fixlet: Remove fs_excl from struct task.
[linux-3.10.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 <linux/cleancache.h>
35 #include "internal.h"
36
37
38 LIST_HEAD(super_blocks);
39 DEFINE_SPINLOCK(sb_lock);
40
41 /**
42  *      alloc_super     -       create new superblock
43  *      @type:  filesystem type superblock should belong to
44  *
45  *      Allocates and initializes a new &struct super_block.  alloc_super()
46  *      returns a pointer new superblock or %NULL if allocation had failed.
47  */
48 static struct super_block *alloc_super(struct file_system_type *type)
49 {
50         struct super_block *s = kzalloc(sizeof(struct super_block),  GFP_USER);
51         static const struct super_operations default_op;
52
53         if (s) {
54                 if (security_sb_alloc(s)) {
55                         kfree(s);
56                         s = NULL;
57                         goto out;
58                 }
59 #ifdef CONFIG_SMP
60                 s->s_files = alloc_percpu(struct list_head);
61                 if (!s->s_files) {
62                         security_sb_free(s);
63                         kfree(s);
64                         s = NULL;
65                         goto out;
66                 } else {
67                         int i;
68
69                         for_each_possible_cpu(i)
70                                 INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i));
71                 }
72 #else
73                 INIT_LIST_HEAD(&s->s_files);
74 #endif
75                 s->s_bdi = &default_backing_dev_info;
76                 INIT_LIST_HEAD(&s->s_instances);
77                 INIT_HLIST_BL_HEAD(&s->s_anon);
78                 INIT_LIST_HEAD(&s->s_inodes);
79                 INIT_LIST_HEAD(&s->s_dentry_lru);
80                 init_rwsem(&s->s_umount);
81                 mutex_init(&s->s_lock);
82                 lockdep_set_class(&s->s_umount, &type->s_umount_key);
83                 /*
84                  * The locking rules for s_lock are up to the
85                  * filesystem. For example ext3fs has different
86                  * lock ordering than usbfs:
87                  */
88                 lockdep_set_class(&s->s_lock, &type->s_lock_key);
89                 /*
90                  * sget() can have s_umount recursion.
91                  *
92                  * When it cannot find a suitable sb, it allocates a new
93                  * one (this one), and tries again to find a suitable old
94                  * one.
95                  *
96                  * In case that succeeds, it will acquire the s_umount
97                  * lock of the old one. Since these are clearly distrinct
98                  * locks, and this object isn't exposed yet, there's no
99                  * risk of deadlocks.
100                  *
101                  * Annotate this by putting this lock in a different
102                  * subclass.
103                  */
104                 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
105                 s->s_count = 1;
106                 atomic_set(&s->s_active, 1);
107                 mutex_init(&s->s_vfs_rename_mutex);
108                 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
109                 mutex_init(&s->s_dquot.dqio_mutex);
110                 mutex_init(&s->s_dquot.dqonoff_mutex);
111                 init_rwsem(&s->s_dquot.dqptr_sem);
112                 init_waitqueue_head(&s->s_wait_unfrozen);
113                 s->s_maxbytes = MAX_NON_LFS;
114                 s->s_op = &default_op;
115                 s->s_time_gran = 1000000000;
116                 s->cleancache_poolid = -1;
117         }
118 out:
119         return s;
120 }
121
122 /**
123  *      destroy_super   -       frees a superblock
124  *      @s: superblock to free
125  *
126  *      Frees a superblock.
127  */
128 static inline void destroy_super(struct super_block *s)
129 {
130 #ifdef CONFIG_SMP
131         free_percpu(s->s_files);
132 #endif
133         security_sb_free(s);
134         kfree(s->s_subtype);
135         kfree(s->s_options);
136         kfree(s);
137 }
138
139 /* Superblock refcounting  */
140
141 /*
142  * Drop a superblock's refcount.  The caller must hold sb_lock.
