1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
9 #include <linux/init.h>
11 #include <linux/fcntl.h>
12 #include <linux/slab.h>
13 #include <linux/kmod.h>
14 #include <linux/major.h>
15 #include <linux/device_cgroup.h>
16 #include <linux/highmem.h>
17 #include <linux/blkdev.h>
18 #include <linux/backing-dev.h>
19 #include <linux/module.h>
20 #include <linux/blkpg.h>
21 #include <linux/magic.h>
22 #include <linux/buffer_head.h>
23 #include <linux/swap.h>
24 #include <linux/pagevec.h>
25 #include <linux/writeback.h>
26 #include <linux/mpage.h>
27 #include <linux/mount.h>
28 #include <linux/pseudo_fs.h>
29 #include <linux/uio.h>
30 #include <linux/namei.h>
31 #include <linux/log2.h>
32 #include <linux/cleancache.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/falloc.h>
35 #include <linux/uaccess.h>
36 #include <linux/suspend.h>
40 struct block_device bdev;
41 struct inode vfs_inode;
44 static const struct address_space_operations def_blk_aops;
46 static inline struct bdev_inode *BDEV_I(struct inode *inode)
48 return container_of(inode, struct bdev_inode, vfs_inode);
51 struct block_device *I_BDEV(struct inode *inode)
53 return &BDEV_I(inode)->bdev;
55 EXPORT_SYMBOL(I_BDEV);
57 static void bdev_write_inode(struct block_device *bdev)
59 struct inode *inode = bdev->bd_inode;
62 spin_lock(&inode->i_lock);
63 while (inode->i_state & I_DIRTY) {
64 spin_unlock(&inode->i_lock);
65 ret = write_inode_now(inode, true);
67 char name[BDEVNAME_SIZE];
68 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
69 "for block device %s (err=%d).\n",
70 bdevname(bdev, name), ret);
72 spin_lock(&inode->i_lock);
74 spin_unlock(&inode->i_lock);
77 /* Kill _all_ buffers and pagecache , dirty or not.. */
78 static void kill_bdev(struct block_device *bdev)
80 struct address_space *mapping = bdev->bd_inode->i_mapping;
82 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
86 truncate_inode_pages(mapping, 0);
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) {
96 lru_add_drain_all(); /* make sure all lru add caches are flushed */
97 invalidate_mapping_pages(mapping, 0, -1);
99 /* 99% of the time, we don't need to flush the cleancache on the bdev.
100 * But, for the strange corners, lets be cautious
102 cleancache_invalidate_inode(mapping);
104 EXPORT_SYMBOL(invalidate_bdev);
107 * Drop all buffers & page cache for given bdev range. This function bails
108 * with error if bdev has other exclusive owner (such as filesystem).
110 int truncate_bdev_range(struct block_device *bdev, fmode_t mode,
111 loff_t lstart, loff_t lend)
114 * If we don't hold exclusive handle for the device, upgrade to it
115 * while we discard the buffer cache to avoid discarding buffers
116 * under live filesystem.
118 if (!(mode & FMODE_EXCL)) {
119 int err = bd_prepare_to_claim(bdev, truncate_bdev_range);
124 truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend);
125 if (!(mode & FMODE_EXCL))
126 bd_abort_claiming(bdev, truncate_bdev_range);
129 EXPORT_SYMBOL(truncate_bdev_range);
131 static void set_init_blocksize(struct block_device *bdev)
133 bdev->bd_inode->i_blkbits = blksize_bits(bdev_logical_block_size(bdev));
136 int set_blocksize(struct block_device *bdev, int size)
138 /* Size must be a power of two, and between 512 and PAGE_SIZE */
139 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
142 /* Size cannot be smaller than the size supported by the device */
143 if (size < bdev_logical_block_size(bdev))
146 /* Don't change the size if it is same as current */
147 if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
149 bdev->bd_inode->i_blkbits = blksize_bits(size);
155 EXPORT_SYMBOL(set_blocksize);
157 int sb_set_blocksize(struct super_block *sb, int size)
159 if (set_blocksize(sb->s_bdev, size))
161 /* If we get here, we know size is power of two
162 * and it's value is between 512 and PAGE_SIZE */
163 sb->s_blocksize = size;
164 sb->s_blocksize_bits = blksize_bits(size);
165 return sb->s_blocksize;
168 EXPORT_SYMBOL(sb_set_blocksize);
170 int sb_min_blocksize(struct super_block *sb, int size)
172 int minsize = bdev_logical_block_size(sb->s_bdev);
175 return sb_set_blocksize(sb, size);
178 EXPORT_SYMBOL(sb_min_blocksize);
181 blkdev_get_block(struct inode *inode, sector_t iblock,
182 struct buffer_head *bh, int create)
184 bh->b_bdev = I_BDEV(inode);
185 bh->b_blocknr = iblock;
186 set_buffer_mapped(bh);
190 static struct inode *bdev_file_inode(struct file *file)
192 return file->f_mapping->host;
195 static unsigned int dio_bio_write_op(struct kiocb *iocb)
197 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
199 /* avoid the need for a I/O completion work item */
200 if (iocb->ki_flags & IOCB_DSYNC)
205 #define DIO_INLINE_BIO_VECS 4
207 static void blkdev_bio_end_io_simple(struct bio *bio)
209 struct task_struct *waiter = bio->bi_private;
211 WRITE_ONCE(bio->bi_private, NULL);
212 blk_wake_io_task(waiter);
216 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
219 struct file *file = iocb->ki_filp;
220 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
221 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
222 loff_t pos = iocb->ki_pos;
223 bool should_dirty = false;
228 if ((pos | iov_iter_alignment(iter)) &
229 (bdev_logical_block_size(bdev) - 1))
232 if (nr_pages <= DIO_INLINE_BIO_VECS)
235 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
241 bio_init(&bio, vecs, nr_pages);
242 bio_set_dev(&bio, bdev);
243 bio.bi_iter.bi_sector = pos >> 9;
