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 if (sb->s_op->freeze_super)
239 error = sb->s_op->freeze_super(sb);
241 error = freeze_super(sb);
243 deactivate_super(sb);
244 bdev->bd_fsfreeze_count--;
245 mutex_unlock(&bdev->bd_fsfreeze_mutex);
246 return ERR_PTR(error);
248 deactivate_super(sb);
251 mutex_unlock(&bdev->bd_fsfreeze_mutex);
252 return sb; /* thaw_bdev releases s->s_umount */
254 EXPORT_SYMBOL(freeze_bdev);
257 * thaw_bdev -- unlock filesystem
258 * @bdev: blockdevice to unlock
259 * @sb: associated superblock
261 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
263 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
267 mutex_lock(&bdev->bd_fsfreeze_mutex);
268 if (!bdev->bd_fsfreeze_count)
272 if (--bdev->bd_fsfreeze_count > 0)
278 if (sb->s_op->thaw_super)
279 error = sb->s_op->thaw_super(sb);
281 error = thaw_super(sb);
283 bdev->bd_fsfreeze_count++;
284 mutex_unlock(&bdev->bd_fsfreeze_mutex);
288 mutex_unlock(&bdev->bd_fsfreeze_mutex);
291 EXPORT_SYMBOL(thaw_bdev);
293 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
295 return block_write_full_page(page, blkdev_get_block, wbc);
298 static int blkdev_readpage(struct file * file, struct page * page)
300 return block_read_full_page(page, blkdev_get_block);
303 static int blkdev_readpages(struct file *file, struct address_space *mapping,
304 struct list_head *pages, unsigned nr_pages)
306 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
309 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
310 loff_t pos, unsigned len, unsigned flags,
311 struct page **pagep, void **fsdata)
313 return block_write_begin(mapping, pos, len, flags, pagep,
317 static int blkdev_write_end(struct file *file, struct address_space *mapping,
318 loff_t pos, unsigned len, unsigned copied,
319 struct page *page, void *fsdata)
322 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
325 page_cache_release(page);
332 * for a block special file file_inode(file)->i_size is zero
333 * so we compute the size by hand (just as in block_read/write above)
335 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
337 struct inode *bd_inode = file->f_mapping->host;
340 mutex_lock(&bd_inode->i_mutex);
341 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
342 mutex_unlock(&bd_inode->i_mutex);
346 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
348 struct inode *bd_inode = filp->f_mapping->host;
349 struct block_device *bdev = I_BDEV(bd_inode);
352 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
357 * There is no need to serialise calls to blkdev_issue_flush with
358 * i_mutex and doing so causes performance issues with concurrent
359 * O_SYNC writers to a block device.
361 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
362 if (error == -EOPNOTSUPP)
367 EXPORT_SYMBOL(blkdev_fsync);
370 * bdev_read_page() - Start reading a page from a block device
371 * @bdev: The device to read the page from
372 * @sector: The offset on the device to read the page to (need not be aligned)
373 * @page: The page to read
375 * On entry, the page should be locked. It will be unlocked when the page
376 * has been read. If the block driver implements rw_page synchronously,
377 * that will be true on exit from this function, but it need not be.
379 * Errors returned by this function are usually "soft", eg out of memory, or
380 * queue full; callers should try a different route to read this page rather
381 * than propagate an error back up the stack.
383 * Return: negative errno if an error occurs, 0 if submission was successful.
385 int bdev_read_page(struct block_device *bdev, sector_t sector,
388 const struct block_device_operations *ops = bdev->bd_disk->fops;
391 return ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
393 EXPORT_SYMBOL_GPL(bdev_read_page);
396 * bdev_write_page() - Start writing a page to a block device
397 * @bdev: The device to write the page to
398 * @sector: The offset on the device to write the page to (need not be aligned)
399 * @page: The page to write
400 * @wbc: The writeback_control for the write
402 * On entry, the page should be locked and not currently under writeback.
403 * On exit, if the write started successfully, the page will be unlocked and
404 * under writeback. If the write failed already (eg the driver failed to
405 * queue the page to the device), the page will still be locked. If the
406 * caller is a ->writepage implementation, it will need to unlock the page.
408 * Errors returned by this function are usually "soft", eg out of memory, or
409 * queue full; callers should try a different route to write this page rather
410 * than propagate an error back up the stack.
412 * Return: negative errno if an error occurs, 0 if submission was successful.
