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 struct super_block *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) {
559 * We don't even need to grab a reference - the first call
560 * to freeze_bdev grab an active reference and only the last
561 * thaw_bdev drops it.
563 sb = get_super(bdev);
566 mutex_unlock(&bdev->bd_fsfreeze_mutex);
570 sb = get_active_super(bdev);
573 if (sb->s_op->freeze_super)
574 error = sb->s_op->freeze_super(sb);
576 error = freeze_super(sb);
578 deactivate_super(sb);
579 bdev->bd_fsfreeze_count--;
580 mutex_unlock(&bdev->bd_fsfreeze_mutex);
581 return ERR_PTR(error);
583 deactivate_super(sb);
586 mutex_unlock(&bdev->bd_fsfreeze_mutex);
587 return sb; /* thaw_bdev releases s->s_umount */
589 EXPORT_SYMBOL(freeze_bdev);
592 * thaw_bdev -- unlock filesystem
593 * @bdev: blockdevice to unlock
594 * @sb: associated superblock
596 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
598 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
602 mutex_lock(&bdev->bd_fsfreeze_mutex);
603 if (!bdev->bd_fsfreeze_count)
607 if (--bdev->bd_fsfreeze_count > 0)
613 if (sb->s_op->thaw_super)
614 error = sb->s_op->thaw_super(sb);
616 error = thaw_super(sb);
618 bdev->bd_fsfreeze_count++;
620 mutex_unlock(&bdev->bd_fsfreeze_mutex);
623 EXPORT_SYMBOL(thaw_bdev);
625 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
627 return block_write_full_page(page, blkdev_get_block, wbc);
630 static int blkdev_readpage(struct file * file, struct page * page)
632 return block_read_full_page(page, blkdev_get_block);
635 static void blkdev_readahead(struct readahead_control *rac)
637 mpage_readahead(rac, blkdev_get_block);
640 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
641 loff_t pos, unsigned len, unsigned flags,
642 struct page **pagep, void **fsdata)
644 return block_write_begin(mapping, pos, len, flags, pagep,
648 static int blkdev_write_end(struct file *file, struct address_space *mapping,
649 loff_t pos, unsigned len, unsigned copied,
650 struct page *page, void *fsdata)
653 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
663 * for a block special file file_inode(file)->i_size is zero
664 * so we compute the size by hand (just as in block_read/write above)
666 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
668 struct inode *bd_inode = bdev_file_inode(file);
671 inode_lock(bd_inode);
672 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
673 inode_unlock(bd_inode);
677 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
679 struct inode *bd_inode = bdev_file_inode(filp);
680 struct block_device *bdev = I_BDEV(bd_inode);
683 error = file_write_and_wait_range(filp, start, end);
688 * There is no need to serialise calls to blkdev_issue_flush with
689 * i_mutex and doing so causes performance issues with concurrent
690 * O_SYNC writers to a block device.
692 error = blkdev_issue_flush(bdev, GFP_KERNEL);
693 if (error == -EOPNOTSUPP)
698 EXPORT_SYMBOL(blkdev_fsync);
701 * bdev_read_page() - Start reading a page from a block device
702 * @bdev: The device to read the page from
703 * @sector: The offset on the device to read the page to (need not be aligned)
704 * @page: The page to read
706 * On entry, the page should be locked. It will be unlocked when the page
707 * has been read. If the block driver implements rw_page synchronously,
708 * that will be true on exit from this function, but it need not be.
710 * Errors returned by this function are usually "soft", eg out of memory, or
711 * queue full; callers should try a different route to read this page rather
712 * than propagate an error back up the stack.
714 * Return: negative errno if an error occurs, 0 if submission was successful.
716 int bdev_read_page(struct block_device *bdev, sector_t sector,
719 const struct block_device_operations *ops = bdev->bd_disk->fops;
720 int result = -EOPNOTSUPP;
722 if (!ops->rw_page || bdev_get_integrity(bdev))
725 result = blk_queue_enter(bdev->bd_disk->queue, 0);
728 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
730 blk_queue_exit(bdev->bd_disk->queue);
735 * bdev_write_page() - Start writing a page to a block device
736 * @bdev: The device to write the page to
737 * @sector: The offset on the device to write the page to (need not be aligned)
738 * @page: The page to write
739 * @wbc: The writeback_control for the write
741 * On entry, the page should be locked and not currently under writeback.
742 * On exit, if the write started successfully, the page will be unlocked and
743 * under writeback. If the write failed already (eg the driver failed to
744 * queue the page to the device), the page will still be locked. If the
745 * caller is a ->writepage implementation, it will need to unlock the page.
747 * Errors returned by this function are usually "soft", eg out of memory, or
748 * queue full; callers should try a different route to write this page rather
749 * than propagate an error back up the stack.
