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/dax.h>
23 #include <linux/buffer_head.h>
24 #include <linux/swap.h>
25 #include <linux/pagevec.h>
26 #include <linux/writeback.h>
27 #include <linux/mpage.h>
28 #include <linux/mount.h>
29 #include <linux/pseudo_fs.h>
30 #include <linux/uio.h>
31 #include <linux/namei.h>
32 #include <linux/log2.h>
33 #include <linux/cleancache.h>
34 #include <linux/task_io_accounting_ops.h>
35 #include <linux/falloc.h>
36 #include <linux/uaccess.h>
37 #include <linux/suspend.h>
41 struct block_device bdev;
42 struct inode vfs_inode;
45 static const struct address_space_operations def_blk_aops;
47 static inline struct bdev_inode *BDEV_I(struct inode *inode)
49 return container_of(inode, struct bdev_inode, vfs_inode);
52 struct block_device *I_BDEV(struct inode *inode)
54 return &BDEV_I(inode)->bdev;
56 EXPORT_SYMBOL(I_BDEV);
58 static void bdev_write_inode(struct block_device *bdev)
60 struct inode *inode = bdev->bd_inode;
63 spin_lock(&inode->i_lock);
64 while (inode->i_state & I_DIRTY) {
65 spin_unlock(&inode->i_lock);
66 ret = write_inode_now(inode, true);
68 char name[BDEVNAME_SIZE];
69 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
70 "for block device %s (err=%d).\n",
71 bdevname(bdev, name), ret);
73 spin_lock(&inode->i_lock);
75 spin_unlock(&inode->i_lock);
78 /* Kill _all_ buffers and pagecache , dirty or not.. */
79 void kill_bdev(struct block_device *bdev)
81 struct address_space *mapping = bdev->bd_inode->i_mapping;
83 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
87 truncate_inode_pages(mapping, 0);
89 EXPORT_SYMBOL(kill_bdev);
91 /* Invalidate clean unused buffers and pagecache. */
92 void invalidate_bdev(struct block_device *bdev)
94 struct address_space *mapping = bdev->bd_inode->i_mapping;
96 if (mapping->nrpages) {
98 lru_add_drain_all(); /* make sure all lru add caches are flushed */
99 invalidate_mapping_pages(mapping, 0, -1);
101 /* 99% of the time, we don't need to flush the cleancache on the bdev.
102 * But, for the strange corners, lets be cautious
104 cleancache_invalidate_inode(mapping);
106 EXPORT_SYMBOL(invalidate_bdev);
108 static void set_init_blocksize(struct block_device *bdev)
110 unsigned bsize = bdev_logical_block_size(bdev);
111 loff_t size = i_size_read(bdev->bd_inode);
113 while (bsize < PAGE_SIZE) {
118 bdev->bd_block_size = bsize;
119 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
122 int set_blocksize(struct block_device *bdev, int size)
124 /* Size must be a power of two, and between 512 and PAGE_SIZE */
125 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
128 /* Size cannot be smaller than the size supported by the device */
129 if (size < bdev_logical_block_size(bdev))
132 /* Don't change the size if it is same as current */
133 if (bdev->bd_block_size != size) {
135 bdev->bd_block_size = size;
136 bdev->bd_inode->i_blkbits = blksize_bits(size);
142 EXPORT_SYMBOL(set_blocksize);
144 int sb_set_blocksize(struct super_block *sb, int size)
146 if (set_blocksize(sb->s_bdev, size))
148 /* If we get here, we know size is power of two
149 * and it's value is between 512 and PAGE_SIZE */
150 sb->s_blocksize = size;
151 sb->s_blocksize_bits = blksize_bits(size);
152 return sb->s_blocksize;
155 EXPORT_SYMBOL(sb_set_blocksize);
157 int sb_min_blocksize(struct super_block *sb, int size)
159 int minsize = bdev_logical_block_size(sb->s_bdev);
162 return sb_set_blocksize(sb, size);
165 EXPORT_SYMBOL(sb_min_blocksize);
168 blkdev_get_block(struct inode *inode, sector_t iblock,
169 struct buffer_head *bh, int create)
171 bh->b_bdev = I_BDEV(inode);
172 bh->b_blocknr = iblock;
173 set_buffer_mapped(bh);
177 static struct inode *bdev_file_inode(struct file *file)
179 return file->f_mapping->host;
182 static unsigned int dio_bio_write_op(struct kiocb *iocb)
184 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
186 /* avoid the need for a I/O completion work item */
187 if (iocb->ki_flags & IOCB_DSYNC)
192 #define DIO_INLINE_BIO_VECS 4
194 static void blkdev_bio_end_io_simple(struct bio *bio)
196 struct task_struct *waiter = bio->bi_private;
198 WRITE_ONCE(bio->bi_private, NULL);
199 blk_wake_io_task(waiter);
203 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
206 struct file *file = iocb->ki_filp;
207 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
208 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
209 loff_t pos = iocb->ki_pos;
210 bool should_dirty = false;
215 if ((pos | iov_iter_alignment(iter)) &
216 (bdev_logical_block_size(bdev) - 1))
219 if (nr_pages <= DIO_INLINE_BIO_VECS)
222 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
228 bio_init(&bio, vecs, nr_pages);
229 bio_set_dev(&bio, bdev);
230 bio.bi_iter.bi_sector = pos >> 9;
231 bio.bi_write_hint = iocb->ki_hint;
232 bio.bi_private = current;
233 bio.bi_end_io = blkdev_bio_end_io_simple;
234 bio.bi_ioprio = iocb->ki_ioprio;
236 ret = bio_iov_iter_get_pages(&bio, iter);
239 ret = bio.bi_iter.bi_size;
241 if (iov_iter_rw(iter) == READ) {
242 bio.bi_opf = REQ_OP_READ;
243 if (iter_is_iovec(iter))
246 bio.bi_opf = dio_bio_write_op(iocb);
247 task_io_account_write(ret);
249 if (iocb->ki_flags & IOCB_NOWAIT)
250 bio.bi_opf |= REQ_NOWAIT;
251 if (iocb->ki_flags & IOCB_HIPRI)
252 bio_set_polled(&bio, iocb);
254 qc = submit_bio(&bio);
256 set_current_state(TASK_UNINTERRUPTIBLE);
257 if (!READ_ONCE(bio.bi_private))
259 if (!(iocb->ki_flags & IOCB_HIPRI) ||
260 !blk_poll(bdev_get_queue(bdev), qc, true))
