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);
135 * Someone else has handle exclusively open. Try invalidating instead.
136 * The 'end' argument is inclusive so the rounding is safe.
138 return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
139 lstart >> PAGE_SHIFT,
142 EXPORT_SYMBOL(truncate_bdev_range);
144 static void set_init_blocksize(struct block_device *bdev)
146 unsigned int bsize = bdev_logical_block_size(bdev);
147 loff_t size = i_size_read(bdev->bd_inode);
149 while (bsize < PAGE_SIZE) {
154 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
157 int set_blocksize(struct block_device *bdev, int size)
159 /* Size must be a power of two, and between 512 and PAGE_SIZE */
160 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
163 /* Size cannot be smaller than the size supported by the device */
164 if (size < bdev_logical_block_size(bdev))
167 /* Don't change the size if it is same as current */
168 if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
170 bdev->bd_inode->i_blkbits = blksize_bits(size);
176 EXPORT_SYMBOL(set_blocksize);
178 int sb_set_blocksize(struct super_block *sb, int size)
180 if (set_blocksize(sb->s_bdev, size))
182 /* If we get here, we know size is power of two
183 * and it's value is between 512 and PAGE_SIZE */
184 sb->s_blocksize = size;
185 sb->s_blocksize_bits = blksize_bits(size);
186 return sb->s_blocksize;
189 EXPORT_SYMBOL(sb_set_blocksize);
191 int sb_min_blocksize(struct super_block *sb, int size)
193 int minsize = bdev_logical_block_size(sb->s_bdev);
196 return sb_set_blocksize(sb, size);
199 EXPORT_SYMBOL(sb_min_blocksize);
202 blkdev_get_block(struct inode *inode, sector_t iblock,
203 struct buffer_head *bh, int create)
205 bh->b_bdev = I_BDEV(inode);
206 bh->b_blocknr = iblock;
207 set_buffer_mapped(bh);
211 static struct inode *bdev_file_inode(struct file *file)
213 return file->f_mapping->host;
216 static unsigned int dio_bio_write_op(struct kiocb *iocb)
218 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
220 /* avoid the need for a I/O completion work item */
221 if (iocb->ki_flags & IOCB_DSYNC)
226 #define DIO_INLINE_BIO_VECS 4
228 static void blkdev_bio_end_io_simple(struct bio *bio)
230 struct task_struct *waiter = bio->bi_private;
232 WRITE_ONCE(bio->bi_private, NULL);
233 blk_wake_io_task(waiter);
237 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
240 struct file *file = iocb->ki_filp;
241 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
242 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
243 loff_t pos = iocb->ki_pos;
244 bool should_dirty = false;
249 if ((pos | iov_iter_alignment(iter)) &
250 (bdev_logical_block_size(bdev) - 1))
253 if (nr_pages <= DIO_INLINE_BIO_VECS)
256 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
262 bio_init(&bio, vecs, nr_pages);
263 bio_set_dev(&bio, bdev);
264 bio.bi_iter.bi_sector = pos >> 9;
265 bio.bi_write_hint = iocb->ki_hint;
266 bio.bi_private = current;
267 bio.bi_end_io = blkdev_bio_end_io_simple;
268 bio.bi_ioprio = iocb->ki_ioprio;
270 ret = bio_iov_iter_get_pages(&bio, iter);
273 ret = bio.bi_iter.bi_size;
275 if (iov_iter_rw(iter) == READ) {
276 bio.bi_opf = REQ_OP_READ;
277 if (iter_is_iovec(iter))
280 bio.bi_opf = dio_bio_write_op(iocb);
281 task_io_account_write(ret);
283 if (iocb->ki_flags & IOCB_NOWAIT)
284 bio.bi_opf |= REQ_NOWAIT;
285 if (iocb->ki_flags & IOCB_HIPRI)
286 bio_set_polled(&bio, iocb);
288 qc = submit_bio(&bio);
290 set_current_state(TASK_UNINTERRUPTIBLE);
291 if (!READ_ONCE(bio.bi_private))
293 if (!(iocb->ki_flags & IOCB_HIPRI) ||
294 !blk_poll(bdev_get_queue(bdev), qc, true))
297 __set_current_state(TASK_RUNNING);
299 bio_release_pages(&bio, should_dirty);
300 if (unlikely(bio.bi_status))
301 ret = blk_status_to_errno(bio.bi_status);
304 if (vecs != inline_vecs)
315 struct task_struct *waiter;
320 bool should_dirty : 1;
325 static struct bio_set blkdev_dio_pool;
327 static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
329 struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
330 struct request_queue *q = bdev_get_queue(bdev);
332 return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
335 static void blkdev_bio_end_io(struct bio *bio)
337 struct blkdev_dio *dio = bio->bi_private;
338 bool should_dirty = dio->should_dirty;
340 if (bio->bi_status && !dio->bio.bi_status)
341 dio->bio.bi_status = bio->bi_status;
343 if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
345 struct kiocb *iocb = dio->iocb;
348 if (likely(!dio->bio.bi_status)) {
352 ret = blk_status_to_errno(dio->bio.bi_status);
355 dio->iocb->ki_complete(iocb, ret, 0);
359 struct task_struct *waiter = dio->waiter;
361 WRITE_ONCE(dio->waiter, NULL);
362 blk_wake_io_task(waiter);
367 bio_check_pages_dirty(bio);
369 bio_release_pages(bio, false);
375 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
377 struct file *file = iocb->ki_filp;
378 struct inode *inode = bdev_file_inode(file);
379 struct block_device *bdev = I_BDEV(inode);
380 struct blk_plug plug;
381 struct blkdev_dio *dio;
383 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
384 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
385 loff_t pos = iocb->ki_pos;
386 blk_qc_t qc = BLK_QC_T_NONE;
389 if ((pos | iov_iter_alignment(iter)) &
390 (bdev_logical_block_size(bdev) - 1))
393 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
395 dio = container_of(bio, struct blkdev_dio, bio);
396 dio->is_sync = is_sync = is_sync_kiocb(iocb);
398 dio->waiter = current;
405 dio->multi_bio = false;
406 dio->should_dirty = is_read && iter_is_iovec(iter);
409 * Don't plug for HIPRI/polled IO, as those should go straight
413 blk_start_plug(&plug);
416 bio_set_dev(bio, bdev);
417 bio->bi_iter.bi_sector = pos >> 9;
418 bio->bi_write_hint = iocb->ki_hint;
419 bio->bi_private = dio;
420 bio->bi_end_io = blkdev_bio_end_io;
