4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/device_cgroup.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/backing-dev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/magic.h>
21 #include <linux/dax.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/uio.h>
29 #include <linux/namei.h>
30 #include <linux/log2.h>
31 #include <linux/cleancache.h>
32 #include <linux/dax.h>
33 #include <linux/badblocks.h>
34 #include <linux/task_io_accounting_ops.h>
35 #include <linux/falloc.h>
36 #include <linux/uaccess.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 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);
88 EXPORT_SYMBOL(kill_bdev);
90 /* Invalidate clean unused buffers and pagecache. */
91 void invalidate_bdev(struct block_device *bdev)
93 struct address_space *mapping = bdev->bd_inode->i_mapping;
95 if (mapping->nrpages) {
97 lru_add_drain_all(); /* make sure all lru add caches are flushed */
98 invalidate_mapping_pages(mapping, 0, -1);
100 /* 99% of the time, we don't need to flush the cleancache on the bdev.
101 * But, for the strange corners, lets be cautious
103 cleancache_invalidate_inode(mapping);
105 EXPORT_SYMBOL(invalidate_bdev);
107 int set_blocksize(struct block_device *bdev, int size)
109 /* Size must be a power of two, and between 512 and PAGE_SIZE */
110 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
113 /* Size cannot be smaller than the size supported by the device */
114 if (size < bdev_logical_block_size(bdev))
117 /* Don't change the size if it is same as current */
118 if (bdev->bd_block_size != size) {
120 bdev->bd_block_size = size;
121 bdev->bd_inode->i_blkbits = blksize_bits(size);
127 EXPORT_SYMBOL(set_blocksize);
129 int sb_set_blocksize(struct super_block *sb, int size)
131 if (set_blocksize(sb->s_bdev, size))
133 /* If we get here, we know size is power of two
134 * and it's value is between 512 and PAGE_SIZE */
135 sb->s_blocksize = size;
136 sb->s_blocksize_bits = blksize_bits(size);
137 return sb->s_blocksize;
140 EXPORT_SYMBOL(sb_set_blocksize);
142 int sb_min_blocksize(struct super_block *sb, int size)
144 int minsize = bdev_logical_block_size(sb->s_bdev);
147 return sb_set_blocksize(sb, size);
150 EXPORT_SYMBOL(sb_min_blocksize);
153 blkdev_get_block(struct inode *inode, sector_t iblock,
154 struct buffer_head *bh, int create)
156 bh->b_bdev = I_BDEV(inode);
157 bh->b_blocknr = iblock;
158 set_buffer_mapped(bh);
162 static struct inode *bdev_file_inode(struct file *file)
164 return file->f_mapping->host;
167 static unsigned int dio_bio_write_op(struct kiocb *iocb)
169 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
171 /* avoid the need for a I/O completion work item */
172 if (iocb->ki_flags & IOCB_DSYNC)
177 #define DIO_INLINE_BIO_VECS 4
179 static void blkdev_bio_end_io_simple(struct bio *bio)
181 struct task_struct *waiter = bio->bi_private;
183 WRITE_ONCE(bio->bi_private, NULL);
184 blk_wake_io_task(waiter);
188 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
191 struct file *file = iocb->ki_filp;
192 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
193 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec;
194 loff_t pos = iocb->ki_pos;
195 bool should_dirty = false;
201 if ((pos | iov_iter_alignment(iter)) &
202 (bdev_logical_block_size(bdev) - 1))
205 if (nr_pages <= DIO_INLINE_BIO_VECS)
208 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
214 bio_init(&bio, vecs, nr_pages);
215 bio_set_dev(&bio, bdev);
216 bio.bi_iter.bi_sector = pos >> 9;
217 bio.bi_write_hint = iocb->ki_hint;
218 bio.bi_private = current;
219 bio.bi_end_io = blkdev_bio_end_io_simple;
220 bio.bi_ioprio = iocb->ki_ioprio;
222 ret = bio_iov_iter_get_pages(&bio, iter);
225 ret = bio.bi_iter.bi_size;
227 if (iov_iter_rw(iter) == READ) {
228 bio.bi_opf = REQ_OP_READ;
229 if (iter_is_iovec(iter))
232 bio.bi_opf = dio_bio_write_op(iocb);
233 task_io_account_write(ret);
235 if (iocb->ki_flags & IOCB_HIPRI)
236 bio.bi_opf |= REQ_HIPRI;
238 qc = submit_bio(&bio);
240 __set_current_state(TASK_UNINTERRUPTIBLE);
242 if (!READ_ONCE(bio.bi_private))
245 if (!(iocb->ki_flags & IOCB_HIPRI) ||
246 !blk_poll(bdev_get_queue(bdev), qc, true))
249 __set_current_state(TASK_RUNNING);
251 bio_for_each_segment_all(bvec, &bio, i) {
252 if (should_dirty && !PageCompound(bvec->bv_page))
253 set_page_dirty_lock(bvec->bv_page);
254 put_page(bvec->bv_page);
257 if (unlikely(bio.bi_status))
258 ret = blk_status_to_errno(bio.bi_status);
261 if (vecs != inline_vecs)
272 struct task_struct *waiter;
277 bool should_dirty : 1;
282 static struct bio_set blkdev_dio_pool;
284 static void blkdev_bio_end_io(struct bio *bio)
286 struct blkdev_dio *dio = bio->bi_private;
287 bool should_dirty = dio->should_dirty;
