Root/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/export.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 <linux/fsnotify.h>
36#include <linux/lockdep.h>
37#include "internal.h"
38
39
40LIST_HEAD(super_blocks);
41DEFINE_SPINLOCK(sb_lock);
42
43static char *sb_writers_name[SB_FREEZE_LEVELS] = {
44    "sb_writers",
45    "sb_pagefaults",
46    "sb_internal",
47};
48
49/*
50 * One thing we have to be careful of with a per-sb shrinker is that we don't
51 * drop the last active reference to the superblock from within the shrinker.
52 * If that happens we could trigger unregistering the shrinker from within the
53 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
54 * take a passive reference to the superblock to avoid this from occurring.
55 */
56static int prune_super(struct shrinker *shrink, struct shrink_control *sc)
57{
58    struct super_block *sb;
59    int fs_objects = 0;
60    int total_objects;
61
62    sb = container_of(shrink, struct super_block, s_shrink);
63
64    /*
65     * Deadlock avoidance. We may hold various FS locks, and we don't want
66     * to recurse into the FS that called us in clear_inode() and friends..
67     */
68    if (sc->nr_to_scan && !(sc->gfp_mask & __GFP_FS))
69        return -1;
70
71    if (!grab_super_passive(sb))
72        return -1;
73
74    if (sb->s_op && sb->s_op->nr_cached_objects)
75        fs_objects = sb->s_op->nr_cached_objects(sb);
76
77    total_objects = sb->s_nr_dentry_unused +
78            sb->s_nr_inodes_unused + fs_objects + 1;
79
80    if (sc->nr_to_scan) {
81        int dentries;
82        int inodes;
83
84        /* proportion the scan between the caches */
85        dentries = (sc->nr_to_scan * sb->s_nr_dentry_unused) /
86                            total_objects;
87        inodes = (sc->nr_to_scan * sb->s_nr_inodes_unused) /
88                            total_objects;
89        if (fs_objects)
90            fs_objects = (sc->nr_to_scan * fs_objects) /
91                            total_objects;
92        /*
93         * prune the dcache first as the icache is pinned by it, then
94         * prune the icache, followed by the filesystem specific caches
95         */
96        prune_dcache_sb(sb, dentries);
97        prune_icache_sb(sb, inodes);
98
99        if (fs_objects && sb->s_op->free_cached_objects) {
100            sb->s_op->free_cached_objects(sb, fs_objects);
101            fs_objects = sb->s_op->nr_cached_objects(sb);
102        }
103        total_objects = sb->s_nr_dentry_unused +
104                sb->s_nr_inodes_unused + fs_objects;
105    }
106
107    total_objects = (total_objects / 100) * sysctl_vfs_cache_pressure;
108    drop_super(sb);
109    return total_objects;
110}
111
112static int init_sb_writers(struct super_block *s, struct file_system_type *type)
113{
114    int err;
115    int i;
116
117    for (i = 0; i < SB_FREEZE_LEVELS; i++) {
118        err = percpu_counter_init(&s->s_writers.counter[i], 0);
119        if (err < 0)
120            goto err_out;
121        lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i],
122                 &type->s_writers_key[i], 0);
123    }
124    init_waitqueue_head(&s->s_writers.wait);
125    init_waitqueue_head(&s->s_writers.wait_unfrozen);
126    return 0;
127err_out:
128    while (--i >= 0)
129        percpu_counter_destroy(&s->s_writers.counter[i]);
130    return err;
131}
132
133static void destroy_sb_writers(struct super_block *s)
134{
135    int i;
136
137    for (i = 0; i < SB_FREEZE_LEVELS; i++)
138        percpu_counter_destroy(&s->s_writers.counter[i]);
139}
140
141/**
142 * alloc_super - create new superblock
143 * @type: filesystem type superblock should belong to
144 * @flags: the mount flags
145 *
146 * Allocates and initializes a new &struct super_block. alloc_super()
147 * returns a pointer new superblock or %NULL if allocation had failed.
148 */
149static struct super_block *alloc_super(struct file_system_type *type, int flags)
150{
151    struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
152    static const struct super_operations default_op;
153
154    if (s) {
155        if (security_sb_alloc(s)) {
156            /*
157             * We cannot call security_sb_free() without
158             * security_sb_alloc() succeeding. So bail out manually
159             */
160            kfree(s);
161            s = NULL;
162            goto out;
163        }
164#ifdef CONFIG_SMP
165        s->s_files = alloc_percpu(struct list_head);
166        if (!s->s_files)
167            goto err_out;
168        else {
169            int i;
170
171            for_each_possible_cpu(i)
172                INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i));
173        }
174#else
175        INIT_LIST_HEAD(&s->s_files);
176#endif
177        if (init_sb_writers(s, type))
178            goto err_out;
179        s->s_flags = flags;
180        s->s_bdi = &default_backing_dev_info;
181        INIT_HLIST_NODE(&s->s_instances);
182        INIT_HLIST_BL_HEAD(&s->s_anon);
183        INIT_LIST_HEAD(&s->s_inodes);
184        INIT_LIST_HEAD(&s->s_dentry_lru);
185        INIT_LIST_HEAD(&s->s_inode_lru);
186        spin_lock_init(&s->s_inode_lru_lock);
187        INIT_LIST_HEAD(&s->s_mounts);
188        init_rwsem(&s->s_umount);
189        mutex_init(&s->s_lock);
190        lockdep_set_class(&s->s_umount, &type->s_umount_key);
191        /*
192         * The locking rules for s_lock are up to the
193         * filesystem. For example ext3fs has different
194         * lock ordering than usbfs:
195         */
196        lockdep_set_class(&s->s_lock, &type->s_lock_key);
197        /*
198         * sget() can have s_umount recursion.
