more conservative S_NOSEC handling
[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 <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         get_fs_excl();
249         mutex_lock(&sb->s_lock);
250 }
251
252 void unlock_super(struct super_block * sb)
253 {
254         put_fs_excl();
255         mutex_unlock(&sb->s_lock);
256 }
257
258 EXPORT_SYMBOL(lock_super);
259 EXPORT_SYMBOL(unlock_super);
260
261 /**
262  *      generic_shutdown_super  -       common helper for ->kill_sb()
263  *      @sb: superblock to kill
264  *
265  *      generic_shutdown_super() does all fs-independent work on superblock
266  *      shutdown.  Typical ->kill_sb() should pick all fs-specific objects
267  *      that need destruction out of superblock, call generic_shutdown_super()
268  *      and release aforementioned objects.  Note: dentries and inodes _are_
269  *      taken care of and do not need specific handling.
270  *
271  *      Upon calling this function, the filesystem may no longer alter or
272  *      rearrange the set of dentries belonging to this super_block, nor may it
273  *      change the attachments of dentries to inodes.
274  */
275 void generic_shutdown_super(struct super_block *sb)
276 {
277         const struct super_operations *sop = sb->s_op;
278
279
280         if (sb->s_root) {
281                 shrink_dcache_for_umount(sb);
282                 sync_filesystem(sb);
283                 get_fs_excl();
284                 sb->s_flags &= ~MS_ACTIVE;
285
286                 fsnotify_unmount_inodes(&sb->s_inodes);
287
288                 evict_inodes(sb);
289
290                 if (sop->put_super)
291                         sop->put_super(sb);
292
293                 if (!list_empty(&sb->s_inodes)) {
294                         printk("VFS: Busy inodes after unmount of %s. "
295                            "Self-destruct in 5 seconds.  Have a nice day...\n",
296                            sb->s_id);
297                 }
298                 put_fs_excl();
299         }
300         spin_lock(&sb_lock);
301         /* should be initialized for __put_super_and_need_restart() */
302         list_del_init(&sb->s_instances);
303         spin_unlock(&sb_lock);
304         up_write(&sb->s_umount);
305 }
306
307 EXPORT_SYMBOL(generic_shutdown_super);
308
309 /**
310  *      sget    -       find or create a superblock
311  *      @type:  filesystem type superblock should belong to
312  *      @test:  comparison callback
313  *      @set:   setup callback
314  *      @data:  argument to each of them
315  */
316 struct super_block *sget(struct file_system_type *type,
317                         int (*test)(struct super_block *,void *),
318                         int (*set)(struct super_block *,void *),
319                         void *data)
320 {
321         struct super_block *s = NULL;
322         struct super_block *old;
323         int err;
324
325 retry:
326         spin_lock(&sb_lock);
327         if (test) {
328                 list_for_each_entry(old, &type->fs_supers, s_instances) {
329                         if (!test(old, data))
330                                 continue;
331                         if (!grab_super(old))
332                                 goto retry;
333                         if (s) {
334                                 up_write(&s->s_umount);
335                                 destroy_super(s);
336                                 s = NULL;
337                         }
338                         down_write(&old->s_umount);
339                         if (unlikely(!(old->s_flags & MS_BORN))) {
340                                 deactivate_locked_super(old);
341                                 goto retry;
342                         }
343                         return old;
344                 }
345         }
346         if (!s) {
347                 spin_unlock(&sb_lock);
348                 s = alloc_super(type);
349                 if (!s)
350                         return ERR_PTR(-ENOMEM);
351                 goto retry;
352         }
353                 
354         err = set(s, data);
355         if (err) {
356                 spin_unlock(&sb_lock);
357                 up_write(&s->s_umount);
358                 destroy_super(s);
359                 return ERR_PTR(err);
360         }
361         s->s_type = type;
362         strlcpy(s->s_id, type->name, sizeof(s->s_id));
363         list_add_tail(&s->s_list, &super_blocks);
364         list_add(&s->s_instances, &type->fs_supers);
365         spin_unlock(&sb_lock);
366         get_filesystem(type);
367         return s;
368 }
369
370 EXPORT_SYMBOL(sget);
371
372 void drop_super(struct super_block *sb)
373 {
374         up_read(&sb->s_umount);
375         put_super(sb);
376 }
377
378 EXPORT_SYMBOL(drop_super);
379
380 /**
381  * sync_supers - helper for periodic superblock writeback
382  *
383  * Call the write_super method if present on all dirty superblocks in
384  * the system.  This is for the periodic writeback used by most older
385  * filesystems.  For data integrity superblock writeback use
386  * sync_filesystems() instead.
