Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6.git] / fs / super.c
1 /*
2  *  linux/fs/super.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  *
6  *  super.c contains code to handle: - mount structures
7  *                                   - super-block tables
8  *                                   - filesystem drivers list
9  *                                   - mount system call
10  *                                   - umount system call
11  *                                   - ustat system call
12  *
13  * GK 2/5/95  -  Changed to support mounting the root fs via NFS
14  *
15  *  Added kerneld support: Jacques Gelinas and Bjorn Ekwall
16  *  Added change_root: Werner Almesberger & Hans Lermen, Feb '96
17  *  Added options to /proc/mounts:
18  *    Torbj√∂rn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
19  *  Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
20  *  Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
21  */
22
23 #include <linux/module.h>
24 #include <linux/slab.h>
25 #include <linux/acct.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 "internal.h"
35
36
37 LIST_HEAD(super_blocks);
38 DEFINE_SPINLOCK(sb_lock);
39
40 /**
41  *      alloc_super     -       create new superblock
42  *      @type:  filesystem type superblock should belong to
43  *
44  *      Allocates and initializes a new &struct super_block.  alloc_super()
45  *      returns a pointer new superblock or %NULL if allocation had failed.
46  */
47 static struct super_block *alloc_super(struct file_system_type *type)
48 {
49         struct super_block *s = kzalloc(sizeof(struct super_block),  GFP_USER);
50         static const struct super_operations default_op;
51
52         if (s) {
53                 if (security_sb_alloc(s)) {
54                         kfree(s);
55                         s = NULL;
56                         goto out;
57                 }
58 #ifdef CONFIG_SMP
59                 s->s_files = alloc_percpu(struct list_head);
60                 if (!s->s_files) {
61                         security_sb_free(s);
62                         kfree(s);
63                         s = NULL;
64                         goto out;
65                 } else {
66                         int i;
67
68                         for_each_possible_cpu(i)
69                                 INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i));
70                 }
71 #else
72                 INIT_LIST_HEAD(&s->s_files);
73 #endif
74                 s->s_bdi = &default_backing_dev_info;
75                 INIT_LIST_HEAD(&s->s_instances);
76                 INIT_HLIST_BL_HEAD(&s->s_anon);
77                 INIT_LIST_HEAD(&s->s_inodes);
78                 INIT_LIST_HEAD(&s->s_dentry_lru);
79                 init_rwsem(&s->s_umount);
80                 mutex_init(&s->s_lock);
81                 lockdep_set_class(&s->s_umount, &type->s_umount_key);
82                 /*
83                  * The locking rules for s_lock are up to the
84                  * filesystem. For example ext3fs has different
85                  * lock ordering than usbfs:
86                  */
87                 lockdep_set_class(&s->s_lock, &type->s_lock_key);
88                 /*
89                  * sget() can have s_umount recursion.
90                  *
91                  * When it cannot find a suitable sb, it allocates a new
92                  * one (this one), and tries again to find a suitable old
93                  * one.
94                  *
95                  * In case that succeeds, it will acquire the s_umount
96                  * lock of the old one. Since these are clearly distrinct
97                  * locks, and this object isn't exposed yet, there's no
98                  * risk of deadlocks.
99                  *
100                  * Annotate this by putting this lock in a different
101                  * subclass.
102                  */
103                 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
104                 s->s_count = 1;
105                 atomic_set(&s->s_active, 1);
106                 mutex_init(&s->s_vfs_rename_mutex);
107                 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
108                 mutex_init(&s->s_dquot.dqio_mutex);
109                 mutex_init(&s->s_dquot.dqonoff_mutex);
110                 init_rwsem(&s->s_dquot.dqptr_sem);
111                 init_waitqueue_head(&s->s_wait_unfrozen);
112                 s->s_maxbytes = MAX_NON_LFS;
113                 s->s_op = &default_op;
114                 s->s_time_gran = 1000000000;
115         }
116 out:
117         return s;
118 }
119
120 /**
121  *      destroy_super   -       frees a superblock
122  *      @s: superblock to free
123  *
124  *      Frees a superblock.
