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/backing-dev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/magic.h>
21 #include <linux/buffer_head.h>
22 #include <linux/swap.h>
23 #include <linux/pagevec.h>
24 #include <linux/writeback.h>
25 #include <linux/mpage.h>
26 #include <linux/mount.h>
27 #include <linux/uio.h>
28 #include <linux/namei.h>
29 #include <linux/log2.h>
30 #include <linux/cleancache.h>
31 #include <linux/dax.h>
32 #include <asm/uaccess.h>
36 struct block_device bdev;
37 struct inode vfs_inode;
40 static const struct address_space_operations def_blk_aops;
42 static inline struct bdev_inode *BDEV_I(struct inode *inode)
44 return container_of(inode, struct bdev_inode, vfs_inode);
47 struct block_device *I_BDEV(struct inode *inode)
49 return &BDEV_I(inode)->bdev;
51 EXPORT_SYMBOL(I_BDEV);
53 static void bdev_write_inode(struct block_device *bdev)
55 struct inode *inode = bdev->bd_inode;
58 spin_lock(&inode->i_lock);
59 while (inode->i_state & I_DIRTY) {
60 spin_unlock(&inode->i_lock);
61 ret = write_inode_now(inode, true);
63 char name[BDEVNAME_SIZE];
64 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
65 "for block device %s (err=%d).\n",
66 bdevname(bdev, name), ret);
68 spin_lock(&inode->i_lock);
70 spin_unlock(&inode->i_lock);
73 /* Kill _all_ buffers and pagecache , dirty or not.. */
74 void kill_bdev(struct block_device *bdev)
76 struct address_space *mapping = bdev->bd_inode->i_mapping;
78 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
82 truncate_inode_pages(mapping, 0);
84 EXPORT_SYMBOL(kill_bdev);
86 /* Invalidate clean unused buffers and pagecache. */
87 void invalidate_bdev(struct block_device *bdev)
89 struct address_space *mapping = bdev->bd_inode->i_mapping;
91 if (mapping->nrpages) {
93 lru_add_drain_all(); /* make sure all lru add caches are flushed */
94 invalidate_mapping_pages(mapping, 0, -1);
96 /* 99% of the time, we don't need to flush the cleancache on the bdev.
97 * But, for the strange corners, lets be cautious
99 cleancache_invalidate_inode(mapping);
101 EXPORT_SYMBOL(invalidate_bdev);
103 int set_blocksize(struct block_device *bdev, int size)
105 /* Size must be a power of two, and between 512 and PAGE_SIZE */
106 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
109 /* Size cannot be smaller than the size supported by the device */
110 if (size < bdev_logical_block_size(bdev))
113 /* Don't change the size if it is same as current */
114 if (bdev->bd_block_size != size) {
116 bdev->bd_block_size = size;
117 bdev->bd_inode->i_blkbits = blksize_bits(size);
123 EXPORT_SYMBOL(set_blocksize);
125 int sb_set_blocksize(struct super_block *sb, int size)
127 if (set_blocksize(sb->s_bdev, size))
129 /* If we get here, we know size is power of two
130 * and it's value is between 512 and PAGE_SIZE */
131 sb->s_blocksize = size;
132 sb->s_blocksize_bits = blksize_bits(size);
133 return sb->s_blocksize;
136 EXPORT_SYMBOL(sb_set_blocksize);
138 int sb_min_blocksize(struct super_block *sb, int size)
140 int minsize = bdev_logical_block_size(sb->s_bdev);
143 return sb_set_blocksize(sb, size);
146 EXPORT_SYMBOL(sb_min_blocksize);
149 blkdev_get_block(struct inode *inode, sector_t iblock,
150 struct buffer_head *bh, int create)
152 bh->b_bdev = I_BDEV(inode);
153 bh->b_blocknr = iblock;
154 set_buffer_mapped(bh);
159 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
161 struct file *file = iocb->ki_filp;
162 struct inode *inode = file->f_mapping->host;
165 return dax_do_io(iocb, inode, iter, offset, blkdev_get_block,
166 NULL, DIO_SKIP_DIO_COUNT);
167 return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter, offset,
168 blkdev_get_block, NULL, NULL,
172 int __sync_blockdev(struct block_device *bdev, int wait)
177 return filemap_flush(bdev->bd_inode->i_mapping);
178 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
182 * Write out and wait upon all the dirty data associated with a block
183 * device via its mapping. Does not take the superblock lock.
185 int sync_blockdev(struct block_device *bdev)
187 return __sync_blockdev(bdev, 1);
189 EXPORT_SYMBOL(sync_blockdev);
192 * Write out and wait upon all dirty data associated with this
193 * device. Filesystem data as well as the underlying block
194 * device. Takes the superblock lock.
196 int fsync_bdev(struct block_device *bdev)
198 struct super_block *sb = get_super(bdev);
200 int res = sync_filesystem(sb);
204 return sync_blockdev(bdev);
206 EXPORT_SYMBOL(fsync_bdev);
209 * freeze_bdev -- lock a filesystem and force it into a consistent state
210 * @bdev: blockdevice to lock
212 * If a superblock is found on this device, we take the s_umount semaphore
213 * on it to make sure nobody unmounts until the snapshot creation is done.
214 * The reference counter (bd_fsfreeze_count) guarantees that only the last
215 * unfreeze process can unfreeze the frozen filesystem actually when multiple
216 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
217 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
220 struct super_block *freeze_bdev(struct block_device *bdev)
222 struct super_block *sb;
225 mutex_lock(&bdev->bd_fsfreeze_mutex);
226 if (++bdev->bd_fsfreeze_count > 1) {
228 * We don't even need to grab a reference - the first call
229 * to freeze_bdev grab an active reference and only the last
230 * thaw_bdev drops it.