199         *
200         * When it cannot find a suitable sb, it allocates a new
201         * one (this one), and tries again to find a suitable old
202         * one.
203         *
204         * In case that succeeds, it will acquire the s_umount
205         * lock of the old one. Since these are clearly distrinct
206         * locks, and this object isn't exposed yet, there's no
207         * risk of deadlocks.
208         *
209         * Annotate this by putting this lock in a different
210         * subclass.
211         */
212        down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
213        s->s_count = 1;
214        atomic_set(&s->s_active, 1);
215        mutex_init(&s->s_vfs_rename_mutex);
216        lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
217        mutex_init(&s->s_dquot.dqio_mutex);
218        mutex_init(&s->s_dquot.dqonoff_mutex);
219        init_rwsem(&s->s_dquot.dqptr_sem);
220        s->s_maxbytes = MAX_NON_LFS;
221        s->s_op = &default_op;
222        s->s_time_gran = 1000000000;
223        s->cleancache_poolid = -1;
224
225        s->s_shrink.seeks = DEFAULT_SEEKS;
226        s->s_shrink.shrink = prune_super;
227        s->s_shrink.batch = 1024;
228    }
229out:
230    return s;
231err_out:
232    security_sb_free(s);
233#ifdef CONFIG_SMP
234    if (s->s_files)
235        free_percpu(s->s_files);
236#endif
237    destroy_sb_writers(s);
238    kfree(s);
239    s = NULL;
240    goto out;
241}
242
243/**
244 * destroy_super - frees a superblock
245 * @s: superblock to free
246 *
247 * Frees a superblock.
248 */
249static inline void destroy_super(struct super_block *s)
250{
251#ifdef CONFIG_SMP
252    free_percpu(s->s_files);
253#endif
254    destroy_sb_writers(s);
255    security_sb_free(s);
256    WARN_ON(!list_empty(&s->s_mounts));
257    kfree(s->s_subtype);
258    kfree(s->s_options);
259    kfree(s);
260}
261
262/* Superblock refcounting */
263
264/*
265 * Drop a superblock's refcount. The caller must hold sb_lock.
266 */
267static void __put_super(struct super_block *sb)
268{
269    if (!--sb->s_count) {
270        list_del_init(&sb->s_list);
271        destroy_super(sb);
272    }
273}
274
275/**
276 * put_super - drop a temporary reference to superblock
277 * @sb: superblock in question
278 *
279 * Drops a temporary reference, frees superblock if there's no
280 * references left.
281 */
282static void put_super(struct super_block *sb)
283{
284    spin_lock(&sb_lock);
285    __put_super(sb);
286    spin_unlock(&sb_lock);
287}
288
289
290/**
291 * deactivate_locked_super - drop an active reference to superblock
292 * @s: superblock to deactivate
293 *
294 * Drops an active reference to superblock, converting it into a temprory
295 * one if there is no other active references left. In that case we
296 * tell fs driver to shut it down and drop the temporary reference we
297 * had just acquired.
298 *
299 * Caller holds exclusive lock on superblock; that lock is released.
300 */
301void deactivate_locked_super(struct super_block *s)
302{
303    struct file_system_type *fs = s->s_type;
304    if (atomic_dec_and_test(&s->s_active)) {
305        cleancache_invalidate_fs(s);
306        fs->kill_sb(s);
307
308        /* caches are now gone, we can safely kill the shrinker now */
309        unregister_shrinker(&s->s_shrink);
310
311        /*
312         * We need to call rcu_barrier so all the delayed rcu free
313         * inodes are flushed before we release the fs module.
314         */
315        rcu_barrier();
316        put_filesystem(fs);
317        put_super(s);
318    } else {
319        up_write(&s->s_umount);
320    }
321}
322
323EXPORT_SYMBOL(deactivate_locked_super);
324
325/**
326 * deactivate_super - drop an active reference to superblock
327 * @s: superblock to deactivate
328 *
329 * Variant of deactivate_locked_super(), except that superblock is *not*
330 * locked by caller. If we are going to drop the final active reference,
331 * lock will be acquired prior to that.
332 */
333void deactivate_super(struct super_block *s)
334{
335        if (!atomic_add_unless(&s->s_active, -1, 1)) {
336        down_write(&s->s_umount);
337        deactivate_locked_super(s);
338    }
339}
340
341EXPORT_SYMBOL(deactivate_super);
342
343/**
344 * grab_super - acquire an active reference
345 * @s: reference we are trying to make active
346 *
347 * Tries to acquire an active reference. grab_super() is used when we
348 * had just found a superblock in super_blocks or fs_type->fs_supers
349 * and want to turn it into a full-blown active reference. grab_super()
350 * is called with sb_lock held and drops it. Returns 1 in case of
351 * success, 0 if we had failed (superblock contents was already dead or
352 * dying when grab_super() had been called).