143  */
144 void __put_super(struct super_block *sb)
145 {
146         if (!--sb->s_count) {
147                 list_del_init(&sb->s_list);
148                 destroy_super(sb);
149         }
150 }
151
152 /**
153  *      put_super       -       drop a temporary reference to superblock
154  *      @sb: superblock in question
155  *
156  *      Drops a temporary reference, frees superblock if there's no
157  *      references left.
158  */
159 void put_super(struct super_block *sb)
160 {
161         spin_lock(&sb_lock);
162         __put_super(sb);
163         spin_unlock(&sb_lock);
164 }
165
166
167 /**
168  *      deactivate_locked_super -       drop an active reference to superblock
169  *      @s: superblock to deactivate
170  *
171  *      Drops an active reference to superblock, converting it into a temprory
172  *      one if there is no other active references left.  In that case we
173  *      tell fs driver to shut it down and drop the temporary reference we
174  *      had just acquired.
175  *
176  *      Caller holds exclusive lock on superblock; that lock is released.
177  */
178 void deactivate_locked_super(struct super_block *s)
179 {
180         struct file_system_type *fs = s->s_type;
181         if (atomic_dec_and_test(&s->s_active)) {
182                 cleancache_flush_fs(s);
183                 fs->kill_sb(s);
184                 /*
185                  * We need to call rcu_barrier so all the delayed rcu free
186                  * inodes are flushed before we release the fs module.
187                  */
188                 rcu_barrier();
189                 put_filesystem(fs);
190                 put_super(s);
191         } else {
192                 up_write(&s->s_umount);
193         }
194 }
195
196 EXPORT_SYMBOL(deactivate_locked_super);
197
198 /**
199  *      deactivate_super        -       drop an active reference to superblock
200  *      @s: superblock to deactivate
201  *
202  *      Variant of deactivate_locked_super(), except that superblock is *not*
203  *      locked by caller.  If we are going to drop the final active reference,
204  *      lock will be acquired prior to that.
205  */
206 void deactivate_super(struct super_block *s)
207 {
208         if (!atomic_add_unless(&s->s_active, -1, 1)) {
209                 down_write(&s->s_umount);
210                 deactivate_locked_super(s);
211         }
212 }
213
214 EXPORT_SYMBOL(deactivate_super);
215
216 /**
217  *      grab_super - acquire an active reference
218  *      @s: reference we are trying to make active
219  *
220  *      Tries to acquire an active reference.  grab_super() is used when we
221  *      had just found a superblock in super_blocks or fs_type->fs_supers
222  *      and want to turn it into a full-blown active reference.  grab_super()
223  *      is called with sb_lock held and drops it.  Returns 1 in case of
224  *      success, 0 if we had failed (superblock contents was already dead or
225  *      dying when grab_super() had been called).
226  */
227 static int grab_super(struct super_block *s) __releases(sb_lock)
228 {
229         if (atomic_inc_not_zero(&s->s_active)) {
230                 spin_unlock(&sb_lock);
231                 return 1;
232         }
233         /* it's going away */
234         s->s_count++;
235         spin_unlock(&sb_lock);
236         /* wait for it to die */
237         down_write(&s->s_umount);
238         up_write(&s->s_umount);
239         put_super(s);
240         return 0;
241 }
242
243 /*
244  * Superblock locking.  We really ought to get rid of these two.
245  */
246 void lock_super(struct super_block * sb)
247 {
248         mutex_lock(&sb->s_lock);
249 }
250
251 void unlock_super(struct super_block * sb)
252 {
253         mutex_unlock(&sb->s_lock);
254 }
255
256 EXPORT_SYMBOL(lock_super);
257 EXPORT_SYMBOL(unlock_super);
258
259 /**
260  *      generic_shutdown_super  -       common helper for ->kill_sb()
261  *      @sb: superblock to kill
262  *
263  *      generic_shutdown_super() does all fs-independent work on superblock
264  *      shutdown.  Typical ->kill_sb() should pick all fs-specific objects
265  *      that need destruction out of superblock, call generic_shutdown_super()
266  *      and release aforementioned objects.  Note: dentries and inodes _are_
267  *      taken care of and do not need specific handling.