232 sb = get_super(bdev);
234 mutex_unlock(&bdev->bd_fsfreeze_mutex);
238 sb = get_active_super(bdev);
241 if (sb->s_op->freeze_super)
242 error = sb->s_op->freeze_super(sb);
244 error = freeze_super(sb);
246 deactivate_super(sb);
247 bdev->bd_fsfreeze_count--;
248 mutex_unlock(&bdev->bd_fsfreeze_mutex);
249 return ERR_PTR(error);
251 deactivate_super(sb);
254 mutex_unlock(&bdev->bd_fsfreeze_mutex);
255 return sb; /* thaw_bdev releases s->s_umount */
257 EXPORT_SYMBOL(freeze_bdev);
260 * thaw_bdev -- unlock filesystem
261 * @bdev: blockdevice to unlock
262 * @sb: associated superblock
264 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
266 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
270 mutex_lock(&bdev->bd_fsfreeze_mutex);
271 if (!bdev->bd_fsfreeze_count)
275 if (--bdev->bd_fsfreeze_count > 0)
281 if (sb->s_op->thaw_super)
282 error = sb->s_op->thaw_super(sb);
284 error = thaw_super(sb);
286 bdev->bd_fsfreeze_count++;
287 mutex_unlock(&bdev->bd_fsfreeze_mutex);
291 mutex_unlock(&bdev->bd_fsfreeze_mutex);
294 EXPORT_SYMBOL(thaw_bdev);
296 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
298 return block_write_full_page(page, blkdev_get_block, wbc);
301 static int blkdev_readpage(struct file * file, struct page * page)
303 return block_read_full_page(page, blkdev_get_block);
306 static int blkdev_readpages(struct file *file, struct address_space *mapping,
307 struct list_head *pages, unsigned nr_pages)
309 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
312 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
313 loff_t pos, unsigned len, unsigned flags,
314 struct page **pagep, void **fsdata)
316 return block_write_begin(mapping, pos, len, flags, pagep,
320 static int blkdev_write_end(struct file *file, struct address_space *mapping,
321 loff_t pos, unsigned len, unsigned copied,
322 struct page *page, void *fsdata)
325 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
328 page_cache_release(page);
335 * for a block special file file_inode(file)->i_size is zero
336 * so we compute the size by hand (just as in block_read/write above)
338 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
340 struct inode *bd_inode = file->f_mapping->host;
343 mutex_lock(&bd_inode->i_mutex);
344 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
345 mutex_unlock(&bd_inode->i_mutex);
349 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
351 struct inode *bd_inode = filp->f_mapping->host;
352 struct block_device *bdev = I_BDEV(bd_inode);
355 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
360 * There is no need to serialise calls to blkdev_issue_flush with
361 * i_mutex and doing so causes performance issues with concurrent
362 * O_SYNC writers to a block device.
364 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
365 if (error == -EOPNOTSUPP)
370 EXPORT_SYMBOL(blkdev_fsync);
373 * bdev_read_page() - Start reading a page from a block device
374 * @bdev: The device to read the page from
375 * @sector: The offset on the device to read the page to (need not be aligned)
376 * @page: The page to read
378 * On entry, the page should be locked. It will be unlocked when the page
379 * has been read. If the block driver implements rw_page synchronously,
380 * that will be true on exit from this function, but it need not be.
382 * Errors returned by this function are usually "soft", eg out of memory, or
383 * queue full; callers should try a different route to read this page rather
384 * than propagate an error back up the stack.
386 * Return: negative errno if an error occurs, 0 if submission was successful.
388 int bdev_read_page(struct block_device *bdev, sector_t sector,
391 const struct block_device_operations *ops = bdev->bd_disk->fops;
392 int result = -EOPNOTSUPP;
394 if (!ops->rw_page || bdev_get_integrity(bdev))
397 result = blk_queue_enter(bdev->bd_queue, GFP_KERNEL);
400 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
401 blk_queue_exit(bdev->bd_queue);
404 EXPORT_SYMBOL_GPL(bdev_read_page);
407 * bdev_write_page() - Start writing a page to a block device
408 * @bdev: The device to write the page to
409 * @sector: The offset on the device to write the page to (need not be aligned)
410 * @page: The page to write
411 * @wbc: The writeback_control for the write
413 * On entry, the page should be locked and not currently under writeback.
414 * On exit, if the write started successfully, the page will be unlocked and
415 * under writeback. If the write failed already (eg the driver failed to
416 * queue the page to the device), the page will still be locked. If the
417 * caller is a ->writepage implementation, it will need to unlock the page.
419 * Errors returned by this function are usually "soft", eg out of memory, or
420 * queue full; callers should try a different route to write this page rather
421 * than propagate an error back up the stack.
423 * Return: negative errno if an error occurs, 0 if submission was successful.
425 int bdev_write_page(struct block_device *bdev, sector_t sector,
426 struct page *page, struct writeback_control *wbc)
429 int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
430 const struct block_device_operations *ops = bdev->bd_disk->fops;
432 if (!ops->rw_page || bdev_get_integrity(bdev))
434 result = blk_queue_enter(bdev->bd_queue, GFP_KERNEL);
438 set_page_writeback(page);
439 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
441 end_page_writeback(page);
444 blk_queue_exit(bdev->bd_queue);
447 EXPORT_SYMBOL_GPL(bdev_write_page);
450 * bdev_direct_access() - Get the address for directly-accessibly memory
451 * @bdev: The device containing the memory
452 * @sector: The offset within the device
453 * @addr: Where to put the address of the memory
454 * @pfn: The Page Frame Number for the memory
455 * @size: The number of bytes requested
457 * If a block device is made up of directly addressable memory, this function
458 * will tell the caller the PFN and the address of the memory. The address
459 * may be directly dereferenced within the kernel without the need to call
460 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
463 * Return: negative errno if an error occurs, otherwise the number of bytes
464 * accessible at this address.
