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 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;
914 inode->i_mode = S_IFBLK;
916 inode->i_bdev = bdev;
917 inode->i_data.a_ops = &def_blk_aops;
918 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
919 unlock_new_inode(inode);
924 EXPORT_SYMBOL(bdget);
927 * bdgrab -- Grab a reference to an already referenced block device
928 * @bdev: Block device to grab a reference to.
930 struct block_device *bdgrab(struct block_device *bdev)
932 ihold(bdev->bd_inode);
935 EXPORT_SYMBOL(bdgrab);
937 long nr_blockdev_pages(void)
942 spin_lock(&blockdev_superblock->s_inode_list_lock);
943 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
944 ret += inode->i_mapping->nrpages;
945 spin_unlock(&blockdev_superblock->s_inode_list_lock);
950 void bdput(struct block_device *bdev)
952 iput(bdev->bd_inode);
955 EXPORT_SYMBOL(bdput);
957 static struct block_device *bd_acquire(struct inode *inode)
959 struct block_device *bdev;
961 spin_lock(&bdev_lock);
962 bdev = inode->i_bdev;
963 if (bdev && !inode_unhashed(bdev->bd_inode)) {
965 spin_unlock(&bdev_lock);
968 spin_unlock(&bdev_lock);
971 * i_bdev references block device inode that was already shut down
972 * (corresponding device got removed). Remove the reference and look
973 * up block device inode again just in case new device got
974 * reestablished under the same device number.
979 bdev = bdget(inode->i_rdev);
981 spin_lock(&bdev_lock);
982 if (!inode->i_bdev) {
984 * We take an additional reference to bd_inode,
985 * and it's released in clear_inode() of inode.
986 * So, we can access it via ->i_mapping always
990 inode->i_bdev = bdev;
991 inode->i_mapping = bdev->bd_inode->i_mapping;
993 spin_unlock(&bdev_lock);
998 /* Call when you free inode */
1000 void bd_forget(struct inode *inode)
1002 struct block_device *bdev = NULL;
1004 spin_lock(&bdev_lock);
1005 if (!sb_is_blkdev_sb(inode->i_sb))
1006 bdev = inode->i_bdev;
1007 inode->i_bdev = NULL;
1008 inode->i_mapping = &inode->i_data;
1009 spin_unlock(&bdev_lock);
1016 * bd_may_claim - test whether a block device can be claimed
1017 * @bdev: block device of interest
1018 * @whole: whole block device containing @bdev, may equal @bdev
1019 * @holder: holder trying to claim @bdev
1021 * Test whether @bdev can be claimed by @holder.
1024 * spin_lock(&bdev_lock).
1027 * %true if @bdev can be claimed, %false otherwise.
1029 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1032 if (bdev->bd_holder == holder)
1033 return true; /* already a holder */
1034 else if (bdev->bd_holder != NULL)
1035 return false; /* held by someone else */
1036 else if (whole == bdev)
1037 return true; /* is a whole device which isn't held */
1039 else if (whole->bd_holder == bd_may_claim)
1040 return true; /* is a partition of a device that is being partitioned */
1041 else if (whole->bd_holder != NULL)
1042 return false; /* is a partition of a held device */
1044 return true; /* is a partition of an un-held device */
1048 * bd_prepare_to_claim - claim a block device
1049 * @bdev: block device of interest
1050 * @whole: the whole device containing @bdev, may equal @bdev
1051 * @holder: holder trying to claim @bdev
1053 * Claim @bdev. This function fails if @bdev is already claimed by another
1054 * holder and waits if another claiming is in progress. return, the caller
1055 * has ownership of bd_claiming and bd_holder[s].
1058 * 0 if @bdev can be claimed, -EBUSY otherwise.