387  *
388  * Note: check the dirty flag before waiting, so we don't
389  * hold up the sync while mounting a device. (The newly
390  * mounted device won't need syncing.)
391  */
392 void sync_supers(void)
393 {
394         struct super_block *sb, *p = NULL;
395
396         spin_lock(&sb_lock);
397         list_for_each_entry(sb, &super_blocks, s_list) {
398                 if (list_empty(&sb->s_instances))
399                         continue;
400                 if (sb->s_op->write_super && sb->s_dirt) {
401                         sb->s_count++;
402                         spin_unlock(&sb_lock);
403
404                         down_read(&sb->s_umount);
405                         if (sb->s_root && sb->s_dirt)
406                                 sb->s_op->write_super(sb);
407                         up_read(&sb->s_umount);
408
409                         spin_lock(&sb_lock);
410                         if (p)
411                                 __put_super(p);
412                         p = sb;
413                 }
414         }
415         if (p)
416                 __put_super(p);
417         spin_unlock(&sb_lock);
418 }
419
420 /**
421  *      iterate_supers - call function for all active superblocks
422  *      @f: function to call
423  *      @arg: argument to pass to it
424  *
425  *      Scans the superblock list and calls given function, passing it
426  *      locked superblock and given argument.
427  */
428 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
429 {
430         struct super_block *sb, *p = NULL;
431
432         spin_lock(&sb_lock);
433         list_for_each_entry(sb, &super_blocks, s_list) {
434                 if (list_empty(&sb->s_instances))
435                         continue;
436                 sb->s_count++;
437                 spin_unlock(&sb_lock);
438
439                 down_read(&sb->s_umount);
440                 if (sb->s_root)
441                         f(sb, arg);
442                 up_read(&sb->s_umount);
443
444                 spin_lock(&sb_lock);
445                 if (p)
446                         __put_super(p);
447                 p = sb;
448         }
449         if (p)
450                 __put_super(p);
451         spin_unlock(&sb_lock);
452 }
453
454 /**
455  *      get_super - get the superblock of a device
456  *      @bdev: device to get the superblock for
457  *      
458  *      Scans the superblock list and finds the superblock of the file system
459  *      mounted on the device given. %NULL is returned if no match is found.
460  */
461
462 struct super_block *get_super(struct block_device *bdev)
463 {
464         struct super_block *sb;
465
466         if (!bdev)
467                 return NULL;
468
469         spin_lock(&sb_lock);
470 rescan:
471         list_for_each_entry(sb, &super_blocks, s_list) {
472                 if (list_empty(&sb->s_instances))
473                         continue;
474                 if (sb->s_bdev == bdev) {
475                         sb->s_count++;
476                         spin_unlock(&sb_lock);
477                         down_read(&sb->s_umount);
478                         /* still alive? */
479                         if (sb->s_root)
480                                 return sb;
481                         up_read(&sb->s_umount);
482                         /* nope, got unmounted */
483                         spin_lock(&sb_lock);
484                         __put_super(sb);
485                         goto rescan;
486                 }
487         }
488         spin_unlock(&sb_lock);
489         return NULL;
490 }
491
492 EXPORT_SYMBOL(get_super);
493
494 /**
495  * get_active_super - get an active reference to the superblock of a device
496  * @bdev: device to get the superblock for
497  *
498  * Scans the superblock list and finds the superblock of the file system
499  * mounted on the device given.  Returns the superblock with an active
500  * reference or %NULL if none was found.