353 */
354static int grab_super(struct super_block *s) __releases(sb_lock)
355{
356    if (atomic_inc_not_zero(&s->s_active)) {
357        spin_unlock(&sb_lock);
358        return 1;
359    }
360    /* it's going away */
361    s->s_count++;
362    spin_unlock(&sb_lock);
363    /* wait for it to die */
364    down_write(&s->s_umount);
365    up_write(&s->s_umount);
366    put_super(s);
367    return 0;
368}
369
370/*
371 * grab_super_passive - acquire a passive reference
372 * @sb: reference we are trying to grab
373 *
374 * Tries to acquire a passive reference. This is used in places where we
375 * cannot take an active reference but we need to ensure that the
376 * superblock does not go away while we are working on it. It returns
377 * false if a reference was not gained, and returns true with the s_umount
378 * lock held in read mode if a reference is gained. On successful return,
379 * the caller must drop the s_umount lock and the passive reference when
380 * done.
381 */
382bool grab_super_passive(struct super_block *sb)
383{
384    spin_lock(&sb_lock);
385    if (hlist_unhashed(&sb->s_instances)) {
386        spin_unlock(&sb_lock);
387        return false;
388    }
389
390    sb->s_count++;
391    spin_unlock(&sb_lock);
392
393    if (down_read_trylock(&sb->s_umount)) {
394        if (sb->s_root && (sb->s_flags & MS_BORN))
395            return true;
396        up_read(&sb->s_umount);
397    }
398
399    put_super(sb);
400    return false;
401}
402
403/*
404 * Superblock locking. We really ought to get rid of these two.
405 */
406void lock_super(struct super_block * sb)
407{
408    mutex_lock(&sb->s_lock);
409}
410
411void unlock_super(struct super_block * sb)
412{
413    mutex_unlock(&sb->s_lock);
414}
415
416EXPORT_SYMBOL(lock_super);
417EXPORT_SYMBOL(unlock_super);
418
419/**
420 * generic_shutdown_super - common helper for ->kill_sb()
421 * @sb: superblock to kill
422 *
423 * generic_shutdown_super() does all fs-independent work on superblock
424 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
425 * that need destruction out of superblock, call generic_shutdown_super()
426 * and release aforementioned objects. Note: dentries and inodes _are_
427 * taken care of and do not need specific handling.
428 *
429 * Upon calling this function, the filesystem may no longer alter or
430 * rearrange the set of dentries belonging to this super_block, nor may it
431 * change the attachments of dentries to inodes.
432 */
433void generic_shutdown_super(struct super_block *sb)
434{
435    const struct super_operations *sop = sb->s_op;
436
437    if (sb->s_root) {
438        shrink_dcache_for_umount(sb);
439        sync_filesystem(sb);
440        sb->s_flags &= ~MS_ACTIVE;
441
442        fsnotify_unmount_inodes(&sb->s_inodes);
443
444        evict_inodes(sb);
445
446        if (sop->put_super)
447            sop->put_super(sb);
448
449        if (!list_empty(&sb->s_inodes)) {
450            printk("VFS: Busy inodes after unmount of %s. "
451               "Self-destruct in 5 seconds. Have a nice day...\n",
452               sb->s_id);
453        }
454    }
455    spin_lock(&sb_lock);
456    /* should be initialized for __put_super_and_need_restart() */
457    hlist_del_init(&sb->s_instances);
458    spin_unlock(&sb_lock);
459    up_write(&sb->s_umount);
460}
461
462EXPORT_SYMBOL(generic_shutdown_super);
463
464/**
465 * sget - find or create a superblock
466 * @type: filesystem type superblock should belong to
467 * @test: comparison callback
468 * @set: setup callback
469 * @flags: mount flags
470 * @data: argument to each of them
471 */
472struct super_block *sget(struct file_system_type *type,
473            int (*test)(struct super_block *,void *),
474            int (*set)(struct super_block *,void *),
475            int flags,
476            void *data)
477{
478    struct super_block *s = NULL;
479    struct hlist_node *node;
480    struct super_block *old;
481    int err;
482
483retry:
484    spin_lock(&sb_lock);
485    if (test) {
486        hlist_for_each_entry(old, node, &type->fs_supers, s_instances) {
487            if (!test(old, data))
488                continue;
489            if (!grab_super(old))
490                goto retry;
491            if (s) {
492                up_write(&s->s_umount);
493                destroy_super(s);
494                s = NULL;
495            }
496            down_write(&old->s_umount);
497            if (unlikely(!(old->s_flags & MS_BORN))) {
498                deactivate_locked_super(old);
499                goto retry;
500            }
501            return old;
502        }
503    }
504    if (!s) {
505        spin_unlock(&sb_lock);
506        s = alloc_super(type, flags);
507        if (!s)
508            return ERR_PTR(-ENOMEM);
509        goto retry;
510    }
511        
512    err = set(s, data);
513    if (err) {
514        spin_unlock(&sb_lock);
515        up_write(&s->s_umount);
516        destroy_super(s);
517        return ERR_PTR(err);
518    }
519    s->s_type = type;
520    strlcpy(s->s_id, type->name, sizeof(s->s_id));
521    list_add_tail(&s->s_list, &super_blocks);
522    hlist_add_head(&s->s_instances, &type->fs_supers);
523    spin_unlock(&sb_lock);
524    get_filesystem(type);
525    register_shrinker(&s->s_shrink);
526    return s;
527}
528
529EXPORT_SYMBOL(sget);
530
531void drop_super(struct super_block *sb)
532{
533    up_read(&sb->s_umount);
534    put_super(sb);
535}
536
537EXPORT_SYMBOL(drop_super);
538
539/**
540 * iterate_supers - call function for all active superblocks
541 * @f: function to call
542 * @arg: argument to pass to it
543 *
544 * Scans the superblock list and calls given function, passing it
545 * locked superblock and given argument.