268  *
269  *      Upon calling this function, the filesystem may no longer alter or
270  *      rearrange the set of dentries belonging to this super_block, nor may it
271  *      change the attachments of dentries to inodes.
272  */
273 void generic_shutdown_super(struct super_block *sb)
274 {
275         const struct super_operations *sop = sb->s_op;
276
277
278         if (sb->s_root) {
279                 shrink_dcache_for_umount(sb);
280                 sync_filesystem(sb);
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         }
296         spin_lock(&sb_lock);
297         /* should be initialized for __put_super_and_need_restart() */
298         list_del_init(&sb->s_instances);
299         spin_unlock(&sb_lock);
300         up_write(&sb->s_umount);
301 }
302
303 EXPORT_SYMBOL(generic_shutdown_super);
304
305 /**
306  *      sget    -       find or create a superblock
307  *      @type:  filesystem type superblock should belong to
308  *      @test:  comparison callback
309  *      @set:   setup callback
310  *      @data:  argument to each of them
311  */
312 struct super_block *sget(struct file_system_type *type,
313                         int (*test)(struct super_block *,void *),
314                         int (*set)(struct super_block *,void *),
315                         void *data)
316 {
317         struct super_block *s = NULL;
318         struct super_block *old;
319         int err;
320
321 retry:
322         spin_lock(&sb_lock);
323         if (test) {
324                 list_for_each_entry(old, &type->fs_supers, s_instances) {
325                         if (!test(old, data))
326                                 continue;
327                         if (!grab_super(old))
328                                 goto retry;
329                         if (s) {
330                                 up_write(&s->s_umount);
331                                 destroy_super(s);
332                                 s = NULL;
333                         }
334                         down_write(&old->s_umount);
335                         if (unlikely(!(old->s_flags & MS_BORN))) {
336                                 deactivate_locked_super(old);
337                                 goto retry;
338                         }
339                         return old;
340                 }
341         }
342         if (!s) {
343                 spin_unlock(&sb_lock);
344                 s = alloc_super(type);
345                 if (!s)
346                         return ERR_PTR(-ENOMEM);
347                 goto retry;
348         }
349                 
350         err = set(s, data);
351         if (err) {
352                 spin_unlock(&sb_lock);
353                 up_write(&s->s_umount);
354                 destroy_super(s);
355                 return ERR_PTR(err);
356         }
357         s->s_type = type;
358         strlcpy(s->s_id, type->name, sizeof(s->s_id));
359         list_add_tail(&s->s_list, &super_blocks);
360         list_add(&s->s_instances, &type->fs_supers);
361         spin_unlock(&sb_lock);
362         get_filesystem(type);
363         return s;
364 }
365
366 EXPORT_SYMBOL(sget);
367
368 void drop_super(struct super_block *sb)
369 {
370         up_read(&sb->s_umount);
371         put_super(sb);
372 }
373
374 EXPORT_SYMBOL(drop_super);
375
376 /**
377  * sync_supers - helper for periodic superblock writeback
378  *
379  * Call the write_super method if present on all dirty superblocks in
380  * the system.  This is for the periodic writeback used by most older
381  * filesystems.  For data integrity superblock writeback use
382  * sync_filesystems() instead.
383  *
384  * Note: check the dirty flag before waiting, so we don't
385  * hold up the sync while mounting a device. (The newly
386  * mounted device won't need syncing.)