421 bio->bi_ioprio = iocb->ki_ioprio;
423 ret = bio_iov_iter_get_pages(bio, iter);
425 bio->bi_status = BLK_STS_IOERR;
431 bio->bi_opf = REQ_OP_READ;
432 if (dio->should_dirty)
433 bio_set_pages_dirty(bio);
435 bio->bi_opf = dio_bio_write_op(iocb);
436 task_io_account_write(bio->bi_iter.bi_size);
438 if (iocb->ki_flags & IOCB_NOWAIT)
439 bio->bi_opf |= REQ_NOWAIT;
441 dio->size += bio->bi_iter.bi_size;
442 pos += bio->bi_iter.bi_size;
444 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
448 if (iocb->ki_flags & IOCB_HIPRI) {
449 bio_set_polled(bio, iocb);
453 qc = submit_bio(bio);
456 WRITE_ONCE(iocb->ki_cookie, qc);
460 if (!dio->multi_bio) {
462 * AIO needs an extra reference to ensure the dio
463 * structure which is embedded into the first bio
468 dio->multi_bio = true;
469 atomic_set(&dio->ref, 2);
471 atomic_inc(&dio->ref);
475 bio = bio_alloc(GFP_KERNEL, nr_pages);
479 blk_finish_plug(&plug);
485 set_current_state(TASK_UNINTERRUPTIBLE);
486 if (!READ_ONCE(dio->waiter))
489 if (!(iocb->ki_flags & IOCB_HIPRI) ||
490 !blk_poll(bdev_get_queue(bdev), qc, true))
493 __set_current_state(TASK_RUNNING);
496 ret = blk_status_to_errno(dio->bio.bi_status);
505 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
509 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
512 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
513 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
515 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
518 static __init int blkdev_init(void)
520 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
522 module_init(blkdev_init);
524 int __sync_blockdev(struct block_device *bdev, int wait)
529 return filemap_flush(bdev->bd_inode->i_mapping);
530 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
534 * Write out and wait upon all the dirty data associated with a block
535 * device via its mapping. Does not take the superblock lock.
537 int sync_blockdev(struct block_device *bdev)
539 return __sync_blockdev(bdev, 1);
541 EXPORT_SYMBOL(sync_blockdev);
544 * Write out and wait upon all dirty data associated with this
545 * device. Filesystem data as well as the underlying block
546 * device. Takes the superblock lock.
548 int fsync_bdev(struct block_device *bdev)
550 struct super_block *sb = get_super(bdev);
552 int res = sync_filesystem(sb);
556 return sync_blockdev(bdev);
558 EXPORT_SYMBOL(fsync_bdev);
561 * freeze_bdev -- lock a filesystem and force it into a consistent state
562 * @bdev: blockdevice to lock
564 * If a superblock is found on this device, we take the s_umount semaphore
565 * on it to make sure nobody unmounts until the snapshot creation is done.
566 * The reference counter (bd_fsfreeze_count) guarantees that only the last
567 * unfreeze process can unfreeze the frozen filesystem actually when multiple
568 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
569 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
572 struct super_block *freeze_bdev(struct block_device *bdev)
574 struct super_block *sb;
577 mutex_lock(&bdev->bd_fsfreeze_mutex);
578 if (++bdev->bd_fsfreeze_count > 1) {
580 * We don't even need to grab a reference - the first call
581 * to freeze_bdev grab an active reference and only the last
582 * thaw_bdev drops it.
584 sb = get_super(bdev);
587 mutex_unlock(&bdev->bd_fsfreeze_mutex);
591 sb = get_active_super(bdev);
594 if (sb->s_op->freeze_super)
595 error = sb->s_op->freeze_super(sb);
597 error = freeze_super(sb);
599 deactivate_super(sb);
600 bdev->bd_fsfreeze_count--;
601 mutex_unlock(&bdev->bd_fsfreeze_mutex);
602 return ERR_PTR(error);
604 deactivate_super(sb);
607 mutex_unlock(&bdev->bd_fsfreeze_mutex);
608 return sb; /* thaw_bdev releases s->s_umount */
610 EXPORT_SYMBOL(freeze_bdev);
613 * thaw_bdev -- unlock filesystem
614 * @bdev: blockdevice to unlock
615 * @sb: associated superblock
617 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
619 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
623 mutex_lock(&bdev->bd_fsfreeze_mutex);
624 if (!bdev->bd_fsfreeze_count)
628 if (--bdev->bd_fsfreeze_count > 0)
634 if (sb->s_op->thaw_super)
635 error = sb->s_op->thaw_super(sb);
637 error = thaw_super(sb);
639 bdev->bd_fsfreeze_count++;
641 mutex_unlock(&bdev->bd_fsfreeze_mutex);
644 EXPORT_SYMBOL(thaw_bdev);
646 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
648 return block_write_full_page(page, blkdev_get_block, wbc);
651 static int blkdev_readpage(struct file * file, struct page * page)
653 return block_read_full_page(page, blkdev_get_block);
656 static void blkdev_readahead(struct readahead_control *rac)
658 mpage_readahead(rac, blkdev_get_block);
661 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
662 loff_t pos, unsigned len, unsigned flags,
663 struct page **pagep, void **fsdata)
665 return block_write_begin(mapping, pos, len, flags, pagep,
669 static int blkdev_write_end(struct file *file, struct address_space *mapping,
670 loff_t pos, unsigned len, unsigned copied,
671 struct page *page, void *fsdata)
674 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
684 * for a block special file file_inode(file)->i_size is zero
685 * so we compute the size by hand (just as in block_read/write above)
687 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
689 struct inode *bd_inode = bdev_file_inode(file);
692 inode_lock(bd_inode);
693 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
694 inode_unlock(bd_inode);
698 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
700 struct inode *bd_inode = bdev_file_inode(filp);
701 struct block_device *bdev = I_BDEV(bd_inode);
704 error = file_write_and_wait_range(filp, start, end);
709 * There is no need to serialise calls to blkdev_issue_flush with
710 * i_mutex and doing so causes performance issues with concurrent
711 * O_SYNC writers to a block device.