546 */
547void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
548{
549    struct super_block *sb, *p = NULL;
550
551    spin_lock(&sb_lock);
552    list_for_each_entry(sb, &super_blocks, s_list) {
553        if (hlist_unhashed(&sb->s_instances))
554            continue;
555        sb->s_count++;
556        spin_unlock(&sb_lock);
557
558        down_read(&sb->s_umount);
559        if (sb->s_root && (sb->s_flags & MS_BORN))
560            f(sb, arg);
561        up_read(&sb->s_umount);
562
563        spin_lock(&sb_lock);
564        if (p)
565            __put_super(p);
566        p = sb;
567    }
568    if (p)
569        __put_super(p);
570    spin_unlock(&sb_lock);
571}
572
573/**
574 * iterate_supers_type - call function for superblocks of given type
575 * @type: fs type
576 * @f: function to call
577 * @arg: argument to pass to it
578 *
579 * Scans the superblock list and calls given function, passing it
580 * locked superblock and given argument.
581 */
582void iterate_supers_type(struct file_system_type *type,
583    void (*f)(struct super_block *, void *), void *arg)
584{
585    struct super_block *sb, *p = NULL;
586    struct hlist_node *node;
587
588    spin_lock(&sb_lock);
589    hlist_for_each_entry(sb, node, &type->fs_supers, s_instances) {
590        sb->s_count++;
591        spin_unlock(&sb_lock);
592
593        down_read(&sb->s_umount);
594        if (sb->s_root && (sb->s_flags & MS_BORN))
595            f(sb, arg);
596        up_read(&sb->s_umount);
597
598        spin_lock(&sb_lock);
599        if (p)
600            __put_super(p);
601        p = sb;
602    }
603    if (p)
604        __put_super(p);
605    spin_unlock(&sb_lock);
606}
607
608EXPORT_SYMBOL(iterate_supers_type);
609
610/**
611 * get_super - get the superblock of a device
612 * @bdev: device to get the superblock for
613 *
614 * Scans the superblock list and finds the superblock of the file system
615 * mounted on the device given. %NULL is returned if no match is found.
616 */
617
618struct super_block *get_super(struct block_device *bdev)
619{
620    struct super_block *sb;
621
622    if (!bdev)
623        return NULL;
624
625    spin_lock(&sb_lock);
626rescan:
627    list_for_each_entry(sb, &super_blocks, s_list) {
628        if (hlist_unhashed(&sb->s_instances))
629            continue;
630        if (sb->s_bdev == bdev) {
631            sb->s_count++;
632            spin_unlock(&sb_lock);
633            down_read(&sb->s_umount);
634            /* still alive? */
635            if (sb->s_root && (sb->s_flags & MS_BORN))
636                return sb;
637            up_read(&sb->s_umount);
638            /* nope, got unmounted */
639            spin_lock(&sb_lock);
640            __put_super(sb);
641            goto rescan;
642        }
643    }
644    spin_unlock(&sb_lock);
645    return NULL;
646}
647
648EXPORT_SYMBOL(get_super);
649
650/**
651 * get_super_thawed - get thawed superblock of a device
652 * @bdev: device to get the superblock for
653 *
654 * Scans the superblock list and finds the superblock of the file system
655 * mounted on the device. The superblock is returned once it is thawed
656 * (or immediately if it was not frozen). %NULL is returned if no match
657 * is found.
658 */
659struct super_block *get_super_thawed(struct block_device *bdev)
660{
661    while (1) {
662        struct super_block *s = get_super(bdev);
663        if (!s || s->s_writers.frozen == SB_UNFROZEN)
664            return s;
665        up_read(&s->s_umount);
666        wait_event(s->s_writers.wait_unfrozen,
667               s->s_writers.frozen == SB_UNFROZEN);
668        put_super(s);
669    }
670}
671EXPORT_SYMBOL(get_super_thawed);
672
673/**
674 * get_active_super - get an active reference to the superblock of a device
675 * @bdev: device to get the superblock for
676 *
677 * Scans the superblock list and finds the superblock of the file system
678 * mounted on the device given. Returns the superblock with an active
679 * reference or %NULL if none was found.
680 */
681struct super_block *get_active_super(struct block_device *bdev)
682{
683    struct super_block *sb;
684
685    if (!bdev)
686        return NULL;
687
688restart:
689    spin_lock(&sb_lock);
690    list_for_each_entry(sb, &super_blocks, s_list) {
691        if (hlist_unhashed(&sb->s_instances))
692            continue;
693        if (sb->s_bdev == bdev) {
694            if (grab_super(sb)) /* drops sb_lock */
695                return sb;
696            else
697                goto restart;
698        }
699    }
700    spin_unlock(&sb_lock);
701    return NULL;
702}
703 
704struct super_block *user_get_super(dev_t dev)
705{
706    struct super_block *sb;
707
708    spin_lock(&sb_lock);
709rescan:
710    list_for_each_entry(sb, &super_blocks, s_list) {
711        if (hlist_unhashed(&sb->s_instances))
712            continue;
713        if (sb->s_dev == dev) {
714            sb->s_count++;
715            spin_unlock(&sb_lock);
716            down_read(&sb->s_umount);
717            /* still alive? */
718            if (sb->s_root && (sb->s_flags & MS_BORN))
719                return sb;
720            up_read(&sb->s_umount);
721            /* nope, got unmounted */
722            spin_lock(&sb_lock);
723            __put_super(sb);
724            goto rescan;
725        }
726    }
727    spin_unlock(&sb_lock);
728    return NULL;
729}
730
731/**
732 * do_remount_sb - asks filesystem to change mount options.