751 * Return: negative errno if an error occurs, 0 if submission was successful.
753 int bdev_write_page(struct block_device *bdev, sector_t sector,
754 struct page *page, struct writeback_control *wbc)
757 const struct block_device_operations *ops = bdev->bd_disk->fops;
759 if (!ops->rw_page || bdev_get_integrity(bdev))
761 result = blk_queue_enter(bdev->bd_disk->queue, 0);
765 set_page_writeback(page);
766 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
769 end_page_writeback(page);
771 clean_page_buffers(page);
774 blk_queue_exit(bdev->bd_disk->queue);
782 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
783 static struct kmem_cache * bdev_cachep __read_mostly;
785 static struct inode *bdev_alloc_inode(struct super_block *sb)
787 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
790 return &ei->vfs_inode;
793 static void bdev_free_inode(struct inode *inode)
795 kmem_cache_free(bdev_cachep, BDEV_I(inode));
798 static void init_once(void *foo)
800 struct bdev_inode *ei = (struct bdev_inode *) foo;
801 struct block_device *bdev = &ei->bdev;
803 memset(bdev, 0, sizeof(*bdev));
804 mutex_init(&bdev->bd_mutex);
806 INIT_LIST_HEAD(&bdev->bd_holder_disks);
808 bdev->bd_bdi = &noop_backing_dev_info;
809 inode_init_once(&ei->vfs_inode);
810 /* Initialize mutex for freeze. */
811 mutex_init(&bdev->bd_fsfreeze_mutex);
814 static void bdev_evict_inode(struct inode *inode)
816 struct block_device *bdev = &BDEV_I(inode)->bdev;
817 truncate_inode_pages_final(&inode->i_data);
818 invalidate_inode_buffers(inode); /* is it needed here? */
820 /* Detach inode from wb early as bdi_put() may free bdi->wb */
821 inode_detach_wb(inode);
822 if (bdev->bd_bdi != &noop_backing_dev_info) {
823 bdi_put(bdev->bd_bdi);
824 bdev->bd_bdi = &noop_backing_dev_info;
828 static const struct super_operations bdev_sops = {
829 .statfs = simple_statfs,
830 .alloc_inode = bdev_alloc_inode,
831 .free_inode = bdev_free_inode,
832 .drop_inode = generic_delete_inode,
833 .evict_inode = bdev_evict_inode,
836 static int bd_init_fs_context(struct fs_context *fc)
838 struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
841 fc->s_iflags |= SB_I_CGROUPWB;
842 ctx->ops = &bdev_sops;
846 static struct file_system_type bd_type = {
848 .init_fs_context = bd_init_fs_context,
849 .kill_sb = kill_anon_super,
852 struct super_block *blockdev_superblock __read_mostly;
853 EXPORT_SYMBOL_GPL(blockdev_superblock);
855 void __init bdev_cache_init(void)
858 static struct vfsmount *bd_mnt;
860 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
861 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
862 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
864 err = register_filesystem(&bd_type);
866 panic("Cannot register bdev pseudo-fs");
867 bd_mnt = kern_mount(&bd_type);
869 panic("Cannot create bdev pseudo-fs");
870 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
874 * Most likely _very_ bad one - but then it's hardly critical for small
875 * /dev and can be fixed when somebody will need really large one.
876 * Keep in mind that it will be fed through icache hash function too.
878 static inline unsigned long hash(dev_t dev)
880 return MAJOR(dev)+MINOR(dev);
883 static int bdev_test(struct inode *inode, void *data)
885 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
888 static int bdev_set(struct inode *inode, void *data)
890 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
894 static struct block_device *bdget(dev_t dev)
896 struct block_device *bdev;
899 inode = iget5_locked(blockdev_superblock, hash(dev),
900 bdev_test, bdev_set, &dev);
905 bdev = &BDEV_I(inode)->bdev;
907 if (inode->i_state & I_NEW) {
908 spin_lock_init(&bdev->bd_size_lock);
909 bdev->bd_contains = NULL;
910 bdev->bd_super = NULL;
911 bdev->bd_inode = inode;
912 bdev->bd_part_count = 0;
913 inode->i_mode = S_IFBLK;
915 inode->i_bdev = bdev;
916 inode->i_data.a_ops = &def_blk_aops;
917 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
918 unlock_new_inode(inode);
924 * bdgrab -- Grab a reference to an already referenced block device
925 * @bdev: Block device to grab a reference to.