289 if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) {
290 if (bio->bi_status && !dio->bio.bi_status)
291 dio->bio.bi_status = bio->bi_status;
294 struct kiocb *iocb = dio->iocb;
297 if (likely(!dio->bio.bi_status)) {
301 ret = blk_status_to_errno(dio->bio.bi_status);
304 dio->iocb->ki_complete(iocb, ret, 0);
308 struct task_struct *waiter = dio->waiter;
310 WRITE_ONCE(dio->waiter, NULL);
311 blk_wake_io_task(waiter);
316 bio_check_pages_dirty(bio);
318 struct bio_vec *bvec;
321 bio_for_each_segment_all(bvec, bio, i)
322 put_page(bvec->bv_page);
328 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
330 struct file *file = iocb->ki_filp;
331 struct inode *inode = bdev_file_inode(file);
332 struct block_device *bdev = I_BDEV(inode);
333 struct blk_plug plug;
334 struct blkdev_dio *dio;
336 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
337 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
338 loff_t pos = iocb->ki_pos;
339 blk_qc_t qc = BLK_QC_T_NONE;
342 if ((pos | iov_iter_alignment(iter)) &
343 (bdev_logical_block_size(bdev) - 1))
346 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
348 dio = container_of(bio, struct blkdev_dio, bio);
349 dio->is_sync = is_sync = is_sync_kiocb(iocb);
351 dio->waiter = current;
358 dio->multi_bio = false;
359 dio->should_dirty = is_read && iter_is_iovec(iter);
362 * Don't plug for HIPRI/polled IO, as those should go straight
366 blk_start_plug(&plug);
369 bio_set_dev(bio, bdev);
370 bio->bi_iter.bi_sector = pos >> 9;
371 bio->bi_write_hint = iocb->ki_hint;
372 bio->bi_private = dio;
373 bio->bi_end_io = blkdev_bio_end_io;
374 bio->bi_ioprio = iocb->ki_ioprio;
376 ret = bio_iov_iter_get_pages(bio, iter);
378 bio->bi_status = BLK_STS_IOERR;
384 bio->bi_opf = REQ_OP_READ;
385 if (dio->should_dirty)
386 bio_set_pages_dirty(bio);
388 bio->bi_opf = dio_bio_write_op(iocb);
389 task_io_account_write(bio->bi_iter.bi_size);
392 dio->size += bio->bi_iter.bi_size;
393 pos += bio->bi_iter.bi_size;
395 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
397 if (iocb->ki_flags & IOCB_HIPRI)
398 bio->bi_opf |= REQ_HIPRI;
400 qc = submit_bio(bio);
404 if (!dio->multi_bio) {
406 * AIO needs an extra reference to ensure the dio
407 * structure which is embedded into the first bio
412 dio->multi_bio = true;
413 atomic_set(&dio->ref, 2);
415 atomic_inc(&dio->ref);
419 bio = bio_alloc(GFP_KERNEL, nr_pages);
423 blk_finish_plug(&plug);
429 __set_current_state(TASK_UNINTERRUPTIBLE);
431 if (!READ_ONCE(dio->waiter))
434 if (!(iocb->ki_flags & IOCB_HIPRI) ||
435 !blk_poll(bdev_get_queue(bdev), qc, true))
438 __set_current_state(TASK_RUNNING);
441 ret = blk_status_to_errno(dio->bio.bi_status);
450 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
454 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
457 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
458 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
460 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
463 static __init int blkdev_init(void)
465 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
467 module_init(blkdev_init);
469 int __sync_blockdev(struct block_device *bdev, int wait)
474 return filemap_flush(bdev->bd_inode->i_mapping);
475 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
479 * Write out and wait upon all the dirty data associated with a block
480 * device via its mapping. Does not take the superblock lock.
482 int sync_blockdev(struct block_device *bdev)
484 return __sync_blockdev(bdev, 1);
486 EXPORT_SYMBOL(sync_blockdev);
489 * Write out and wait upon all dirty data associated with this
490 * device. Filesystem data as well as the underlying block
491 * device. Takes the superblock lock.
493 int fsync_bdev(struct block_device *bdev)
495 struct super_block *sb = get_super(bdev);
497 int res = sync_filesystem(sb);
501 return sync_blockdev(bdev);
503 EXPORT_SYMBOL(fsync_bdev);
506 * freeze_bdev -- lock a filesystem and force it into a consistent state
507 * @bdev: blockdevice to lock
509 * If a superblock is found on this device, we take the s_umount semaphore
510 * on it to make sure nobody unmounts until the snapshot creation is done.
511 * The reference counter (bd_fsfreeze_count) guarantees that only the last
512 * unfreeze process can unfreeze the frozen filesystem actually when multiple
513 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
514 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
517 struct super_block *freeze_bdev(struct block_device *bdev)
519 struct super_block *sb;
522 mutex_lock(&bdev->bd_fsfreeze_mutex);
523 if (++bdev->bd_fsfreeze_count > 1) {
525 * We don't even need to grab a reference - the first call
526 * to freeze_bdev grab an active reference and only the last
527 * thaw_bdev drops it.