501  */
502 struct super_block *get_active_super(struct block_device *bdev)
503 {
504         struct super_block *sb;
505
506         if (!bdev)
507                 return NULL;
508
509 restart:
510         spin_lock(&sb_lock);
511         list_for_each_entry(sb, &super_blocks, s_list) {
512                 if (list_empty(&sb->s_instances))
513                         continue;
514                 if (sb->s_bdev == bdev) {
515                         if (grab_super(sb)) /* drops sb_lock */
516                                 return sb;
517                         else
518                                 goto restart;
519                 }
520         }
521         spin_unlock(&sb_lock);
522         return NULL;
523 }
524  
525 struct super_block *user_get_super(dev_t dev)
526 {
527         struct super_block *sb;
528
529         spin_lock(&sb_lock);
530 rescan:
531         list_for_each_entry(sb, &super_blocks, s_list) {
532                 if (list_empty(&sb->s_instances))
533                         continue;
534                 if (sb->s_dev ==  dev) {
535                         sb->s_count++;
536                         spin_unlock(&sb_lock);
537                         down_read(&sb->s_umount);
538                         /* still alive? */
539                         if (sb->s_root)
540                                 return sb;
541                         up_read(&sb->s_umount);
542                         /* nope, got unmounted */
543                         spin_lock(&sb_lock);
544                         __put_super(sb);
545                         goto rescan;
546                 }
547         }
548         spin_unlock(&sb_lock);
549         return NULL;
550 }
551
552 /**
553  *      do_remount_sb - asks filesystem to change mount options.
554  *      @sb:    superblock in question
555  *      @flags: numeric part of options
556  *      @data:  the rest of options
557  *      @force: whether or not to force the change
558  *
559  *      Alters the mount options of a mounted file system.
560  */
561 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
562 {
563         int retval;
564         int remount_ro;
565
566         if (sb->s_frozen != SB_UNFROZEN)
567                 return -EBUSY;
568
569 #ifdef CONFIG_BLOCK
570         if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
571                 return -EACCES;
572 #endif
573
574         if (flags & MS_RDONLY)
575                 acct_auto_close(sb);
576         shrink_dcache_sb(sb);
577         sync_filesystem(sb);
578
579         remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
580
581         /* If we are remounting RDONLY and current sb is read/write,
582            make sure there are no rw files opened */
583         if (remount_ro) {
584                 if (force)
585                         mark_files_ro(sb);
586                 else if (!fs_may_remount_ro(sb))
587                         return -EBUSY;
588         }
589
590         if (sb->s_op->remount_fs) {
591                 retval = sb->s_op->remount_fs(sb, &flags, data);
592                 if (retval)
593                         return retval;
594         }
595         sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
596
597         /*
598          * Some filesystems modify their metadata via some other path than the
599          * bdev buffer cache (eg. use a private mapping, or directories in
600          * pagecache, etc). Also file data modifications go via their own
601          * mappings. So If we try to mount readonly then copy the filesystem
602          * from bdev, we could get stale data, so invalidate it to give a best
603          * effort at coherency.
604          */
605         if (remount_ro && sb->s_bdev)
606                 invalidate_bdev(sb->s_bdev);
607         return 0;
608 }
609
610 static void do_emergency_remount(struct work_struct *work)
611 {
612         struct super_block *sb, *p = NULL;
613
614         spin_lock(&sb_lock);
615         list_for_each_entry(sb, &super_blocks, s_list) {
616                 if (list_empty(&sb->s_instances))
617                         continue;
618                 sb->s_count++;
619                 spin_unlock(&sb_lock);
620                 down_write(&sb->s_umount);
621                 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
622                         /*
623                          * What lock protects sb->s_flags??
