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;
62 if (unlikely(dst == old)) /* deadlock avoidance */
64 bdi_lock_two(&old->wb, &dst->wb);
65 spin_lock(&inode->i_lock);
66 inode->i_data.backing_dev_info = dst;
67 if (inode->i_state & I_DIRTY)
68 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
69 spin_unlock(&inode->i_lock);
70 spin_unlock(&old->wb.list_lock);
71 spin_unlock(&dst->wb.list_lock);
74 /* Kill _all_ buffers and pagecache , dirty or not.. */
75 void kill_bdev(struct block_device *bdev)
77 struct address_space *mapping = bdev->bd_inode->i_mapping;
79 if (mapping->nrpages == 0)
83 truncate_inode_pages(mapping, 0);
85 EXPORT_SYMBOL(kill_bdev);
87 /* Invalidate clean unused buffers and pagecache. */
88 void invalidate_bdev(struct block_device *bdev)
90 struct address_space *mapping = bdev->bd_inode->i_mapping;
92 if (mapping->nrpages == 0)
96 lru_add_drain_all(); /* make sure all lru add caches are flushed */
97 invalidate_mapping_pages(mapping, 0, -1);
98 /* 99% of the time, we don't need to flush the cleancache on the bdev.
99 * But, for the strange corners, lets be cautious
101 cleancache_invalidate_inode(mapping);
103 EXPORT_SYMBOL(invalidate_bdev);
105 int set_blocksize(struct block_device *bdev, int size)
107 /* Size must be a power of two, and between 512 and PAGE_SIZE */
108 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
111 /* Size cannot be smaller than the size supported by the device */
112 if (size < bdev_logical_block_size(bdev))
115 /* Don't change the size if it is same as current */
116 if (bdev->bd_block_size != size) {
118 bdev->bd_block_size = size;
119 bdev->bd_inode->i_blkbits = blksize_bits(size);
125 EXPORT_SYMBOL(set_blocksize);
127 int sb_set_blocksize(struct super_block *sb, int size)
129 if (set_blocksize(sb->s_bdev, size))
131 /* If we get here, we know size is power of two
132 * and it's value is between 512 and PAGE_SIZE */
133 sb->s_blocksize = size;
134 sb->s_blocksize_bits = blksize_bits(size);
135 return sb->s_blocksize;
138 EXPORT_SYMBOL(sb_set_blocksize);
140 int sb_min_blocksize(struct super_block *sb, int size)
142 int minsize = bdev_logical_block_size(sb->s_bdev);
145 return sb_set_blocksize(sb, size);
148 EXPORT_SYMBOL(sb_min_blocksize);
151 blkdev_get_block(struct inode *inode, sector_t iblock,
152 struct buffer_head *bh, int create)
154 bh->b_bdev = I_BDEV(inode);
155 bh->b_blocknr = iblock;
156 set_buffer_mapped(bh);
161 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
162 loff_t offset, unsigned long nr_segs)
164 struct file *file = iocb->ki_filp;
165 struct inode *inode = file->f_mapping->host;
167 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
168 nr_segs, blkdev_get_block, NULL, NULL, 0);
171 int __sync_blockdev(struct block_device *bdev, int wait)
176 return filemap_flush(bdev->bd_inode->i_mapping);
177 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
181 * Write out and wait upon all the dirty data associated with a block
182 * device via its mapping. Does not take the superblock lock.
184 int sync_blockdev(struct block_device *bdev)
186 return __sync_blockdev(bdev, 1);
188 EXPORT_SYMBOL(sync_blockdev);
191 * Write out and wait upon all dirty data associated with this
192 * device. Filesystem data as well as the underlying block
193 * device. Takes the superblock lock.
195 int fsync_bdev(struct block_device *bdev)
197 struct super_block *sb = get_super(bdev);
199 int res = sync_filesystem(sb);
203 return sync_blockdev(bdev);
205 EXPORT_SYMBOL(fsync_bdev);
208 * freeze_bdev -- lock a filesystem and force it into a consistent state
209 * @bdev: blockdevice to lock
211 * If a superblock is found on this device, we take the s_umount semaphore
212 * on it to make sure nobody unmounts until the snapshot creation is done.
213 * The reference counter (bd_fsfreeze_count) guarantees that only the last
214 * unfreeze process can unfreeze the frozen filesystem actually when multiple
215 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
216 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
219 struct super_block *freeze_bdev(struct block_device *bdev)
221 struct super_block *sb;
224 mutex_lock(&bdev->bd_fsfreeze_mutex);
225 if (++bdev->bd_fsfreeze_count > 1) {
227 * We don't even need to grab a reference - the first call
228 * to freeze_bdev grab an active reference and only the last
229 * thaw_bdev drops it.
