1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 1991, 1992 Linus Torvalds
4 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Copyright (C) 2016 - 2020 Christoph Hellwig
8 #include <linux/init.h>
10 #include <linux/slab.h>
11 #include <linux/kmod.h>
12 #include <linux/major.h>
13 #include <linux/device_cgroup.h>
14 #include <linux/blkdev.h>
15 #include <linux/blk-integrity.h>
16 #include <linux/backing-dev.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/writeback.h>
23 #include <linux/mount.h>
24 #include <linux/pseudo_fs.h>
25 #include <linux/uio.h>
26 #include <linux/namei.h>
27 #include <linux/part_stat.h>
28 #include <linux/uaccess.h>
29 #include <linux/stat.h>
30 #include "../fs/internal.h"
34 struct block_device bdev;
35 struct inode vfs_inode;
38 static inline struct bdev_inode *BDEV_I(struct inode *inode)
40 return container_of(inode, struct bdev_inode, vfs_inode);
43 struct block_device *I_BDEV(struct inode *inode)
45 return &BDEV_I(inode)->bdev;
47 EXPORT_SYMBOL(I_BDEV);
49 static void bdev_write_inode(struct block_device *bdev)
51 struct inode *inode = bdev->bd_inode;
54 spin_lock(&inode->i_lock);
55 while (inode->i_state & I_DIRTY) {
56 spin_unlock(&inode->i_lock);
57 ret = write_inode_now(inode, true);
60 "VFS: Dirty inode writeback failed for block device %pg (err=%d).\n",
62 spin_lock(&inode->i_lock);
64 spin_unlock(&inode->i_lock);
67 /* Kill _all_ buffers and pagecache , dirty or not.. */
68 static void kill_bdev(struct block_device *bdev)
70 struct address_space *mapping = bdev->bd_inode->i_mapping;
72 if (mapping_empty(mapping))
76 truncate_inode_pages(mapping, 0);
79 /* Invalidate clean unused buffers and pagecache. */
80 void invalidate_bdev(struct block_device *bdev)
82 struct address_space *mapping = bdev->bd_inode->i_mapping;
84 if (mapping->nrpages) {
86 lru_add_drain_all(); /* make sure all lru add caches are flushed */
87 invalidate_mapping_pages(mapping, 0, -1);
90 EXPORT_SYMBOL(invalidate_bdev);
93 * Drop all buffers & page cache for given bdev range. This function bails
94 * with error if bdev has other exclusive owner (such as filesystem).
96 int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode,
97 loff_t lstart, loff_t lend)
100 * If we don't hold exclusive handle for the device, upgrade to it
101 * while we discard the buffer cache to avoid discarding buffers
102 * under live filesystem.
104 if (!(mode & BLK_OPEN_EXCL)) {
105 int err = bd_prepare_to_claim(bdev, truncate_bdev_range, NULL);
110 truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend);
111 if (!(mode & BLK_OPEN_EXCL))
112 bd_abort_claiming(bdev, truncate_bdev_range);
117 * Someone else has handle exclusively open. Try invalidating instead.
118 * The 'end' argument is inclusive so the rounding is safe.