125  */
126 static inline void destroy_super(struct super_block *s)
127 {
128 #ifdef CONFIG_SMP
129         free_percpu(s->s_files);
130 #endif
131         security_sb_free(s);
132         kfree(s->s_subtype);
133         kfree(s->s_options);
134         kfree(s);
135 }
136
137 /* Superblock refcounting  */
138
139 /*
140  * Drop a superblock's refcount.  The caller must hold sb_lock.
141  */
142 void __put_super(struct super_block *sb)
143 {
144         if (!--sb->s_count) {
145                 list_del_init(&sb->s_list);
146                 destroy_super(sb);
147         }
148 }
149
150 /**
151  *      put_super       -       drop a temporary reference to superblock
152  *      @sb: superblock in question
153  *
154  *      Drops a temporary reference, frees superblock if there's no
155  *      references left.
156  */
157 void put_super(struct super_block *sb)
158 {
159         spin_lock(&sb_lock);
160         __put_super(sb);
161         spin_unlock(&sb_lock);
162 }
163
164
165 /**
166  *      deactivate_locked_super -       drop an active reference to superblock
167  *      @s: superblock to deactivate
168  *
169  *      Drops an active reference to superblock, converting it into a temprory
170  *      one if there is no other active references left.  In that case we
171  *      tell fs driver to shut it down and drop the temporary reference we
172  *      had just acquired.
173  *
174  *      Caller holds exclusive lock on superblock; that lock is released.
175  */
176 void deactivate_locked_super(struct super_block *s)
177 {
178         struct file_system_type *fs = s->s_type;
179         if (atomic_dec_and_test(&s->s_active)) {
180                 fs->kill_sb(s);
181                 /*
182                  * We need to call rcu_barrier so all the delayed rcu free
183                  * inodes are flushed before we release the fs module.
184                  */
185                 rcu_barrier();
186                 put_filesystem(fs);
187                 put_super(s);
188         } else {
189                 up_write(&s->s_umount);
190         }
191 }
192
193 EXPORT_SYMBOL(deactivate_locked_super);
194
195 /**
196  *      deactivate_super        -       drop an active reference to superblock
197  *      @s: superblock to deactivate
198  *
199  *      Variant of deactivate_locked_super(), except that superblock is *not*
200  *      locked by caller.  If we are going to drop the final active reference,
201  *      lock will be acquired prior to that.
202  */
203 void deactivate_super(struct super_block *s)
204 {
205         if (!atomic_add_unless(&s->s_active, -1, 1)) {
206                 down_write(&s->s_umount);
207                 deactivate_locked_super(s);
208         }
209 }
210
211 EXPORT_SYMBOL(deactivate_super);
212
213 /**
214  *      grab_super - acquire an active reference
215  *      @s: reference we are trying to make active
216  *
217  *      Tries to acquire an active reference.  grab_super() is used when we
218  *      had just found a superblock in super_blocks or fs_type->fs_supers
219  *      and want to turn it into a full-blown active reference.  grab_super()
220  *      is called with sb_lock held and drops it.  Returns 1 in case of
221  *      success, 0 if we had failed (superblock contents was already dead or
222  *      dying when grab_super() had been called).
223  */
224 static int grab_super(struct super_block *s) __releases(sb_lock)
225 {
226         if (atomic_inc_not_zero(&s->s_active)) {
227                 spin_unlock(&sb_lock);
228                 return 1;
229         }
230         /* it's going away */
231         s->s_count++;
232         spin_unlock(&sb_lock);
233         /* wait for it to die */
234         down_write(&s->s_umount);
235         up_write(&s->s_umount);
236         put_super(s);
237         return 0;
238 }
239
240 /*
241  * Superblock locking.  We really ought to get rid of these two.