231 sb = get_super(bdev);
233 mutex_unlock(&bdev->bd_fsfreeze_mutex);
237 sb = get_active_super(bdev);
240 error = freeze_super(sb);
242 deactivate_super(sb);
243 bdev->bd_fsfreeze_count--;
244 mutex_unlock(&bdev->bd_fsfreeze_mutex);
245 return ERR_PTR(error);
247 deactivate_super(sb);
250 mutex_unlock(&bdev->bd_fsfreeze_mutex);
251 return sb; /* thaw_bdev releases s->s_umount */
253 EXPORT_SYMBOL(freeze_bdev);
256 * thaw_bdev -- unlock filesystem
257 * @bdev: blockdevice to unlock
258 * @sb: associated superblock
260 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
262 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
266 mutex_lock(&bdev->bd_fsfreeze_mutex);
267 if (!bdev->bd_fsfreeze_count)
271 if (--bdev->bd_fsfreeze_count > 0)
277 error = thaw_super(sb);
279 bdev->bd_fsfreeze_count++;
280 mutex_unlock(&bdev->bd_fsfreeze_mutex);
284 mutex_unlock(&bdev->bd_fsfreeze_mutex);
287 EXPORT_SYMBOL(thaw_bdev);
289 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
291 return block_write_full_page(page, blkdev_get_block, wbc);
294 static int blkdev_readpage(struct file * file, struct page * page)
296 return block_read_full_page(page, blkdev_get_block);
299 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
300 loff_t pos, unsigned len, unsigned flags,
301 struct page **pagep, void **fsdata)
303 return block_write_begin(mapping, pos, len, flags, pagep,
307 static int blkdev_write_end(struct file *file, struct address_space *mapping,
308 loff_t pos, unsigned len, unsigned copied,
309 struct page *page, void *fsdata)
312 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
315 page_cache_release(page);
322 * for a block special file file_inode(file)->i_size is zero
323 * so we compute the size by hand (just as in block_read/write above)
325 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
327 struct inode *bd_inode = file->f_mapping->host;
330 mutex_lock(&bd_inode->i_mutex);
331 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
332 mutex_unlock(&bd_inode->i_mutex);
336 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
338 struct inode *bd_inode = filp->f_mapping->host;
339 struct block_device *bdev = I_BDEV(bd_inode);
342 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
347 * There is no need to serialise calls to blkdev_issue_flush with
348 * i_mutex and doing so causes performance issues with concurrent
349 * O_SYNC writers to a block device.
351 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
352 if (error == -EOPNOTSUPP)
357 EXPORT_SYMBOL(blkdev_fsync);
363 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
364 static struct kmem_cache * bdev_cachep __read_mostly;
366 static struct inode *bdev_alloc_inode(struct super_block *sb)
368 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
371 return &ei->vfs_inode;
374 static void bdev_i_callback(struct rcu_head *head)
376 struct inode *inode = container_of(head, struct inode, i_rcu);
377 struct bdev_inode *bdi = BDEV_I(inode);
379 kmem_cache_free(bdev_cachep, bdi);
382 static void bdev_destroy_inode(struct inode *inode)
384 call_rcu(&inode->i_rcu, bdev_i_callback);
387 static void init_once(void *foo)
389 struct bdev_inode *ei = (struct bdev_inode *) foo;
390 struct block_device *bdev = &ei->bdev;
392 memset(bdev, 0, sizeof(*bdev));
393 mutex_init(&bdev->bd_mutex);
394 INIT_LIST_HEAD(&bdev->bd_inodes);
395 INIT_LIST_HEAD(&bdev->bd_list);
397 INIT_LIST_HEAD(&bdev->bd_holder_disks);
399 inode_init_once(&ei->vfs_inode);
400 /* Initialize mutex for freeze. */
401 mutex_init(&bdev->bd_fsfreeze_mutex);
404 static inline void __bd_forget(struct inode *inode)
406 list_del_init(&inode->i_devices);
407 inode->i_bdev = NULL;
408 inode->i_mapping = &inode->i_data;
411 static void bdev_evict_inode(struct inode *inode)
413 struct block_device *bdev = &BDEV_I(inode)->bdev;
415 truncate_inode_pages(&inode->i_data, 0);
416 invalidate_inode_buffers(inode); /* is it needed here? */
418 spin_lock(&bdev_lock);
419 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
420 __bd_forget(list_entry(p, struct inode, i_devices));
422 list_del_init(&bdev->bd_list);
423 spin_unlock(&bdev_lock);
426 static const struct super_operations bdev_sops = {
427 .statfs = simple_statfs,
428 .alloc_inode = bdev_alloc_inode,
429 .destroy_inode = bdev_destroy_inode,
430 .drop_inode = generic_delete_inode,
431 .evict_inode = bdev_evict_inode,
434 static struct dentry *bd_mount(struct file_system_type *fs_type,
435 int flags, const char *dev_name, void *data)
437 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
440 static struct file_system_type bd_type = {
443 .kill_sb = kill_anon_super,
446 static struct super_block *blockdev_superblock __read_mostly;
448 void __init bdev_cache_init(void)
451 static struct vfsmount *bd_mnt;
453 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
454 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
455 SLAB_MEM_SPREAD|SLAB_PANIC),
457 err = register_filesystem(&bd_type);
459 panic("Cannot register bdev pseudo-fs");
460 bd_mnt = kern_mount(&bd_type);
462 panic("Cannot create bdev pseudo-fs");
463 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
467 * Most likely _very_ bad one - but then it's hardly critical for small
468 * /dev and can be fixed when somebody will need really large one.