1060 int bd_prepare_to_claim(struct block_device *bdev, struct block_device *whole,
1064 spin_lock(&bdev_lock);
1065 /* if someone else claimed, fail */
1066 if (!bd_may_claim(bdev, whole, holder)) {
1067 spin_unlock(&bdev_lock);
1071 /* if claiming is already in progress, wait for it to finish */
1072 if (whole->bd_claiming) {
1073 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1076 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1077 spin_unlock(&bdev_lock);
1079 finish_wait(wq, &wait);
1084 whole->bd_claiming = holder;
1085 spin_unlock(&bdev_lock);
1088 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
1090 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1092 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1097 * Now that we hold gendisk reference we make sure bdev we looked up is
1098 * not stale. If it is, it means device got removed and created before
1099 * we looked up gendisk and we fail open in such case. Associating
1100 * unhashed bdev with newly created gendisk could lead to two bdevs
1101 * (and thus two independent caches) being associated with one device
1104 if (inode_unhashed(bdev->bd_inode)) {
1105 put_disk_and_module(disk);
1111 static void bd_clear_claiming(struct block_device *whole, void *holder)
1113 lockdep_assert_held(&bdev_lock);
1114 /* tell others that we're done */
1115 BUG_ON(whole->bd_claiming != holder);
1116 whole->bd_claiming = NULL;
1117 wake_up_bit(&whole->bd_claiming, 0);
1121 * bd_finish_claiming - finish claiming of a block device
1122 * @bdev: block device of interest
1123 * @whole: whole block device
1124 * @holder: holder that has claimed @bdev
1126 * Finish exclusive open of a block device. Mark the device as exlusively
1127 * open by the holder and wake up all waiters for exclusive open to finish.
1129 static void bd_finish_claiming(struct block_device *bdev,
1130 struct block_device *whole, void *holder)
1132 spin_lock(&bdev_lock);
1133 BUG_ON(!bd_may_claim(bdev, whole, holder));
1135 * Note that for a whole device bd_holders will be incremented twice,
1136 * and bd_holder will be set to bd_may_claim before being set to holder
1138 whole->bd_holders++;
1139 whole->bd_holder = bd_may_claim;
1141 bdev->bd_holder = holder;
1142 bd_clear_claiming(whole, holder);
1143 spin_unlock(&bdev_lock);
1147 * bd_abort_claiming - abort claiming of a block device
1148 * @bdev: block device of interest
1149 * @whole: whole block device
1150 * @holder: holder that has claimed @bdev
1152 * Abort claiming of a block device when the exclusive open failed. This can be
1153 * also used when exclusive open is not actually desired and we just needed
1154 * to block other exclusive openers for a while.
1156 void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
1159 spin_lock(&bdev_lock);
1160 bd_clear_claiming(whole, holder);
1161 spin_unlock(&bdev_lock);
1163 EXPORT_SYMBOL(bd_abort_claiming);
1166 struct bd_holder_disk {
1167 struct list_head list;
1168 struct gendisk *disk;
1172 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1173 struct gendisk *disk)
1175 struct bd_holder_disk *holder;
1177 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1178 if (holder->disk == disk)
1183 static int add_symlink(struct kobject *from, struct kobject *to)
1185 return sysfs_create_link(from, to, kobject_name(to));
1188 static void del_symlink(struct kobject *from, struct kobject *to)
1190 sysfs_remove_link(from, kobject_name(to));
1194 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1195 * @bdev: the claimed slave bdev
1196 * @disk: the holding disk
1198 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1200 * This functions creates the following sysfs symlinks.
1202 * - from "slaves" directory of the holder @disk to the claimed @bdev
1203 * - from "holders" directory of the @bdev to the holder @disk
1205 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1206 * passed to bd_link_disk_holder(), then:
1208 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1209 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1211 * The caller must have claimed @bdev before calling this function and
1212 * ensure that both @bdev and @disk are valid during the creation and
1213 * lifetime of these symlinks.
1219 * 0 on success, -errno on failure.
1221 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1223 struct bd_holder_disk *holder;
1226 mutex_lock(&bdev->bd_mutex);
1228 WARN_ON_ONCE(!bdev->bd_holder);
1230 /* FIXME: remove the following once add_disk() handles errors */
1231 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1234 holder = bd_find_holder_disk(bdev, disk);
1240 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1246 INIT_LIST_HEAD(&holder->list);
1247 holder->disk = disk;
1250 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1254 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1258 * bdev could be deleted beneath us which would implicitly destroy
1259 * the holder directory. Hold on to it.
