1151                  */
1152                 up_write(&s->s_umount);
1153                 blkdev_put(bdev, mode);
1154                 down_write(&s->s_umount);
1155         } else {
1156                 s->s_mode = mode;
1157                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1158                 sb_set_blocksize(s, block_size(bdev));
1159                 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1160                 if (error) {
1161                         deactivate_locked_super(s);
1162                         goto error;
1163                 }
1164 
1165                 s->s_flags |= SB_ACTIVE;
1166                 bdev->bd_super = s;
1167         }
1168 
1169         return dget(s->s_root);
1170 
1171 error_s:
1172         error = PTR_ERR(s);
1173 error_bdev:
1174         blkdev_put(bdev, mode);
1175 error:
1176         return ERR_PTR(error);
1177 }
1178 EXPORT_SYMBOL(mount_bdev);
1179 
1180 void kill_block_super(struct super_block *sb)
1181 {
1182         struct block_device *bdev = sb->s_bdev;
1183         fmode_t mode = sb->s_mode;
1184 
1185         bdev->bd_super = NULL;
1186         generic_shutdown_super(sb);
1187         sync_blockdev(bdev);
1188         WARN_ON_ONCE(!(mode & FMODE_EXCL));
1189         blkdev_put(bdev, mode | FMODE_EXCL);
1190 }
1191 
1192 EXPORT_SYMBOL(kill_block_super);
1193 #endif
1194 
1195 struct dentry *mount_nodev(struct file_system_type *fs_type,
1196         int flags, void *data,
1197         int (*fill_super)(struct super_block *, void *, int))
1198 {
1199         int error;
1200         struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1201 
1202         if (IS_ERR(s))
1203                 return ERR_CAST(s);
1204 
1205         error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1206         if (error) {
1207                 deactivate_locked_super(s);
1208                 return ERR_PTR(error);
1209         }
1210         s->s_flags |= SB_ACTIVE;
1211         return dget(s->s_root);
1212 }
1213 EXPORT_SYMBOL(mount_nodev);
1214 
1215 static int compare_single(struct super_block *s, void *p)
1216 {
1217         return 1;
1218 }
1219 
1220 struct dentry *mount_single(struct file_system_type *fs_type,
1221         int flags, void *data,
1222         int (*fill_super)(struct super_block *, void *, int))
1223 {
1224         struct super_block *s;
1225         int error;
1226 
1227         s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1228         if (IS_ERR(s))
1229                 return ERR_CAST(s);
1230         if (!s->s_root) {
1231                 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1232                 if (error) {
1233                         deactivate_locked_super(s);
1234                         return ERR_PTR(error);
1235                 }
1236                 s->s_flags |= SB_ACTIVE;
1237         } else {
1238                 do_remount_sb(s, flags, data, 0);
1239         }
1240         return dget(s->s_root);
1241 }
1242 EXPORT_SYMBOL(mount_single);
1243 
1244 struct dentry *
1245 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1246 {
1247         struct dentry *root;
1248         struct super_block *sb;
1249         int error = -ENOMEM;
1250         void *sec_opts = NULL;
1251 
1252         if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1253                 error = security_sb_eat_lsm_opts(data, &sec_opts);
1254                 if (error)
1255                         return ERR_PTR(error);
1256         }
1257 
1258         root = type->mount(type, flags, name, data);
1259         if (IS_ERR(root)) {
1260                 error = PTR_ERR(root);
1261                 goto out_free_secdata;
1262         }
1263         sb = root->d_sb;
1264         BUG_ON(!sb);
1265         WARN_ON(!sb->s_bdi);
1266 
1267         /*
1268          * Write barrier is for super_cache_count(). We place it before setting
1269          * SB_BORN as the data dependency between the two functions is the
1270          * superblock structure contents that we just set up, not the SB_BORN
1271          * flag.
1272          */
1273         smp_wmb();
1274         sb->s_flags |= SB_BORN;
1275 
1276         error = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1277         if (error)
1278                 goto out_sb;
1279 
1280         if (!(flags & (MS_KERNMOUNT|MS_SUBMOUNT))) {
1281                 error = security_sb_kern_mount(sb);
1282                 if (error)
1283                         goto out_sb;
1284         }
1285 
1286         /*
1287          * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1288          * but s_maxbytes was an unsigned long long for many releases. Throw
1289          * this warning for a little while to try and catch filesystems that
1290          * violate this rule.
1291          */
1292         WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1293                 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1294 
1295         up_write(&sb->s_umount);
1296         security_free_mnt_opts(&sec_opts);
1297         return root;
1298 out_sb:
1299         dput(root);
1300         deactivate_locked_super(sb);
1301 out_free_secdata:
1302         security_free_mnt_opts(&sec_opts);
1303         return ERR_PTR(error);
1304 }
1305 
1306 /*
1307  * Setup private BDI for given superblock. It gets automatically cleaned up
1308  * in generic_shutdown_super().
