4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/device_cgroup.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/module.h>
18 #include <linux/blkpg.h>
19 #include <linux/magic.h>
20 #include <linux/buffer_head.h>
21 #include <linux/swap.h>
22 #include <linux/pagevec.h>
23 #include <linux/writeback.h>
24 #include <linux/mpage.h>
25 #include <linux/mount.h>
26 #include <linux/uio.h>
27 #include <linux/namei.h>
28 #include <linux/log2.h>
29 #include <linux/cleancache.h>
30 #include <linux/aio.h>
31 #include <asm/uaccess.h>
35 struct block_device bdev;
36 struct inode vfs_inode;
39 static const struct address_space_operations def_blk_aops;
41 static inline struct bdev_inode *BDEV_I(struct inode *inode)
43 return container_of(inode, struct bdev_inode, vfs_inode);
46 inline struct block_device *I_BDEV(struct inode *inode)
48 return &BDEV_I(inode)->bdev;
50 EXPORT_SYMBOL(I_BDEV);
53 * Move the inode from its current bdi to a new bdi. If the inode is dirty we
54 * need to move it onto the dirty list of @dst so that the inode is always on
57 static void bdev_inode_switch_bdi(struct inode *inode,
58 struct backing_dev_info *dst)
60 struct backing_dev_info *old = inode->i_data.backing_dev_info;
61 bool wakeup_bdi = false;
63 if (unlikely(dst == old)) /* deadlock avoidance */
65 bdi_lock_two(&old->wb, &dst->wb);
66 spin_lock(&inode->i_lock);
67 inode->i_data.backing_dev_info = dst;
68 if (inode->i_state & I_DIRTY) {
69 if (bdi_cap_writeback_dirty(dst) && !wb_has_dirty_io(&dst->wb))
71 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
73 spin_unlock(&inode->i_lock);
74 spin_unlock(&old->wb.list_lock);
75 spin_unlock(&dst->wb.list_lock);
78 bdi_wakeup_thread_delayed(dst);
81 /* Kill _all_ buffers and pagecache , dirty or not.. */
82 void kill_bdev(struct block_device *bdev)
84 struct address_space *mapping = bdev->bd_inode->i_mapping;
86 if (mapping->nrpages == 0)
90 truncate_inode_pages(mapping, 0);
92 EXPORT_SYMBOL(kill_bdev);
94 /* Invalidate clean unused buffers and pagecache. */
95 void invalidate_bdev(struct block_device *bdev)
97 struct address_space *mapping = bdev->bd_inode->i_mapping;
99 if (mapping->nrpages == 0)
102 invalidate_bh_lrus();
103 lru_add_drain_all(); /* make sure all lru add caches are flushed */
104 invalidate_mapping_pages(mapping, 0, -1);
105 /* 99% of the time, we don't need to flush the cleancache on the bdev.
106 * But, for the strange corners, lets be cautious
108 cleancache_invalidate_inode(mapping);
110 EXPORT_SYMBOL(invalidate_bdev);
112 int set_blocksize(struct block_device *bdev, int size)
114 /* Size must be a power of two, and between 512 and PAGE_SIZE */
115 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
118 /* Size cannot be smaller than the size supported by the device */
119 if (size < bdev_logical_block_size(bdev))
122 /* Don't change the size if it is same as current */
123 if (bdev->bd_block_size != size) {
125 bdev->bd_block_size = size;
126 bdev->bd_inode->i_blkbits = blksize_bits(size);
132 EXPORT_SYMBOL(set_blocksize);
134 int sb_set_blocksize(struct super_block *sb, int size)
136 if (set_blocksize(sb->s_bdev, size))
138 /* If we get here, we know size is power of two
139 * and it's value is between 512 and PAGE_SIZE */
140 sb->s_blocksize = size;
141 sb->s_blocksize_bits = blksize_bits(size);
142 return sb->s_blocksize;
145 EXPORT_SYMBOL(sb_set_blocksize);
147 int sb_min_blocksize(struct super_block *sb, int size)
149 int minsize = bdev_logical_block_size(sb->s_bdev);
152 return sb_set_blocksize(sb, size);
155 EXPORT_SYMBOL(sb_min_blocksize);
158 blkdev_get_block(struct inode *inode, sector_t iblock,
159 struct buffer_head *bh, int create)
161 bh->b_bdev = I_BDEV(inode);
162 bh->b_blocknr = iblock;
163 set_buffer_mapped(bh);
168 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
169 loff_t offset, unsigned long nr_segs)
171 struct file *file = iocb->ki_filp;
172 struct inode *inode = file->f_mapping->host;
174 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
175 nr_segs, blkdev_get_block, NULL, NULL, 0);
178 int __sync_blockdev(struct block_device *bdev, int wait)
183 return filemap_flush(bdev->bd_inode->i_mapping);
184 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
188 * Write out and wait upon all the dirty data associated with a block
189 * device via its mapping. Does not take the superblock lock.
191 int sync_blockdev(struct block_device *bdev)
193 return __sync_blockdev(bdev, 1);
195 EXPORT_SYMBOL(sync_blockdev);
198 * Write out and wait upon all dirty data associated with this
199 * device. Filesystem data as well as the underlying block
200 * device. Takes the superblock lock.