242  */
243 void lock_super(struct super_block * sb)
244 {
245         get_fs_excl();
246         mutex_lock(&sb->s_lock);
247 }
248
249 void unlock_super(struct super_block * sb)
250 {
251         put_fs_excl();
252         mutex_unlock(&sb->s_lock);
253 }
254
255 EXPORT_SYMBOL(lock_super);
256 EXPORT_SYMBOL(unlock_super);
257
258 /**
259  *      generic_shutdown_super  -       common helper for ->kill_sb()
260  *      @sb: superblock to kill
261  *
262  *      generic_shutdown_super() does all fs-independent work on superblock
263  *      shutdown.  Typical ->kill_sb() should pick all fs-specific objects
264  *      that need destruction out of superblock, call generic_shutdown_super()
265  *      and release aforementioned objects.  Note: dentries and inodes _are_
266  *      taken care of and do not need specific handling.
267  *
268  *      Upon calling this function, the filesystem may no longer alter or
269  *      rearrange the set of dentries belonging to this super_block, nor may it
270  *      change the attachments of dentries to inodes.
271  */
272 void generic_shutdown_super(struct super_block *sb)
273 {
274         const struct super_operations *sop = sb->s_op;
275
276
277         if (sb->s_root) {
278                 shrink_dcache_for_umount(sb);
279                 sync_filesystem(sb);
280                 get_fs_excl();
281                 sb->s_flags &= ~MS_ACTIVE;
282
283                 fsnotify_unmount_inodes(&sb->s_inodes);
284
285                 evict_inodes(sb);
286
287                 if (sop->put_super)
288                         sop->put_super(sb);
289
290                 if (!list_empty(&sb->s_inodes)) {
291                         printk("VFS: Busy inodes after unmount of %s. "
292                            "Self-destruct in 5 seconds.  Have a nice day...\n",
293                            sb->s_id);
294                 }
295                 put_fs_excl();
296         }
297         spin_lock(&sb_lock);
298         /* should be initialized for __put_super_and_need_restart() */
299         list_del_init(&sb->s_instances);
300         spin_unlock(&sb_lock);
301         up_write(&sb->s_umount);
302 }
303
304 EXPORT_SYMBOL(generic_shutdown_super);
305
306 /**
307  *      sget    -       find or create a superblock
308  *      @type:  filesystem type superblock should belong to
309  *      @test:  comparison callback
310  *      @set:   setup callback
311  *      @data:  argument to each of them
312  */
313 struct super_block *sget(struct file_system_type *type,
314                         int (*test)(struct super_block *,void *),
315                         int (*set)(struct super_block *,void *),
316                         void *data)
317 {
318         struct super_block *s = NULL;
319         struct super_block *old;
320         int err;
321
322 retry:
323         spin_lock(&sb_lock);
324         if (test) {
325                 list_for_each_entry(old, &type->fs_supers, s_instances) {
326                         if (!test(old, data))
327                                 continue;
328                         if (!grab_super(old))
329                                 goto retry;
330                         if (s) {
331                                 up_write(&s->s_umount);
332                                 destroy_super(s);
333                                 s = NULL;
334                         }
335                         down_write(&old->s_umount);
336                         if (unlikely(!(old->s_flags & MS_BORN))) {
337                                 deactivate_locked_super(old);
338                                 goto retry;
339                         }
340                         return old;
341                 }
342         }
343         if (!s) {
344                 spin_unlock(&sb_lock);
345                 s = alloc_super(type);
346                 if (!s)
347                         return ERR_PTR(-ENOMEM);
348                 goto retry;
349         }
350                 
351         err = set(s, data);
352         if (err) {
353                 spin_unlock(&sb_lock);
354                 up_write(&s->s_umount);
355                 destroy_super(s);
356                 return ERR_PTR(err);
357         }
358         s->s_type = type;
359         strlcpy(s->s_id, type->name, sizeof(s->s_id));
360         list_add_tail(&s->s_list, &super_blocks);
361         list_add(&s->s_instances, &type->fs_supers);
362         spin_unlock(&sb_lock);
363         get_filesystem(type);
364         return s;
365 }
366
367 EXPORT_SYMBOL(sget);
368
369 void drop_super(struct super_block *sb)
370 {
371         up_read(&sb->s_umount);
372         put_super(sb);
373 }
374
375 EXPORT_SYMBOL(drop_super);
376
377 /**
378  * sync_supers - helper for periodic superblock writeback
379  *
380  * Call the write_super method if present on all dirty superblocks in
381  * the system.  This is for the periodic writeback used by most older
382  * filesystems.  For data integrity superblock writeback use
383  * sync_filesystems() instead.