733 * @sb: superblock in question
734 * @flags: numeric part of options
735 * @data: the rest of options
736 * @force: whether or not to force the change
737 *
738 * Alters the mount options of a mounted file system.
739 */
740int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
741{
742    int retval;
743    int remount_ro;
744
745    if (sb->s_writers.frozen != SB_UNFROZEN)
746        return -EBUSY;
747
748#ifdef CONFIG_BLOCK
749    if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
750        return -EACCES;
751#endif
752
753    if (flags & MS_RDONLY)
754        acct_auto_close(sb);
755    shrink_dcache_sb(sb);
756    sync_filesystem(sb);
757
758    remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
759
760    /* If we are remounting RDONLY and current sb is read/write,
761       make sure there are no rw files opened */
762    if (remount_ro) {
763        if (force) {
764            mark_files_ro(sb);
765        } else {
766            retval = sb_prepare_remount_readonly(sb);
767            if (retval)
768                return retval;
769        }
770    }
771
772    if (sb->s_op->remount_fs) {
773        retval = sb->s_op->remount_fs(sb, &flags, data);
774        if (retval) {
775            if (!force)
776                goto cancel_readonly;
777            /* If forced remount, go ahead despite any errors */
778            WARN(1, "forced remount of a %s fs returned %i\n",
779                 sb->s_type->name, retval);
780        }
781    }
782    sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
783    /* Needs to be ordered wrt mnt_is_readonly() */
784    smp_wmb();
785    sb->s_readonly_remount = 0;
786
787    /*
788     * Some filesystems modify their metadata via some other path than the
789     * bdev buffer cache (eg. use a private mapping, or directories in
790     * pagecache, etc). Also file data modifications go via their own
791     * mappings. So If we try to mount readonly then copy the filesystem
792     * from bdev, we could get stale data, so invalidate it to give a best
793     * effort at coherency.
794     */
795    if (remount_ro && sb->s_bdev)
796        invalidate_bdev(sb->s_bdev);
797    return 0;
798
799cancel_readonly:
800    sb->s_readonly_remount = 0;
801    return retval;
802}
803
804static void do_emergency_remount(struct work_struct *work)
805{
806    struct super_block *sb, *p = NULL;
807
808    spin_lock(&sb_lock);
809    list_for_each_entry(sb, &super_blocks, s_list) {
810        if (hlist_unhashed(&sb->s_instances))
811            continue;
812        sb->s_count++;
813        spin_unlock(&sb_lock);
814        down_write(&sb->s_umount);
815        if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
816            !(sb->s_flags & MS_RDONLY)) {
817            /*
818             * What lock protects sb->s_flags??
819             */
820            do_remount_sb(sb, MS_RDONLY, NULL, 1);
821        }
822        up_write(&sb->s_umount);
823        spin_lock(&sb_lock);
824        if (p)
825            __put_super(p);
826        p = sb;
827    }
828    if (p)
829        __put_super(p);
830    spin_unlock(&sb_lock);
831    kfree(work);
832    printk("Emergency Remount complete\n");
833}
834
835void emergency_remount(void)
836{
837    struct work_struct *work;
838
839    work = kmalloc(sizeof(*work), GFP_ATOMIC);
840    if (work) {
841        INIT_WORK(work, do_emergency_remount);
842        schedule_work(work);
843    }
844}
845
846/*
847 * Unnamed block devices are dummy devices used by virtual
848 * filesystems which don't use real block-devices. -- jrs
849 */
850
851static DEFINE_IDA(unnamed_dev_ida);
852static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
853static int unnamed_dev_start = 0; /* don't bother trying below it */
854
855int get_anon_bdev(dev_t *p)
856{
857    int dev;
858    int error;
859
860 retry:
861    if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
862        return -ENOMEM;
863    spin_lock(&unnamed_dev_lock);
864    error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
865    if (!error)
866        unnamed_dev_start = dev + 1;
867    spin_unlock(&unnamed_dev_lock);
868    if (error == -EAGAIN)
869        /* We raced and lost with another CPU. */
870        goto retry;
871    else if (error)
872        return -EAGAIN;
873
874    if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
875        spin_lock(&unnamed_dev_lock);
876        ida_remove(&unnamed_dev_ida, dev);
877        if (unnamed_dev_start > dev)
878            unnamed_dev_start = dev;
879        spin_unlock(&unnamed_dev_lock);
880        return -EMFILE;
881    }
882    *p = MKDEV(0, dev & MINORMASK);
883    return 0;
884}
885EXPORT_SYMBOL(get_anon_bdev);
886
887void free_anon_bdev(dev_t dev)
888{
889    int slot = MINOR(dev);
890    spin_lock(&unnamed_dev_lock);
891    ida_remove(&unnamed_dev_ida, slot);
892    if (slot < unnamed_dev_start)
893        unnamed_dev_start = slot;
894    spin_unlock(&unnamed_dev_lock);
895}
896EXPORT_SYMBOL(free_anon_bdev);
897
898int set_anon_super(struct super_block *s, void *data)
899{
900    int error = get_anon_bdev(&s->s_dev);
901    if (!error)
902        s->s_bdi = &noop_backing_dev_info;
903    return error;
904}
905
906EXPORT_SYMBOL(set_anon_super);
907
908void kill_anon_super(struct super_block *sb)