202 int fsync_bdev(struct block_device *bdev)
204 struct super_block *sb = get_super(bdev);
206 int res = sync_filesystem(sb);
210 return sync_blockdev(bdev);
212 EXPORT_SYMBOL(fsync_bdev);
215 * freeze_bdev -- lock a filesystem and force it into a consistent state
216 * @bdev: blockdevice to lock
218 * If a superblock is found on this device, we take the s_umount semaphore
219 * on it to make sure nobody unmounts until the snapshot creation is done.
220 * The reference counter (bd_fsfreeze_count) guarantees that only the last
221 * unfreeze process can unfreeze the frozen filesystem actually when multiple
222 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
223 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
226 struct super_block *freeze_bdev(struct block_device *bdev)
228 struct super_block *sb;
231 mutex_lock(&bdev->bd_fsfreeze_mutex);
232 if (++bdev->bd_fsfreeze_count > 1) {
234 * We don't even need to grab a reference - the first call
235 * to freeze_bdev grab an active reference and only the last
236 * thaw_bdev drops it.
238 sb = get_super(bdev);
240 mutex_unlock(&bdev->bd_fsfreeze_mutex);
244 sb = get_active_super(bdev);
247 error = freeze_super(sb);
249 deactivate_super(sb);
250 bdev->bd_fsfreeze_count--;
251 mutex_unlock(&bdev->bd_fsfreeze_mutex);
252 return ERR_PTR(error);
254 deactivate_super(sb);
257 mutex_unlock(&bdev->bd_fsfreeze_mutex);
258 return sb; /* thaw_bdev releases s->s_umount */
260 EXPORT_SYMBOL(freeze_bdev);
263 * thaw_bdev -- unlock filesystem
264 * @bdev: blockdevice to unlock
265 * @sb: associated superblock
267 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
269 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
273 mutex_lock(&bdev->bd_fsfreeze_mutex);
274 if (!bdev->bd_fsfreeze_count)
278 if (--bdev->bd_fsfreeze_count > 0)
284 error = thaw_super(sb);
286 bdev->bd_fsfreeze_count++;
287 mutex_unlock(&bdev->bd_fsfreeze_mutex);
291 mutex_unlock(&bdev->bd_fsfreeze_mutex);
294 EXPORT_SYMBOL(thaw_bdev);
296 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
298 return block_write_full_page(page, blkdev_get_block, wbc);
301 static int blkdev_readpage(struct file * file, struct page * page)
303 return block_read_full_page(page, blkdev_get_block);
306 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
307 loff_t pos, unsigned len, unsigned flags,
308 struct page **pagep, void **fsdata)
310 return block_write_begin(mapping, pos, len, flags, pagep,
314 static int blkdev_write_end(struct file *file, struct address_space *mapping,
315 loff_t pos, unsigned len, unsigned copied,
316 struct page *page, void *fsdata)
319 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
322 page_cache_release(page);
329 * for a block special file file_inode(file)->i_size is zero
330 * so we compute the size by hand (just as in block_read/write above)
332 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
334 struct inode *bd_inode = file->f_mapping->host;
337 mutex_lock(&bd_inode->i_mutex);
338 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
339 mutex_unlock(&bd_inode->i_mutex);
343 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
345 struct inode *bd_inode = filp->f_mapping->host;
346 struct block_device *bdev = I_BDEV(bd_inode);
349 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
354 * There is no need to serialise calls to blkdev_issue_flush with
355 * i_mutex and doing so causes performance issues with concurrent
356 * O_SYNC writers to a block device.