466 long bdev_direct_access(struct block_device *bdev, sector_t sector,
467 void __pmem **addr, unsigned long *pfn, long size)
470 const struct block_device_operations *ops = bdev->bd_disk->fops;
473 * The device driver is allowed to sleep, in order to make the
474 * memory directly accessible.
480 if (!ops->direct_access)
482 if ((sector + DIV_ROUND_UP(size, 512)) >
483 part_nr_sects_read(bdev->bd_part))
485 sector += get_start_sect(bdev);
486 if (sector % (PAGE_SIZE / 512))
488 avail = ops->direct_access(bdev, sector, addr, pfn);
491 return min(avail, size);
493 EXPORT_SYMBOL_GPL(bdev_direct_access);
499 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
500 static struct kmem_cache * bdev_cachep __read_mostly;
502 static struct inode *bdev_alloc_inode(struct super_block *sb)
504 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
507 return &ei->vfs_inode;
510 static void bdev_i_callback(struct rcu_head *head)
512 struct inode *inode = container_of(head, struct inode, i_rcu);
513 struct bdev_inode *bdi = BDEV_I(inode);
515 kmem_cache_free(bdev_cachep, bdi);
518 static void bdev_destroy_inode(struct inode *inode)
520 call_rcu(&inode->i_rcu, bdev_i_callback);
523 static void init_once(void *foo)
525 struct bdev_inode *ei = (struct bdev_inode *) foo;
526 struct block_device *bdev = &ei->bdev;
528 memset(bdev, 0, sizeof(*bdev));
529 mutex_init(&bdev->bd_mutex);
530 INIT_LIST_HEAD(&bdev->bd_inodes);
531 INIT_LIST_HEAD(&bdev->bd_list);
533 INIT_LIST_HEAD(&bdev->bd_holder_disks);
535 inode_init_once(&ei->vfs_inode);
536 /* Initialize mutex for freeze. */
537 mutex_init(&bdev->bd_fsfreeze_mutex);
540 static inline void __bd_forget(struct inode *inode)
542 list_del_init(&inode->i_devices);
543 inode->i_bdev = NULL;
544 inode->i_mapping = &inode->i_data;
547 static void bdev_evict_inode(struct inode *inode)
549 struct block_device *bdev = &BDEV_I(inode)->bdev;
551 truncate_inode_pages_final(&inode->i_data);
552 invalidate_inode_buffers(inode); /* is it needed here? */
554 spin_lock(&bdev_lock);
555 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
556 __bd_forget(list_entry(p, struct inode, i_devices));
558 list_del_init(&bdev->bd_list);
559 spin_unlock(&bdev_lock);
562 static const struct super_operations bdev_sops = {
563 .statfs = simple_statfs,
564 .alloc_inode = bdev_alloc_inode,
565 .destroy_inode = bdev_destroy_inode,
566 .drop_inode = generic_delete_inode,
567 .evict_inode = bdev_evict_inode,
570 static struct dentry *bd_mount(struct file_system_type *fs_type,
571 int flags, const char *dev_name, void *data)
573 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
576 static struct file_system_type bd_type = {
579 .kill_sb = kill_anon_super,
582 struct super_block *blockdev_superblock __read_mostly;
583 EXPORT_SYMBOL_GPL(blockdev_superblock);
585 void __init bdev_cache_init(void)
588 static struct vfsmount *bd_mnt;
590 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
591 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
592 SLAB_MEM_SPREAD|SLAB_PANIC),
594 err = register_filesystem(&bd_type);
596 panic("Cannot register bdev pseudo-fs");
597 bd_mnt = kern_mount(&bd_type);
599 panic("Cannot create bdev pseudo-fs");
600 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
604 * Most likely _very_ bad one - but then it's hardly critical for small
605 * /dev and can be fixed when somebody will need really large one.
606 * Keep in mind that it will be fed through icache hash function too.
608 static inline unsigned long hash(dev_t dev)
610 return MAJOR(dev)+MINOR(dev);
613 static int bdev_test(struct inode *inode, void *data)
615 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
618 static int bdev_set(struct inode *inode, void *data)
620 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
624 static LIST_HEAD(all_bdevs);
626 struct block_device *bdget(dev_t dev)
628 struct block_device *bdev;
631 inode = iget5_locked(blockdev_superblock, hash(dev),
632 bdev_test, bdev_set, &dev);
637 bdev = &BDEV_I(inode)->bdev;
639 if (inode->i_state & I_NEW) {
640 bdev->bd_contains = NULL;
641 bdev->bd_super = NULL;
642 bdev->bd_inode = inode;
643 bdev->bd_block_size = (1 << inode->i_blkbits);
644 bdev->bd_part_count = 0;
645 bdev->bd_invalidated = 0;
646 inode->i_mode = S_IFBLK;
648 inode->i_bdev = bdev;
649 inode->i_data.a_ops = &def_blk_aops;
650 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
651 spin_lock(&bdev_lock);
652 list_add(&bdev->bd_list, &all_bdevs);
653 spin_unlock(&bdev_lock);
654 unlock_new_inode(inode);
659 EXPORT_SYMBOL(bdget);
662 * bdgrab -- Grab a reference to an already referenced block device
663 * @bdev: Block device to grab a reference to.