244 bio.bi_write_hint = iocb->ki_hint;
245 bio.bi_private = current;
246 bio.bi_end_io = blkdev_bio_end_io_simple;
247 bio.bi_ioprio = iocb->ki_ioprio;
249 ret = bio_iov_iter_get_pages(&bio, iter);
252 ret = bio.bi_iter.bi_size;
254 if (iov_iter_rw(iter) == READ) {
255 bio.bi_opf = REQ_OP_READ;
256 if (iter_is_iovec(iter))
259 bio.bi_opf = dio_bio_write_op(iocb);
260 task_io_account_write(ret);
262 if (iocb->ki_flags & IOCB_HIPRI)
263 bio_set_polled(&bio, iocb);
265 qc = submit_bio(&bio);
267 set_current_state(TASK_UNINTERRUPTIBLE);
268 if (!READ_ONCE(bio.bi_private))
270 if (!(iocb->ki_flags & IOCB_HIPRI) ||
271 !blk_poll(bdev_get_queue(bdev), qc, true))
274 __set_current_state(TASK_RUNNING);
276 bio_release_pages(&bio, should_dirty);
277 if (unlikely(bio.bi_status))
278 ret = blk_status_to_errno(bio.bi_status);
281 if (vecs != inline_vecs)
292 struct task_struct *waiter;
297 bool should_dirty : 1;
302 static struct bio_set blkdev_dio_pool;
304 static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
306 struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
307 struct request_queue *q = bdev_get_queue(bdev);
309 return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
312 static void blkdev_bio_end_io(struct bio *bio)
314 struct blkdev_dio *dio = bio->bi_private;
315 bool should_dirty = dio->should_dirty;
317 if (bio->bi_status && !dio->bio.bi_status)
318 dio->bio.bi_status = bio->bi_status;
320 if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
322 struct kiocb *iocb = dio->iocb;
325 if (likely(!dio->bio.bi_status)) {
329 ret = blk_status_to_errno(dio->bio.bi_status);
332 dio->iocb->ki_complete(iocb, ret, 0);
336 struct task_struct *waiter = dio->waiter;
338 WRITE_ONCE(dio->waiter, NULL);
339 blk_wake_io_task(waiter);
344 bio_check_pages_dirty(bio);
346 bio_release_pages(bio, false);
352 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
354 struct file *file = iocb->ki_filp;
355 struct inode *inode = bdev_file_inode(file);
356 struct block_device *bdev = I_BDEV(inode);
357 struct blk_plug plug;
358 struct blkdev_dio *dio;
360 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
361 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
362 loff_t pos = iocb->ki_pos;
363 blk_qc_t qc = BLK_QC_T_NONE;
366 if ((pos | iov_iter_alignment(iter)) &
367 (bdev_logical_block_size(bdev) - 1))
370 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
372 dio = container_of(bio, struct blkdev_dio, bio);
373 dio->is_sync = is_sync = is_sync_kiocb(iocb);
375 dio->waiter = current;
382 dio->multi_bio = false;
383 dio->should_dirty = is_read && iter_is_iovec(iter);
386 * Don't plug for HIPRI/polled IO, as those should go straight
390 blk_start_plug(&plug);
393 bio_set_dev(bio, bdev);
394 bio->bi_iter.bi_sector = pos >> 9;
395 bio->bi_write_hint = iocb->ki_hint;
396 bio->bi_private = dio;
397 bio->bi_end_io = blkdev_bio_end_io;
398 bio->bi_ioprio = iocb->ki_ioprio;
400 ret = bio_iov_iter_get_pages(bio, iter);
402 bio->bi_status = BLK_STS_IOERR;
408 bio->bi_opf = REQ_OP_READ;
409 if (dio->should_dirty)
410 bio_set_pages_dirty(bio);
412 bio->bi_opf = dio_bio_write_op(iocb);
413 task_io_account_write(bio->bi_iter.bi_size);
416 dio->size += bio->bi_iter.bi_size;
417 pos += bio->bi_iter.bi_size;
419 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
423 if (iocb->ki_flags & IOCB_HIPRI) {
424 bio_set_polled(bio, iocb);
428 qc = submit_bio(bio);
431 WRITE_ONCE(iocb->ki_cookie, qc);
435 if (!dio->multi_bio) {
437 * AIO needs an extra reference to ensure the dio
438 * structure which is embedded into the first bio
443 dio->multi_bio = true;
444 atomic_set(&dio->ref, 2);
446 atomic_inc(&dio->ref);
450 bio = bio_alloc(GFP_KERNEL, nr_pages);
454 blk_finish_plug(&plug);
460 set_current_state(TASK_UNINTERRUPTIBLE);
461 if (!READ_ONCE(dio->waiter))
464 if (!(iocb->ki_flags & IOCB_HIPRI) ||
465 !blk_poll(bdev_get_queue(bdev), qc, true))
468 __set_current_state(TASK_RUNNING);
471 ret = blk_status_to_errno(dio->bio.bi_status);
480 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
484 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
487 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
488 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
490 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
493 static __init int blkdev_init(void)
495 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
497 module_init(blkdev_init);
499 int __sync_blockdev(struct block_device *bdev, int wait)
504 return filemap_flush(bdev->bd_inode->i_mapping);
505 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
509 * Write out and wait upon all the dirty data associated with a block
510 * device via its mapping. Does not take the superblock lock.
512 int sync_blockdev(struct block_device *bdev)
514 return __sync_blockdev(bdev, 1);
516 EXPORT_SYMBOL(sync_blockdev);
519 * Write out and wait upon all dirty data associated with this
520 * device. Filesystem data as well as the underlying block
521 * device. Takes the superblock lock.
523 int fsync_bdev(struct block_device *bdev)
525 struct super_block *sb = get_super(bdev);
527 int res = sync_filesystem(sb);
531 return sync_blockdev(bdev);
533 EXPORT_SYMBOL(fsync_bdev);
536 * freeze_bdev -- lock a filesystem and force it into a consistent state
537 * @bdev: blockdevice to lock
539 * If a superblock is found on this device, we take the s_umount semaphore
540 * on it to make sure nobody unmounts until the snapshot creation is done.