358 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
359 if (error == -EOPNOTSUPP)
364 EXPORT_SYMBOL(blkdev_fsync);
370 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
371 static struct kmem_cache * bdev_cachep __read_mostly;
373 static struct inode *bdev_alloc_inode(struct super_block *sb)
375 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
378 return &ei->vfs_inode;
381 static void bdev_i_callback(struct rcu_head *head)
383 struct inode *inode = container_of(head, struct inode, i_rcu);
384 struct bdev_inode *bdi = BDEV_I(inode);
386 kmem_cache_free(bdev_cachep, bdi);
389 static void bdev_destroy_inode(struct inode *inode)
391 call_rcu(&inode->i_rcu, bdev_i_callback);
394 static void init_once(void *foo)
396 struct bdev_inode *ei = (struct bdev_inode *) foo;
397 struct block_device *bdev = &ei->bdev;
399 memset(bdev, 0, sizeof(*bdev));
400 mutex_init(&bdev->bd_mutex);
401 INIT_LIST_HEAD(&bdev->bd_inodes);
402 INIT_LIST_HEAD(&bdev->bd_list);
404 INIT_LIST_HEAD(&bdev->bd_holder_disks);
406 inode_init_once(&ei->vfs_inode);
407 /* Initialize mutex for freeze. */
408 mutex_init(&bdev->bd_fsfreeze_mutex);
411 static inline void __bd_forget(struct inode *inode)
413 list_del_init(&inode->i_devices);
414 inode->i_bdev = NULL;
415 inode->i_mapping = &inode->i_data;
418 static void bdev_evict_inode(struct inode *inode)
420 struct block_device *bdev = &BDEV_I(inode)->bdev;
422 truncate_inode_pages(&inode->i_data, 0);
423 invalidate_inode_buffers(inode); /* is it needed here? */
425 spin_lock(&bdev_lock);
426 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
427 __bd_forget(list_entry(p, struct inode, i_devices));
429 list_del_init(&bdev->bd_list);
430 spin_unlock(&bdev_lock);
433 static const struct super_operations bdev_sops = {
434 .statfs = simple_statfs,
435 .alloc_inode = bdev_alloc_inode,
436 .destroy_inode = bdev_destroy_inode,
437 .drop_inode = generic_delete_inode,
438 .evict_inode = bdev_evict_inode,
441 static struct dentry *bd_mount(struct file_system_type *fs_type,
442 int flags, const char *dev_name, void *data)
444 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
447 static struct file_system_type bd_type = {
450 .kill_sb = kill_anon_super,
453 static struct super_block *blockdev_superblock __read_mostly;
455 void __init bdev_cache_init(void)
458 static struct vfsmount *bd_mnt;
460 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
461 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
462 SLAB_MEM_SPREAD|SLAB_PANIC),
464 err = register_filesystem(&bd_type);
466 panic("Cannot register bdev pseudo-fs");
467 bd_mnt = kern_mount(&bd_type);
469 panic("Cannot create bdev pseudo-fs");
470 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
474 * Most likely _very_ bad one - but then it's hardly critical for small
475 * /dev and can be fixed when somebody will need really large one.
476 * Keep in mind that it will be fed through icache hash function too.
478 static inline unsigned long hash(dev_t dev)
480 return MAJOR(dev)+MINOR(dev);
483 static int bdev_test(struct inode *inode, void *data)
485 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
488 static int bdev_set(struct inode *inode, void *data)
490 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
494 static LIST_HEAD(all_bdevs);
496 struct block_device *bdget(dev_t dev)
498 struct block_device *bdev;
501 inode = iget5_locked(blockdev_superblock, hash(dev),
502 bdev_test, bdev_set, &dev);
507 bdev = &BDEV_I(inode)->bdev;
509 if (inode->i_state & I_NEW) {
510 bdev->bd_contains = NULL;
511 bdev->bd_super = NULL;
512 bdev->bd_inode = inode;
513 bdev->bd_block_size = (1 << inode->i_blkbits);
514 bdev->bd_part_count = 0;
515 bdev->bd_invalidated = 0;
516 inode->i_mode = S_IFBLK;
518 inode->i_bdev = bdev;
519 inode->i_data.a_ops = &def_blk_aops;
520 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
521 inode->i_data.backing_dev_info = &default_backing_dev_info;
522 spin_lock(&bdev_lock);
523 list_add(&bdev->bd_list, &all_bdevs);
524 spin_unlock(&bdev_lock);
525 unlock_new_inode(inode);
530 EXPORT_SYMBOL(bdget);
533 * bdgrab -- Grab a reference to an already referenced block device
534 * @bdev: Block device to grab a reference to.
536 struct block_device *bdgrab(struct block_device *bdev)
538 ihold(bdev->bd_inode);
541 EXPORT_SYMBOL(bdgrab);
543 long nr_blockdev_pages(void)
545 struct block_device *bdev;
547 spin_lock(&bdev_lock);
548 list_for_each_entry(bdev, &all_bdevs, bd_list) {
549 ret += bdev->bd_inode->i_mapping->nrpages;
551 spin_unlock(&bdev_lock);
555 void bdput(struct block_device *bdev)
557 iput(bdev->bd_inode);
560 EXPORT_SYMBOL(bdput);
562 static struct block_device *bd_acquire(struct inode *inode)
564 struct block_device *bdev;
566 spin_lock(&bdev_lock);
567 bdev = inode->i_bdev;
569 ihold(bdev->bd_inode);
570 spin_unlock(&bdev_lock);
573 spin_unlock(&bdev_lock);
575 bdev = bdget(inode->i_rdev);
577 spin_lock(&bdev_lock);
578 if (!inode->i_bdev) {
580 * We take an additional reference to bd_inode,
581 * and it's released in clear_inode() of inode.
582 * So, we can access it via ->i_mapping always
585 ihold(bdev->bd_inode);
586 inode->i_bdev = bdev;
587 inode->i_mapping = bdev->bd_inode->i_mapping;
588 list_add(&inode->i_devices, &bdev->bd_inodes);
590 spin_unlock(&bdev_lock);
595 static inline int sb_is_blkdev_sb(struct super_block *sb)
597 return sb == blockdev_superblock;
600 /* Call when you free inode */
602 void bd_forget(struct inode *inode)
604 struct block_device *bdev = NULL;
606 spin_lock(&bdev_lock);
607 if (!sb_is_blkdev_sb(inode->i_sb))
608 bdev = inode->i_bdev;
610 spin_unlock(&bdev_lock);
613 iput(bdev->bd_inode);
617 * bd_may_claim - test whether a block device can be claimed
618 * @bdev: block device of interest
619 * @whole: whole block device containing @bdev, may equal @bdev
620 * @holder: holder trying to claim @bdev
622 * Test whether @bdev can be claimed by @holder.
