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);
52 static void bdev_write_inode(struct inode *inode)
54 spin_lock(&inode->i_lock);
55 while (inode->i_state & I_DIRTY) {
56 spin_unlock(&inode->i_lock);
57 WARN_ON_ONCE(write_inode_now(inode, true));
58 spin_lock(&inode->i_lock);
60 spin_unlock(&inode->i_lock);
64 * Move the inode from its current bdi to a new bdi. Make sure the inode
65 * is clean before moving so that it doesn't linger on the old bdi.
67 static void bdev_inode_switch_bdi(struct inode *inode,
68 struct backing_dev_info *dst)
70 spin_lock(&inode->i_lock);
71 WARN_ON_ONCE(inode->i_state & I_DIRTY);
72 inode->i_data.backing_dev_info = dst;
73 spin_unlock(&inode->i_lock);
76 /* Kill _all_ buffers and pagecache , dirty or not.. */
77 void kill_bdev(struct block_device *bdev)
79 struct address_space *mapping = bdev->bd_inode->i_mapping;
81 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
85 truncate_inode_pages(mapping, 0);
87 EXPORT_SYMBOL(kill_bdev);
89 /* Invalidate clean unused buffers and pagecache. */
90 void invalidate_bdev(struct block_device *bdev)
92 struct address_space *mapping = bdev->bd_inode->i_mapping;
94 if (mapping->nrpages == 0)
98 lru_add_drain_all(); /* make sure all lru add caches are flushed */
99 invalidate_mapping_pages(mapping, 0, -1);
100 /* 99% of the time, we don't need to flush the cleancache on the bdev.
101 * But, for the strange corners, lets be cautious
103 cleancache_invalidate_inode(mapping);
105 EXPORT_SYMBOL(invalidate_bdev);
107 int set_blocksize(struct block_device *bdev, int size)
109 /* Size must be a power of two, and between 512 and PAGE_SIZE */
110 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
113 /* Size cannot be smaller than the size supported by the device */
114 if (size < bdev_logical_block_size(bdev))
117 /* Don't change the size if it is same as current */
118 if (bdev->bd_block_size != size) {
120 bdev->bd_block_size = size;
121 bdev->bd_inode->i_blkbits = blksize_bits(size);
127 EXPORT_SYMBOL(set_blocksize);
129 int sb_set_blocksize(struct super_block *sb, int size)
131 if (set_blocksize(sb->s_bdev, size))
133 /* If we get here, we know size is power of two
134 * and it's value is between 512 and PAGE_SIZE */
135 sb->s_blocksize = size;
136 sb->s_blocksize_bits = blksize_bits(size);
137 return sb->s_blocksize;
140 EXPORT_SYMBOL(sb_set_blocksize);
142 int sb_min_blocksize(struct super_block *sb, int size)
144 int minsize = bdev_logical_block_size(sb->s_bdev);
147 return sb_set_blocksize(sb, size);
150 EXPORT_SYMBOL(sb_min_blocksize);
153 blkdev_get_block(struct inode *inode, sector_t iblock,
154 struct buffer_head *bh, int create)
156 bh->b_bdev = I_BDEV(inode);
157 bh->b_blocknr = iblock;
158 set_buffer_mapped(bh);
163 blkdev_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter,
166 struct file *file = iocb->ki_filp;
167 struct inode *inode = file->f_mapping->host;
169 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iter,
170 offset, blkdev_get_block,
174 int __sync_blockdev(struct block_device *bdev, int wait)
179 return filemap_flush(bdev->bd_inode->i_mapping);
180 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
184 * Write out and wait upon all the dirty data associated with a block
185 * device via its mapping. Does not take the superblock lock.
187 int sync_blockdev(struct block_device *bdev)
189 return __sync_blockdev(bdev, 1);
191 EXPORT_SYMBOL(sync_blockdev);
194 * Write out and wait upon all dirty data associated with this
195 * device. Filesystem data as well as the underlying block
196 * device. Takes the superblock lock.
198 int fsync_bdev(struct block_device *bdev)
200 struct super_block *sb = get_super(bdev);
202 int res = sync_filesystem(sb);
206 return sync_blockdev(bdev);
208 EXPORT_SYMBOL(fsync_bdev);
211 * freeze_bdev -- lock a filesystem and force it into a consistent state
212 * @bdev: blockdevice to lock
214 * If a superblock is found on this device, we take the s_umount semaphore
215 * on it to make sure nobody unmounts until the snapshot creation is done.
216 * The reference counter (bd_fsfreeze_count) guarantees that only the last
217 * unfreeze process can unfreeze the frozen filesystem actually when multiple
218 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
219 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
222 struct super_block *freeze_bdev(struct block_device *bdev)
224 struct super_block *sb;
227 mutex_lock(&bdev->bd_fsfreeze_mutex);
228 if (++bdev->bd_fsfreeze_count > 1) {
230 * We don't even need to grab a reference - the first call
231 * to freeze_bdev grab an active reference and only the last
232 * thaw_bdev drops it.