541 * The reference counter (bd_fsfreeze_count) guarantees that only the last
542 * unfreeze process can unfreeze the frozen filesystem actually when multiple
543 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
544 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
547 int freeze_bdev(struct block_device *bdev)
549 struct super_block *sb;
552 mutex_lock(&bdev->bd_fsfreeze_mutex);
553 if (++bdev->bd_fsfreeze_count > 1)
556 sb = get_active_super(bdev);
559 if (sb->s_op->freeze_super)
560 error = sb->s_op->freeze_super(sb);
562 error = freeze_super(sb);
563 deactivate_super(sb);
566 bdev->bd_fsfreeze_count--;
569 bdev->bd_fsfreeze_sb = sb;
574 mutex_unlock(&bdev->bd_fsfreeze_mutex);
577 EXPORT_SYMBOL(freeze_bdev);
580 * thaw_bdev -- unlock filesystem
581 * @bdev: blockdevice to unlock
583 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
585 int thaw_bdev(struct block_device *bdev)
587 struct super_block *sb;
590 mutex_lock(&bdev->bd_fsfreeze_mutex);
591 if (!bdev->bd_fsfreeze_count)
595 if (--bdev->bd_fsfreeze_count > 0)
598 sb = bdev->bd_fsfreeze_sb;
602 if (sb->s_op->thaw_super)
603 error = sb->s_op->thaw_super(sb);
605 error = thaw_super(sb);
607 bdev->bd_fsfreeze_count++;
609 mutex_unlock(&bdev->bd_fsfreeze_mutex);
612 EXPORT_SYMBOL(thaw_bdev);
614 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
616 return block_write_full_page(page, blkdev_get_block, wbc);
619 static int blkdev_readpage(struct file * file, struct page * page)
621 return block_read_full_page(page, blkdev_get_block);
624 static void blkdev_readahead(struct readahead_control *rac)
626 mpage_readahead(rac, blkdev_get_block);
629 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
630 loff_t pos, unsigned len, unsigned flags,
631 struct page **pagep, void **fsdata)
633 return block_write_begin(mapping, pos, len, flags, pagep,
637 static int blkdev_write_end(struct file *file, struct address_space *mapping,
638 loff_t pos, unsigned len, unsigned copied,
639 struct page *page, void *fsdata)
642 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
652 * for a block special file file_inode(file)->i_size is zero
653 * so we compute the size by hand (just as in block_read/write above)
655 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
657 struct inode *bd_inode = bdev_file_inode(file);
660 inode_lock(bd_inode);
661 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
662 inode_unlock(bd_inode);
666 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
668 struct inode *bd_inode = bdev_file_inode(filp);
669 struct block_device *bdev = I_BDEV(bd_inode);
672 error = file_write_and_wait_range(filp, start, end);
677 * There is no need to serialise calls to blkdev_issue_flush with
678 * i_mutex and doing so causes performance issues with concurrent
679 * O_SYNC writers to a block device.
681 error = blkdev_issue_flush(bdev, GFP_KERNEL);
682 if (error == -EOPNOTSUPP)
687 EXPORT_SYMBOL(blkdev_fsync);
690 * bdev_read_page() - Start reading a page from a block device
691 * @bdev: The device to read the page from
692 * @sector: The offset on the device to read the page to (need not be aligned)
693 * @page: The page to read
695 * On entry, the page should be locked. It will be unlocked when the page
696 * has been read. If the block driver implements rw_page synchronously,
697 * that will be true on exit from this function, but it need not be.
699 * Errors returned by this function are usually "soft", eg out of memory, or
700 * queue full; callers should try a different route to read this page rather
701 * than propagate an error back up the stack.
703 * Return: negative errno if an error occurs, 0 if submission was successful.
705 int bdev_read_page(struct block_device *bdev, sector_t sector,
708 const struct block_device_operations *ops = bdev->bd_disk->fops;
709 int result = -EOPNOTSUPP;
711 if (!ops->rw_page || bdev_get_integrity(bdev))
714 result = blk_queue_enter(bdev->bd_disk->queue, 0);
717 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
719 blk_queue_exit(bdev->bd_disk->queue);
724 * bdev_write_page() - Start writing a page to a block device
725 * @bdev: The device to write the page to
726 * @sector: The offset on the device to write the page to (need not be aligned)
727 * @page: The page to write
728 * @wbc: The writeback_control for the write
730 * On entry, the page should be locked and not currently under writeback.
731 * On exit, if the write started successfully, the page will be unlocked and
732 * under writeback. If the write failed already (eg the driver failed to
733 * queue the page to the device), the page will still be locked. If the
734 * caller is a ->writepage implementation, it will need to unlock the page.
736 * Errors returned by this function are usually "soft", eg out of memory, or
737 * queue full; callers should try a different route to write this page rather
738 * than propagate an error back up the stack.
740 * Return: negative errno if an error occurs, 0 if submission was successful.