665 struct block_device *bdgrab(struct block_device *bdev)
667 ihold(bdev->bd_inode);
670 EXPORT_SYMBOL(bdgrab);
672 long nr_blockdev_pages(void)
674 struct block_device *bdev;
676 spin_lock(&bdev_lock);
677 list_for_each_entry(bdev, &all_bdevs, bd_list) {
678 ret += bdev->bd_inode->i_mapping->nrpages;
680 spin_unlock(&bdev_lock);
684 void bdput(struct block_device *bdev)
686 iput(bdev->bd_inode);
689 EXPORT_SYMBOL(bdput);
691 static struct block_device *bd_acquire(struct inode *inode)
693 struct block_device *bdev;
695 spin_lock(&bdev_lock);
696 bdev = inode->i_bdev;
698 ihold(bdev->bd_inode);
699 spin_unlock(&bdev_lock);
702 spin_unlock(&bdev_lock);
704 bdev = bdget(inode->i_rdev);
706 spin_lock(&bdev_lock);
707 if (!inode->i_bdev) {
709 * We take an additional reference to bd_inode,
710 * and it's released in clear_inode() of inode.
711 * So, we can access it via ->i_mapping always
714 ihold(bdev->bd_inode);
715 inode->i_bdev = bdev;
716 inode->i_mapping = bdev->bd_inode->i_mapping;
717 list_add(&inode->i_devices, &bdev->bd_inodes);
719 spin_unlock(&bdev_lock);
724 /* Call when you free inode */
726 void bd_forget(struct inode *inode)
728 struct block_device *bdev = NULL;
730 spin_lock(&bdev_lock);
731 if (!sb_is_blkdev_sb(inode->i_sb))
732 bdev = inode->i_bdev;
734 spin_unlock(&bdev_lock);
737 iput(bdev->bd_inode);
741 * bd_may_claim - test whether a block device can be claimed
742 * @bdev: block device of interest
743 * @whole: whole block device containing @bdev, may equal @bdev
744 * @holder: holder trying to claim @bdev
746 * Test whether @bdev can be claimed by @holder.
749 * spin_lock(&bdev_lock).
752 * %true if @bdev can be claimed, %false otherwise.
754 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
757 if (bdev->bd_holder == holder)
758 return true; /* already a holder */
759 else if (bdev->bd_holder != NULL)
760 return false; /* held by someone else */
761 else if (whole == bdev)
762 return true; /* is a whole device which isn't held */
764 else if (whole->bd_holder == bd_may_claim)
765 return true; /* is a partition of a device that is being partitioned */
766 else if (whole->bd_holder != NULL)
767 return false; /* is a partition of a held device */
769 return true; /* is a partition of an un-held device */
773 * bd_prepare_to_claim - prepare to claim a block device
774 * @bdev: block device of interest
775 * @whole: the whole device containing @bdev, may equal @bdev
776 * @holder: holder trying to claim @bdev
778 * Prepare to claim @bdev. This function fails if @bdev is already
779 * claimed by another holder and waits if another claiming is in
780 * progress. This function doesn't actually claim. On successful
781 * return, the caller has ownership of bd_claiming and bd_holder[s].
784 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
788 * 0 if @bdev can be claimed, -EBUSY otherwise.
790 static int bd_prepare_to_claim(struct block_device *bdev,
791 struct block_device *whole, void *holder)
794 /* if someone else claimed, fail */
795 if (!bd_may_claim(bdev, whole, holder))
798 /* if claiming is already in progress, wait for it to finish */
799 if (whole->bd_claiming) {
800 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
803 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
804 spin_unlock(&bdev_lock);
806 finish_wait(wq, &wait);
807 spin_lock(&bdev_lock);
816 * bd_start_claiming - start claiming a block device
817 * @bdev: block device of interest
818 * @holder: holder trying to claim @bdev
820 * @bdev is about to be opened exclusively. Check @bdev can be opened
821 * exclusively and mark that an exclusive open is in progress. Each
822 * successful call to this function must be matched with a call to
823 * either bd_finish_claiming() or bd_abort_claiming() (which do not
826 * This function is used to gain exclusive access to the block device
827 * without actually causing other exclusive open attempts to fail. It
828 * should be used when the open sequence itself requires exclusive
829 * access but may subsequently fail.
835 * Pointer to the block device containing @bdev on success, ERR_PTR()
838 static struct block_device *bd_start_claiming(struct block_device *bdev,
841 struct gendisk *disk;
842 struct block_device *whole;
848 * @bdev might not have been initialized properly yet, look up
849 * and grab the outer block device the hard way.
851 disk = get_gendisk(bdev->bd_dev, &partno);
853 return ERR_PTR(-ENXIO);
856 * Normally, @bdev should equal what's returned from bdget_disk()
857 * if partno is 0; however, some drivers (floppy) use multiple
858 * bdev's for the same physical device and @bdev may be one of the
859 * aliases. Keep @bdev if partno is 0. This means claimer
860 * tracking is broken for those devices but it has always been that
864 whole = bdget_disk(disk, 0);
866 whole = bdgrab(bdev);
868 module_put(disk->fops->owner);
871 return ERR_PTR(-ENOMEM);
873 /* prepare to claim, if successful, mark claiming in progress */
874 spin_lock(&bdev_lock);
876 err = bd_prepare_to_claim(bdev, whole, holder);
878 whole->bd_claiming = holder;
879 spin_unlock(&bdev_lock);
882 spin_unlock(&bdev_lock);
889 struct bd_holder_disk {
890 struct list_head list;
891 struct gendisk *disk;
895 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
896 struct gendisk *disk)
898 struct bd_holder_disk *holder;
900 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
901 if (holder->disk == disk)
906 static int add_symlink(struct kobject *from, struct kobject *to)
908 return sysfs_create_link(from, to, kobject_name(to));
911 static void del_symlink(struct kobject *from, struct kobject *to)
913 sysfs_remove_link(from, kobject_name(to));
917 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
918 * @bdev: the claimed slave bdev
919 * @disk: the holding disk
921 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
923 * This functions creates the following sysfs symlinks.