414 int bdev_write_page(struct block_device *bdev, sector_t sector,
415 struct page *page, struct writeback_control *wbc)
418 int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
419 const struct block_device_operations *ops = bdev->bd_disk->fops;
422 set_page_writeback(page);
423 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
425 end_page_writeback(page);
430 EXPORT_SYMBOL_GPL(bdev_write_page);
433 * bdev_direct_access() - Get the address for directly-accessibly memory
434 * @bdev: The device containing the memory
435 * @sector: The offset within the device
436 * @addr: Where to put the address of the memory
437 * @pfn: The Page Frame Number for the memory
438 * @size: The number of bytes requested
440 * If a block device is made up of directly addressable memory, this function
441 * will tell the caller the PFN and the address of the memory. The address
442 * may be directly dereferenced within the kernel without the need to call
443 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
446 * Return: negative errno if an error occurs, otherwise the number of bytes
447 * accessible at this address.
449 long bdev_direct_access(struct block_device *bdev, sector_t sector,
450 void **addr, unsigned long *pfn, long size)
453 const struct block_device_operations *ops = bdev->bd_disk->fops;
457 if (!ops->direct_access)
459 if ((sector + DIV_ROUND_UP(size, 512)) >
460 part_nr_sects_read(bdev->bd_part))
462 sector += get_start_sect(bdev);
463 if (sector % (PAGE_SIZE / 512))
465 avail = ops->direct_access(bdev, sector, addr, pfn, size);
468 return min(avail, size);
470 EXPORT_SYMBOL_GPL(bdev_direct_access);
476 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
477 static struct kmem_cache * bdev_cachep __read_mostly;
479 static struct inode *bdev_alloc_inode(struct super_block *sb)
481 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
484 return &ei->vfs_inode;
487 static void bdev_i_callback(struct rcu_head *head)
489 struct inode *inode = container_of(head, struct inode, i_rcu);
490 struct bdev_inode *bdi = BDEV_I(inode);
492 kmem_cache_free(bdev_cachep, bdi);
495 static void bdev_destroy_inode(struct inode *inode)
497 call_rcu(&inode->i_rcu, bdev_i_callback);
500 static void init_once(void *foo)
502 struct bdev_inode *ei = (struct bdev_inode *) foo;
503 struct block_device *bdev = &ei->bdev;
505 memset(bdev, 0, sizeof(*bdev));
506 mutex_init(&bdev->bd_mutex);
507 INIT_LIST_HEAD(&bdev->bd_inodes);
508 INIT_LIST_HEAD(&bdev->bd_list);
510 INIT_LIST_HEAD(&bdev->bd_holder_disks);
512 inode_init_once(&ei->vfs_inode);
513 /* Initialize mutex for freeze. */
514 mutex_init(&bdev->bd_fsfreeze_mutex);
517 static inline void __bd_forget(struct inode *inode)
519 list_del_init(&inode->i_devices);
520 inode->i_bdev = NULL;
521 inode->i_mapping = &inode->i_data;
524 static void bdev_evict_inode(struct inode *inode)
526 struct block_device *bdev = &BDEV_I(inode)->bdev;
528 truncate_inode_pages_final(&inode->i_data);
529 invalidate_inode_buffers(inode); /* is it needed here? */
531 spin_lock(&bdev_lock);
532 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
533 __bd_forget(list_entry(p, struct inode, i_devices));
535 list_del_init(&bdev->bd_list);
536 spin_unlock(&bdev_lock);
539 static const struct super_operations bdev_sops = {
540 .statfs = simple_statfs,
541 .alloc_inode = bdev_alloc_inode,
542 .destroy_inode = bdev_destroy_inode,
543 .drop_inode = generic_delete_inode,
544 .evict_inode = bdev_evict_inode,
547 static struct dentry *bd_mount(struct file_system_type *fs_type,
548 int flags, const char *dev_name, void *data)
550 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
553 static struct file_system_type bd_type = {
556 .kill_sb = kill_anon_super,
559 static struct super_block *blockdev_superblock __read_mostly;
561 void __init bdev_cache_init(void)
564 static struct vfsmount *bd_mnt;
566 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
567 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
568 SLAB_MEM_SPREAD|SLAB_PANIC),
570 err = register_filesystem(&bd_type);
572 panic("Cannot register bdev pseudo-fs");
573 bd_mnt = kern_mount(&bd_type);
575 panic("Cannot create bdev pseudo-fs");
576 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
580 * Most likely _very_ bad one - but then it's hardly critical for small
581 * /dev and can be fixed when somebody will need really large one.
582 * Keep in mind that it will be fed through icache hash function too.