234 sb = get_super(bdev);
236 mutex_unlock(&bdev->bd_fsfreeze_mutex);
240 sb = get_active_super(bdev);
243 if (sb->s_op->freeze_super)
244 error = sb->s_op->freeze_super(sb);
246 error = freeze_super(sb);
248 deactivate_super(sb);
249 bdev->bd_fsfreeze_count--;
250 mutex_unlock(&bdev->bd_fsfreeze_mutex);
251 return ERR_PTR(error);
253 deactivate_super(sb);
256 mutex_unlock(&bdev->bd_fsfreeze_mutex);
257 return sb; /* thaw_bdev releases s->s_umount */
259 EXPORT_SYMBOL(freeze_bdev);
262 * thaw_bdev -- unlock filesystem
263 * @bdev: blockdevice to unlock
264 * @sb: associated superblock
266 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
268 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
272 mutex_lock(&bdev->bd_fsfreeze_mutex);
273 if (!bdev->bd_fsfreeze_count)
277 if (--bdev->bd_fsfreeze_count > 0)
283 if (sb->s_op->thaw_super)
284 error = sb->s_op->thaw_super(sb);
286 error = thaw_super(sb);
288 bdev->bd_fsfreeze_count++;
289 mutex_unlock(&bdev->bd_fsfreeze_mutex);
293 mutex_unlock(&bdev->bd_fsfreeze_mutex);
296 EXPORT_SYMBOL(thaw_bdev);
298 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
300 return block_write_full_page(page, blkdev_get_block, wbc);
303 static int blkdev_readpage(struct file * file, struct page * page)
305 return block_read_full_page(page, blkdev_get_block);
308 static int blkdev_readpages(struct file *file, struct address_space *mapping,
309 struct list_head *pages, unsigned nr_pages)
311 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
314 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
315 loff_t pos, unsigned len, unsigned flags,
316 struct page **pagep, void **fsdata)
318 return block_write_begin(mapping, pos, len, flags, pagep,
322 static int blkdev_write_end(struct file *file, struct address_space *mapping,
323 loff_t pos, unsigned len, unsigned copied,
324 struct page *page, void *fsdata)
327 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
330 page_cache_release(page);
337 * for a block special file file_inode(file)->i_size is zero
338 * so we compute the size by hand (just as in block_read/write above)
340 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
342 struct inode *bd_inode = file->f_mapping->host;
345 mutex_lock(&bd_inode->i_mutex);
346 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
347 mutex_unlock(&bd_inode->i_mutex);
351 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
353 struct inode *bd_inode = filp->f_mapping->host;
354 struct block_device *bdev = I_BDEV(bd_inode);
357 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
362 * There is no need to serialise calls to blkdev_issue_flush with
363 * i_mutex and doing so causes performance issues with concurrent
364 * O_SYNC writers to a block device.
366 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
367 if (error == -EOPNOTSUPP)
372 EXPORT_SYMBOL(blkdev_fsync);
375 * bdev_read_page() - Start reading a page from a block device
376 * @bdev: The device to read the page from
377 * @sector: The offset on the device to read the page to (need not be aligned)
378 * @page: The page to read
380 * On entry, the page should be locked. It will be unlocked when the page
381 * has been read. If the block driver implements rw_page synchronously,
382 * that will be true on exit from this function, but it need not be.
384 * Errors returned by this function are usually "soft", eg out of memory, or
385 * queue full; callers should try a different route to read this page rather
386 * than propagate an error back up the stack.
388 * Return: negative errno if an error occurs, 0 if submission was successful.
390 int bdev_read_page(struct block_device *bdev, sector_t sector,
393 const struct block_device_operations *ops = bdev->bd_disk->fops;
396 return ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
398 EXPORT_SYMBOL_GPL(bdev_read_page);
401 * bdev_write_page() - Start writing a page to a block device
402 * @bdev: The device to write the page to
403 * @sector: The offset on the device to write the page to (need not be aligned)
404 * @page: The page to write
405 * @wbc: The writeback_control for the write
407 * On entry, the page should be locked and not currently under writeback.
408 * On exit, if the write started successfully, the page will be unlocked and
409 * under writeback. If the write failed already (eg the driver failed to
410 * queue the page to the device), the page will still be locked. If the
411 * caller is a ->writepage implementation, it will need to unlock the page.
413 * Errors returned by this function are usually "soft", eg out of memory, or
414 * queue full; callers should try a different route to write this page rather
415 * than propagate an error back up the stack.
417 * Return: negative errno if an error occurs, 0 if submission was successful.