624                          */
625                         do_remount_sb(sb, MS_RDONLY, NULL, 1);
626                 }
627                 up_write(&sb->s_umount);
628                 spin_lock(&sb_lock);
629                 if (p)
630                         __put_super(p);
631                 p = sb;
632         }
633         if (p)
634                 __put_super(p);
635         spin_unlock(&sb_lock);
636         kfree(work);
637         printk("Emergency Remount complete\n");
638 }
639
640 void emergency_remount(void)
641 {
642         struct work_struct *work;
643
644         work = kmalloc(sizeof(*work), GFP_ATOMIC);
645         if (work) {
646                 INIT_WORK(work, do_emergency_remount);
647                 schedule_work(work);
648         }
649 }
650
651 /*
652  * Unnamed block devices are dummy devices used by virtual
653  * filesystems which don't use real block-devices.  -- jrs
654  */
655
656 static DEFINE_IDA(unnamed_dev_ida);
657 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
658 static int unnamed_dev_start = 0; /* don't bother trying below it */
659
660 int set_anon_super(struct super_block *s, void *data)
661 {
662         int dev;
663         int error;
664
665  retry:
666         if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
667                 return -ENOMEM;
668         spin_lock(&unnamed_dev_lock);
669         error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
670         if (!error)
671                 unnamed_dev_start = dev + 1;
672         spin_unlock(&unnamed_dev_lock);
673         if (error == -EAGAIN)
674                 /* We raced and lost with another CPU. */
675                 goto retry;
676         else if (error)
677                 return -EAGAIN;
678
679         if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
680                 spin_lock(&unnamed_dev_lock);
681                 ida_remove(&unnamed_dev_ida, dev);
682                 if (unnamed_dev_start > dev)
683                         unnamed_dev_start = dev;
684                 spin_unlock(&unnamed_dev_lock);
685                 return -EMFILE;
686         }
687         s->s_dev = MKDEV(0, dev & MINORMASK);
688         s->s_bdi = &noop_backing_dev_info;
689         return 0;
690 }
691
692 EXPORT_SYMBOL(set_anon_super);
693
694 void kill_anon_super(struct super_block *sb)
695 {
696         int slot = MINOR(sb->s_dev);
697
698         generic_shutdown_super(sb);
699         spin_lock(&unnamed_dev_lock);
700         ida_remove(&unnamed_dev_ida, slot);
701         if (slot < unnamed_dev_start)
702                 unnamed_dev_start = slot;
703         spin_unlock(&unnamed_dev_lock);
704 }
705
706 EXPORT_SYMBOL(kill_anon_super);
707
708 void kill_litter_super(struct super_block *sb)
709 {
710         if (sb->s_root)
711                 d_genocide(sb->s_root);
712         kill_anon_super(sb);
713 }
714
715 EXPORT_SYMBOL(kill_litter_super);
716
717 static int ns_test_super(struct super_block *sb, void *data)
718 {
719         return sb->s_fs_info == data;
720 }
721
722 static int ns_set_super(struct super_block *sb, void *data)
723 {
724         sb->s_fs_info = data;
725         return set_anon_super(sb, NULL);
726 }
727
728 struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
729         void *data, int (*fill_super)(struct super_block *, void *, int))
730 {
731         struct super_block *sb;
732
733         sb = sget(fs_type, ns_test_super, ns_set_super, data);
734         if (IS_ERR(sb))
735                 return ERR_CAST(sb);
736
737         if (!sb->s_root) {
738                 int err;
739                 sb->s_flags = flags;
740                 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
741                 if (err) {
742                         deactivate_locked_super(sb);
743                         return ERR_PTR(err);
744                 }
745
746                 sb->s_flags |= MS_ACTIVE;
747         }
748
749         return dget(sb->s_root);
750 }
751
752 EXPORT_SYMBOL(mount_ns);
753
754 #ifdef CONFIG_BLOCK
755 static int set_bdev_super(struct super_block *s, void *data)
756 {
757         s->s_bdev = data;
758         s->s_dev = s->s_bdev->bd_dev;
759
760         /*
761          * We set the bdi here to the queue backing, file systems can
762          * overwrite this in ->fill_super()
763          */
764         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
765         return 0;
766 }
767
768 static int test_bdev_super(struct super_block *s, void *data)
769 {
770         return (void *)s->s_bdev == data;
771 }
772
773 struct dentry *mount_bdev(struct file_system_type *fs_type,
774         int flags, const char *dev_name, void *data,
775         int (*fill_super)(struct super_block *, void *, int))
776 {
777         struct block_device *bdev;
778         struct super_block *s;
779         fmode_t mode = FMODE_READ | FMODE_EXCL;
780         int error = 0;
781
782         if (!(flags & MS_RDONLY))
783                 mode |= FMODE_WRITE;
784
785         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
786         if (IS_ERR(bdev))
787                 return ERR_CAST(bdev);
788
789         /*
790          * once the super is inserted into the list by sget, s_umount
791          * will protect the lockfs code from trying to start a snapshot
792          * while we are mounting
793          */
794         mutex_lock(&bdev->bd_fsfreeze_mutex);
795         if (bdev->bd_fsfreeze_count > 0) {
796                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
797                 error = -EBUSY;
798                 goto error_bdev;
799         }
800         s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
801         mutex_unlock(&bdev->bd_fsfreeze_mutex);
802         if (IS_ERR(s))
803                 goto error_s;
804
805         if (s->s_root) {
806                 if ((flags ^ s->s_flags) & MS_RDONLY) {
807                         deactivate_locked_super(s);
808                         error = -EBUSY;
809                         goto error_bdev;
810                 }
811
812                 /*
813                  * s_umount nests inside bd_mutex during
814                  * __invalidate_device().  blkdev_put() acquires
815                  * bd_mutex and can't be called under s_umount.  Drop
816                  * s_umount temporarily.  This is safe as we're
817                  * holding an active reference.