584 static inline unsigned long hash(dev_t dev)
586 return MAJOR(dev)+MINOR(dev);
589 static int bdev_test(struct inode *inode, void *data)
591 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
594 static int bdev_set(struct inode *inode, void *data)
596 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
600 static LIST_HEAD(all_bdevs);
602 struct block_device *bdget(dev_t dev)
604 struct block_device *bdev;
607 inode = iget5_locked(blockdev_superblock, hash(dev),
608 bdev_test, bdev_set, &dev);
613 bdev = &BDEV_I(inode)->bdev;
615 if (inode->i_state & I_NEW) {
616 bdev->bd_contains = NULL;
617 bdev->bd_super = NULL;
618 bdev->bd_inode = inode;
619 bdev->bd_block_size = (1 << inode->i_blkbits);
620 bdev->bd_part_count = 0;
621 bdev->bd_invalidated = 0;
622 inode->i_mode = S_IFBLK;
624 inode->i_bdev = bdev;
625 inode->i_data.a_ops = &def_blk_aops;
626 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
627 inode->i_data.backing_dev_info = &default_backing_dev_info;
628 spin_lock(&bdev_lock);
629 list_add(&bdev->bd_list, &all_bdevs);
630 spin_unlock(&bdev_lock);
631 unlock_new_inode(inode);
636 EXPORT_SYMBOL(bdget);
639 * bdgrab -- Grab a reference to an already referenced block device
640 * @bdev: Block device to grab a reference to.
642 struct block_device *bdgrab(struct block_device *bdev)
644 ihold(bdev->bd_inode);
647 EXPORT_SYMBOL(bdgrab);
649 long nr_blockdev_pages(void)
651 struct block_device *bdev;
653 spin_lock(&bdev_lock);
654 list_for_each_entry(bdev, &all_bdevs, bd_list) {
655 ret += bdev->bd_inode->i_mapping->nrpages;
657 spin_unlock(&bdev_lock);
661 void bdput(struct block_device *bdev)
663 iput(bdev->bd_inode);
666 EXPORT_SYMBOL(bdput);
668 static struct block_device *bd_acquire(struct inode *inode)
670 struct block_device *bdev;
672 spin_lock(&bdev_lock);
673 bdev = inode->i_bdev;
675 ihold(bdev->bd_inode);
676 spin_unlock(&bdev_lock);
679 spin_unlock(&bdev_lock);
681 bdev = bdget(inode->i_rdev);
683 spin_lock(&bdev_lock);
684 if (!inode->i_bdev) {
686 * We take an additional reference to bd_inode,
687 * and it's released in clear_inode() of inode.
688 * So, we can access it via ->i_mapping always
691 ihold(bdev->bd_inode);
692 inode->i_bdev = bdev;
693 inode->i_mapping = bdev->bd_inode->i_mapping;
694 list_add(&inode->i_devices, &bdev->bd_inodes);
696 spin_unlock(&bdev_lock);
701 int sb_is_blkdev_sb(struct super_block *sb)
703 return sb == blockdev_superblock;
706 /* Call when you free inode */
708 void bd_forget(struct inode *inode)
710 struct block_device *bdev = NULL;
712 spin_lock(&bdev_lock);
713 if (!sb_is_blkdev_sb(inode->i_sb))
714 bdev = inode->i_bdev;
716 spin_unlock(&bdev_lock);
719 iput(bdev->bd_inode);
723 * bd_may_claim - test whether a block device can be claimed
724 * @bdev: block device of interest
725 * @whole: whole block device containing @bdev, may equal @bdev
726 * @holder: holder trying to claim @bdev
728 * Test whether @bdev can be claimed by @holder.
731 * spin_lock(&bdev_lock).
734 * %true if @bdev can be claimed, %false otherwise.
736 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
739 if (bdev->bd_holder == holder)
740 return true; /* already a holder */
741 else if (bdev->bd_holder != NULL)
742 return false; /* held by someone else */
743 else if (bdev->bd_contains == bdev)
744 return true; /* is a whole device which isn't held */
746 else if (whole->bd_holder == bd_may_claim)
747 return true; /* is a partition of a device that is being partitioned */
748 else if (whole->bd_holder != NULL)
749 return false; /* is a partition of a held device */
751 return true; /* is a partition of an un-held device */
755 * bd_prepare_to_claim - prepare to claim a block device
756 * @bdev: block device of interest
757 * @whole: the whole device containing @bdev, may equal @bdev
758 * @holder: holder trying to claim @bdev
760 * Prepare to claim @bdev. This function fails if @bdev is already
761 * claimed by another holder and waits if another claiming is in
762 * progress. This function doesn't actually claim. On successful
763 * return, the caller has ownership of bd_claiming and bd_holder[s].
766 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
770 * 0 if @bdev can be claimed, -EBUSY otherwise.
