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. Make sure the inode
54 * is clean before moving so that it doesn't linger on the old bdi.
56 static void bdev_inode_switch_bdi(struct inode *inode,
57 struct backing_dev_info *dst)
60 spin_lock(&inode->i_lock);
61 if (!(inode->i_state & I_DIRTY)) {
62 inode->i_data.backing_dev_info = dst;
63 spin_unlock(&inode->i_lock);
66 spin_unlock(&inode->i_lock);
67 WARN_ON_ONCE(write_inode_now(inode, true));
71 /* Kill _all_ buffers and pagecache , dirty or not.. */
72 void kill_bdev(struct block_device *bdev)
74 struct address_space *mapping = bdev->bd_inode->i_mapping;
76 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
80 truncate_inode_pages(mapping, 0);
82 EXPORT_SYMBOL(kill_bdev);
84 /* Invalidate clean unused buffers and pagecache. */
85 void invalidate_bdev(struct block_device *bdev)
87 struct address_space *mapping = bdev->bd_inode->i_mapping;
89 if (mapping->nrpages == 0)
93 lru_add_drain_all(); /* make sure all lru add caches are flushed */
94 invalidate_mapping_pages(mapping, 0, -1);
95 /* 99% of the time, we don't need to flush the cleancache on the bdev.
96 * But, for the strange corners, lets be cautious
98 cleancache_invalidate_inode(mapping);
100 EXPORT_SYMBOL(invalidate_bdev);
102 int set_blocksize(struct block_device *bdev, int size)
104 /* Size must be a power of two, and between 512 and PAGE_SIZE */
105 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
108 /* Size cannot be smaller than the size supported by the device */
109 if (size < bdev_logical_block_size(bdev))
112 /* Don't change the size if it is same as current */
113 if (bdev->bd_block_size != size) {
115 bdev->bd_block_size = size;
116 bdev->bd_inode->i_blkbits = blksize_bits(size);
122 EXPORT_SYMBOL(set_blocksize);
124 int sb_set_blocksize(struct super_block *sb, int size)
126 if (set_blocksize(sb->s_bdev, size))
128 /* If we get here, we know size is power of two
129 * and it's value is between 512 and PAGE_SIZE */
130 sb->s_blocksize = size;
131 sb->s_blocksize_bits = blksize_bits(size);
132 return sb->s_blocksize;
135 EXPORT_SYMBOL(sb_set_blocksize);
137 int sb_min_blocksize(struct super_block *sb, int size)
139 int minsize = bdev_logical_block_size(sb->s_bdev);
142 return sb_set_blocksize(sb, size);
145 EXPORT_SYMBOL(sb_min_blocksize);
148 blkdev_get_block(struct inode *inode, sector_t iblock,
149 struct buffer_head *bh, int create)
151 bh->b_bdev = I_BDEV(inode);
152 bh->b_blocknr = iblock;
153 set_buffer_mapped(bh);
158 blkdev_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter,
161 struct file *file = iocb->ki_filp;
162 struct inode *inode = file->f_mapping->host;
164 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iter,
165 offset, blkdev_get_block,
169 int __sync_blockdev(struct block_device *bdev, int wait)
174 return filemap_flush(bdev->bd_inode->i_mapping);
175 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
179 * Write out and wait upon all the dirty data associated with a block
180 * device via its mapping. Does not take the superblock lock.
182 int sync_blockdev(struct block_device *bdev)
184 return __sync_blockdev(bdev, 1);
186 EXPORT_SYMBOL(sync_blockdev);
189 * Write out and wait upon all dirty data associated with this
190 * device. Filesystem data as well as the underlying block
191 * device. Takes the superblock lock.
193 int fsync_bdev(struct block_device *bdev)
195 struct super_block *sb = get_super(bdev);
197 int res = sync_filesystem(sb);
201 return sync_blockdev(bdev);
203 EXPORT_SYMBOL(fsync_bdev);
206 * freeze_bdev -- lock a filesystem and force it into a consistent state
207 * @bdev: blockdevice to lock
209 * If a superblock is found on this device, we take the s_umount semaphore
210 * on it to make sure nobody unmounts until the snapshot creation is done.
211 * The reference counter (bd_fsfreeze_count) guarantees that only the last
212 * unfreeze process can unfreeze the frozen filesystem actually when multiple
213 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
214 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
217 struct super_block *freeze_bdev(struct block_device *bdev)
219 struct super_block *sb;
222 mutex_lock(&bdev->bd_fsfreeze_mutex);
223 if (++bdev->bd_fsfreeze_count > 1) {
225 * We don't even need to grab a reference - the first call
226 * to freeze_bdev grab an active reference and only the last
227 * thaw_bdev drops it.