625 * spin_lock(&bdev_lock).
628 * %true if @bdev can be claimed, %false otherwise.
630 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
633 if (bdev->bd_holder == holder)
634 return true; /* already a holder */
635 else if (bdev->bd_holder != NULL)
636 return false; /* held by someone else */
637 else if (bdev->bd_contains == bdev)
638 return true; /* is a whole device which isn't held */
640 else if (whole->bd_holder == bd_may_claim)
641 return true; /* is a partition of a device that is being partitioned */
642 else if (whole->bd_holder != NULL)
643 return false; /* is a partition of a held device */
645 return true; /* is a partition of an un-held device */
649 * bd_prepare_to_claim - prepare to claim a block device
650 * @bdev: block device of interest
651 * @whole: the whole device containing @bdev, may equal @bdev
652 * @holder: holder trying to claim @bdev
654 * Prepare to claim @bdev. This function fails if @bdev is already
655 * claimed by another holder and waits if another claiming is in
656 * progress. This function doesn't actually claim. On successful
657 * return, the caller has ownership of bd_claiming and bd_holder[s].
660 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
664 * 0 if @bdev can be claimed, -EBUSY otherwise.
666 static int bd_prepare_to_claim(struct block_device *bdev,
667 struct block_device *whole, void *holder)
670 /* if someone else claimed, fail */
671 if (!bd_may_claim(bdev, whole, holder))
674 /* if claiming is already in progress, wait for it to finish */
675 if (whole->bd_claiming) {
676 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
679 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
680 spin_unlock(&bdev_lock);
682 finish_wait(wq, &wait);
683 spin_lock(&bdev_lock);
692 * bd_start_claiming - start claiming a block device
693 * @bdev: block device of interest
694 * @holder: holder trying to claim @bdev
696 * @bdev is about to be opened exclusively. Check @bdev can be opened
697 * exclusively and mark that an exclusive open is in progress. Each
698 * successful call to this function must be matched with a call to
699 * either bd_finish_claiming() or bd_abort_claiming() (which do not
702 * This function is used to gain exclusive access to the block device
703 * without actually causing other exclusive open attempts to fail. It
704 * should be used when the open sequence itself requires exclusive
705 * access but may subsequently fail.
711 * Pointer to the block device containing @bdev on success, ERR_PTR()
714 static struct block_device *bd_start_claiming(struct block_device *bdev,
717 struct gendisk *disk;
718 struct block_device *whole;
724 * @bdev might not have been initialized properly yet, look up
725 * and grab the outer block device the hard way.
727 disk = get_gendisk(bdev->bd_dev, &partno);
729 return ERR_PTR(-ENXIO);
732 * Normally, @bdev should equal what's returned from bdget_disk()
733 * if partno is 0; however, some drivers (floppy) use multiple
734 * bdev's for the same physical device and @bdev may be one of the
735 * aliases. Keep @bdev if partno is 0. This means claimer
736 * tracking is broken for those devices but it has always been that
740 whole = bdget_disk(disk, 0);
742 whole = bdgrab(bdev);
744 module_put(disk->fops->owner);
747 return ERR_PTR(-ENOMEM);
749 /* prepare to claim, if successful, mark claiming in progress */
750 spin_lock(&bdev_lock);
752 err = bd_prepare_to_claim(bdev, whole, holder);
754 whole->bd_claiming = holder;
755 spin_unlock(&bdev_lock);
758 spin_unlock(&bdev_lock);
765 struct bd_holder_disk {
766 struct list_head list;
767 struct gendisk *disk;
771 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
772 struct gendisk *disk)
774 struct bd_holder_disk *holder;
776 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
777 if (holder->disk == disk)
782 static int add_symlink(struct kobject *from, struct kobject *to)
784 return sysfs_create_link(from, to, kobject_name(to));
787 static void del_symlink(struct kobject *from, struct kobject *to)
789 sysfs_remove_link(from, kobject_name(to));
793 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
794 * @bdev: the claimed slave bdev
795 * @disk: the holding disk
797 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
799 * This functions creates the following sysfs symlinks.
801 * - from "slaves" directory of the holder @disk to the claimed @bdev
802 * - from "holders" directory of the @bdev to the holder @disk
804 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
805 * passed to bd_link_disk_holder(), then:
807 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
808 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
810 * The caller must have claimed @bdev before calling this function and
811 * ensure that both @bdev and @disk are valid during the creation and
812 * lifetime of these symlinks.
818 * 0 on success, -errno on failure.
820 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
822 struct bd_holder_disk *holder;
825 mutex_lock(&bdev->bd_mutex);
827 WARN_ON_ONCE(!bdev->bd_holder);
829 /* FIXME: remove the following once add_disk() handles errors */
830 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
833 holder = bd_find_holder_disk(bdev, disk);
839 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
845 INIT_LIST_HEAD(&holder->list);
849 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
853 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
857 * bdev could be deleted beneath us which would implicitly destroy
858 * the holder directory. Hold on to it.