713 error = blkdev_issue_flush(bdev, GFP_KERNEL);
714 if (error == -EOPNOTSUPP)
719 EXPORT_SYMBOL(blkdev_fsync);
722 * bdev_read_page() - Start reading a page from a block device
723 * @bdev: The device to read the page from
724 * @sector: The offset on the device to read the page to (need not be aligned)
725 * @page: The page to read
727 * On entry, the page should be locked. It will be unlocked when the page
728 * has been read. If the block driver implements rw_page synchronously,
729 * that will be true on exit from this function, but it need not be.
731 * Errors returned by this function are usually "soft", eg out of memory, or
732 * queue full; callers should try a different route to read this page rather
733 * than propagate an error back up the stack.
735 * Return: negative errno if an error occurs, 0 if submission was successful.
737 int bdev_read_page(struct block_device *bdev, sector_t sector,
740 const struct block_device_operations *ops = bdev->bd_disk->fops;
741 int result = -EOPNOTSUPP;
743 if (!ops->rw_page || bdev_get_integrity(bdev))
746 result = blk_queue_enter(bdev->bd_disk->queue, 0);
749 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
751 blk_queue_exit(bdev->bd_disk->queue);
756 * bdev_write_page() - Start writing a page to a block device
757 * @bdev: The device to write the page to
758 * @sector: The offset on the device to write the page to (need not be aligned)
759 * @page: The page to write
760 * @wbc: The writeback_control for the write
762 * On entry, the page should be locked and not currently under writeback.
763 * On exit, if the write started successfully, the page will be unlocked and
764 * under writeback. If the write failed already (eg the driver failed to
765 * queue the page to the device), the page will still be locked. If the
766 * caller is a ->writepage implementation, it will need to unlock the page.
768 * Errors returned by this function are usually "soft", eg out of memory, or
769 * queue full; callers should try a different route to write this page rather
770 * than propagate an error back up the stack.
772 * Return: negative errno if an error occurs, 0 if submission was successful.
774 int bdev_write_page(struct block_device *bdev, sector_t sector,
775 struct page *page, struct writeback_control *wbc)
778 const struct block_device_operations *ops = bdev->bd_disk->fops;
780 if (!ops->rw_page || bdev_get_integrity(bdev))
782 result = blk_queue_enter(bdev->bd_disk->queue, 0);
786 set_page_writeback(page);
787 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
790 end_page_writeback(page);
792 clean_page_buffers(page);
795 blk_queue_exit(bdev->bd_disk->queue);
803 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
804 static struct kmem_cache * bdev_cachep __read_mostly;
806 static struct inode *bdev_alloc_inode(struct super_block *sb)
808 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
811 return &ei->vfs_inode;
814 static void bdev_free_inode(struct inode *inode)
816 kmem_cache_free(bdev_cachep, BDEV_I(inode));
819 static void init_once(void *foo)
821 struct bdev_inode *ei = (struct bdev_inode *) foo;
822 struct block_device *bdev = &ei->bdev;
824 memset(bdev, 0, sizeof(*bdev));
825 mutex_init(&bdev->bd_mutex);
827 INIT_LIST_HEAD(&bdev->bd_holder_disks);
829 bdev->bd_bdi = &noop_backing_dev_info;
830 inode_init_once(&ei->vfs_inode);
831 /* Initialize mutex for freeze. */
832 mutex_init(&bdev->bd_fsfreeze_mutex);
835 static void bdev_evict_inode(struct inode *inode)
837 struct block_device *bdev = &BDEV_I(inode)->bdev;
838 truncate_inode_pages_final(&inode->i_data);
839 invalidate_inode_buffers(inode); /* is it needed here? */
841 /* Detach inode from wb early as bdi_put() may free bdi->wb */
842 inode_detach_wb(inode);
843 if (bdev->bd_bdi != &noop_backing_dev_info) {
844 bdi_put(bdev->bd_bdi);
845 bdev->bd_bdi = &noop_backing_dev_info;
849 static const struct super_operations bdev_sops = {
850 .statfs = simple_statfs,
851 .alloc_inode = bdev_alloc_inode,
852 .free_inode = bdev_free_inode,
853 .drop_inode = generic_delete_inode,
854 .evict_inode = bdev_evict_inode,
857 static int bd_init_fs_context(struct fs_context *fc)
859 struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
862 fc->s_iflags |= SB_I_CGROUPWB;
863 ctx->ops = &bdev_sops;
867 static struct file_system_type bd_type = {
869 .init_fs_context = bd_init_fs_context,
870 .kill_sb = kill_anon_super,
873 struct super_block *blockdev_superblock __read_mostly;
874 EXPORT_SYMBOL_GPL(blockdev_superblock);
876 void __init bdev_cache_init(void)
879 static struct vfsmount *bd_mnt;
881 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
882 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
883 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
885 err = register_filesystem(&bd_type);
887 panic("Cannot register bdev pseudo-fs");
888 bd_mnt = kern_mount(&bd_type);
890 panic("Cannot create bdev pseudo-fs");
891 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
895 * Most likely _very_ bad one - but then it's hardly critical for small
896 * /dev and can be fixed when somebody will need really large one.
897 * Keep in mind that it will be fed through icache hash function too.