927 struct block_device *bdgrab(struct block_device *bdev)
929 ihold(bdev->bd_inode);
932 EXPORT_SYMBOL(bdgrab);
934 struct block_device *bdget_part(struct hd_struct *part)
936 return bdget(part_devt(part));
939 long nr_blockdev_pages(void)
944 spin_lock(&blockdev_superblock->s_inode_list_lock);
945 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
946 ret += inode->i_mapping->nrpages;
947 spin_unlock(&blockdev_superblock->s_inode_list_lock);
952 void bdput(struct block_device *bdev)
954 iput(bdev->bd_inode);
957 EXPORT_SYMBOL(bdput);
959 static struct block_device *bd_acquire(struct inode *inode)
961 struct block_device *bdev;
963 spin_lock(&bdev_lock);
964 bdev = inode->i_bdev;
965 if (bdev && !inode_unhashed(bdev->bd_inode)) {
967 spin_unlock(&bdev_lock);
970 spin_unlock(&bdev_lock);
973 * i_bdev references block device inode that was already shut down
974 * (corresponding device got removed). Remove the reference and look
975 * up block device inode again just in case new device got
976 * reestablished under the same device number.
981 bdev = bdget(inode->i_rdev);
983 spin_lock(&bdev_lock);
984 if (!inode->i_bdev) {
986 * We take an additional reference to bd_inode,
987 * and it's released in clear_inode() of inode.
988 * So, we can access it via ->i_mapping always
992 inode->i_bdev = bdev;
993 inode->i_mapping = bdev->bd_inode->i_mapping;
995 spin_unlock(&bdev_lock);
1000 /* Call when you free inode */
1002 void bd_forget(struct inode *inode)
1004 struct block_device *bdev = NULL;
1006 spin_lock(&bdev_lock);
1007 if (!sb_is_blkdev_sb(inode->i_sb))
1008 bdev = inode->i_bdev;
1009 inode->i_bdev = NULL;
1010 inode->i_mapping = &inode->i_data;
1011 spin_unlock(&bdev_lock);
1018 * bd_may_claim - test whether a block device can be claimed
1019 * @bdev: block device of interest
1020 * @whole: whole block device containing @bdev, may equal @bdev
1021 * @holder: holder trying to claim @bdev
1023 * Test whether @bdev can be claimed by @holder.
1026 * spin_lock(&bdev_lock).
1029 * %true if @bdev can be claimed, %false otherwise.
1031 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1034 if (bdev->bd_holder == holder)
1035 return true; /* already a holder */
1036 else if (bdev->bd_holder != NULL)
1037 return false; /* held by someone else */
1038 else if (whole == bdev)
1039 return true; /* is a whole device which isn't held */
1041 else if (whole->bd_holder == bd_may_claim)
1042 return true; /* is a partition of a device that is being partitioned */
1043 else if (whole->bd_holder != NULL)
1044 return false; /* is a partition of a held device */
1046 return true; /* is a partition of an un-held device */
1050 * bd_prepare_to_claim - claim a block device
1051 * @bdev: block device of interest
1052 * @whole: the whole device containing @bdev, may equal @bdev
1053 * @holder: holder trying to claim @bdev
1055 * Claim @bdev. This function fails if @bdev is already claimed by another
1056 * holder and waits if another claiming is in progress. return, the caller
1057 * has ownership of bd_claiming and bd_holder[s].
1060 * 0 if @bdev can be claimed, -EBUSY otherwise.
1062 int bd_prepare_to_claim(struct block_device *bdev, struct block_device *whole,
1066 spin_lock(&bdev_lock);
1067 /* if someone else claimed, fail */
1068 if (!bd_may_claim(bdev, whole, holder)) {
1069 spin_unlock(&bdev_lock);
1073 /* if claiming is already in progress, wait for it to finish */
1074 if (whole->bd_claiming) {
1075 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1078 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1079 spin_unlock(&bdev_lock);
1081 finish_wait(wq, &wait);
1086 whole->bd_claiming = holder;
1087 spin_unlock(&bdev_lock);
1090 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
1092 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1094 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1099 * Now that we hold gendisk reference we make sure bdev we looked up is
1100 * not stale. If it is, it means device got removed and created before
1101 * we looked up gendisk and we fail open in such case. Associating
1102 * unhashed bdev with newly created gendisk could lead to two bdevs
1103 * (and thus two independent caches) being associated with one device
1106 if (inode_unhashed(bdev->bd_inode)) {
1107 put_disk_and_module(disk);
1113 static void bd_clear_claiming(struct block_device *whole, void *holder)
1115 lockdep_assert_held(&bdev_lock);
1116 /* tell others that we're done */
1117 BUG_ON(whole->bd_claiming != holder);
1118 whole->bd_claiming = NULL;
1119 wake_up_bit(&whole->bd_claiming, 0);
1123 * bd_finish_claiming - finish claiming of a block device
1124 * @bdev: block device of interest
1125 * @whole: whole block device
1126 * @holder: holder that has claimed @bdev
1128 * Finish exclusive open of a block device. Mark the device as exlusively
1129 * open by the holder and wake up all waiters for exclusive open to finish.