860 kobject_get(bdev->bd_part->holder_dir);
862 list_add(&holder->list, &bdev->bd_holder_disks);
866 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
870 mutex_unlock(&bdev->bd_mutex);
873 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
876 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
877 * @bdev: the calimed slave bdev
878 * @disk: the holding disk
880 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
885 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
887 struct bd_holder_disk *holder;
889 mutex_lock(&bdev->bd_mutex);
891 holder = bd_find_holder_disk(bdev, disk);
893 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
894 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
895 del_symlink(bdev->bd_part->holder_dir,
896 &disk_to_dev(disk)->kobj);
897 kobject_put(bdev->bd_part->holder_dir);
898 list_del_init(&holder->list);
902 mutex_unlock(&bdev->bd_mutex);
904 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
908 * flush_disk - invalidates all buffer-cache entries on a disk
910 * @bdev: struct block device to be flushed
911 * @kill_dirty: flag to guide handling of dirty inodes
913 * Invalidates all buffer-cache entries on a disk. It should be called
914 * when a disk has been changed -- either by a media change or online
917 static void flush_disk(struct block_device *bdev, bool kill_dirty)
919 if (__invalidate_device(bdev, kill_dirty)) {
920 char name[BDEVNAME_SIZE] = "";
923 disk_name(bdev->bd_disk, 0, name);
924 printk(KERN_WARNING "VFS: busy inodes on changed media or "
925 "resized disk %s\n", name);
930 if (disk_part_scan_enabled(bdev->bd_disk))
931 bdev->bd_invalidated = 1;
935 * check_disk_size_change - checks for disk size change and adjusts bdev size.
936 * @disk: struct gendisk to check
937 * @bdev: struct bdev to adjust.
939 * This routine checks to see if the bdev size does not match the disk size
940 * and adjusts it if it differs.
942 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
944 loff_t disk_size, bdev_size;
946 disk_size = (loff_t)get_capacity(disk) << 9;
947 bdev_size = i_size_read(bdev->bd_inode);
948 if (disk_size != bdev_size) {
949 char name[BDEVNAME_SIZE];
951 disk_name(disk, 0, name);
953 "%s: detected capacity change from %lld to %lld\n",
954 name, bdev_size, disk_size);
955 i_size_write(bdev->bd_inode, disk_size);
956 flush_disk(bdev, false);
959 EXPORT_SYMBOL(check_disk_size_change);
962 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
963 * @disk: struct gendisk to be revalidated
965 * This routine is a wrapper for lower-level driver's revalidate_disk
966 * call-backs. It is used to do common pre and post operations needed
967 * for all revalidate_disk operations.
969 int revalidate_disk(struct gendisk *disk)
971 struct block_device *bdev;
974 if (disk->fops->revalidate_disk)
975 ret = disk->fops->revalidate_disk(disk);
977 bdev = bdget_disk(disk, 0);
981 mutex_lock(&bdev->bd_mutex);
982 check_disk_size_change(disk, bdev);
983 bdev->bd_invalidated = 0;
984 mutex_unlock(&bdev->bd_mutex);
988 EXPORT_SYMBOL(revalidate_disk);
991 * This routine checks whether a removable media has been changed,
992 * and invalidates all buffer-cache-entries in that case. This
993 * is a relatively slow routine, so we have to try to minimize using
994 * it. Thus it is called only upon a 'mount' or 'open'. This
995 * is the best way of combining speed and utility, I think.
996 * People changing diskettes in the middle of an operation deserve
999 int check_disk_change(struct block_device *bdev)
1001 struct gendisk *disk = bdev->bd_disk;
1002 const struct block_device_operations *bdops = disk->fops;
1003 unsigned int events;
1005 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1006 DISK_EVENT_EJECT_REQUEST);
1007 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1010 flush_disk(bdev, true);
1011 if (bdops->revalidate_disk)
1012 bdops->revalidate_disk(bdev->bd_disk);
1016 EXPORT_SYMBOL(check_disk_change);
1018 void bd_set_size(struct block_device *bdev, loff_t size)
1020 unsigned bsize = bdev_logical_block_size(bdev);
1022 mutex_lock(&bdev->bd_inode->i_mutex);
1023 i_size_write(bdev->bd_inode, size);
1024 mutex_unlock(&bdev->bd_inode->i_mutex);
1025 while (bsize < PAGE_CACHE_SIZE) {
1030 bdev->bd_block_size = bsize;
1031 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1033 EXPORT_SYMBOL(bd_set_size);
1035 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1040 * mutex_lock(part->bd_mutex)
1041 * mutex_lock_nested(whole->bd_mutex, 1)
1044 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1046 struct gendisk *disk;
1047 struct module *owner;
1052 if (mode & FMODE_READ)
1054 if (mode & FMODE_WRITE)
1057 * hooks: /n/, see "layering violations".