772 static int bd_prepare_to_claim(struct block_device *bdev,
773 struct block_device *whole, void *holder)
776 /* if someone else claimed, fail */
777 if (!bd_may_claim(bdev, whole, holder))
780 /* if claiming is already in progress, wait for it to finish */
781 if (whole->bd_claiming) {
782 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
785 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
786 spin_unlock(&bdev_lock);
788 finish_wait(wq, &wait);
789 spin_lock(&bdev_lock);
798 * bd_start_claiming - start claiming a block device
799 * @bdev: block device of interest
800 * @holder: holder trying to claim @bdev
802 * @bdev is about to be opened exclusively. Check @bdev can be opened
803 * exclusively and mark that an exclusive open is in progress. Each
804 * successful call to this function must be matched with a call to
805 * either bd_finish_claiming() or bd_abort_claiming() (which do not
808 * This function is used to gain exclusive access to the block device
809 * without actually causing other exclusive open attempts to fail. It
810 * should be used when the open sequence itself requires exclusive
811 * access but may subsequently fail.
817 * Pointer to the block device containing @bdev on success, ERR_PTR()
820 static struct block_device *bd_start_claiming(struct block_device *bdev,
823 struct gendisk *disk;
824 struct block_device *whole;
830 * @bdev might not have been initialized properly yet, look up
831 * and grab the outer block device the hard way.
833 disk = get_gendisk(bdev->bd_dev, &partno);
835 return ERR_PTR(-ENXIO);
838 * Normally, @bdev should equal what's returned from bdget_disk()
839 * if partno is 0; however, some drivers (floppy) use multiple
840 * bdev's for the same physical device and @bdev may be one of the
841 * aliases. Keep @bdev if partno is 0. This means claimer
842 * tracking is broken for those devices but it has always been that
846 whole = bdget_disk(disk, 0);
848 whole = bdgrab(bdev);
850 module_put(disk->fops->owner);
853 return ERR_PTR(-ENOMEM);
855 /* prepare to claim, if successful, mark claiming in progress */
856 spin_lock(&bdev_lock);
858 err = bd_prepare_to_claim(bdev, whole, holder);
860 whole->bd_claiming = holder;
861 spin_unlock(&bdev_lock);
864 spin_unlock(&bdev_lock);
871 struct bd_holder_disk {
872 struct list_head list;
873 struct gendisk *disk;
877 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
878 struct gendisk *disk)
880 struct bd_holder_disk *holder;
882 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
883 if (holder->disk == disk)
888 static int add_symlink(struct kobject *from, struct kobject *to)
890 return sysfs_create_link(from, to, kobject_name(to));
893 static void del_symlink(struct kobject *from, struct kobject *to)
895 sysfs_remove_link(from, kobject_name(to));
899 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
900 * @bdev: the claimed slave bdev
901 * @disk: the holding disk
903 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
905 * This functions creates the following sysfs symlinks.
907 * - from "slaves" directory of the holder @disk to the claimed @bdev
908 * - from "holders" directory of the @bdev to the holder @disk
910 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
911 * passed to bd_link_disk_holder(), then:
913 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
914 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
916 * The caller must have claimed @bdev before calling this function and
917 * ensure that both @bdev and @disk are valid during the creation and
918 * lifetime of these symlinks.
924 * 0 on success, -errno on failure.
926 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
928 struct bd_holder_disk *holder;
931 mutex_lock(&bdev->bd_mutex);
933 WARN_ON_ONCE(!bdev->bd_holder);
935 /* FIXME: remove the following once add_disk() handles errors */
936 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
939 holder = bd_find_holder_disk(bdev, disk);
945 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
951 INIT_LIST_HEAD(&holder->list);
955 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
959 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
963 * bdev could be deleted beneath us which would implicitly destroy
964 * the holder directory. Hold on to it.
966 kobject_get(bdev->bd_part->holder_dir);
968 list_add(&holder->list, &bdev->bd_holder_disks);
972 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
976 mutex_unlock(&bdev->bd_mutex);
979 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
982 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
983 * @bdev: the calimed slave bdev
984 * @disk: the holding disk
986 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
991 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
993 struct bd_holder_disk *holder;
995 mutex_lock(&bdev->bd_mutex);
997 holder = bd_find_holder_disk(bdev, disk);
999 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1000 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1001 del_symlink(bdev->bd_part->holder_dir,
1002 &disk_to_dev(disk)->kobj);
1003 kobject_put(bdev->bd_part->holder_dir);
1004 list_del_init(&holder->list);
1008 mutex_unlock(&bdev->bd_mutex);
1010 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1014 * flush_disk - invalidates all buffer-cache entries on a disk
1016 * @bdev: struct block device to be flushed
1017 * @kill_dirty: flag to guide handling of dirty inodes
1019 * Invalidates all buffer-cache entries on a disk. It should be called
1020 * when a disk has been changed -- either by a media change or online
1023 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1025 if (__invalidate_device(bdev, kill_dirty)) {
1026 char name[BDEVNAME_SIZE] = "";
1029 disk_name(bdev->bd_disk, 0, name);
1030 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1031 "resized disk %s\n", name);
1036 if (disk_part_scan_enabled(bdev->bd_disk))
1037 bdev->bd_invalidated = 1;
1041 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1042 * @disk: struct gendisk to check
1043 * @bdev: struct bdev to adjust.