1131 static void bd_finish_claiming(struct block_device *bdev,
1132 struct block_device *whole, void *holder)
1134 spin_lock(&bdev_lock);
1135 BUG_ON(!bd_may_claim(bdev, whole, holder));
1137 * Note that for a whole device bd_holders will be incremented twice,
1138 * and bd_holder will be set to bd_may_claim before being set to holder
1140 whole->bd_holders++;
1141 whole->bd_holder = bd_may_claim;
1143 bdev->bd_holder = holder;
1144 bd_clear_claiming(whole, holder);
1145 spin_unlock(&bdev_lock);
1149 * bd_abort_claiming - abort claiming of a block device
1150 * @bdev: block device of interest
1151 * @whole: whole block device
1152 * @holder: holder that has claimed @bdev
1154 * Abort claiming of a block device when the exclusive open failed. This can be
1155 * also used when exclusive open is not actually desired and we just needed
1156 * to block other exclusive openers for a while.
1158 void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
1161 spin_lock(&bdev_lock);
1162 bd_clear_claiming(whole, holder);
1163 spin_unlock(&bdev_lock);
1165 EXPORT_SYMBOL(bd_abort_claiming);
1168 struct bd_holder_disk {
1169 struct list_head list;
1170 struct gendisk *disk;
1174 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1175 struct gendisk *disk)
1177 struct bd_holder_disk *holder;
1179 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1180 if (holder->disk == disk)
1185 static int add_symlink(struct kobject *from, struct kobject *to)
1187 return sysfs_create_link(from, to, kobject_name(to));
1190 static void del_symlink(struct kobject *from, struct kobject *to)
1192 sysfs_remove_link(from, kobject_name(to));
1196 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1197 * @bdev: the claimed slave bdev
1198 * @disk: the holding disk
1200 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1202 * This functions creates the following sysfs symlinks.
1204 * - from "slaves" directory of the holder @disk to the claimed @bdev
1205 * - from "holders" directory of the @bdev to the holder @disk
1207 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1208 * passed to bd_link_disk_holder(), then:
1210 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1211 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1213 * The caller must have claimed @bdev before calling this function and
1214 * ensure that both @bdev and @disk are valid during the creation and
1215 * lifetime of these symlinks.
1221 * 0 on success, -errno on failure.
1223 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1225 struct bd_holder_disk *holder;
1228 mutex_lock(&bdev->bd_mutex);
1230 WARN_ON_ONCE(!bdev->bd_holder);
1232 /* FIXME: remove the following once add_disk() handles errors */
1233 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1236 holder = bd_find_holder_disk(bdev, disk);
1242 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1248 INIT_LIST_HEAD(&holder->list);
1249 holder->disk = disk;
1252 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1256 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1260 * bdev could be deleted beneath us which would implicitly destroy
1261 * the holder directory. Hold on to it.
1263 kobject_get(bdev->bd_part->holder_dir);
1265 list_add(&holder->list, &bdev->bd_holder_disks);
1269 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1273 mutex_unlock(&bdev->bd_mutex);
1276 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1279 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1280 * @bdev: the calimed slave bdev
1281 * @disk: the holding disk
1283 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1288 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1290 struct bd_holder_disk *holder;
1292 mutex_lock(&bdev->bd_mutex);
1294 holder = bd_find_holder_disk(bdev, disk);
1296 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1297 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1298 del_symlink(bdev->bd_part->holder_dir,
1299 &disk_to_dev(disk)->kobj);
1300 kobject_put(bdev->bd_part->holder_dir);
1301 list_del_init(&holder->list);
1305 mutex_unlock(&bdev->bd_mutex);
1307 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1311 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1312 * @disk: struct gendisk to check
1313 * @bdev: struct bdev to adjust.
1314 * @verbose: if %true log a message about a size change if there is any
1316 * This routine checks to see if the bdev size does not match the disk size
1317 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1320 static void check_disk_size_change(struct gendisk *disk,
1321 struct block_device *bdev, bool verbose)
1323 loff_t disk_size, bdev_size;
1325 spin_lock(&bdev->bd_size_lock);
1326 disk_size = (loff_t)get_capacity(disk) << 9;
1327 bdev_size = i_size_read(bdev->bd_inode);
1328 if (disk_size != bdev_size) {
1331 "%s: detected capacity change from %lld to %lld\n",
1332 disk->disk_name, bdev_size, disk_size);
1334 i_size_write(bdev->bd_inode, disk_size);
1336 spin_unlock(&bdev->bd_size_lock);
1338 if (bdev_size > disk_size) {
1339 if (__invalidate_device(bdev, false))
1340 pr_warn("VFS: busy inodes on resized disk %s\n",
1346 * revalidate_disk_size - checks for disk size change and adjusts bdev size.