469 * Keep in mind that it will be fed through icache hash function too.
471 static inline unsigned long hash(dev_t dev)
473 return MAJOR(dev)+MINOR(dev);
476 static int bdev_test(struct inode *inode, void *data)
478 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
481 static int bdev_set(struct inode *inode, void *data)
483 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
487 static LIST_HEAD(all_bdevs);
489 struct block_device *bdget(dev_t dev)
491 struct block_device *bdev;
494 inode = iget5_locked(blockdev_superblock, hash(dev),
495 bdev_test, bdev_set, &dev);
500 bdev = &BDEV_I(inode)->bdev;
502 if (inode->i_state & I_NEW) {
503 bdev->bd_contains = NULL;
504 bdev->bd_super = NULL;
505 bdev->bd_inode = inode;
506 bdev->bd_block_size = (1 << inode->i_blkbits);
507 bdev->bd_part_count = 0;
508 bdev->bd_invalidated = 0;
509 inode->i_mode = S_IFBLK;
511 inode->i_bdev = bdev;
512 inode->i_data.a_ops = &def_blk_aops;
513 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
514 inode->i_data.backing_dev_info = &default_backing_dev_info;
515 spin_lock(&bdev_lock);
516 list_add(&bdev->bd_list, &all_bdevs);
517 spin_unlock(&bdev_lock);
518 unlock_new_inode(inode);
523 EXPORT_SYMBOL(bdget);
526 * bdgrab -- Grab a reference to an already referenced block device
527 * @bdev: Block device to grab a reference to.
529 struct block_device *bdgrab(struct block_device *bdev)
531 ihold(bdev->bd_inode);
534 EXPORT_SYMBOL(bdgrab);
536 long nr_blockdev_pages(void)
538 struct block_device *bdev;
540 spin_lock(&bdev_lock);
541 list_for_each_entry(bdev, &all_bdevs, bd_list) {
542 ret += bdev->bd_inode->i_mapping->nrpages;
544 spin_unlock(&bdev_lock);
548 void bdput(struct block_device *bdev)
550 iput(bdev->bd_inode);
553 EXPORT_SYMBOL(bdput);
555 static struct block_device *bd_acquire(struct inode *inode)
557 struct block_device *bdev;
559 spin_lock(&bdev_lock);
560 bdev = inode->i_bdev;
562 ihold(bdev->bd_inode);
563 spin_unlock(&bdev_lock);
566 spin_unlock(&bdev_lock);
568 bdev = bdget(inode->i_rdev);
570 spin_lock(&bdev_lock);
571 if (!inode->i_bdev) {
573 * We take an additional reference to bd_inode,
574 * and it's released in clear_inode() of inode.
575 * So, we can access it via ->i_mapping always
578 ihold(bdev->bd_inode);
579 inode->i_bdev = bdev;
580 inode->i_mapping = bdev->bd_inode->i_mapping;
581 list_add(&inode->i_devices, &bdev->bd_inodes);
583 spin_unlock(&bdev_lock);
588 static inline int sb_is_blkdev_sb(struct super_block *sb)
590 return sb == blockdev_superblock;
593 /* Call when you free inode */
595 void bd_forget(struct inode *inode)
597 struct block_device *bdev = NULL;
599 spin_lock(&bdev_lock);
600 if (!sb_is_blkdev_sb(inode->i_sb))
601 bdev = inode->i_bdev;
603 spin_unlock(&bdev_lock);
606 iput(bdev->bd_inode);
610 * bd_may_claim - test whether a block device can be claimed
611 * @bdev: block device of interest
612 * @whole: whole block device containing @bdev, may equal @bdev
613 * @holder: holder trying to claim @bdev
615 * Test whether @bdev can be claimed by @holder.
618 * spin_lock(&bdev_lock).
621 * %true if @bdev can be claimed, %false otherwise.
623 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
626 if (bdev->bd_holder == holder)
627 return true; /* already a holder */
628 else if (bdev->bd_holder != NULL)
629 return false; /* held by someone else */
630 else if (bdev->bd_contains == bdev)
631 return true; /* is a whole device which isn't held */
633 else if (whole->bd_holder == bd_may_claim)
634 return true; /* is a partition of a device that is being partitioned */
635 else if (whole->bd_holder != NULL)
636 return false; /* is a partition of a held device */
638 return true; /* is a partition of an un-held device */
642 * bd_prepare_to_claim - prepare to claim a block device
643 * @bdev: block device of interest
644 * @whole: the whole device containing @bdev, may equal @bdev
645 * @holder: holder trying to claim @bdev
647 * Prepare to claim @bdev. This function fails if @bdev is already
648 * claimed by another holder and waits if another claiming is in
649 * progress. This function doesn't actually claim. On successful
650 * return, the caller has ownership of bd_claiming and bd_holder[s].
