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)
113 struct block_device *claimed_bdev = NULL;
117 * If we don't hold exclusive handle for the device, upgrade to it
118 * while we discard the buffer cache to avoid discarding buffers
119 * under live filesystem.
121 if (!(mode & FMODE_EXCL)) {
122 claimed_bdev = bdev->bd_contains;
123 err = bd_prepare_to_claim(bdev, claimed_bdev,
124 truncate_bdev_range);
128 truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend);
130 bd_abort_claiming(bdev, claimed_bdev, truncate_bdev_range);
133 EXPORT_SYMBOL(truncate_bdev_range);
135 static void set_init_blocksize(struct block_device *bdev)
137 bdev->bd_inode->i_blkbits = blksize_bits(bdev_logical_block_size(bdev));
140 int set_blocksize(struct block_device *bdev, int size)
142 /* Size must be a power of two, and between 512 and PAGE_SIZE */
143 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
146 /* Size cannot be smaller than the size supported by the device */
147 if (size < bdev_logical_block_size(bdev))
150 /* Don't change the size if it is same as current */
151 if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
153 bdev->bd_inode->i_blkbits = blksize_bits(size);
159 EXPORT_SYMBOL(set_blocksize);
161 int sb_set_blocksize(struct super_block *sb, int size)
163 if (set_blocksize(sb->s_bdev, size))
165 /* If we get here, we know size is power of two
166 * and it's value is between 512 and PAGE_SIZE */
167 sb->s_blocksize = size;
168 sb->s_blocksize_bits = blksize_bits(size);
169 return sb->s_blocksize;
172 EXPORT_SYMBOL(sb_set_blocksize);
174 int sb_min_blocksize(struct super_block *sb, int size)
176 int minsize = bdev_logical_block_size(sb->s_bdev);
179 return sb_set_blocksize(sb, size);
182 EXPORT_SYMBOL(sb_min_blocksize);
185 blkdev_get_block(struct inode *inode, sector_t iblock,
186 struct buffer_head *bh, int create)
188 bh->b_bdev = I_BDEV(inode);
189 bh->b_blocknr = iblock;
190 set_buffer_mapped(bh);
194 static struct inode *bdev_file_inode(struct file *file)
196 return file->f_mapping->host;
199 static unsigned int dio_bio_write_op(struct kiocb *iocb)
201 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
203 /* avoid the need for a I/O completion work item */
204 if (iocb->ki_flags & IOCB_DSYNC)
209 #define DIO_INLINE_BIO_VECS 4
211 static void blkdev_bio_end_io_simple(struct bio *bio)
213 struct task_struct *waiter = bio->bi_private;
215 WRITE_ONCE(bio->bi_private, NULL);
216 blk_wake_io_task(waiter);
220 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
223 struct file *file = iocb->ki_filp;
224 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
225 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
226 loff_t pos = iocb->ki_pos;
227 bool should_dirty = false;
232 if ((pos | iov_iter_alignment(iter)) &
233 (bdev_logical_block_size(bdev) - 1))
236 if (nr_pages <= DIO_INLINE_BIO_VECS)
239 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
245 bio_init(&bio, vecs, nr_pages);
246 bio_set_dev(&bio, bdev);
247 bio.bi_iter.bi_sector = pos >> 9;
248 bio.bi_write_hint = iocb->ki_hint;
249 bio.bi_private = current;
250 bio.bi_end_io = blkdev_bio_end_io_simple;
251 bio.bi_ioprio = iocb->ki_ioprio;
253 ret = bio_iov_iter_get_pages(&bio, iter);
256 ret = bio.bi_iter.bi_size;
258 if (iov_iter_rw(iter) == READ) {
259 bio.bi_opf = REQ_OP_READ;
260 if (iter_is_iovec(iter))
263 bio.bi_opf = dio_bio_write_op(iocb);
264 task_io_account_write(ret);
266 if (iocb->ki_flags & IOCB_HIPRI)
267 bio_set_polled(&bio, iocb);
269 qc = submit_bio(&bio);
271 set_current_state(TASK_UNINTERRUPTIBLE);
272 if (!READ_ONCE(bio.bi_private))
274 if (!(iocb->ki_flags & IOCB_HIPRI) ||
275 !blk_poll(bdev_get_queue(bdev), qc, true))
278 __set_current_state(TASK_RUNNING);
280 bio_release_pages(&bio, should_dirty);
281 if (unlikely(bio.bi_status))
282 ret = blk_status_to_errno(bio.bi_status);
285 if (vecs != inline_vecs)
296 struct task_struct *waiter;
301 bool should_dirty : 1;
306 static struct bio_set blkdev_dio_pool;
308 static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
310 struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
311 struct request_queue *q = bdev_get_queue(bdev);
313 return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
316 static void blkdev_bio_end_io(struct bio *bio)
318 struct blkdev_dio *dio = bio->bi_private;
319 bool should_dirty = dio->should_dirty;
321 if (bio->bi_status && !dio->bio.bi_status)
322 dio->bio.bi_status = bio->bi_status;
324 if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
326 struct kiocb *iocb = dio->iocb;
329 if (likely(!dio->bio.bi_status)) {
333 ret = blk_status_to_errno(dio->bio.bi_status);
336 dio->iocb->ki_complete(iocb, ret, 0);
340 struct task_struct *waiter = dio->waiter;
342 WRITE_ONCE(dio->waiter, NULL);
343 blk_wake_io_task(waiter);
348 bio_check_pages_dirty(bio);
350 bio_release_pages(bio, false);
356 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
358 struct file *file = iocb->ki_filp;
359 struct inode *inode = bdev_file_inode(file);
360 struct block_device *bdev = I_BDEV(inode);
361 struct blk_plug plug;
362 struct blkdev_dio *dio;
364 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
365 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
366 loff_t pos = iocb->ki_pos;
367 blk_qc_t qc = BLK_QC_T_NONE;
370 if ((pos | iov_iter_alignment(iter)) &
371 (bdev_logical_block_size(bdev) - 1))
374 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
376 dio = container_of(bio, struct blkdev_dio, bio);
377 dio->is_sync = is_sync = is_sync_kiocb(iocb);
379 dio->waiter = current;
386 dio->multi_bio = false;
387 dio->should_dirty = is_read && iter_is_iovec(iter);