120 return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
121 lstart >> PAGE_SHIFT,
125 static void set_init_blocksize(struct block_device *bdev)
127 unsigned int bsize = bdev_logical_block_size(bdev);
128 loff_t size = i_size_read(bdev->bd_inode);
130 while (bsize < PAGE_SIZE) {
135 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
138 int set_blocksize(struct block_device *bdev, int size)
140 /* Size must be a power of two, and between 512 and PAGE_SIZE */
141 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
144 /* Size cannot be smaller than the size supported by the device */
145 if (size < bdev_logical_block_size(bdev))
148 /* Don't change the size if it is same as current */
149 if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
151 bdev->bd_inode->i_blkbits = blksize_bits(size);
157 EXPORT_SYMBOL(set_blocksize);
159 int sb_set_blocksize(struct super_block *sb, int size)
161 if (set_blocksize(sb->s_bdev, size))
163 /* If we get here, we know size is power of two
164 * and it's value is between 512 and PAGE_SIZE */
165 sb->s_blocksize = size;
166 sb->s_blocksize_bits = blksize_bits(size);
167 return sb->s_blocksize;
170 EXPORT_SYMBOL(sb_set_blocksize);
172 int sb_min_blocksize(struct super_block *sb, int size)
174 int minsize = bdev_logical_block_size(sb->s_bdev);
177 return sb_set_blocksize(sb, size);
180 EXPORT_SYMBOL(sb_min_blocksize);
182 int sync_blockdev_nowait(struct block_device *bdev)
186 return filemap_flush(bdev->bd_inode->i_mapping);
188 EXPORT_SYMBOL_GPL(sync_blockdev_nowait);
191 * Write out and wait upon all the dirty data associated with a block
192 * device via its mapping. Does not take the superblock lock.
194 int sync_blockdev(struct block_device *bdev)
198 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
200 EXPORT_SYMBOL(sync_blockdev);
202 int sync_blockdev_range(struct block_device *bdev, loff_t lstart, loff_t lend)
204 return filemap_write_and_wait_range(bdev->bd_inode->i_mapping,
207 EXPORT_SYMBOL(sync_blockdev_range);
210 * Write out and wait upon all dirty data associated with this
211 * device. Filesystem data as well as the underlying block
212 * device. Takes the superblock lock.
214 int fsync_bdev(struct block_device *bdev)
216 struct super_block *sb = get_super(bdev);
218 int res = sync_filesystem(sb);
222 return sync_blockdev(bdev);
224 EXPORT_SYMBOL(fsync_bdev);
227 * freeze_bdev - lock a filesystem and force it into a consistent state
228 * @bdev: blockdevice to lock
230 * If a superblock is found on this device, we take the s_umount semaphore
231 * on it to make sure nobody unmounts until the snapshot creation is done.
232 * The reference counter (bd_fsfreeze_count) guarantees that only the last
233 * unfreeze process can unfreeze the frozen filesystem actually when multiple
234 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
235 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
238 int freeze_bdev(struct block_device *bdev)
240 struct super_block *sb;
243 mutex_lock(&bdev->bd_fsfreeze_mutex);
244 if (++bdev->bd_fsfreeze_count > 1)
247 sb = get_active_super(bdev);
250 if (sb->s_op->freeze_super)
251 error = sb->s_op->freeze_super(sb);
253 error = freeze_super(sb);
254 deactivate_super(sb);
257 bdev->bd_fsfreeze_count--;
260 bdev->bd_fsfreeze_sb = sb;
265 mutex_unlock(&bdev->bd_fsfreeze_mutex);
268 EXPORT_SYMBOL(freeze_bdev);
271 * thaw_bdev - unlock filesystem
272 * @bdev: blockdevice to unlock
274 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
276 int thaw_bdev(struct block_device *bdev)
278 struct super_block *sb;
281 mutex_lock(&bdev->bd_fsfreeze_mutex);
282 if (!bdev->bd_fsfreeze_count)
286 if (--bdev->bd_fsfreeze_count > 0)
289 sb = bdev->bd_fsfreeze_sb;
293 if (sb->s_op->thaw_super)
294 error = sb->s_op->thaw_super(sb);
296 error = thaw_super(sb);
298 bdev->bd_fsfreeze_count++;