653 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
657 * 0 if @bdev can be claimed, -EBUSY otherwise.
659 static int bd_prepare_to_claim(struct block_device *bdev,
660 struct block_device *whole, void *holder)
663 /* if someone else claimed, fail */
664 if (!bd_may_claim(bdev, whole, holder))
667 /* if claiming is already in progress, wait for it to finish */
668 if (whole->bd_claiming) {
669 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
672 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
673 spin_unlock(&bdev_lock);
675 finish_wait(wq, &wait);
676 spin_lock(&bdev_lock);
685 * bd_start_claiming - start claiming a block device
686 * @bdev: block device of interest
687 * @holder: holder trying to claim @bdev
689 * @bdev is about to be opened exclusively. Check @bdev can be opened
690 * exclusively and mark that an exclusive open is in progress. Each
691 * successful call to this function must be matched with a call to
692 * either bd_finish_claiming() or bd_abort_claiming() (which do not
695 * This function is used to gain exclusive access to the block device
696 * without actually causing other exclusive open attempts to fail. It
697 * should be used when the open sequence itself requires exclusive
698 * access but may subsequently fail.
704 * Pointer to the block device containing @bdev on success, ERR_PTR()
707 static struct block_device *bd_start_claiming(struct block_device *bdev,
710 struct gendisk *disk;
711 struct block_device *whole;
717 * @bdev might not have been initialized properly yet, look up
718 * and grab the outer block device the hard way.
720 disk = get_gendisk(bdev->bd_dev, &partno);
722 return ERR_PTR(-ENXIO);
725 * Normally, @bdev should equal what's returned from bdget_disk()
726 * if partno is 0; however, some drivers (floppy) use multiple
727 * bdev's for the same physical device and @bdev may be one of the
728 * aliases. Keep @bdev if partno is 0. This means claimer
729 * tracking is broken for those devices but it has always been that
733 whole = bdget_disk(disk, 0);
735 whole = bdgrab(bdev);
737 module_put(disk->fops->owner);
740 return ERR_PTR(-ENOMEM);
742 /* prepare to claim, if successful, mark claiming in progress */
743 spin_lock(&bdev_lock);
745 err = bd_prepare_to_claim(bdev, whole, holder);
747 whole->bd_claiming = holder;
748 spin_unlock(&bdev_lock);
751 spin_unlock(&bdev_lock);
758 struct bd_holder_disk {
759 struct list_head list;
760 struct gendisk *disk;
764 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
765 struct gendisk *disk)
767 struct bd_holder_disk *holder;
769 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
770 if (holder->disk == disk)
775 static int add_symlink(struct kobject *from, struct kobject *to)
777 return sysfs_create_link(from, to, kobject_name(to));
780 static void del_symlink(struct kobject *from, struct kobject *to)
782 sysfs_remove_link(from, kobject_name(to));
786 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
787 * @bdev: the claimed slave bdev
788 * @disk: the holding disk
790 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
792 * This functions creates the following sysfs symlinks.
794 * - from "slaves" directory of the holder @disk to the claimed @bdev
795 * - from "holders" directory of the @bdev to the holder @disk
797 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
798 * passed to bd_link_disk_holder(), then:
800 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
801 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
803 * The caller must have claimed @bdev before calling this function and
804 * ensure that both @bdev and @disk are valid during the creation and
805 * lifetime of these symlinks.
811 * 0 on success, -errno on failure.
813 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
815 struct bd_holder_disk *holder;
818 mutex_lock(&bdev->bd_mutex);
820 WARN_ON_ONCE(!bdev->bd_holder);
822 /* FIXME: remove the following once add_disk() handles errors */
823 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
826 holder = bd_find_holder_disk(bdev, disk);
832 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
838 INIT_LIST_HEAD(&holder->list);
842 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
846 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
850 * bdev could be deleted beneath us which would implicitly destroy
851 * the holder directory. Hold on to it.