263 __set_current_state(TASK_RUNNING);
265 bio_release_pages(&bio, should_dirty);
266 if (unlikely(bio.bi_status))
267 ret = blk_status_to_errno(bio.bi_status);
270 if (vecs != inline_vecs)
281 struct task_struct *waiter;
286 bool should_dirty : 1;
291 static struct bio_set blkdev_dio_pool;
293 static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
295 struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
296 struct request_queue *q = bdev_get_queue(bdev);
298 return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
301 static void blkdev_bio_end_io(struct bio *bio)
303 struct blkdev_dio *dio = bio->bi_private;
304 bool should_dirty = dio->should_dirty;
306 if (bio->bi_status && !dio->bio.bi_status)
307 dio->bio.bi_status = bio->bi_status;
309 if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
311 struct kiocb *iocb = dio->iocb;
314 if (likely(!dio->bio.bi_status)) {
318 ret = blk_status_to_errno(dio->bio.bi_status);
321 dio->iocb->ki_complete(iocb, ret, 0);
325 struct task_struct *waiter = dio->waiter;
327 WRITE_ONCE(dio->waiter, NULL);
328 blk_wake_io_task(waiter);
333 bio_check_pages_dirty(bio);
335 bio_release_pages(bio, false);
341 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
343 struct file *file = iocb->ki_filp;
344 struct inode *inode = bdev_file_inode(file);
345 struct block_device *bdev = I_BDEV(inode);
346 struct blk_plug plug;
347 struct blkdev_dio *dio;
349 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
350 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
351 loff_t pos = iocb->ki_pos;
352 blk_qc_t qc = BLK_QC_T_NONE;
355 if ((pos | iov_iter_alignment(iter)) &
356 (bdev_logical_block_size(bdev) - 1))
359 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
361 dio = container_of(bio, struct blkdev_dio, bio);
362 dio->is_sync = is_sync = is_sync_kiocb(iocb);
364 dio->waiter = current;
371 dio->multi_bio = false;
372 dio->should_dirty = is_read && iter_is_iovec(iter);
375 * Don't plug for HIPRI/polled IO, as those should go straight
379 blk_start_plug(&plug);
382 bio_set_dev(bio, bdev);
383 bio->bi_iter.bi_sector = pos >> 9;
384 bio->bi_write_hint = iocb->ki_hint;
385 bio->bi_private = dio;
386 bio->bi_end_io = blkdev_bio_end_io;
387 bio->bi_ioprio = iocb->ki_ioprio;
389 ret = bio_iov_iter_get_pages(bio, iter);
391 bio->bi_status = BLK_STS_IOERR;
397 bio->bi_opf = REQ_OP_READ;
398 if (dio->should_dirty)
399 bio_set_pages_dirty(bio);
401 bio->bi_opf = dio_bio_write_op(iocb);
402 task_io_account_write(bio->bi_iter.bi_size);
404 if (iocb->ki_flags & IOCB_NOWAIT)
405 bio->bi_opf |= REQ_NOWAIT;
407 dio->size += bio->bi_iter.bi_size;
408 pos += bio->bi_iter.bi_size;
410 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
414 if (iocb->ki_flags & IOCB_HIPRI) {
415 bio_set_polled(bio, iocb);
419 qc = submit_bio(bio);
422 WRITE_ONCE(iocb->ki_cookie, qc);
426 if (!dio->multi_bio) {
428 * AIO needs an extra reference to ensure the dio
429 * structure which is embedded into the first bio
434 dio->multi_bio = true;
435 atomic_set(&dio->ref, 2);
437 atomic_inc(&dio->ref);
441 bio = bio_alloc(GFP_KERNEL, nr_pages);
445 blk_finish_plug(&plug);
451 set_current_state(TASK_UNINTERRUPTIBLE);
452 if (!READ_ONCE(dio->waiter))
455 if (!(iocb->ki_flags & IOCB_HIPRI) ||
456 !blk_poll(bdev_get_queue(bdev), qc, true))
459 __set_current_state(TASK_RUNNING);
462 ret = blk_status_to_errno(dio->bio.bi_status);
471 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
475 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
478 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
479 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
481 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
484 static __init int blkdev_init(void)
486 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
488 module_init(blkdev_init);
490 int __sync_blockdev(struct block_device *bdev, int wait)
495 return filemap_flush(bdev->bd_inode->i_mapping);
496 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
500 * Write out and wait upon all the dirty data associated with a block
501 * device via its mapping. Does not take the superblock lock.
503 int sync_blockdev(struct block_device *bdev)
505 return __sync_blockdev(bdev, 1);
507 EXPORT_SYMBOL(sync_blockdev);
510 * Write out and wait upon all dirty data associated with this
511 * device. Filesystem data as well as the underlying block
512 * device. Takes the superblock lock.
514 int fsync_bdev(struct block_device *bdev)
516 struct super_block *sb = get_super(bdev);
518 int res = sync_filesystem(sb);
522 return sync_blockdev(bdev);
524 EXPORT_SYMBOL(fsync_bdev);
527 * freeze_bdev -- lock a filesystem and force it into a consistent state
528 * @bdev: blockdevice to lock
530 * If a superblock is found on this device, we take the s_umount semaphore
531 * on it to make sure nobody unmounts until the snapshot creation is done.
532 * The reference counter (bd_fsfreeze_count) guarantees that only the last
533 * unfreeze process can unfreeze the frozen filesystem actually when multiple
534 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
535 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
538 struct super_block *freeze_bdev(struct block_device *bdev)
540 struct super_block *sb;
543 mutex_lock(&bdev->bd_fsfreeze_mutex);
544 if (++bdev->bd_fsfreeze_count > 1) {
546 * We don't even need to grab a reference - the first call
547 * to freeze_bdev grab an active reference and only the last
548 * thaw_bdev drops it.