818                  */
819                 up_write(&s->s_umount);
820                 blkdev_put(bdev, mode);
821                 down_write(&s->s_umount);
822         } else {
823                 char b[BDEVNAME_SIZE];
824
825                 s->s_flags = flags | MS_NOSEC;
826                 s->s_mode = mode;
827                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
828                 sb_set_blocksize(s, block_size(bdev));
829                 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
830                 if (error) {
831                         deactivate_locked_super(s);
832                         goto error;
833                 }
834
835                 s->s_flags |= MS_ACTIVE;
836                 bdev->bd_super = s;
837         }
838
839         return dget(s->s_root);
840
841 error_s:
842         error = PTR_ERR(s);
843 error_bdev:
844         blkdev_put(bdev, mode);
845 error:
846         return ERR_PTR(error);
847 }
848 EXPORT_SYMBOL(mount_bdev);
849
850 void kill_block_super(struct super_block *sb)
851 {
852         struct block_device *bdev = sb->s_bdev;
853         fmode_t mode = sb->s_mode;
854
855         bdev->bd_super = NULL;
856         generic_shutdown_super(sb);
857         sync_blockdev(bdev);
858         WARN_ON_ONCE(!(mode & FMODE_EXCL));
859         blkdev_put(bdev, mode | FMODE_EXCL);
860 }
861
862 EXPORT_SYMBOL(kill_block_super);
863 #endif
864
865 struct dentry *mount_nodev(struct file_system_type *fs_type,
866         int flags, void *data,
867         int (*fill_super)(struct super_block *, void *, int))
868 {
869         int error;
870         struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
871
872         if (IS_ERR(s))
873                 return ERR_CAST(s);
874
875         s->s_flags = flags;
876
877         error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
878         if (error) {
879                 deactivate_locked_super(s);
880                 return ERR_PTR(error);
881         }
882         s->s_flags |= MS_ACTIVE;
883         return dget(s->s_root);
884 }
885 EXPORT_SYMBOL(mount_nodev);
886
887 static int compare_single(struct super_block *s, void *p)
888 {
889         return 1;
890 }
891
892 struct dentry *mount_single(struct file_system_type *fs_type,
893         int flags, void *data,
894         int (*fill_super)(struct super_block *, void *, int))
895 {
896         struct super_block *s;
897         int error;
898
899         s = sget(fs_type, compare_single, set_anon_super, NULL);
900         if (IS_ERR(s))
901                 return ERR_CAST(s);
902         if (!s->s_root) {
903                 s->s_flags = flags;
904                 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
905                 if (error) {
906                         deactivate_locked_super(s);
907                         return ERR_PTR(error);
908                 }
909                 s->s_flags |= MS_ACTIVE;
910         } else {
911                 do_remount_sb(s, flags, data, 0);
912         }
913         return dget(s->s_root);
914 }
915 EXPORT_SYMBOL(mount_single);
916
917 struct dentry *
918 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
919 {
920         struct dentry *root;
921         struct super_block *sb;
922         char *secdata = NULL;
923         int error = -ENOMEM;
924
925         if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
926                 secdata = alloc_secdata();
927                 if (!secdata)
928                         goto out;
929
930                 error = security_sb_copy_data(data, secdata);
931                 if (error)
932                         goto out_free_secdata;
933         }
934
935         root = type->mount(type, flags, name, data);
936         if (IS_ERR(root)) {
937                 error = PTR_ERR(root);
938                 goto out_free_secdata;
939         }
940         sb = root->d_sb;
941         BUG_ON(!sb);
942         WARN_ON(!sb->s_bdi);
943         WARN_ON(sb->s_bdi == &default_backing_dev_info);
944         sb->s_flags |= MS_BORN;
945
946         error = security_sb_kern_mount(sb, flags, secdata);
947         if (error)
948                 goto out_sb;
949
950         /*
951          * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
952          * but s_maxbytes was an unsigned long long for many releases. Throw
953          * this warning for a little while to try and catch filesystems that
954          * violate this rule.