300 bdev->bd_fsfreeze_sb = NULL;
302 mutex_unlock(&bdev->bd_fsfreeze_mutex);
305 EXPORT_SYMBOL(thaw_bdev);
311 static __cacheline_aligned_in_smp DEFINE_MUTEX(bdev_lock);
312 static struct kmem_cache * bdev_cachep __read_mostly;
314 static struct inode *bdev_alloc_inode(struct super_block *sb)
316 struct bdev_inode *ei = alloc_inode_sb(sb, bdev_cachep, GFP_KERNEL);
320 memset(&ei->bdev, 0, sizeof(ei->bdev));
321 return &ei->vfs_inode;
324 static void bdev_free_inode(struct inode *inode)
326 struct block_device *bdev = I_BDEV(inode);
328 free_percpu(bdev->bd_stats);
329 kfree(bdev->bd_meta_info);
331 if (!bdev_is_partition(bdev)) {
332 if (bdev->bd_disk && bdev->bd_disk->bdi)
333 bdi_put(bdev->bd_disk->bdi);
334 kfree(bdev->bd_disk);
337 if (MAJOR(bdev->bd_dev) == BLOCK_EXT_MAJOR)
338 blk_free_ext_minor(MINOR(bdev->bd_dev));
340 kmem_cache_free(bdev_cachep, BDEV_I(inode));
343 static void init_once(void *data)
345 struct bdev_inode *ei = data;
347 inode_init_once(&ei->vfs_inode);
350 static void bdev_evict_inode(struct inode *inode)
352 truncate_inode_pages_final(&inode->i_data);
353 invalidate_inode_buffers(inode); /* is it needed here? */
357 static const struct super_operations bdev_sops = {
358 .statfs = simple_statfs,
359 .alloc_inode = bdev_alloc_inode,
360 .free_inode = bdev_free_inode,
361 .drop_inode = generic_delete_inode,
362 .evict_inode = bdev_evict_inode,
365 static int bd_init_fs_context(struct fs_context *fc)
367 struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
370 fc->s_iflags |= SB_I_CGROUPWB;
371 ctx->ops = &bdev_sops;
375 static struct file_system_type bd_type = {
377 .init_fs_context = bd_init_fs_context,
378 .kill_sb = kill_anon_super,
381 struct super_block *blockdev_superblock __read_mostly;
382 EXPORT_SYMBOL_GPL(blockdev_superblock);
384 void __init bdev_cache_init(void)
387 static struct vfsmount *bd_mnt;
389 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
390 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
391 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
393 err = register_filesystem(&bd_type);
395 panic("Cannot register bdev pseudo-fs");
396 bd_mnt = kern_mount(&bd_type);
398 panic("Cannot create bdev pseudo-fs");
399 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
402 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
404 struct block_device *bdev;
407 inode = new_inode(blockdev_superblock);
410 inode->i_mode = S_IFBLK;
412 inode->i_data.a_ops = &def_blk_aops;
413 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
415 bdev = I_BDEV(inode);
416 mutex_init(&bdev->bd_fsfreeze_mutex);
417 spin_lock_init(&bdev->bd_size_lock);
418 mutex_init(&bdev->bd_holder_lock);
419 bdev->bd_partno = partno;
420 bdev->bd_inode = inode;
421 bdev->bd_queue = disk->queue;
423 bdev->bd_has_submit_bio = disk->part0->bd_has_submit_bio;
425 bdev->bd_has_submit_bio = false;
426 bdev->bd_stats = alloc_percpu(struct disk_stats);
427 if (!bdev->bd_stats) {
431 bdev->bd_disk = disk;
435 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors)
437 spin_lock(&bdev->bd_size_lock);
438 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
439 bdev->bd_nr_sectors = sectors;
440 spin_unlock(&bdev->bd_size_lock);
443 void bdev_add(struct block_device *bdev, dev_t dev)
446 bdev->bd_inode->i_rdev = dev;
447 bdev->bd_inode->i_ino = dev;
448 insert_inode_hash(bdev->bd_inode);
451 long nr_blockdev_pages(void)
456 spin_lock(&blockdev_superblock->s_inode_list_lock);
457 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
458 ret += inode->i_mapping->nrpages;
459 spin_unlock(&blockdev_superblock->s_inode_list_lock);
465 * bd_may_claim - test whether a block device can be claimed
466 * @bdev: block device of interest
467 * @holder: holder trying to claim @bdev
470 * Test whether @bdev can be claimed by @holder.
473 * %true if @bdev can be claimed, %false otherwise.