853 kobject_get(bdev->bd_part->holder_dir);
855 list_add(&holder->list, &bdev->bd_holder_disks);
859 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
863 mutex_unlock(&bdev->bd_mutex);
866 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
869 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
870 * @bdev: the calimed slave bdev
871 * @disk: the holding disk
873 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
878 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
880 struct bd_holder_disk *holder;
882 mutex_lock(&bdev->bd_mutex);
884 holder = bd_find_holder_disk(bdev, disk);
886 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
887 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
888 del_symlink(bdev->bd_part->holder_dir,
889 &disk_to_dev(disk)->kobj);
890 kobject_put(bdev->bd_part->holder_dir);
891 list_del_init(&holder->list);
895 mutex_unlock(&bdev->bd_mutex);
897 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
901 * flush_disk - invalidates all buffer-cache entries on a disk
903 * @bdev: struct block device to be flushed
904 * @kill_dirty: flag to guide handling of dirty inodes
906 * Invalidates all buffer-cache entries on a disk. It should be called
907 * when a disk has been changed -- either by a media change or online
910 static void flush_disk(struct block_device *bdev, bool kill_dirty)
912 if (__invalidate_device(bdev, kill_dirty)) {
913 char name[BDEVNAME_SIZE] = "";
916 disk_name(bdev->bd_disk, 0, name);
917 printk(KERN_WARNING "VFS: busy inodes on changed media or "
918 "resized disk %s\n", name);
923 if (disk_part_scan_enabled(bdev->bd_disk))
924 bdev->bd_invalidated = 1;
928 * check_disk_size_change - checks for disk size change and adjusts bdev size.
929 * @disk: struct gendisk to check
930 * @bdev: struct bdev to adjust.
932 * This routine checks to see if the bdev size does not match the disk size
933 * and adjusts it if it differs.
935 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
937 loff_t disk_size, bdev_size;
939 disk_size = (loff_t)get_capacity(disk) << 9;
940 bdev_size = i_size_read(bdev->bd_inode);
941 if (disk_size != bdev_size) {
942 char name[BDEVNAME_SIZE];
944 disk_name(disk, 0, name);
946 "%s: detected capacity change from %lld to %lld\n",
947 name, bdev_size, disk_size);
948 i_size_write(bdev->bd_inode, disk_size);
949 flush_disk(bdev, false);
952 EXPORT_SYMBOL(check_disk_size_change);
955 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
956 * @disk: struct gendisk to be revalidated
958 * This routine is a wrapper for lower-level driver's revalidate_disk
959 * call-backs. It is used to do common pre and post operations needed
960 * for all revalidate_disk operations.
962 int revalidate_disk(struct gendisk *disk)
964 struct block_device *bdev;
967 if (disk->fops->revalidate_disk)
968 ret = disk->fops->revalidate_disk(disk);
970 bdev = bdget_disk(disk, 0);
974 mutex_lock(&bdev->bd_mutex);
975 check_disk_size_change(disk, bdev);
976 bdev->bd_invalidated = 0;
977 mutex_unlock(&bdev->bd_mutex);
981 EXPORT_SYMBOL(revalidate_disk);
984 * This routine checks whether a removable media has been changed,
985 * and invalidates all buffer-cache-entries in that case. This
986 * is a relatively slow routine, so we have to try to minimize using
987 * it. Thus it is called only upon a 'mount' or 'open'. This
988 * is the best way of combining speed and utility, I think.
989 * People changing diskettes in the middle of an operation deserve
992 int check_disk_change(struct block_device *bdev)
994 struct gendisk *disk = bdev->bd_disk;
995 const struct block_device_operations *bdops = disk->fops;
998 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
999 DISK_EVENT_EJECT_REQUEST);
1000 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1003 flush_disk(bdev, true);
1004 if (bdops->revalidate_disk)
1005 bdops->revalidate_disk(bdev->bd_disk);
1009 EXPORT_SYMBOL(check_disk_change);
1011 void bd_set_size(struct block_device *bdev, loff_t size)
1013 unsigned bsize = bdev_logical_block_size(bdev);
1015 mutex_lock(&bdev->bd_inode->i_mutex);
1016 i_size_write(bdev->bd_inode, size);
1017 mutex_unlock(&bdev->bd_inode->i_mutex);
1018 while (bsize < PAGE_CACHE_SIZE) {
1023 bdev->bd_block_size = bsize;
1024 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1026 EXPORT_SYMBOL(bd_set_size);
1028 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1033 * mutex_lock(part->bd_mutex)
1034 * mutex_lock_nested(whole->bd_mutex, 1)
1037 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1039 struct gendisk *disk;
1040 struct module *owner;
1045 if (mode & FMODE_READ)
1047 if (mode & FMODE_WRITE)
1050 * hooks: /n/, see "layering violations".
1053 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1063 disk = get_gendisk(bdev->bd_dev, &partno);
1066 owner = disk->fops->owner;
1068 disk_block_events(disk);
1069 mutex_lock_nested(&bdev->bd_mutex, for_part);
1070 if (!bdev->bd_openers) {
1071 bdev->bd_disk = disk;
1072 bdev->bd_queue = disk->queue;
1073 bdev->bd_contains = bdev;
1075 struct backing_dev_info *bdi;
1078 bdev->bd_part = disk_get_part(disk, partno);
1083 if (disk->fops->open) {
1084 ret = disk->fops->open(bdev, mode);
1085 if (ret == -ERESTARTSYS) {
1086 /* Lost a race with 'disk' being
1087 * deleted, try again.