229 sb = get_super(bdev);
231 mutex_unlock(&bdev->bd_fsfreeze_mutex);
235 sb = get_active_super(bdev);
238 error = freeze_super(sb);
240 deactivate_super(sb);
241 bdev->bd_fsfreeze_count--;
242 mutex_unlock(&bdev->bd_fsfreeze_mutex);
243 return ERR_PTR(error);
245 deactivate_super(sb);
248 mutex_unlock(&bdev->bd_fsfreeze_mutex);
249 return sb; /* thaw_bdev releases s->s_umount */
251 EXPORT_SYMBOL(freeze_bdev);
254 * thaw_bdev -- unlock filesystem
255 * @bdev: blockdevice to unlock
256 * @sb: associated superblock
258 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
260 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
264 mutex_lock(&bdev->bd_fsfreeze_mutex);
265 if (!bdev->bd_fsfreeze_count)
269 if (--bdev->bd_fsfreeze_count > 0)
275 error = thaw_super(sb);
277 bdev->bd_fsfreeze_count++;
278 mutex_unlock(&bdev->bd_fsfreeze_mutex);
282 mutex_unlock(&bdev->bd_fsfreeze_mutex);
285 EXPORT_SYMBOL(thaw_bdev);
287 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
289 return block_write_full_page(page, blkdev_get_block, wbc);
292 static int blkdev_readpage(struct file * file, struct page * page)
294 return block_read_full_page(page, blkdev_get_block);
297 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
298 loff_t pos, unsigned len, unsigned flags,
299 struct page **pagep, void **fsdata)
301 return block_write_begin(mapping, pos, len, flags, pagep,
305 static int blkdev_write_end(struct file *file, struct address_space *mapping,
306 loff_t pos, unsigned len, unsigned copied,
307 struct page *page, void *fsdata)
310 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
313 page_cache_release(page);
320 * for a block special file file_inode(file)->i_size is zero
321 * so we compute the size by hand (just as in block_read/write above)
323 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
325 struct inode *bd_inode = file->f_mapping->host;
328 mutex_lock(&bd_inode->i_mutex);
329 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
330 mutex_unlock(&bd_inode->i_mutex);
334 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
336 struct inode *bd_inode = filp->f_mapping->host;
337 struct block_device *bdev = I_BDEV(bd_inode);
340 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
345 * There is no need to serialise calls to blkdev_issue_flush with
346 * i_mutex and doing so causes performance issues with concurrent
347 * O_SYNC writers to a block device.
349 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
350 if (error == -EOPNOTSUPP)
355 EXPORT_SYMBOL(blkdev_fsync);
358 * bdev_read_page() - Start reading a page from a block device
359 * @bdev: The device to read the page from
360 * @sector: The offset on the device to read the page to (need not be aligned)
361 * @page: The page to read
363 * On entry, the page should be locked. It will be unlocked when the page
364 * has been read. If the block driver implements rw_page synchronously,
365 * that will be true on exit from this function, but it need not be.
367 * Errors returned by this function are usually "soft", eg out of memory, or
368 * queue full; callers should try a different route to read this page rather
369 * than propagate an error back up the stack.
371 * Return: negative errno if an error occurs, 0 if submission was successful.
373 int bdev_read_page(struct block_device *bdev, sector_t sector,
376 const struct block_device_operations *ops = bdev->bd_disk->fops;
379 return ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
381 EXPORT_SYMBOL_GPL(bdev_read_page);
384 * bdev_write_page() - Start writing a page to a block device
385 * @bdev: The device to write the page to
386 * @sector: The offset on the device to write the page to (need not be aligned)
387 * @page: The page to write
388 * @wbc: The writeback_control for the write
390 * On entry, the page should be locked and not currently under writeback.
391 * On exit, if the write started successfully, the page will be unlocked and
392 * under writeback. If the write failed already (eg the driver failed to
393 * queue the page to the device), the page will still be locked. If the
394 * caller is a ->writepage implementation, it will need to unlock the page.
396 * Errors returned by this function are usually "soft", eg out of memory, or
397 * queue full; callers should try a different route to write this page rather
398 * than propagate an error back up the stack.
400 * Return: negative errno if an error occurs, 0 if submission was successful.