529 sb = get_super(bdev);
532 mutex_unlock(&bdev->bd_fsfreeze_mutex);
536 sb = get_active_super(bdev);
539 if (sb->s_op->freeze_super)
540 error = sb->s_op->freeze_super(sb);
542 error = freeze_super(sb);
544 deactivate_super(sb);
545 bdev->bd_fsfreeze_count--;
546 mutex_unlock(&bdev->bd_fsfreeze_mutex);
547 return ERR_PTR(error);
549 deactivate_super(sb);
552 mutex_unlock(&bdev->bd_fsfreeze_mutex);
553 return sb; /* thaw_bdev releases s->s_umount */
555 EXPORT_SYMBOL(freeze_bdev);
558 * thaw_bdev -- unlock filesystem
559 * @bdev: blockdevice to unlock
560 * @sb: associated superblock
562 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
564 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
568 mutex_lock(&bdev->bd_fsfreeze_mutex);
569 if (!bdev->bd_fsfreeze_count)
573 if (--bdev->bd_fsfreeze_count > 0)
579 if (sb->s_op->thaw_super)
580 error = sb->s_op->thaw_super(sb);
582 error = thaw_super(sb);
584 bdev->bd_fsfreeze_count++;
586 mutex_unlock(&bdev->bd_fsfreeze_mutex);
589 EXPORT_SYMBOL(thaw_bdev);
591 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
593 return block_write_full_page(page, blkdev_get_block, wbc);
596 static int blkdev_readpage(struct file * file, struct page * page)
598 return block_read_full_page(page, blkdev_get_block);
601 static int blkdev_readpages(struct file *file, struct address_space *mapping,
602 struct list_head *pages, unsigned nr_pages)
604 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
607 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
608 loff_t pos, unsigned len, unsigned flags,
609 struct page **pagep, void **fsdata)
611 return block_write_begin(mapping, pos, len, flags, pagep,
615 static int blkdev_write_end(struct file *file, struct address_space *mapping,
616 loff_t pos, unsigned len, unsigned copied,
617 struct page *page, void *fsdata)
620 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
630 * for a block special file file_inode(file)->i_size is zero
631 * so we compute the size by hand (just as in block_read/write above)
633 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
635 struct inode *bd_inode = bdev_file_inode(file);
638 inode_lock(bd_inode);
639 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
640 inode_unlock(bd_inode);
644 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
646 struct inode *bd_inode = bdev_file_inode(filp);
647 struct block_device *bdev = I_BDEV(bd_inode);
650 error = file_write_and_wait_range(filp, start, end);
655 * There is no need to serialise calls to blkdev_issue_flush with
656 * i_mutex and doing so causes performance issues with concurrent
657 * O_SYNC writers to a block device.
659 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
660 if (error == -EOPNOTSUPP)
665 EXPORT_SYMBOL(blkdev_fsync);
668 * bdev_read_page() - Start reading a page from a block device
669 * @bdev: The device to read the page from
670 * @sector: The offset on the device to read the page to (need not be aligned)
671 * @page: The page to read
673 * On entry, the page should be locked. It will be unlocked when the page
674 * has been read. If the block driver implements rw_page synchronously,
675 * that will be true on exit from this function, but it need not be.
677 * Errors returned by this function are usually "soft", eg out of memory, or
678 * queue full; callers should try a different route to read this page rather
679 * than propagate an error back up the stack.
681 * Return: negative errno if an error occurs, 0 if submission was successful.
683 int bdev_read_page(struct block_device *bdev, sector_t sector,
686 const struct block_device_operations *ops = bdev->bd_disk->fops;
687 int result = -EOPNOTSUPP;
689 if (!ops->rw_page || bdev_get_integrity(bdev))
692 result = blk_queue_enter(bdev->bd_queue, 0);
695 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
697 blk_queue_exit(bdev->bd_queue);
700 EXPORT_SYMBOL_GPL(bdev_read_page);
703 * bdev_write_page() - Start writing a page to a block device
704 * @bdev: The device to write the page to
705 * @sector: The offset on the device to write the page to (need not be aligned)
706 * @page: The page to write
707 * @wbc: The writeback_control for the write
709 * On entry, the page should be locked and not currently under writeback.
710 * On exit, if the write started successfully, the page will be unlocked and
711 * under writeback. If the write failed already (eg the driver failed to
712 * queue the page to the device), the page will still be locked. If the
713 * caller is a ->writepage implementation, it will need to unlock the page.
715 * Errors returned by this function are usually "soft", eg out of memory, or
716 * queue full; callers should try a different route to write this page rather
717 * than propagate an error back up the stack.
719 * Return: negative errno if an error occurs, 0 if submission was successful.
721 int bdev_write_page(struct block_device *bdev, sector_t sector,
722 struct page *page, struct writeback_control *wbc)
725 const struct block_device_operations *ops = bdev->bd_disk->fops;
727 if (!ops->rw_page || bdev_get_integrity(bdev))
729 result = blk_queue_enter(bdev->bd_queue, 0);
733 set_page_writeback(page);
734 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
737 end_page_writeback(page);
739 clean_page_buffers(page);
742 blk_queue_exit(bdev->bd_queue);
745 EXPORT_SYMBOL_GPL(bdev_write_page);
751 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
752 static struct kmem_cache * bdev_cachep __read_mostly;
754 static struct inode *bdev_alloc_inode(struct super_block *sb)
756 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
759 return &ei->vfs_inode;
762 static void bdev_i_callback(struct rcu_head *head)
764 struct inode *inode = container_of(head, struct inode, i_rcu);
765 struct bdev_inode *bdi = BDEV_I(inode);
767 kmem_cache_free(bdev_cachep, bdi);
770 static void bdev_destroy_inode(struct inode *inode)
772 call_rcu(&inode->i_rcu, bdev_i_callback);
775 static void init_once(void *foo)
777 struct bdev_inode *ei = (struct bdev_inode *) foo;
778 struct block_device *bdev = &ei->bdev;
780 memset(bdev, 0, sizeof(*bdev));
781 mutex_init(&bdev->bd_mutex);
782 INIT_LIST_HEAD(&bdev->bd_list);
784 INIT_LIST_HEAD(&bdev->bd_holder_disks);
786 bdev->bd_bdi = &noop_backing_dev_info;
787 inode_init_once(&ei->vfs_inode);
788 /* Initialize mutex for freeze. */
789 mutex_init(&bdev->bd_fsfreeze_mutex);
792 static void bdev_evict_inode(struct inode *inode)
794 struct block_device *bdev = &BDEV_I(inode)->bdev;
795 truncate_inode_pages_final(&inode->i_data);
796 invalidate_inode_buffers(inode); /* is it needed here? */
798 spin_lock(&bdev_lock);
799 list_del_init(&bdev->bd_list);
800 spin_unlock(&bdev_lock);
801 /* Detach inode from wb early as bdi_put() may free bdi->wb */
802 inode_detach_wb(inode);
803 if (bdev->bd_bdi != &noop_backing_dev_info) {
804 bdi_put(bdev->bd_bdi);
805 bdev->bd_bdi = &noop_backing_dev_info;
809 static const struct super_operations bdev_sops = {
810 .statfs = simple_statfs,
811 .alloc_inode = bdev_alloc_inode,
812 .destroy_inode = bdev_destroy_inode,
813 .drop_inode = generic_delete_inode,
814 .evict_inode = bdev_evict_inode,
817 static struct dentry *bd_mount(struct file_system_type *fs_type,
818 int flags, const char *dev_name, void *data)
821 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
823 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
827 static struct file_system_type bd_type = {
830 .kill_sb = kill_anon_super,
833 struct super_block *blockdev_superblock __read_mostly;
834 EXPORT_SYMBOL_GPL(blockdev_superblock);
836 void __init bdev_cache_init(void)
839 static struct vfsmount *bd_mnt;
841 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
842 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
843 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
845 err = register_filesystem(&bd_type);
847 panic("Cannot register bdev pseudo-fs");
848 bd_mnt = kern_mount(&bd_type);
850 panic("Cannot create bdev pseudo-fs");
851 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
855 * Most likely _very_ bad one - but then it's hardly critical for small
856 * /dev and can be fixed when somebody will need really large one.