384  *
385  * Note: check the dirty flag before waiting, so we don't
386  * hold up the sync while mounting a device. (The newly
387  * mounted device won't need syncing.)
388  */
389 void sync_supers(void)
390 {
391         struct super_block *sb, *p = NULL;
392
393         spin_lock(&sb_lock);
394         list_for_each_entry(sb, &super_blocks, s_list) {
395                 if (list_empty(&sb->s_instances))
396                         continue;
397                 if (sb->s_op->write_super && sb->s_dirt) {
398                         sb->s_count++;
399                         spin_unlock(&sb_lock);
400
401                         down_read(&sb->s_umount);
402                         if (sb->s_root && sb->s_dirt)
403                                 sb->s_op->write_super(sb);
404                         up_read(&sb->s_umount);
405
406                         spin_lock(&sb_lock);
407                         if (p)
408                                 __put_super(p);
409                         p = sb;
410                 }
411         }
412         if (p)
413                 __put_super(p);
414         spin_unlock(&sb_lock);
415 }
416
417 /**
418  *      iterate_supers - call function for all active superblocks
419  *      @f: function to call
420  *      @arg: argument to pass to it
421  *
422  *      Scans the superblock list and calls given function, passing it
423  *      locked superblock and given argument.
424  */
425 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
426 {
427         struct super_block *sb, *p = NULL;
428
429         spin_lock(&sb_lock);
430         list_for_each_entry(sb, &super_blocks, s_list) {
431                 if (list_empty(&sb->s_instances))
432                         continue;
433                 sb->s_count++;
434                 spin_unlock(&sb_lock);
435
436                 down_read(&sb->s_umount);
437                 if (sb->s_root)
438                         f(sb, arg);
439                 up_read(&sb->s_umount);
440
441                 spin_lock(&sb_lock);
442                 if (p)
443                         __put_super(p);
444                 p = sb;
445         }
446         if (p)
447                 __put_super(p);
448         spin_unlock(&sb_lock);
449 }
450
451 /**
452  *      get_super - get the superblock of a device
453  *      @bdev: device to get the superblock for
454  *      
455  *      Scans the superblock list and finds the superblock of the file system
456  *      mounted on the device given. %NULL is returned if no match is found.
457  */
458
459 struct super_block *get_super(struct block_device *bdev)
460 {
461         struct super_block *sb;
462
463         if (!bdev)
464                 return NULL;
465
466         spin_lock(&sb_lock);
467 rescan:
468         list_for_each_entry(sb, &super_blocks, s_list) {
469                 if (list_empty(&sb->s_instances))
470                         continue;
471                 if (sb->s_bdev == bdev) {
472                         sb->s_count++;
473                         spin_unlock(&sb_lock);
474                         down_read(&sb->s_umount);
475                         /* still alive? */
476                         if (sb->s_root)
477                                 return sb;
478                         up_read(&sb->s_umount);
479                         /* nope, got unmounted */
480                         spin_lock(&sb_lock);
481                         __put_super(sb);
482                         goto rescan;
483                 }
484         }
485         spin_unlock(&sb_lock);
486         return NULL;
487 }
488
489 EXPORT_SYMBOL(get_super);
490
491 /**
492  * get_active_super - get an active reference to the superblock of a device
493  * @bdev: device to get the superblock for
494  *
495  * Scans the superblock list and finds the superblock of the file system
496  * mounted on the device given.  Returns the superblock with an active
497  * reference or %NULL if none was found.