390 * Don't plug for HIPRI/polled IO, as those should go straight
394 blk_start_plug(&plug);
397 bio_set_dev(bio, bdev);
398 bio->bi_iter.bi_sector = pos >> 9;
399 bio->bi_write_hint = iocb->ki_hint;
400 bio->bi_private = dio;
401 bio->bi_end_io = blkdev_bio_end_io;
402 bio->bi_ioprio = iocb->ki_ioprio;
404 ret = bio_iov_iter_get_pages(bio, iter);
406 bio->bi_status = BLK_STS_IOERR;
412 bio->bi_opf = REQ_OP_READ;
413 if (dio->should_dirty)
414 bio_set_pages_dirty(bio);
416 bio->bi_opf = dio_bio_write_op(iocb);
417 task_io_account_write(bio->bi_iter.bi_size);
420 dio->size += bio->bi_iter.bi_size;
421 pos += bio->bi_iter.bi_size;
423 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
427 if (iocb->ki_flags & IOCB_HIPRI) {
428 bio_set_polled(bio, iocb);
432 qc = submit_bio(bio);
435 WRITE_ONCE(iocb->ki_cookie, qc);
439 if (!dio->multi_bio) {
441 * AIO needs an extra reference to ensure the dio
442 * structure which is embedded into the first bio
447 dio->multi_bio = true;
448 atomic_set(&dio->ref, 2);
450 atomic_inc(&dio->ref);
454 bio = bio_alloc(GFP_KERNEL, nr_pages);
458 blk_finish_plug(&plug);
464 set_current_state(TASK_UNINTERRUPTIBLE);
465 if (!READ_ONCE(dio->waiter))
468 if (!(iocb->ki_flags & IOCB_HIPRI) ||
469 !blk_poll(bdev_get_queue(bdev), qc, true))
472 __set_current_state(TASK_RUNNING);
475 ret = blk_status_to_errno(dio->bio.bi_status);
484 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
488 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
491 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
492 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
494 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
497 static __init int blkdev_init(void)
499 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
501 module_init(blkdev_init);
503 int __sync_blockdev(struct block_device *bdev, int wait)
508 return filemap_flush(bdev->bd_inode->i_mapping);
509 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
513 * Write out and wait upon all the dirty data associated with a block
514 * device via its mapping. Does not take the superblock lock.
516 int sync_blockdev(struct block_device *bdev)
518 return __sync_blockdev(bdev, 1);
520 EXPORT_SYMBOL(sync_blockdev);
523 * Write out and wait upon all dirty data associated with this
524 * device. Filesystem data as well as the underlying block
525 * device. Takes the superblock lock.
527 int fsync_bdev(struct block_device *bdev)
529 struct super_block *sb = get_super(bdev);
531 int res = sync_filesystem(sb);
535 return sync_blockdev(bdev);
537 EXPORT_SYMBOL(fsync_bdev);
540 * freeze_bdev -- lock a filesystem and force it into a consistent state
541 * @bdev: blockdevice to lock
543 * If a superblock is found on this device, we take the s_umount semaphore
544 * on it to make sure nobody unmounts until the snapshot creation is done.
545 * The reference counter (bd_fsfreeze_count) guarantees that only the last
546 * unfreeze process can unfreeze the frozen filesystem actually when multiple
547 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
548 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
551 int freeze_bdev(struct block_device *bdev)
553 struct super_block *sb;
556 mutex_lock(&bdev->bd_fsfreeze_mutex);
557 if (++bdev->bd_fsfreeze_count > 1)
560 sb = get_active_super(bdev);
563 if (sb->s_op->freeze_super)
564 error = sb->s_op->freeze_super(sb);
566 error = freeze_super(sb);
567 deactivate_super(sb);
570 bdev->bd_fsfreeze_count--;
573 bdev->bd_fsfreeze_sb = sb;
578 mutex_unlock(&bdev->bd_fsfreeze_mutex);
581 EXPORT_SYMBOL(freeze_bdev);
584 * thaw_bdev -- unlock filesystem
585 * @bdev: blockdevice to unlock
587 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
589 int thaw_bdev(struct block_device *bdev)
591 struct super_block *sb;
594 mutex_lock(&bdev->bd_fsfreeze_mutex);
595 if (!bdev->bd_fsfreeze_count)
599 if (--bdev->bd_fsfreeze_count > 0)
602 sb = bdev->bd_fsfreeze_sb;
606 if (sb->s_op->thaw_super)
607 error = sb->s_op->thaw_super(sb);
609 error = thaw_super(sb);
611 bdev->bd_fsfreeze_count++;
613 mutex_unlock(&bdev->bd_fsfreeze_mutex);
616 EXPORT_SYMBOL(thaw_bdev);
618 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
620 return block_write_full_page(page, blkdev_get_block, wbc);
623 static int blkdev_readpage(struct file * file, struct page * page)
625 return block_read_full_page(page, blkdev_get_block);
628 static void blkdev_readahead(struct readahead_control *rac)
630 mpage_readahead(rac, blkdev_get_block);
633 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
634 loff_t pos, unsigned len, unsigned flags,
635 struct page **pagep, void **fsdata)
637 return block_write_begin(mapping, pos, len, flags, pagep,
641 static int blkdev_write_end(struct file *file, struct address_space *mapping,
642 loff_t pos, unsigned len, unsigned copied,
643 struct page *page, void *fsdata)
646 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
656 * for a block special file file_inode(file)->i_size is zero
657 * so we compute the size by hand (just as in block_read/write above)
659 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
661 struct inode *bd_inode = bdev_file_inode(file);
664 inode_lock(bd_inode);
665 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
666 inode_unlock(bd_inode);
670 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
672 struct inode *bd_inode = bdev_file_inode(filp);
673 struct block_device *bdev = I_BDEV(bd_inode);
676 error = file_write_and_wait_range(filp, start, end);
681 * There is no need to serialise calls to blkdev_issue_flush with
682 * i_mutex and doing so causes performance issues with concurrent
683 * O_SYNC writers to a block device.