742 int bdev_write_page(struct block_device *bdev, sector_t sector,
743 struct page *page, struct writeback_control *wbc)
746 const struct block_device_operations *ops = bdev->bd_disk->fops;
748 if (!ops->rw_page || bdev_get_integrity(bdev))
750 result = blk_queue_enter(bdev->bd_disk->queue, 0);
754 set_page_writeback(page);
755 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
758 end_page_writeback(page);
760 clean_page_buffers(page);
763 blk_queue_exit(bdev->bd_disk->queue);
771 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
772 static struct kmem_cache * bdev_cachep __read_mostly;
774 static struct inode *bdev_alloc_inode(struct super_block *sb)
776 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
779 return &ei->vfs_inode;
782 static void bdev_free_inode(struct inode *inode)
784 kmem_cache_free(bdev_cachep, BDEV_I(inode));
787 static void init_once(void *foo)
789 struct bdev_inode *ei = (struct bdev_inode *) foo;
790 struct block_device *bdev = &ei->bdev;
792 memset(bdev, 0, sizeof(*bdev));
793 mutex_init(&bdev->bd_mutex);
795 INIT_LIST_HEAD(&bdev->bd_holder_disks);
797 bdev->bd_bdi = &noop_backing_dev_info;
798 inode_init_once(&ei->vfs_inode);
799 /* Initialize mutex for freeze. */
800 mutex_init(&bdev->bd_fsfreeze_mutex);
803 static void bdev_evict_inode(struct inode *inode)
805 struct block_device *bdev = &BDEV_I(inode)->bdev;
806 truncate_inode_pages_final(&inode->i_data);
807 invalidate_inode_buffers(inode); /* is it needed here? */
809 /* Detach inode from wb early as bdi_put() may free bdi->wb */
810 inode_detach_wb(inode);
811 if (bdev->bd_bdi != &noop_backing_dev_info) {
812 bdi_put(bdev->bd_bdi);
813 bdev->bd_bdi = &noop_backing_dev_info;
817 static const struct super_operations bdev_sops = {
818 .statfs = simple_statfs,
819 .alloc_inode = bdev_alloc_inode,
820 .free_inode = bdev_free_inode,
821 .drop_inode = generic_delete_inode,
822 .evict_inode = bdev_evict_inode,
825 static int bd_init_fs_context(struct fs_context *fc)
827 struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
830 fc->s_iflags |= SB_I_CGROUPWB;
831 ctx->ops = &bdev_sops;
835 static struct file_system_type bd_type = {
837 .init_fs_context = bd_init_fs_context,
838 .kill_sb = kill_anon_super,
841 struct super_block *blockdev_superblock __read_mostly;
842 EXPORT_SYMBOL_GPL(blockdev_superblock);
844 void __init bdev_cache_init(void)
847 static struct vfsmount *bd_mnt;
849 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
850 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
851 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
853 err = register_filesystem(&bd_type);
855 panic("Cannot register bdev pseudo-fs");
856 bd_mnt = kern_mount(&bd_type);
858 panic("Cannot create bdev pseudo-fs");
859 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
862 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
864 struct block_device *bdev;
867 inode = new_inode(blockdev_superblock);
870 inode->i_mode = S_IFBLK;
872 inode->i_data.a_ops = &def_blk_aops;
873 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
875 bdev = I_BDEV(inode);
876 spin_lock_init(&bdev->bd_size_lock);
877 bdev->bd_disk = disk;
878 bdev->bd_partno = partno;
879 bdev->bd_super = NULL;
880 bdev->bd_inode = inode;
881 bdev->bd_part_count = 0;
885 void bdev_add(struct block_device *bdev, dev_t dev)
888 bdev->bd_inode->i_rdev = dev;
889 bdev->bd_inode->i_ino = dev;
890 insert_inode_hash(bdev->bd_inode);
893 static struct block_device *bdget(dev_t dev)
897 inode = ilookup(blockdev_superblock, dev);
900 return &BDEV_I(inode)->bdev;
904 * bdgrab -- Grab a reference to an already referenced block device
905 * @bdev: Block device to grab a reference to.
907 * Returns the block_device with an additional reference when successful,
908 * or NULL if the inode is already beeing freed.
910 struct block_device *bdgrab(struct block_device *bdev)
912 if (!igrab(bdev->bd_inode))
916 EXPORT_SYMBOL(bdgrab);
918 struct block_device *bdget_part(struct hd_struct *part)
920 return bdget(part_devt(part));
923 long nr_blockdev_pages(void)
928 spin_lock(&blockdev_superblock->s_inode_list_lock);
929 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
930 ret += inode->i_mapping->nrpages;
931 spin_unlock(&blockdev_superblock->s_inode_list_lock);
936 void bdput(struct block_device *bdev)
938 iput(bdev->bd_inode);
940 EXPORT_SYMBOL(bdput);
943 * bd_may_claim - test whether a block device can be claimed
944 * @bdev: block device of interest
945 * @whole: whole block device containing @bdev, may equal @bdev
946 * @holder: holder trying to claim @bdev
948 * Test whether @bdev can be claimed by @holder.
951 * spin_lock(&bdev_lock).
954 * %true if @bdev can be claimed, %false otherwise.
956 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
959 if (bdev->bd_holder == holder)
960 return true; /* already a holder */
961 else if (bdev->bd_holder != NULL)
962 return false; /* held by someone else */
963 else if (whole == bdev)
964 return true; /* is a whole device which isn't held */
966 else if (whole->bd_holder == bd_may_claim)
967 return true; /* is a partition of a device that is being partitioned */
968 else if (whole->bd_holder != NULL)
969 return false; /* is a partition of a held device */
971 return true; /* is a partition of an un-held device */
975 * bd_prepare_to_claim - claim a block device
976 * @bdev: block device of interest
977 * @holder: holder trying to claim @bdev
979 * Claim @bdev. This function fails if @bdev is already claimed by another
980 * holder and waits if another claiming is in progress. return, the caller
981 * has ownership of bd_claiming and bd_holder[s].
984 * 0 if @bdev can be claimed, -EBUSY otherwise.
986 int bd_prepare_to_claim(struct block_device *bdev, void *holder)
988 struct block_device *whole = bdev_whole(bdev);
990 if (WARN_ON_ONCE(!holder))
993 spin_lock(&bdev_lock);
994 /* if someone else claimed, fail */
995 if (!bd_may_claim(bdev, whole, holder)) {
996 spin_unlock(&bdev_lock);
1000 /* if claiming is already in progress, wait for it to finish */
1001 if (whole->bd_claiming) {
1002 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1005 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1006 spin_unlock(&bdev_lock);
1008 finish_wait(wq, &wait);
1013 whole->bd_claiming = holder;
1014 spin_unlock(&bdev_lock);
1017 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
1019 static void bd_clear_claiming(struct block_device *whole, void *holder)
1021 lockdep_assert_held(&bdev_lock);
1022 /* tell others that we're done */
1023 BUG_ON(whole->bd_claiming != holder);
1024 whole->bd_claiming = NULL;
1025 wake_up_bit(&whole->bd_claiming, 0);
1029 * bd_finish_claiming - finish claiming of a block device
1030 * @bdev: block device of interest
1031 * @holder: holder that has claimed @bdev
1033 * Finish exclusive open of a block device. Mark the device as exlusively
1034 * open by the holder and wake up all waiters for exclusive open to finish.