925 * - from "slaves" directory of the holder @disk to the claimed @bdev
926 * - from "holders" directory of the @bdev to the holder @disk
928 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
929 * passed to bd_link_disk_holder(), then:
931 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
932 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
934 * The caller must have claimed @bdev before calling this function and
935 * ensure that both @bdev and @disk are valid during the creation and
936 * lifetime of these symlinks.
942 * 0 on success, -errno on failure.
944 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
946 struct bd_holder_disk *holder;
949 mutex_lock(&bdev->bd_mutex);
951 WARN_ON_ONCE(!bdev->bd_holder);
953 /* FIXME: remove the following once add_disk() handles errors */
954 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
957 holder = bd_find_holder_disk(bdev, disk);
963 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
969 INIT_LIST_HEAD(&holder->list);
973 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
977 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
981 * bdev could be deleted beneath us which would implicitly destroy
982 * the holder directory. Hold on to it.
984 kobject_get(bdev->bd_part->holder_dir);
986 list_add(&holder->list, &bdev->bd_holder_disks);
990 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
994 mutex_unlock(&bdev->bd_mutex);
997 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1000 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1001 * @bdev: the calimed slave bdev
1002 * @disk: the holding disk
1004 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1009 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1011 struct bd_holder_disk *holder;
1013 mutex_lock(&bdev->bd_mutex);
1015 holder = bd_find_holder_disk(bdev, disk);
1017 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1018 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1019 del_symlink(bdev->bd_part->holder_dir,
1020 &disk_to_dev(disk)->kobj);
1021 kobject_put(bdev->bd_part->holder_dir);
1022 list_del_init(&holder->list);
1026 mutex_unlock(&bdev->bd_mutex);
1028 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1032 * flush_disk - invalidates all buffer-cache entries on a disk
1034 * @bdev: struct block device to be flushed
1035 * @kill_dirty: flag to guide handling of dirty inodes
1037 * Invalidates all buffer-cache entries on a disk. It should be called
1038 * when a disk has been changed -- either by a media change or online
1041 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1043 if (__invalidate_device(bdev, kill_dirty)) {
1044 char name[BDEVNAME_SIZE] = "";
1047 disk_name(bdev->bd_disk, 0, name);
1048 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1049 "resized disk %s\n", name);
1054 if (disk_part_scan_enabled(bdev->bd_disk))
1055 bdev->bd_invalidated = 1;
1059 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1060 * @disk: struct gendisk to check
1061 * @bdev: struct bdev to adjust.
1063 * This routine checks to see if the bdev size does not match the disk size
1064 * and adjusts it if it differs.
1066 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1068 loff_t disk_size, bdev_size;
1070 disk_size = (loff_t)get_capacity(disk) << 9;
1071 bdev_size = i_size_read(bdev->bd_inode);
1072 if (disk_size != bdev_size) {
1073 char name[BDEVNAME_SIZE];
1075 disk_name(disk, 0, name);
1077 "%s: detected capacity change from %lld to %lld\n",
1078 name, bdev_size, disk_size);
1079 i_size_write(bdev->bd_inode, disk_size);
1080 flush_disk(bdev, false);
1083 EXPORT_SYMBOL(check_disk_size_change);
1086 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1087 * @disk: struct gendisk to be revalidated
1089 * This routine is a wrapper for lower-level driver's revalidate_disk
1090 * call-backs. It is used to do common pre and post operations needed
1091 * for all revalidate_disk operations.
1093 int revalidate_disk(struct gendisk *disk)
1095 struct block_device *bdev;
1098 if (disk->fops->revalidate_disk)
1099 ret = disk->fops->revalidate_disk(disk);
1100 bdev = bdget_disk(disk, 0);
1104 mutex_lock(&bdev->bd_mutex);
1105 check_disk_size_change(disk, bdev);
1106 bdev->bd_invalidated = 0;
1107 mutex_unlock(&bdev->bd_mutex);
1111 EXPORT_SYMBOL(revalidate_disk);
1114 * This routine checks whether a removable media has been changed,
1115 * and invalidates all buffer-cache-entries in that case. This
1116 * is a relatively slow routine, so we have to try to minimize using
1117 * it. Thus it is called only upon a 'mount' or 'open'. This
1118 * is the best way of combining speed and utility, I think.
1119 * People changing diskettes in the middle of an operation deserve
1122 int check_disk_change(struct block_device *bdev)
1124 struct gendisk *disk = bdev->bd_disk;
1125 const struct block_device_operations *bdops = disk->fops;
1126 unsigned int events;
1128 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1129 DISK_EVENT_EJECT_REQUEST);
1130 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1133 flush_disk(bdev, true);
1134 if (bdops->revalidate_disk)
1135 bdops->revalidate_disk(bdev->bd_disk);
1139 EXPORT_SYMBOL(check_disk_change);
1141 void bd_set_size(struct block_device *bdev, loff_t size)
1143 unsigned bsize = bdev_logical_block_size(bdev);
1145 mutex_lock(&bdev->bd_inode->i_mutex);
1146 i_size_write(bdev->bd_inode, size);
1147 mutex_unlock(&bdev->bd_inode->i_mutex);
1148 while (bsize < PAGE_CACHE_SIZE) {
1153 bdev->bd_block_size = bsize;
1154 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1156 EXPORT_SYMBOL(bd_set_size);
1158 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1163 * mutex_lock(part->bd_mutex)
1164 * mutex_lock_nested(whole->bd_mutex, 1)
1167 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1169 struct gendisk *disk;
1170 struct module *owner;
1175 if (mode & FMODE_READ)
1177 if (mode & FMODE_WRITE)
1180 * hooks: /n/, see "layering violations".