1347 * @disk: struct gendisk to check
1348 * @verbose: if %true log a message about a size change if there is any
1350 * This routine checks to see if the bdev size does not match the disk size
1351 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1354 void revalidate_disk_size(struct gendisk *disk, bool verbose)
1356 struct block_device *bdev;
1359 * Hidden disks don't have associated bdev so there's no point in
1360 * revalidating them.
1362 if (disk->flags & GENHD_FL_HIDDEN)
1365 bdev = bdget_disk(disk, 0);
1367 check_disk_size_change(disk, bdev, verbose);
1371 EXPORT_SYMBOL(revalidate_disk_size);
1373 void bd_set_nr_sectors(struct block_device *bdev, sector_t sectors)
1375 spin_lock(&bdev->bd_size_lock);
1376 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
1377 spin_unlock(&bdev->bd_size_lock);
1379 EXPORT_SYMBOL(bd_set_nr_sectors);
1381 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1383 int bdev_disk_changed(struct block_device *bdev, bool invalidate)
1385 struct gendisk *disk = bdev->bd_disk;
1388 lockdep_assert_held(&bdev->bd_mutex);
1390 clear_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1393 ret = blk_drop_partitions(bdev);
1398 * Historically we only set the capacity to zero for devices that
1399 * support partitions (independ of actually having partitions created).
1400 * Doing that is rather inconsistent, but changing it broke legacy
1401 * udisks polling for legacy ide-cdrom devices. Use the crude check
1402 * below to get the sane behavior for most device while not breaking
1403 * userspace for this particular setup.
1406 if (disk_part_scan_enabled(disk) ||
1407 !(disk->flags & GENHD_FL_REMOVABLE))
1408 set_capacity(disk, 0);
1410 if (disk->fops->revalidate_disk)
1411 disk->fops->revalidate_disk(disk);
1414 check_disk_size_change(disk, bdev, !invalidate);
1416 if (get_capacity(disk)) {
1417 ret = blk_add_partitions(disk, bdev);
1420 } else if (invalidate) {
1422 * Tell userspace that the media / partition table may have
1425 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
1431 * Only exported for for loop and dasd for historic reasons. Don't use in new
1434 EXPORT_SYMBOL_GPL(bdev_disk_changed);
1439 * mutex_lock(part->bd_mutex)
1440 * mutex_lock_nested(whole->bd_mutex, 1)
1443 static int __blkdev_get(struct block_device *bdev, fmode_t mode, void *holder,
1446 struct block_device *whole = NULL, *claiming = NULL;
1447 struct gendisk *disk;
1450 bool first_open = false, unblock_events = true, need_restart;
1453 need_restart = false;
1455 disk = bdev_get_gendisk(bdev, &partno);
1460 whole = bdget_disk(disk, 0);
1467 if (!for_part && (mode & FMODE_EXCL)) {
1468 WARN_ON_ONCE(!holder);
1473 ret = bd_prepare_to_claim(bdev, claiming, holder);
1478 disk_block_events(disk);
1479 mutex_lock_nested(&bdev->bd_mutex, for_part);
1480 if (!bdev->bd_openers) {
1482 bdev->bd_disk = disk;
1483 bdev->bd_contains = bdev;
1484 bdev->bd_partno = partno;
1488 bdev->bd_part = disk_get_part(disk, partno);
1493 if (disk->fops->open) {
1494 ret = disk->fops->open(bdev, mode);
1496 * If we lost a race with 'disk' being deleted,
1497 * try again. See md.c
1499 if (ret == -ERESTARTSYS)
1500 need_restart = true;
1504 bd_set_nr_sectors(bdev, get_capacity(disk));
1505 set_init_blocksize(bdev);
1509 * If the device is invalidated, rescan partition
1510 * if open succeeded or failed with -ENOMEDIUM.
1511 * The latter is necessary to prevent ghost
1512 * partitions on a removed medium.