402 int bdev_write_page(struct block_device *bdev, sector_t sector,
403 struct page *page, struct writeback_control *wbc)
406 int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
407 const struct block_device_operations *ops = bdev->bd_disk->fops;
410 set_page_writeback(page);
411 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
413 end_page_writeback(page);
418 EXPORT_SYMBOL_GPL(bdev_write_page);
424 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
425 static struct kmem_cache * bdev_cachep __read_mostly;
427 static struct inode *bdev_alloc_inode(struct super_block *sb)
429 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
432 return &ei->vfs_inode;
435 static void bdev_i_callback(struct rcu_head *head)
437 struct inode *inode = container_of(head, struct inode, i_rcu);
438 struct bdev_inode *bdi = BDEV_I(inode);
440 kmem_cache_free(bdev_cachep, bdi);
443 static void bdev_destroy_inode(struct inode *inode)
445 call_rcu(&inode->i_rcu, bdev_i_callback);
448 static void init_once(void *foo)
450 struct bdev_inode *ei = (struct bdev_inode *) foo;
451 struct block_device *bdev = &ei->bdev;
453 memset(bdev, 0, sizeof(*bdev));
454 mutex_init(&bdev->bd_mutex);
455 INIT_LIST_HEAD(&bdev->bd_inodes);
456 INIT_LIST_HEAD(&bdev->bd_list);
458 INIT_LIST_HEAD(&bdev->bd_holder_disks);
460 inode_init_once(&ei->vfs_inode);
461 /* Initialize mutex for freeze. */
462 mutex_init(&bdev->bd_fsfreeze_mutex);
465 static inline void __bd_forget(struct inode *inode)
467 list_del_init(&inode->i_devices);
468 inode->i_bdev = NULL;
469 inode->i_mapping = &inode->i_data;
472 static void bdev_evict_inode(struct inode *inode)
474 struct block_device *bdev = &BDEV_I(inode)->bdev;
476 truncate_inode_pages_final(&inode->i_data);
477 invalidate_inode_buffers(inode); /* is it needed here? */
479 spin_lock(&bdev_lock);
480 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
481 __bd_forget(list_entry(p, struct inode, i_devices));
483 list_del_init(&bdev->bd_list);
484 spin_unlock(&bdev_lock);
487 static const struct super_operations bdev_sops = {
488 .statfs = simple_statfs,
489 .alloc_inode = bdev_alloc_inode,
490 .destroy_inode = bdev_destroy_inode,
491 .drop_inode = generic_delete_inode,
492 .evict_inode = bdev_evict_inode,
495 static struct dentry *bd_mount(struct file_system_type *fs_type,
496 int flags, const char *dev_name, void *data)
498 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
501 static struct file_system_type bd_type = {
504 .kill_sb = kill_anon_super,
507 static struct super_block *blockdev_superblock __read_mostly;
509 void __init bdev_cache_init(void)
512 static struct vfsmount *bd_mnt;
514 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
515 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
516 SLAB_MEM_SPREAD|SLAB_PANIC),
518 err = register_filesystem(&bd_type);
520 panic("Cannot register bdev pseudo-fs");
521 bd_mnt = kern_mount(&bd_type);
523 panic("Cannot create bdev pseudo-fs");
524 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
528 * Most likely _very_ bad one - but then it's hardly critical for small
529 * /dev and can be fixed when somebody will need really large one.
530 * Keep in mind that it will be fed through icache hash function too.
532 static inline unsigned long hash(dev_t dev)
534 return MAJOR(dev)+MINOR(dev);
537 static int bdev_test(struct inode *inode, void *data)
539 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
542 static int bdev_set(struct inode *inode, void *data)
544 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
548 static LIST_HEAD(all_bdevs);
550 struct block_device *bdget(dev_t dev)
552 struct block_device *bdev;
555 inode = iget5_locked(blockdev_superblock, hash(dev),
556 bdev_test, bdev_set, &dev);
561 bdev = &BDEV_I(inode)->bdev;
563 if (inode->i_state & I_NEW) {
564 bdev->bd_contains = NULL;
565 bdev->bd_super = NULL;
566 bdev->bd_inode = inode;
567 bdev->bd_block_size = (1 << inode->i_blkbits);
568 bdev->bd_part_count = 0;
569 bdev->bd_invalidated = 0;
570 inode->i_mode = S_IFBLK;
572 inode->i_bdev = bdev;
573 inode->i_data.a_ops = &def_blk_aops;
574 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
575 inode->i_data.backing_dev_info = &default_backing_dev_info;
576 spin_lock(&bdev_lock);
577 list_add(&bdev->bd_list, &all_bdevs);
578 spin_unlock(&bdev_lock);
579 unlock_new_inode(inode);
584 EXPORT_SYMBOL(bdget);
587 * bdgrab -- Grab a reference to an already referenced block device
588 * @bdev: Block device to grab a reference to.
590 struct block_device *bdgrab(struct block_device *bdev)
592 ihold(bdev->bd_inode);
595 EXPORT_SYMBOL(bdgrab);
597 long nr_blockdev_pages(void)
599 struct block_device *bdev;
601 spin_lock(&bdev_lock);
602 list_for_each_entry(bdev, &all_bdevs, bd_list) {
603 ret += bdev->bd_inode->i_mapping->nrpages;
605 spin_unlock(&bdev_lock);
609 void bdput(struct block_device *bdev)
611 iput(bdev->bd_inode);
614 EXPORT_SYMBOL(bdput);
616 static struct block_device *bd_acquire(struct inode *inode)
618 struct block_device *bdev;
620 spin_lock(&bdev_lock);
621 bdev = inode->i_bdev;
623 ihold(bdev->bd_inode);
624 spin_unlock(&bdev_lock);
627 spin_unlock(&bdev_lock);
629 bdev = bdget(inode->i_rdev);
631 spin_lock(&bdev_lock);
632 if (!inode->i_bdev) {
634 * We take an additional reference to bd_inode,
635 * and it's released in clear_inode() of inode.