1060 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1070 disk = get_gendisk(bdev->bd_dev, &partno);
1073 owner = disk->fops->owner;
1075 disk_block_events(disk);
1076 mutex_lock_nested(&bdev->bd_mutex, for_part);
1077 if (!bdev->bd_openers) {
1078 bdev->bd_disk = disk;
1079 bdev->bd_queue = disk->queue;
1080 bdev->bd_contains = bdev;
1082 struct backing_dev_info *bdi;
1085 bdev->bd_part = disk_get_part(disk, partno);
1090 if (disk->fops->open) {
1091 ret = disk->fops->open(bdev, mode);
1092 if (ret == -ERESTARTSYS) {
1093 /* Lost a race with 'disk' being
1094 * deleted, try again.
1097 disk_put_part(bdev->bd_part);
1098 bdev->bd_part = NULL;
1099 bdev->bd_disk = NULL;
1100 bdev->bd_queue = NULL;
1101 mutex_unlock(&bdev->bd_mutex);
1102 disk_unblock_events(disk);
1110 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1111 bdi = blk_get_backing_dev_info(bdev);
1113 bdi = &default_backing_dev_info;
1114 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1118 * If the device is invalidated, rescan partition
1119 * if open succeeded or failed with -ENOMEDIUM.
1120 * The latter is necessary to prevent ghost
1121 * partitions on a removed medium.
1123 if (bdev->bd_invalidated) {
1125 rescan_partitions(disk, bdev);
1126 else if (ret == -ENOMEDIUM)
1127 invalidate_partitions(disk, bdev);
1132 struct block_device *whole;
1133 whole = bdget_disk(disk, 0);
1138 ret = __blkdev_get(whole, mode, 1);
1141 bdev->bd_contains = whole;
1142 bdev_inode_switch_bdi(bdev->bd_inode,
1143 whole->bd_inode->i_data.backing_dev_info);
1144 bdev->bd_part = disk_get_part(disk, partno);
1145 if (!(disk->flags & GENHD_FL_UP) ||
1146 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1150 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1153 if (bdev->bd_contains == bdev) {
1155 if (bdev->bd_disk->fops->open)
1156 ret = bdev->bd_disk->fops->open(bdev, mode);
1157 /* the same as first opener case, read comment there */
1158 if (bdev->bd_invalidated) {
1160 rescan_partitions(bdev->bd_disk, bdev);
1161 else if (ret == -ENOMEDIUM)
1162 invalidate_partitions(bdev->bd_disk, bdev);
1165 goto out_unlock_bdev;
1167 /* only one opener holds refs to the module and disk */
1173 bdev->bd_part_count++;
1174 mutex_unlock(&bdev->bd_mutex);
1175 disk_unblock_events(disk);
1179 disk_put_part(bdev->bd_part);
1180 bdev->bd_disk = NULL;
1181 bdev->bd_part = NULL;
1182 bdev->bd_queue = NULL;
1183 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1184 if (bdev != bdev->bd_contains)
1185 __blkdev_put(bdev->bd_contains, mode, 1);
1186 bdev->bd_contains = NULL;
1188 mutex_unlock(&bdev->bd_mutex);
1189 disk_unblock_events(disk);
1199 * blkdev_get - open a block device
1200 * @bdev: block_device to open
1201 * @mode: FMODE_* mask
1202 * @holder: exclusive holder identifier
1204 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1205 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1206 * @holder is invalid. Exclusive opens may nest for the same @holder.
1208 * On success, the reference count of @bdev is unchanged. On failure,
1215 * 0 on success, -errno on failure.
1217 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1219 struct block_device *whole = NULL;
1222 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1224 if ((mode & FMODE_EXCL) && holder) {
1225 whole = bd_start_claiming(bdev, holder);
1226 if (IS_ERR(whole)) {
1228 return PTR_ERR(whole);
1232 res = __blkdev_get(bdev, mode, 0);
1235 struct gendisk *disk = whole->bd_disk;
1237 /* finish claiming */
1238 mutex_lock(&bdev->bd_mutex);
1239 spin_lock(&bdev_lock);
1242 BUG_ON(!bd_may_claim(bdev, whole, holder));
1244 * Note that for a whole device bd_holders
1245 * will be incremented twice, and bd_holder
1246 * will be set to bd_may_claim before being
1249 whole->bd_holders++;
1250 whole->bd_holder = bd_may_claim;
1252 bdev->bd_holder = holder;
1255 /* tell others that we're done */
1256 BUG_ON(whole->bd_claiming != holder);
1257 whole->bd_claiming = NULL;
1258 wake_up_bit(&whole->bd_claiming, 0);
1260 spin_unlock(&bdev_lock);
1263 * Block event polling for write claims if requested. Any
1264 * write holder makes the write_holder state stick until
1265 * all are released. This is good enough and tracking
1266 * individual writeable reference is too fragile given the
1267 * way @mode is used in blkdev_get/put().
1269 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1270 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1271 bdev->bd_write_holder = true;
1272 disk_block_events(disk);
1275 mutex_unlock(&bdev->bd_mutex);
1281 EXPORT_SYMBOL(blkdev_get);
1284 * blkdev_get_by_path - open a block device by name
1285 * @path: path to the block device to open
1286 * @mode: FMODE_* mask
1287 * @holder: exclusive holder identifier
1289 * Open the blockdevice described by the device file at @path. @mode
1290 * and @holder are identical to blkdev_get().