419 int bdev_write_page(struct block_device *bdev, sector_t sector,
420 struct page *page, struct writeback_control *wbc)
423 int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
424 const struct block_device_operations *ops = bdev->bd_disk->fops;
427 set_page_writeback(page);
428 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
430 end_page_writeback(page);
435 EXPORT_SYMBOL_GPL(bdev_write_page);
441 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
442 static struct kmem_cache * bdev_cachep __read_mostly;
444 static struct inode *bdev_alloc_inode(struct super_block *sb)
446 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
449 return &ei->vfs_inode;
452 static void bdev_i_callback(struct rcu_head *head)
454 struct inode *inode = container_of(head, struct inode, i_rcu);
455 struct bdev_inode *bdi = BDEV_I(inode);
457 kmem_cache_free(bdev_cachep, bdi);
460 static void bdev_destroy_inode(struct inode *inode)
462 call_rcu(&inode->i_rcu, bdev_i_callback);
465 static void init_once(void *foo)
467 struct bdev_inode *ei = (struct bdev_inode *) foo;
468 struct block_device *bdev = &ei->bdev;
470 memset(bdev, 0, sizeof(*bdev));
471 mutex_init(&bdev->bd_mutex);
472 INIT_LIST_HEAD(&bdev->bd_inodes);
473 INIT_LIST_HEAD(&bdev->bd_list);
475 INIT_LIST_HEAD(&bdev->bd_holder_disks);
477 inode_init_once(&ei->vfs_inode);
478 /* Initialize mutex for freeze. */
479 mutex_init(&bdev->bd_fsfreeze_mutex);
482 static inline void __bd_forget(struct inode *inode)
484 list_del_init(&inode->i_devices);
485 inode->i_bdev = NULL;
486 inode->i_mapping = &inode->i_data;
489 static void bdev_evict_inode(struct inode *inode)
491 struct block_device *bdev = &BDEV_I(inode)->bdev;
493 truncate_inode_pages_final(&inode->i_data);
494 invalidate_inode_buffers(inode); /* is it needed here? */
496 spin_lock(&bdev_lock);
497 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
498 __bd_forget(list_entry(p, struct inode, i_devices));
500 list_del_init(&bdev->bd_list);
501 spin_unlock(&bdev_lock);
504 static const struct super_operations bdev_sops = {
505 .statfs = simple_statfs,
506 .alloc_inode = bdev_alloc_inode,
507 .destroy_inode = bdev_destroy_inode,
508 .drop_inode = generic_delete_inode,
509 .evict_inode = bdev_evict_inode,
512 static struct dentry *bd_mount(struct file_system_type *fs_type,
513 int flags, const char *dev_name, void *data)
515 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
518 static struct file_system_type bd_type = {
521 .kill_sb = kill_anon_super,
524 static struct super_block *blockdev_superblock __read_mostly;
526 void __init bdev_cache_init(void)
529 static struct vfsmount *bd_mnt;
531 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
532 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
533 SLAB_MEM_SPREAD|SLAB_PANIC),
535 err = register_filesystem(&bd_type);
537 panic("Cannot register bdev pseudo-fs");
538 bd_mnt = kern_mount(&bd_type);
540 panic("Cannot create bdev pseudo-fs");
541 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
545 * Most likely _very_ bad one - but then it's hardly critical for small
546 * /dev and can be fixed when somebody will need really large one.
547 * Keep in mind that it will be fed through icache hash function too.
549 static inline unsigned long hash(dev_t dev)
551 return MAJOR(dev)+MINOR(dev);
554 static int bdev_test(struct inode *inode, void *data)
556 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
559 static int bdev_set(struct inode *inode, void *data)
561 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
565 static LIST_HEAD(all_bdevs);
567 struct block_device *bdget(dev_t dev)
569 struct block_device *bdev;
572 inode = iget5_locked(blockdev_superblock, hash(dev),
573 bdev_test, bdev_set, &dev);
578 bdev = &BDEV_I(inode)->bdev;
580 if (inode->i_state & I_NEW) {
581 bdev->bd_contains = NULL;
582 bdev->bd_super = NULL;
583 bdev->bd_inode = inode;
584 bdev->bd_block_size = (1 << inode->i_blkbits);
585 bdev->bd_part_count = 0;
586 bdev->bd_invalidated = 0;
587 inode->i_mode = S_IFBLK;
589 inode->i_bdev = bdev;
590 inode->i_data.a_ops = &def_blk_aops;
591 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
592 inode->i_data.backing_dev_info = &default_backing_dev_info;
593 spin_lock(&bdev_lock);
594 list_add(&bdev->bd_list, &all_bdevs);
595 spin_unlock(&bdev_lock);
596 unlock_new_inode(inode);
601 EXPORT_SYMBOL(bdget);
604 * bdgrab -- Grab a reference to an already referenced block device
605 * @bdev: Block device to grab a reference to.
607 struct block_device *bdgrab(struct block_device *bdev)
609 ihold(bdev->bd_inode);
612 EXPORT_SYMBOL(bdgrab);
614 long nr_blockdev_pages(void)
616 struct block_device *bdev;
618 spin_lock(&bdev_lock);
619 list_for_each_entry(bdev, &all_bdevs, bd_list) {
620 ret += bdev->bd_inode->i_mapping->nrpages;
622 spin_unlock(&bdev_lock);
626 void bdput(struct block_device *bdev)
628 iput(bdev->bd_inode);
631 EXPORT_SYMBOL(bdput);
633 static struct block_device *bd_acquire(struct inode *inode)
635 struct block_device *bdev;
637 spin_lock(&bdev_lock);
638 bdev = inode->i_bdev;
640 ihold(bdev->bd_inode);
641 spin_unlock(&bdev_lock);
644 spin_unlock(&bdev_lock);
646 bdev = bdget(inode->i_rdev);
648 spin_lock(&bdev_lock);
649 if (!inode->i_bdev) {
651 * We take an additional reference to bd_inode,
652 * and it's released in clear_inode() of inode.