1045 * This routine checks to see if the bdev size does not match the disk size
1046 * and adjusts it if it differs.
1048 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1050 loff_t disk_size, bdev_size;
1052 disk_size = (loff_t)get_capacity(disk) << 9;
1053 bdev_size = i_size_read(bdev->bd_inode);
1054 if (disk_size != bdev_size) {
1055 char name[BDEVNAME_SIZE];
1057 disk_name(disk, 0, name);
1059 "%s: detected capacity change from %lld to %lld\n",
1060 name, bdev_size, disk_size);
1061 i_size_write(bdev->bd_inode, disk_size);
1062 flush_disk(bdev, false);
1065 EXPORT_SYMBOL(check_disk_size_change);
1068 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1069 * @disk: struct gendisk to be revalidated
1071 * This routine is a wrapper for lower-level driver's revalidate_disk
1072 * call-backs. It is used to do common pre and post operations needed
1073 * for all revalidate_disk operations.
1075 int revalidate_disk(struct gendisk *disk)
1077 struct block_device *bdev;
1080 if (disk->fops->revalidate_disk)
1081 ret = disk->fops->revalidate_disk(disk);
1083 bdev = bdget_disk(disk, 0);
1087 mutex_lock(&bdev->bd_mutex);
1088 check_disk_size_change(disk, bdev);
1089 bdev->bd_invalidated = 0;
1090 mutex_unlock(&bdev->bd_mutex);
1094 EXPORT_SYMBOL(revalidate_disk);
1097 * This routine checks whether a removable media has been changed,
1098 * and invalidates all buffer-cache-entries in that case. This
1099 * is a relatively slow routine, so we have to try to minimize using
1100 * it. Thus it is called only upon a 'mount' or 'open'. This
1101 * is the best way of combining speed and utility, I think.
1102 * People changing diskettes in the middle of an operation deserve
1105 int check_disk_change(struct block_device *bdev)
1107 struct gendisk *disk = bdev->bd_disk;
1108 const struct block_device_operations *bdops = disk->fops;
1109 unsigned int events;
1111 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1112 DISK_EVENT_EJECT_REQUEST);
1113 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1116 flush_disk(bdev, true);
1117 if (bdops->revalidate_disk)
1118 bdops->revalidate_disk(bdev->bd_disk);
1122 EXPORT_SYMBOL(check_disk_change);
1124 void bd_set_size(struct block_device *bdev, loff_t size)
1126 unsigned bsize = bdev_logical_block_size(bdev);
1128 mutex_lock(&bdev->bd_inode->i_mutex);
1129 i_size_write(bdev->bd_inode, size);
1130 mutex_unlock(&bdev->bd_inode->i_mutex);
1131 while (bsize < PAGE_CACHE_SIZE) {
1136 bdev->bd_block_size = bsize;
1137 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1139 EXPORT_SYMBOL(bd_set_size);
1141 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1146 * mutex_lock(part->bd_mutex)
1147 * mutex_lock_nested(whole->bd_mutex, 1)
1150 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1152 struct gendisk *disk;
1153 struct module *owner;
1158 if (mode & FMODE_READ)
1160 if (mode & FMODE_WRITE)
1163 * hooks: /n/, see "layering violations".
1166 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1176 disk = get_gendisk(bdev->bd_dev, &partno);
1179 owner = disk->fops->owner;
1181 disk_block_events(disk);
1182 mutex_lock_nested(&bdev->bd_mutex, for_part);
1183 if (!bdev->bd_openers) {
1184 bdev->bd_disk = disk;
1185 bdev->bd_queue = disk->queue;
1186 bdev->bd_contains = bdev;
1188 struct backing_dev_info *bdi;
1191 bdev->bd_part = disk_get_part(disk, partno);
1196 if (disk->fops->open) {
1197 ret = disk->fops->open(bdev, mode);
1198 if (ret == -ERESTARTSYS) {
1199 /* Lost a race with 'disk' being
1200 * deleted, try again.
1203 disk_put_part(bdev->bd_part);
1204 bdev->bd_part = NULL;
1205 bdev->bd_disk = NULL;
1206 bdev->bd_queue = NULL;
1207 mutex_unlock(&bdev->bd_mutex);
1208 disk_unblock_events(disk);
1216 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1217 bdi = blk_get_backing_dev_info(bdev);
1218 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1222 * If the device is invalidated, rescan partition
1223 * if open succeeded or failed with -ENOMEDIUM.