475 static bool bd_may_claim(struct block_device *bdev, void *holder,
476 const struct blk_holder_ops *hops)
478 struct block_device *whole = bdev_whole(bdev);
480 lockdep_assert_held(&bdev_lock);
482 if (bdev->bd_holder) {
484 * The same holder can always re-claim.
486 if (bdev->bd_holder == holder) {
487 if (WARN_ON_ONCE(bdev->bd_holder_ops != hops))
495 * If the whole devices holder is set to bd_may_claim, a partition on
496 * the device is claimed, but not the whole device.
499 whole->bd_holder && whole->bd_holder != bd_may_claim)
505 * bd_prepare_to_claim - claim a block device
506 * @bdev: block device of interest
507 * @holder: holder trying to claim @bdev
510 * Claim @bdev. This function fails if @bdev is already claimed by another
511 * holder and waits if another claiming is in progress. return, the caller
512 * has ownership of bd_claiming and bd_holder[s].
515 * 0 if @bdev can be claimed, -EBUSY otherwise.
517 int bd_prepare_to_claim(struct block_device *bdev, void *holder,
518 const struct blk_holder_ops *hops)
520 struct block_device *whole = bdev_whole(bdev);
522 if (WARN_ON_ONCE(!holder))
525 mutex_lock(&bdev_lock);
526 /* if someone else claimed, fail */
527 if (!bd_may_claim(bdev, holder, hops)) {
528 mutex_unlock(&bdev_lock);
532 /* if claiming is already in progress, wait for it to finish */
533 if (whole->bd_claiming) {
534 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
537 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
538 mutex_unlock(&bdev_lock);
540 finish_wait(wq, &wait);
545 whole->bd_claiming = holder;
546 mutex_unlock(&bdev_lock);
549 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
551 static void bd_clear_claiming(struct block_device *whole, void *holder)
553 lockdep_assert_held(&bdev_lock);
554 /* tell others that we're done */
555 BUG_ON(whole->bd_claiming != holder);
556 whole->bd_claiming = NULL;
557 wake_up_bit(&whole->bd_claiming, 0);
561 * bd_finish_claiming - finish claiming of a block device
562 * @bdev: block device of interest
563 * @holder: holder that has claimed @bdev
564 * @hops: block device holder operations
566 * Finish exclusive open of a block device. Mark the device as exlusively
567 * open by the holder and wake up all waiters for exclusive open to finish.
569 static void bd_finish_claiming(struct block_device *bdev, void *holder,
570 const struct blk_holder_ops *hops)
572 struct block_device *whole = bdev_whole(bdev);
574 mutex_lock(&bdev_lock);
575 BUG_ON(!bd_may_claim(bdev, holder, hops));
577 * Note that for a whole device bd_holders will be incremented twice,
578 * and bd_holder will be set to bd_may_claim before being set to holder
581 whole->bd_holder = bd_may_claim;
583 mutex_lock(&bdev->bd_holder_lock);
584 bdev->bd_holder = holder;
585 bdev->bd_holder_ops = hops;
586 mutex_unlock(&bdev->bd_holder_lock);
587 bd_clear_claiming(whole, holder);
588 mutex_unlock(&bdev_lock);
592 * bd_abort_claiming - abort claiming of a block device
593 * @bdev: block device of interest
594 * @holder: holder that has claimed @bdev
596 * Abort claiming of a block device when the exclusive open failed. This can be
597 * also used when exclusive open is not actually desired and we just needed
598 * to block other exclusive openers for a while.
600 void bd_abort_claiming(struct block_device *bdev, void *holder)
602 mutex_lock(&bdev_lock);
603 bd_clear_claiming(bdev_whole(bdev), holder);
604 mutex_unlock(&bdev_lock);
606 EXPORT_SYMBOL(bd_abort_claiming);
608 static void bd_end_claim(struct block_device *bdev, void *holder)
610 struct block_device *whole = bdev_whole(bdev);
611 bool unblock = false;
614 * Release a claim on the device. The holder fields are protected with
615 * bdev_lock. open_mutex is used to synchronize disk_holder unlinking.