857 * Keep in mind that it will be fed through icache hash function too.
859 static inline unsigned long hash(dev_t dev)
861 return MAJOR(dev)+MINOR(dev);
864 static int bdev_test(struct inode *inode, void *data)
866 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
869 static int bdev_set(struct inode *inode, void *data)
871 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
875 static LIST_HEAD(all_bdevs);
878 * If there is a bdev inode for this device, unhash it so that it gets evicted
879 * as soon as last inode reference is dropped.
881 void bdev_unhash_inode(dev_t dev)
885 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
887 remove_inode_hash(inode);
892 struct block_device *bdget(dev_t dev)
894 struct block_device *bdev;
897 inode = iget5_locked(blockdev_superblock, hash(dev),
898 bdev_test, bdev_set, &dev);
903 bdev = &BDEV_I(inode)->bdev;
905 if (inode->i_state & I_NEW) {
906 bdev->bd_contains = NULL;
907 bdev->bd_super = NULL;
908 bdev->bd_inode = inode;
909 bdev->bd_block_size = i_blocksize(inode);
910 bdev->bd_part_count = 0;
911 bdev->bd_invalidated = 0;
912 inode->i_mode = S_IFBLK;
914 inode->i_bdev = bdev;
915 inode->i_data.a_ops = &def_blk_aops;
916 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
917 spin_lock(&bdev_lock);
918 list_add(&bdev->bd_list, &all_bdevs);
919 spin_unlock(&bdev_lock);
920 unlock_new_inode(inode);
925 EXPORT_SYMBOL(bdget);
928 * bdgrab -- Grab a reference to an already referenced block device
929 * @bdev: Block device to grab a reference to.
931 struct block_device *bdgrab(struct block_device *bdev)
933 ihold(bdev->bd_inode);
936 EXPORT_SYMBOL(bdgrab);
938 long nr_blockdev_pages(void)
940 struct block_device *bdev;
942 spin_lock(&bdev_lock);
943 list_for_each_entry(bdev, &all_bdevs, bd_list) {
944 ret += bdev->bd_inode->i_mapping->nrpages;
946 spin_unlock(&bdev_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 - prepare to 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 * Prepare to claim @bdev. This function fails if @bdev is already
1054 * claimed by another holder and waits if another claiming is in
1055 * progress. This function doesn't actually claim. On successful
1056 * return, the caller has ownership of bd_claiming and bd_holder[s].
1059 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1060 * it multiple times.
1063 * 0 if @bdev can be claimed, -EBUSY otherwise.
1065 static int bd_prepare_to_claim(struct block_device *bdev,
1066 struct block_device *whole, void *holder)
1069 /* if someone else claimed, fail */
1070 if (!bd_may_claim(bdev, whole, holder))
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);
1082 spin_lock(&bdev_lock);
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);
1112 * bd_start_claiming - start claiming a block device
1113 * @bdev: block device of interest
1114 * @holder: holder trying to claim @bdev
1116 * @bdev is about to be opened exclusively. Check @bdev can be opened
1117 * exclusively and mark that an exclusive open is in progress. Each
1118 * successful call to this function must be matched with a call to
1119 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1122 * This function is used to gain exclusive access to the block device
1123 * without actually causing other exclusive open attempts to fail. It
1124 * should be used when the open sequence itself requires exclusive
1125 * access but may subsequently fail.
1131 * Pointer to the block device containing @bdev on success, ERR_PTR()
1134 static struct block_device *bd_start_claiming(struct block_device *bdev,
1137 struct gendisk *disk;
1138 struct block_device *whole;
1144 * @bdev might not have been initialized properly yet, look up
1145 * and grab the outer block device the hard way.
1147 disk = bdev_get_gendisk(bdev, &partno);
1149 return ERR_PTR(-ENXIO);
1152 * Normally, @bdev should equal what's returned from bdget_disk()
1153 * if partno is 0; however, some drivers (floppy) use multiple
1154 * bdev's for the same physical device and @bdev may be one of the
1155 * aliases. Keep @bdev if partno is 0. This means claimer
1156 * tracking is broken for those devices but it has always been that
1160 whole = bdget_disk(disk, 0);
1162 whole = bdgrab(bdev);
1164 put_disk_and_module(disk);
1166 return ERR_PTR(-ENOMEM);
1168 /* prepare to claim, if successful, mark claiming in progress */
1169 spin_lock(&bdev_lock);
1171 err = bd_prepare_to_claim(bdev, whole, holder);
1173 whole->bd_claiming = holder;
1174 spin_unlock(&bdev_lock);
1177 spin_unlock(&bdev_lock);
1179 return ERR_PTR(err);
1184 struct bd_holder_disk {
1185 struct list_head list;
1186 struct gendisk *disk;
1190 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1191 struct gendisk *disk)
1193 struct bd_holder_disk *holder;
1195 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1196 if (holder->disk == disk)
1201 static int add_symlink(struct kobject *from, struct kobject *to)
1203 return sysfs_create_link(from, to, kobject_name(to));
1206 static void del_symlink(struct kobject *from, struct kobject *to)
1208 sysfs_remove_link(from, kobject_name(to));
1212 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1213 * @bdev: the claimed slave bdev
1214 * @disk: the holding disk
1216 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1218 * This functions creates the following sysfs symlinks.