1261 kobject_get(bdev->bd_part->holder_dir);
1263 list_add(&holder->list, &bdev->bd_holder_disks);
1267 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1271 mutex_unlock(&bdev->bd_mutex);
1274 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1277 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1278 * @bdev: the calimed slave bdev
1279 * @disk: the holding disk
1281 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1286 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1288 struct bd_holder_disk *holder;
1290 mutex_lock(&bdev->bd_mutex);
1292 holder = bd_find_holder_disk(bdev, disk);
1294 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1295 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1296 del_symlink(bdev->bd_part->holder_dir,
1297 &disk_to_dev(disk)->kobj);
1298 kobject_put(bdev->bd_part->holder_dir);
1299 list_del_init(&holder->list);
1303 mutex_unlock(&bdev->bd_mutex);
1305 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1309 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1310 * @disk: struct gendisk to check
1311 * @bdev: struct bdev to adjust.
1312 * @verbose: if %true log a message about a size change if there is any
1314 * This routine checks to see if the bdev size does not match the disk size
1315 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1318 static void check_disk_size_change(struct gendisk *disk,
1319 struct block_device *bdev, bool verbose)
1321 loff_t disk_size, bdev_size;
1323 spin_lock(&bdev->bd_size_lock);
1324 disk_size = (loff_t)get_capacity(disk) << 9;
1325 bdev_size = i_size_read(bdev->bd_inode);
1326 if (disk_size != bdev_size) {
1329 "%s: detected capacity change from %lld to %lld\n",
1330 disk->disk_name, bdev_size, disk_size);
1332 i_size_write(bdev->bd_inode, disk_size);
1334 spin_unlock(&bdev->bd_size_lock);
1336 if (bdev_size > disk_size) {
1337 if (__invalidate_device(bdev, false))
1338 pr_warn("VFS: busy inodes on resized disk %s\n",
1344 * revalidate_disk_size - checks for disk size change and adjusts bdev size.
1345 * @disk: struct gendisk to check
1346 * @verbose: if %true log a message about a size change if there is any
1348 * This routine checks to see if the bdev size does not match the disk size
1349 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1352 void revalidate_disk_size(struct gendisk *disk, bool verbose)
1354 struct block_device *bdev;
1357 * Hidden disks don't have associated bdev so there's no point in
1358 * revalidating them.
1360 if (disk->flags & GENHD_FL_HIDDEN)
1363 bdev = bdget_disk(disk, 0);
1365 check_disk_size_change(disk, bdev, verbose);
1369 EXPORT_SYMBOL(revalidate_disk_size);
1371 void bd_set_nr_sectors(struct block_device *bdev, sector_t sectors)
1373 spin_lock(&bdev->bd_size_lock);
1374 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
1375 spin_unlock(&bdev->bd_size_lock);
1377 EXPORT_SYMBOL(bd_set_nr_sectors);
1379 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1381 int bdev_disk_changed(struct block_device *bdev, bool invalidate)
1383 struct gendisk *disk = bdev->bd_disk;
1386 lockdep_assert_held(&bdev->bd_mutex);
1388 clear_bit(BDEV_NEED_PART_SCAN, &bdev->bd_flags);
1391 ret = blk_drop_partitions(bdev);
1396 * Historically we only set the capacity to zero for devices that
1397 * support partitions (independ of actually having partitions created).
1398 * Doing that is rather inconsistent, but changing it broke legacy
1399 * udisks polling for legacy ide-cdrom devices. Use the crude check
1400 * below to get the sane behavior for most device while not breaking
1401 * userspace for this particular setup.