653 * So, we can access it via ->i_mapping always
656 ihold(bdev->bd_inode);
657 inode->i_bdev = bdev;
658 inode->i_mapping = bdev->bd_inode->i_mapping;
659 list_add(&inode->i_devices, &bdev->bd_inodes);
661 spin_unlock(&bdev_lock);
666 int sb_is_blkdev_sb(struct super_block *sb)
668 return sb == blockdev_superblock;
671 /* Call when you free inode */
673 void bd_forget(struct inode *inode)
675 struct block_device *bdev = NULL;
677 spin_lock(&bdev_lock);
678 if (!sb_is_blkdev_sb(inode->i_sb))
679 bdev = inode->i_bdev;
681 spin_unlock(&bdev_lock);
684 iput(bdev->bd_inode);
688 * bd_may_claim - test whether a block device can be claimed
689 * @bdev: block device of interest
690 * @whole: whole block device containing @bdev, may equal @bdev
691 * @holder: holder trying to claim @bdev
693 * Test whether @bdev can be claimed by @holder.
696 * spin_lock(&bdev_lock).
699 * %true if @bdev can be claimed, %false otherwise.
701 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
704 if (bdev->bd_holder == holder)
705 return true; /* already a holder */
706 else if (bdev->bd_holder != NULL)
707 return false; /* held by someone else */
708 else if (bdev->bd_contains == bdev)
709 return true; /* is a whole device which isn't held */
711 else if (whole->bd_holder == bd_may_claim)
712 return true; /* is a partition of a device that is being partitioned */
713 else if (whole->bd_holder != NULL)
714 return false; /* is a partition of a held device */
716 return true; /* is a partition of an un-held device */
720 * bd_prepare_to_claim - prepare to claim a block device
721 * @bdev: block device of interest
722 * @whole: the whole device containing @bdev, may equal @bdev
723 * @holder: holder trying to claim @bdev
725 * Prepare to claim @bdev. This function fails if @bdev is already
726 * claimed by another holder and waits if another claiming is in
727 * progress. This function doesn't actually claim. On successful
728 * return, the caller has ownership of bd_claiming and bd_holder[s].
731 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
735 * 0 if @bdev can be claimed, -EBUSY otherwise.
737 static int bd_prepare_to_claim(struct block_device *bdev,
738 struct block_device *whole, void *holder)
741 /* if someone else claimed, fail */
742 if (!bd_may_claim(bdev, whole, holder))
745 /* if claiming is already in progress, wait for it to finish */
746 if (whole->bd_claiming) {
747 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
750 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
751 spin_unlock(&bdev_lock);
753 finish_wait(wq, &wait);
754 spin_lock(&bdev_lock);
763 * bd_start_claiming - start claiming a block device
764 * @bdev: block device of interest
765 * @holder: holder trying to claim @bdev
767 * @bdev is about to be opened exclusively. Check @bdev can be opened
768 * exclusively and mark that an exclusive open is in progress. Each
769 * successful call to this function must be matched with a call to
770 * either bd_finish_claiming() or bd_abort_claiming() (which do not
773 * This function is used to gain exclusive access to the block device
774 * without actually causing other exclusive open attempts to fail. It
775 * should be used when the open sequence itself requires exclusive
776 * access but may subsequently fail.
782 * Pointer to the block device containing @bdev on success, ERR_PTR()
785 static struct block_device *bd_start_claiming(struct block_device *bdev,
788 struct gendisk *disk;
789 struct block_device *whole;
795 * @bdev might not have been initialized properly yet, look up
796 * and grab the outer block device the hard way.
798 disk = get_gendisk(bdev->bd_dev, &partno);
800 return ERR_PTR(-ENXIO);
803 * Normally, @bdev should equal what's returned from bdget_disk()
804 * if partno is 0; however, some drivers (floppy) use multiple
805 * bdev's for the same physical device and @bdev may be one of the
806 * aliases. Keep @bdev if partno is 0. This means claimer
807 * tracking is broken for those devices but it has always been that
811 whole = bdget_disk(disk, 0);
813 whole = bdgrab(bdev);
815 module_put(disk->fops->owner);
818 return ERR_PTR(-ENOMEM);
820 /* prepare to claim, if successful, mark claiming in progress */
821 spin_lock(&bdev_lock);
823 err = bd_prepare_to_claim(bdev, whole, holder);
825 whole->bd_claiming = holder;
826 spin_unlock(&bdev_lock);
829 spin_unlock(&bdev_lock);
836 struct bd_holder_disk {
837 struct list_head list;
838 struct gendisk *disk;
842 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
843 struct gendisk *disk)
845 struct bd_holder_disk *holder;
847 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
848 if (holder->disk == disk)
853 static int add_symlink(struct kobject *from, struct kobject *to)
855 return sysfs_create_link(from, to, kobject_name(to));
858 static void del_symlink(struct kobject *from, struct kobject *to)
860 sysfs_remove_link(from, kobject_name(to));
864 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
865 * @bdev: the claimed slave bdev
866 * @disk: the holding disk
868 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
870 * This functions creates the following sysfs symlinks.
872 * - from "slaves" directory of the holder @disk to the claimed @bdev
873 * - from "holders" directory of the @bdev to the holder @disk
875 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
876 * passed to bd_link_disk_holder(), then:
878 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
879 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
881 * The caller must have claimed @bdev before calling this function and
882 * ensure that both @bdev and @disk are valid during the creation and
883 * lifetime of these symlinks.