498  */
499 struct super_block *get_active_super(struct block_device *bdev)
500 {
501         struct super_block *sb;
502
503         if (!bdev)
504                 return NULL;
505
506 restart:
507         spin_lock(&sb_lock);
508         list_for_each_entry(sb, &super_blocks, s_list) {
509                 if (list_empty(&sb->s_instances))
510                         continue;
511                 if (sb->s_bdev == bdev) {
512                         if (grab_super(sb)) /* drops sb_lock */
513                                 return sb;
514                         else
515                                 goto restart;
516                 }
517         }
518         spin_unlock(&sb_lock);
519         return NULL;
520 }
521  
522 struct super_block *user_get_super(dev_t dev)
523 {
524         struct super_block *sb;
525
526         spin_lock(&sb_lock);
527 rescan:
528         list_for_each_entry(sb, &super_blocks, s_list) {
529                 if (list_empty(&sb->s_instances))
530                         continue;
531                 if (sb->s_dev ==  dev) {
532                         sb->s_count++;
533                         spin_unlock(&sb_lock);
534                         down_read(&sb->s_umount);
535                         /* still alive? */
536                         if (sb->s_root)
537                                 return sb;
538                         up_read(&sb->s_umount);
539                         /* nope, got unmounted */
540                         spin_lock(&sb_lock);
541                         __put_super(sb);
542                         goto rescan;
543                 }
544         }
545         spin_unlock(&sb_lock);
546         return NULL;
547 }
548
549 /**
550  *      do_remount_sb - asks filesystem to change mount options.
551  *      @sb:    superblock in question
552  *      @flags: numeric part of options
553  *      @data:  the rest of options
554  *      @force: whether or not to force the change
555  *
556  *      Alters the mount options of a mounted file system.
557  */
558 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
559 {
560         int retval;
561         int remount_ro;
562
563         if (sb->s_frozen != SB_UNFROZEN)
564                 return -EBUSY;
565
566 #ifdef CONFIG_BLOCK
567         if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
568                 return -EACCES;
569 #endif
570
571         if (flags & MS_RDONLY)
572                 acct_auto_close(sb);
573         shrink_dcache_sb(sb);
574         sync_filesystem(sb);
575
576         remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
577
578         /* If we are remounting RDONLY and current sb is read/write,
579            make sure there are no rw files opened */
580         if (remount_ro) {
581                 if (force)
582                         mark_files_ro(sb);
583                 else if (!fs_may_remount_ro(sb))
584                         return -EBUSY;
585         }
586
587         if (sb->s_op->remount_fs) {
588                 retval = sb->s_op->remount_fs(sb, &flags, data);
589                 if (retval)
590                         return retval;
591         }
592         sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
593
594         /*
595          * Some filesystems modify their metadata via some other path than the
596          * bdev buffer cache (eg. use a private mapping, or directories in
597          * pagecache, etc). Also file data modifications go via their own
598          * mappings. So If we try to mount readonly then copy the filesystem
599          * from bdev, we could get stale data, so invalidate it to give a best
600          * effort at coherency.
601          */
602         if (remount_ro && sb->s_bdev)
603                 invalidate_bdev(sb->s_bdev);
604         return 0;
605 }
606
607 static void do_emergency_remount(struct work_struct *work)
608 {
609         struct super_block *sb, *p = NULL;
610
611         spin_lock(&sb_lock);
612         list_for_each_entry(sb, &super_blocks, s_list) {
613                 if (list_empty(&sb->s_instances))
614                         continue;
615                 sb->s_count++;
616                 spin_unlock(&sb_lock);
617                 down_write(&sb->s_umount);
618                 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
619                         /*
620                          * What lock protects sb->s_flags??