1220 * - from "slaves" directory of the holder @disk to the claimed @bdev
1221 * - from "holders" directory of the @bdev to the holder @disk
1223 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1224 * passed to bd_link_disk_holder(), then:
1226 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1227 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1229 * The caller must have claimed @bdev before calling this function and
1230 * ensure that both @bdev and @disk are valid during the creation and
1231 * lifetime of these symlinks.
1237 * 0 on success, -errno on failure.
1239 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1241 struct bd_holder_disk *holder;
1244 mutex_lock(&bdev->bd_mutex);
1246 WARN_ON_ONCE(!bdev->bd_holder);
1248 /* FIXME: remove the following once add_disk() handles errors */
1249 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1252 holder = bd_find_holder_disk(bdev, disk);
1258 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1264 INIT_LIST_HEAD(&holder->list);
1265 holder->disk = disk;
1268 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1272 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1276 * bdev could be deleted beneath us which would implicitly destroy
1277 * the holder directory. Hold on to it.
1279 kobject_get(bdev->bd_part->holder_dir);
1281 list_add(&holder->list, &bdev->bd_holder_disks);
1285 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1289 mutex_unlock(&bdev->bd_mutex);
1292 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1295 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1296 * @bdev: the calimed slave bdev
1297 * @disk: the holding disk
1299 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1304 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1306 struct bd_holder_disk *holder;
1308 mutex_lock(&bdev->bd_mutex);
1310 holder = bd_find_holder_disk(bdev, disk);
1312 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1313 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1314 del_symlink(bdev->bd_part->holder_dir,
1315 &disk_to_dev(disk)->kobj);
1316 kobject_put(bdev->bd_part->holder_dir);
1317 list_del_init(&holder->list);
1321 mutex_unlock(&bdev->bd_mutex);
1323 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1327 * flush_disk - invalidates all buffer-cache entries on a disk
1329 * @bdev: struct block device to be flushed
1330 * @kill_dirty: flag to guide handling of dirty inodes
1332 * Invalidates all buffer-cache entries on a disk. It should be called
1333 * when a disk has been changed -- either by a media change or online
1336 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1338 if (__invalidate_device(bdev, kill_dirty)) {
1339 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1340 "resized disk %s\n",
1341 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1346 if (disk_part_scan_enabled(bdev->bd_disk))
1347 bdev->bd_invalidated = 1;
1351 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1352 * @disk: struct gendisk to check
1353 * @bdev: struct bdev to adjust.
1354 * @verbose: if %true log a message about a size change if there is any
1356 * This routine checks to see if the bdev size does not match the disk size
1357 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1360 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev,
1363 loff_t disk_size, bdev_size;
1365 disk_size = (loff_t)get_capacity(disk) << 9;
1366 bdev_size = i_size_read(bdev->bd_inode);
1367 if (disk_size != bdev_size) {
1370 "%s: detected capacity change from %lld to %lld\n",
1371 disk->disk_name, bdev_size, disk_size);
1373 i_size_write(bdev->bd_inode, disk_size);
1374 if (bdev_size > disk_size)
1375 flush_disk(bdev, false);
1380 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1381 * @disk: struct gendisk to be revalidated
1383 * This routine is a wrapper for lower-level driver's revalidate_disk
1384 * call-backs. It is used to do common pre and post operations needed
1385 * for all revalidate_disk operations.
1387 int revalidate_disk(struct gendisk *disk)
1389 struct block_device *bdev;
1392 if (disk->fops->revalidate_disk)
1393 ret = disk->fops->revalidate_disk(disk);
1394 bdev = bdget_disk(disk, 0);
1398 mutex_lock(&bdev->bd_mutex);
1399 check_disk_size_change(disk, bdev, ret == 0);
1400 bdev->bd_invalidated = 0;
1401 mutex_unlock(&bdev->bd_mutex);
1405 EXPORT_SYMBOL(revalidate_disk);
1408 * This routine checks whether a removable media has been changed,
1409 * and invalidates all buffer-cache-entries in that case. This
1410 * is a relatively slow routine, so we have to try to minimize using
1411 * it. Thus it is called only upon a 'mount' or 'open'. This
1412 * is the best way of combining speed and utility, I think.
1413 * People changing diskettes in the middle of an operation deserve
1416 int check_disk_change(struct block_device *bdev)
1418 struct gendisk *disk = bdev->bd_disk;
1419 const struct block_device_operations *bdops = disk->fops;
1420 unsigned int events;
1422 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1423 DISK_EVENT_EJECT_REQUEST);
1424 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1427 flush_disk(bdev, true);
1428 if (bdops->revalidate_disk)
1429 bdops->revalidate_disk(bdev->bd_disk);
1433 EXPORT_SYMBOL(check_disk_change);
1435 void bd_set_size(struct block_device *bdev, loff_t size)
1437 unsigned bsize = bdev_logical_block_size(bdev);
1439 inode_lock(bdev->bd_inode);
1440 i_size_write(bdev->bd_inode, size);
1441 inode_unlock(bdev->bd_inode);
1442 while (bsize < PAGE_SIZE) {
1447 bdev->bd_block_size = bsize;
1448 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1450 EXPORT_SYMBOL(bd_set_size);
1452 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1457 * mutex_lock(part->bd_mutex)
1458 * mutex_lock_nested(whole->bd_mutex, 1)
1461 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1463 struct gendisk *disk;
1467 bool first_open = false;
1469 if (mode & FMODE_READ)
1471 if (mode & FMODE_WRITE)
1474 * hooks: /n/, see "layering violations".