889 * 0 on success, -errno on failure.
891 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
893 struct bd_holder_disk *holder;
896 mutex_lock(&bdev->bd_mutex);
898 WARN_ON_ONCE(!bdev->bd_holder);
900 /* FIXME: remove the following once add_disk() handles errors */
901 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
904 holder = bd_find_holder_disk(bdev, disk);
910 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
916 INIT_LIST_HEAD(&holder->list);
920 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
924 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
928 * bdev could be deleted beneath us which would implicitly destroy
929 * the holder directory. Hold on to it.
931 kobject_get(bdev->bd_part->holder_dir);
933 list_add(&holder->list, &bdev->bd_holder_disks);
937 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
941 mutex_unlock(&bdev->bd_mutex);
944 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
947 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
948 * @bdev: the calimed slave bdev
949 * @disk: the holding disk
951 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
956 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
958 struct bd_holder_disk *holder;
960 mutex_lock(&bdev->bd_mutex);
962 holder = bd_find_holder_disk(bdev, disk);
964 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
965 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
966 del_symlink(bdev->bd_part->holder_dir,
967 &disk_to_dev(disk)->kobj);
968 kobject_put(bdev->bd_part->holder_dir);
969 list_del_init(&holder->list);
973 mutex_unlock(&bdev->bd_mutex);
975 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
979 * flush_disk - invalidates all buffer-cache entries on a disk
981 * @bdev: struct block device to be flushed
982 * @kill_dirty: flag to guide handling of dirty inodes
984 * Invalidates all buffer-cache entries on a disk. It should be called
985 * when a disk has been changed -- either by a media change or online
988 static void flush_disk(struct block_device *bdev, bool kill_dirty)
990 if (__invalidate_device(bdev, kill_dirty)) {
991 char name[BDEVNAME_SIZE] = "";
994 disk_name(bdev->bd_disk, 0, name);
995 printk(KERN_WARNING "VFS: busy inodes on changed media or "
996 "resized disk %s\n", name);
1001 if (disk_part_scan_enabled(bdev->bd_disk))
1002 bdev->bd_invalidated = 1;
1006 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1007 * @disk: struct gendisk to check
1008 * @bdev: struct bdev to adjust.
1010 * This routine checks to see if the bdev size does not match the disk size
1011 * and adjusts it if it differs.
1013 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1015 loff_t disk_size, bdev_size;
1017 disk_size = (loff_t)get_capacity(disk) << 9;
1018 bdev_size = i_size_read(bdev->bd_inode);
1019 if (disk_size != bdev_size) {
1020 char name[BDEVNAME_SIZE];
1022 disk_name(disk, 0, name);
1024 "%s: detected capacity change from %lld to %lld\n",
1025 name, bdev_size, disk_size);
1026 i_size_write(bdev->bd_inode, disk_size);
1027 flush_disk(bdev, false);
1030 EXPORT_SYMBOL(check_disk_size_change);
1033 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1034 * @disk: struct gendisk to be revalidated
1036 * This routine is a wrapper for lower-level driver's revalidate_disk
1037 * call-backs. It is used to do common pre and post operations needed
1038 * for all revalidate_disk operations.
1040 int revalidate_disk(struct gendisk *disk)
1042 struct block_device *bdev;
1045 if (disk->fops->revalidate_disk)
1046 ret = disk->fops->revalidate_disk(disk);
1048 bdev = bdget_disk(disk, 0);
1052 mutex_lock(&bdev->bd_mutex);
1053 check_disk_size_change(disk, bdev);
1054 bdev->bd_invalidated = 0;
1055 mutex_unlock(&bdev->bd_mutex);
1059 EXPORT_SYMBOL(revalidate_disk);
1062 * This routine checks whether a removable media has been changed,
1063 * and invalidates all buffer-cache-entries in that case. This
1064 * is a relatively slow routine, so we have to try to minimize using
1065 * it. Thus it is called only upon a 'mount' or 'open'. This
1066 * is the best way of combining speed and utility, I think.
1067 * People changing diskettes in the middle of an operation deserve
1070 int check_disk_change(struct block_device *bdev)
1072 struct gendisk *disk = bdev->bd_disk;
1073 const struct block_device_operations *bdops = disk->fops;
1074 unsigned int events;
1076 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1077 DISK_EVENT_EJECT_REQUEST);
1078 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1081 flush_disk(bdev, true);
1082 if (bdops->revalidate_disk)
1083 bdops->revalidate_disk(bdev->bd_disk);
1087 EXPORT_SYMBOL(check_disk_change);
1089 void bd_set_size(struct block_device *bdev, loff_t size)
1091 unsigned bsize = bdev_logical_block_size(bdev);
1093 mutex_lock(&bdev->bd_inode->i_mutex);
1094 i_size_write(bdev->bd_inode, size);
1095 mutex_unlock(&bdev->bd_inode->i_mutex);
1096 while (bsize < PAGE_CACHE_SIZE) {
1101 bdev->bd_block_size = bsize;
1102 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1104 EXPORT_SYMBOL(bd_set_size);
1106 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1111 * mutex_lock(part->bd_mutex)
1112 * mutex_lock_nested(whole->bd_mutex, 1)
1115 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1117 struct gendisk *disk;
1118 struct module *owner;
1123 if (mode & FMODE_READ)
1125 if (mode & FMODE_WRITE)
1128 * hooks: /n/, see "layering violations".