685 error = blkdev_issue_flush(bdev, GFP_KERNEL);
686 if (error == -EOPNOTSUPP)
691 EXPORT_SYMBOL(blkdev_fsync);
694 * bdev_read_page() - Start reading a page from a block device
695 * @bdev: The device to read the page from
696 * @sector: The offset on the device to read the page to (need not be aligned)
697 * @page: The page to read
699 * On entry, the page should be locked. It will be unlocked when the page
700 * has been read. If the block driver implements rw_page synchronously,
701 * that will be true on exit from this function, but it need not be.
703 * Errors returned by this function are usually "soft", eg out of memory, or
704 * queue full; callers should try a different route to read this page rather
705 * than propagate an error back up the stack.
707 * Return: negative errno if an error occurs, 0 if submission was successful.
709 int bdev_read_page(struct block_device *bdev, sector_t sector,
712 const struct block_device_operations *ops = bdev->bd_disk->fops;
713 int result = -EOPNOTSUPP;
715 if (!ops->rw_page || bdev_get_integrity(bdev))
718 result = blk_queue_enter(bdev->bd_disk->queue, 0);
721 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
723 blk_queue_exit(bdev->bd_disk->queue);
728 * bdev_write_page() - Start writing a page to a block device
729 * @bdev: The device to write the page to
730 * @sector: The offset on the device to write the page to (need not be aligned)
731 * @page: The page to write
732 * @wbc: The writeback_control for the write
734 * On entry, the page should be locked and not currently under writeback.
735 * On exit, if the write started successfully, the page will be unlocked and
736 * under writeback. If the write failed already (eg the driver failed to
737 * queue the page to the device), the page will still be locked. If the
738 * caller is a ->writepage implementation, it will need to unlock the page.
740 * Errors returned by this function are usually "soft", eg out of memory, or
741 * queue full; callers should try a different route to write this page rather
742 * than propagate an error back up the stack.
744 * Return: negative errno if an error occurs, 0 if submission was successful.
746 int bdev_write_page(struct block_device *bdev, sector_t sector,
747 struct page *page, struct writeback_control *wbc)
750 const struct block_device_operations *ops = bdev->bd_disk->fops;
752 if (!ops->rw_page || bdev_get_integrity(bdev))
754 result = blk_queue_enter(bdev->bd_disk->queue, 0);
758 set_page_writeback(page);
759 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
762 end_page_writeback(page);
764 clean_page_buffers(page);
767 blk_queue_exit(bdev->bd_disk->queue);
775 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
776 static struct kmem_cache * bdev_cachep __read_mostly;
778 static struct inode *bdev_alloc_inode(struct super_block *sb)
780 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
783 return &ei->vfs_inode;
786 static void bdev_free_inode(struct inode *inode)
788 kmem_cache_free(bdev_cachep, BDEV_I(inode));
791 static void init_once(void *foo)
793 struct bdev_inode *ei = (struct bdev_inode *) foo;
794 struct block_device *bdev = &ei->bdev;
796 memset(bdev, 0, sizeof(*bdev));
797 mutex_init(&bdev->bd_mutex);
799 INIT_LIST_HEAD(&bdev->bd_holder_disks);
801 bdev->bd_bdi = &noop_backing_dev_info;
802 inode_init_once(&ei->vfs_inode);
803 /* Initialize mutex for freeze. */
804 mutex_init(&bdev->bd_fsfreeze_mutex);
807 static void bdev_evict_inode(struct inode *inode)
809 struct block_device *bdev = &BDEV_I(inode)->bdev;
810 truncate_inode_pages_final(&inode->i_data);
811 invalidate_inode_buffers(inode); /* is it needed here? */
813 /* Detach inode from wb early as bdi_put() may free bdi->wb */
814 inode_detach_wb(inode);
815 if (bdev->bd_bdi != &noop_backing_dev_info) {
816 bdi_put(bdev->bd_bdi);
817 bdev->bd_bdi = &noop_backing_dev_info;
821 static const struct super_operations bdev_sops = {
822 .statfs = simple_statfs,
823 .alloc_inode = bdev_alloc_inode,
824 .free_inode = bdev_free_inode,
825 .drop_inode = generic_delete_inode,
826 .evict_inode = bdev_evict_inode,
829 static int bd_init_fs_context(struct fs_context *fc)
831 struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
834 fc->s_iflags |= SB_I_CGROUPWB;
835 ctx->ops = &bdev_sops;
839 static struct file_system_type bd_type = {
841 .init_fs_context = bd_init_fs_context,
842 .kill_sb = kill_anon_super,
845 struct super_block *blockdev_superblock __read_mostly;
846 EXPORT_SYMBOL_GPL(blockdev_superblock);
848 void __init bdev_cache_init(void)
851 static struct vfsmount *bd_mnt;
853 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
854 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
855 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
857 err = register_filesystem(&bd_type);
859 panic("Cannot register bdev pseudo-fs");
860 bd_mnt = kern_mount(&bd_type);
862 panic("Cannot create bdev pseudo-fs");
863 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
866 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
868 struct block_device *bdev;
871 inode = new_inode(blockdev_superblock);
874 inode->i_mode = S_IFBLK;
876 inode->i_data.a_ops = &def_blk_aops;
877 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
879 bdev = I_BDEV(inode);
880 spin_lock_init(&bdev->bd_size_lock);
881 bdev->bd_disk = disk;
882 bdev->bd_partno = partno;
883 bdev->bd_contains = NULL;
884 bdev->bd_super = NULL;
885 bdev->bd_inode = inode;
886 bdev->bd_part_count = 0;
890 void bdev_add(struct block_device *bdev, dev_t dev)
893 bdev->bd_inode->i_rdev = dev;
894 bdev->bd_inode->i_ino = dev;
895 insert_inode_hash(bdev->bd_inode);
898 static struct block_device *bdget(dev_t dev)
902 inode = ilookup(blockdev_superblock, dev);
905 return &BDEV_I(inode)->bdev;
909 * bdgrab -- Grab a reference to an already referenced block device
910 * @bdev: Block device to grab a reference to.