1404 if (disk_part_scan_enabled(disk) ||
1405 !(disk->flags & GENHD_FL_REMOVABLE))
1406 set_capacity(disk, 0);
1408 if (disk->fops->revalidate_disk)
1409 disk->fops->revalidate_disk(disk);
1412 check_disk_size_change(disk, bdev, !invalidate);
1414 if (get_capacity(disk)) {
1415 ret = blk_add_partitions(disk, bdev);
1418 } else if (invalidate) {
1420 * Tell userspace that the media / partition table may have
1423 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
1429 * Only exported for for loop and dasd for historic reasons. Don't use in new
1432 EXPORT_SYMBOL_GPL(bdev_disk_changed);
1437 * mutex_lock(part->bd_mutex)
1438 * mutex_lock_nested(whole->bd_mutex, 1)
1441 static int __blkdev_get(struct block_device *bdev, fmode_t mode, void *holder,
1444 struct block_device *whole = NULL, *claiming = NULL;
1445 struct gendisk *disk;
1448 bool first_open = false, unblock_events = true, need_restart;
1451 need_restart = false;
1453 disk = bdev_get_gendisk(bdev, &partno);
1458 whole = bdget_disk(disk, 0);
1465 if (!for_part && (mode & FMODE_EXCL)) {
1466 WARN_ON_ONCE(!holder);
1471 ret = bd_prepare_to_claim(bdev, claiming, holder);
1476 disk_block_events(disk);
1477 mutex_lock_nested(&bdev->bd_mutex, for_part);
1478 if (!bdev->bd_openers) {
1480 bdev->bd_disk = disk;
1481 bdev->bd_contains = bdev;
1482 bdev->bd_partno = partno;
1486 bdev->bd_part = disk_get_part(disk, partno);
1491 if (disk->fops->open) {
1492 ret = disk->fops->open(bdev, mode);
1494 * If we lost a race with 'disk' being deleted,
1495 * try again. See md.c
1497 if (ret == -ERESTARTSYS)
1498 need_restart = true;
1502 bd_set_nr_sectors(bdev, get_capacity(disk));
1503 set_init_blocksize(bdev);
1507 * If the device is invalidated, rescan partition
1508 * if open succeeded or failed with -ENOMEDIUM.
1509 * The latter is necessary to prevent ghost
1510 * partitions on a removed medium.
1512 if (test_bit(BDEV_NEED_PART_SCAN, &bdev->bd_flags) &&
1513 (!ret || ret == -ENOMEDIUM))
1514 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1520 ret = __blkdev_get(whole, mode, NULL, 1);
1523 bdev->bd_contains = bdgrab(whole);
1524 bdev->bd_part = disk_get_part(disk, partno);
1525 if (!(disk->flags & GENHD_FL_UP) ||
1526 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1530 bd_set_nr_sectors(bdev, bdev->bd_part->nr_sects);
1531 set_init_blocksize(bdev);
1534 if (bdev->bd_bdi == &noop_backing_dev_info)
1535 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1537 if (bdev->bd_contains == bdev) {
1539 if (bdev->bd_disk->fops->open)
1540 ret = bdev->bd_disk->fops->open(bdev, mode);
1541 /* the same as first opener case, read comment there */
1542 if (test_bit(BDEV_NEED_PART_SCAN, &bdev->bd_flags) &&
1543 (!ret || ret == -ENOMEDIUM))
1544 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1546 goto out_unlock_bdev;
1551 bdev->bd_part_count++;
1553 bd_finish_claiming(bdev, claiming, holder);
1556 * Block event polling for write claims if requested. Any write holder
1557 * makes the write_holder state stick until all are released. This is
1558 * good enough and tracking individual writeable reference is too
1559 * fragile given the way @mode is used in blkdev_get/put().
1561 if (claiming && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1562 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1563 bdev->bd_write_holder = true;
1564 unblock_events = false;
1566 mutex_unlock(&bdev->bd_mutex);
1569 disk_unblock_events(disk);
1571 /* only one opener holds refs to the module and disk */
1573 put_disk_and_module(disk);
1579 disk_put_part(bdev->bd_part);
1580 bdev->bd_disk = NULL;
1581 bdev->bd_part = NULL;
1582 if (bdev != bdev->bd_contains)
1583 __blkdev_put(bdev->bd_contains, mode, 1);
1584 bdev->bd_contains = NULL;
1587 bd_abort_claiming(bdev, claiming, holder);
1588 mutex_unlock(&bdev->bd_mutex);
1589 disk_unblock_events(disk);
1594 put_disk_and_module(disk);
1602 * blkdev_get - open a block device
1603 * @bdev: block_device to open
1604 * @mode: FMODE_* mask
1605 * @holder: exclusive holder identifier
1607 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1608 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1609 * @holder is invalid. Exclusive opens may nest for the same @holder.