1514 if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
1515 (!ret || ret == -ENOMEDIUM))
1516 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1522 ret = __blkdev_get(whole, mode, NULL, 1);
1525 bdev->bd_contains = bdgrab(whole);
1526 bdev->bd_part = disk_get_part(disk, partno);
1527 if (!(disk->flags & GENHD_FL_UP) ||
1528 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1532 bd_set_nr_sectors(bdev, bdev->bd_part->nr_sects);
1533 set_init_blocksize(bdev);
1536 if (bdev->bd_bdi == &noop_backing_dev_info)
1537 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1539 if (bdev->bd_contains == bdev) {
1541 if (bdev->bd_disk->fops->open)
1542 ret = bdev->bd_disk->fops->open(bdev, mode);
1543 /* the same as first opener case, read comment there */
1544 if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
1545 (!ret || ret == -ENOMEDIUM))
1546 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1548 goto out_unlock_bdev;
1553 bdev->bd_part_count++;
1555 bd_finish_claiming(bdev, claiming, holder);
1558 * Block event polling for write claims if requested. Any write holder
1559 * makes the write_holder state stick until all are released. This is
1560 * good enough and tracking individual writeable reference is too
1561 * fragile given the way @mode is used in blkdev_get/put().
1563 if (claiming && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1564 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1565 bdev->bd_write_holder = true;
1566 unblock_events = false;
1568 mutex_unlock(&bdev->bd_mutex);
1571 disk_unblock_events(disk);
1573 /* only one opener holds refs to the module and disk */
1575 put_disk_and_module(disk);
1581 disk_put_part(bdev->bd_part);
1582 bdev->bd_disk = NULL;
1583 bdev->bd_part = NULL;
1584 if (bdev != bdev->bd_contains)
1585 __blkdev_put(bdev->bd_contains, mode, 1);
1586 bdev->bd_contains = NULL;
1589 bd_abort_claiming(bdev, claiming, holder);
1590 mutex_unlock(&bdev->bd_mutex);
1591 disk_unblock_events(disk);
1596 put_disk_and_module(disk);
1604 * blkdev_get - open a block device
1605 * @bdev: block_device to open
1606 * @mode: FMODE_* mask
1607 * @holder: exclusive holder identifier
1609 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1610 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1611 * @holder is invalid. Exclusive opens may nest for the same @holder.
1613 * On success, the reference count of @bdev is unchanged. On failure,
1620 * 0 on success, -errno on failure.
1622 static int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1626 if (mode & FMODE_READ)
1628 if (mode & FMODE_WRITE)
1630 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1634 ret =__blkdev_get(bdev, mode, holder, 0);
1645 * blkdev_get_by_path - open a block device by name
1646 * @path: path to the block device to open
1647 * @mode: FMODE_* mask
1648 * @holder: exclusive holder identifier
1650 * Open the blockdevice described by the device file at @path. @mode
1651 * and @holder are identical to blkdev_get().
1653 * On success, the returned block_device has reference count of one.
1659 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1661 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1664 struct block_device *bdev;
1667 bdev = lookup_bdev(path);
1671 err = blkdev_get(bdev, mode, holder);
1673 return ERR_PTR(err);
1675 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1676 blkdev_put(bdev, mode);
1677 return ERR_PTR(-EACCES);
1682 EXPORT_SYMBOL(blkdev_get_by_path);
1685 * blkdev_get_by_dev - open a block device by device number
1686 * @dev: device number of block device to open
1687 * @mode: FMODE_* mask
1688 * @holder: exclusive holder identifier
1690 * Open the blockdevice described by device number @dev. @mode and
1691 * @holder are identical to blkdev_get().
1693 * Use it ONLY if you really do not have anything better - i.e. when
1694 * you are behind a truly sucky interface and all you are given is a
1695 * device number. _Never_ to be used for internal purposes. If you
1696 * ever need it - reconsider your API.
1698 * On success, the returned block_device has reference count of one.
1704 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1706 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1708 struct block_device *bdev;
1713 return ERR_PTR(-ENOMEM);
1715 err = blkdev_get(bdev, mode, holder);
1717 return ERR_PTR(err);
1721 EXPORT_SYMBOL(blkdev_get_by_dev);
1723 static int blkdev_open(struct inode * inode, struct file * filp)
1725 struct block_device *bdev;
1728 * Preserve backwards compatibility and allow large file access
1729 * even if userspace doesn't ask for it explicitly. Some mkfs
1730 * binary needs it. We might want to drop this workaround
1731 * during an unstable branch.
1733 filp->f_flags |= O_LARGEFILE;
1735 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
1737 if (filp->f_flags & O_NDELAY)
1738 filp->f_mode |= FMODE_NDELAY;
1739 if (filp->f_flags & O_EXCL)
1740 filp->f_mode |= FMODE_EXCL;
1741 if ((filp->f_flags & O_ACCMODE) == 3)
1742 filp->f_mode |= FMODE_WRITE_IOCTL;
1744 bdev = bd_acquire(inode);
1748 filp->f_mapping = bdev->bd_inode->i_mapping;
1749 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1751 return blkdev_get(bdev, filp->f_mode, filp);
1754 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1756 struct gendisk *disk = bdev->bd_disk;
1757 struct block_device *victim = NULL;
1760 * Sync early if it looks like we're the last one. If someone else
1761 * opens the block device between now and the decrement of bd_openers
1762 * then we did a sync that we didn't need to, but that's not the end
1763 * of the world and we want to avoid long (could be several minute)
1764 * syncs while holding the mutex.