912 * Returns the block_device with an additional reference when successful,
913 * or NULL if the inode is already beeing freed.
915 struct block_device *bdgrab(struct block_device *bdev)
917 if (!igrab(bdev->bd_inode))
921 EXPORT_SYMBOL(bdgrab);
923 struct block_device *bdget_part(struct hd_struct *part)
925 return bdget(part_devt(part));
928 long nr_blockdev_pages(void)
933 spin_lock(&blockdev_superblock->s_inode_list_lock);
934 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
935 ret += inode->i_mapping->nrpages;
936 spin_unlock(&blockdev_superblock->s_inode_list_lock);
941 void bdput(struct block_device *bdev)
943 iput(bdev->bd_inode);
945 EXPORT_SYMBOL(bdput);
948 * bd_may_claim - test whether a block device can be claimed
949 * @bdev: block device of interest
950 * @whole: whole block device containing @bdev, may equal @bdev
951 * @holder: holder trying to claim @bdev
953 * Test whether @bdev can be claimed by @holder.
956 * spin_lock(&bdev_lock).
959 * %true if @bdev can be claimed, %false otherwise.
961 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
964 if (bdev->bd_holder == holder)
965 return true; /* already a holder */
966 else if (bdev->bd_holder != NULL)
967 return false; /* held by someone else */
968 else if (whole == bdev)
969 return true; /* is a whole device which isn't held */
971 else if (whole->bd_holder == bd_may_claim)
972 return true; /* is a partition of a device that is being partitioned */
973 else if (whole->bd_holder != NULL)
974 return false; /* is a partition of a held device */
976 return true; /* is a partition of an un-held device */
980 * bd_prepare_to_claim - claim a block device
981 * @bdev: block device of interest
982 * @whole: the whole device containing @bdev, may equal @bdev
983 * @holder: holder trying to claim @bdev
985 * Claim @bdev. This function fails if @bdev is already claimed by another
986 * holder and waits if another claiming is in progress. return, the caller
987 * has ownership of bd_claiming and bd_holder[s].
990 * 0 if @bdev can be claimed, -EBUSY otherwise.
992 int bd_prepare_to_claim(struct block_device *bdev, struct block_device *whole,
996 spin_lock(&bdev_lock);
997 /* if someone else claimed, fail */
998 if (!bd_may_claim(bdev, whole, holder)) {
999 spin_unlock(&bdev_lock);
1003 /* if claiming is already in progress, wait for it to finish */
1004 if (whole->bd_claiming) {
1005 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1008 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1009 spin_unlock(&bdev_lock);
1011 finish_wait(wq, &wait);
1016 whole->bd_claiming = holder;
1017 spin_unlock(&bdev_lock);
1020 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
1022 static void bd_clear_claiming(struct block_device *whole, void *holder)
1024 lockdep_assert_held(&bdev_lock);
1025 /* tell others that we're done */
1026 BUG_ON(whole->bd_claiming != holder);
1027 whole->bd_claiming = NULL;
1028 wake_up_bit(&whole->bd_claiming, 0);
1032 * bd_finish_claiming - finish claiming of a block device
1033 * @bdev: block device of interest
1034 * @whole: whole block device
1035 * @holder: holder that has claimed @bdev
1037 * Finish exclusive open of a block device. Mark the device as exlusively
1038 * open by the holder and wake up all waiters for exclusive open to finish.
1040 static void bd_finish_claiming(struct block_device *bdev,
1041 struct block_device *whole, void *holder)
1043 spin_lock(&bdev_lock);
1044 BUG_ON(!bd_may_claim(bdev, whole, holder));
1046 * Note that for a whole device bd_holders will be incremented twice,
1047 * and bd_holder will be set to bd_may_claim before being set to holder
1049 whole->bd_holders++;
1050 whole->bd_holder = bd_may_claim;
1052 bdev->bd_holder = holder;
1053 bd_clear_claiming(whole, holder);
1054 spin_unlock(&bdev_lock);
1058 * bd_abort_claiming - abort claiming of a block device
1059 * @bdev: block device of interest
1060 * @whole: whole block device
1061 * @holder: holder that has claimed @bdev
1063 * Abort claiming of a block device when the exclusive open failed. This can be
1064 * also used when exclusive open is not actually desired and we just needed
1065 * to block other exclusive openers for a while.