1183 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1191 disk = get_gendisk(bdev->bd_dev, &partno);
1194 owner = disk->fops->owner;
1196 disk_block_events(disk);
1197 mutex_lock_nested(&bdev->bd_mutex, for_part);
1198 if (!bdev->bd_openers) {
1199 bdev->bd_disk = disk;
1200 bdev->bd_queue = disk->queue;
1201 bdev->bd_contains = bdev;
1202 bdev->bd_inode->i_flags = disk->fops->direct_access ? S_DAX : 0;
1205 bdev->bd_part = disk_get_part(disk, partno);
1210 if (disk->fops->open) {
1211 ret = disk->fops->open(bdev, mode);
1212 if (ret == -ERESTARTSYS) {
1213 /* Lost a race with 'disk' being
1214 * deleted, try again.
1217 disk_put_part(bdev->bd_part);
1218 bdev->bd_part = NULL;
1219 bdev->bd_disk = NULL;
1220 bdev->bd_queue = NULL;
1221 mutex_unlock(&bdev->bd_mutex);
1222 disk_unblock_events(disk);
1230 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1233 * If the device is invalidated, rescan partition
1234 * if open succeeded or failed with -ENOMEDIUM.
1235 * The latter is necessary to prevent ghost
1236 * partitions on a removed medium.
1238 if (bdev->bd_invalidated) {
1240 rescan_partitions(disk, bdev);
1241 else if (ret == -ENOMEDIUM)
1242 invalidate_partitions(disk, bdev);
1247 struct block_device *whole;
1248 whole = bdget_disk(disk, 0);
1253 ret = __blkdev_get(whole, mode, 1);
1258 bdev->bd_contains = whole;
1259 bdev->bd_part = disk_get_part(disk, partno);
1260 if (!(disk->flags & GENHD_FL_UP) ||
1261 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1265 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1267 * If the partition is not aligned on a page
1268 * boundary, we can't do dax I/O to it.
1270 if ((bdev->bd_part->start_sect % (PAGE_SIZE / 512)) ||
1271 (bdev->bd_part->nr_sects % (PAGE_SIZE / 512)))
1272 bdev->bd_inode->i_flags &= ~S_DAX;
1275 if (bdev->bd_contains == bdev) {
1277 if (bdev->bd_disk->fops->open)
1278 ret = bdev->bd_disk->fops->open(bdev, mode);
1279 /* the same as first opener case, read comment there */
1280 if (bdev->bd_invalidated) {
1282 rescan_partitions(bdev->bd_disk, bdev);
1283 else if (ret == -ENOMEDIUM)
1284 invalidate_partitions(bdev->bd_disk, bdev);
1287 goto out_unlock_bdev;
1289 /* only one opener holds refs to the module and disk */
1295 bdev->bd_part_count++;
1296 mutex_unlock(&bdev->bd_mutex);
1297 disk_unblock_events(disk);
1301 disk_put_part(bdev->bd_part);
1302 bdev->bd_disk = NULL;
1303 bdev->bd_part = NULL;
1304 bdev->bd_queue = NULL;
1305 if (bdev != bdev->bd_contains)
1306 __blkdev_put(bdev->bd_contains, mode, 1);
1307 bdev->bd_contains = NULL;
1309 mutex_unlock(&bdev->bd_mutex);
1310 disk_unblock_events(disk);
1319 * blkdev_get - open a block device
1320 * @bdev: block_device to open
1321 * @mode: FMODE_* mask
1322 * @holder: exclusive holder identifier
1324 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1325 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1326 * @holder is invalid. Exclusive opens may nest for the same @holder.
1328 * On success, the reference count of @bdev is unchanged. On failure,
1335 * 0 on success, -errno on failure.
1337 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1339 struct block_device *whole = NULL;
1342 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1344 if ((mode & FMODE_EXCL) && holder) {
1345 whole = bd_start_claiming(bdev, holder);
1346 if (IS_ERR(whole)) {
1348 return PTR_ERR(whole);
1352 res = __blkdev_get(bdev, mode, 0);
1355 struct gendisk *disk = whole->bd_disk;
1357 /* finish claiming */
1358 mutex_lock(&bdev->bd_mutex);
1359 spin_lock(&bdev_lock);
1362 BUG_ON(!bd_may_claim(bdev, whole, holder));
1364 * Note that for a whole device bd_holders
1365 * will be incremented twice, and bd_holder
1366 * will be set to bd_may_claim before being
1369 whole->bd_holders++;
1370 whole->bd_holder = bd_may_claim;
1372 bdev->bd_holder = holder;
1375 /* tell others that we're done */
1376 BUG_ON(whole->bd_claiming != holder);
1377 whole->bd_claiming = NULL;
1378 wake_up_bit(&whole->bd_claiming, 0);
1380 spin_unlock(&bdev_lock);
1383 * Block event polling for write claims if requested. Any
1384 * write holder makes the write_holder state stick until
1385 * all are released. This is good enough and tracking
1386 * individual writeable reference is too fragile given the
1387 * way @mode is used in blkdev_get/put().