1224 * The latter is necessary to prevent ghost
1225 * partitions on a removed medium.
1227 if (bdev->bd_invalidated) {
1229 rescan_partitions(disk, bdev);
1230 else if (ret == -ENOMEDIUM)
1231 invalidate_partitions(disk, bdev);
1236 struct block_device *whole;
1237 whole = bdget_disk(disk, 0);
1242 ret = __blkdev_get(whole, mode, 1);
1245 bdev->bd_contains = whole;
1246 bdev_inode_switch_bdi(bdev->bd_inode,
1247 whole->bd_inode->i_data.backing_dev_info);
1248 bdev->bd_part = disk_get_part(disk, partno);
1249 if (!(disk->flags & GENHD_FL_UP) ||
1250 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1254 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1257 if (bdev->bd_contains == bdev) {
1259 if (bdev->bd_disk->fops->open)
1260 ret = bdev->bd_disk->fops->open(bdev, mode);
1261 /* the same as first opener case, read comment there */
1262 if (bdev->bd_invalidated) {
1264 rescan_partitions(bdev->bd_disk, bdev);
1265 else if (ret == -ENOMEDIUM)
1266 invalidate_partitions(bdev->bd_disk, bdev);
1269 goto out_unlock_bdev;
1271 /* only one opener holds refs to the module and disk */
1277 bdev->bd_part_count++;
1278 mutex_unlock(&bdev->bd_mutex);
1279 disk_unblock_events(disk);
1283 disk_put_part(bdev->bd_part);
1284 bdev->bd_disk = NULL;
1285 bdev->bd_part = NULL;
1286 bdev->bd_queue = NULL;
1287 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1288 if (bdev != bdev->bd_contains)
1289 __blkdev_put(bdev->bd_contains, mode, 1);
1290 bdev->bd_contains = NULL;
1292 mutex_unlock(&bdev->bd_mutex);
1293 disk_unblock_events(disk);
1303 * blkdev_get - open a block device
1304 * @bdev: block_device to open
1305 * @mode: FMODE_* mask
1306 * @holder: exclusive holder identifier
1308 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1309 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1310 * @holder is invalid. Exclusive opens may nest for the same @holder.
1312 * On success, the reference count of @bdev is unchanged. On failure,
1319 * 0 on success, -errno on failure.
1321 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1323 struct block_device *whole = NULL;
1326 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1328 if ((mode & FMODE_EXCL) && holder) {
1329 whole = bd_start_claiming(bdev, holder);
1330 if (IS_ERR(whole)) {
1332 return PTR_ERR(whole);
1336 res = __blkdev_get(bdev, mode, 0);
1339 struct gendisk *disk = whole->bd_disk;
1341 /* finish claiming */
1342 mutex_lock(&bdev->bd_mutex);
1343 spin_lock(&bdev_lock);
1346 BUG_ON(!bd_may_claim(bdev, whole, holder));
1348 * Note that for a whole device bd_holders
1349 * will be incremented twice, and bd_holder
1350 * will be set to bd_may_claim before being
1353 whole->bd_holders++;
1354 whole->bd_holder = bd_may_claim;
1356 bdev->bd_holder = holder;
1359 /* tell others that we're done */
1360 BUG_ON(whole->bd_claiming != holder);
1361 whole->bd_claiming = NULL;
1362 wake_up_bit(&whole->bd_claiming, 0);
1364 spin_unlock(&bdev_lock);
1367 * Block event polling for write claims if requested. Any
1368 * write holder makes the write_holder state stick until
1369 * all are released. This is good enough and tracking
1370 * individual writeable reference is too fragile given the
1371 * way @mode is used in blkdev_get/put().
1373 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1374 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1375 bdev->bd_write_holder = true;
1376 disk_block_events(disk);
1379 mutex_unlock(&bdev->bd_mutex);
1385 EXPORT_SYMBOL(blkdev_get);
1388 * blkdev_get_by_path - open a block device by name
1389 * @path: path to the block device to open
1390 * @mode: FMODE_* mask
1391 * @holder: exclusive holder identifier
1393 * Open the blockdevice described by the device file at @path. @mode
1394 * and @holder are identical to blkdev_get().
1396 * On success, the returned block_device has reference count of one.
1402 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1404 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1407 struct block_device *bdev;
1410 bdev = lookup_bdev(path);
1414 err = blkdev_get(bdev, mode, holder);
1416 return ERR_PTR(err);
1418 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1419 blkdev_put(bdev, mode);
1420 return ERR_PTR(-EACCES);
1425 EXPORT_SYMBOL(blkdev_get_by_path);
1428 * blkdev_get_by_dev - open a block device by device number
1429 * @dev: device number of block device to open
1430 * @mode: FMODE_* mask
1431 * @holder: exclusive holder identifier
1433 * Open the blockdevice described by device number @dev. @mode and
1434 * @holder are identical to blkdev_get().