387  */
388 void sync_supers(void)
389 {
390         struct super_block *sb, *p = NULL;
391
392         spin_lock(&sb_lock);
393         list_for_each_entry(sb, &super_blocks, s_list) {
394                 if (list_empty(&sb->s_instances))
395                         continue;
396                 if (sb->s_op->write_super && sb->s_dirt) {
397                         sb->s_count++;
398                         spin_unlock(&sb_lock);
399
400                         down_read(&sb->s_umount);
401                         if (sb->s_root && sb->s_dirt)
402                                 sb->s_op->write_super(sb);
403                         up_read(&sb->s_umount);
404
405                         spin_lock(&sb_lock);
406                         if (p)
407                                 __put_super(p);
408                         p = sb;
409                 }
410         }
411         if (p)
412                 __put_super(p);
413         spin_unlock(&sb_lock);
414 }
415
416 /**
417  *      iterate_supers - call function for all active superblocks
418  *      @f: function to call
419  *      @arg: argument to pass to it
420  *
421  *      Scans the superblock list and calls given function, passing it
422  *      locked superblock and given argument.
423  */
424 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
425 {
426         struct super_block *sb, *p = NULL;
427
428         spin_lock(&sb_lock);
429         list_for_each_entry(sb, &super_blocks, s_list) {
430                 if (list_empty(&sb->s_instances))
431                         continue;
432                 sb->s_count++;
433                 spin_unlock(&sb_lock);
434
435                 down_read(&sb->s_umount);
436                 if (sb->s_root)
437                         f(sb, arg);
438                 up_read(&sb->s_umount);
439
440                 spin_lock(&sb_lock);
441                 if (p)
442                         __put_super(p);
443                 p = sb;
444         }
445         if (p)
446                 __put_super(p);
447         spin_unlock(&sb_lock);
448 }
449
450 /**
451  *      get_super - get the superblock of a device
452  *      @bdev: device to get the superblock for
453  *      
454  *      Scans the superblock list and finds the superblock of the file system
455  *      mounted on the device given. %NULL is returned if no match is found.
456  */
457
458 struct super_block *get_super(struct block_device *bdev)
459 {
460         struct super_block *sb;
461
462         if (!bdev)
463                 return NULL;
464
465         spin_lock(&sb_lock);
466 rescan:
467         list_for_each_entry(sb, &super_blocks, s_list) {
468                 if (list_empty(&sb->s_instances))
469                         continue;
470                 if (sb->s_bdev == bdev) {
471                         sb->s_count++;
472                         spin_unlock(&sb_lock);
473                         down_read(&sb->s_umount);
474                         /* still alive? */
475                         if (sb->s_root)
476                                 return sb;
477                         up_read(&sb->s_umount);
478                         /* nope, got unmounted */
479                         spin_lock(&sb_lock);
480                         __put_super(sb);
481                         goto rescan;
482                 }
483         }
484         spin_unlock(&sb_lock);
485         return NULL;
486 }
487
488 EXPORT_SYMBOL(get_super);
489
490 /**
491  * get_active_super - get an active reference to the superblock of a device
492  * @bdev: device to get the superblock for
493  *
494  * Scans the superblock list and finds the superblock of the file system
495  * mounted on the device given.  Returns the superblock with an active
496  * reference or %NULL if none was found.
497  */
498 struct super_block *get_active_super(struct block_device *bdev)
499 {
500         struct super_block *sb;
501
502         if (!bdev)
503                 return NULL;
504
505 restart:
506         spin_lock(&sb_lock);
507         list_for_each_entry(sb, &super_blocks, s_list) {
508                 if (list_empty(&sb->s_instances))
509                         continue;
510                 if (sb->s_bdev == bdev) {
511                         if (grab_super(sb)) /* drops sb_lock */
512                                 return sb;
513                         else
514                                 goto restart;
515                 }
516         }
517         spin_unlock(&sb_lock);
518         return NULL;
519 }
520  
521 struct super_block *user_get_super(dev_t dev)
522 {
523         struct super_block *sb;
524
525         spin_lock(&sb_lock);
526 rescan:
527         list_for_each_entry(sb, &super_blocks, s_list) {
528                 if (list_empty(&sb->s_instances))
529                         continue;
530                 if (sb->s_dev ==  dev) {
531                         sb->s_count++;
532                         spin_unlock(&sb_lock);
533                         down_read(&sb->s_umount);
534                         /* still alive? */
535                         if (sb->s_root)
536                                 return sb;
537                         up_read(&sb->s_umount);
538                         /* nope, got unmounted */
539                         spin_lock(&sb_lock);
540                         __put_super(sb);
541                         goto rescan;
542                 }
543         }
544         spin_unlock(&sb_lock);
545         return NULL;
546 }
547
548 /**
549  *      do_remount_sb - asks filesystem to change mount options.