1389 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1390 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1391 bdev->bd_write_holder = true;
1392 disk_block_events(disk);
1395 mutex_unlock(&bdev->bd_mutex);
1404 EXPORT_SYMBOL(blkdev_get);
1407 * blkdev_get_by_path - open a block device by name
1408 * @path: path to the block device to open
1409 * @mode: FMODE_* mask
1410 * @holder: exclusive holder identifier
1412 * Open the blockdevice described by the device file at @path. @mode
1413 * and @holder are identical to blkdev_get().
1415 * On success, the returned block_device has reference count of one.
1421 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1423 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1426 struct block_device *bdev;
1429 bdev = lookup_bdev(path);
1433 err = blkdev_get(bdev, mode, holder);
1435 return ERR_PTR(err);
1437 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1438 blkdev_put(bdev, mode);
1439 return ERR_PTR(-EACCES);
1444 EXPORT_SYMBOL(blkdev_get_by_path);
1447 * blkdev_get_by_dev - open a block device by device number
1448 * @dev: device number of block device to open
1449 * @mode: FMODE_* mask
1450 * @holder: exclusive holder identifier
1452 * Open the blockdevice described by device number @dev. @mode and
1453 * @holder are identical to blkdev_get().
1455 * Use it ONLY if you really do not have anything better - i.e. when
1456 * you are behind a truly sucky interface and all you are given is a
1457 * device number. _Never_ to be used for internal purposes. If you
1458 * ever need it - reconsider your API.
1460 * On success, the returned block_device has reference count of one.
1466 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1468 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1470 struct block_device *bdev;
1475 return ERR_PTR(-ENOMEM);
1477 err = blkdev_get(bdev, mode, holder);
1479 return ERR_PTR(err);
1483 EXPORT_SYMBOL(blkdev_get_by_dev);
1485 static int blkdev_open(struct inode * inode, struct file * filp)
1487 struct block_device *bdev;
1490 * Preserve backwards compatibility and allow large file access
1491 * even if userspace doesn't ask for it explicitly. Some mkfs
1492 * binary needs it. We might want to drop this workaround
1493 * during an unstable branch.
1495 filp->f_flags |= O_LARGEFILE;
1497 if (filp->f_flags & O_NDELAY)
1498 filp->f_mode |= FMODE_NDELAY;
1499 if (filp->f_flags & O_EXCL)
1500 filp->f_mode |= FMODE_EXCL;
1501 if ((filp->f_flags & O_ACCMODE) == 3)
1502 filp->f_mode |= FMODE_WRITE_IOCTL;
1504 bdev = bd_acquire(inode);
1508 filp->f_mapping = bdev->bd_inode->i_mapping;
1510 return blkdev_get(bdev, filp->f_mode, filp);
1513 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1515 struct gendisk *disk = bdev->bd_disk;
1516 struct block_device *victim = NULL;
1519 * Sync early if it looks like we're the last one. If someone else
1520 * opens the block device between now and the decrement of bd_openers
1521 * then we did a sync that we didn't need to, but that's not the end
1522 * of the world and we want to avoid long (could be several minute)
1523 * syncs while holding the mutex.
1525 if (bdev->bd_openers == 1)
1526 sync_blockdev(bdev);
1528 mutex_lock_nested(&bdev->bd_mutex, for_part);
1530 bdev->bd_part_count--;
1532 if (!--bdev->bd_openers) {
1533 WARN_ON_ONCE(bdev->bd_holders);
1534 sync_blockdev(bdev);
1537 bdev_write_inode(bdev);
1539 * Detaching bdev inode from its wb in __destroy_inode()
1540 * is too late: the queue which embeds its bdi (along with
1541 * root wb) can be gone as soon as we put_disk() below.
1543 inode_detach_wb(bdev->bd_inode);
1545 if (bdev->bd_contains == bdev) {
1546 if (disk->fops->release)
1547 disk->fops->release(disk, mode);
1549 if (!bdev->bd_openers) {
1550 struct module *owner = disk->fops->owner;
1552 disk_put_part(bdev->bd_part);
1553 bdev->bd_part = NULL;
1554 bdev->bd_disk = NULL;
1555 if (bdev != bdev->bd_contains)
1556 victim = bdev->bd_contains;
1557 bdev->bd_contains = NULL;
1562 mutex_unlock(&bdev->bd_mutex);
1565 __blkdev_put(victim, mode, 1);
1568 void blkdev_put(struct block_device *bdev, fmode_t mode)
1570 mutex_lock(&bdev->bd_mutex);
1572 if (mode & FMODE_EXCL) {
1576 * Release a claim on the device. The holder fields
1577 * are protected with bdev_lock. bd_mutex is to
1578 * synchronize disk_holder unlinking.
1580 spin_lock(&bdev_lock);
1582 WARN_ON_ONCE(--bdev->bd_holders < 0);
1583 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1585 /* bd_contains might point to self, check in a separate step */
1586 if ((bdev_free = !bdev->bd_holders))
1587 bdev->bd_holder = NULL;
1588 if (!bdev->bd_contains->bd_holders)
1589 bdev->bd_contains->bd_holder = NULL;
1591 spin_unlock(&bdev_lock);
1594 * If this was the last claim, remove holder link and
1595 * unblock evpoll if it was a write holder.
1597 if (bdev_free && bdev->bd_write_holder) {
1598 disk_unblock_events(bdev->bd_disk);
1599 bdev->bd_write_holder = false;
1604 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1605 * event. This is to ensure detection of media removal commanded
1606 * from userland - e.g. eject(1).
1608 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1610 mutex_unlock(&bdev->bd_mutex);
1612 __blkdev_put(bdev, mode, 0);
1614 EXPORT_SYMBOL(blkdev_put);
1616 static int blkdev_close(struct inode * inode, struct file * filp)
1618 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1619 blkdev_put(bdev, filp->f_mode);
1623 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1625 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1626 fmode_t mode = file->f_mode;
1629 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1630 * to updated it before every ioctl.