1090 disk_put_part(bdev->bd_part);
1091 bdev->bd_part = NULL;
1092 bdev->bd_disk = NULL;
1093 bdev->bd_queue = NULL;
1094 mutex_unlock(&bdev->bd_mutex);
1095 disk_unblock_events(disk);
1103 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1104 bdi = blk_get_backing_dev_info(bdev);
1106 bdi = &default_backing_dev_info;
1107 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1111 * If the device is invalidated, rescan partition
1112 * if open succeeded or failed with -ENOMEDIUM.
1113 * The latter is necessary to prevent ghost
1114 * partitions on a removed medium.
1116 if (bdev->bd_invalidated) {
1118 rescan_partitions(disk, bdev);
1119 else if (ret == -ENOMEDIUM)
1120 invalidate_partitions(disk, bdev);
1125 struct block_device *whole;
1126 whole = bdget_disk(disk, 0);
1131 ret = __blkdev_get(whole, mode, 1);
1134 bdev->bd_contains = whole;
1135 bdev_inode_switch_bdi(bdev->bd_inode,
1136 whole->bd_inode->i_data.backing_dev_info);
1137 bdev->bd_part = disk_get_part(disk, partno);
1138 if (!(disk->flags & GENHD_FL_UP) ||
1139 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1143 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1146 if (bdev->bd_contains == bdev) {
1148 if (bdev->bd_disk->fops->open)
1149 ret = bdev->bd_disk->fops->open(bdev, mode);
1150 /* the same as first opener case, read comment there */
1151 if (bdev->bd_invalidated) {
1153 rescan_partitions(bdev->bd_disk, bdev);
1154 else if (ret == -ENOMEDIUM)
1155 invalidate_partitions(bdev->bd_disk, bdev);
1158 goto out_unlock_bdev;
1160 /* only one opener holds refs to the module and disk */
1166 bdev->bd_part_count++;
1167 mutex_unlock(&bdev->bd_mutex);
1168 disk_unblock_events(disk);
1172 disk_put_part(bdev->bd_part);
1173 bdev->bd_disk = NULL;
1174 bdev->bd_part = NULL;
1175 bdev->bd_queue = NULL;
1176 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1177 if (bdev != bdev->bd_contains)
1178 __blkdev_put(bdev->bd_contains, mode, 1);
1179 bdev->bd_contains = NULL;
1181 mutex_unlock(&bdev->bd_mutex);
1182 disk_unblock_events(disk);
1192 * blkdev_get - open a block device
1193 * @bdev: block_device to open
1194 * @mode: FMODE_* mask
1195 * @holder: exclusive holder identifier
1197 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1198 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1199 * @holder is invalid. Exclusive opens may nest for the same @holder.
1201 * On success, the reference count of @bdev is unchanged. On failure,
1208 * 0 on success, -errno on failure.
1210 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1212 struct block_device *whole = NULL;
1215 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1217 if ((mode & FMODE_EXCL) && holder) {
1218 whole = bd_start_claiming(bdev, holder);
1219 if (IS_ERR(whole)) {
1221 return PTR_ERR(whole);
1225 res = __blkdev_get(bdev, mode, 0);
1228 struct gendisk *disk = whole->bd_disk;
1230 /* finish claiming */
1231 mutex_lock(&bdev->bd_mutex);
1232 spin_lock(&bdev_lock);
1235 BUG_ON(!bd_may_claim(bdev, whole, holder));
1237 * Note that for a whole device bd_holders
1238 * will be incremented twice, and bd_holder
1239 * will be set to bd_may_claim before being
1242 whole->bd_holders++;
1243 whole->bd_holder = bd_may_claim;
1245 bdev->bd_holder = holder;
1248 /* tell others that we're done */
1249 BUG_ON(whole->bd_claiming != holder);
1250 whole->bd_claiming = NULL;
1251 wake_up_bit(&whole->bd_claiming, 0);
1253 spin_unlock(&bdev_lock);
1256 * Block event polling for write claims if requested. Any
1257 * write holder makes the write_holder state stick until
1258 * all are released. This is good enough and tracking
1259 * individual writeable reference is too fragile given the
1260 * way @mode is used in blkdev_get/put().
1262 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1263 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1264 bdev->bd_write_holder = true;
1265 disk_block_events(disk);
1268 mutex_unlock(&bdev->bd_mutex);
1274 EXPORT_SYMBOL(blkdev_get);
1277 * blkdev_get_by_path - open a block device by name
1278 * @path: path to the block device to open
1279 * @mode: FMODE_* mask
1280 * @holder: exclusive holder identifier
1282 * Open the blockdevice described by the device file at @path. @mode
1283 * and @holder are identical to blkdev_get().
1285 * On success, the returned block_device has reference count of one.