1766 if (bdev->bd_openers == 1)
1767 sync_blockdev(bdev);
1769 mutex_lock_nested(&bdev->bd_mutex, for_part);
1771 bdev->bd_part_count--;
1773 if (!--bdev->bd_openers) {
1774 WARN_ON_ONCE(bdev->bd_holders);
1775 sync_blockdev(bdev);
1778 bdev_write_inode(bdev);
1780 if (bdev->bd_contains == bdev) {
1781 if (disk->fops->release)
1782 disk->fops->release(disk, mode);
1784 if (!bdev->bd_openers) {
1785 disk_put_part(bdev->bd_part);
1786 bdev->bd_part = NULL;
1787 bdev->bd_disk = NULL;
1788 if (bdev != bdev->bd_contains)
1789 victim = bdev->bd_contains;
1790 bdev->bd_contains = NULL;
1792 put_disk_and_module(disk);
1794 mutex_unlock(&bdev->bd_mutex);
1797 __blkdev_put(victim, mode, 1);
1800 void blkdev_put(struct block_device *bdev, fmode_t mode)
1802 mutex_lock(&bdev->bd_mutex);
1804 if (mode & FMODE_EXCL) {
1808 * Release a claim on the device. The holder fields
1809 * are protected with bdev_lock. bd_mutex is to
1810 * synchronize disk_holder unlinking.
1812 spin_lock(&bdev_lock);
1814 WARN_ON_ONCE(--bdev->bd_holders < 0);
1815 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1817 /* bd_contains might point to self, check in a separate step */
1818 if ((bdev_free = !bdev->bd_holders))
1819 bdev->bd_holder = NULL;
1820 if (!bdev->bd_contains->bd_holders)
1821 bdev->bd_contains->bd_holder = NULL;
1823 spin_unlock(&bdev_lock);
1826 * If this was the last claim, remove holder link and
1827 * unblock evpoll if it was a write holder.
1829 if (bdev_free && bdev->bd_write_holder) {
1830 disk_unblock_events(bdev->bd_disk);
1831 bdev->bd_write_holder = false;
1836 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1837 * event. This is to ensure detection of media removal commanded
1838 * from userland - e.g. eject(1).
1840 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1842 mutex_unlock(&bdev->bd_mutex);
1844 __blkdev_put(bdev, mode, 0);
1846 EXPORT_SYMBOL(blkdev_put);
1848 static int blkdev_close(struct inode * inode, struct file * filp)
1850 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1851 blkdev_put(bdev, filp->f_mode);
1855 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1857 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1858 fmode_t mode = file->f_mode;
1861 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1862 * to updated it before every ioctl.
1864 if (file->f_flags & O_NDELAY)
1865 mode |= FMODE_NDELAY;
1867 mode &= ~FMODE_NDELAY;
1869 return blkdev_ioctl(bdev, mode, cmd, arg);
1873 * Write data to the block device. Only intended for the block device itself
1874 * and the raw driver which basically is a fake block device.
1876 * Does not take i_mutex for the write and thus is not for general purpose
1879 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1881 struct file *file = iocb->ki_filp;
1882 struct inode *bd_inode = bdev_file_inode(file);
1883 loff_t size = i_size_read(bd_inode);
1884 struct blk_plug plug;
1887 if (bdev_read_only(I_BDEV(bd_inode)))
1890 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
1893 if (!iov_iter_count(from))
1896 if (iocb->ki_pos >= size)
1899 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1902 iov_iter_truncate(from, size - iocb->ki_pos);
1904 blk_start_plug(&plug);
1905 ret = __generic_file_write_iter(iocb, from);
1907 ret = generic_write_sync(iocb, ret);
1908 blk_finish_plug(&plug);
1911 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1913 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1915 struct file *file = iocb->ki_filp;
1916 struct inode *bd_inode = bdev_file_inode(file);
1917 loff_t size = i_size_read(bd_inode);
1918 loff_t pos = iocb->ki_pos;
1924 iov_iter_truncate(to, size);
1925 return generic_file_read_iter(iocb, to);
1927 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1930 * Try to release a page associated with block device when the system
1931 * is under memory pressure.