617 mutex_lock(&bdev_lock);
618 WARN_ON_ONCE(bdev->bd_holder != holder);
619 WARN_ON_ONCE(--bdev->bd_holders < 0);
620 WARN_ON_ONCE(--whole->bd_holders < 0);
621 if (!bdev->bd_holders) {
622 mutex_lock(&bdev->bd_holder_lock);
623 bdev->bd_holder = NULL;
624 bdev->bd_holder_ops = NULL;
625 mutex_unlock(&bdev->bd_holder_lock);
626 if (bdev->bd_write_holder)
629 if (!whole->bd_holders)
630 whole->bd_holder = NULL;
631 mutex_unlock(&bdev_lock);
634 * If this was the last claim, remove holder link and unblock evpoll if
635 * it was a write holder.
638 disk_unblock_events(bdev->bd_disk);
639 bdev->bd_write_holder = false;
643 static void blkdev_flush_mapping(struct block_device *bdev)
645 WARN_ON_ONCE(bdev->bd_holders);
648 bdev_write_inode(bdev);
651 static int blkdev_get_whole(struct block_device *bdev, blk_mode_t mode)
653 struct gendisk *disk = bdev->bd_disk;
656 if (disk->fops->open) {
657 ret = disk->fops->open(disk, mode);
659 /* avoid ghost partitions on a removed medium */
660 if (ret == -ENOMEDIUM &&
661 test_bit(GD_NEED_PART_SCAN, &disk->state))
662 bdev_disk_changed(disk, true);
667 if (!atomic_read(&bdev->bd_openers))
668 set_init_blocksize(bdev);
669 if (test_bit(GD_NEED_PART_SCAN, &disk->state))
670 bdev_disk_changed(disk, false);
671 atomic_inc(&bdev->bd_openers);
675 static void blkdev_put_whole(struct block_device *bdev)
677 if (atomic_dec_and_test(&bdev->bd_openers))
678 blkdev_flush_mapping(bdev);
679 if (bdev->bd_disk->fops->release)
680 bdev->bd_disk->fops->release(bdev->bd_disk);
683 static int blkdev_get_part(struct block_device *part, blk_mode_t mode)
685 struct gendisk *disk = part->bd_disk;
688 ret = blkdev_get_whole(bdev_whole(part), mode);
693 if (!bdev_nr_sectors(part))
696 if (!atomic_read(&part->bd_openers)) {
697 disk->open_partitions++;
698 set_init_blocksize(part);
700 atomic_inc(&part->bd_openers);
704 blkdev_put_whole(bdev_whole(part));
708 static void blkdev_put_part(struct block_device *part)
710 struct block_device *whole = bdev_whole(part);
712 if (atomic_dec_and_test(&part->bd_openers)) {
713 blkdev_flush_mapping(part);
714 whole->bd_disk->open_partitions--;
716 blkdev_put_whole(whole);
719 struct block_device *blkdev_get_no_open(dev_t dev)
721 struct block_device *bdev;
724 inode = ilookup(blockdev_superblock, dev);
725 if (!inode && IS_ENABLED(CONFIG_BLOCK_LEGACY_AUTOLOAD)) {
726 blk_request_module(dev);
727 inode = ilookup(blockdev_superblock, dev);
730 "block device autoloading is deprecated and will be removed.\n");
735 /* switch from the inode reference to a device mode one: */
736 bdev = &BDEV_I(inode)->bdev;
737 if (!kobject_get_unless_zero(&bdev->bd_device.kobj))
743 void blkdev_put_no_open(struct block_device *bdev)
745 put_device(&bdev->bd_device);
749 * blkdev_get_by_dev - open a block device by device number
750 * @dev: device number of block device to open
751 * @mode: open mode (BLK_OPEN_*)
752 * @holder: exclusive holder identifier
753 * @hops: holder operations
755 * Open the block device described by device number @dev. If @holder is not
756 * %NULL, the block device is opened with exclusive access. Exclusive opens may
757 * nest for the same @holder.