550  *      @sb:    superblock in question
551  *      @flags: numeric part of options
552  *      @data:  the rest of options
553  *      @force: whether or not to force the change
554  *
555  *      Alters the mount options of a mounted file system.
556  */
557 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
558 {
559         int retval;
560         int remount_ro;
561
562         if (sb->s_frozen != SB_UNFROZEN)
563                 return -EBUSY;
564
565 #ifdef CONFIG_BLOCK
566         if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
567                 return -EACCES;
568 #endif
569
570         if (flags & MS_RDONLY)
571                 acct_auto_close(sb);
572         shrink_dcache_sb(sb);
573         sync_filesystem(sb);
574
575         remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
576
577         /* If we are remounting RDONLY and current sb is read/write,
578            make sure there are no rw files opened */
579         if (remount_ro) {
580                 if (force)
581                         mark_files_ro(sb);
582                 else if (!fs_may_remount_ro(sb))
583                         return -EBUSY;
584         }
585
586         if (sb->s_op->remount_fs) {
587                 retval = sb->s_op->remount_fs(sb, &flags, data);
588                 if (retval)
589                         return retval;
590         }
591         sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
592
593         /*
594          * Some filesystems modify their metadata via some other path than the
595          * bdev buffer cache (eg. use a private mapping, or directories in
596          * pagecache, etc). Also file data modifications go via their own
597          * mappings. So If we try to mount readonly then copy the filesystem
598          * from bdev, we could get stale data, so invalidate it to give a best
599          * effort at coherency.
600          */
601         if (remount_ro && sb->s_bdev)
602                 invalidate_bdev(sb->s_bdev);
603         return 0;
604 }
605
606 static void do_emergency_remount(struct work_struct *work)
607 {
608         struct super_block *sb, *p = NULL;
609
610         spin_lock(&sb_lock);
611         list_for_each_entry(sb, &super_blocks, s_list) {
612                 if (list_empty(&sb->s_instances))
613                         continue;
614                 sb->s_count++;
615                 spin_unlock(&sb_lock);
616                 down_write(&sb->s_umount);
617                 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
618                         /*
619                          * What lock protects sb->s_flags??
620                          */
621                         do_remount_sb(sb, MS_RDONLY, NULL, 1);
622                 }
623                 up_write(&sb->s_umount);
624                 spin_lock(&sb_lock);
625                 if (p)
626                         __put_super(p);
627                 p = sb;
628         }
629         if (p)
630                 __put_super(p);
631         spin_unlock(&sb_lock);
632         kfree(work);
633         printk("Emergency Remount complete\n");
634 }
635
636 void emergency_remount(void)
637 {
638         struct work_struct *work;
639
640         work = kmalloc(sizeof(*work), GFP_ATOMIC);
641         if (work) {
642                 INIT_WORK(work, do_emergency_remount);
643                 schedule_work(work);
644         }
645 }
646
647 /*
648  * Unnamed block devices are dummy devices used by virtual
649  * filesystems which don't use real block-devices.  -- jrs
650  */
651
652 static DEFINE_IDA(unnamed_dev_ida);
653 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
654 static int unnamed_dev_start = 0; /* don't bother trying below it */
655
656 int set_anon_super(struct super_block *s, void *data)
657 {
658         int dev;
659         int error;
660
661  retry:
662         if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
663                 return -ENOMEM;
664         spin_lock(&unnamed_dev_lock);
665         error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
666         if (!error)
667                 unnamed_dev_start = dev + 1;
668         spin_unlock(&unnamed_dev_lock);
669         if (error == -EAGAIN)
670                 /* We raced and lost with another CPU. */
671                 goto retry;
672         else if (error)
673                 return -EAGAIN;
674
675         if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
676                 spin_lock(&unnamed_dev_lock);