621                          */
622                         do_remount_sb(sb, MS_RDONLY, NULL, 1);
623                 }
624                 up_write(&sb->s_umount);
625                 spin_lock(&sb_lock);
626                 if (p)
627                         __put_super(p);
628                 p = sb;
629         }
630         if (p)
631                 __put_super(p);
632         spin_unlock(&sb_lock);
633         kfree(work);
634         printk("Emergency Remount complete\n");
635 }
636
637 void emergency_remount(void)
638 {
639         struct work_struct *work;
640
641         work = kmalloc(sizeof(*work), GFP_ATOMIC);
642         if (work) {
643                 INIT_WORK(work, do_emergency_remount);
644                 schedule_work(work);
645         }
646 }
647
648 /*
649  * Unnamed block devices are dummy devices used by virtual
650  * filesystems which don't use real block-devices.  -- jrs
651  */
652
653 static DEFINE_IDA(unnamed_dev_ida);
654 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
655 static int unnamed_dev_start = 0; /* don't bother trying below it */
656
657 int set_anon_super(struct super_block *s, void *data)
658 {
659         int dev;
660         int error;
661
662  retry:
663         if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
664                 return -ENOMEM;
665         spin_lock(&unnamed_dev_lock);
666         error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
667         if (!error)
668                 unnamed_dev_start = dev + 1;
669         spin_unlock(&unnamed_dev_lock);
670         if (error == -EAGAIN)
671                 /* We raced and lost with another CPU. */
672                 goto retry;
673         else if (error)
674                 return -EAGAIN;
675
676         if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
677                 spin_lock(&unnamed_dev_lock);
678                 ida_remove(&unnamed_dev_ida, dev);
679                 if (unnamed_dev_start > dev)
680                         unnamed_dev_start = dev;
681                 spin_unlock(&unnamed_dev_lock);
682                 return -EMFILE;
683         }
684         s->s_dev = MKDEV(0, dev & MINORMASK);
685         s->s_bdi = &noop_backing_dev_info;
686         return 0;
687 }
688
689 EXPORT_SYMBOL(set_anon_super);
690
691 void kill_anon_super(struct super_block *sb)
692 {
693         int slot = MINOR(sb->s_dev);
694
695         generic_shutdown_super(sb);
696         spin_lock(&unnamed_dev_lock);
697         ida_remove(&unnamed_dev_ida, slot);
698         if (slot < unnamed_dev_start)
699                 unnamed_dev_start = slot;
700         spin_unlock(&unnamed_dev_lock);
701 }
702
703 EXPORT_SYMBOL(kill_anon_super);
704
705 void kill_litter_super(struct super_block *sb)
706 {
707         if (sb->s_root)
708                 d_genocide(sb->s_root);
709         kill_anon_super(sb);
710 }
711
712 EXPORT_SYMBOL(kill_litter_super);
713
714 static int ns_test_super(struct super_block *sb, void *data)
715 {
716         return sb->s_fs_info == data;
717 }
718
719 static int ns_set_super(struct super_block *sb, void *data)
720 {
721         sb->s_fs_info = data;
722         return set_anon_super(sb, NULL);
723 }
724
725 struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
726         void *data, int (*fill_super)(struct super_block *, void *, int))
727 {
728         struct super_block *sb;
729
730         sb = sget(fs_type, ns_test_super, ns_set_super, data);
731         if (IS_ERR(sb))
732                 return ERR_CAST(sb);
733
734         if (!sb->s_root) {
735                 int err;
736                 sb->s_flags = flags;
737                 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
738                 if (err) {
739                         deactivate_locked_super(sb);
740                         return ERR_PTR(err);
741                 }
742
743                 sb->s_flags |= MS_ACTIVE;
744         }
745
746         return dget(sb->s_root);
747 }
748
749 EXPORT_SYMBOL(mount_ns);
750
751 #ifdef CONFIG_BLOCK
752 static int set_bdev_super(struct super_block *s, void *data)
753 {
754         s->s_bdev = data;
755         s->s_dev = s->s_bdev->bd_dev;
756
757         /*
758          * We set the bdi here to the queue backing, file systems can
759          * overwrite this in ->fill_super()
760          */
761         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
762         return 0;
763 }
764
765 static int test_bdev_super(struct super_block *s, void *data)
766 {
767         return (void *)s->s_bdev == data;
768 }
769
770 struct dentry *mount_bdev(struct file_system_type *fs_type,
771         int flags, const char *dev_name, void *data,
772         int (*fill_super)(struct super_block *, void *, int))
773 {
774         struct block_device *bdev;
775         struct super_block *s;
776         fmode_t mode = FMODE_READ | FMODE_EXCL;
777         int error = 0;
778
779         if (!(flags & MS_RDONLY))
780                 mode |= FMODE_WRITE;
781
782         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
783         if (IS_ERR(bdev))
784                 return ERR_CAST(bdev);
785
786         /*
787          * once the super is inserted into the list by sget, s_umount
788          * will protect the lockfs code from trying to start a snapshot
789          * while we are mounting
790          */
791         mutex_lock(&bdev->bd_fsfreeze_mutex);
792         if (bdev->bd_fsfreeze_count > 0) {
793                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
794                 error = -EBUSY;
795                 goto error_bdev;
796         }
797         s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
798         mutex_unlock(&bdev->bd_fsfreeze_mutex);
799         if (IS_ERR(s))
800                 goto error_s;
801
802         if (s->s_root) {
803                 if ((flags ^ s->s_flags) & MS_RDONLY) {
804                         deactivate_locked_super(s);
805                         error = -EBUSY;
806                         goto error_bdev;
807                 }
808
809                 /*
810                  * s_umount nests inside bd_mutex during
811                  * __invalidate_device().  blkdev_put() acquires
812                  * bd_mutex and can't be called under s_umount.  Drop
813                  * s_umount temporarily.  This is safe as we're
814                  * holding an active reference.