1291 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1293 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1296 struct block_device *bdev;
1299 bdev = lookup_bdev(path);
1303 err = blkdev_get(bdev, mode, holder);
1305 return ERR_PTR(err);
1307 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1308 blkdev_put(bdev, mode);
1309 return ERR_PTR(-EACCES);
1314 EXPORT_SYMBOL(blkdev_get_by_path);
1317 * blkdev_get_by_dev - open a block device by device number
1318 * @dev: device number of block device to open
1319 * @mode: FMODE_* mask
1320 * @holder: exclusive holder identifier
1322 * Open the blockdevice described by device number @dev. @mode and
1323 * @holder are identical to blkdev_get().
1325 * Use it ONLY if you really do not have anything better - i.e. when
1326 * you are behind a truly sucky interface and all you are given is a
1327 * device number. _Never_ to be used for internal purposes. If you
1328 * ever need it - reconsider your API.
1330 * On success, the returned block_device has reference count of one.
1336 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1338 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1340 struct block_device *bdev;
1345 return ERR_PTR(-ENOMEM);
1347 err = blkdev_get(bdev, mode, holder);
1349 return ERR_PTR(err);
1353 EXPORT_SYMBOL(blkdev_get_by_dev);
1355 static int blkdev_open(struct inode * inode, struct file * filp)
1357 struct block_device *bdev;
1360 * Preserve backwards compatibility and allow large file access
1361 * even if userspace doesn't ask for it explicitly. Some mkfs
1362 * binary needs it. We might want to drop this workaround
1363 * during an unstable branch.
1365 filp->f_flags |= O_LARGEFILE;
1367 if (filp->f_flags & O_NDELAY)
1368 filp->f_mode |= FMODE_NDELAY;
1369 if (filp->f_flags & O_EXCL)
1370 filp->f_mode |= FMODE_EXCL;
1371 if ((filp->f_flags & O_ACCMODE) == 3)
1372 filp->f_mode |= FMODE_WRITE_IOCTL;
1374 bdev = bd_acquire(inode);
1378 filp->f_mapping = bdev->bd_inode->i_mapping;
1380 return blkdev_get(bdev, filp->f_mode, filp);
1383 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1385 struct gendisk *disk = bdev->bd_disk;
1386 struct block_device *victim = NULL;
1388 mutex_lock_nested(&bdev->bd_mutex, for_part);
1390 bdev->bd_part_count--;
1392 if (!--bdev->bd_openers) {
1393 WARN_ON_ONCE(bdev->bd_holders);
1394 sync_blockdev(bdev);
1396 /* ->release can cause the old bdi to disappear,
1397 * so must switch it out first
1399 bdev_inode_switch_bdi(bdev->bd_inode,
1400 &default_backing_dev_info);
1402 if (bdev->bd_contains == bdev) {
1403 if (disk->fops->release)
1404 disk->fops->release(disk, mode);
1406 if (!bdev->bd_openers) {
1407 struct module *owner = disk->fops->owner;
1409 disk_put_part(bdev->bd_part);
1410 bdev->bd_part = NULL;
1411 bdev->bd_disk = NULL;
1412 if (bdev != bdev->bd_contains)
1413 victim = bdev->bd_contains;
1414 bdev->bd_contains = NULL;
1419 mutex_unlock(&bdev->bd_mutex);
1422 __blkdev_put(victim, mode, 1);
1425 void blkdev_put(struct block_device *bdev, fmode_t mode)
1427 mutex_lock(&bdev->bd_mutex);
1429 if (mode & FMODE_EXCL) {
1433 * Release a claim on the device. The holder fields
1434 * are protected with bdev_lock. bd_mutex is to
1435 * synchronize disk_holder unlinking.
1437 spin_lock(&bdev_lock);
1439 WARN_ON_ONCE(--bdev->bd_holders < 0);
1440 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1442 /* bd_contains might point to self, check in a separate step */
1443 if ((bdev_free = !bdev->bd_holders))
1444 bdev->bd_holder = NULL;
1445 if (!bdev->bd_contains->bd_holders)
1446 bdev->bd_contains->bd_holder = NULL;
1448 spin_unlock(&bdev_lock);
1451 * If this was the last claim, remove holder link and
1452 * unblock evpoll if it was a write holder.
1454 if (bdev_free && bdev->bd_write_holder) {
1455 disk_unblock_events(bdev->bd_disk);
1456 bdev->bd_write_holder = false;
1461 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1462 * event. This is to ensure detection of media removal commanded
1463 * from userland - e.g. eject(1).
1465 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1467 mutex_unlock(&bdev->bd_mutex);
1469 __blkdev_put(bdev, mode, 0);
1471 EXPORT_SYMBOL(blkdev_put);
1473 static int blkdev_close(struct inode * inode, struct file * filp)
1475 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1476 blkdev_put(bdev, filp->f_mode);
1480 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1482 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1483 fmode_t mode = file->f_mode;
1486 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1487 * to updated it before every ioctl.
1489 if (file->f_flags & O_NDELAY)
1490 mode |= FMODE_NDELAY;
1492 mode &= ~FMODE_NDELAY;
1494 return blkdev_ioctl(bdev, mode, cmd, arg);
1498 * Write data to the block device. Only intended for the block device itself
1499 * and the raw driver which basically is a fake block device.