1477 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1487 disk = bdev_get_gendisk(bdev, &partno);
1491 disk_block_events(disk);
1492 mutex_lock_nested(&bdev->bd_mutex, for_part);
1493 if (!bdev->bd_openers) {
1495 bdev->bd_disk = disk;
1496 bdev->bd_queue = disk->queue;
1497 bdev->bd_contains = bdev;
1498 bdev->bd_partno = partno;
1502 bdev->bd_part = disk_get_part(disk, partno);
1507 if (disk->fops->open) {
1508 ret = disk->fops->open(bdev, mode);
1509 if (ret == -ERESTARTSYS) {
1510 /* Lost a race with 'disk' being
1511 * deleted, try again.
1514 disk_put_part(bdev->bd_part);
1515 bdev->bd_part = NULL;
1516 bdev->bd_disk = NULL;
1517 bdev->bd_queue = NULL;
1518 mutex_unlock(&bdev->bd_mutex);
1519 disk_unblock_events(disk);
1520 put_disk_and_module(disk);
1526 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1529 * If the device is invalidated, rescan partition
1530 * if open succeeded or failed with -ENOMEDIUM.
1531 * The latter is necessary to prevent ghost
1532 * partitions on a removed medium.
1534 if (bdev->bd_invalidated) {
1536 rescan_partitions(disk, bdev);
1537 else if (ret == -ENOMEDIUM)
1538 invalidate_partitions(disk, bdev);
1544 struct block_device *whole;
1545 whole = bdget_disk(disk, 0);
1550 ret = __blkdev_get(whole, mode, 1);
1553 bdev->bd_contains = whole;
1554 bdev->bd_part = disk_get_part(disk, partno);
1555 if (!(disk->flags & GENHD_FL_UP) ||
1556 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1560 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1563 if (bdev->bd_bdi == &noop_backing_dev_info)
1564 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1566 if (bdev->bd_contains == bdev) {
1568 if (bdev->bd_disk->fops->open)
1569 ret = bdev->bd_disk->fops->open(bdev, mode);
1570 /* the same as first opener case, read comment there */
1571 if (bdev->bd_invalidated) {
1573 rescan_partitions(bdev->bd_disk, bdev);
1574 else if (ret == -ENOMEDIUM)
1575 invalidate_partitions(bdev->bd_disk, bdev);
1578 goto out_unlock_bdev;
1583 bdev->bd_part_count++;
1584 mutex_unlock(&bdev->bd_mutex);
1585 disk_unblock_events(disk);
1586 /* only one opener holds refs to the module and disk */
1588 put_disk_and_module(disk);
1592 disk_put_part(bdev->bd_part);
1593 bdev->bd_disk = NULL;
1594 bdev->bd_part = NULL;
1595 bdev->bd_queue = NULL;
1596 if (bdev != bdev->bd_contains)
1597 __blkdev_put(bdev->bd_contains, mode, 1);
1598 bdev->bd_contains = NULL;
1600 mutex_unlock(&bdev->bd_mutex);
1601 disk_unblock_events(disk);
1602 put_disk_and_module(disk);
1610 * blkdev_get - open a block device
1611 * @bdev: block_device to open
1612 * @mode: FMODE_* mask
1613 * @holder: exclusive holder identifier
1615 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1616 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1617 * @holder is invalid. Exclusive opens may nest for the same @holder.
1619 * On success, the reference count of @bdev is unchanged. On failure,
1626 * 0 on success, -errno on failure.
1628 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1630 struct block_device *whole = NULL;
1633 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1635 if ((mode & FMODE_EXCL) && holder) {
1636 whole = bd_start_claiming(bdev, holder);
1637 if (IS_ERR(whole)) {
1639 return PTR_ERR(whole);
1643 res = __blkdev_get(bdev, mode, 0);
1646 struct gendisk *disk = whole->bd_disk;
1648 /* finish claiming */
1649 mutex_lock(&bdev->bd_mutex);
1650 spin_lock(&bdev_lock);
1653 BUG_ON(!bd_may_claim(bdev, whole, holder));
1655 * Note that for a whole device bd_holders
1656 * will be incremented twice, and bd_holder
1657 * will be set to bd_may_claim before being
1660 whole->bd_holders++;
1661 whole->bd_holder = bd_may_claim;
1663 bdev->bd_holder = holder;
1666 /* tell others that we're done */
1667 BUG_ON(whole->bd_claiming != holder);
1668 whole->bd_claiming = NULL;
1669 wake_up_bit(&whole->bd_claiming, 0);
1671 spin_unlock(&bdev_lock);
1674 * Block event polling for write claims if requested. Any
1675 * write holder makes the write_holder state stick until
1676 * all are released. This is good enough and tracking
1677 * individual writeable reference is too fragile given the
1678 * way @mode is used in blkdev_get/put().
1680 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1681 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1682 bdev->bd_write_holder = true;
1683 disk_block_events(disk);
1686 mutex_unlock(&bdev->bd_mutex);
1692 EXPORT_SYMBOL(blkdev_get);
1695 * blkdev_get_by_path - open a block device by name
1696 * @path: path to the block device to open
1697 * @mode: FMODE_* mask
1698 * @holder: exclusive holder identifier
1700 * Open the blockdevice described by the device file at @path. @mode
1701 * and @holder are identical to blkdev_get().
1703 * On success, the returned block_device has reference count of one.