1611 * On success, the reference count of @bdev is unchanged. On failure,
1618 * 0 on success, -errno on failure.
1620 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1624 if (mode & FMODE_READ)
1626 if (mode & FMODE_WRITE)
1628 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1632 ret =__blkdev_get(bdev, mode, holder, 0);
1641 EXPORT_SYMBOL(blkdev_get);
1644 * blkdev_get_by_path - open a block device by name
1645 * @path: path to the block device to open
1646 * @mode: FMODE_* mask
1647 * @holder: exclusive holder identifier
1649 * Open the blockdevice described by the device file at @path. @mode
1650 * and @holder are identical to blkdev_get().
1652 * On success, the returned block_device has reference count of one.
1658 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1660 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1663 struct block_device *bdev;
1666 bdev = lookup_bdev(path);
1670 err = blkdev_get(bdev, mode, holder);
1672 return ERR_PTR(err);
1674 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1675 blkdev_put(bdev, mode);
1676 return ERR_PTR(-EACCES);
1681 EXPORT_SYMBOL(blkdev_get_by_path);
1684 * blkdev_get_by_dev - open a block device by device number
1685 * @dev: device number of block device to open
1686 * @mode: FMODE_* mask
1687 * @holder: exclusive holder identifier
1689 * Open the blockdevice described by device number @dev. @mode and
1690 * @holder are identical to blkdev_get().
1692 * Use it ONLY if you really do not have anything better - i.e. when
1693 * you are behind a truly sucky interface and all you are given is a
1694 * device number. _Never_ to be used for internal purposes. If you
1695 * ever need it - reconsider your API.
1697 * On success, the returned block_device has reference count of one.
1703 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1705 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1707 struct block_device *bdev;
1712 return ERR_PTR(-ENOMEM);
1714 err = blkdev_get(bdev, mode, holder);
1716 return ERR_PTR(err);
1720 EXPORT_SYMBOL(blkdev_get_by_dev);
1722 static int blkdev_open(struct inode * inode, struct file * filp)
1724 struct block_device *bdev;
1727 * Preserve backwards compatibility and allow large file access
1728 * even if userspace doesn't ask for it explicitly. Some mkfs
1729 * binary needs it. We might want to drop this workaround
1730 * during an unstable branch.
1732 filp->f_flags |= O_LARGEFILE;
1734 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
1736 if (filp->f_flags & O_NDELAY)
1737 filp->f_mode |= FMODE_NDELAY;
1738 if (filp->f_flags & O_EXCL)
1739 filp->f_mode |= FMODE_EXCL;
1740 if ((filp->f_flags & O_ACCMODE) == 3)
1741 filp->f_mode |= FMODE_WRITE_IOCTL;
1743 bdev = bd_acquire(inode);
1747 filp->f_mapping = bdev->bd_inode->i_mapping;
1748 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1750 return blkdev_get(bdev, filp->f_mode, filp);
1753 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1755 struct gendisk *disk = bdev->bd_disk;
1756 struct block_device *victim = NULL;
1759 * Sync early if it looks like we're the last one. If someone else
1760 * opens the block device between now and the decrement of bd_openers
1761 * then we did a sync that we didn't need to, but that's not the end
1762 * of the world and we want to avoid long (could be several minute)
1763 * syncs while holding the mutex.
1765 if (bdev->bd_openers == 1)
1766 sync_blockdev(bdev);
1768 mutex_lock_nested(&bdev->bd_mutex, for_part);
1770 bdev->bd_part_count--;
1772 if (!--bdev->bd_openers) {
1773 WARN_ON_ONCE(bdev->bd_holders);
1774 sync_blockdev(bdev);
1777 bdev_write_inode(bdev);
1779 if (bdev->bd_contains == bdev) {
1780 if (disk->fops->release)
1781 disk->fops->release(disk, mode);
1783 if (!bdev->bd_openers) {
1784 disk_put_part(bdev->bd_part);
1785 bdev->bd_part = NULL;
1786 bdev->bd_disk = NULL;
1787 if (bdev != bdev->bd_contains)
1788 victim = bdev->bd_contains;
1789 bdev->bd_contains = NULL;
1791 put_disk_and_module(disk);
1793 mutex_unlock(&bdev->bd_mutex);
1796 __blkdev_put(victim, mode, 1);
1799 void blkdev_put(struct block_device *bdev, fmode_t mode)
1801 mutex_lock(&bdev->bd_mutex);
1803 if (mode & FMODE_EXCL) {
1807 * Release a claim on the device. The holder fields
1808 * are protected with bdev_lock. bd_mutex is to
1809 * synchronize disk_holder unlinking.