899 static inline unsigned long hash(dev_t dev)
901 return MAJOR(dev)+MINOR(dev);
904 static int bdev_test(struct inode *inode, void *data)
906 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
909 static int bdev_set(struct inode *inode, void *data)
911 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
915 static struct block_device *bdget(dev_t dev)
917 struct block_device *bdev;
920 inode = iget5_locked(blockdev_superblock, hash(dev),
921 bdev_test, bdev_set, &dev);
926 bdev = &BDEV_I(inode)->bdev;
928 if (inode->i_state & I_NEW) {
929 spin_lock_init(&bdev->bd_size_lock);
930 bdev->bd_contains = NULL;
931 bdev->bd_super = NULL;
932 bdev->bd_inode = inode;
933 bdev->bd_part_count = 0;
934 inode->i_mode = S_IFBLK;
936 inode->i_bdev = bdev;
937 inode->i_data.a_ops = &def_blk_aops;
938 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
939 unlock_new_inode(inode);
945 * bdgrab -- Grab a reference to an already referenced block device
946 * @bdev: Block device to grab a reference to.
948 struct block_device *bdgrab(struct block_device *bdev)
950 ihold(bdev->bd_inode);
953 EXPORT_SYMBOL(bdgrab);
955 struct block_device *bdget_part(struct hd_struct *part)
957 return bdget(part_devt(part));
960 long nr_blockdev_pages(void)
965 spin_lock(&blockdev_superblock->s_inode_list_lock);
966 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
967 ret += inode->i_mapping->nrpages;
968 spin_unlock(&blockdev_superblock->s_inode_list_lock);
973 void bdput(struct block_device *bdev)
975 iput(bdev->bd_inode);
978 EXPORT_SYMBOL(bdput);
980 static struct block_device *bd_acquire(struct inode *inode)
982 struct block_device *bdev;
984 spin_lock(&bdev_lock);
985 bdev = inode->i_bdev;
986 if (bdev && !inode_unhashed(bdev->bd_inode)) {
988 spin_unlock(&bdev_lock);
991 spin_unlock(&bdev_lock);
994 * i_bdev references block device inode that was already shut down
995 * (corresponding device got removed). Remove the reference and look
996 * up block device inode again just in case new device got
997 * reestablished under the same device number.
1002 bdev = bdget(inode->i_rdev);
1004 spin_lock(&bdev_lock);
1005 if (!inode->i_bdev) {
1007 * We take an additional reference to bd_inode,
1008 * and it's released in clear_inode() of inode.
1009 * So, we can access it via ->i_mapping always
1013 inode->i_bdev = bdev;
1014 inode->i_mapping = bdev->bd_inode->i_mapping;
1016 spin_unlock(&bdev_lock);
1021 /* Call when you free inode */
1023 void bd_forget(struct inode *inode)
1025 struct block_device *bdev = NULL;
1027 spin_lock(&bdev_lock);
1028 if (!sb_is_blkdev_sb(inode->i_sb))
1029 bdev = inode->i_bdev;
1030 inode->i_bdev = NULL;
1031 inode->i_mapping = &inode->i_data;
1032 spin_unlock(&bdev_lock);
1039 * bd_may_claim - test whether a block device can be claimed
1040 * @bdev: block device of interest
1041 * @whole: whole block device containing @bdev, may equal @bdev
1042 * @holder: holder trying to claim @bdev
1044 * Test whether @bdev can be claimed by @holder.
1047 * spin_lock(&bdev_lock).
1050 * %true if @bdev can be claimed, %false otherwise.
1052 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1055 if (bdev->bd_holder == holder)
1056 return true; /* already a holder */
1057 else if (bdev->bd_holder != NULL)
1058 return false; /* held by someone else */
1059 else if (whole == bdev)
1060 return true; /* is a whole device which isn't held */
1062 else if (whole->bd_holder == bd_may_claim)
1063 return true; /* is a partition of a device that is being partitioned */
1064 else if (whole->bd_holder != NULL)
1065 return false; /* is a partition of a held device */
1067 return true; /* is a partition of an un-held device */
1071 * bd_prepare_to_claim - claim a block device
1072 * @bdev: block device of interest
1073 * @whole: the whole device containing @bdev, may equal @bdev
1074 * @holder: holder trying to claim @bdev
1076 * Claim @bdev. This function fails if @bdev is already claimed by another
1077 * holder and waits if another claiming is in progress. return, the caller
1078 * has ownership of bd_claiming and bd_holder[s].
1081 * 0 if @bdev can be claimed, -EBUSY otherwise.
1083 int bd_prepare_to_claim(struct block_device *bdev, struct block_device *whole,
1087 spin_lock(&bdev_lock);
1088 /* if someone else claimed, fail */
1089 if (!bd_may_claim(bdev, whole, holder)) {
1090 spin_unlock(&bdev_lock);
1094 /* if claiming is already in progress, wait for it to finish */
1095 if (whole->bd_claiming) {
1096 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1099 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1100 spin_unlock(&bdev_lock);
1102 finish_wait(wq, &wait);
1107 whole->bd_claiming = holder;
1108 spin_unlock(&bdev_lock);
1111 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
1113 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1115 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1120 * Now that we hold gendisk reference we make sure bdev we looked up is
1121 * not stale. If it is, it means device got removed and created before
1122 * we looked up gendisk and we fail open in such case. Associating
1123 * unhashed bdev with newly created gendisk could lead to two bdevs
1124 * (and thus two independent caches) being associated with one device
1127 if (inode_unhashed(bdev->bd_inode)) {
1128 put_disk_and_module(disk);
1134 static void bd_clear_claiming(struct block_device *whole, void *holder)
1136 lockdep_assert_held(&bdev_lock);
1137 /* tell others that we're done */
1138 BUG_ON(whole->bd_claiming != holder);
1139 whole->bd_claiming = NULL;
1140 wake_up_bit(&whole->bd_claiming, 0);
1144 * bd_finish_claiming - finish claiming of a block device
1145 * @bdev: block device of interest
1146 * @whole: whole block device
1147 * @holder: holder that has claimed @bdev
1149 * Finish exclusive open of a block device. Mark the device as exlusively
1150 * open by the holder and wake up all waiters for exclusive open to finish.