955          */
956         WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
957                 "negative value (%lld)\n", type->name, sb->s_maxbytes);
958
959         up_write(&sb->s_umount);
960         free_secdata(secdata);
961         return root;
962 out_sb:
963         dput(root);
964         deactivate_locked_super(sb);
965 out_free_secdata:
966         free_secdata(secdata);
967 out:
968         return ERR_PTR(error);
969 }
970
971 /**
972  * freeze_super - lock the filesystem and force it into a consistent state
973  * @sb: the super to lock
974  *
975  * Syncs the super to make sure the filesystem is consistent and calls the fs's
976  * freeze_fs.  Subsequent calls to this without first thawing the fs will return
977  * -EBUSY.
978  */
979 int freeze_super(struct super_block *sb)
980 {
981         int ret;
982
983         atomic_inc(&sb->s_active);
984         down_write(&sb->s_umount);
985         if (sb->s_frozen) {
986                 deactivate_locked_super(sb);
987                 return -EBUSY;
988         }
989
990         if (sb->s_flags & MS_RDONLY) {
991                 sb->s_frozen = SB_FREEZE_TRANS;
992                 smp_wmb();
993                 up_write(&sb->s_umount);
994                 return 0;
995         }
996
997         sb->s_frozen = SB_FREEZE_WRITE;
998         smp_wmb();
999
1000         sync_filesystem(sb);
1001
1002         sb->s_frozen = SB_FREEZE_TRANS;
1003         smp_wmb();
1004
1005         sync_blockdev(sb->s_bdev);
1006         if (sb->s_op->freeze_fs) {
1007                 ret = sb->s_op->freeze_fs(sb);
1008                 if (ret) {
1009                         printk(KERN_ERR
1010                                 "VFS:Filesystem freeze failed\n");
1011                         sb->s_frozen = SB_UNFROZEN;
1012                         deactivate_locked_super(sb);
1013                         return ret;
1014                 }
1015         }
1016         up_write(&sb->s_umount);
1017         return 0;
1018 }
1019 EXPORT_SYMBOL(freeze_super);
1020
1021 /**
1022  * thaw_super -- unlock filesystem
1023  * @sb: the super to thaw
1024  *
1025  * Unlocks the filesystem and marks it writeable again after freeze_super().
1026  */
1027 int thaw_super(struct super_block *sb)
1028 {
1029         int error;
1030
1031         down_write(&sb->s_umount);
1032         if (sb->s_frozen == SB_UNFROZEN) {
1033                 up_write(&sb->s_umount);
1034                 return -EINVAL;
1035         }
1036
1037         if (sb->s_flags & MS_RDONLY)
1038                 goto out;
1039
1040         if (sb->s_op->unfreeze_fs) {
1041                 error = sb->s_op->unfreeze_fs(sb);
1042                 if (error) {
1043                         printk(KERN_ERR
1044                                 "VFS:Filesystem thaw failed\n");
1045                         sb->s_frozen = SB_FREEZE_TRANS;
1046                         up_write(&sb->s_umount);
1047                         return error;
1048                 }
1049         }
1050
1051 out:
1052         sb->s_frozen = SB_UNFROZEN;
1053         smp_wmb();
1054         wake_up(&sb->s_wait_unfrozen);
1055         deactivate_locked_super(sb);
1056
1057         return 0;
1058 }
1059 EXPORT_SYMBOL(thaw_super);