1036 static void bd_finish_claiming(struct block_device *bdev, void *holder)
1038 struct block_device *whole = bdev_whole(bdev);
1040 spin_lock(&bdev_lock);
1041 BUG_ON(!bd_may_claim(bdev, whole, holder));
1043 * Note that for a whole device bd_holders will be incremented twice,
1044 * and bd_holder will be set to bd_may_claim before being set to holder
1046 whole->bd_holders++;
1047 whole->bd_holder = bd_may_claim;
1049 bdev->bd_holder = holder;
1050 bd_clear_claiming(whole, holder);
1051 spin_unlock(&bdev_lock);
1055 * bd_abort_claiming - abort claiming of a block device
1056 * @bdev: block device of interest
1057 * @whole: whole block device
1058 * @holder: holder that has claimed @bdev
1060 * Abort claiming of a block device when the exclusive open failed. This can be
1061 * also used when exclusive open is not actually desired and we just needed
1062 * to block other exclusive openers for a while.
1064 void bd_abort_claiming(struct block_device *bdev, void *holder)
1066 spin_lock(&bdev_lock);
1067 bd_clear_claiming(bdev_whole(bdev), holder);
1068 spin_unlock(&bdev_lock);
1070 EXPORT_SYMBOL(bd_abort_claiming);
1073 struct bd_holder_disk {
1074 struct list_head list;
1075 struct gendisk *disk;
1079 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1080 struct gendisk *disk)
1082 struct bd_holder_disk *holder;
1084 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1085 if (holder->disk == disk)
1090 static int add_symlink(struct kobject *from, struct kobject *to)
1092 return sysfs_create_link(from, to, kobject_name(to));
1095 static void del_symlink(struct kobject *from, struct kobject *to)
1097 sysfs_remove_link(from, kobject_name(to));
1101 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1102 * @bdev: the claimed slave bdev
1103 * @disk: the holding disk
1105 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1107 * This functions creates the following sysfs symlinks.
1109 * - from "slaves" directory of the holder @disk to the claimed @bdev
1110 * - from "holders" directory of the @bdev to the holder @disk
1112 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1113 * passed to bd_link_disk_holder(), then:
1115 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1116 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1118 * The caller must have claimed @bdev before calling this function and
1119 * ensure that both @bdev and @disk are valid during the creation and
1120 * lifetime of these symlinks.
1126 * 0 on success, -errno on failure.
1128 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1130 struct bd_holder_disk *holder;
1133 mutex_lock(&bdev->bd_mutex);
1135 WARN_ON_ONCE(!bdev->bd_holder);
1137 /* FIXME: remove the following once add_disk() handles errors */
1138 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1141 holder = bd_find_holder_disk(bdev, disk);
1147 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1153 INIT_LIST_HEAD(&holder->list);
1154 holder->disk = disk;
1157 ret = add_symlink(disk->slave_dir, bdev_kobj(bdev));
1161 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1165 * bdev could be deleted beneath us which would implicitly destroy
1166 * the holder directory. Hold on to it.
1168 kobject_get(bdev->bd_part->holder_dir);
1170 list_add(&holder->list, &bdev->bd_holder_disks);
1174 del_symlink(disk->slave_dir, bdev_kobj(bdev));
1178 mutex_unlock(&bdev->bd_mutex);
1181 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1184 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1185 * @bdev: the calimed slave bdev
1186 * @disk: the holding disk
1188 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1193 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1195 struct bd_holder_disk *holder;
1197 mutex_lock(&bdev->bd_mutex);
1199 holder = bd_find_holder_disk(bdev, disk);
1201 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1202 del_symlink(disk->slave_dir, bdev_kobj(bdev));
1203 del_symlink(bdev->bd_part->holder_dir,
1204 &disk_to_dev(disk)->kobj);
1205 kobject_put(bdev->bd_part->holder_dir);
1206 list_del_init(&holder->list);
1210 mutex_unlock(&bdev->bd_mutex);
1212 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1216 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1217 * @disk: struct gendisk to check
1218 * @bdev: struct bdev to adjust.
1219 * @verbose: if %true log a message about a size change if there is any
1221 * This routine checks to see if the bdev size does not match the disk size
1222 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1225 static void check_disk_size_change(struct gendisk *disk,
1226 struct block_device *bdev, bool verbose)
1228 loff_t disk_size, bdev_size;
1230 spin_lock(&bdev->bd_size_lock);
1231 disk_size = (loff_t)get_capacity(disk) << 9;
1232 bdev_size = i_size_read(bdev->bd_inode);
1233 if (disk_size != bdev_size) {
1236 "%s: detected capacity change from %lld to %lld\n",
1237 disk->disk_name, bdev_size, disk_size);
1239 i_size_write(bdev->bd_inode, disk_size);
1241 spin_unlock(&bdev->bd_size_lock);
1245 * revalidate_disk_size - checks for disk size change and adjusts bdev size.
1246 * @disk: struct gendisk to check
1247 * @verbose: if %true log a message about a size change if there is any
1249 * This routine checks to see if the bdev size does not match the disk size
1250 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1253 void revalidate_disk_size(struct gendisk *disk, bool verbose)
1255 struct block_device *bdev;
1258 * Hidden disks don't have associated bdev so there's no point in
1259 * revalidating them.
1261 if (disk->flags & GENHD_FL_HIDDEN)
1264 bdev = bdget_disk(disk, 0);
1266 check_disk_size_change(disk, bdev, verbose);
1271 void bd_set_nr_sectors(struct block_device *bdev, sector_t sectors)
1273 spin_lock(&bdev->bd_size_lock);
1274 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
1275 spin_unlock(&bdev->bd_size_lock);
1277 EXPORT_SYMBOL(bd_set_nr_sectors);
1279 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1281 int bdev_disk_changed(struct block_device *bdev, bool invalidate)
1283 struct gendisk *disk = bdev->bd_disk;
1286 lockdep_assert_held(&bdev->bd_mutex);
1288 clear_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1291 ret = blk_drop_partitions(bdev);
1296 * Historically we only set the capacity to zero for devices that
1297 * support partitions (independ of actually having partitions created).