677                 ida_remove(&unnamed_dev_ida, dev);
678                 if (unnamed_dev_start > dev)
679                         unnamed_dev_start = dev;
680                 spin_unlock(&unnamed_dev_lock);
681                 return -EMFILE;
682         }
683         s->s_dev = MKDEV(0, dev & MINORMASK);
684         s->s_bdi = &noop_backing_dev_info;
685         return 0;
686 }
687
688 EXPORT_SYMBOL(set_anon_super);
689
690 void kill_anon_super(struct super_block *sb)
691 {
692         int slot = MINOR(sb->s_dev);
693
694         generic_shutdown_super(sb);
695         spin_lock(&unnamed_dev_lock);
696         ida_remove(&unnamed_dev_ida, slot);
697         if (slot < unnamed_dev_start)
698                 unnamed_dev_start = slot;
699         spin_unlock(&unnamed_dev_lock);
700 }
701
702 EXPORT_SYMBOL(kill_anon_super);
703
704 void kill_litter_super(struct super_block *sb)
705 {
706         if (sb->s_root)
707                 d_genocide(sb->s_root);
708         kill_anon_super(sb);
709 }
710
711 EXPORT_SYMBOL(kill_litter_super);
712
713 static int ns_test_super(struct super_block *sb, void *data)
714 {
715         return sb->s_fs_info == data;
716 }
717
718 static int ns_set_super(struct super_block *sb, void *data)
719 {
720         sb->s_fs_info = data;
721         return set_anon_super(sb, NULL);
722 }
723
724 struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
725         void *data, int (*fill_super)(struct super_block *, void *, int))
726 {
727         struct super_block *sb;
728
729         sb = sget(fs_type, ns_test_super, ns_set_super, data);
730         if (IS_ERR(sb))
731                 return ERR_CAST(sb);
732
733         if (!sb->s_root) {
734                 int err;
735                 sb->s_flags = flags;
736                 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
737                 if (err) {
738                         deactivate_locked_super(sb);
739                         return ERR_PTR(err);
740                 }
741
742                 sb->s_flags |= MS_ACTIVE;
743         }
744
745         return dget(sb->s_root);
746 }
747
748 EXPORT_SYMBOL(mount_ns);
749
750 #ifdef CONFIG_BLOCK
751 static int set_bdev_super(struct super_block *s, void *data)
752 {
753         s->s_bdev = data;
754         s->s_dev = s->s_bdev->bd_dev;
755
756         /*
757          * We set the bdi here to the queue backing, file systems can
758          * overwrite this in ->fill_super()
759          */
760         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
761         return 0;
762 }
763
764 static int test_bdev_super(struct super_block *s, void *data)
765 {
766         return (void *)s->s_bdev == data;
767 }
768
769 struct dentry *mount_bdev(struct file_system_type *fs_type,
770         int flags, const char *dev_name, void *data,
771         int (*fill_super)(struct super_block *, void *, int))
772 {
773         struct block_device *bdev;
774         struct super_block *s;
775         fmode_t mode = FMODE_READ | FMODE_EXCL;
776         int error = 0;
777
778         if (!(flags & MS_RDONLY))
779                 mode |= FMODE_WRITE;
780
781         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
782         if (IS_ERR(bdev))
783                 return ERR_CAST(bdev);
784
785         /*
786          * once the super is inserted into the list by sget, s_umount
787          * will protect the lockfs code from trying to start a snapshot
788          * while we are mounting
789          */
790         mutex_lock(&bdev->bd_fsfreeze_mutex);
791         if (bdev->bd_fsfreeze_count > 0) {
792                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
793                 error = -EBUSY;
794                 goto error_bdev;
795         }
796         s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
797         mutex_unlock(&bdev->bd_fsfreeze_mutex);
798         if (IS_ERR(s))
799                 goto error_s;
800
801         if (s->s_root) {
802                 if ((flags ^ s->s_flags) & MS_RDONLY) {
803                         deactivate_locked_super(s);
804                         error = -EBUSY;
805                         goto error_bdev;
806                 }
807
808                 /*
809                  * s_umount nests inside bd_mutex during
810                  * __invalidate_device().  blkdev_put() acquires
811                  * bd_mutex and can't be called under s_umount.  Drop
812                  * s_umount temporarily.  This is safe as we're
813                  * holding an active reference.