636 * So, we can access it via ->i_mapping always
639 ihold(bdev->bd_inode);
640 inode->i_bdev = bdev;
641 inode->i_mapping = bdev->bd_inode->i_mapping;
642 list_add(&inode->i_devices, &bdev->bd_inodes);
644 spin_unlock(&bdev_lock);
649 int sb_is_blkdev_sb(struct super_block *sb)
651 return sb == blockdev_superblock;
654 /* Call when you free inode */
656 void bd_forget(struct inode *inode)
658 struct block_device *bdev = NULL;
660 spin_lock(&bdev_lock);
661 if (!sb_is_blkdev_sb(inode->i_sb))
662 bdev = inode->i_bdev;
664 spin_unlock(&bdev_lock);
667 iput(bdev->bd_inode);
671 * bd_may_claim - test whether a block device can be claimed
672 * @bdev: block device of interest
673 * @whole: whole block device containing @bdev, may equal @bdev
674 * @holder: holder trying to claim @bdev
676 * Test whether @bdev can be claimed by @holder.
679 * spin_lock(&bdev_lock).
682 * %true if @bdev can be claimed, %false otherwise.
684 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
687 if (bdev->bd_holder == holder)
688 return true; /* already a holder */
689 else if (bdev->bd_holder != NULL)
690 return false; /* held by someone else */
691 else if (bdev->bd_contains == bdev)
692 return true; /* is a whole device which isn't held */
694 else if (whole->bd_holder == bd_may_claim)
695 return true; /* is a partition of a device that is being partitioned */
696 else if (whole->bd_holder != NULL)
697 return false; /* is a partition of a held device */
699 return true; /* is a partition of an un-held device */
703 * bd_prepare_to_claim - prepare to claim a block device
704 * @bdev: block device of interest
705 * @whole: the whole device containing @bdev, may equal @bdev
706 * @holder: holder trying to claim @bdev
708 * Prepare to claim @bdev. This function fails if @bdev is already
709 * claimed by another holder and waits if another claiming is in
710 * progress. This function doesn't actually claim. On successful
711 * return, the caller has ownership of bd_claiming and bd_holder[s].
714 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
718 * 0 if @bdev can be claimed, -EBUSY otherwise.
720 static int bd_prepare_to_claim(struct block_device *bdev,
721 struct block_device *whole, void *holder)
724 /* if someone else claimed, fail */
725 if (!bd_may_claim(bdev, whole, holder))
728 /* if claiming is already in progress, wait for it to finish */
729 if (whole->bd_claiming) {
730 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
733 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
734 spin_unlock(&bdev_lock);
736 finish_wait(wq, &wait);
737 spin_lock(&bdev_lock);
746 * bd_start_claiming - start claiming a block device
747 * @bdev: block device of interest
748 * @holder: holder trying to claim @bdev
750 * @bdev is about to be opened exclusively. Check @bdev can be opened
751 * exclusively and mark that an exclusive open is in progress. Each
752 * successful call to this function must be matched with a call to
753 * either bd_finish_claiming() or bd_abort_claiming() (which do not
756 * This function is used to gain exclusive access to the block device
757 * without actually causing other exclusive open attempts to fail. It
758 * should be used when the open sequence itself requires exclusive
759 * access but may subsequently fail.
765 * Pointer to the block device containing @bdev on success, ERR_PTR()
768 static struct block_device *bd_start_claiming(struct block_device *bdev,
771 struct gendisk *disk;
772 struct block_device *whole;
778 * @bdev might not have been initialized properly yet, look up
779 * and grab the outer block device the hard way.
781 disk = get_gendisk(bdev->bd_dev, &partno);
783 return ERR_PTR(-ENXIO);
786 * Normally, @bdev should equal what's returned from bdget_disk()
787 * if partno is 0; however, some drivers (floppy) use multiple
788 * bdev's for the same physical device and @bdev may be one of the
789 * aliases. Keep @bdev if partno is 0. This means claimer
790 * tracking is broken for those devices but it has always been that
794 whole = bdget_disk(disk, 0);
796 whole = bdgrab(bdev);
798 module_put(disk->fops->owner);
801 return ERR_PTR(-ENOMEM);
803 /* prepare to claim, if successful, mark claiming in progress */
804 spin_lock(&bdev_lock);
806 err = bd_prepare_to_claim(bdev, whole, holder);
808 whole->bd_claiming = holder;
809 spin_unlock(&bdev_lock);
812 spin_unlock(&bdev_lock);
819 struct bd_holder_disk {
820 struct list_head list;
821 struct gendisk *disk;
825 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
826 struct gendisk *disk)
828 struct bd_holder_disk *holder;
830 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
831 if (holder->disk == disk)
836 static int add_symlink(struct kobject *from, struct kobject *to)
838 return sysfs_create_link(from, to, kobject_name(to));
841 static void del_symlink(struct kobject *from, struct kobject *to)
843 sysfs_remove_link(from, kobject_name(to));
847 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
848 * @bdev: the claimed slave bdev
849 * @disk: the holding disk
851 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
853 * This functions creates the following sysfs symlinks.
855 * - from "slaves" directory of the holder @disk to the claimed @bdev
856 * - from "holders" directory of the @bdev to the holder @disk
858 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
859 * passed to bd_link_disk_holder(), then:
861 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
862 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
864 * The caller must have claimed @bdev before calling this function and
865 * ensure that both @bdev and @disk are valid during the creation and
866 * lifetime of these symlinks.
872 * 0 on success, -errno on failure.