1067 void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
1070 spin_lock(&bdev_lock);
1071 bd_clear_claiming(whole, holder);
1072 spin_unlock(&bdev_lock);
1074 EXPORT_SYMBOL(bd_abort_claiming);
1077 struct bd_holder_disk {
1078 struct list_head list;
1079 struct gendisk *disk;
1083 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1084 struct gendisk *disk)
1086 struct bd_holder_disk *holder;
1088 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1089 if (holder->disk == disk)
1094 static int add_symlink(struct kobject *from, struct kobject *to)
1096 return sysfs_create_link(from, to, kobject_name(to));
1099 static void del_symlink(struct kobject *from, struct kobject *to)
1101 sysfs_remove_link(from, kobject_name(to));
1105 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1106 * @bdev: the claimed slave bdev
1107 * @disk: the holding disk
1109 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1111 * This functions creates the following sysfs symlinks.
1113 * - from "slaves" directory of the holder @disk to the claimed @bdev
1114 * - from "holders" directory of the @bdev to the holder @disk
1116 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1117 * passed to bd_link_disk_holder(), then:
1119 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1120 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1122 * The caller must have claimed @bdev before calling this function and
1123 * ensure that both @bdev and @disk are valid during the creation and
1124 * lifetime of these symlinks.
1130 * 0 on success, -errno on failure.
1132 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1134 struct bd_holder_disk *holder;
1137 mutex_lock(&bdev->bd_mutex);
1139 WARN_ON_ONCE(!bdev->bd_holder);
1141 /* FIXME: remove the following once add_disk() handles errors */
1142 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1145 holder = bd_find_holder_disk(bdev, disk);
1151 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1157 INIT_LIST_HEAD(&holder->list);
1158 holder->disk = disk;
1161 ret = add_symlink(disk->slave_dir, bdev_kobj(bdev));
1165 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1169 * bdev could be deleted beneath us which would implicitly destroy
1170 * the holder directory. Hold on to it.
1172 kobject_get(bdev->bd_part->holder_dir);
1174 list_add(&holder->list, &bdev->bd_holder_disks);
1178 del_symlink(disk->slave_dir, bdev_kobj(bdev));
1182 mutex_unlock(&bdev->bd_mutex);
1185 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1188 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1189 * @bdev: the calimed slave bdev
1190 * @disk: the holding disk
1192 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1197 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1199 struct bd_holder_disk *holder;
1201 mutex_lock(&bdev->bd_mutex);
1203 holder = bd_find_holder_disk(bdev, disk);
1205 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1206 del_symlink(disk->slave_dir, bdev_kobj(bdev));
1207 del_symlink(bdev->bd_part->holder_dir,
1208 &disk_to_dev(disk)->kobj);
1209 kobject_put(bdev->bd_part->holder_dir);
1210 list_del_init(&holder->list);
1214 mutex_unlock(&bdev->bd_mutex);
1216 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1220 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1221 * @disk: struct gendisk to check
1222 * @bdev: struct bdev to adjust.
1223 * @verbose: if %true log a message about a size change if there is any
1225 * This routine checks to see if the bdev size does not match the disk size
1226 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1229 static void check_disk_size_change(struct gendisk *disk,
1230 struct block_device *bdev, bool verbose)
1232 loff_t disk_size, bdev_size;
1234 spin_lock(&bdev->bd_size_lock);
1235 disk_size = (loff_t)get_capacity(disk) << 9;
1236 bdev_size = i_size_read(bdev->bd_inode);
1237 if (disk_size != bdev_size) {
1240 "%s: detected capacity change from %lld to %lld\n",
1241 disk->disk_name, bdev_size, disk_size);
1243 i_size_write(bdev->bd_inode, disk_size);
1245 spin_unlock(&bdev->bd_size_lock);
1249 * revalidate_disk_size - checks for disk size change and adjusts bdev size.
1250 * @disk: struct gendisk to check
1251 * @verbose: if %true log a message about a size change if there is any
1253 * This routine checks to see if the bdev size does not match the disk size
1254 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1257 void revalidate_disk_size(struct gendisk *disk, bool verbose)
1259 struct block_device *bdev;
1262 * Hidden disks don't have associated bdev so there's no point in
1263 * revalidating them.
1265 if (disk->flags & GENHD_FL_HIDDEN)
1268 bdev = bdget_disk(disk, 0);
1270 check_disk_size_change(disk, bdev, verbose);
1275 void bd_set_nr_sectors(struct block_device *bdev, sector_t sectors)
1277 spin_lock(&bdev->bd_size_lock);
1278 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
1279 spin_unlock(&bdev->bd_size_lock);
1281 EXPORT_SYMBOL(bd_set_nr_sectors);
1283 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1285 int bdev_disk_changed(struct block_device *bdev, bool invalidate)
1287 struct gendisk *disk = bdev->bd_disk;
1290 lockdep_assert_held(&bdev->bd_mutex);
1292 clear_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1295 ret = blk_drop_partitions(bdev);
1300 * Historically we only set the capacity to zero for devices that
1301 * support partitions (independ of actually having partitions created).
1302 * Doing that is rather inconsistent, but changing it broke legacy
1303 * udisks polling for legacy ide-cdrom devices. Use the crude check
1304 * below to get the sane behavior for most device while not breaking
1305 * userspace for this particular setup.
1308 if (disk_part_scan_enabled(disk) ||
1309 !(disk->flags & GENHD_FL_REMOVABLE))
1310 set_capacity(disk, 0);
1312 if (disk->fops->revalidate_disk)
1313 disk->fops->revalidate_disk(disk);
1316 check_disk_size_change(disk, bdev, !invalidate);
1318 if (get_capacity(disk)) {
1319 ret = blk_add_partitions(disk, bdev);
1322 } else if (invalidate) {
1324 * Tell userspace that the media / partition table may have
1327 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
1333 * Only exported for for loop and dasd for historic reasons. Don't use in new
1336 EXPORT_SYMBOL_GPL(bdev_disk_changed);
1341 * mutex_lock(part->bd_mutex)
1342 * mutex_lock_nested(whole->bd_mutex, 1)
1344 static int __blkdev_get(struct block_device *bdev, fmode_t mode)
1346 struct gendisk *disk = bdev->bd_disk;
1349 if (!bdev->bd_openers) {
1350 bdev->bd_contains = bdev;
1352 if (!bdev->bd_partno) {
1354 bdev->bd_part = disk_get_part(disk, 0);
1359 if (disk->fops->open)
1360 ret = disk->fops->open(bdev, mode);
1363 bd_set_nr_sectors(bdev, get_capacity(disk));
1364 set_init_blocksize(bdev);
1368 * If the device is invalidated, rescan partition
1369 * if open succeeded or failed with -ENOMEDIUM.