814                  */
815                 up_write(&s->s_umount);
816                 blkdev_put(bdev, mode);
817                 down_write(&s->s_umount);
818         } else {
819                 char b[BDEVNAME_SIZE];
820
821                 s->s_flags = flags | MS_NOSEC;
822                 s->s_mode = mode;
823                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
824                 sb_set_blocksize(s, block_size(bdev));
825                 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
826                 if (error) {
827                         deactivate_locked_super(s);
828                         goto error;
829                 }
830
831                 s->s_flags |= MS_ACTIVE;
832                 bdev->bd_super = s;
833         }
834
835         return dget(s->s_root);
836
837 error_s:
838         error = PTR_ERR(s);
839 error_bdev:
840         blkdev_put(bdev, mode);
841 error:
842         return ERR_PTR(error);
843 }
844 EXPORT_SYMBOL(mount_bdev);
845
846 void kill_block_super(struct super_block *sb)
847 {
848         struct block_device *bdev = sb->s_bdev;
849         fmode_t mode = sb->s_mode;
850
851         bdev->bd_super = NULL;
852         generic_shutdown_super(sb);
853         sync_blockdev(bdev);
854         WARN_ON_ONCE(!(mode & FMODE_EXCL));
855         blkdev_put(bdev, mode | FMODE_EXCL);
856 }
857
858 EXPORT_SYMBOL(kill_block_super);
859 #endif
860
861 struct dentry *mount_nodev(struct file_system_type *fs_type,
862         int flags, void *data,
863         int (*fill_super)(struct super_block *, void *, int))
864 {
865         int error;
866         struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
867
868         if (IS_ERR(s))
869                 return ERR_CAST(s);
870
871         s->s_flags = flags;
872
873         error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
874         if (error) {
875                 deactivate_locked_super(s);
876                 return ERR_PTR(error);
877         }
878         s->s_flags |= MS_ACTIVE;
879         return dget(s->s_root);
880 }
881 EXPORT_SYMBOL(mount_nodev);
882
883 static int compare_single(struct super_block *s, void *p)
884 {
885         return 1;
886 }
887
888 struct dentry *mount_single(struct file_system_type *fs_type,
889         int flags, void *data,
890         int (*fill_super)(struct super_block *, void *, int))
891 {
892         struct super_block *s;
893         int error;
894
895         s = sget(fs_type, compare_single, set_anon_super, NULL);
896         if (IS_ERR(s))
897                 return ERR_CAST(s);
898         if (!s->s_root) {
899                 s->s_flags = flags;
900                 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
901                 if (error) {
902                         deactivate_locked_super(s);
903                         return ERR_PTR(error);
904                 }
905                 s->s_flags |= MS_ACTIVE;
906         } else {
907                 do_remount_sb(s, flags, data, 0);
908         }
909         return dget(s->s_root);
910 }
911 EXPORT_SYMBOL(mount_single);
912
913 struct dentry *
914 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
915 {
916         struct dentry *root;
917         struct super_block *sb;
918         char *secdata = NULL;
919         int error = -ENOMEM;
920
921         if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
922                 secdata = alloc_secdata();
923                 if (!secdata)
924                         goto out;
925
926                 error = security_sb_copy_data(data, secdata);
927                 if (error)
928                         goto out_free_secdata;
929         }
930
931         root = type->mount(type, flags, name, data);
932         if (IS_ERR(root)) {
933                 error = PTR_ERR(root);
934                 goto out_free_secdata;
935         }
936         sb = root->d_sb;
937         BUG_ON(!sb);
938         WARN_ON(!sb->s_bdi);
939         WARN_ON(sb->s_bdi == &default_backing_dev_info);
940         sb->s_flags |= MS_BORN;
941
942         error = security_sb_kern_mount(sb, flags, secdata);
943         if (error)
944                 goto out_sb;
945
946         /*
947          * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
948          * but s_maxbytes was an unsigned long long for many releases. Throw
949          * this warning for a little while to try and catch filesystems that
950          * violate this rule.