1436 * Use it ONLY if you really do not have anything better - i.e. when
1437 * you are behind a truly sucky interface and all you are given is a
1438 * device number. _Never_ to be used for internal purposes. If you
1439 * ever need it - reconsider your API.
1441 * On success, the returned block_device has reference count of one.
1447 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1449 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1451 struct block_device *bdev;
1456 return ERR_PTR(-ENOMEM);
1458 err = blkdev_get(bdev, mode, holder);
1460 return ERR_PTR(err);
1464 EXPORT_SYMBOL(blkdev_get_by_dev);
1466 static int blkdev_open(struct inode * inode, struct file * filp)
1468 struct block_device *bdev;
1471 * Preserve backwards compatibility and allow large file access
1472 * even if userspace doesn't ask for it explicitly. Some mkfs
1473 * binary needs it. We might want to drop this workaround
1474 * during an unstable branch.
1476 filp->f_flags |= O_LARGEFILE;
1478 if (filp->f_flags & O_NDELAY)
1479 filp->f_mode |= FMODE_NDELAY;
1480 if (filp->f_flags & O_EXCL)
1481 filp->f_mode |= FMODE_EXCL;
1482 if ((filp->f_flags & O_ACCMODE) == 3)
1483 filp->f_mode |= FMODE_WRITE_IOCTL;
1485 bdev = bd_acquire(inode);
1489 filp->f_mapping = bdev->bd_inode->i_mapping;
1491 return blkdev_get(bdev, filp->f_mode, filp);
1494 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1496 struct gendisk *disk = bdev->bd_disk;
1497 struct block_device *victim = NULL;
1499 mutex_lock_nested(&bdev->bd_mutex, for_part);
1501 bdev->bd_part_count--;
1503 if (!--bdev->bd_openers) {
1504 WARN_ON_ONCE(bdev->bd_holders);
1505 sync_blockdev(bdev);
1507 /* ->release can cause the old bdi to disappear,
1508 * so must switch it out first
1510 bdev_inode_switch_bdi(bdev->bd_inode,
1511 &default_backing_dev_info);
1513 if (bdev->bd_contains == bdev) {
1514 if (disk->fops->release)
1515 disk->fops->release(disk, mode);
1517 if (!bdev->bd_openers) {
1518 struct module *owner = disk->fops->owner;
1520 disk_put_part(bdev->bd_part);
1521 bdev->bd_part = NULL;
1522 bdev->bd_disk = NULL;
1523 if (bdev != bdev->bd_contains)
1524 victim = bdev->bd_contains;
1525 bdev->bd_contains = NULL;
1530 mutex_unlock(&bdev->bd_mutex);
1533 __blkdev_put(victim, mode, 1);
1536 void blkdev_put(struct block_device *bdev, fmode_t mode)
1538 mutex_lock(&bdev->bd_mutex);
1540 if (mode & FMODE_EXCL) {
1544 * Release a claim on the device. The holder fields
1545 * are protected with bdev_lock. bd_mutex is to
1546 * synchronize disk_holder unlinking.
1548 spin_lock(&bdev_lock);
1550 WARN_ON_ONCE(--bdev->bd_holders < 0);
1551 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1553 /* bd_contains might point to self, check in a separate step */
1554 if ((bdev_free = !bdev->bd_holders))
1555 bdev->bd_holder = NULL;
1556 if (!bdev->bd_contains->bd_holders)
1557 bdev->bd_contains->bd_holder = NULL;
1559 spin_unlock(&bdev_lock);
1562 * If this was the last claim, remove holder link and
1563 * unblock evpoll if it was a write holder.
1565 if (bdev_free && bdev->bd_write_holder) {
1566 disk_unblock_events(bdev->bd_disk);
1567 bdev->bd_write_holder = false;
1572 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1573 * event. This is to ensure detection of media removal commanded
1574 * from userland - e.g. eject(1).
1576 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1578 mutex_unlock(&bdev->bd_mutex);
1580 __blkdev_put(bdev, mode, 0);
1582 EXPORT_SYMBOL(blkdev_put);
1584 static int blkdev_close(struct inode * inode, struct file * filp)
1586 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1587 blkdev_put(bdev, filp->f_mode);
1591 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1593 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1594 fmode_t mode = file->f_mode;
1597 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1598 * to updated it before every ioctl.
1600 if (file->f_flags & O_NDELAY)
1601 mode |= FMODE_NDELAY;
1603 mode &= ~FMODE_NDELAY;
1605 return blkdev_ioctl(bdev, mode, cmd, arg);
1609 * Write data to the block device. Only intended for the block device itself
1610 * and the raw driver which basically is a fake block device.