550 sb = get_super(bdev);
553 mutex_unlock(&bdev->bd_fsfreeze_mutex);
557 sb = get_active_super(bdev);
560 if (sb->s_op->freeze_super)
561 error = sb->s_op->freeze_super(sb);
563 error = freeze_super(sb);
565 deactivate_super(sb);
566 bdev->bd_fsfreeze_count--;
567 mutex_unlock(&bdev->bd_fsfreeze_mutex);
568 return ERR_PTR(error);
570 deactivate_super(sb);
573 mutex_unlock(&bdev->bd_fsfreeze_mutex);
574 return sb; /* thaw_bdev releases s->s_umount */
576 EXPORT_SYMBOL(freeze_bdev);
579 * thaw_bdev -- unlock filesystem
580 * @bdev: blockdevice to unlock
581 * @sb: associated superblock
583 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
585 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
589 mutex_lock(&bdev->bd_fsfreeze_mutex);
590 if (!bdev->bd_fsfreeze_count)
594 if (--bdev->bd_fsfreeze_count > 0)
600 if (sb->s_op->thaw_super)
601 error = sb->s_op->thaw_super(sb);
603 error = thaw_super(sb);
605 bdev->bd_fsfreeze_count++;
607 mutex_unlock(&bdev->bd_fsfreeze_mutex);
610 EXPORT_SYMBOL(thaw_bdev);
612 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
614 return block_write_full_page(page, blkdev_get_block, wbc);
617 static int blkdev_readpage(struct file * file, struct page * page)
619 return block_read_full_page(page, blkdev_get_block);
622 static int blkdev_readpages(struct file *file, struct address_space *mapping,
623 struct list_head *pages, unsigned nr_pages)
625 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
628 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
629 loff_t pos, unsigned len, unsigned flags,
630 struct page **pagep, void **fsdata)
632 return block_write_begin(mapping, pos, len, flags, pagep,
636 static int blkdev_write_end(struct file *file, struct address_space *mapping,
637 loff_t pos, unsigned len, unsigned copied,
638 struct page *page, void *fsdata)
641 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
651 * for a block special file file_inode(file)->i_size is zero
652 * so we compute the size by hand (just as in block_read/write above)
654 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
656 struct inode *bd_inode = bdev_file_inode(file);
659 inode_lock(bd_inode);
660 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
661 inode_unlock(bd_inode);
665 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
667 struct inode *bd_inode = bdev_file_inode(filp);
668 struct block_device *bdev = I_BDEV(bd_inode);
671 error = file_write_and_wait_range(filp, start, end);
676 * There is no need to serialise calls to blkdev_issue_flush with
677 * i_mutex and doing so causes performance issues with concurrent
678 * O_SYNC writers to a block device.
680 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
681 if (error == -EOPNOTSUPP)
686 EXPORT_SYMBOL(blkdev_fsync);
689 * bdev_read_page() - Start reading a page from a block device
690 * @bdev: The device to read the page from
691 * @sector: The offset on the device to read the page to (need not be aligned)
692 * @page: The page to read
694 * On entry, the page should be locked. It will be unlocked when the page
695 * has been read. If the block driver implements rw_page synchronously,
696 * that will be true on exit from this function, but it need not be.
698 * Errors returned by this function are usually "soft", eg out of memory, or
699 * queue full; callers should try a different route to read this page rather
700 * than propagate an error back up the stack.
702 * Return: negative errno if an error occurs, 0 if submission was successful.
704 int bdev_read_page(struct block_device *bdev, sector_t sector,
707 const struct block_device_operations *ops = bdev->bd_disk->fops;
708 int result = -EOPNOTSUPP;
710 if (!ops->rw_page || bdev_get_integrity(bdev))
713 result = blk_queue_enter(bdev->bd_queue, 0);
716 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
718 blk_queue_exit(bdev->bd_queue);
721 EXPORT_SYMBOL_GPL(bdev_read_page);
724 * bdev_write_page() - Start writing a page to a block device
725 * @bdev: The device to write the page to
726 * @sector: The offset on the device to write the page to (need not be aligned)
727 * @page: The page to write
728 * @wbc: The writeback_control for the write
730 * On entry, the page should be locked and not currently under writeback.
731 * On exit, if the write started successfully, the page will be unlocked and
732 * under writeback. If the write failed already (eg the driver failed to
733 * queue the page to the device), the page will still be locked. If the
734 * caller is a ->writepage implementation, it will need to unlock the page.
736 * Errors returned by this function are usually "soft", eg out of memory, or
737 * queue full; callers should try a different route to write this page rather
738 * than propagate an error back up the stack.
740 * Return: negative errno if an error occurs, 0 if submission was successful.
742 int bdev_write_page(struct block_device *bdev, sector_t sector,
743 struct page *page, struct writeback_control *wbc)
746 const struct block_device_operations *ops = bdev->bd_disk->fops;
748 if (!ops->rw_page || bdev_get_integrity(bdev))
750 result = blk_queue_enter(bdev->bd_queue, 0);
754 set_page_writeback(page);
755 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
758 end_page_writeback(page);
760 clean_page_buffers(page);
763 blk_queue_exit(bdev->bd_queue);
766 EXPORT_SYMBOL_GPL(bdev_write_page);
772 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
773 static struct kmem_cache * bdev_cachep __read_mostly;
775 static struct inode *bdev_alloc_inode(struct super_block *sb)
777 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
780 return &ei->vfs_inode;
783 static void bdev_free_inode(struct inode *inode)
785 kmem_cache_free(bdev_cachep, BDEV_I(inode));
788 static void init_once(void *foo)
790 struct bdev_inode *ei = (struct bdev_inode *) foo;
791 struct block_device *bdev = &ei->bdev;
793 memset(bdev, 0, sizeof(*bdev));
794 mutex_init(&bdev->bd_mutex);
795 INIT_LIST_HEAD(&bdev->bd_list);
797 INIT_LIST_HEAD(&bdev->bd_holder_disks);
799 bdev->bd_bdi = &noop_backing_dev_info;
800 inode_init_once(&ei->vfs_inode);
801 /* Initialize mutex for freeze. */
802 mutex_init(&bdev->bd_fsfreeze_mutex);
805 static void bdev_evict_inode(struct inode *inode)
807 struct block_device *bdev = &BDEV_I(inode)->bdev;
808 truncate_inode_pages_final(&inode->i_data);
809 invalidate_inode_buffers(inode); /* is it needed here? */
811 spin_lock(&bdev_lock);
812 list_del_init(&bdev->bd_list);
813 spin_unlock(&bdev_lock);
814 /* Detach inode from wb early as bdi_put() may free bdi->wb */
815 inode_detach_wb(inode);
816 if (bdev->bd_bdi != &noop_backing_dev_info) {
817 bdi_put(bdev->bd_bdi);
818 bdev->bd_bdi = &noop_backing_dev_info;
822 static const struct super_operations bdev_sops = {
823 .statfs = simple_statfs,
824 .alloc_inode = bdev_alloc_inode,
825 .free_inode = bdev_free_inode,
826 .drop_inode = generic_delete_inode,
827 .evict_inode = bdev_evict_inode,
830 static int bd_init_fs_context(struct fs_context *fc)
832 struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
835 fc->s_iflags |= SB_I_CGROUPWB;
836 ctx->ops = &bdev_sops;
840 static struct file_system_type bd_type = {
842 .init_fs_context = bd_init_fs_context,
843 .kill_sb = kill_anon_super,
846 struct super_block *blockdev_superblock __read_mostly;
847 EXPORT_SYMBOL_GPL(blockdev_superblock);
849 void __init bdev_cache_init(void)
852 static struct vfsmount *bd_mnt;
854 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
855 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
856 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
858 err = register_filesystem(&bd_type);
860 panic("Cannot register bdev pseudo-fs");
861 bd_mnt = kern_mount(&bd_type);
863 panic("Cannot create bdev pseudo-fs");
864 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
868 * Most likely _very_ bad one - but then it's hardly critical for small
869 * /dev and can be fixed when somebody will need really large one.