1152 static void bd_finish_claiming(struct block_device *bdev,
1153 struct block_device *whole, void *holder)
1155 spin_lock(&bdev_lock);
1156 BUG_ON(!bd_may_claim(bdev, whole, holder));
1158 * Note that for a whole device bd_holders will be incremented twice,
1159 * and bd_holder will be set to bd_may_claim before being set to holder
1161 whole->bd_holders++;
1162 whole->bd_holder = bd_may_claim;
1164 bdev->bd_holder = holder;
1165 bd_clear_claiming(whole, holder);
1166 spin_unlock(&bdev_lock);
1170 * bd_abort_claiming - abort claiming of a block device
1171 * @bdev: block device of interest
1172 * @whole: whole block device
1173 * @holder: holder that has claimed @bdev
1175 * Abort claiming of a block device when the exclusive open failed. This can be
1176 * also used when exclusive open is not actually desired and we just needed
1177 * to block other exclusive openers for a while.
1179 void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
1182 spin_lock(&bdev_lock);
1183 bd_clear_claiming(whole, holder);
1184 spin_unlock(&bdev_lock);
1186 EXPORT_SYMBOL(bd_abort_claiming);
1189 struct bd_holder_disk {
1190 struct list_head list;
1191 struct gendisk *disk;
1195 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1196 struct gendisk *disk)
1198 struct bd_holder_disk *holder;
1200 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1201 if (holder->disk == disk)
1206 static int add_symlink(struct kobject *from, struct kobject *to)
1208 return sysfs_create_link(from, to, kobject_name(to));
1211 static void del_symlink(struct kobject *from, struct kobject *to)
1213 sysfs_remove_link(from, kobject_name(to));
1217 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1218 * @bdev: the claimed slave bdev
1219 * @disk: the holding disk
1221 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1223 * This functions creates the following sysfs symlinks.
1225 * - from "slaves" directory of the holder @disk to the claimed @bdev
1226 * - from "holders" directory of the @bdev to the holder @disk
1228 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1229 * passed to bd_link_disk_holder(), then:
1231 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1232 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1234 * The caller must have claimed @bdev before calling this function and
1235 * ensure that both @bdev and @disk are valid during the creation and
1236 * lifetime of these symlinks.
1242 * 0 on success, -errno on failure.
1244 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1246 struct bd_holder_disk *holder;
1249 mutex_lock(&bdev->bd_mutex);
1251 WARN_ON_ONCE(!bdev->bd_holder);
1253 /* FIXME: remove the following once add_disk() handles errors */
1254 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1257 holder = bd_find_holder_disk(bdev, disk);
1263 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1269 INIT_LIST_HEAD(&holder->list);
1270 holder->disk = disk;
1273 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1277 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1281 * bdev could be deleted beneath us which would implicitly destroy
1282 * the holder directory. Hold on to it.
1284 kobject_get(bdev->bd_part->holder_dir);
1286 list_add(&holder->list, &bdev->bd_holder_disks);
1290 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1294 mutex_unlock(&bdev->bd_mutex);
1297 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1300 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1301 * @bdev: the calimed slave bdev
1302 * @disk: the holding disk
1304 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1309 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1311 struct bd_holder_disk *holder;
1313 mutex_lock(&bdev->bd_mutex);
1315 holder = bd_find_holder_disk(bdev, disk);
1317 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1318 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1319 del_symlink(bdev->bd_part->holder_dir,
1320 &disk_to_dev(disk)->kobj);
1321 kobject_put(bdev->bd_part->holder_dir);
1322 list_del_init(&holder->list);
1326 mutex_unlock(&bdev->bd_mutex);
1328 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1332 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1333 * @disk: struct gendisk to check
1334 * @bdev: struct bdev to adjust.
1335 * @verbose: if %true log a message about a size change if there is any
1337 * This routine checks to see if the bdev size does not match the disk size
1338 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1341 static void check_disk_size_change(struct gendisk *disk,
1342 struct block_device *bdev, bool verbose)
1344 loff_t disk_size, bdev_size;
1346 spin_lock(&bdev->bd_size_lock);
1347 disk_size = (loff_t)get_capacity(disk) << 9;
1348 bdev_size = i_size_read(bdev->bd_inode);
1349 if (disk_size != bdev_size) {
1352 "%s: detected capacity change from %lld to %lld\n",
1353 disk->disk_name, bdev_size, disk_size);
1355 i_size_write(bdev->bd_inode, disk_size);
1357 spin_unlock(&bdev->bd_size_lock);
1359 if (bdev_size > disk_size) {
1360 if (__invalidate_device(bdev, false))
1361 pr_warn("VFS: busy inodes on resized disk %s\n",
1367 * revalidate_disk_size - checks for disk size change and adjusts bdev size.
1368 * @disk: struct gendisk to check
1369 * @verbose: if %true log a message about a size change if there is any
1371 * This routine checks to see if the bdev size does not match the disk size
1372 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1375 void revalidate_disk_size(struct gendisk *disk, bool verbose)
1377 struct block_device *bdev;
1380 * Hidden disks don't have associated bdev so there's no point in
1381 * revalidating them.
1383 if (disk->flags & GENHD_FL_HIDDEN)
1386 bdev = bdget_disk(disk, 0);
1388 check_disk_size_change(disk, bdev, verbose);
1392 EXPORT_SYMBOL(revalidate_disk_size);
1394 void bd_set_nr_sectors(struct block_device *bdev, sector_t sectors)
1396 spin_lock(&bdev->bd_size_lock);
1397 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
1398 spin_unlock(&bdev->bd_size_lock);
1400 EXPORT_SYMBOL(bd_set_nr_sectors);
1402 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1404 int bdev_disk_changed(struct block_device *bdev, bool invalidate)
1406 struct gendisk *disk = bdev->bd_disk;
1409 lockdep_assert_held(&bdev->bd_mutex);
1411 if (!(disk->flags & GENHD_FL_UP))
1415 ret = blk_drop_partitions(bdev);
1419 clear_bit(GD_NEED_PART_SCAN, &disk->state);
1422 * Historically we only set the capacity to zero for devices that
1423 * support partitions (independ of actually having partitions created).