1370 * The latter is necessary to prevent ghost
1371 * partitions on a removed medium.
1373 if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
1374 (!ret || ret == -ENOMEDIUM))
1375 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1380 struct block_device *whole = bdget_disk(disk, 0);
1382 mutex_lock_nested(&whole->bd_mutex, 1);
1383 ret = __blkdev_get(whole, mode);
1385 mutex_unlock(&whole->bd_mutex);
1389 whole->bd_part_count++;
1390 mutex_unlock(&whole->bd_mutex);
1392 bdev->bd_contains = whole;
1393 bdev->bd_part = disk_get_part(disk, bdev->bd_partno);
1394 if (!(disk->flags & GENHD_FL_UP) ||
1395 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1396 __blkdev_put(whole, mode, 1);
1401 bd_set_nr_sectors(bdev, bdev->bd_part->nr_sects);
1402 set_init_blocksize(bdev);
1405 if (bdev->bd_bdi == &noop_backing_dev_info)
1406 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1408 if (bdev->bd_contains == bdev) {
1410 if (bdev->bd_disk->fops->open)
1411 ret = bdev->bd_disk->fops->open(bdev, mode);
1412 /* the same as first opener case, read comment there */
1413 if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
1414 (!ret || ret == -ENOMEDIUM))
1415 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1424 disk_put_part(bdev->bd_part);
1425 bdev->bd_part = NULL;
1426 bdev->bd_contains = NULL;
1430 struct block_device *blkdev_get_no_open(dev_t dev)
1432 struct block_device *bdev;
1433 struct gendisk *disk;
1435 down_read(&bdev_lookup_sem);
1438 up_read(&bdev_lookup_sem);
1439 blk_request_module(dev);
1440 down_read(&bdev_lookup_sem);
1447 disk = bdev->bd_disk;
1448 if (!kobject_get_unless_zero(&disk_to_dev(disk)->kobj))
1450 if ((disk->flags & (GENHD_FL_UP | GENHD_FL_HIDDEN)) != GENHD_FL_UP)
1452 if (!try_module_get(bdev->bd_disk->fops->owner))
1454 up_read(&bdev_lookup_sem);
1461 up_read(&bdev_lookup_sem);
1465 void blkdev_put_no_open(struct block_device *bdev)
1467 module_put(bdev->bd_disk->fops->owner);
1468 put_disk(bdev->bd_disk);
1473 * blkdev_get_by_dev - open a block device by device number
1474 * @dev: device number of block device to open
1475 * @mode: FMODE_* mask
1476 * @holder: exclusive holder identifier
1478 * Open the block device described by device number @dev. If @mode includes
1479 * %FMODE_EXCL, the block device is opened with exclusive access. Specifying
1480 * %FMODE_EXCL with a %NULL @holder is invalid. Exclusive opens may nest for
1483 * Use this interface ONLY if you really do not have anything better - i.e. when
1484 * you are behind a truly sucky interface and all you are given is a device
1485 * number. Everything else should use blkdev_get_by_path().
1491 * Reference to the block_device on success, ERR_PTR(-errno) on failure.
1493 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1495 struct block_device *claiming;
1496 bool unblock_events = true;
1497 struct block_device *bdev;
1498 struct gendisk *disk;
1501 ret = devcgroup_check_permission(DEVCG_DEV_BLOCK,
1502 MAJOR(dev), MINOR(dev),
1503 ((mode & FMODE_READ) ? DEVCG_ACC_READ : 0) |
1504 ((mode & FMODE_WRITE) ? DEVCG_ACC_WRITE : 0));
1506 return ERR_PTR(ret);
1509 * If we lost a race with 'disk' being deleted, try again. See md.c.
1512 bdev = blkdev_get_no_open(dev);
1514 return ERR_PTR(-ENXIO);
1515 disk = bdev->bd_disk;
1517 if (mode & FMODE_EXCL) {
1518 WARN_ON_ONCE(!holder);
1521 claiming = bdget_disk(disk, 0);
1524 ret = bd_prepare_to_claim(bdev, claiming, holder);
1529 disk_block_events(disk);
1531 mutex_lock(&bdev->bd_mutex);
1532 ret =__blkdev_get(bdev, mode);
1534 goto abort_claiming;
1535 if (mode & FMODE_EXCL) {
1536 bd_finish_claiming(bdev, claiming, holder);
1539 * Block event polling for write claims if requested. Any write
1540 * holder makes the write_holder state stick until all are
1541 * released. This is good enough and tracking individual
1542 * writeable reference is too fragile given the way @mode is
1543 * used in blkdev_get/put().
1545 if ((mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1546 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1547 bdev->bd_write_holder = true;
1548 unblock_events = false;
1551 mutex_unlock(&bdev->bd_mutex);
1554 disk_unblock_events(disk);
1555 if (mode & FMODE_EXCL)
1560 if (mode & FMODE_EXCL)
1561 bd_abort_claiming(bdev, claiming, holder);
1562 mutex_unlock(&bdev->bd_mutex);
1563 disk_unblock_events(disk);
1565 if (mode & FMODE_EXCL)
1568 blkdev_put_no_open(bdev);
1569 if (ret == -ERESTARTSYS)
1571 return ERR_PTR(ret);
1573 EXPORT_SYMBOL(blkdev_get_by_dev);
1576 * blkdev_get_by_path - open a block device by name
1577 * @path: path to the block device to open
1578 * @mode: FMODE_* mask
1579 * @holder: exclusive holder identifier
1581 * Open the block device described by the device file at @path. If @mode
1582 * includes %FMODE_EXCL, the block device is opened with exclusive access.