870 * Keep in mind that it will be fed through icache hash function too.
872 static inline unsigned long hash(dev_t dev)
874 return MAJOR(dev)+MINOR(dev);
877 static int bdev_test(struct inode *inode, void *data)
879 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
882 static int bdev_set(struct inode *inode, void *data)
884 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
888 static LIST_HEAD(all_bdevs);
891 * If there is a bdev inode for this device, unhash it so that it gets evicted
892 * as soon as last inode reference is dropped.
894 void bdev_unhash_inode(dev_t dev)
898 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
900 remove_inode_hash(inode);
905 struct block_device *bdget(dev_t dev)
907 struct block_device *bdev;
910 inode = iget5_locked(blockdev_superblock, hash(dev),
911 bdev_test, bdev_set, &dev);
916 bdev = &BDEV_I(inode)->bdev;
918 if (inode->i_state & I_NEW) {
919 bdev->bd_contains = NULL;
920 bdev->bd_super = NULL;
921 bdev->bd_inode = inode;
922 bdev->bd_block_size = i_blocksize(inode);
923 bdev->bd_part_count = 0;
924 bdev->bd_invalidated = 0;
925 inode->i_mode = S_IFBLK;
927 inode->i_bdev = bdev;
928 inode->i_data.a_ops = &def_blk_aops;
929 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
930 spin_lock(&bdev_lock);
931 list_add(&bdev->bd_list, &all_bdevs);
932 spin_unlock(&bdev_lock);
933 unlock_new_inode(inode);
938 EXPORT_SYMBOL(bdget);
941 * bdgrab -- Grab a reference to an already referenced block device
942 * @bdev: Block device to grab a reference to.
944 struct block_device *bdgrab(struct block_device *bdev)
946 ihold(bdev->bd_inode);
949 EXPORT_SYMBOL(bdgrab);
951 long nr_blockdev_pages(void)
953 struct block_device *bdev;
955 spin_lock(&bdev_lock);
956 list_for_each_entry(bdev, &all_bdevs, bd_list) {
957 ret += bdev->bd_inode->i_mapping->nrpages;
959 spin_unlock(&bdev_lock);
963 void bdput(struct block_device *bdev)
965 iput(bdev->bd_inode);
968 EXPORT_SYMBOL(bdput);
970 static struct block_device *bd_acquire(struct inode *inode)
972 struct block_device *bdev;
974 spin_lock(&bdev_lock);
975 bdev = inode->i_bdev;
976 if (bdev && !inode_unhashed(bdev->bd_inode)) {
978 spin_unlock(&bdev_lock);
981 spin_unlock(&bdev_lock);
984 * i_bdev references block device inode that was already shut down
985 * (corresponding device got removed). Remove the reference and look
986 * up block device inode again just in case new device got
987 * reestablished under the same device number.
992 bdev = bdget(inode->i_rdev);
994 spin_lock(&bdev_lock);
995 if (!inode->i_bdev) {
997 * We take an additional reference to bd_inode,
998 * and it's released in clear_inode() of inode.
999 * So, we can access it via ->i_mapping always
1003 inode->i_bdev = bdev;
1004 inode->i_mapping = bdev->bd_inode->i_mapping;
1006 spin_unlock(&bdev_lock);
1011 /* Call when you free inode */
1013 void bd_forget(struct inode *inode)
1015 struct block_device *bdev = NULL;
1017 spin_lock(&bdev_lock);
1018 if (!sb_is_blkdev_sb(inode->i_sb))
1019 bdev = inode->i_bdev;
1020 inode->i_bdev = NULL;
1021 inode->i_mapping = &inode->i_data;
1022 spin_unlock(&bdev_lock);
1029 * bd_may_claim - test whether a block device can be claimed
1030 * @bdev: block device of interest
1031 * @whole: whole block device containing @bdev, may equal @bdev
1032 * @holder: holder trying to claim @bdev
1034 * Test whether @bdev can be claimed by @holder.
1037 * spin_lock(&bdev_lock).
1040 * %true if @bdev can be claimed, %false otherwise.
1042 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1045 if (bdev->bd_holder == holder)
1046 return true; /* already a holder */
1047 else if (bdev->bd_holder != NULL)
1048 return false; /* held by someone else */
1049 else if (whole == bdev)
1050 return true; /* is a whole device which isn't held */
1052 else if (whole->bd_holder == bd_may_claim)
1053 return true; /* is a partition of a device that is being partitioned */
1054 else if (whole->bd_holder != NULL)
1055 return false; /* is a partition of a held device */
1057 return true; /* is a partition of an un-held device */
1061 * bd_prepare_to_claim - prepare to claim a block device
1062 * @bdev: block device of interest
1063 * @whole: the whole device containing @bdev, may equal @bdev
1064 * @holder: holder trying to claim @bdev
1066 * Prepare to claim @bdev. This function fails if @bdev is already
1067 * claimed by another holder and waits if another claiming is in
1068 * progress. This function doesn't actually claim. On successful
1069 * return, the caller has ownership of bd_claiming and bd_holder[s].
1072 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1073 * it multiple times.
1076 * 0 if @bdev can be claimed, -EBUSY otherwise.