1424 * Doing that is rather inconsistent, but changing it broke legacy
1425 * udisks polling for legacy ide-cdrom devices. Use the crude check
1426 * below to get the sane behavior for most device while not breaking
1427 * userspace for this particular setup.
1430 if (disk_part_scan_enabled(disk) ||
1431 !(disk->flags & GENHD_FL_REMOVABLE))
1432 set_capacity(disk, 0);
1434 if (disk->fops->revalidate_disk)
1435 disk->fops->revalidate_disk(disk);
1438 check_disk_size_change(disk, bdev, !invalidate);
1440 if (get_capacity(disk)) {
1441 ret = blk_add_partitions(disk, bdev);
1444 } else if (invalidate) {
1446 * Tell userspace that the media / partition table may have
1449 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
1455 * Only exported for for loop and dasd for historic reasons. Don't use in new
1458 EXPORT_SYMBOL_GPL(bdev_disk_changed);
1463 * mutex_lock(part->bd_mutex)
1464 * mutex_lock_nested(whole->bd_mutex, 1)
1467 static int __blkdev_get(struct block_device *bdev, fmode_t mode, void *holder,
1470 struct block_device *whole = NULL, *claiming = NULL;
1471 struct gendisk *disk;
1474 bool first_open = false, unblock_events = true, need_restart;
1477 need_restart = false;
1479 disk = bdev_get_gendisk(bdev, &partno);
1484 whole = bdget_disk(disk, 0);
1491 if (!for_part && (mode & FMODE_EXCL)) {
1492 WARN_ON_ONCE(!holder);
1497 ret = bd_prepare_to_claim(bdev, claiming, holder);
1502 disk_block_events(disk);
1503 mutex_lock_nested(&bdev->bd_mutex, for_part);
1504 if (!bdev->bd_openers) {
1506 bdev->bd_disk = disk;
1507 bdev->bd_contains = bdev;
1508 bdev->bd_partno = partno;
1512 bdev->bd_part = disk_get_part(disk, partno);
1517 if (disk->fops->open) {
1518 ret = disk->fops->open(bdev, mode);
1520 * If we lost a race with 'disk' being deleted,
1521 * try again. See md.c
1523 if (ret == -ERESTARTSYS)
1524 need_restart = true;
1528 bd_set_nr_sectors(bdev, get_capacity(disk));
1529 set_init_blocksize(bdev);
1533 * If the device is invalidated, rescan partition
1534 * if open succeeded or failed with -ENOMEDIUM.
1535 * The latter is necessary to prevent ghost
1536 * partitions on a removed medium.
1538 if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
1539 (!ret || ret == -ENOMEDIUM))
1540 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1546 ret = __blkdev_get(whole, mode, NULL, 1);
1549 bdev->bd_contains = bdgrab(whole);
1550 bdev->bd_part = disk_get_part(disk, partno);
1551 if (!(disk->flags & GENHD_FL_UP) ||
1552 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1556 bd_set_nr_sectors(bdev, bdev->bd_part->nr_sects);
1557 set_init_blocksize(bdev);
1560 if (bdev->bd_bdi == &noop_backing_dev_info)
1561 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1563 if (bdev->bd_contains == bdev) {
1565 if (bdev->bd_disk->fops->open)
1566 ret = bdev->bd_disk->fops->open(bdev, mode);
1567 /* the same as first opener case, read comment there */
1568 if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
1569 (!ret || ret == -ENOMEDIUM))
1570 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1572 goto out_unlock_bdev;
1577 bdev->bd_part_count++;
1579 bd_finish_claiming(bdev, claiming, holder);
1582 * Block event polling for write claims if requested. Any write holder
1583 * makes the write_holder state stick until all are released. This is
1584 * good enough and tracking individual writeable reference is too
1585 * fragile given the way @mode is used in blkdev_get/put().
1587 if (claiming && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1588 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1589 bdev->bd_write_holder = true;
1590 unblock_events = false;
1592 mutex_unlock(&bdev->bd_mutex);
1595 disk_unblock_events(disk);
1597 /* only one opener holds refs to the module and disk */
1599 put_disk_and_module(disk);
1605 disk_put_part(bdev->bd_part);
1606 bdev->bd_disk = NULL;
1607 bdev->bd_part = NULL;
1608 if (bdev != bdev->bd_contains)
1609 __blkdev_put(bdev->bd_contains, mode, 1);
1610 bdev->bd_contains = NULL;
1613 bd_abort_claiming(bdev, claiming, holder);
1614 mutex_unlock(&bdev->bd_mutex);
1615 disk_unblock_events(disk);
1620 put_disk_and_module(disk);
1628 * blkdev_get - open a block device
1629 * @bdev: block_device to open
1630 * @mode: FMODE_* mask
1631 * @holder: exclusive holder identifier
1633 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1634 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1635 * @holder is invalid. Exclusive opens may nest for the same @holder.
1637 * On success, the reference count of @bdev is unchanged. On failure,
1644 * 0 on success, -errno on failure.
1646 static int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1650 if (mode & FMODE_READ)
1652 if (mode & FMODE_WRITE)
1654 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1658 ret =__blkdev_get(bdev, mode, holder, 0);
1669 * blkdev_get_by_path - open a block device by name
1670 * @path: path to the block device to open
1671 * @mode: FMODE_* mask
1672 * @holder: exclusive holder identifier
1674 * Open the blockdevice described by the device file at @path. @mode
1675 * and @holder are identical to blkdev_get().
1677 * On success, the returned block_device has reference count of one.