874 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
876 struct bd_holder_disk *holder;
879 mutex_lock(&bdev->bd_mutex);
881 WARN_ON_ONCE(!bdev->bd_holder);
883 /* FIXME: remove the following once add_disk() handles errors */
884 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
887 holder = bd_find_holder_disk(bdev, disk);
893 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
899 INIT_LIST_HEAD(&holder->list);
903 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
907 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
911 * bdev could be deleted beneath us which would implicitly destroy
912 * the holder directory. Hold on to it.
914 kobject_get(bdev->bd_part->holder_dir);
916 list_add(&holder->list, &bdev->bd_holder_disks);
920 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
924 mutex_unlock(&bdev->bd_mutex);
927 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
930 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
931 * @bdev: the calimed slave bdev
932 * @disk: the holding disk
934 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
939 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
941 struct bd_holder_disk *holder;
943 mutex_lock(&bdev->bd_mutex);
945 holder = bd_find_holder_disk(bdev, disk);
947 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
948 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
949 del_symlink(bdev->bd_part->holder_dir,
950 &disk_to_dev(disk)->kobj);
951 kobject_put(bdev->bd_part->holder_dir);
952 list_del_init(&holder->list);
956 mutex_unlock(&bdev->bd_mutex);
958 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
962 * flush_disk - invalidates all buffer-cache entries on a disk
964 * @bdev: struct block device to be flushed
965 * @kill_dirty: flag to guide handling of dirty inodes
967 * Invalidates all buffer-cache entries on a disk. It should be called
968 * when a disk has been changed -- either by a media change or online
971 static void flush_disk(struct block_device *bdev, bool kill_dirty)
973 if (__invalidate_device(bdev, kill_dirty)) {
974 char name[BDEVNAME_SIZE] = "";
977 disk_name(bdev->bd_disk, 0, name);
978 printk(KERN_WARNING "VFS: busy inodes on changed media or "
979 "resized disk %s\n", name);
984 if (disk_part_scan_enabled(bdev->bd_disk))
985 bdev->bd_invalidated = 1;
989 * check_disk_size_change - checks for disk size change and adjusts bdev size.
990 * @disk: struct gendisk to check
991 * @bdev: struct bdev to adjust.
993 * This routine checks to see if the bdev size does not match the disk size
994 * and adjusts it if it differs.
996 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
998 loff_t disk_size, bdev_size;
1000 disk_size = (loff_t)get_capacity(disk) << 9;
1001 bdev_size = i_size_read(bdev->bd_inode);
1002 if (disk_size != bdev_size) {
1003 char name[BDEVNAME_SIZE];
1005 disk_name(disk, 0, name);
1007 "%s: detected capacity change from %lld to %lld\n",
1008 name, bdev_size, disk_size);
1009 i_size_write(bdev->bd_inode, disk_size);
1010 flush_disk(bdev, false);
1013 EXPORT_SYMBOL(check_disk_size_change);
1016 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1017 * @disk: struct gendisk to be revalidated
1019 * This routine is a wrapper for lower-level driver's revalidate_disk
1020 * call-backs. It is used to do common pre and post operations needed
1021 * for all revalidate_disk operations.
1023 int revalidate_disk(struct gendisk *disk)
1025 struct block_device *bdev;
1028 if (disk->fops->revalidate_disk)
1029 ret = disk->fops->revalidate_disk(disk);
1031 bdev = bdget_disk(disk, 0);
1035 mutex_lock(&bdev->bd_mutex);
1036 check_disk_size_change(disk, bdev);
1037 bdev->bd_invalidated = 0;
1038 mutex_unlock(&bdev->bd_mutex);
1042 EXPORT_SYMBOL(revalidate_disk);
1045 * This routine checks whether a removable media has been changed,
1046 * and invalidates all buffer-cache-entries in that case. This
1047 * is a relatively slow routine, so we have to try to minimize using
1048 * it. Thus it is called only upon a 'mount' or 'open'. This
1049 * is the best way of combining speed and utility, I think.
1050 * People changing diskettes in the middle of an operation deserve
1053 int check_disk_change(struct block_device *bdev)
1055 struct gendisk *disk = bdev->bd_disk;
1056 const struct block_device_operations *bdops = disk->fops;
1057 unsigned int events;
1059 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1060 DISK_EVENT_EJECT_REQUEST);
1061 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1064 flush_disk(bdev, true);
1065 if (bdops->revalidate_disk)
1066 bdops->revalidate_disk(bdev->bd_disk);
1070 EXPORT_SYMBOL(check_disk_change);
1072 void bd_set_size(struct block_device *bdev, loff_t size)
1074 unsigned bsize = bdev_logical_block_size(bdev);
1076 mutex_lock(&bdev->bd_inode->i_mutex);
1077 i_size_write(bdev->bd_inode, size);
1078 mutex_unlock(&bdev->bd_inode->i_mutex);
1079 while (bsize < PAGE_CACHE_SIZE) {
1084 bdev->bd_block_size = bsize;
1085 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1087 EXPORT_SYMBOL(bd_set_size);
1089 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1094 * mutex_lock(part->bd_mutex)
1095 * mutex_lock_nested(whole->bd_mutex, 1)
1098 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1100 struct gendisk *disk;
1101 struct module *owner;
1106 if (mode & FMODE_READ)
1108 if (mode & FMODE_WRITE)
1111 * hooks: /n/, see "layering violations".