1078 static int bd_prepare_to_claim(struct block_device *bdev,
1079 struct block_device *whole, void *holder)
1082 /* if someone else claimed, fail */
1083 if (!bd_may_claim(bdev, whole, holder))
1086 /* if claiming is already in progress, wait for it to finish */
1087 if (whole->bd_claiming) {
1088 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1091 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1092 spin_unlock(&bdev_lock);
1094 finish_wait(wq, &wait);
1095 spin_lock(&bdev_lock);
1103 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1105 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1110 * Now that we hold gendisk reference we make sure bdev we looked up is
1111 * not stale. If it is, it means device got removed and created before
1112 * we looked up gendisk and we fail open in such case. Associating
1113 * unhashed bdev with newly created gendisk could lead to two bdevs
1114 * (and thus two independent caches) being associated with one device
1117 if (inode_unhashed(bdev->bd_inode)) {
1118 put_disk_and_module(disk);
1125 * bd_start_claiming - start claiming a block device
1126 * @bdev: block device of interest
1127 * @holder: holder trying to claim @bdev
1129 * @bdev is about to be opened exclusively. Check @bdev can be opened
1130 * exclusively and mark that an exclusive open is in progress. Each
1131 * successful call to this function must be matched with a call to
1132 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1135 * This function is used to gain exclusive access to the block device
1136 * without actually causing other exclusive open attempts to fail. It
1137 * should be used when the open sequence itself requires exclusive
1138 * access but may subsequently fail.
1144 * Pointer to the block device containing @bdev on success, ERR_PTR()
1147 struct block_device *bd_start_claiming(struct block_device *bdev, void *holder)
1149 struct gendisk *disk;
1150 struct block_device *whole;
1156 * @bdev might not have been initialized properly yet, look up
1157 * and grab the outer block device the hard way.
1159 disk = bdev_get_gendisk(bdev, &partno);
1161 return ERR_PTR(-ENXIO);
1164 * Normally, @bdev should equal what's returned from bdget_disk()
1165 * if partno is 0; however, some drivers (floppy) use multiple
1166 * bdev's for the same physical device and @bdev may be one of the
1167 * aliases. Keep @bdev if partno is 0. This means claimer
1168 * tracking is broken for those devices but it has always been that
1172 whole = bdget_disk(disk, 0);
1174 whole = bdgrab(bdev);
1176 put_disk_and_module(disk);
1178 return ERR_PTR(-ENOMEM);
1180 /* prepare to claim, if successful, mark claiming in progress */
1181 spin_lock(&bdev_lock);
1183 err = bd_prepare_to_claim(bdev, whole, holder);
1185 whole->bd_claiming = holder;
1186 spin_unlock(&bdev_lock);
1189 spin_unlock(&bdev_lock);
1191 return ERR_PTR(err);
1194 EXPORT_SYMBOL(bd_start_claiming);
1196 static void bd_clear_claiming(struct block_device *whole, void *holder)
1198 lockdep_assert_held(&bdev_lock);
1199 /* tell others that we're done */
1200 BUG_ON(whole->bd_claiming != holder);
1201 whole->bd_claiming = NULL;
1202 wake_up_bit(&whole->bd_claiming, 0);
1206 * bd_finish_claiming - finish claiming of a block device
1207 * @bdev: block device of interest
1208 * @whole: whole block device (returned from bd_start_claiming())
1209 * @holder: holder that has claimed @bdev
1211 * Finish exclusive open of a block device. Mark the device as exlusively
1212 * open by the holder and wake up all waiters for exclusive open to finish.
1214 void bd_finish_claiming(struct block_device *bdev, struct block_device *whole,
1217 spin_lock(&bdev_lock);
1218 BUG_ON(!bd_may_claim(bdev, whole, holder));
1220 * Note that for a whole device bd_holders will be incremented twice,
1221 * and bd_holder will be set to bd_may_claim before being set to holder
1223 whole->bd_holders++;
1224 whole->bd_holder = bd_may_claim;
1226 bdev->bd_holder = holder;
1227 bd_clear_claiming(whole, holder);
1228 spin_unlock(&bdev_lock);
1230 EXPORT_SYMBOL(bd_finish_claiming);
1233 * bd_abort_claiming - abort claiming of a block device
1234 * @bdev: block device of interest
1235 * @whole: whole block device (returned from bd_start_claiming())
1236 * @holder: holder that has claimed @bdev
1238 * Abort claiming of a block device when the exclusive open failed. This can be
1239 * also used when exclusive open is not actually desired and we just needed
1240 * to block other exclusive openers for a while.
1242 void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
1245 spin_lock(&bdev_lock);
1246 bd_clear_claiming(whole, holder);
1247 spin_unlock(&bdev_lock);
1249 EXPORT_SYMBOL(bd_abort_claiming);
1252 struct bd_holder_disk {
1253 struct list_head list;
1254 struct gendisk *disk;
1258 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1259 struct gendisk *disk)
1261 struct bd_holder_disk *holder;
1263 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1264 if (holder->disk == disk)
1269 static int add_symlink(struct kobject *from, struct kobject *to)
1271 return sysfs_create_link(from, to, kobject_name(to));
1274 static void del_symlink(struct kobject *from, struct kobject *to)
1276 sysfs_remove_link(from, kobject_name(to));
1280 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1281 * @bdev: the claimed slave bdev
1282 * @disk: the holding disk
1284 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1286 * This functions creates the following sysfs symlinks.
1288 * - from "slaves" directory of the holder @disk to the claimed @bdev
1289 * - from "holders" directory of the @bdev to the holder @disk
1291 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1292 * passed to bd_link_disk_holder(), then:
1294 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1295 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1297 * The caller must have claimed @bdev before calling this function and
1298 * ensure that both @bdev and @disk are valid during the creation and
1299 * lifetime of these symlinks.
1305 * 0 on success, -errno on failure.
1307 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1309 struct bd_holder_disk *holder;
1312 mutex_lock(&bdev->bd_mutex);
1314 WARN_ON_ONCE(!bdev->bd_holder);
1316 /* FIXME: remove the following once add_disk() handles errors */
1317 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1320 holder = bd_find_holder_disk(bdev, disk);
1326 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1332 INIT_LIST_HEAD(&holder->list);
1333 holder->disk = disk;
1336 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1340 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1344 * bdev could be deleted beneath us which would implicitly destroy
1345 * the holder directory. Hold on to it.
1347 kobject_get(bdev->bd_part->holder_dir);
1349 list_add(&holder->list, &bdev->bd_holder_disks);
1353 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1357 mutex_unlock(&bdev->bd_mutex);
1360 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1363 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1364 * @bdev: the calimed slave bdev
1365 * @disk: the holding disk
1367 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1372 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1374 struct bd_holder_disk *holder;
1376 mutex_lock(&bdev->bd_mutex);
1378 holder = bd_find_holder_disk(bdev, disk);
1380 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1381 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1382 del_symlink(bdev->bd_part->holder_dir,
1383 &disk_to_dev(disk)->kobj);
1384 kobject_put(bdev->bd_part->holder_dir);
1385 list_del_init(&holder->list);
1389 mutex_unlock(&bdev->bd_mutex);
1391 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1395 * flush_disk - invalidates all buffer-cache entries on a disk
1397 * @bdev: struct block device to be flushed
1398 * @kill_dirty: flag to guide handling of dirty inodes
1400 * Invalidates all buffer-cache entries on a disk. It should be called
1401 * when a disk has been changed -- either by a media change or online
1404 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1406 if (__invalidate_device(bdev, kill_dirty)) {
1407 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1408 "resized disk %s\n",
1409 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1411 bdev->bd_invalidated = 1;
1415 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1416 * @disk: struct gendisk to check
1417 * @bdev: struct bdev to adjust.