1298 * Doing that is rather inconsistent, but changing it broke legacy
1299 * udisks polling for legacy ide-cdrom devices. Use the crude check
1300 * below to get the sane behavior for most device while not breaking
1301 * userspace for this particular setup.
1304 if (disk_part_scan_enabled(disk) ||
1305 !(disk->flags & GENHD_FL_REMOVABLE))
1306 set_capacity(disk, 0);
1308 if (disk->fops->revalidate_disk)
1309 disk->fops->revalidate_disk(disk);
1312 check_disk_size_change(disk, bdev, !invalidate);
1314 if (get_capacity(disk)) {
1315 ret = blk_add_partitions(disk, bdev);
1318 } else if (invalidate) {
1320 * Tell userspace that the media / partition table may have
1323 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
1329 * Only exported for for loop and dasd for historic reasons. Don't use in new
1332 EXPORT_SYMBOL_GPL(bdev_disk_changed);
1337 * mutex_lock(part->bd_mutex)
1338 * mutex_lock_nested(whole->bd_mutex, 1)
1340 static int __blkdev_get(struct block_device *bdev, fmode_t mode)
1342 struct gendisk *disk = bdev->bd_disk;
1345 if (!bdev->bd_openers) {
1346 if (!bdev_is_partition(bdev)) {
1348 bdev->bd_part = disk_get_part(disk, 0);
1353 if (disk->fops->open)
1354 ret = disk->fops->open(bdev, mode);
1357 bd_set_nr_sectors(bdev, get_capacity(disk));
1358 set_init_blocksize(bdev);
1362 * If the device is invalidated, rescan partition
1363 * if open succeeded or failed with -ENOMEDIUM.
1364 * The latter is necessary to prevent ghost
1365 * partitions on a removed medium.
1367 if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
1368 (!ret || ret == -ENOMEDIUM))
1369 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1374 struct block_device *whole = bdget_disk(disk, 0);
1376 mutex_lock_nested(&whole->bd_mutex, 1);
1377 ret = __blkdev_get(whole, mode);
1379 mutex_unlock(&whole->bd_mutex);
1383 whole->bd_part_count++;
1384 mutex_unlock(&whole->bd_mutex);
1386 bdev->bd_part = disk_get_part(disk, bdev->bd_partno);
1387 if (!(disk->flags & GENHD_FL_UP) ||
1388 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1389 __blkdev_put(whole, mode, 1);
1394 bd_set_nr_sectors(bdev, bdev->bd_part->nr_sects);
1395 set_init_blocksize(bdev);
1398 if (bdev->bd_bdi == &noop_backing_dev_info)
1399 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1401 if (!bdev_is_partition(bdev)) {
1403 if (bdev->bd_disk->fops->open)
1404 ret = bdev->bd_disk->fops->open(bdev, mode);
1405 /* the same as first opener case, read comment there */
1406 if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
1407 (!ret || ret == -ENOMEDIUM))
1408 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1417 disk_put_part(bdev->bd_part);
1418 bdev->bd_part = NULL;
1422 struct block_device *blkdev_get_no_open(dev_t dev)
1424 struct block_device *bdev;
1425 struct gendisk *disk;
1427 down_read(&bdev_lookup_sem);
1430 up_read(&bdev_lookup_sem);
1431 blk_request_module(dev);
1432 down_read(&bdev_lookup_sem);
1439 disk = bdev->bd_disk;
1440 if (!kobject_get_unless_zero(&disk_to_dev(disk)->kobj))
1442 if ((disk->flags & (GENHD_FL_UP | GENHD_FL_HIDDEN)) != GENHD_FL_UP)
1444 if (!try_module_get(bdev->bd_disk->fops->owner))
1446 up_read(&bdev_lookup_sem);
1453 up_read(&bdev_lookup_sem);
1457 void blkdev_put_no_open(struct block_device *bdev)
1459 module_put(bdev->bd_disk->fops->owner);
1460 put_disk(bdev->bd_disk);
1465 * blkdev_get_by_dev - open a block device by device number
1466 * @dev: device number of block device to open
1467 * @mode: FMODE_* mask
1468 * @holder: exclusive holder identifier
1470 * Open the block device described by device number @dev. If @mode includes
1471 * %FMODE_EXCL, the block device is opened with exclusive access. Specifying
1472 * %FMODE_EXCL with a %NULL @holder is invalid. Exclusive opens may nest for
1475 * Use this interface ONLY if you really do not have anything better - i.e. when
1476 * you are behind a truly sucky interface and all you are given is a device
1477 * number. Everything else should use blkdev_get_by_path().
1483 * Reference to the block_device on success, ERR_PTR(-errno) on failure.
1485 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1487 bool unblock_events = true;
1488 struct block_device *bdev;
1489 struct gendisk *disk;
1492 ret = devcgroup_check_permission(DEVCG_DEV_BLOCK,
1493 MAJOR(dev), MINOR(dev),
1494 ((mode & FMODE_READ) ? DEVCG_ACC_READ : 0) |
1495 ((mode & FMODE_WRITE) ? DEVCG_ACC_WRITE : 0));
1497 return ERR_PTR(ret);
1500 * If we lost a race with 'disk' being deleted, try again. See md.c.
1503 bdev = blkdev_get_no_open(dev);
1505 return ERR_PTR(-ENXIO);
1506 disk = bdev->bd_disk;
1508 if (mode & FMODE_EXCL) {
1509 ret = bd_prepare_to_claim(bdev, holder);
1514 disk_block_events(disk);
1516 mutex_lock(&bdev->bd_mutex);
1517 ret =__blkdev_get(bdev, mode);
1519 goto abort_claiming;
1520 if (mode & FMODE_EXCL) {
1521 bd_finish_claiming(bdev, holder);
1524 * Block event polling for write claims if requested. Any write
1525 * holder makes the write_holder state stick until all are
1526 * released. This is good enough and tracking individual
1527 * writeable reference is too fragile given the way @mode is
1528 * used in blkdev_get/put().