1632 if (file->f_flags & O_NDELAY)
1633 mode |= FMODE_NDELAY;
1635 mode &= ~FMODE_NDELAY;
1637 return blkdev_ioctl(bdev, mode, cmd, arg);
1641 * Write data to the block device. Only intended for the block device itself
1642 * and the raw driver which basically is a fake block device.
1644 * Does not take i_mutex for the write and thus is not for general purpose
1647 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1649 struct file *file = iocb->ki_filp;
1650 struct inode *bd_inode = file->f_mapping->host;
1651 loff_t size = i_size_read(bd_inode);
1652 struct blk_plug plug;
1655 if (bdev_read_only(I_BDEV(bd_inode)))
1658 if (!iov_iter_count(from))
1661 if (iocb->ki_pos >= size)
1664 iov_iter_truncate(from, size - iocb->ki_pos);
1666 blk_start_plug(&plug);
1667 ret = __generic_file_write_iter(iocb, from);
1670 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
1674 blk_finish_plug(&plug);
1677 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1679 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1681 struct file *file = iocb->ki_filp;
1682 struct inode *bd_inode = file->f_mapping->host;
1683 loff_t size = i_size_read(bd_inode);
1684 loff_t pos = iocb->ki_pos;
1690 iov_iter_truncate(to, size);
1691 return generic_file_read_iter(iocb, to);
1693 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1696 * Try to release a page associated with block device when the system
1697 * is under memory pressure.
1699 static int blkdev_releasepage(struct page *page, gfp_t wait)
1701 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1703 if (super && super->s_op->bdev_try_to_free_page)
1704 return super->s_op->bdev_try_to_free_page(super, page, wait);
1706 return try_to_free_buffers(page);
1709 static const struct address_space_operations def_blk_aops = {
1710 .readpage = blkdev_readpage,
1711 .readpages = blkdev_readpages,
1712 .writepage = blkdev_writepage,
1713 .write_begin = blkdev_write_begin,
1714 .write_end = blkdev_write_end,
1715 .writepages = generic_writepages,
1716 .releasepage = blkdev_releasepage,
1717 .direct_IO = blkdev_direct_IO,
1718 .is_dirty_writeback = buffer_check_dirty_writeback,
1721 const struct file_operations def_blk_fops = {
1722 .open = blkdev_open,
1723 .release = blkdev_close,
1724 .llseek = block_llseek,
1725 .read_iter = blkdev_read_iter,
1726 .write_iter = blkdev_write_iter,
1727 .mmap = generic_file_mmap,
1728 .fsync = blkdev_fsync,
1729 .unlocked_ioctl = block_ioctl,
1730 #ifdef CONFIG_COMPAT
1731 .compat_ioctl = compat_blkdev_ioctl,
1733 .splice_read = generic_file_splice_read,
1734 .splice_write = iter_file_splice_write,
1737 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1740 mm_segment_t old_fs = get_fs();
1742 res = blkdev_ioctl(bdev, 0, cmd, arg);
1747 EXPORT_SYMBOL(ioctl_by_bdev);
1750 * lookup_bdev - lookup a struct block_device by name
1751 * @pathname: special file representing the block device
1753 * Get a reference to the blockdevice at @pathname in the current
1754 * namespace if possible and return it. Return ERR_PTR(error)
1757 struct block_device *lookup_bdev(const char *pathname)
1759 struct block_device *bdev;
1760 struct inode *inode;
1764 if (!pathname || !*pathname)
1765 return ERR_PTR(-EINVAL);
1767 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1769 return ERR_PTR(error);
1771 inode = d_backing_inode(path.dentry);
1773 if (!S_ISBLK(inode->i_mode))
1776 if (path.mnt->mnt_flags & MNT_NODEV)
1779 bdev = bd_acquire(inode);
1786 bdev = ERR_PTR(error);
1789 EXPORT_SYMBOL(lookup_bdev);
1791 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1793 struct super_block *sb = get_super(bdev);
1798 * no need to lock the super, get_super holds the
1799 * read mutex so the filesystem cannot go away
1800 * under us (->put_super runs with the write lock
1803 shrink_dcache_sb(sb);
1804 res = invalidate_inodes(sb, kill_dirty);
1807 invalidate_bdev(bdev);
1810 EXPORT_SYMBOL(__invalidate_device);
1812 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1814 struct inode *inode, *old_inode = NULL;
1816 spin_lock(&blockdev_superblock->s_inode_list_lock);
1817 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1818 struct address_space *mapping = inode->i_mapping;
1819 struct block_device *bdev;
1821 spin_lock(&inode->i_lock);
1822 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1823 mapping->nrpages == 0) {
1824 spin_unlock(&inode->i_lock);
1828 spin_unlock(&inode->i_lock);
1829 spin_unlock(&blockdev_superblock->s_inode_list_lock);
1831 * We hold a reference to 'inode' so it couldn't have been
1832 * removed from s_inodes list while we dropped the
1833 * s_inode_list_lock We cannot iput the inode now as we can
1834 * be holding the last reference and we cannot iput it under
1835 * s_inode_list_lock. So we keep the reference and iput it
1840 bdev = I_BDEV(inode);
1842 mutex_lock(&bdev->bd_mutex);
1843 if (bdev->bd_openers)
1845 mutex_unlock(&bdev->bd_mutex);
1847 spin_lock(&blockdev_superblock->s_inode_list_lock);
1849 spin_unlock(&blockdev_superblock->s_inode_list_lock);