909{
910    dev_t dev = sb->s_dev;
911    generic_shutdown_super(sb);
912    free_anon_bdev(dev);
913}
914
915EXPORT_SYMBOL(kill_anon_super);
916
917void kill_litter_super(struct super_block *sb)
918{
919    if (sb->s_root)
920        d_genocide(sb->s_root);
921    kill_anon_super(sb);
922}
923
924EXPORT_SYMBOL(kill_litter_super);
925
926static int ns_test_super(struct super_block *sb, void *data)
927{
928    return sb->s_fs_info == data;
929}
930
931static int ns_set_super(struct super_block *sb, void *data)
932{
933    sb->s_fs_info = data;
934    return set_anon_super(sb, NULL);
935}
936
937struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
938    void *data, int (*fill_super)(struct super_block *, void *, int))
939{
940    struct super_block *sb;
941
942    sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
943    if (IS_ERR(sb))
944        return ERR_CAST(sb);
945
946    if (!sb->s_root) {
947        int err;
948        err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
949        if (err) {
950            deactivate_locked_super(sb);
951            return ERR_PTR(err);
952        }
953
954        sb->s_flags |= MS_ACTIVE;
955    }
956
957    return dget(sb->s_root);
958}
959
960EXPORT_SYMBOL(mount_ns);
961
962#ifdef CONFIG_BLOCK
963static int set_bdev_super(struct super_block *s, void *data)
964{
965    s->s_bdev = data;
966    s->s_dev = s->s_bdev->bd_dev;
967
968    /*
969     * We set the bdi here to the queue backing, file systems can
970     * overwrite this in ->fill_super()
971     */
972    s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
973    return 0;
974}
975
976static int test_bdev_super(struct super_block *s, void *data)
977{
978    return (void *)s->s_bdev == data;
979}
980
981struct dentry *mount_bdev(struct file_system_type *fs_type,
982    int flags, const char *dev_name, void *data,
983    int (*fill_super)(struct super_block *, void *, int))
984{
985    struct block_device *bdev;
986    struct super_block *s;
987    fmode_t mode = FMODE_READ | FMODE_EXCL;
988    int error = 0;
989
990    if (!(flags & MS_RDONLY))
991        mode |= FMODE_WRITE;
992
993    bdev = blkdev_get_by_path(dev_name, mode, fs_type);
994    if (IS_ERR(bdev))
995        return ERR_CAST(bdev);
996
997    /*
998     * once the super is inserted into the list by sget, s_umount
999     * will protect the lockfs code from trying to start a snapshot
1000     * while we are mounting
1001     */
1002    mutex_lock(&bdev->bd_fsfreeze_mutex);
1003    if (bdev->bd_fsfreeze_count > 0) {
1004        mutex_unlock(&bdev->bd_fsfreeze_mutex);
1005        error = -EBUSY;
1006        goto error_bdev;
1007    }
1008    s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
1009         bdev);
1010    mutex_unlock(&bdev->bd_fsfreeze_mutex);
1011    if (IS_ERR(s))
1012        goto error_s;
1013
1014    if (s->s_root) {
1015        if ((flags ^ s->s_flags) & MS_RDONLY) {
1016            deactivate_locked_super(s);
1017            error = -EBUSY;
1018            goto error_bdev;
1019        }
1020
1021        /*
1022         * s_umount nests inside bd_mutex during
1023         * __invalidate_device(). blkdev_put() acquires
1024         * bd_mutex and can't be called under s_umount. Drop
1025         * s_umount temporarily. This is safe as we're
1026         * holding an active reference.
1027         */
1028        up_write(&s->s_umount);
1029        blkdev_put(bdev, mode);
1030        down_write(&s->s_umount);
1031    } else {
1032        char b[BDEVNAME_SIZE];
1033
1034        s->s_mode = mode;
1035        strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1036        sb_set_blocksize(s, block_size(bdev));
1037        error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1038        if (error) {
1039            deactivate_locked_super(s);
1040            goto error;
1041        }
1042
1043        s->s_flags |= MS_ACTIVE;
1044        bdev->bd_super = s;
1045    }
1046
1047    return dget(s->s_root);
1048
1049error_s:
1050    error = PTR_ERR(s);
1051error_bdev:
1052    blkdev_put(bdev, mode);
1053error:
1054    return ERR_PTR(error);
1055}
1056EXPORT_SYMBOL(mount_bdev);
1057
1058void kill_block_super(struct super_block *sb)
1059{
1060    struct block_device *bdev = sb->s_bdev;
1061    fmode_t mode = sb->s_mode;
1062
1063    bdev->bd_super = NULL;
1064    generic_shutdown_super(sb);
1065    sync_blockdev(bdev);
1066    WARN_ON_ONCE(!(mode & FMODE_EXCL));
1067    blkdev_put(bdev, mode | FMODE_EXCL);
1068}
1069
1070EXPORT_SYMBOL(kill_block_super);
1071#endif
1072
1073struct dentry *mount_nodev(struct file_system_type *fs_type,
1074    int flags, void *data,
1075    int (*fill_super)(struct super_block *, void *, int))
1076{
1077    int error;
1078    struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1079
1080    if (IS_ERR(s))
1081        return ERR_CAST(s);
1082
1083    error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1084    if (error) {
1085        deactivate_locked_super(s);
1086        return ERR_PTR(error);
1087    }
1088    s->s_flags |= MS_ACTIVE;
1089    return dget(s->s_root);
1090}
1091EXPORT_SYMBOL(mount_nodev);
1092
1093static int compare_single(struct super_block *s, void *p)
1094{
1095    return 1;
1096}
1097
1098struct dentry *mount_single(struct file_system_type *fs_type,