1131 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1141 disk = get_gendisk(bdev->bd_dev, &partno);
1144 owner = disk->fops->owner;
1146 disk_block_events(disk);
1147 mutex_lock_nested(&bdev->bd_mutex, for_part);
1148 if (!bdev->bd_openers) {
1149 bdev->bd_disk = disk;
1150 bdev->bd_queue = disk->queue;
1151 bdev->bd_contains = bdev;
1153 struct backing_dev_info *bdi;
1156 bdev->bd_part = disk_get_part(disk, partno);
1161 if (disk->fops->open) {
1162 ret = disk->fops->open(bdev, mode);
1163 if (ret == -ERESTARTSYS) {
1164 /* Lost a race with 'disk' being
1165 * deleted, try again.
1168 disk_put_part(bdev->bd_part);
1169 bdev->bd_part = NULL;
1170 bdev->bd_disk = NULL;
1171 bdev->bd_queue = NULL;
1172 mutex_unlock(&bdev->bd_mutex);
1173 disk_unblock_events(disk);
1181 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1182 bdi = blk_get_backing_dev_info(bdev);
1183 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1187 * If the device is invalidated, rescan partition
1188 * if open succeeded or failed with -ENOMEDIUM.
1189 * The latter is necessary to prevent ghost
1190 * partitions on a removed medium.
1192 if (bdev->bd_invalidated) {
1194 rescan_partitions(disk, bdev);
1195 else if (ret == -ENOMEDIUM)
1196 invalidate_partitions(disk, bdev);
1201 struct block_device *whole;
1202 whole = bdget_disk(disk, 0);
1207 ret = __blkdev_get(whole, mode, 1);
1210 bdev->bd_contains = whole;
1211 bdev_inode_switch_bdi(bdev->bd_inode,
1212 whole->bd_inode->i_data.backing_dev_info);
1213 bdev->bd_part = disk_get_part(disk, partno);
1214 if (!(disk->flags & GENHD_FL_UP) ||
1215 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1219 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1222 if (bdev->bd_contains == bdev) {
1224 if (bdev->bd_disk->fops->open)
1225 ret = bdev->bd_disk->fops->open(bdev, mode);
1226 /* the same as first opener case, read comment there */
1227 if (bdev->bd_invalidated) {
1229 rescan_partitions(bdev->bd_disk, bdev);
1230 else if (ret == -ENOMEDIUM)
1231 invalidate_partitions(bdev->bd_disk, bdev);
1234 goto out_unlock_bdev;
1236 /* only one opener holds refs to the module and disk */
1242 bdev->bd_part_count++;
1243 mutex_unlock(&bdev->bd_mutex);
1244 disk_unblock_events(disk);
1248 disk_put_part(bdev->bd_part);
1249 bdev->bd_disk = NULL;
1250 bdev->bd_part = NULL;
1251 bdev->bd_queue = NULL;
1252 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1253 if (bdev != bdev->bd_contains)
1254 __blkdev_put(bdev->bd_contains, mode, 1);
1255 bdev->bd_contains = NULL;
1257 mutex_unlock(&bdev->bd_mutex);
1258 disk_unblock_events(disk);
1268 * blkdev_get - open a block device
1269 * @bdev: block_device to open
1270 * @mode: FMODE_* mask
1271 * @holder: exclusive holder identifier
1273 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1274 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1275 * @holder is invalid. Exclusive opens may nest for the same @holder.
1277 * On success, the reference count of @bdev is unchanged. On failure,
1284 * 0 on success, -errno on failure.
1286 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1288 struct block_device *whole = NULL;
1291 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1293 if ((mode & FMODE_EXCL) && holder) {
1294 whole = bd_start_claiming(bdev, holder);
1295 if (IS_ERR(whole)) {
1297 return PTR_ERR(whole);
1301 res = __blkdev_get(bdev, mode, 0);
1304 struct gendisk *disk = whole->bd_disk;
1306 /* finish claiming */
1307 mutex_lock(&bdev->bd_mutex);
1308 spin_lock(&bdev_lock);
1311 BUG_ON(!bd_may_claim(bdev, whole, holder));
1313 * Note that for a whole device bd_holders
1314 * will be incremented twice, and bd_holder
1315 * will be set to bd_may_claim before being
1318 whole->bd_holders++;
1319 whole->bd_holder = bd_may_claim;
1321 bdev->bd_holder = holder;
1324 /* tell others that we're done */
1325 BUG_ON(whole->bd_claiming != holder);
1326 whole->bd_claiming = NULL;
1327 wake_up_bit(&whole->bd_claiming, 0);
1329 spin_unlock(&bdev_lock);
1332 * Block event polling for write claims if requested. Any
1333 * write holder makes the write_holder state stick until
1334 * all are released. This is good enough and tracking
1335 * individual writeable reference is too fragile given the
1336 * way @mode is used in blkdev_get/put().