1530 if ((mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1531 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1532 bdev->bd_write_holder = true;
1533 unblock_events = false;
1536 mutex_unlock(&bdev->bd_mutex);
1539 disk_unblock_events(disk);
1543 if (mode & FMODE_EXCL)
1544 bd_abort_claiming(bdev, holder);
1545 mutex_unlock(&bdev->bd_mutex);
1546 disk_unblock_events(disk);
1548 blkdev_put_no_open(bdev);
1549 if (ret == -ERESTARTSYS)
1551 return ERR_PTR(ret);
1553 EXPORT_SYMBOL(blkdev_get_by_dev);
1556 * blkdev_get_by_path - open a block device by name
1557 * @path: path to the block device to open
1558 * @mode: FMODE_* mask
1559 * @holder: exclusive holder identifier
1561 * Open the block device described by the device file at @path. If @mode
1562 * includes %FMODE_EXCL, the block device is opened with exclusive access.
1563 * Specifying %FMODE_EXCL with a %NULL @holder is invalid. Exclusive opens may
1564 * nest for the same @holder.
1570 * Reference to the block_device on success, ERR_PTR(-errno) on failure.
1572 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1575 struct block_device *bdev;
1579 error = lookup_bdev(path, &dev);
1581 return ERR_PTR(error);
1583 bdev = blkdev_get_by_dev(dev, mode, holder);
1584 if (!IS_ERR(bdev) && (mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1585 blkdev_put(bdev, mode);
1586 return ERR_PTR(-EACCES);
1591 EXPORT_SYMBOL(blkdev_get_by_path);
1593 static int blkdev_open(struct inode * inode, struct file * filp)
1595 struct block_device *bdev;
1598 * Preserve backwards compatibility and allow large file access
1599 * even if userspace doesn't ask for it explicitly. Some mkfs
1600 * binary needs it. We might want to drop this workaround
1601 * during an unstable branch.
1603 filp->f_flags |= O_LARGEFILE;
1605 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
1607 if (filp->f_flags & O_NDELAY)
1608 filp->f_mode |= FMODE_NDELAY;
1609 if (filp->f_flags & O_EXCL)
1610 filp->f_mode |= FMODE_EXCL;
1611 if ((filp->f_flags & O_ACCMODE) == 3)
1612 filp->f_mode |= FMODE_WRITE_IOCTL;
1614 bdev = blkdev_get_by_dev(inode->i_rdev, filp->f_mode, filp);
1616 return PTR_ERR(bdev);
1617 filp->f_mapping = bdev->bd_inode->i_mapping;
1618 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1622 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1624 struct gendisk *disk = bdev->bd_disk;
1625 struct block_device *victim = NULL;
1628 * Sync early if it looks like we're the last one. If someone else
1629 * opens the block device between now and the decrement of bd_openers
1630 * then we did a sync that we didn't need to, but that's not the end
1631 * of the world and we want to avoid long (could be several minute)
1632 * syncs while holding the mutex.
1634 if (bdev->bd_openers == 1)
1635 sync_blockdev(bdev);
1637 mutex_lock_nested(&bdev->bd_mutex, for_part);
1639 bdev->bd_part_count--;
1641 if (!--bdev->bd_openers) {
1642 WARN_ON_ONCE(bdev->bd_holders);
1643 sync_blockdev(bdev);
1645 bdev_write_inode(bdev);
1647 if (!bdev_is_partition(bdev) && disk->fops->release)
1648 disk->fops->release(disk, mode);
1650 disk_put_part(bdev->bd_part);
1651 bdev->bd_part = NULL;
1652 if (bdev_is_partition(bdev))
1653 victim = bdev_whole(bdev);
1655 if (!bdev_is_partition(bdev) && disk->fops->release)
1656 disk->fops->release(disk, mode);
1658 mutex_unlock(&bdev->bd_mutex);
1660 __blkdev_put(victim, mode, 1);
1665 void blkdev_put(struct block_device *bdev, fmode_t mode)
1667 struct gendisk *disk = bdev->bd_disk;
1669 mutex_lock(&bdev->bd_mutex);
1671 if (mode & FMODE_EXCL) {
1672 struct block_device *whole = bdev_whole(bdev);
1676 * Release a claim on the device. The holder fields
1677 * are protected with bdev_lock. bd_mutex is to
1678 * synchronize disk_holder unlinking.
1680 spin_lock(&bdev_lock);
1682 WARN_ON_ONCE(--bdev->bd_holders < 0);
1683 WARN_ON_ONCE(--whole->bd_holders < 0);
1685 if ((bdev_free = !bdev->bd_holders))
1686 bdev->bd_holder = NULL;
1687 if (!whole->bd_holders)
1688 whole->bd_holder = NULL;
1690 spin_unlock(&bdev_lock);
1693 * If this was the last claim, remove holder link and
1694 * unblock evpoll if it was a write holder.
1696 if (bdev_free && bdev->bd_write_holder) {
1697 disk_unblock_events(disk);
1698 bdev->bd_write_holder = false;
1703 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1704 * event. This is to ensure detection of media removal commanded
1705 * from userland - e.g. eject(1).
1707 disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE);
1708 mutex_unlock(&bdev->bd_mutex);
1710 __blkdev_put(bdev, mode, 0);
1711 blkdev_put_no_open(bdev);
1713 EXPORT_SYMBOL(blkdev_put);
1715 static int blkdev_close(struct inode * inode, struct file * filp)
1717 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1718 blkdev_put(bdev, filp->f_mode);
1722 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1724 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1725 fmode_t mode = file->f_mode;
1728 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1729 * to updated it before every ioctl.
1731 if (file->f_flags & O_NDELAY)
1732 mode |= FMODE_NDELAY;
1734 mode &= ~FMODE_NDELAY;
1736 return blkdev_ioctl(bdev, mode, cmd, arg);
1740 * Write data to the block device. Only intended for the block device itself
1741 * and the raw driver which basically is a fake block device.