1418 * @verbose: if %true log a message about a size change if there is any
1420 * This routine checks to see if the bdev size does not match the disk size
1421 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1424 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev,
1427 loff_t disk_size, bdev_size;
1429 disk_size = (loff_t)get_capacity(disk) << 9;
1430 bdev_size = i_size_read(bdev->bd_inode);
1431 if (disk_size != bdev_size) {
1434 "%s: detected capacity change from %lld to %lld\n",
1435 disk->disk_name, bdev_size, disk_size);
1437 i_size_write(bdev->bd_inode, disk_size);
1438 if (bdev_size > disk_size)
1439 flush_disk(bdev, false);
1444 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1445 * @disk: struct gendisk to be revalidated
1447 * This routine is a wrapper for lower-level driver's revalidate_disk
1448 * call-backs. It is used to do common pre and post operations needed
1449 * for all revalidate_disk operations.
1451 int revalidate_disk(struct gendisk *disk)
1455 if (disk->fops->revalidate_disk)
1456 ret = disk->fops->revalidate_disk(disk);
1459 * Hidden disks don't have associated bdev so there's no point in
1462 if (!(disk->flags & GENHD_FL_HIDDEN)) {
1463 struct block_device *bdev = bdget_disk(disk, 0);
1468 mutex_lock(&bdev->bd_mutex);
1469 check_disk_size_change(disk, bdev, ret == 0);
1470 bdev->bd_invalidated = 0;
1471 mutex_unlock(&bdev->bd_mutex);
1476 EXPORT_SYMBOL(revalidate_disk);
1479 * This routine checks whether a removable media has been changed,
1480 * and invalidates all buffer-cache-entries in that case. This
1481 * is a relatively slow routine, so we have to try to minimize using
1482 * it. Thus it is called only upon a 'mount' or 'open'. This
1483 * is the best way of combining speed and utility, I think.
1484 * People changing diskettes in the middle of an operation deserve
1487 int check_disk_change(struct block_device *bdev)
1489 struct gendisk *disk = bdev->bd_disk;
1490 const struct block_device_operations *bdops = disk->fops;
1491 unsigned int events;
1493 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1494 DISK_EVENT_EJECT_REQUEST);
1495 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1498 flush_disk(bdev, true);
1499 if (bdops->revalidate_disk)
1500 bdops->revalidate_disk(bdev->bd_disk);
1504 EXPORT_SYMBOL(check_disk_change);
1506 void bd_set_size(struct block_device *bdev, loff_t size)
1508 inode_lock(bdev->bd_inode);
1509 i_size_write(bdev->bd_inode, size);
1510 inode_unlock(bdev->bd_inode);
1512 EXPORT_SYMBOL(bd_set_size);
1514 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1516 static void bdev_disk_changed(struct block_device *bdev, bool invalidate)
1518 if (disk_part_scan_enabled(bdev->bd_disk)) {
1520 invalidate_partitions(bdev->bd_disk, bdev);
1522 rescan_partitions(bdev->bd_disk, bdev);
1524 check_disk_size_change(bdev->bd_disk, bdev, !invalidate);
1525 bdev->bd_invalidated = 0;
1532 * mutex_lock(part->bd_mutex)
1533 * mutex_lock_nested(whole->bd_mutex, 1)
1536 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1538 struct gendisk *disk;
1542 bool first_open = false;
1544 if (mode & FMODE_READ)
1546 if (mode & FMODE_WRITE)
1549 * hooks: /n/, see "layering violations".
1552 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1560 disk = bdev_get_gendisk(bdev, &partno);
1564 disk_block_events(disk);
1565 mutex_lock_nested(&bdev->bd_mutex, for_part);
1566 if (!bdev->bd_openers) {
1568 bdev->bd_disk = disk;
1569 bdev->bd_queue = disk->queue;
1570 bdev->bd_contains = bdev;
1571 bdev->bd_partno = partno;
1575 bdev->bd_part = disk_get_part(disk, partno);
1580 if (disk->fops->open) {
1581 ret = disk->fops->open(bdev, mode);
1582 if (ret == -ERESTARTSYS) {
1583 /* Lost a race with 'disk' being
1584 * deleted, try again.
1587 disk_put_part(bdev->bd_part);
1588 bdev->bd_part = NULL;
1589 bdev->bd_disk = NULL;
1590 bdev->bd_queue = NULL;
1591 mutex_unlock(&bdev->bd_mutex);
1592 disk_unblock_events(disk);
1593 put_disk_and_module(disk);
1599 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1600 set_init_blocksize(bdev);
1604 * If the device is invalidated, rescan partition
1605 * if open succeeded or failed with -ENOMEDIUM.
1606 * The latter is necessary to prevent ghost
1607 * partitions on a removed medium.
1609 if (bdev->bd_invalidated &&
1610 (!ret || ret == -ENOMEDIUM))
1611 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1616 struct block_device *whole;
1617 whole = bdget_disk(disk, 0);
1622 ret = __blkdev_get(whole, mode, 1);
1627 bdev->bd_contains = whole;
1628 bdev->bd_part = disk_get_part(disk, partno);
1629 if (!(disk->flags & GENHD_FL_UP) ||
1630 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1634 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1635 set_init_blocksize(bdev);
1638 if (bdev->bd_bdi == &noop_backing_dev_info)
1639 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1641 if (bdev->bd_contains == bdev) {
1643 if (bdev->bd_disk->fops->open)
1644 ret = bdev->bd_disk->fops->open(bdev, mode);
1645 /* the same as first opener case, read comment there */
1646 if (bdev->bd_invalidated &&
1647 (!ret || ret == -ENOMEDIUM))
1648 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1650 goto out_unlock_bdev;
1655 bdev->bd_part_count++;
1656 mutex_unlock(&bdev->bd_mutex);
1657 disk_unblock_events(disk);
1658 /* only one opener holds refs to the module and disk */
1660 put_disk_and_module(disk);
1664 disk_put_part(bdev->bd_part);
1665 bdev->bd_disk = NULL;
1666 bdev->bd_part = NULL;
1667 bdev->bd_queue = NULL;
1668 if (bdev != bdev->bd_contains)
1669 __blkdev_put(bdev->bd_contains, mode, 1);
1670 bdev->bd_contains = NULL;
1672 mutex_unlock(&bdev->bd_mutex);
1673 disk_unblock_events(disk);
1674 put_disk_and_module(disk);
1681 * blkdev_get - open a block device
1682 * @bdev: block_device to open
1683 * @mode: FMODE_* mask
1684 * @holder: exclusive holder identifier
1686 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1687 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1688 * @holder is invalid. Exclusive opens may nest for the same @holder.