1338 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1339 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1340 bdev->bd_write_holder = true;
1341 disk_block_events(disk);
1344 mutex_unlock(&bdev->bd_mutex);
1350 EXPORT_SYMBOL(blkdev_get);
1353 * blkdev_get_by_path - open a block device by name
1354 * @path: path to the block device to open
1355 * @mode: FMODE_* mask
1356 * @holder: exclusive holder identifier
1358 * Open the blockdevice described by the device file at @path. @mode
1359 * and @holder are identical to blkdev_get().
1361 * On success, the returned block_device has reference count of one.
1367 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1369 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1372 struct block_device *bdev;
1375 bdev = lookup_bdev(path);
1379 err = blkdev_get(bdev, mode, holder);
1381 return ERR_PTR(err);
1383 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1384 blkdev_put(bdev, mode);
1385 return ERR_PTR(-EACCES);
1390 EXPORT_SYMBOL(blkdev_get_by_path);
1393 * blkdev_get_by_dev - open a block device by device number
1394 * @dev: device number of block device to open
1395 * @mode: FMODE_* mask
1396 * @holder: exclusive holder identifier
1398 * Open the blockdevice described by device number @dev. @mode and
1399 * @holder are identical to blkdev_get().
1401 * Use it ONLY if you really do not have anything better - i.e. when
1402 * you are behind a truly sucky interface and all you are given is a
1403 * device number. _Never_ to be used for internal purposes. If you
1404 * ever need it - reconsider your API.
1406 * On success, the returned block_device has reference count of one.
1412 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1414 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1416 struct block_device *bdev;
1421 return ERR_PTR(-ENOMEM);
1423 err = blkdev_get(bdev, mode, holder);
1425 return ERR_PTR(err);
1429 EXPORT_SYMBOL(blkdev_get_by_dev);
1431 static int blkdev_open(struct inode * inode, struct file * filp)
1433 struct block_device *bdev;
1436 * Preserve backwards compatibility and allow large file access
1437 * even if userspace doesn't ask for it explicitly. Some mkfs
1438 * binary needs it. We might want to drop this workaround
1439 * during an unstable branch.
1441 filp->f_flags |= O_LARGEFILE;
1443 if (filp->f_flags & O_NDELAY)
1444 filp->f_mode |= FMODE_NDELAY;
1445 if (filp->f_flags & O_EXCL)
1446 filp->f_mode |= FMODE_EXCL;
1447 if ((filp->f_flags & O_ACCMODE) == 3)
1448 filp->f_mode |= FMODE_WRITE_IOCTL;
1450 bdev = bd_acquire(inode);
1454 filp->f_mapping = bdev->bd_inode->i_mapping;
1456 return blkdev_get(bdev, filp->f_mode, filp);
1459 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1461 struct gendisk *disk = bdev->bd_disk;
1462 struct block_device *victim = NULL;
1464 mutex_lock_nested(&bdev->bd_mutex, for_part);
1466 bdev->bd_part_count--;
1468 if (!--bdev->bd_openers) {
1469 WARN_ON_ONCE(bdev->bd_holders);
1470 sync_blockdev(bdev);
1473 * ->release can cause the queue to disappear, so flush all
1474 * dirty data before.
1476 bdev_write_inode(bdev->bd_inode);
1477 bdev_inode_switch_bdi(bdev->bd_inode,
1478 &default_backing_dev_info);
1480 if (bdev->bd_contains == bdev) {
1481 if (disk->fops->release)
1482 disk->fops->release(disk, mode);
1484 if (!bdev->bd_openers) {
1485 struct module *owner = disk->fops->owner;
1487 disk_put_part(bdev->bd_part);
1488 bdev->bd_part = NULL;
1489 bdev->bd_disk = NULL;
1490 if (bdev != bdev->bd_contains)
1491 victim = bdev->bd_contains;
1492 bdev->bd_contains = NULL;
1497 mutex_unlock(&bdev->bd_mutex);
1500 __blkdev_put(victim, mode, 1);
1503 void blkdev_put(struct block_device *bdev, fmode_t mode)
1505 mutex_lock(&bdev->bd_mutex);
1507 if (mode & FMODE_EXCL) {
1511 * Release a claim on the device. The holder fields
1512 * are protected with bdev_lock. bd_mutex is to
1513 * synchronize disk_holder unlinking.
1515 spin_lock(&bdev_lock);
1517 WARN_ON_ONCE(--bdev->bd_holders < 0);
1518 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1520 /* bd_contains might point to self, check in a separate step */
1521 if ((bdev_free = !bdev->bd_holders))
1522 bdev->bd_holder = NULL;
1523 if (!bdev->bd_contains->bd_holders)
1524 bdev->bd_contains->bd_holder = NULL;
1526 spin_unlock(&bdev_lock);
1529 * If this was the last claim, remove holder link and
1530 * unblock evpoll if it was a write holder.
1532 if (bdev_free && bdev->bd_write_holder) {
1533 disk_unblock_events(bdev->bd_disk);
1534 bdev->bd_write_holder = false;
1539 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1540 * event. This is to ensure detection of media removal commanded
1541 * from userland - e.g. eject(1).