1709 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1711 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1714 struct block_device *bdev;
1717 bdev = lookup_bdev(path);
1721 err = blkdev_get(bdev, mode, holder);
1723 return ERR_PTR(err);
1725 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1726 blkdev_put(bdev, mode);
1727 return ERR_PTR(-EACCES);
1732 EXPORT_SYMBOL(blkdev_get_by_path);
1735 * blkdev_get_by_dev - open a block device by device number
1736 * @dev: device number of block device to open
1737 * @mode: FMODE_* mask
1738 * @holder: exclusive holder identifier
1740 * Open the blockdevice described by device number @dev. @mode and
1741 * @holder are identical to blkdev_get().
1743 * Use it ONLY if you really do not have anything better - i.e. when
1744 * you are behind a truly sucky interface and all you are given is a
1745 * device number. _Never_ to be used for internal purposes. If you
1746 * ever need it - reconsider your API.
1748 * On success, the returned block_device has reference count of one.
1754 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1756 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1758 struct block_device *bdev;
1763 return ERR_PTR(-ENOMEM);
1765 err = blkdev_get(bdev, mode, holder);
1767 return ERR_PTR(err);
1771 EXPORT_SYMBOL(blkdev_get_by_dev);
1773 static int blkdev_open(struct inode * inode, struct file * filp)
1775 struct block_device *bdev;
1778 * Preserve backwards compatibility and allow large file access
1779 * even if userspace doesn't ask for it explicitly. Some mkfs
1780 * binary needs it. We might want to drop this workaround
1781 * during an unstable branch.
1783 filp->f_flags |= O_LARGEFILE;
1785 filp->f_mode |= FMODE_NOWAIT;
1787 if (filp->f_flags & O_NDELAY)
1788 filp->f_mode |= FMODE_NDELAY;
1789 if (filp->f_flags & O_EXCL)
1790 filp->f_mode |= FMODE_EXCL;
1791 if ((filp->f_flags & O_ACCMODE) == 3)
1792 filp->f_mode |= FMODE_WRITE_IOCTL;
1794 bdev = bd_acquire(inode);
1798 filp->f_mapping = bdev->bd_inode->i_mapping;
1799 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1801 return blkdev_get(bdev, filp->f_mode, filp);
1804 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1806 struct gendisk *disk = bdev->bd_disk;
1807 struct block_device *victim = NULL;
1809 mutex_lock_nested(&bdev->bd_mutex, for_part);
1811 bdev->bd_part_count--;
1813 if (!--bdev->bd_openers) {
1814 WARN_ON_ONCE(bdev->bd_holders);
1815 sync_blockdev(bdev);
1818 bdev_write_inode(bdev);
1820 if (bdev->bd_contains == bdev) {
1821 if (disk->fops->release)
1822 disk->fops->release(disk, mode);
1824 if (!bdev->bd_openers) {
1825 disk_put_part(bdev->bd_part);
1826 bdev->bd_part = NULL;
1827 bdev->bd_disk = NULL;
1828 if (bdev != bdev->bd_contains)
1829 victim = bdev->bd_contains;
1830 bdev->bd_contains = NULL;
1832 put_disk_and_module(disk);
1834 mutex_unlock(&bdev->bd_mutex);
1837 __blkdev_put(victim, mode, 1);
1840 void blkdev_put(struct block_device *bdev, fmode_t mode)
1842 mutex_lock(&bdev->bd_mutex);
1844 if (mode & FMODE_EXCL) {
1848 * Release a claim on the device. The holder fields
1849 * are protected with bdev_lock. bd_mutex is to
1850 * synchronize disk_holder unlinking.
1852 spin_lock(&bdev_lock);
1854 WARN_ON_ONCE(--bdev->bd_holders < 0);
1855 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1857 /* bd_contains might point to self, check in a separate step */
1858 if ((bdev_free = !bdev->bd_holders))
1859 bdev->bd_holder = NULL;
1860 if (!bdev->bd_contains->bd_holders)
1861 bdev->bd_contains->bd_holder = NULL;
1863 spin_unlock(&bdev_lock);
1866 * If this was the last claim, remove holder link and
1867 * unblock evpoll if it was a write holder.
1869 if (bdev_free && bdev->bd_write_holder) {
1870 disk_unblock_events(bdev->bd_disk);
1871 bdev->bd_write_holder = false;
1876 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1877 * event. This is to ensure detection of media removal commanded
1878 * from userland - e.g. eject(1).
1880 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1882 mutex_unlock(&bdev->bd_mutex);
1884 __blkdev_put(bdev, mode, 0);
1886 EXPORT_SYMBOL(blkdev_put);
1888 static int blkdev_close(struct inode * inode, struct file * filp)
1890 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1891 blkdev_put(bdev, filp->f_mode);
1895 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1897 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1898 fmode_t mode = file->f_mode;
1901 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1902 * to updated it before every ioctl.
1904 if (file->f_flags & O_NDELAY)
1905 mode |= FMODE_NDELAY;
1907 mode &= ~FMODE_NDELAY;
1909 return blkdev_ioctl(bdev, mode, cmd, arg);
1913 * Write data to the block device. Only intended for the block device itself
1914 * and the raw driver which basically is a fake block device.
1916 * Does not take i_mutex for the write and thus is not for general purpose
1919 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1921 struct file *file = iocb->ki_filp;
1922 struct inode *bd_inode = bdev_file_inode(file);
1923 loff_t size = i_size_read(bd_inode);
1924 struct blk_plug plug;
1927 if (bdev_read_only(I_BDEV(bd_inode)))
1930 if (!iov_iter_count(from))
1933 if (iocb->ki_pos >= size)
1936 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1939 iov_iter_truncate(from, size - iocb->ki_pos);
1941 blk_start_plug(&plug);
1942 ret = __generic_file_write_iter(iocb, from);
1944 ret = generic_write_sync(iocb, ret);
1945 blk_finish_plug(&plug);
1948 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1950 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1952 struct file *file = iocb->ki_filp;
1953 struct inode *bd_inode = bdev_file_inode(file);
1954 loff_t size = i_size_read(bd_inode);
1955 loff_t pos = iocb->ki_pos;
1961 iov_iter_truncate(to, size);
1962 return generic_file_read_iter(iocb, to);
1964 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1967 * Try to release a page associated with block device when the system
1968 * is under memory pressure.