1114 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1124 disk = get_gendisk(bdev->bd_dev, &partno);
1127 owner = disk->fops->owner;
1129 disk_block_events(disk);
1130 mutex_lock_nested(&bdev->bd_mutex, for_part);
1131 if (!bdev->bd_openers) {
1132 bdev->bd_disk = disk;
1133 bdev->bd_queue = disk->queue;
1134 bdev->bd_contains = bdev;
1136 struct backing_dev_info *bdi;
1139 bdev->bd_part = disk_get_part(disk, partno);
1144 if (disk->fops->open) {
1145 ret = disk->fops->open(bdev, mode);
1146 if (ret == -ERESTARTSYS) {
1147 /* Lost a race with 'disk' being
1148 * deleted, try again.
1151 disk_put_part(bdev->bd_part);
1152 bdev->bd_part = NULL;
1153 bdev->bd_disk = NULL;
1154 bdev->bd_queue = NULL;
1155 mutex_unlock(&bdev->bd_mutex);
1156 disk_unblock_events(disk);
1164 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1165 bdi = blk_get_backing_dev_info(bdev);
1166 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1170 * If the device is invalidated, rescan partition
1171 * if open succeeded or failed with -ENOMEDIUM.
1172 * The latter is necessary to prevent ghost
1173 * partitions on a removed medium.
1175 if (bdev->bd_invalidated) {
1177 rescan_partitions(disk, bdev);
1178 else if (ret == -ENOMEDIUM)
1179 invalidate_partitions(disk, bdev);
1184 struct block_device *whole;
1185 whole = bdget_disk(disk, 0);
1190 ret = __blkdev_get(whole, mode, 1);
1193 bdev->bd_contains = whole;
1194 bdev_inode_switch_bdi(bdev->bd_inode,
1195 whole->bd_inode->i_data.backing_dev_info);
1196 bdev->bd_part = disk_get_part(disk, partno);
1197 if (!(disk->flags & GENHD_FL_UP) ||
1198 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1202 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1205 if (bdev->bd_contains == bdev) {
1207 if (bdev->bd_disk->fops->open)
1208 ret = bdev->bd_disk->fops->open(bdev, mode);
1209 /* the same as first opener case, read comment there */
1210 if (bdev->bd_invalidated) {
1212 rescan_partitions(bdev->bd_disk, bdev);
1213 else if (ret == -ENOMEDIUM)
1214 invalidate_partitions(bdev->bd_disk, bdev);
1217 goto out_unlock_bdev;
1219 /* only one opener holds refs to the module and disk */
1225 bdev->bd_part_count++;
1226 mutex_unlock(&bdev->bd_mutex);
1227 disk_unblock_events(disk);
1231 disk_put_part(bdev->bd_part);
1232 bdev->bd_disk = NULL;
1233 bdev->bd_part = NULL;
1234 bdev->bd_queue = NULL;
1235 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1236 if (bdev != bdev->bd_contains)
1237 __blkdev_put(bdev->bd_contains, mode, 1);
1238 bdev->bd_contains = NULL;
1240 mutex_unlock(&bdev->bd_mutex);
1241 disk_unblock_events(disk);
1251 * blkdev_get - open a block device
1252 * @bdev: block_device to open
1253 * @mode: FMODE_* mask
1254 * @holder: exclusive holder identifier
1256 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1257 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1258 * @holder is invalid. Exclusive opens may nest for the same @holder.
1260 * On success, the reference count of @bdev is unchanged. On failure,
1267 * 0 on success, -errno on failure.
1269 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1271 struct block_device *whole = NULL;
1274 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1276 if ((mode & FMODE_EXCL) && holder) {
1277 whole = bd_start_claiming(bdev, holder);
1278 if (IS_ERR(whole)) {
1280 return PTR_ERR(whole);
1284 res = __blkdev_get(bdev, mode, 0);
1287 struct gendisk *disk = whole->bd_disk;
1289 /* finish claiming */
1290 mutex_lock(&bdev->bd_mutex);
1291 spin_lock(&bdev_lock);
1294 BUG_ON(!bd_may_claim(bdev, whole, holder));
1296 * Note that for a whole device bd_holders
1297 * will be incremented twice, and bd_holder
1298 * will be set to bd_may_claim before being
1301 whole->bd_holders++;
1302 whole->bd_holder = bd_may_claim;
1304 bdev->bd_holder = holder;
1307 /* tell others that we're done */
1308 BUG_ON(whole->bd_claiming != holder);
1309 whole->bd_claiming = NULL;
1310 wake_up_bit(&whole->bd_claiming, 0);
1312 spin_unlock(&bdev_lock);
1315 * Block event polling for write claims if requested. Any
1316 * write holder makes the write_holder state stick until
1317 * all are released. This is good enough and tracking
1318 * individual writeable reference is too fragile given the
1319 * way @mode is used in blkdev_get/put().