1292 * On success, the returned block_device has reference count of one.
1298 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1300 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1303 struct block_device *bdev;
1306 bdev = lookup_bdev(path);
1310 err = blkdev_get(bdev, mode, holder);
1312 return ERR_PTR(err);
1314 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1315 blkdev_put(bdev, mode);
1316 return ERR_PTR(-EACCES);
1321 EXPORT_SYMBOL(blkdev_get_by_path);
1324 * blkdev_get_by_dev - open a block device by device number
1325 * @dev: device number of block device to open
1326 * @mode: FMODE_* mask
1327 * @holder: exclusive holder identifier
1329 * Open the blockdevice described by device number @dev. @mode and
1330 * @holder are identical to blkdev_get().
1332 * Use it ONLY if you really do not have anything better - i.e. when
1333 * you are behind a truly sucky interface and all you are given is a
1334 * device number. _Never_ to be used for internal purposes. If you
1335 * ever need it - reconsider your API.
1337 * On success, the returned block_device has reference count of one.
1343 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1345 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1347 struct block_device *bdev;
1352 return ERR_PTR(-ENOMEM);
1354 err = blkdev_get(bdev, mode, holder);
1356 return ERR_PTR(err);
1360 EXPORT_SYMBOL(blkdev_get_by_dev);
1362 static int blkdev_open(struct inode * inode, struct file * filp)
1364 struct block_device *bdev;
1367 * Preserve backwards compatibility and allow large file access
1368 * even if userspace doesn't ask for it explicitly. Some mkfs
1369 * binary needs it. We might want to drop this workaround
1370 * during an unstable branch.
1372 filp->f_flags |= O_LARGEFILE;
1374 if (filp->f_flags & O_NDELAY)
1375 filp->f_mode |= FMODE_NDELAY;
1376 if (filp->f_flags & O_EXCL)
1377 filp->f_mode |= FMODE_EXCL;
1378 if ((filp->f_flags & O_ACCMODE) == 3)
1379 filp->f_mode |= FMODE_WRITE_IOCTL;
1381 bdev = bd_acquire(inode);
1385 filp->f_mapping = bdev->bd_inode->i_mapping;
1387 return blkdev_get(bdev, filp->f_mode, filp);
1390 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1392 struct gendisk *disk = bdev->bd_disk;
1393 struct block_device *victim = NULL;
1395 mutex_lock_nested(&bdev->bd_mutex, for_part);
1397 bdev->bd_part_count--;
1399 if (!--bdev->bd_openers) {
1400 WARN_ON_ONCE(bdev->bd_holders);
1401 sync_blockdev(bdev);
1403 /* ->release can cause the old bdi to disappear,
1404 * so must switch it out first
1406 bdev_inode_switch_bdi(bdev->bd_inode,
1407 &default_backing_dev_info);
1409 if (bdev->bd_contains == bdev) {
1410 if (disk->fops->release)
1411 disk->fops->release(disk, mode);
1413 if (!bdev->bd_openers) {
1414 struct module *owner = disk->fops->owner;
1416 disk_put_part(bdev->bd_part);
1417 bdev->bd_part = NULL;
1418 bdev->bd_disk = NULL;
1419 if (bdev != bdev->bd_contains)
1420 victim = bdev->bd_contains;
1421 bdev->bd_contains = NULL;
1426 mutex_unlock(&bdev->bd_mutex);
1429 __blkdev_put(victim, mode, 1);
1432 void blkdev_put(struct block_device *bdev, fmode_t mode)
1434 mutex_lock(&bdev->bd_mutex);
1436 if (mode & FMODE_EXCL) {
1440 * Release a claim on the device. The holder fields
1441 * are protected with bdev_lock. bd_mutex is to
1442 * synchronize disk_holder unlinking.
1444 spin_lock(&bdev_lock);
1446 WARN_ON_ONCE(--bdev->bd_holders < 0);
1447 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1449 /* bd_contains might point to self, check in a separate step */
1450 if ((bdev_free = !bdev->bd_holders))
1451 bdev->bd_holder = NULL;
1452 if (!bdev->bd_contains->bd_holders)
1453 bdev->bd_contains->bd_holder = NULL;
1455 spin_unlock(&bdev_lock);
1458 * If this was the last claim, remove holder link and
1459 * unblock evpoll if it was a write holder.
1461 if (bdev_free && bdev->bd_write_holder) {
1462 disk_unblock_events(bdev->bd_disk);
1463 bdev->bd_write_holder = false;
1468 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1469 * event. This is to ensure detection of media removal commanded
1470 * from userland - e.g. eject(1).
1472 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1474 mutex_unlock(&bdev->bd_mutex);
1476 __blkdev_put(bdev, mode, 0);
1478 EXPORT_SYMBOL(blkdev_put);
1480 static int blkdev_close(struct inode * inode, struct file * filp)
1482 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1483 blkdev_put(bdev, filp->f_mode);
1487 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1489 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1490 fmode_t mode = file->f_mode;
1493 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1494 * to updated it before every ioctl.