1683 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1685 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1688 struct block_device *bdev;
1691 bdev = lookup_bdev(path);
1695 err = blkdev_get(bdev, mode, holder);
1697 return ERR_PTR(err);
1699 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1700 blkdev_put(bdev, mode);
1701 return ERR_PTR(-EACCES);
1706 EXPORT_SYMBOL(blkdev_get_by_path);
1709 * blkdev_get_by_dev - open a block device by device number
1710 * @dev: device number of block device to open
1711 * @mode: FMODE_* mask
1712 * @holder: exclusive holder identifier
1714 * Open the blockdevice described by device number @dev. @mode and
1715 * @holder are identical to blkdev_get().
1717 * Use it ONLY if you really do not have anything better - i.e. when
1718 * you are behind a truly sucky interface and all you are given is a
1719 * device number. _Never_ to be used for internal purposes. If you
1720 * ever need it - reconsider your API.
1722 * On success, the returned block_device has reference count of one.
1728 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1730 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1732 struct block_device *bdev;
1737 return ERR_PTR(-ENOMEM);
1739 err = blkdev_get(bdev, mode, holder);
1741 return ERR_PTR(err);
1745 EXPORT_SYMBOL(blkdev_get_by_dev);
1747 static int blkdev_open(struct inode * inode, struct file * filp)
1749 struct block_device *bdev;
1752 * Preserve backwards compatibility and allow large file access
1753 * even if userspace doesn't ask for it explicitly. Some mkfs
1754 * binary needs it. We might want to drop this workaround
1755 * during an unstable branch.
1757 filp->f_flags |= O_LARGEFILE;
1759 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
1761 if (filp->f_flags & O_NDELAY)
1762 filp->f_mode |= FMODE_NDELAY;
1763 if (filp->f_flags & O_EXCL)
1764 filp->f_mode |= FMODE_EXCL;
1765 if ((filp->f_flags & O_ACCMODE) == 3)
1766 filp->f_mode |= FMODE_WRITE_IOCTL;
1768 bdev = bd_acquire(inode);
1772 filp->f_mapping = bdev->bd_inode->i_mapping;
1773 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1775 return blkdev_get(bdev, filp->f_mode, filp);
1778 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1780 struct gendisk *disk = bdev->bd_disk;
1781 struct block_device *victim = NULL;
1784 * Sync early if it looks like we're the last one. If someone else
1785 * opens the block device between now and the decrement of bd_openers
1786 * then we did a sync that we didn't need to, but that's not the end
1787 * of the world and we want to avoid long (could be several minute)
1788 * syncs while holding the mutex.
1790 if (bdev->bd_openers == 1)
1791 sync_blockdev(bdev);
1793 mutex_lock_nested(&bdev->bd_mutex, for_part);
1795 bdev->bd_part_count--;
1797 if (!--bdev->bd_openers) {
1798 WARN_ON_ONCE(bdev->bd_holders);
1799 sync_blockdev(bdev);
1802 bdev_write_inode(bdev);
1804 if (bdev->bd_contains == bdev) {
1805 if (disk->fops->release)
1806 disk->fops->release(disk, mode);
1808 if (!bdev->bd_openers) {
1809 disk_put_part(bdev->bd_part);
1810 bdev->bd_part = NULL;
1811 bdev->bd_disk = NULL;
1812 if (bdev != bdev->bd_contains)
1813 victim = bdev->bd_contains;
1814 bdev->bd_contains = NULL;
1816 put_disk_and_module(disk);
1818 mutex_unlock(&bdev->bd_mutex);
1821 __blkdev_put(victim, mode, 1);
1824 void blkdev_put(struct block_device *bdev, fmode_t mode)
1826 mutex_lock(&bdev->bd_mutex);
1828 if (mode & FMODE_EXCL) {
1832 * Release a claim on the device. The holder fields
1833 * are protected with bdev_lock. bd_mutex is to
1834 * synchronize disk_holder unlinking.
1836 spin_lock(&bdev_lock);
1838 WARN_ON_ONCE(--bdev->bd_holders < 0);
1839 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1841 /* bd_contains might point to self, check in a separate step */
1842 if ((bdev_free = !bdev->bd_holders))
1843 bdev->bd_holder = NULL;
1844 if (!bdev->bd_contains->bd_holders)
1845 bdev->bd_contains->bd_holder = NULL;
1847 spin_unlock(&bdev_lock);
1850 * If this was the last claim, remove holder link and
1851 * unblock evpoll if it was a write holder.
1853 if (bdev_free && bdev->bd_write_holder) {
1854 disk_unblock_events(bdev->bd_disk);
1855 bdev->bd_write_holder = false;
1860 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1861 * event. This is to ensure detection of media removal commanded
1862 * from userland - e.g. eject(1).
1864 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1866 mutex_unlock(&bdev->bd_mutex);
1868 __blkdev_put(bdev, mode, 0);
1870 EXPORT_SYMBOL(blkdev_put);
1872 static int blkdev_close(struct inode * inode, struct file * filp)
1874 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1875 blkdev_put(bdev, filp->f_mode);
1879 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1881 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1882 fmode_t mode = file->f_mode;
1885 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1886 * to updated it before every ioctl.
1888 if (file->f_flags & O_NDELAY)
1889 mode |= FMODE_NDELAY;
1891 mode &= ~FMODE_NDELAY;
1893 return blkdev_ioctl(bdev, mode, cmd, arg);
1897 * Write data to the block device. Only intended for the block device itself
1898 * and the raw driver which basically is a fake block device.
1900 * Does not take i_mutex for the write and thus is not for general purpose
1903 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1905 struct file *file = iocb->ki_filp;
1906 struct inode *bd_inode = bdev_file_inode(file);
1907 loff_t size = i_size_read(bd_inode);
1908 struct blk_plug plug;
1912 if (bdev_read_only(I_BDEV(bd_inode)))
1915 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
1918 if (!iov_iter_count(from))
1921 if (iocb->ki_pos >= size)
1924 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1927 size -= iocb->ki_pos;
1928 if (iov_iter_count(from) > size) {
1929 shorted = iov_iter_count(from) - size;
1930 iov_iter_truncate(from, size);
1933 blk_start_plug(&plug);
1934 ret = __generic_file_write_iter(iocb, from);
1936 ret = generic_write_sync(iocb, ret);
1937 iov_iter_reexpand(from, iov_iter_count(from) + shorted);
1938 blk_finish_plug(&plug);
1941 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1943 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1945 struct file *file = iocb->ki_filp;
1946 struct inode *bd_inode = bdev_file_inode(file);
1947 loff_t size = i_size_read(bd_inode);
1948 loff_t pos = iocb->ki_pos;
1956 if (iov_iter_count(to) > size) {
1957 shorted = iov_iter_count(to) - size;
1958 iov_iter_truncate(to, size);
1961 ret = generic_file_read_iter(iocb, to);
1962 iov_iter_reexpand(to, iov_iter_count(to) + shorted);
1965 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1968 * Try to release a page associated with block device when the system
1969 * is under memory pressure.