951          */
952         WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
953                 "negative value (%lld)\n", type->name, sb->s_maxbytes);
954
955         up_write(&sb->s_umount);
956         free_secdata(secdata);
957         return root;
958 out_sb:
959         dput(root);
960         deactivate_locked_super(sb);
961 out_free_secdata:
962         free_secdata(secdata);
963 out:
964         return ERR_PTR(error);
965 }
966
967 /**
968  * freeze_super - lock the filesystem and force it into a consistent state
969  * @sb: the super to lock
970  *
971  * Syncs the super to make sure the filesystem is consistent and calls the fs's
972  * freeze_fs.  Subsequent calls to this without first thawing the fs will return
973  * -EBUSY.
974  */
975 int freeze_super(struct super_block *sb)
976 {
977         int ret;
978
979         atomic_inc(&sb->s_active);
980         down_write(&sb->s_umount);
981         if (sb->s_frozen) {
982                 deactivate_locked_super(sb);
983                 return -EBUSY;
984         }
985
986         if (sb->s_flags & MS_RDONLY) {
987                 sb->s_frozen = SB_FREEZE_TRANS;
988                 smp_wmb();
989                 up_write(&sb->s_umount);
990                 return 0;
991         }
992
993         sb->s_frozen = SB_FREEZE_WRITE;
994         smp_wmb();
995
996         sync_filesystem(sb);
997
998         sb->s_frozen = SB_FREEZE_TRANS;
999         smp_wmb();
1000
1001         sync_blockdev(sb->s_bdev);
1002         if (sb->s_op->freeze_fs) {
1003                 ret = sb->s_op->freeze_fs(sb);
1004                 if (ret) {
1005                         printk(KERN_ERR
1006                                 "VFS:Filesystem freeze failed\n");
1007                         sb->s_frozen = SB_UNFROZEN;
1008                         deactivate_locked_super(sb);
1009                         return ret;
1010                 }
1011         }
1012         up_write(&sb->s_umount);
1013         return 0;
1014 }
1015 EXPORT_SYMBOL(freeze_super);
1016
1017 /**
1018  * thaw_super -- unlock filesystem
1019  * @sb: the super to thaw
1020  *
1021  * Unlocks the filesystem and marks it writeable again after freeze_super().
1022  */
1023 int thaw_super(struct super_block *sb)
1024 {
1025         int error;
1026
1027         down_write(&sb->s_umount);
1028         if (sb->s_frozen == SB_UNFROZEN) {
1029                 up_write(&sb->s_umount);
1030                 return -EINVAL;
1031         }
1032
1033         if (sb->s_flags & MS_RDONLY)
1034                 goto out;
1035
1036         if (sb->s_op->unfreeze_fs) {
1037                 error = sb->s_op->unfreeze_fs(sb);
1038                 if (error) {
1039                         printk(KERN_ERR
1040                                 "VFS:Filesystem thaw failed\n");
1041                         sb->s_frozen = SB_FREEZE_TRANS;
1042                         up_write(&sb->s_umount);
1043                         return error;
1044                 }
1045         }
1046
1047 out:
1048         sb->s_frozen = SB_UNFROZEN;
1049         smp_wmb();
1050         wake_up(&sb->s_wait_unfrozen);
1051         deactivate_locked_super(sb);
1052
1053         return 0;
1054 }
1055 EXPORT_SYMBOL(thaw_super);