1690 * On success, the reference count of @bdev is unchanged. On failure,
1697 * 0 on success, -errno on failure.
1699 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1701 struct block_device *whole = NULL;
1704 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1706 if ((mode & FMODE_EXCL) && holder) {
1707 whole = bd_start_claiming(bdev, holder);
1708 if (IS_ERR(whole)) {
1710 return PTR_ERR(whole);
1714 res = __blkdev_get(bdev, mode, 0);
1717 struct gendisk *disk = whole->bd_disk;
1719 /* finish claiming */
1720 mutex_lock(&bdev->bd_mutex);
1722 bd_finish_claiming(bdev, whole, holder);
1724 bd_abort_claiming(bdev, whole, holder);
1726 * Block event polling for write claims if requested. Any
1727 * write holder makes the write_holder state stick until
1728 * all are released. This is good enough and tracking
1729 * individual writeable reference is too fragile given the
1730 * way @mode is used in blkdev_get/put().
1732 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1733 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1734 bdev->bd_write_holder = true;
1735 disk_block_events(disk);
1738 mutex_unlock(&bdev->bd_mutex);
1747 EXPORT_SYMBOL(blkdev_get);
1750 * blkdev_get_by_path - open a block device by name
1751 * @path: path to the block device to open
1752 * @mode: FMODE_* mask
1753 * @holder: exclusive holder identifier
1755 * Open the blockdevice described by the device file at @path. @mode
1756 * and @holder are identical to blkdev_get().
1758 * On success, the returned block_device has reference count of one.
1764 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1766 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1769 struct block_device *bdev;
1772 bdev = lookup_bdev(path);
1776 err = blkdev_get(bdev, mode, holder);
1778 return ERR_PTR(err);
1780 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1781 blkdev_put(bdev, mode);
1782 return ERR_PTR(-EACCES);
1787 EXPORT_SYMBOL(blkdev_get_by_path);
1790 * blkdev_get_by_dev - open a block device by device number
1791 * @dev: device number of block device to open
1792 * @mode: FMODE_* mask
1793 * @holder: exclusive holder identifier
1795 * Open the blockdevice described by device number @dev. @mode and
1796 * @holder are identical to blkdev_get().
1798 * Use it ONLY if you really do not have anything better - i.e. when
1799 * you are behind a truly sucky interface and all you are given is a
1800 * device number. _Never_ to be used for internal purposes. If you
1801 * ever need it - reconsider your API.
1803 * On success, the returned block_device has reference count of one.
1809 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1811 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1813 struct block_device *bdev;
1818 return ERR_PTR(-ENOMEM);
1820 err = blkdev_get(bdev, mode, holder);
1822 return ERR_PTR(err);
1826 EXPORT_SYMBOL(blkdev_get_by_dev);
1828 static int blkdev_open(struct inode * inode, struct file * filp)
1830 struct block_device *bdev;
1833 * Preserve backwards compatibility and allow large file access
1834 * even if userspace doesn't ask for it explicitly. Some mkfs
1835 * binary needs it. We might want to drop this workaround
1836 * during an unstable branch.
1838 filp->f_flags |= O_LARGEFILE;
1840 filp->f_mode |= FMODE_NOWAIT;
1842 if (filp->f_flags & O_NDELAY)
1843 filp->f_mode |= FMODE_NDELAY;
1844 if (filp->f_flags & O_EXCL)
1845 filp->f_mode |= FMODE_EXCL;
1846 if ((filp->f_flags & O_ACCMODE) == 3)
1847 filp->f_mode |= FMODE_WRITE_IOCTL;
1849 bdev = bd_acquire(inode);
1853 filp->f_mapping = bdev->bd_inode->i_mapping;
1854 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1856 return blkdev_get(bdev, filp->f_mode, filp);
1859 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1861 struct gendisk *disk = bdev->bd_disk;
1862 struct block_device *victim = NULL;
1865 * Sync early if it looks like we're the last one. If someone else
1866 * opens the block device between now and the decrement of bd_openers
1867 * then we did a sync that we didn't need to, but that's not the end
1868 * of the world and we want to avoid long (could be several minute)
1869 * syncs while holding the mutex.
1871 if (bdev->bd_openers == 1)
1872 sync_blockdev(bdev);
1874 mutex_lock_nested(&bdev->bd_mutex, for_part);
1876 bdev->bd_part_count--;
1878 if (!--bdev->bd_openers) {
1879 WARN_ON_ONCE(bdev->bd_holders);
1880 sync_blockdev(bdev);
1883 bdev_write_inode(bdev);
1885 if (bdev->bd_contains == bdev) {
1886 if (disk->fops->release)
1887 disk->fops->release(disk, mode);
1889 if (!bdev->bd_openers) {
1890 disk_put_part(bdev->bd_part);
1891 bdev->bd_part = NULL;
1892 bdev->bd_disk = NULL;
1893 if (bdev != bdev->bd_contains)
1894 victim = bdev->bd_contains;
1895 bdev->bd_contains = NULL;
1897 put_disk_and_module(disk);
1899 mutex_unlock(&bdev->bd_mutex);
1902 __blkdev_put(victim, mode, 1);
1905 void blkdev_put(struct block_device *bdev, fmode_t mode)
1907 mutex_lock(&bdev->bd_mutex);
1909 if (mode & FMODE_EXCL) {
1913 * Release a claim on the device. The holder fields
1914 * are protected with bdev_lock. bd_mutex is to
1915 * synchronize disk_holder unlinking.
1917 spin_lock(&bdev_lock);
1919 WARN_ON_ONCE(--bdev->bd_holders < 0);
1920 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1922 /* bd_contains might point to self, check in a separate step */
1923 if ((bdev_free = !bdev->bd_holders))
1924 bdev->bd_holder = NULL;
1925 if (!bdev->bd_contains->bd_holders)
1926 bdev->bd_contains->bd_holder = NULL;
1928 spin_unlock(&bdev_lock);
1931 * If this was the last claim, remove holder link and
1932 * unblock evpoll if it was a write holder.