1099    int flags, void *data,
1100    int (*fill_super)(struct super_block *, void *, int))
1101{
1102    struct super_block *s;
1103    int error;
1104
1105    s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1106    if (IS_ERR(s))
1107        return ERR_CAST(s);
1108    if (!s->s_root) {
1109        error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1110        if (error) {
1111            deactivate_locked_super(s);
1112            return ERR_PTR(error);
1113        }
1114        s->s_flags |= MS_ACTIVE;
1115    } else {
1116        do_remount_sb(s, flags, data, 0);
1117    }
1118    return dget(s->s_root);
1119}
1120EXPORT_SYMBOL(mount_single);
1121
1122struct dentry *
1123mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1124{
1125    struct dentry *root;
1126    struct super_block *sb;
1127    char *secdata = NULL;
1128    int error = -ENOMEM;
1129
1130    if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1131        secdata = alloc_secdata();
1132        if (!secdata)
1133            goto out;
1134
1135        error = security_sb_copy_data(data, secdata);
1136        if (error)
1137            goto out_free_secdata;
1138    }
1139
1140    root = type->mount(type, flags, name, data);
1141    if (IS_ERR(root)) {
1142        error = PTR_ERR(root);
1143        goto out_free_secdata;
1144    }
1145    sb = root->d_sb;
1146    BUG_ON(!sb);
1147    WARN_ON(!sb->s_bdi);
1148    WARN_ON(sb->s_bdi == &default_backing_dev_info);
1149    sb->s_flags |= MS_BORN;
1150
1151    error = security_sb_kern_mount(sb, flags, secdata);
1152    if (error)
1153        goto out_sb;
1154
1155    /*
1156     * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1157     * but s_maxbytes was an unsigned long long for many releases. Throw
1158     * this warning for a little while to try and catch filesystems that
1159     * violate this rule.
1160     */
1161    WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1162        "negative value (%lld)\n", type->name, sb->s_maxbytes);
1163
1164    up_write(&sb->s_umount);
1165    free_secdata(secdata);
1166    return root;
1167out_sb:
1168    dput(root);
1169    deactivate_locked_super(sb);
1170out_free_secdata:
1171    free_secdata(secdata);
1172out:
1173    return ERR_PTR(error);
1174}
1175
1176/*
1177 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1178 * instead.
1179 */
1180void __sb_end_write(struct super_block *sb, int level)
1181{
1182    percpu_counter_dec(&sb->s_writers.counter[level-1]);
1183    /*
1184     * Make sure s_writers are updated before we wake up waiters in
1185     * freeze_super().
1186     */
1187    smp_mb();
1188    if (waitqueue_active(&sb->s_writers.wait))
1189        wake_up(&sb->s_writers.wait);
1190    rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_);
1191}
1192EXPORT_SYMBOL(__sb_end_write);
1193
1194#ifdef CONFIG_LOCKDEP
1195/*
1196 * We want lockdep to tell us about possible deadlocks with freezing but
1197 * it's it bit tricky to properly instrument it. Getting a freeze protection
1198 * works as getting a read lock but there are subtle problems. XFS for example
1199 * gets freeze protection on internal level twice in some cases, which is OK
1200 * only because we already hold a freeze protection also on higher level. Due
1201 * to these cases we have to tell lockdep we are doing trylock when we
1202 * already hold a freeze protection for a higher freeze level.
1203 */
1204static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock,
1205                unsigned long ip)
1206{
1207    int i;
1208
1209    if (!trylock) {
1210        for (i = 0; i < level - 1; i++)
1211            if (lock_is_held(&sb->s_writers.lock_map[i])) {
1212                trylock = true;
1213                break;
1214            }
1215    }
1216    rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip);
1217}
1218#endif
1219
1220/*
1221 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1222 * instead.
1223 */
1224int __sb_start_write(struct super_block *sb, int level, bool wait)
1225{
1226retry:
1227    if (unlikely(sb->s_writers.frozen >= level)) {
1228        if (!wait)
1229            return 0;
1230        wait_event(sb->s_writers.wait_unfrozen,
1231               sb->s_writers.frozen < level);
1232    }
1233
1234#ifdef CONFIG_LOCKDEP
1235    acquire_freeze_lock(sb, level, !wait, _RET_IP_);
1236#endif
1237    percpu_counter_inc(&sb->s_writers.counter[level-1]);
1238    /*
1239     * Make sure counter is updated before we check for frozen.
1240     * freeze_super() first sets frozen and then checks the counter.
1241     */
1242    smp_mb();
1243    if (unlikely(sb->s_writers.frozen >= level)) {
1244        __sb_end_write(sb, level);
1245        goto retry;
1246    }
1247    return 1;
1248}
1249EXPORT_SYMBOL(__sb_start_write);
1250
1251/**
1252 * sb_wait_write - wait until all writers to given file system finish
1253 * @sb: the super for which we wait
1254 * @level: type of writers we wait for (normal vs page fault)
1255 *
1256 * This function waits until there are no writers of given type to given file
1257 * system. Caller of this function should make sure there can be no new writers
1258 * of type @level before calling this function. Otherwise this function can
1259 * livelock.