815                  */
816                 up_write(&s->s_umount);
817                 blkdev_put(bdev, mode);
818                 down_write(&s->s_umount);
819         } else {
820                 char b[BDEVNAME_SIZE];
821
822                 s->s_flags = flags;
823                 s->s_mode = mode;
824                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
825                 sb_set_blocksize(s, block_size(bdev));
826                 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
827                 if (error) {
828                         deactivate_locked_super(s);
829                         goto error;
830                 }
831
832                 s->s_flags |= MS_ACTIVE;
833                 bdev->bd_super = s;
834         }
835
836         return dget(s->s_root);
837
838 error_s:
839         error = PTR_ERR(s);
840 error_bdev:
841         blkdev_put(bdev, mode);
842 error:
843         return ERR_PTR(error);
844 }
845 EXPORT_SYMBOL(mount_bdev);
846
847 void kill_block_super(struct super_block *sb)
848 {
849         struct block_device *bdev = sb->s_bdev;
850         fmode_t mode = sb->s_mode;
851
852         bdev->bd_super = NULL;
853         generic_shutdown_super(sb);
854         sync_blockdev(bdev);
855         WARN_ON_ONCE(!(mode & FMODE_EXCL));
856         blkdev_put(bdev, mode | FMODE_EXCL);
857 }
858
859 EXPORT_SYMBOL(kill_block_super);
860 #endif
861
862 struct dentry *mount_nodev(struct file_system_type *fs_type,
863         int flags, void *data,
864         int (*fill_super)(struct super_block *, void *, int))
865 {
866         int error;
867         struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
868
869         if (IS_ERR(s))
870                 return ERR_CAST(s);
871
872         s->s_flags = flags;
873
874         error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
875         if (error) {
876                 deactivate_locked_super(s);
877                 return ERR_PTR(error);
878         }
879         s->s_flags |= MS_ACTIVE;
880         return dget(s->s_root);
881 }
882 EXPORT_SYMBOL(mount_nodev);
883
884 static int compare_single(struct super_block *s, void *p)
885 {
886         return 1;
887 }
888
889 struct dentry *mount_single(struct file_system_type *fs_type,
890         int flags, void *data,
891         int (*fill_super)(struct super_block *, void *, int))
892 {
893         struct super_block *s;
894         int error;
895
896         s = sget(fs_type, compare_single, set_anon_super, NULL);
897         if (IS_ERR(s))
898                 return ERR_CAST(s);
899         if (!s->s_root) {
900                 s->s_flags = flags;
901                 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
902                 if (error) {
903                         deactivate_locked_super(s);
904                         return ERR_PTR(error);
905                 }
906                 s->s_flags |= MS_ACTIVE;
907         } else {
908                 do_remount_sb(s, flags, data, 0);
909         }
910         return dget(s->s_root);
911 }
912 EXPORT_SYMBOL(mount_single);
913
914 struct dentry *
915 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
916 {
917         struct dentry *root;
918         struct super_block *sb;
919         char *secdata = NULL;
920         int error = -ENOMEM;
921
922         if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
923                 secdata = alloc_secdata();
924                 if (!secdata)
925                         goto out;
926
927                 error = security_sb_copy_data(data, secdata);
928                 if (error)
929                         goto out_free_secdata;
930         }
931
932         root = type->mount(type, flags, name, data);
933         if (IS_ERR(root)) {
934                 error = PTR_ERR(root);
935                 goto out_free_secdata;
936         }
937         sb = root->d_sb;
938         BUG_ON(!sb);
939         WARN_ON(!sb->s_bdi);
940         WARN_ON(sb->s_bdi == &default_backing_dev_info);
941         sb->s_flags |= MS_BORN;
942
943         error = security_sb_kern_mount(sb, flags, secdata);
944         if (error)
945                 goto out_sb;
946
947         /*
948          * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
949          * but s_maxbytes was an unsigned long long for many releases. Throw
950          * this warning for a little while to try and catch filesystems that
951          * violate this rule.