1501 * Does not take i_mutex for the write and thus is not for general purpose
1504 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1505 unsigned long nr_segs, loff_t pos)
1507 struct file *file = iocb->ki_filp;
1508 struct blk_plug plug;
1511 BUG_ON(iocb->ki_pos != pos);
1513 blk_start_plug(&plug);
1514 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1515 if (ret > 0 || ret == -EIOCBQUEUED) {
1518 err = generic_write_sync(file, pos, ret);
1519 if (err < 0 && ret > 0)
1522 blk_finish_plug(&plug);
1525 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1527 static ssize_t blkdev_aio_read(struct kiocb *iocb, const struct iovec *iov,
1528 unsigned long nr_segs, loff_t pos)
1530 struct file *file = iocb->ki_filp;
1531 struct inode *bd_inode = file->f_mapping->host;
1532 loff_t size = i_size_read(bd_inode);
1538 if (size < iocb->ki_left)
1539 nr_segs = iov_shorten((struct iovec *)iov, nr_segs, size);
1540 return generic_file_aio_read(iocb, iov, nr_segs, pos);
1544 * Try to release a page associated with block device when the system
1545 * is under memory pressure.
1547 static int blkdev_releasepage(struct page *page, gfp_t wait)
1549 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1551 if (super && super->s_op->bdev_try_to_free_page)
1552 return super->s_op->bdev_try_to_free_page(super, page, wait);
1554 return try_to_free_buffers(page);
1557 static const struct address_space_operations def_blk_aops = {
1558 .readpage = blkdev_readpage,
1559 .writepage = blkdev_writepage,
1560 .write_begin = blkdev_write_begin,
1561 .write_end = blkdev_write_end,
1562 .writepages = generic_writepages,
1563 .releasepage = blkdev_releasepage,
1564 .direct_IO = blkdev_direct_IO,
1567 const struct file_operations def_blk_fops = {
1568 .open = blkdev_open,
1569 .release = blkdev_close,
1570 .llseek = block_llseek,
1571 .read = do_sync_read,
1572 .write = do_sync_write,
1573 .aio_read = blkdev_aio_read,
1574 .aio_write = blkdev_aio_write,
1575 .mmap = generic_file_mmap,
1576 .fsync = blkdev_fsync,
1577 .unlocked_ioctl = block_ioctl,
1578 #ifdef CONFIG_COMPAT
1579 .compat_ioctl = compat_blkdev_ioctl,
1581 .splice_read = generic_file_splice_read,
1582 .splice_write = generic_file_splice_write,
1585 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1588 mm_segment_t old_fs = get_fs();
1590 res = blkdev_ioctl(bdev, 0, cmd, arg);
1595 EXPORT_SYMBOL(ioctl_by_bdev);
1598 * lookup_bdev - lookup a struct block_device by name
1599 * @pathname: special file representing the block device
1601 * Get a reference to the blockdevice at @pathname in the current
1602 * namespace if possible and return it. Return ERR_PTR(error)
1605 struct block_device *lookup_bdev(const char *pathname)
1607 struct block_device *bdev;
1608 struct inode *inode;
1612 if (!pathname || !*pathname)
1613 return ERR_PTR(-EINVAL);
1615 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1617 return ERR_PTR(error);
1619 inode = path.dentry->d_inode;
1621 if (!S_ISBLK(inode->i_mode))
1624 if (path.mnt->mnt_flags & MNT_NODEV)
1627 bdev = bd_acquire(inode);
1634 bdev = ERR_PTR(error);
1637 EXPORT_SYMBOL(lookup_bdev);
1639 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1641 struct super_block *sb = get_super(bdev);
1646 * no need to lock the super, get_super holds the
1647 * read mutex so the filesystem cannot go away
1648 * under us (->put_super runs with the write lock
1651 shrink_dcache_sb(sb);
1652 res = invalidate_inodes(sb, kill_dirty);
1655 invalidate_bdev(bdev);
1658 EXPORT_SYMBOL(__invalidate_device);
1660 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1662 struct inode *inode, *old_inode = NULL;
1664 spin_lock(&inode_sb_list_lock);
1665 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1666 struct address_space *mapping = inode->i_mapping;
1668 spin_lock(&inode->i_lock);
1669 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1670 mapping->nrpages == 0) {
1671 spin_unlock(&inode->i_lock);
1675 spin_unlock(&inode->i_lock);
1676 spin_unlock(&inode_sb_list_lock);
1678 * We hold a reference to 'inode' so it couldn't have been
1679 * removed from s_inodes list while we dropped the
1680 * inode_sb_list_lock. We cannot iput the inode now as we can
1681 * be holding the last reference and we cannot iput it under
1682 * inode_sb_list_lock. So we keep the reference and iput it
1688 func(I_BDEV(inode), arg);
1690 spin_lock(&inode_sb_list_lock);
1692 spin_unlock(&inode_sb_list_lock);