1811 spin_lock(&bdev_lock);
1813 WARN_ON_ONCE(--bdev->bd_holders < 0);
1814 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1816 /* bd_contains might point to self, check in a separate step */
1817 if ((bdev_free = !bdev->bd_holders))
1818 bdev->bd_holder = NULL;
1819 if (!bdev->bd_contains->bd_holders)
1820 bdev->bd_contains->bd_holder = NULL;
1822 spin_unlock(&bdev_lock);
1825 * If this was the last claim, remove holder link and
1826 * unblock evpoll if it was a write holder.
1828 if (bdev_free && bdev->bd_write_holder) {
1829 disk_unblock_events(bdev->bd_disk);
1830 bdev->bd_write_holder = false;
1835 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1836 * event. This is to ensure detection of media removal commanded
1837 * from userland - e.g. eject(1).
1839 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1841 mutex_unlock(&bdev->bd_mutex);
1843 __blkdev_put(bdev, mode, 0);
1845 EXPORT_SYMBOL(blkdev_put);
1847 static int blkdev_close(struct inode * inode, struct file * filp)
1849 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1850 blkdev_put(bdev, filp->f_mode);
1854 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1856 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1857 fmode_t mode = file->f_mode;
1860 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1861 * to updated it before every ioctl.
1863 if (file->f_flags & O_NDELAY)
1864 mode |= FMODE_NDELAY;
1866 mode &= ~FMODE_NDELAY;
1868 return blkdev_ioctl(bdev, mode, cmd, arg);
1872 * Write data to the block device. Only intended for the block device itself
1873 * and the raw driver which basically is a fake block device.
1875 * Does not take i_mutex for the write and thus is not for general purpose
1878 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1880 struct file *file = iocb->ki_filp;
1881 struct inode *bd_inode = bdev_file_inode(file);
1882 loff_t size = i_size_read(bd_inode);
1883 struct blk_plug plug;
1886 if (bdev_read_only(I_BDEV(bd_inode)))
1889 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode))
1892 if (!iov_iter_count(from))
1895 if (iocb->ki_pos >= size)
1898 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1901 iov_iter_truncate(from, size - iocb->ki_pos);
1903 blk_start_plug(&plug);
1904 ret = __generic_file_write_iter(iocb, from);
1906 ret = generic_write_sync(iocb, ret);
1907 blk_finish_plug(&plug);
1910 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1912 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1914 struct file *file = iocb->ki_filp;
1915 struct inode *bd_inode = bdev_file_inode(file);
1916 loff_t size = i_size_read(bd_inode);
1917 loff_t pos = iocb->ki_pos;
1923 iov_iter_truncate(to, size);
1924 return generic_file_read_iter(iocb, to);
1926 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1929 * Try to release a page associated with block device when the system
1930 * is under memory pressure.
1932 static int blkdev_releasepage(struct page *page, gfp_t wait)
1934 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1936 if (super && super->s_op->bdev_try_to_free_page)
1937 return super->s_op->bdev_try_to_free_page(super, page, wait);
1939 return try_to_free_buffers(page);
1942 static int blkdev_writepages(struct address_space *mapping,
1943 struct writeback_control *wbc)
1945 return generic_writepages(mapping, wbc);
1948 static const struct address_space_operations def_blk_aops = {
1949 .readpage = blkdev_readpage,
1950 .readahead = blkdev_readahead,
1951 .writepage = blkdev_writepage,
1952 .write_begin = blkdev_write_begin,
1953 .write_end = blkdev_write_end,
1954 .writepages = blkdev_writepages,
1955 .releasepage = blkdev_releasepage,
1956 .direct_IO = blkdev_direct_IO,
1957 .migratepage = buffer_migrate_page_norefs,
1958 .is_dirty_writeback = buffer_check_dirty_writeback,
1961 #define BLKDEV_FALLOC_FL_SUPPORTED \
1962 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1963 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1965 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1968 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1969 loff_t end = start + len - 1;
1973 /* Fail if we don't recognize the flags. */
1974 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1977 /* Don't go off the end of the device. */
1978 isize = i_size_read(bdev->bd_inode);
1982 if (mode & FALLOC_FL_KEEP_SIZE) {
1983 len = isize - start;
1984 end = start + len - 1;
1990 * Don't allow IO that isn't aligned to logical block size.