759 * Use this interface ONLY if you really do not have anything better - i.e. when
760 * you are behind a truly sucky interface and all you are given is a device
761 * number. Everything else should use blkdev_get_by_path().
767 * Reference to the block_device on success, ERR_PTR(-errno) on failure.
769 struct block_device *blkdev_get_by_dev(dev_t dev, blk_mode_t mode, void *holder,
770 const struct blk_holder_ops *hops)
772 bool unblock_events = true;
773 struct block_device *bdev;
774 struct gendisk *disk;
777 ret = devcgroup_check_permission(DEVCG_DEV_BLOCK,
778 MAJOR(dev), MINOR(dev),
779 ((mode & BLK_OPEN_READ) ? DEVCG_ACC_READ : 0) |
780 ((mode & BLK_OPEN_WRITE) ? DEVCG_ACC_WRITE : 0));
784 bdev = blkdev_get_no_open(dev);
786 return ERR_PTR(-ENXIO);
787 disk = bdev->bd_disk;
790 mode |= BLK_OPEN_EXCL;
791 ret = bd_prepare_to_claim(bdev, holder, hops);
795 if (WARN_ON_ONCE(mode & BLK_OPEN_EXCL)) {
801 disk_block_events(disk);
803 mutex_lock(&disk->open_mutex);
805 if (!disk_live(disk))
807 if (!try_module_get(disk->fops->owner))
809 if (bdev_is_partition(bdev))
810 ret = blkdev_get_part(bdev, mode);
812 ret = blkdev_get_whole(bdev, mode);
816 bd_finish_claiming(bdev, holder, hops);
819 * Block event polling for write claims if requested. Any write
820 * holder makes the write_holder state stick until all are
821 * released. This is good enough and tracking individual
822 * writeable reference is too fragile given the way @mode is
823 * used in blkdev_get/put().
825 if ((mode & BLK_OPEN_WRITE) && !bdev->bd_write_holder &&
826 (disk->event_flags & DISK_EVENT_FLAG_BLOCK_ON_EXCL_WRITE)) {
827 bdev->bd_write_holder = true;
828 unblock_events = false;
831 mutex_unlock(&disk->open_mutex);
834 disk_unblock_events(disk);
837 module_put(disk->fops->owner);
840 bd_abort_claiming(bdev, holder);
841 mutex_unlock(&disk->open_mutex);
842 disk_unblock_events(disk);
844 blkdev_put_no_open(bdev);
847 EXPORT_SYMBOL(blkdev_get_by_dev);
850 * blkdev_get_by_path - open a block device by name
851 * @path: path to the block device to open
852 * @mode: open mode (BLK_OPEN_*)
853 * @holder: exclusive holder identifier
854 * @hops: holder operations
856 * Open the block device described by the device file at @path. If @holder is
857 * not %NULL, the block device is opened with exclusive access. Exclusive opens
858 * may nest for the same @holder.
864 * Reference to the block_device on success, ERR_PTR(-errno) on failure.
866 struct block_device *blkdev_get_by_path(const char *path, blk_mode_t mode,
867 void *holder, const struct blk_holder_ops *hops)
869 struct block_device *bdev;
873 error = lookup_bdev(path, &dev);
875 return ERR_PTR(error);
877 bdev = blkdev_get_by_dev(dev, mode, holder, hops);
878 if (!IS_ERR(bdev) && (mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) {
879 blkdev_put(bdev, holder);
880 return ERR_PTR(-EACCES);
885 EXPORT_SYMBOL(blkdev_get_by_path);
887 void blkdev_put(struct block_device *bdev, void *holder)
889 struct gendisk *disk = bdev->bd_disk;
892 * Sync early if it looks like we're the last one. If someone else
893 * opens the block device between now and the decrement of bd_openers
894 * then we did a sync that we didn't need to, but that's not the end
895 * of the world and we want to avoid long (could be several minute)
896 * syncs while holding the mutex.