1496 if (file->f_flags & O_NDELAY)
1497 mode |= FMODE_NDELAY;
1499 mode &= ~FMODE_NDELAY;
1501 return blkdev_ioctl(bdev, mode, cmd, arg);
1505 * Write data to the block device. Only intended for the block device itself
1506 * and the raw driver which basically is a fake block device.
1508 * Does not take i_mutex for the write and thus is not for general purpose
1511 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1512 unsigned long nr_segs, loff_t pos)
1514 struct file *file = iocb->ki_filp;
1515 struct blk_plug plug;
1518 BUG_ON(iocb->ki_pos != pos);
1520 blk_start_plug(&plug);
1521 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1525 err = generic_write_sync(file, pos, ret);
1526 if (err < 0 && ret > 0)
1529 blk_finish_plug(&plug);
1532 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1534 static ssize_t blkdev_aio_read(struct kiocb *iocb, const struct iovec *iov,
1535 unsigned long nr_segs, loff_t pos)
1537 struct file *file = iocb->ki_filp;
1538 struct inode *bd_inode = file->f_mapping->host;
1539 loff_t size = i_size_read(bd_inode);
1545 if (size < iocb->ki_left)
1546 nr_segs = iov_shorten((struct iovec *)iov, nr_segs, size);
1547 return generic_file_aio_read(iocb, iov, nr_segs, pos);
1551 * Try to release a page associated with block device when the system
1552 * is under memory pressure.
1554 static int blkdev_releasepage(struct page *page, gfp_t wait)
1556 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1558 if (super && super->s_op->bdev_try_to_free_page)
1559 return super->s_op->bdev_try_to_free_page(super, page, wait);
1561 return try_to_free_buffers(page);
1564 static const struct address_space_operations def_blk_aops = {
1565 .readpage = blkdev_readpage,
1566 .writepage = blkdev_writepage,
1567 .write_begin = blkdev_write_begin,
1568 .write_end = blkdev_write_end,
1569 .writepages = generic_writepages,
1570 .releasepage = blkdev_releasepage,
1571 .direct_IO = blkdev_direct_IO,
1572 .is_dirty_writeback = buffer_check_dirty_writeback,
1575 const struct file_operations def_blk_fops = {
1576 .open = blkdev_open,
1577 .release = blkdev_close,
1578 .llseek = block_llseek,
1579 .read = do_sync_read,
1580 .write = do_sync_write,
1581 .aio_read = blkdev_aio_read,
1582 .aio_write = blkdev_aio_write,
1583 .mmap = generic_file_mmap,
1584 .fsync = blkdev_fsync,
1585 .unlocked_ioctl = block_ioctl,
1586 #ifdef CONFIG_COMPAT
1587 .compat_ioctl = compat_blkdev_ioctl,
1589 .splice_read = generic_file_splice_read,
1590 .splice_write = generic_file_splice_write,
1593 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1596 mm_segment_t old_fs = get_fs();
1598 res = blkdev_ioctl(bdev, 0, cmd, arg);
1603 EXPORT_SYMBOL(ioctl_by_bdev);
1606 * lookup_bdev - lookup a struct block_device by name
1607 * @pathname: special file representing the block device
1609 * Get a reference to the blockdevice at @pathname in the current
1610 * namespace if possible and return it. Return ERR_PTR(error)
1613 struct block_device *lookup_bdev(const char *pathname)
1615 struct block_device *bdev;
1616 struct inode *inode;
1620 if (!pathname || !*pathname)
1621 return ERR_PTR(-EINVAL);
1623 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1625 return ERR_PTR(error);
1627 inode = path.dentry->d_inode;
1629 if (!S_ISBLK(inode->i_mode))
1632 if (path.mnt->mnt_flags & MNT_NODEV)
1635 bdev = bd_acquire(inode);
1642 bdev = ERR_PTR(error);
1645 EXPORT_SYMBOL(lookup_bdev);
1647 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1649 struct super_block *sb = get_super(bdev);
1654 * no need to lock the super, get_super holds the
1655 * read mutex so the filesystem cannot go away
1656 * under us (->put_super runs with the write lock
1659 shrink_dcache_sb(sb);
1660 res = invalidate_inodes(sb, kill_dirty);
1663 invalidate_bdev(bdev);
1666 EXPORT_SYMBOL(__invalidate_device);
1668 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1670 struct inode *inode, *old_inode = NULL;
1672 spin_lock(&inode_sb_list_lock);
1673 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1674 struct address_space *mapping = inode->i_mapping;
1676 spin_lock(&inode->i_lock);
1677 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1678 mapping->nrpages == 0) {
1679 spin_unlock(&inode->i_lock);
1683 spin_unlock(&inode->i_lock);
1684 spin_unlock(&inode_sb_list_lock);
1686 * We hold a reference to 'inode' so it couldn't have been
1687 * removed from s_inodes list while we dropped the
1688 * inode_sb_list_lock. We cannot iput the inode now as we can
1689 * be holding the last reference and we cannot iput it under
1690 * inode_sb_list_lock. So we keep the reference and iput it
1696 func(I_BDEV(inode), arg);
1698 spin_lock(&inode_sb_list_lock);
1700 spin_unlock(&inode_sb_list_lock);