1309  */
1310 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1311 {
1312         struct backing_dev_info *bdi;
1313         int err;
1314         va_list args;
1315 
1316         bdi = bdi_alloc(GFP_KERNEL);
1317         if (!bdi)
1318                 return -ENOMEM;
1319 
1320         bdi->name = sb->s_type->name;
1321 
1322         va_start(args, fmt);
1323         err = bdi_register_va(bdi, fmt, args);
1324         va_end(args);
1325         if (err) {
1326                 bdi_put(bdi);
1327                 return err;
1328         }
1329         WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1330         sb->s_bdi = bdi;
1331 
1332         return 0;
1333 }
1334 EXPORT_SYMBOL(super_setup_bdi_name);
1335 
1336 /*
1337  * Setup private BDI for given superblock. I gets automatically cleaned up
1338  * in generic_shutdown_super().
1339  */
1340 int super_setup_bdi(struct super_block *sb)
1341 {
1342         static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1343 
1344         return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1345                                     atomic_long_inc_return(&bdi_seq));
1346 }
1347 EXPORT_SYMBOL(super_setup_bdi);
1348 
1349 /*
1350  * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1351  * instead.
1352  */
1353 void __sb_end_write(struct super_block *sb, int level)
1354 {
1355         percpu_up_read(sb->s_writers.rw_sem + level-1);
1356 }
1357 EXPORT_SYMBOL(__sb_end_write);
1358 
1359 /*
1360  * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1361  * instead.
1362  */
1363 int __sb_start_write(struct super_block *sb, int level, bool wait)
1364 {
1365         bool force_trylock = false;
1366         int ret = 1;
1367 
1368 #ifdef CONFIG_LOCKDEP
1369         /*
1370          * We want lockdep to tell us about possible deadlocks with freezing
1371          * but it's it bit tricky to properly instrument it. Getting a freeze
1372          * protection works as getting a read lock but there are subtle
1373          * problems. XFS for example gets freeze protection on internal level
1374          * twice in some cases, which is OK only because we already hold a
1375          * freeze protection also on higher level. Due to these cases we have
1376          * to use wait == F (trylock mode) which must not fail.
1377          */
1378         if (wait) {
1379                 int i;
1380 
1381                 for (i = 0; i < level - 1; i++)
1382                         if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
1383                                 force_trylock = true;
1384                                 break;
1385                         }
1386         }
1387 #endif
1388         if (wait && !force_trylock)
1389                 percpu_down_read(sb->s_writers.rw_sem + level-1);
1390         else
1391                 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1392 
1393         WARN_ON(force_trylock && !ret);
1394         return ret;
1395 }
1396 EXPORT_SYMBOL(__sb_start_write);
1397 
1398 /**
1399  * sb_wait_write - wait until all writers to given file system finish
1400  * @sb: the super for which we wait
1401  * @level: type of writers we wait for (normal vs page fault)
1402  *
1403  * This function waits until there are no writers of given type to given file
1404  * system.
1405  */
1406 static void sb_wait_write(struct super_block *sb, int level)
1407 {
1408         percpu_down_write(sb->s_writers.rw_sem + level-1);
1409 }
1410 
1411 /*
1412  * We are going to return to userspace and forget about these locks, the
1413  * ownership goes to the caller of thaw_super() which does unlock().
1414  */
1415 static void lockdep_sb_freeze_release(struct super_block *sb)
1416 {
1417         int level;
1418 
1419         for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1420                 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1421 }
1422 
1423 /*
1424  * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1425  */
1426 static void lockdep_sb_freeze_acquire(struct super_block *sb)
1427 {
1428         int level;
1429 
1430         for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1431                 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1432 }
1433 
1434 static void sb_freeze_unlock(struct super_block *sb)
1435 {
1436         int level;
1437 
1438         for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1439                 percpu_up_write(sb->s_writers.rw_sem + level);
1440 }
1441 
1442 /**
1443  * freeze_super - lock the filesystem and force it into a consistent state
1444  * @sb: the super to lock
1445  *
1446  * Syncs the super to make sure the filesystem is consistent and calls the fs's
1447  * freeze_fs.  Subsequent calls to this without first thawing the fs will return
1448  * -EBUSY.
1449  *
1450  * During this function, sb->s_writers.frozen goes through these values:
1451  *
1452  * SB_UNFROZEN: File system is normal, all writes progress as usual.