1260 */
1261static void sb_wait_write(struct super_block *sb, int level)
1262{
1263    s64 writers;
1264
1265    /*
1266     * We just cycle-through lockdep here so that it does not complain
1267     * about returning with lock to userspace
1268     */
1269    rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_);
1270    rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_);
1271
1272    do {
1273        DEFINE_WAIT(wait);
1274
1275        /*
1276         * We use a barrier in prepare_to_wait() to separate setting
1277         * of frozen and checking of the counter
1278         */
1279        prepare_to_wait(&sb->s_writers.wait, &wait,
1280                TASK_UNINTERRUPTIBLE);
1281
1282        writers = percpu_counter_sum(&sb->s_writers.counter[level-1]);
1283        if (writers)
1284            schedule();
1285
1286        finish_wait(&sb->s_writers.wait, &wait);
1287    } while (writers);
1288}
1289
1290/**
1291 * freeze_super - lock the filesystem and force it into a consistent state
1292 * @sb: the super to lock
1293 *
1294 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1295 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1296 * -EBUSY.
1297 *
1298 * During this function, sb->s_writers.frozen goes through these values:
1299 *
1300 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1301 *
1302 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1303 * writes should be blocked, though page faults are still allowed. We wait for
1304 * all writes to complete and then proceed to the next stage.
1305 *
1306 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1307 * but internal fs threads can still modify the filesystem (although they
1308 * should not dirty new pages or inodes), writeback can run etc. After waiting
1309 * for all running page faults we sync the filesystem which will clean all
1310 * dirty pages and inodes (no new dirty pages or inodes can be created when
1311 * sync is running).
1312 *
1313 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1314 * modification are blocked (e.g. XFS preallocation truncation on inode
1315 * reclaim). This is usually implemented by blocking new transactions for
1316 * filesystems that have them and need this additional guard. After all
1317 * internal writers are finished we call ->freeze_fs() to finish filesystem
1318 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1319 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1320 *
1321 * sb->s_writers.frozen is protected by sb->s_umount.
1322 */
1323int freeze_super(struct super_block *sb)
1324{
1325    int ret;
1326
1327    atomic_inc(&sb->s_active);
1328    down_write(&sb->s_umount);
1329    if (sb->s_writers.frozen != SB_UNFROZEN) {
1330        deactivate_locked_super(sb);
1331        return -EBUSY;
1332    }
1333
1334    if (!(sb->s_flags & MS_BORN)) {
1335        up_write(&sb->s_umount);
1336        return 0; /* sic - it's "nothing to do" */
1337    }
1338
1339    if (sb->s_flags & MS_RDONLY) {
1340        /* Nothing to do really... */
1341        sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1342        up_write(&sb->s_umount);
1343        return 0;
1344    }
1345
1346    /* From now on, no new normal writers can start */
1347    sb->s_writers.frozen = SB_FREEZE_WRITE;
1348    smp_wmb();
1349
1350    /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1351    up_write(&sb->s_umount);
1352
1353    sb_wait_write(sb, SB_FREEZE_WRITE);
1354
1355    /* Now we go and block page faults... */
1356    down_write(&sb->s_umount);
1357    sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1358    smp_wmb();
1359
1360    sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1361
1362    /* All writers are done so after syncing there won't be dirty data */
1363    sync_filesystem(sb);
1364
1365    /* Now wait for internal filesystem counter */
1366    sb->s_writers.frozen = SB_FREEZE_FS;
1367    smp_wmb();
1368    sb_wait_write(sb, SB_FREEZE_FS);
1369
1370    if (sb->s_op->freeze_fs) {
1371        ret = sb->s_op->freeze_fs(sb);
1372        if (ret) {
1373            printk(KERN_ERR
1374                "VFS:Filesystem freeze failed\n");
1375            sb->s_writers.frozen = SB_UNFROZEN;
1376            smp_wmb();
1377            wake_up(&sb->s_writers.wait_unfrozen);
1378            deactivate_locked_super(sb);
1379            return ret;
1380        }
1381    }
1382    /*
1383     * This is just for debugging purposes so that fs can warn if it
1384     * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1385     */
1386    sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1387    up_write(&sb->s_umount);
1388    return 0;
1389}
1390EXPORT_SYMBOL(freeze_super);
1391
1392/**
1393 * thaw_super -- unlock filesystem
1394 * @sb: the super to thaw
1395 *
1396 * Unlocks the filesystem and marks it writeable again after freeze_super().
1397 */
1398int thaw_super(struct super_block *sb)
1399{
1400    int error;
1401
1402    down_write(&sb->s_umount);
1403    if (sb->s_writers.frozen == SB_UNFROZEN) {
1404        up_write(&sb->s_umount);
1405        return -EINVAL;
1406    }
1407
1408    if (sb->s_flags & MS_RDONLY)
1409        goto out;
1410
1411    if (sb->s_op->unfreeze_fs) {
1412        error = sb->s_op->unfreeze_fs(sb);
1413        if (error) {
1414            printk(KERN_ERR
1415                "VFS:Filesystem thaw failed\n");
1416            up_write(&sb->s_umount);
1417            return error;
1418        }
1419    }
1420
1421out:
1422    sb->s_writers.frozen = SB_UNFROZEN;
1423    smp_wmb();
1424    wake_up(&sb->s_writers.wait_unfrozen);
1425    deactivate_locked_super(sb);
1426
1427    return 0;
1428}
1429EXPORT_SYMBOL(thaw_super);
1430

Archive Download this file



interactive