1543 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1545 mutex_unlock(&bdev->bd_mutex);
1547 __blkdev_put(bdev, mode, 0);
1549 EXPORT_SYMBOL(blkdev_put);
1551 static int blkdev_close(struct inode * inode, struct file * filp)
1553 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1554 blkdev_put(bdev, filp->f_mode);
1558 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1560 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1561 fmode_t mode = file->f_mode;
1564 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1565 * to updated it before every ioctl.
1567 if (file->f_flags & O_NDELAY)
1568 mode |= FMODE_NDELAY;
1570 mode &= ~FMODE_NDELAY;
1572 return blkdev_ioctl(bdev, mode, cmd, arg);
1576 * Write data to the block device. Only intended for the block device itself
1577 * and the raw driver which basically is a fake block device.
1579 * Does not take i_mutex for the write and thus is not for general purpose
1582 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1584 struct file *file = iocb->ki_filp;
1585 struct blk_plug plug;
1588 blk_start_plug(&plug);
1589 ret = __generic_file_write_iter(iocb, from);
1592 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
1596 blk_finish_plug(&plug);
1599 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1601 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1603 struct file *file = iocb->ki_filp;
1604 struct inode *bd_inode = file->f_mapping->host;
1605 loff_t size = i_size_read(bd_inode);
1606 loff_t pos = iocb->ki_pos;
1612 iov_iter_truncate(to, size);
1613 return generic_file_read_iter(iocb, to);
1615 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1618 * Try to release a page associated with block device when the system
1619 * is under memory pressure.
1621 static int blkdev_releasepage(struct page *page, gfp_t wait)
1623 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1625 if (super && super->s_op->bdev_try_to_free_page)
1626 return super->s_op->bdev_try_to_free_page(super, page, wait);
1628 return try_to_free_buffers(page);
1631 static const struct address_space_operations def_blk_aops = {
1632 .readpage = blkdev_readpage,
1633 .readpages = blkdev_readpages,
1634 .writepage = blkdev_writepage,
1635 .write_begin = blkdev_write_begin,
1636 .write_end = blkdev_write_end,
1637 .writepages = generic_writepages,
1638 .releasepage = blkdev_releasepage,
1639 .direct_IO = blkdev_direct_IO,
1640 .is_dirty_writeback = buffer_check_dirty_writeback,
1643 const struct file_operations def_blk_fops = {
1644 .open = blkdev_open,
1645 .release = blkdev_close,
1646 .llseek = block_llseek,
1647 .read = new_sync_read,
1648 .write = new_sync_write,
1649 .read_iter = blkdev_read_iter,
1650 .write_iter = blkdev_write_iter,
1651 .mmap = generic_file_mmap,
1652 .fsync = blkdev_fsync,
1653 .unlocked_ioctl = block_ioctl,
1654 #ifdef CONFIG_COMPAT
1655 .compat_ioctl = compat_blkdev_ioctl,
1657 .splice_read = generic_file_splice_read,
1658 .splice_write = iter_file_splice_write,
1661 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1664 mm_segment_t old_fs = get_fs();
1666 res = blkdev_ioctl(bdev, 0, cmd, arg);
1671 EXPORT_SYMBOL(ioctl_by_bdev);
1674 * lookup_bdev - lookup a struct block_device by name
1675 * @pathname: special file representing the block device
1677 * Get a reference to the blockdevice at @pathname in the current
1678 * namespace if possible and return it. Return ERR_PTR(error)
1681 struct block_device *lookup_bdev(const char *pathname)
1683 struct block_device *bdev;
1684 struct inode *inode;
1688 if (!pathname || !*pathname)
1689 return ERR_PTR(-EINVAL);
1691 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1693 return ERR_PTR(error);
1695 inode = path.dentry->d_inode;
1697 if (!S_ISBLK(inode->i_mode))
1700 if (path.mnt->mnt_flags & MNT_NODEV)
1703 bdev = bd_acquire(inode);
1710 bdev = ERR_PTR(error);
1713 EXPORT_SYMBOL(lookup_bdev);
1715 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1717 struct super_block *sb = get_super(bdev);
1722 * no need to lock the super, get_super holds the
1723 * read mutex so the filesystem cannot go away
1724 * under us (->put_super runs with the write lock
1727 shrink_dcache_sb(sb);
1728 res = invalidate_inodes(sb, kill_dirty);
1731 invalidate_bdev(bdev);
1734 EXPORT_SYMBOL(__invalidate_device);
1736 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1738 struct inode *inode, *old_inode = NULL;
1740 spin_lock(&inode_sb_list_lock);
1741 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1742 struct address_space *mapping = inode->i_mapping;
1744 spin_lock(&inode->i_lock);
1745 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1746 mapping->nrpages == 0) {
1747 spin_unlock(&inode->i_lock);
1751 spin_unlock(&inode->i_lock);
1752 spin_unlock(&inode_sb_list_lock);
1754 * We hold a reference to 'inode' so it couldn't have been
1755 * removed from s_inodes list while we dropped the
1756 * inode_sb_list_lock. We cannot iput the inode now as we can
1757 * be holding the last reference and we cannot iput it under
1758 * inode_sb_list_lock. So we keep the reference and iput it
1764 func(I_BDEV(inode), arg);
1766 spin_lock(&inode_sb_list_lock);
1768 spin_unlock(&inode_sb_list_lock);