1583 * Specifying %FMODE_EXCL with a %NULL @holder is invalid. Exclusive opens may
1584 * nest for the same @holder.
1590 * Reference to the block_device on success, ERR_PTR(-errno) on failure.
1592 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1595 struct block_device *bdev;
1599 error = lookup_bdev(path, &dev);
1601 return ERR_PTR(error);
1603 bdev = blkdev_get_by_dev(dev, mode, holder);
1604 if (!IS_ERR(bdev) && (mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1605 blkdev_put(bdev, mode);
1606 return ERR_PTR(-EACCES);
1611 EXPORT_SYMBOL(blkdev_get_by_path);
1613 static int blkdev_open(struct inode * inode, struct file * filp)
1615 struct block_device *bdev;
1618 * Preserve backwards compatibility and allow large file access
1619 * even if userspace doesn't ask for it explicitly. Some mkfs
1620 * binary needs it. We might want to drop this workaround
1621 * during an unstable branch.
1623 filp->f_flags |= O_LARGEFILE;
1625 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
1627 if (filp->f_flags & O_NDELAY)
1628 filp->f_mode |= FMODE_NDELAY;
1629 if (filp->f_flags & O_EXCL)
1630 filp->f_mode |= FMODE_EXCL;
1631 if ((filp->f_flags & O_ACCMODE) == 3)
1632 filp->f_mode |= FMODE_WRITE_IOCTL;
1634 bdev = blkdev_get_by_dev(inode->i_rdev, filp->f_mode, filp);
1636 return PTR_ERR(bdev);
1637 filp->f_mapping = bdev->bd_inode->i_mapping;
1638 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1642 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1644 struct gendisk *disk = bdev->bd_disk;
1645 struct block_device *victim = NULL;
1648 * Sync early if it looks like we're the last one. If someone else
1649 * opens the block device between now and the decrement of bd_openers
1650 * then we did a sync that we didn't need to, but that's not the end
1651 * of the world and we want to avoid long (could be several minute)
1652 * syncs while holding the mutex.
1654 if (bdev->bd_openers == 1)
1655 sync_blockdev(bdev);
1657 mutex_lock_nested(&bdev->bd_mutex, for_part);
1659 bdev->bd_part_count--;
1661 if (!--bdev->bd_openers) {
1662 WARN_ON_ONCE(bdev->bd_holders);
1663 sync_blockdev(bdev);
1665 bdev_write_inode(bdev);
1667 if (!bdev_is_partition(bdev) && disk->fops->release)
1668 disk->fops->release(disk, mode);
1670 disk_put_part(bdev->bd_part);
1671 bdev->bd_part = NULL;
1672 if (bdev_is_partition(bdev))
1673 victim = bdev->bd_contains;
1674 bdev->bd_contains = NULL;
1676 if (!bdev_is_partition(bdev) && disk->fops->release)
1677 disk->fops->release(disk, mode);
1679 mutex_unlock(&bdev->bd_mutex);
1681 __blkdev_put(victim, mode, 1);
1686 void blkdev_put(struct block_device *bdev, fmode_t mode)
1688 struct gendisk *disk = bdev->bd_disk;
1690 mutex_lock(&bdev->bd_mutex);
1692 if (mode & FMODE_EXCL) {
1696 * Release a claim on the device. The holder fields
1697 * are protected with bdev_lock. bd_mutex is to
1698 * synchronize disk_holder unlinking.
1700 spin_lock(&bdev_lock);
1702 WARN_ON_ONCE(--bdev->bd_holders < 0);
1703 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1705 /* bd_contains might point to self, check in a separate step */
1706 if ((bdev_free = !bdev->bd_holders))
1707 bdev->bd_holder = NULL;
1708 if (!bdev->bd_contains->bd_holders)
1709 bdev->bd_contains->bd_holder = NULL;
1711 spin_unlock(&bdev_lock);
1714 * If this was the last claim, remove holder link and
1715 * unblock evpoll if it was a write holder.
1717 if (bdev_free && bdev->bd_write_holder) {
1718 disk_unblock_events(disk);
1719 bdev->bd_write_holder = false;
1724 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1725 * event. This is to ensure detection of media removal commanded
1726 * from userland - e.g. eject(1).
1728 disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE);
1729 mutex_unlock(&bdev->bd_mutex);
1731 __blkdev_put(bdev, mode, 0);
1732 blkdev_put_no_open(bdev);
1734 EXPORT_SYMBOL(blkdev_put);
1736 static int blkdev_close(struct inode * inode, struct file * filp)
1738 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1739 blkdev_put(bdev, filp->f_mode);
1743 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1745 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1746 fmode_t mode = file->f_mode;
1749 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1750 * to updated it before every ioctl.
1752 if (file->f_flags & O_NDELAY)
1753 mode |= FMODE_NDELAY;
1755 mode &= ~FMODE_NDELAY;
1757 return blkdev_ioctl(bdev, mode, cmd, arg);
1761 * Write data to the block device. Only intended for the block device itself
1762 * and the raw driver which basically is a fake block device.