898 if (atomic_read(&bdev->bd_openers) == 1)
901 mutex_lock(&disk->open_mutex);
903 bd_end_claim(bdev, holder);
906 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
907 * event. This is to ensure detection of media removal commanded
908 * from userland - e.g. eject(1).
910 disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE);
912 if (bdev_is_partition(bdev))
913 blkdev_put_part(bdev);
915 blkdev_put_whole(bdev);
916 mutex_unlock(&disk->open_mutex);
918 module_put(disk->fops->owner);
919 blkdev_put_no_open(bdev);
921 EXPORT_SYMBOL(blkdev_put);
924 * lookup_bdev() - Look up a struct block_device by name.
925 * @pathname: Name of the block device in the filesystem.
926 * @dev: Pointer to the block device's dev_t, if found.
928 * Lookup the block device's dev_t at @pathname in the current
929 * namespace if possible and return it in @dev.
931 * Context: May sleep.
932 * Return: 0 if succeeded, negative errno otherwise.
934 int lookup_bdev(const char *pathname, dev_t *dev)
940 if (!pathname || !*pathname)
943 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
947 inode = d_backing_inode(path.dentry);
949 if (!S_ISBLK(inode->i_mode))
952 if (!may_open_dev(&path))
955 *dev = inode->i_rdev;
961 EXPORT_SYMBOL(lookup_bdev);
963 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
965 struct super_block *sb = get_super(bdev);
970 * no need to lock the super, get_super holds the
971 * read mutex so the filesystem cannot go away
972 * under us (->put_super runs with the write lock
975 shrink_dcache_sb(sb);
976 res = invalidate_inodes(sb, kill_dirty);
979 invalidate_bdev(bdev);
982 EXPORT_SYMBOL(__invalidate_device);
984 void sync_bdevs(bool wait)
986 struct inode *inode, *old_inode = NULL;
988 spin_lock(&blockdev_superblock->s_inode_list_lock);
989 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
990 struct address_space *mapping = inode->i_mapping;
991 struct block_device *bdev;
993 spin_lock(&inode->i_lock);
994 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
995 mapping->nrpages == 0) {
996 spin_unlock(&inode->i_lock);
1000 spin_unlock(&inode->i_lock);
1001 spin_unlock(&blockdev_superblock->s_inode_list_lock);
1003 * We hold a reference to 'inode' so it couldn't have been
1004 * removed from s_inodes list while we dropped the
1005 * s_inode_list_lock We cannot iput the inode now as we can
1006 * be holding the last reference and we cannot iput it under
1007 * s_inode_list_lock. So we keep the reference and iput it
1012 bdev = I_BDEV(inode);
1014 mutex_lock(&bdev->bd_disk->open_mutex);
1015 if (!atomic_read(&bdev->bd_openers)) {
1019 * We keep the error status of individual mapping so
1020 * that applications can catch the writeback error using
1021 * fsync(2). See filemap_fdatawait_keep_errors() for
1024 filemap_fdatawait_keep_errors(inode->i_mapping);
1026 filemap_fdatawrite(inode->i_mapping);
1028 mutex_unlock(&bdev->bd_disk->open_mutex);
1030 spin_lock(&blockdev_superblock->s_inode_list_lock);
1032 spin_unlock(&blockdev_superblock->s_inode_list_lock);
1037 * Handle STATX_DIOALIGN for block devices.
1039 * Note that the inode passed to this is the inode of a block device node file,
1040 * not the block device's internal inode. Therefore it is *not* valid to use
1041 * I_BDEV() here; the block device has to be looked up by i_rdev instead.
1043 void bdev_statx_dioalign(struct inode *inode, struct kstat *stat)
1045 struct block_device *bdev;
1047 bdev = blkdev_get_no_open(inode->i_rdev);
1051 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
1052 stat->dio_offset_align = bdev_logical_block_size(bdev);
1053 stat->result_mask |= STATX_DIOALIGN;
1055 blkdev_put_no_open(bdev);