1453  *
1454  * SB_FREEZE_WRITE: The file system is in the process of being frozen.  New
1455  * writes should be blocked, though page faults are still allowed. We wait for
1456  * all writes to complete and then proceed to the next stage.
1457  *
1458  * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1459  * but internal fs threads can still modify the filesystem (although they
1460  * should not dirty new pages or inodes), writeback can run etc. After waiting
1461  * for all running page faults we sync the filesystem which will clean all
1462  * dirty pages and inodes (no new dirty pages or inodes can be created when
1463  * sync is running).
1464  *
1465  * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1466  * modification are blocked (e.g. XFS preallocation truncation on inode
1467  * reclaim). This is usually implemented by blocking new transactions for
1468  * filesystems that have them and need this additional guard. After all
1469  * internal writers are finished we call ->freeze_fs() to finish filesystem
1470  * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1471  * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1472  *
1473  * sb->s_writers.frozen is protected by sb->s_umount.
1474  */
1475 int freeze_super(struct super_block *sb)
1476 {
1477         int ret;
1478 
1479         atomic_inc(&sb->s_active);
1480         down_write(&sb->s_umount);
1481         if (sb->s_writers.frozen != SB_UNFROZEN) {
1482                 deactivate_locked_super(sb);
1483                 return -EBUSY;
1484         }
1485 
1486         if (!(sb->s_flags & SB_BORN)) {
1487                 up_write(&sb->s_umount);
1488                 return 0;       /* sic - it's "nothing to do" */
1489         }
1490 
1491         if (sb_rdonly(sb)) {
1492                 /* Nothing to do really... */
1493                 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1494                 up_write(&sb->s_umount);
1495                 return 0;
1496         }
1497 
1498         sb->s_writers.frozen = SB_FREEZE_WRITE;
1499         /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1500         up_write(&sb->s_umount);
1501         sb_wait_write(sb, SB_FREEZE_WRITE);
1502         down_write(&sb->s_umount);
1503 
1504         /* Now we go and block page faults... */
1505         sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1506         sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1507 
1508         /* All writers are done so after syncing there won't be dirty data */
1509         sync_filesystem(sb);
1510 
1511         /* Now wait for internal filesystem counter */
1512         sb->s_writers.frozen = SB_FREEZE_FS;
1513         sb_wait_write(sb, SB_FREEZE_FS);
1514 
1515         if (sb->s_op->freeze_fs) {
1516                 ret = sb->s_op->freeze_fs(sb);
1517                 if (ret) {
1518                         printk(KERN_ERR
1519                                 "VFS:Filesystem freeze failed\n");
1520                         sb->s_writers.frozen = SB_UNFROZEN;
1521                         sb_freeze_unlock(sb);
1522                         wake_up(&sb->s_writers.wait_unfrozen);
1523                         deactivate_locked_super(sb);
1524                         return ret;
1525                 }
1526         }
1527         /*
1528          * For debugging purposes so that fs can warn if it sees write activity
1529          * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1530          */
1531         sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1532         lockdep_sb_freeze_release(sb);
1533         up_write(&sb->s_umount);
1534         return 0;
1535 }
1536 EXPORT_SYMBOL(freeze_super);
1537 
1538 /**
1539  * thaw_super -- unlock filesystem
1540  * @sb: the super to thaw
1541  *
1542  * Unlocks the filesystem and marks it writeable again after freeze_super().
1543  */
1544 static int thaw_super_locked(struct super_block *sb)
1545 {
1546         int error;
1547 
1548         if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1549                 up_write(&sb->s_umount);
1550                 return -EINVAL;
1551         }
1552 
1553         if (sb_rdonly(sb)) {
1554                 sb->s_writers.frozen = SB_UNFROZEN;
1555                 goto out;
1556         }
1557 
1558         lockdep_sb_freeze_acquire(sb);
1559 
1560         if (sb->s_op->unfreeze_fs) {
1561                 error = sb->s_op->unfreeze_fs(sb);
1562                 if (error) {
1563                         printk(KERN_ERR
1564                                 "VFS:Filesystem thaw failed\n");
1565                         lockdep_sb_freeze_release(sb);
1566                         up_write(&sb->s_umount);
1567                         return error;
1568                 }
1569         }
1570 
1571         sb->s_writers.frozen = SB_UNFROZEN;
1572         sb_freeze_unlock(sb);
1573 out:
1574         wake_up(&sb->s_writers.wait_unfrozen);
1575         deactivate_locked_super(sb);
1576         return 0;
1577 }
1578 
1579 int thaw_super(struct super_block *sb)
1580 {
1581         down_write(&sb->s_umount);
1582         return thaw_super_locked(sb);
1583 }
1584 EXPORT_SYMBOL(thaw_super);
1585 

~ [ 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