1743 * Does not take i_mutex for the write and thus is not for general purpose
1746 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1748 struct file *file = iocb->ki_filp;
1749 struct inode *bd_inode = bdev_file_inode(file);
1750 loff_t size = i_size_read(bd_inode);
1751 struct blk_plug plug;
1754 if (bdev_read_only(I_BDEV(bd_inode)))
1757 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
1760 if (!iov_iter_count(from))
1763 if (iocb->ki_pos >= size)
1766 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1769 iov_iter_truncate(from, size - iocb->ki_pos);
1771 blk_start_plug(&plug);
1772 ret = __generic_file_write_iter(iocb, from);
1774 ret = generic_write_sync(iocb, ret);
1775 blk_finish_plug(&plug);
1778 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1780 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1782 struct file *file = iocb->ki_filp;
1783 struct inode *bd_inode = bdev_file_inode(file);
1784 loff_t size = i_size_read(bd_inode);
1785 loff_t pos = iocb->ki_pos;
1791 iov_iter_truncate(to, size);
1792 return generic_file_read_iter(iocb, to);
1794 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1797 * Try to release a page associated with block device when the system
1798 * is under memory pressure.
1800 static int blkdev_releasepage(struct page *page, gfp_t wait)
1802 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1804 if (super && super->s_op->bdev_try_to_free_page)
1805 return super->s_op->bdev_try_to_free_page(super, page, wait);
1807 return try_to_free_buffers(page);
1810 static int blkdev_writepages(struct address_space *mapping,
1811 struct writeback_control *wbc)
1813 return generic_writepages(mapping, wbc);
1816 static const struct address_space_operations def_blk_aops = {
1817 .readpage = blkdev_readpage,
1818 .readahead = blkdev_readahead,
1819 .writepage = blkdev_writepage,
1820 .write_begin = blkdev_write_begin,
1821 .write_end = blkdev_write_end,
1822 .writepages = blkdev_writepages,
1823 .releasepage = blkdev_releasepage,
1824 .direct_IO = blkdev_direct_IO,
1825 .migratepage = buffer_migrate_page_norefs,
1826 .is_dirty_writeback = buffer_check_dirty_writeback,
1829 #define BLKDEV_FALLOC_FL_SUPPORTED \
1830 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1831 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1833 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1836 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1837 loff_t end = start + len - 1;
1841 /* Fail if we don't recognize the flags. */
1842 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1845 /* Don't go off the end of the device. */
1846 isize = i_size_read(bdev->bd_inode);
1850 if (mode & FALLOC_FL_KEEP_SIZE) {
1851 len = isize - start;
1852 end = start + len - 1;
1858 * Don't allow IO that isn't aligned to logical block size.
1860 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1863 /* Invalidate the page cache, including dirty pages. */
1864 error = truncate_bdev_range(bdev, file->f_mode, start, end);
1869 case FALLOC_FL_ZERO_RANGE:
1870 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
1871 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1872 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
1874 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
1875 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1876 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
1878 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
1879 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1889 * Invalidate again; if someone wandered in and dirtied a page,
1890 * the caller will be given -EBUSY. The third argument is
1891 * inclusive, so the rounding here is safe.
1893 return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
1894 start >> PAGE_SHIFT,
1898 const struct file_operations def_blk_fops = {
1899 .open = blkdev_open,
1900 .release = blkdev_close,
1901 .llseek = block_llseek,
1902 .read_iter = blkdev_read_iter,
1903 .write_iter = blkdev_write_iter,
1904 .iopoll = blkdev_iopoll,
1905 .mmap = generic_file_mmap,
1906 .fsync = blkdev_fsync,
1907 .unlocked_ioctl = block_ioctl,
1908 #ifdef CONFIG_COMPAT
1909 .compat_ioctl = compat_blkdev_ioctl,
1911 .splice_read = generic_file_splice_read,
1912 .splice_write = iter_file_splice_write,
1913 .fallocate = blkdev_fallocate,
1917 * lookup_bdev - lookup a struct block_device by name
1918 * @pathname: special file representing the block device
1920 * Get a reference to the blockdevice at @pathname in the current
1921 * namespace if possible and return it. Return ERR_PTR(error)
1924 int lookup_bdev(const char *pathname, dev_t *dev)
1926 struct inode *inode;
1930 if (!pathname || !*pathname)
1933 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1937 inode = d_backing_inode(path.dentry);
1939 if (!S_ISBLK(inode->i_mode))
1942 if (!may_open_dev(&path))
1945 *dev = inode->i_rdev;
1951 EXPORT_SYMBOL(lookup_bdev);
1953 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1955 struct super_block *sb = get_super(bdev);
1960 * no need to lock the super, get_super holds the
1961 * read mutex so the filesystem cannot go away
1962 * under us (->put_super runs with the write lock
1965 shrink_dcache_sb(sb);
1966 res = invalidate_inodes(sb, kill_dirty);
1969 invalidate_bdev(bdev);
1972 EXPORT_SYMBOL(__invalidate_device);
1974 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1976 struct inode *inode, *old_inode = NULL;
1978 spin_lock(&blockdev_superblock->s_inode_list_lock);
1979 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1980 struct address_space *mapping = inode->i_mapping;
1981 struct block_device *bdev;
1983 spin_lock(&inode->i_lock);
1984 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1985 mapping->nrpages == 0) {
1986 spin_unlock(&inode->i_lock);
1990 spin_unlock(&inode->i_lock);
1991 spin_unlock(&blockdev_superblock->s_inode_list_lock);
1993 * We hold a reference to 'inode' so it couldn't have been
1994 * removed from s_inodes list while we dropped the
1995 * s_inode_list_lock We cannot iput the inode now as we can
1996 * be holding the last reference and we cannot iput it under
1997 * s_inode_list_lock. So we keep the reference and iput it
2002 bdev = I_BDEV(inode);
2004 mutex_lock(&bdev->bd_mutex);
2005 if (bdev->bd_openers)
2007 mutex_unlock(&bdev->bd_mutex);
2009 spin_lock(&blockdev_superblock->s_inode_list_lock);
2011 spin_unlock(&blockdev_superblock->s_inode_list_lock);