1764 * Does not take i_mutex for the write and thus is not for general purpose
1767 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1769 struct file *file = iocb->ki_filp;
1770 struct inode *bd_inode = bdev_file_inode(file);
1771 loff_t size = i_size_read(bd_inode);
1772 struct blk_plug plug;
1775 if (bdev_read_only(I_BDEV(bd_inode)))
1778 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
1781 if (!iov_iter_count(from))
1784 if (iocb->ki_pos >= size)
1787 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1790 iov_iter_truncate(from, size - iocb->ki_pos);
1792 blk_start_plug(&plug);
1793 ret = __generic_file_write_iter(iocb, from);
1795 ret = generic_write_sync(iocb, ret);
1796 blk_finish_plug(&plug);
1799 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1801 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1803 struct file *file = iocb->ki_filp;
1804 struct inode *bd_inode = bdev_file_inode(file);
1805 loff_t size = i_size_read(bd_inode);
1806 loff_t pos = iocb->ki_pos;
1812 iov_iter_truncate(to, size);
1813 return generic_file_read_iter(iocb, to);
1815 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1818 * Try to release a page associated with block device when the system
1819 * is under memory pressure.
1821 static int blkdev_releasepage(struct page *page, gfp_t wait)
1823 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1825 if (super && super->s_op->bdev_try_to_free_page)
1826 return super->s_op->bdev_try_to_free_page(super, page, wait);
1828 return try_to_free_buffers(page);
1831 static int blkdev_writepages(struct address_space *mapping,
1832 struct writeback_control *wbc)
1834 return generic_writepages(mapping, wbc);
1837 static const struct address_space_operations def_blk_aops = {
1838 .readpage = blkdev_readpage,
1839 .readahead = blkdev_readahead,
1840 .writepage = blkdev_writepage,
1841 .write_begin = blkdev_write_begin,
1842 .write_end = blkdev_write_end,
1843 .writepages = blkdev_writepages,
1844 .releasepage = blkdev_releasepage,
1845 .direct_IO = blkdev_direct_IO,
1846 .migratepage = buffer_migrate_page_norefs,
1847 .is_dirty_writeback = buffer_check_dirty_writeback,
1850 #define BLKDEV_FALLOC_FL_SUPPORTED \
1851 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1852 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1854 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1857 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1858 loff_t end = start + len - 1;
1862 /* Fail if we don't recognize the flags. */
1863 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1866 /* Don't go off the end of the device. */
1867 isize = i_size_read(bdev->bd_inode);
1871 if (mode & FALLOC_FL_KEEP_SIZE) {
1872 len = isize - start;
1873 end = start + len - 1;
1879 * Don't allow IO that isn't aligned to logical block size.
1881 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1884 /* Invalidate the page cache, including dirty pages. */
1885 error = truncate_bdev_range(bdev, file->f_mode, start, end);
1890 case FALLOC_FL_ZERO_RANGE:
1891 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
1892 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1893 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
1895 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
1896 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1897 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
1899 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
1900 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1910 * Invalidate again; if someone wandered in and dirtied a page,
1911 * the caller will be given -EBUSY. The third argument is
1912 * inclusive, so the rounding here is safe.
1914 return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
1915 start >> PAGE_SHIFT,
1919 const struct file_operations def_blk_fops = {
1920 .open = blkdev_open,
1921 .release = blkdev_close,
1922 .llseek = block_llseek,
1923 .read_iter = blkdev_read_iter,
1924 .write_iter = blkdev_write_iter,
1925 .iopoll = blkdev_iopoll,
1926 .mmap = generic_file_mmap,
1927 .fsync = blkdev_fsync,
1928 .unlocked_ioctl = block_ioctl,
1929 #ifdef CONFIG_COMPAT
1930 .compat_ioctl = compat_blkdev_ioctl,
1932 .splice_read = generic_file_splice_read,
1933 .splice_write = iter_file_splice_write,
1934 .fallocate = blkdev_fallocate,
1938 * lookup_bdev - lookup a struct block_device by name
1939 * @pathname: special file representing the block device
1941 * Get a reference to the blockdevice at @pathname in the current
1942 * namespace if possible and return it. Return ERR_PTR(error)
1945 int lookup_bdev(const char *pathname, dev_t *dev)
1947 struct inode *inode;
1951 if (!pathname || !*pathname)
1954 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1958 inode = d_backing_inode(path.dentry);
1960 if (!S_ISBLK(inode->i_mode))
1963 if (!may_open_dev(&path))
1966 *dev = inode->i_rdev;
1972 EXPORT_SYMBOL(lookup_bdev);
1974 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1976 struct super_block *sb = get_super(bdev);
1981 * no need to lock the super, get_super holds the
1982 * read mutex so the filesystem cannot go away
1983 * under us (->put_super runs with the write lock
1986 shrink_dcache_sb(sb);
1987 res = invalidate_inodes(sb, kill_dirty);
1990 invalidate_bdev(bdev);
1993 EXPORT_SYMBOL(__invalidate_device);
1995 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1997 struct inode *inode, *old_inode = NULL;
1999 spin_lock(&blockdev_superblock->s_inode_list_lock);
2000 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2001 struct address_space *mapping = inode->i_mapping;
2002 struct block_device *bdev;
2004 spin_lock(&inode->i_lock);
2005 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2006 mapping->nrpages == 0) {
2007 spin_unlock(&inode->i_lock);
2011 spin_unlock(&inode->i_lock);
2012 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2014 * We hold a reference to 'inode' so it couldn't have been
2015 * removed from s_inodes list while we dropped the
2016 * s_inode_list_lock We cannot iput the inode now as we can
2017 * be holding the last reference and we cannot iput it under
2018 * s_inode_list_lock. So we keep the reference and iput it
2023 bdev = I_BDEV(inode);
2025 mutex_lock(&bdev->bd_mutex);
2026 if (bdev->bd_openers)
2028 mutex_unlock(&bdev->bd_mutex);
2030 spin_lock(&blockdev_superblock->s_inode_list_lock);
2032 spin_unlock(&blockdev_superblock->s_inode_list_lock);