1933 static int blkdev_releasepage(struct page *page, gfp_t wait)
1935 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1937 if (super && super->s_op->bdev_try_to_free_page)
1938 return super->s_op->bdev_try_to_free_page(super, page, wait);
1940 return try_to_free_buffers(page);
1943 static int blkdev_writepages(struct address_space *mapping,
1944 struct writeback_control *wbc)
1946 return generic_writepages(mapping, wbc);
1949 static const struct address_space_operations def_blk_aops = {
1950 .readpage = blkdev_readpage,
1951 .readahead = blkdev_readahead,
1952 .writepage = blkdev_writepage,
1953 .write_begin = blkdev_write_begin,
1954 .write_end = blkdev_write_end,
1955 .writepages = blkdev_writepages,
1956 .releasepage = blkdev_releasepage,
1957 .direct_IO = blkdev_direct_IO,
1958 .migratepage = buffer_migrate_page_norefs,
1959 .is_dirty_writeback = buffer_check_dirty_writeback,
1962 #define BLKDEV_FALLOC_FL_SUPPORTED \
1963 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1964 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1966 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1969 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1970 loff_t end = start + len - 1;
1974 /* Fail if we don't recognize the flags. */
1975 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1978 /* Don't go off the end of the device. */
1979 isize = i_size_read(bdev->bd_inode);
1983 if (mode & FALLOC_FL_KEEP_SIZE) {
1984 len = isize - start;
1985 end = start + len - 1;
1991 * Don't allow IO that isn't aligned to logical block size.
1993 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1996 /* Invalidate the page cache, including dirty pages. */
1997 error = truncate_bdev_range(bdev, file->f_mode, start, end);
2002 case FALLOC_FL_ZERO_RANGE:
2003 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2004 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2005 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2007 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2008 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2009 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2011 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2012 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2022 * Invalidate again; if someone wandered in and dirtied a page,
2023 * the caller will be given -EBUSY. The third argument is
2024 * inclusive, so the rounding here is safe.
2026 return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
2027 start >> PAGE_SHIFT,
2031 const struct file_operations def_blk_fops = {
2032 .open = blkdev_open,
2033 .release = blkdev_close,
2034 .llseek = block_llseek,
2035 .read_iter = blkdev_read_iter,
2036 .write_iter = blkdev_write_iter,
2037 .iopoll = blkdev_iopoll,
2038 .mmap = generic_file_mmap,
2039 .fsync = blkdev_fsync,
2040 .unlocked_ioctl = block_ioctl,
2041 #ifdef CONFIG_COMPAT
2042 .compat_ioctl = compat_blkdev_ioctl,
2044 .splice_read = generic_file_splice_read,
2045 .splice_write = iter_file_splice_write,
2046 .fallocate = blkdev_fallocate,
2050 * lookup_bdev - lookup a struct block_device by name
2051 * @pathname: special file representing the block device
2053 * Get a reference to the blockdevice at @pathname in the current
2054 * namespace if possible and return it. Return ERR_PTR(error)
2057 struct block_device *lookup_bdev(const char *pathname)
2059 struct block_device *bdev;
2060 struct inode *inode;
2064 if (!pathname || !*pathname)
2065 return ERR_PTR(-EINVAL);
2067 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2069 return ERR_PTR(error);
2071 inode = d_backing_inode(path.dentry);
2073 if (!S_ISBLK(inode->i_mode))
2076 if (!may_open_dev(&path))
2079 bdev = bd_acquire(inode);
2086 bdev = ERR_PTR(error);
2089 EXPORT_SYMBOL(lookup_bdev);
2091 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2093 struct super_block *sb = get_super(bdev);
2098 * no need to lock the super, get_super holds the
2099 * read mutex so the filesystem cannot go away
2100 * under us (->put_super runs with the write lock
2103 shrink_dcache_sb(sb);
2104 res = invalidate_inodes(sb, kill_dirty);
2107 invalidate_bdev(bdev);
2110 EXPORT_SYMBOL(__invalidate_device);
2112 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2114 struct inode *inode, *old_inode = NULL;
2116 spin_lock(&blockdev_superblock->s_inode_list_lock);
2117 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2118 struct address_space *mapping = inode->i_mapping;
2119 struct block_device *bdev;
2121 spin_lock(&inode->i_lock);
2122 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2123 mapping->nrpages == 0) {
2124 spin_unlock(&inode->i_lock);
2128 spin_unlock(&inode->i_lock);
2129 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2131 * We hold a reference to 'inode' so it couldn't have been
2132 * removed from s_inodes list while we dropped the
2133 * s_inode_list_lock We cannot iput the inode now as we can
2134 * be holding the last reference and we cannot iput it under
2135 * s_inode_list_lock. So we keep the reference and iput it
2140 bdev = I_BDEV(inode);
2142 mutex_lock(&bdev->bd_mutex);
2143 if (bdev->bd_openers)
2145 mutex_unlock(&bdev->bd_mutex);
2147 spin_lock(&blockdev_superblock->s_inode_list_lock);
2149 spin_unlock(&blockdev_superblock->s_inode_list_lock);