1321 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1322 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1323 bdev->bd_write_holder = true;
1324 disk_block_events(disk);
1327 mutex_unlock(&bdev->bd_mutex);
1333 EXPORT_SYMBOL(blkdev_get);
1336 * blkdev_get_by_path - open a block device by name
1337 * @path: path to the block device to open
1338 * @mode: FMODE_* mask
1339 * @holder: exclusive holder identifier
1341 * Open the blockdevice described by the device file at @path. @mode
1342 * and @holder are identical to blkdev_get().
1344 * On success, the returned block_device has reference count of one.
1350 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1352 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1355 struct block_device *bdev;
1358 bdev = lookup_bdev(path);
1362 err = blkdev_get(bdev, mode, holder);
1364 return ERR_PTR(err);
1366 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1367 blkdev_put(bdev, mode);
1368 return ERR_PTR(-EACCES);
1373 EXPORT_SYMBOL(blkdev_get_by_path);
1376 * blkdev_get_by_dev - open a block device by device number
1377 * @dev: device number of block device to open
1378 * @mode: FMODE_* mask
1379 * @holder: exclusive holder identifier
1381 * Open the blockdevice described by device number @dev. @mode and
1382 * @holder are identical to blkdev_get().
1384 * Use it ONLY if you really do not have anything better - i.e. when
1385 * you are behind a truly sucky interface and all you are given is a
1386 * device number. _Never_ to be used for internal purposes. If you
1387 * ever need it - reconsider your API.
1389 * On success, the returned block_device has reference count of one.
1395 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1397 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1399 struct block_device *bdev;
1404 return ERR_PTR(-ENOMEM);
1406 err = blkdev_get(bdev, mode, holder);
1408 return ERR_PTR(err);
1412 EXPORT_SYMBOL(blkdev_get_by_dev);
1414 static int blkdev_open(struct inode * inode, struct file * filp)
1416 struct block_device *bdev;
1419 * Preserve backwards compatibility and allow large file access
1420 * even if userspace doesn't ask for it explicitly. Some mkfs
1421 * binary needs it. We might want to drop this workaround
1422 * during an unstable branch.
1424 filp->f_flags |= O_LARGEFILE;
1426 if (filp->f_flags & O_NDELAY)
1427 filp->f_mode |= FMODE_NDELAY;
1428 if (filp->f_flags & O_EXCL)
1429 filp->f_mode |= FMODE_EXCL;
1430 if ((filp->f_flags & O_ACCMODE) == 3)
1431 filp->f_mode |= FMODE_WRITE_IOCTL;
1433 bdev = bd_acquire(inode);
1437 filp->f_mapping = bdev->bd_inode->i_mapping;
1439 return blkdev_get(bdev, filp->f_mode, filp);
1442 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1444 struct gendisk *disk = bdev->bd_disk;
1445 struct block_device *victim = NULL;
1447 mutex_lock_nested(&bdev->bd_mutex, for_part);
1449 bdev->bd_part_count--;
1451 if (!--bdev->bd_openers) {
1452 WARN_ON_ONCE(bdev->bd_holders);
1453 sync_blockdev(bdev);
1455 /* ->release can cause the old bdi to disappear,
1456 * so must switch it out first
1458 bdev_inode_switch_bdi(bdev->bd_inode,
1459 &default_backing_dev_info);
1461 if (bdev->bd_contains == bdev) {
1462 if (disk->fops->release)
1463 disk->fops->release(disk, mode);
1465 if (!bdev->bd_openers) {
1466 struct module *owner = disk->fops->owner;
1468 disk_put_part(bdev->bd_part);
1469 bdev->bd_part = NULL;
1470 bdev->bd_disk = NULL;
1471 if (bdev != bdev->bd_contains)
1472 victim = bdev->bd_contains;
1473 bdev->bd_contains = NULL;
1478 mutex_unlock(&bdev->bd_mutex);
1481 __blkdev_put(victim, mode, 1);
1484 void blkdev_put(struct block_device *bdev, fmode_t mode)
1486 mutex_lock(&bdev->bd_mutex);
1488 if (mode & FMODE_EXCL) {
1492 * Release a claim on the device. The holder fields
1493 * are protected with bdev_lock. bd_mutex is to
1494 * synchronize disk_holder unlinking.
1496 spin_lock(&bdev_lock);
1498 WARN_ON_ONCE(--bdev->bd_holders < 0);
1499 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1501 /* bd_contains might point to self, check in a separate step */
1502 if ((bdev_free = !bdev->bd_holders))
1503 bdev->bd_holder = NULL;
1504 if (!bdev->bd_contains->bd_holders)
1505 bdev->bd_contains->bd_holder = NULL;
1507 spin_unlock(&bdev_lock);
1510 * If this was the last claim, remove holder link and
1511 * unblock evpoll if it was a write holder.
1513 if (bdev_free && bdev->bd_write_holder) {
1514 disk_unblock_events(bdev->bd_disk);
1515 bdev->bd_write_holder = false;
1520 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1521 * event. This is to ensure detection of media removal commanded
1522 * from userland - e.g. eject(1).