952          */
953         WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
954                 "negative value (%lld)\n", type->name, sb->s_maxbytes);
955
956         up_write(&sb->s_umount);
957         free_secdata(secdata);
958         return root;
959 out_sb:
960         dput(root);
961         deactivate_locked_super(sb);
962 out_free_secdata:
963         free_secdata(secdata);
964 out:
965         return ERR_PTR(error);
966 }
967
968 /**
969  * freeze_super - lock the filesystem and force it into a consistent state
970  * @sb: the super to lock
971  *
972  * Syncs the super to make sure the filesystem is consistent and calls the fs's
973  * freeze_fs.  Subsequent calls to this without first thawing the fs will return
974  * -EBUSY.
975  */
976 int freeze_super(struct super_block *sb)
977 {
978         int ret;
979
980         atomic_inc(&sb->s_active);
981         down_write(&sb->s_umount);
982         if (sb->s_frozen) {
983                 deactivate_locked_super(sb);
984                 return -EBUSY;
985         }
986
987         if (sb->s_flags & MS_RDONLY) {
988                 sb->s_frozen = SB_FREEZE_TRANS;
989                 smp_wmb();
990                 up_write(&sb->s_umount);
991                 return 0;
992         }
993
994         sb->s_frozen = SB_FREEZE_WRITE;
995         smp_wmb();
996
997         sync_filesystem(sb);
998
999         sb->s_frozen = SB_FREEZE_TRANS;
1000         smp_wmb();
1001
1002         sync_blockdev(sb->s_bdev);
1003         if (sb->s_op->freeze_fs) {
1004                 ret = sb->s_op->freeze_fs(sb);
1005                 if (ret) {
1006                         printk(KERN_ERR
1007                                 "VFS:Filesystem freeze failed\n");
1008                         sb->s_frozen = SB_UNFROZEN;
1009                         deactivate_locked_super(sb);
1010                         return ret;
1011                 }
1012         }
1013         up_write(&sb->s_umount);
1014         return 0;
1015 }
1016 EXPORT_SYMBOL(freeze_super);
1017
1018 /**
1019  * thaw_super -- unlock filesystem
1020  * @sb: the super to thaw
1021  *
1022  * Unlocks the filesystem and marks it writeable again after freeze_super().
1023  */
1024 int thaw_super(struct super_block *sb)
1025 {
1026         int error;
1027
1028         down_write(&sb->s_umount);
1029         if (sb->s_frozen == SB_UNFROZEN) {
1030                 up_write(&sb->s_umount);
1031                 return -EINVAL;
1032         }
1033
1034         if (sb->s_flags & MS_RDONLY)
1035                 goto out;
1036
1037         if (sb->s_op->unfreeze_fs) {
1038                 error = sb->s_op->unfreeze_fs(sb);
1039                 if (error) {
1040                         printk(KERN_ERR
1041                                 "VFS:Filesystem thaw failed\n");
1042                         sb->s_frozen = SB_FREEZE_TRANS;
1043                         up_write(&sb->s_umount);
1044                         return error;
1045                 }
1046         }
1047
1048 out:
1049         sb->s_frozen = SB_UNFROZEN;
1050         smp_wmb();
1051         wake_up(&sb->s_wait_unfrozen);
1052         deactivate_locked_super(sb);
1053
1054         return 0;
1055 }
1056 EXPORT_SYMBOL(thaw_super);