1970 static int blkdev_releasepage(struct page *page, gfp_t wait)
1972 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1974 if (super && super->s_op->bdev_try_to_free_page)
1975 return super->s_op->bdev_try_to_free_page(super, page, wait);
1977 return try_to_free_buffers(page);
1980 static int blkdev_writepages(struct address_space *mapping,
1981 struct writeback_control *wbc)
1983 return generic_writepages(mapping, wbc);
1986 static const struct address_space_operations def_blk_aops = {
1987 .readpage = blkdev_readpage,
1988 .readpages = blkdev_readpages,
1989 .writepage = blkdev_writepage,
1990 .write_begin = blkdev_write_begin,
1991 .write_end = blkdev_write_end,
1992 .writepages = blkdev_writepages,
1993 .releasepage = blkdev_releasepage,
1994 .direct_IO = blkdev_direct_IO,
1995 .is_dirty_writeback = buffer_check_dirty_writeback,
1998 #define BLKDEV_FALLOC_FL_SUPPORTED \
1999 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2000 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2002 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2005 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2006 struct address_space *mapping;
2007 loff_t end = start + len - 1;
2011 /* Fail if we don't recognize the flags. */
2012 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2015 /* Don't go off the end of the device. */
2016 isize = i_size_read(bdev->bd_inode);
2020 if (mode & FALLOC_FL_KEEP_SIZE) {
2021 len = isize - start;
2022 end = start + len - 1;
2028 * Don't allow IO that isn't aligned to logical block size.
2030 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2033 /* Invalidate the page cache, including dirty pages. */
2034 mapping = bdev->bd_inode->i_mapping;
2035 truncate_inode_pages_range(mapping, start, end);
2038 case FALLOC_FL_ZERO_RANGE:
2039 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2040 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2041 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2043 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2044 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2045 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2047 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2048 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2058 * Invalidate again; if someone wandered in and dirtied a page,
2059 * the caller will be given -EBUSY. The third argument is
2060 * inclusive, so the rounding here is safe.
2062 return invalidate_inode_pages2_range(mapping,
2063 start >> PAGE_SHIFT,
2067 const struct file_operations def_blk_fops = {
2068 .open = blkdev_open,
2069 .release = blkdev_close,
2070 .llseek = block_llseek,
2071 .read_iter = blkdev_read_iter,
2072 .write_iter = blkdev_write_iter,
2073 .mmap = generic_file_mmap,
2074 .fsync = blkdev_fsync,
2075 .unlocked_ioctl = block_ioctl,
2076 #ifdef CONFIG_COMPAT
2077 .compat_ioctl = compat_blkdev_ioctl,
2079 .splice_read = generic_file_splice_read,
2080 .splice_write = iter_file_splice_write,
2081 .fallocate = blkdev_fallocate,
2084 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2087 mm_segment_t old_fs = get_fs();
2089 res = blkdev_ioctl(bdev, 0, cmd, arg);
2094 EXPORT_SYMBOL(ioctl_by_bdev);
2097 * lookup_bdev - lookup a struct block_device by name
2098 * @pathname: special file representing the block device
2100 * Get a reference to the blockdevice at @pathname in the current
2101 * namespace if possible and return it. Return ERR_PTR(error)
2104 struct block_device *lookup_bdev(const char *pathname)
2106 struct block_device *bdev;
2107 struct inode *inode;
2111 if (!pathname || !*pathname)
2112 return ERR_PTR(-EINVAL);
2114 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2116 return ERR_PTR(error);
2118 inode = d_backing_inode(path.dentry);
2120 if (!S_ISBLK(inode->i_mode))
2123 if (!may_open_dev(&path))
2126 bdev = bd_acquire(inode);
2133 bdev = ERR_PTR(error);
2136 EXPORT_SYMBOL(lookup_bdev);
2138 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2140 struct super_block *sb = get_super(bdev);
2145 * no need to lock the super, get_super holds the
2146 * read mutex so the filesystem cannot go away
2147 * under us (->put_super runs with the write lock
2150 shrink_dcache_sb(sb);
2151 res = invalidate_inodes(sb, kill_dirty);
2154 invalidate_bdev(bdev);
2157 EXPORT_SYMBOL(__invalidate_device);
2159 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2161 struct inode *inode, *old_inode = NULL;
2163 spin_lock(&blockdev_superblock->s_inode_list_lock);
2164 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2165 struct address_space *mapping = inode->i_mapping;
2166 struct block_device *bdev;
2168 spin_lock(&inode->i_lock);
2169 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2170 mapping->nrpages == 0) {
2171 spin_unlock(&inode->i_lock);
2175 spin_unlock(&inode->i_lock);
2176 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2178 * We hold a reference to 'inode' so it couldn't have been
2179 * removed from s_inodes list while we dropped the
2180 * s_inode_list_lock We cannot iput the inode now as we can
2181 * be holding the last reference and we cannot iput it under
2182 * s_inode_list_lock. So we keep the reference and iput it
2187 bdev = I_BDEV(inode);
2189 mutex_lock(&bdev->bd_mutex);
2190 if (bdev->bd_openers)
2192 mutex_unlock(&bdev->bd_mutex);
2194 spin_lock(&blockdev_superblock->s_inode_list_lock);
2196 spin_unlock(&blockdev_superblock->s_inode_list_lock);