1524 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1526 mutex_unlock(&bdev->bd_mutex);
1528 __blkdev_put(bdev, mode, 0);
1530 EXPORT_SYMBOL(blkdev_put);
1532 static int blkdev_close(struct inode * inode, struct file * filp)
1534 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1535 blkdev_put(bdev, filp->f_mode);
1539 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1541 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1542 fmode_t mode = file->f_mode;
1545 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1546 * to updated it before every ioctl.
1548 if (file->f_flags & O_NDELAY)
1549 mode |= FMODE_NDELAY;
1551 mode &= ~FMODE_NDELAY;
1553 return blkdev_ioctl(bdev, mode, cmd, arg);
1557 * Write data to the block device. Only intended for the block device itself
1558 * and the raw driver which basically is a fake block device.
1560 * Does not take i_mutex for the write and thus is not for general purpose
1563 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1565 struct file *file = iocb->ki_filp;
1566 struct blk_plug plug;
1569 blk_start_plug(&plug);
1570 ret = __generic_file_write_iter(iocb, from);
1573 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
1577 blk_finish_plug(&plug);
1580 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1582 static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1584 struct file *file = iocb->ki_filp;
1585 struct inode *bd_inode = file->f_mapping->host;
1586 loff_t size = i_size_read(bd_inode);
1587 loff_t pos = iocb->ki_pos;
1593 iov_iter_truncate(to, size);
1594 return generic_file_read_iter(iocb, to);
1598 * Try to release a page associated with block device when the system
1599 * is under memory pressure.
1601 static int blkdev_releasepage(struct page *page, gfp_t wait)
1603 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1605 if (super && super->s_op->bdev_try_to_free_page)
1606 return super->s_op->bdev_try_to_free_page(super, page, wait);
1608 return try_to_free_buffers(page);
1611 static const struct address_space_operations def_blk_aops = {
1612 .readpage = blkdev_readpage,
1613 .writepage = blkdev_writepage,
1614 .write_begin = blkdev_write_begin,
1615 .write_end = blkdev_write_end,
1616 .writepages = generic_writepages,
1617 .releasepage = blkdev_releasepage,
1618 .direct_IO = blkdev_direct_IO,
1619 .is_dirty_writeback = buffer_check_dirty_writeback,
1622 const struct file_operations def_blk_fops = {
1623 .open = blkdev_open,
1624 .release = blkdev_close,
1625 .llseek = block_llseek,
1626 .read = new_sync_read,
1627 .write = new_sync_write,
1628 .read_iter = blkdev_read_iter,
1629 .write_iter = blkdev_write_iter,
1630 .mmap = generic_file_mmap,
1631 .fsync = blkdev_fsync,
1632 .unlocked_ioctl = block_ioctl,
1633 #ifdef CONFIG_COMPAT
1634 .compat_ioctl = compat_blkdev_ioctl,
1636 .splice_read = generic_file_splice_read,
1637 .splice_write = iter_file_splice_write,
1640 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1643 mm_segment_t old_fs = get_fs();
1645 res = blkdev_ioctl(bdev, 0, cmd, arg);
1650 EXPORT_SYMBOL(ioctl_by_bdev);
1653 * lookup_bdev - lookup a struct block_device by name
1654 * @pathname: special file representing the block device
1656 * Get a reference to the blockdevice at @pathname in the current
1657 * namespace if possible and return it. Return ERR_PTR(error)
1660 struct block_device *lookup_bdev(const char *pathname)
1662 struct block_device *bdev;
1663 struct inode *inode;
1667 if (!pathname || !*pathname)
1668 return ERR_PTR(-EINVAL);
1670 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1672 return ERR_PTR(error);
1674 inode = path.dentry->d_inode;
1676 if (!S_ISBLK(inode->i_mode))
1679 if (path.mnt->mnt_flags & MNT_NODEV)
1682 bdev = bd_acquire(inode);
1689 bdev = ERR_PTR(error);
1692 EXPORT_SYMBOL(lookup_bdev);
1694 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1696 struct super_block *sb = get_super(bdev);
1701 * no need to lock the super, get_super holds the
1702 * read mutex so the filesystem cannot go away
1703 * under us (->put_super runs with the write lock
1706 shrink_dcache_sb(sb);
1707 res = invalidate_inodes(sb, kill_dirty);
1710 invalidate_bdev(bdev);
1713 EXPORT_SYMBOL(__invalidate_device);
1715 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1717 struct inode *inode, *old_inode = NULL;
1719 spin_lock(&inode_sb_list_lock);
1720 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1721 struct address_space *mapping = inode->i_mapping;
1723 spin_lock(&inode->i_lock);
1724 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1725 mapping->nrpages == 0) {
1726 spin_unlock(&inode->i_lock);
1730 spin_unlock(&inode->i_lock);
1731 spin_unlock(&inode_sb_list_lock);
1733 * We hold a reference to 'inode' so it couldn't have been
1734 * removed from s_inodes list while we dropped the
1735 * inode_sb_list_lock. We cannot iput the inode now as we can
1736 * be holding the last reference and we cannot iput it under
1737 * inode_sb_list_lock. So we keep the reference and iput it
1743 func(I_BDEV(inode), arg);
1745 spin_lock(&inode_sb_list_lock);
1747 spin_unlock(&inode_sb_list_lock);