1971 static int blkdev_releasepage(struct page *page, gfp_t wait)
1973 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1975 if (super && super->s_op->bdev_try_to_free_page)
1976 return super->s_op->bdev_try_to_free_page(super, page, wait);
1978 return try_to_free_buffers(page);
1981 static int blkdev_writepages(struct address_space *mapping,
1982 struct writeback_control *wbc)
1984 return generic_writepages(mapping, wbc);
1987 static const struct address_space_operations def_blk_aops = {
1988 .readpage = blkdev_readpage,
1989 .readahead = blkdev_readahead,
1990 .writepage = blkdev_writepage,
1991 .write_begin = blkdev_write_begin,
1992 .write_end = blkdev_write_end,
1993 .writepages = blkdev_writepages,
1994 .releasepage = blkdev_releasepage,
1995 .direct_IO = blkdev_direct_IO,
1996 .migratepage = buffer_migrate_page_norefs,
1997 .is_dirty_writeback = buffer_check_dirty_writeback,
2000 #define BLKDEV_FALLOC_FL_SUPPORTED \
2001 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2002 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2004 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2007 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2008 loff_t end = start + len - 1;
2012 /* Fail if we don't recognize the flags. */
2013 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2016 /* Don't go off the end of the device. */
2017 isize = i_size_read(bdev->bd_inode);
2021 if (mode & FALLOC_FL_KEEP_SIZE) {
2022 len = isize - start;
2023 end = start + len - 1;
2029 * Don't allow IO that isn't aligned to logical block size.
2031 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2035 * Invalidate the page cache, including dirty pages, for valid
2036 * de-allocate mode calls to fallocate().
2039 case FALLOC_FL_ZERO_RANGE:
2040 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2041 error = truncate_bdev_range(bdev, file->f_mode, start, end);
2045 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2046 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2048 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2049 error = truncate_bdev_range(bdev, file->f_mode, start, end);
2053 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2054 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2056 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2057 error = truncate_bdev_range(bdev, file->f_mode, start, end);
2061 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2071 * Invalidate again; if someone wandered in and dirtied a page,
2072 * the caller will be given -EBUSY. The third argument is
2073 * inclusive, so the rounding here is safe.
2075 return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
2076 start >> PAGE_SHIFT,
2080 const struct file_operations def_blk_fops = {
2081 .open = blkdev_open,
2082 .release = blkdev_close,
2083 .llseek = block_llseek,
2084 .read_iter = blkdev_read_iter,
2085 .write_iter = blkdev_write_iter,
2086 .iopoll = blkdev_iopoll,
2087 .mmap = generic_file_mmap,
2088 .fsync = blkdev_fsync,
2089 .unlocked_ioctl = block_ioctl,
2090 #ifdef CONFIG_COMPAT
2091 .compat_ioctl = compat_blkdev_ioctl,
2093 .splice_read = generic_file_splice_read,
2094 .splice_write = iter_file_splice_write,
2095 .fallocate = blkdev_fallocate,
2099 * lookup_bdev - lookup a struct block_device by name
2100 * @pathname: special file representing the block device
2102 * Get a reference to the blockdevice at @pathname in the current
2103 * namespace if possible and return it. Return ERR_PTR(error)
2106 struct block_device *lookup_bdev(const char *pathname)
2108 struct block_device *bdev;
2109 struct inode *inode;
2113 if (!pathname || !*pathname)
2114 return ERR_PTR(-EINVAL);
2116 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2118 return ERR_PTR(error);
2120 inode = d_backing_inode(path.dentry);
2122 if (!S_ISBLK(inode->i_mode))
2125 if (!may_open_dev(&path))
2128 bdev = bd_acquire(inode);
2135 bdev = ERR_PTR(error);
2138 EXPORT_SYMBOL(lookup_bdev);
2140 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2142 struct super_block *sb = get_super(bdev);
2147 * no need to lock the super, get_super holds the
2148 * read mutex so the filesystem cannot go away
2149 * under us (->put_super runs with the write lock
2152 shrink_dcache_sb(sb);
2153 res = invalidate_inodes(sb, kill_dirty);
2156 invalidate_bdev(bdev);
2159 EXPORT_SYMBOL(__invalidate_device);
2161 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2163 struct inode *inode, *old_inode = NULL;
2165 spin_lock(&blockdev_superblock->s_inode_list_lock);
2166 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2167 struct address_space *mapping = inode->i_mapping;
2168 struct block_device *bdev;
2170 spin_lock(&inode->i_lock);
2171 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2172 mapping->nrpages == 0) {
2173 spin_unlock(&inode->i_lock);
2177 spin_unlock(&inode->i_lock);
2178 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2180 * We hold a reference to 'inode' so it couldn't have been
2181 * removed from s_inodes list while we dropped the
2182 * s_inode_list_lock We cannot iput the inode now as we can
2183 * be holding the last reference and we cannot iput it under
2184 * s_inode_list_lock. So we keep the reference and iput it
2189 bdev = I_BDEV(inode);
2191 mutex_lock(&bdev->bd_mutex);
2192 if (bdev->bd_openers)
2194 mutex_unlock(&bdev->bd_mutex);
2196 spin_lock(&blockdev_superblock->s_inode_list_lock);
2198 spin_unlock(&blockdev_superblock->s_inode_list_lock);