1612 * Does not take i_mutex for the write and thus is not for general purpose
1615 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1617 struct file *file = iocb->ki_filp;
1618 struct blk_plug plug;
1621 blk_start_plug(&plug);
1622 ret = __generic_file_write_iter(iocb, from);
1625 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
1629 blk_finish_plug(&plug);
1632 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1634 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1636 struct file *file = iocb->ki_filp;
1637 struct inode *bd_inode = file->f_mapping->host;
1638 loff_t size = i_size_read(bd_inode);
1639 loff_t pos = iocb->ki_pos;
1645 iov_iter_truncate(to, size);
1646 return generic_file_read_iter(iocb, to);
1648 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1651 * Try to release a page associated with block device when the system
1652 * is under memory pressure.
1654 static int blkdev_releasepage(struct page *page, gfp_t wait)
1656 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1658 if (super && super->s_op->bdev_try_to_free_page)
1659 return super->s_op->bdev_try_to_free_page(super, page, wait);
1661 return try_to_free_buffers(page);
1664 static const struct address_space_operations def_blk_aops = {
1665 .readpage = blkdev_readpage,
1666 .readpages = blkdev_readpages,
1667 .writepage = blkdev_writepage,
1668 .write_begin = blkdev_write_begin,
1669 .write_end = blkdev_write_end,
1670 .writepages = generic_writepages,
1671 .releasepage = blkdev_releasepage,
1672 .direct_IO = blkdev_direct_IO,
1673 .is_dirty_writeback = buffer_check_dirty_writeback,
1676 const struct file_operations def_blk_fops = {
1677 .open = blkdev_open,
1678 .release = blkdev_close,
1679 .llseek = block_llseek,
1680 .read = new_sync_read,
1681 .write = new_sync_write,
1682 .read_iter = blkdev_read_iter,
1683 .write_iter = blkdev_write_iter,
1684 .mmap = generic_file_mmap,
1685 .fsync = blkdev_fsync,
1686 .unlocked_ioctl = block_ioctl,
1687 #ifdef CONFIG_COMPAT
1688 .compat_ioctl = compat_blkdev_ioctl,
1690 .splice_read = generic_file_splice_read,
1691 .splice_write = iter_file_splice_write,
1694 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1697 mm_segment_t old_fs = get_fs();
1699 res = blkdev_ioctl(bdev, 0, cmd, arg);
1704 EXPORT_SYMBOL(ioctl_by_bdev);
1707 * lookup_bdev - lookup a struct block_device by name
1708 * @pathname: special file representing the block device
1710 * Get a reference to the blockdevice at @pathname in the current
1711 * namespace if possible and return it. Return ERR_PTR(error)
1714 struct block_device *lookup_bdev(const char *pathname)
1716 struct block_device *bdev;
1717 struct inode *inode;
1721 if (!pathname || !*pathname)
1722 return ERR_PTR(-EINVAL);
1724 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1726 return ERR_PTR(error);
1728 inode = path.dentry->d_inode;
1730 if (!S_ISBLK(inode->i_mode))
1733 if (path.mnt->mnt_flags & MNT_NODEV)
1736 bdev = bd_acquire(inode);
1743 bdev = ERR_PTR(error);
1746 EXPORT_SYMBOL(lookup_bdev);
1748 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1750 struct super_block *sb = get_super(bdev);
1755 * no need to lock the super, get_super holds the
1756 * read mutex so the filesystem cannot go away
1757 * under us (->put_super runs with the write lock
1760 shrink_dcache_sb(sb);
1761 res = invalidate_inodes(sb, kill_dirty);
1764 invalidate_bdev(bdev);
1767 EXPORT_SYMBOL(__invalidate_device);
1769 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1771 struct inode *inode, *old_inode = NULL;
1773 spin_lock(&inode_sb_list_lock);
1774 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1775 struct address_space *mapping = inode->i_mapping;
1777 spin_lock(&inode->i_lock);
1778 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1779 mapping->nrpages == 0) {
1780 spin_unlock(&inode->i_lock);
1784 spin_unlock(&inode->i_lock);
1785 spin_unlock(&inode_sb_list_lock);
1787 * We hold a reference to 'inode' so it couldn't have been
1788 * removed from s_inodes list while we dropped the
1789 * inode_sb_list_lock. We cannot iput the inode now as we can
1790 * be holding the last reference and we cannot iput it under
1791 * inode_sb_list_lock. So we keep the reference and iput it
1797 func(I_BDEV(inode), arg);
1799 spin_lock(&inode_sb_list_lock);
1801 spin_unlock(&inode_sb_list_lock);