1992 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1995 /* Invalidate the page cache, including dirty pages. */
1996 error = truncate_bdev_range(bdev, file->f_mode, start, end);
2001 case FALLOC_FL_ZERO_RANGE:
2002 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2003 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2004 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2006 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2007 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2008 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2010 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2011 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2021 * Invalidate again; if someone wandered in and dirtied a page,
2022 * the caller will be given -EBUSY. The third argument is
2023 * inclusive, so the rounding here is safe.
2025 return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
2026 start >> PAGE_SHIFT,
2030 const struct file_operations def_blk_fops = {
2031 .open = blkdev_open,
2032 .release = blkdev_close,
2033 .llseek = block_llseek,
2034 .read_iter = blkdev_read_iter,
2035 .write_iter = blkdev_write_iter,
2036 .iopoll = blkdev_iopoll,
2037 .mmap = generic_file_mmap,
2038 .fsync = blkdev_fsync,
2039 .unlocked_ioctl = block_ioctl,
2040 #ifdef CONFIG_COMPAT
2041 .compat_ioctl = compat_blkdev_ioctl,
2043 .splice_read = generic_file_splice_read,
2044 .splice_write = iter_file_splice_write,
2045 .fallocate = blkdev_fallocate,
2049 * lookup_bdev - lookup a struct block_device by name
2050 * @pathname: special file representing the block device
2052 * Get a reference to the blockdevice at @pathname in the current
2053 * namespace if possible and return it. Return ERR_PTR(error)
2056 struct block_device *lookup_bdev(const char *pathname)
2058 struct block_device *bdev;
2059 struct inode *inode;
2063 if (!pathname || !*pathname)
2064 return ERR_PTR(-EINVAL);
2066 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2068 return ERR_PTR(error);
2070 inode = d_backing_inode(path.dentry);
2072 if (!S_ISBLK(inode->i_mode))
2075 if (!may_open_dev(&path))
2078 bdev = bd_acquire(inode);
2085 bdev = ERR_PTR(error);
2088 EXPORT_SYMBOL(lookup_bdev);
2090 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2092 struct super_block *sb = get_super(bdev);
2097 * no need to lock the super, get_super holds the
2098 * read mutex so the filesystem cannot go away
2099 * under us (->put_super runs with the write lock
2102 shrink_dcache_sb(sb);
2103 res = invalidate_inodes(sb, kill_dirty);
2106 invalidate_bdev(bdev);
2109 EXPORT_SYMBOL(__invalidate_device);
2111 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2113 struct inode *inode, *old_inode = NULL;
2115 spin_lock(&blockdev_superblock->s_inode_list_lock);
2116 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2117 struct address_space *mapping = inode->i_mapping;
2118 struct block_device *bdev;
2120 spin_lock(&inode->i_lock);
2121 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2122 mapping->nrpages == 0) {
2123 spin_unlock(&inode->i_lock);
2127 spin_unlock(&inode->i_lock);
2128 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2130 * We hold a reference to 'inode' so it couldn't have been
2131 * removed from s_inodes list while we dropped the
2132 * s_inode_list_lock We cannot iput the inode now as we can
2133 * be holding the last reference and we cannot iput it under
2134 * s_inode_list_lock. So we keep the reference and iput it
2139 bdev = I_BDEV(inode);
2141 mutex_lock(&bdev->bd_mutex);
2142 if (bdev->bd_openers)
2144 mutex_unlock(&bdev->bd_mutex);
2146 spin_lock(&blockdev_superblock->s_inode_list_lock);
2148 spin_unlock(&blockdev_superblock->s_inode_list_lock);