1934 if (bdev_free && bdev->bd_write_holder) {
1935 disk_unblock_events(bdev->bd_disk);
1936 bdev->bd_write_holder = false;
1941 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1942 * event. This is to ensure detection of media removal commanded
1943 * from userland - e.g. eject(1).
1945 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1947 mutex_unlock(&bdev->bd_mutex);
1949 __blkdev_put(bdev, mode, 0);
1951 EXPORT_SYMBOL(blkdev_put);
1953 static int blkdev_close(struct inode * inode, struct file * filp)
1955 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1956 blkdev_put(bdev, filp->f_mode);
1960 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1962 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1963 fmode_t mode = file->f_mode;
1966 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1967 * to updated it before every ioctl.
1969 if (file->f_flags & O_NDELAY)
1970 mode |= FMODE_NDELAY;
1972 mode &= ~FMODE_NDELAY;
1974 return blkdev_ioctl(bdev, mode, cmd, arg);
1978 * Write data to the block device. Only intended for the block device itself
1979 * and the raw driver which basically is a fake block device.
1981 * Does not take i_mutex for the write and thus is not for general purpose
1984 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1986 struct file *file = iocb->ki_filp;
1987 struct inode *bd_inode = bdev_file_inode(file);
1988 loff_t size = i_size_read(bd_inode);
1989 struct blk_plug plug;
1993 if (bdev_read_only(I_BDEV(bd_inode)))
1996 /* uswsusp needs write permission to the swap */
1997 if (IS_SWAPFILE(bd_inode) && !hibernation_available())
2000 if (!iov_iter_count(from))
2003 if (iocb->ki_pos >= size)
2006 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
2009 size -= iocb->ki_pos;
2010 if (iov_iter_count(from) > size) {
2011 shorted = iov_iter_count(from) - size;
2012 iov_iter_truncate(from, size);
2015 blk_start_plug(&plug);
2016 ret = __generic_file_write_iter(iocb, from);
2018 ret = generic_write_sync(iocb, ret);
2019 iov_iter_reexpand(from, iov_iter_count(from) + shorted);
2020 blk_finish_plug(&plug);
2023 EXPORT_SYMBOL_GPL(blkdev_write_iter);
2025 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
2027 struct file *file = iocb->ki_filp;
2028 struct inode *bd_inode = bdev_file_inode(file);
2029 loff_t size = i_size_read(bd_inode);
2030 loff_t pos = iocb->ki_pos;
2038 if (iov_iter_count(to) > size) {
2039 shorted = iov_iter_count(to) - size;
2040 iov_iter_truncate(to, size);
2043 ret = generic_file_read_iter(iocb, to);
2044 iov_iter_reexpand(to, iov_iter_count(to) + shorted);
2047 EXPORT_SYMBOL_GPL(blkdev_read_iter);
2050 * Try to release a page associated with block device when the system
2051 * is under memory pressure.
2053 static int blkdev_releasepage(struct page *page, gfp_t wait)
2055 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
2057 if (super && super->s_op->bdev_try_to_free_page)
2058 return super->s_op->bdev_try_to_free_page(super, page, wait);
2060 return try_to_free_buffers(page);
2063 static int blkdev_writepages(struct address_space *mapping,
2064 struct writeback_control *wbc)
2066 return generic_writepages(mapping, wbc);
2069 static const struct address_space_operations def_blk_aops = {
2070 .readpage = blkdev_readpage,
2071 .readpages = blkdev_readpages,
2072 .writepage = blkdev_writepage,
2073 .write_begin = blkdev_write_begin,
2074 .write_end = blkdev_write_end,
2075 .writepages = blkdev_writepages,
2076 .releasepage = blkdev_releasepage,
2077 .direct_IO = blkdev_direct_IO,
2078 .migratepage = buffer_migrate_page_norefs,
2079 .is_dirty_writeback = buffer_check_dirty_writeback,
2082 #define BLKDEV_FALLOC_FL_SUPPORTED \
2083 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2084 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2086 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2089 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2090 struct address_space *mapping;
2091 loff_t end = start + len - 1;
2095 /* Fail if we don't recognize the flags. */
2096 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2099 /* Don't go off the end of the device. */
2100 isize = i_size_read(bdev->bd_inode);
2104 if (mode & FALLOC_FL_KEEP_SIZE) {
2105 len = isize - start;
2106 end = start + len - 1;
2112 * Don't allow IO that isn't aligned to logical block size.
2114 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2117 /* Invalidate the page cache, including dirty pages. */
2118 mapping = bdev->bd_inode->i_mapping;
2119 truncate_inode_pages_range(mapping, start, end);
2122 case FALLOC_FL_ZERO_RANGE:
2123 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2124 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2125 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2127 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2128 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2129 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2131 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2132 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2142 * Invalidate again; if someone wandered in and dirtied a page,
2143 * the caller will be given -EBUSY. The third argument is
2144 * inclusive, so the rounding here is safe.
2146 return invalidate_inode_pages2_range(mapping,
2147 start >> PAGE_SHIFT,
2151 const struct file_operations def_blk_fops = {
2152 .open = blkdev_open,
2153 .release = blkdev_close,
2154 .llseek = block_llseek,
2155 .read_iter = blkdev_read_iter,
2156 .write_iter = blkdev_write_iter,
2157 .iopoll = blkdev_iopoll,
2158 .mmap = generic_file_mmap,
2159 .fsync = blkdev_fsync,
2160 .unlocked_ioctl = block_ioctl,
2161 #ifdef CONFIG_COMPAT
2162 .compat_ioctl = compat_blkdev_ioctl,
2164 .splice_read = generic_file_splice_read,
2165 .splice_write = iter_file_splice_write,
2166 .fallocate = blkdev_fallocate,
2169 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2172 mm_segment_t old_fs = get_fs();
2174 res = blkdev_ioctl(bdev, 0, cmd, arg);
2179 EXPORT_SYMBOL(ioctl_by_bdev);
2182 * lookup_bdev - lookup a struct block_device by name
2183 * @pathname: special file representing the block device
2185 * Get a reference to the blockdevice at @pathname in the current
2186 * namespace if possible and return it. Return ERR_PTR(error)
2189 struct block_device *lookup_bdev(const char *pathname)
2191 struct block_device *bdev;
2192 struct inode *inode;
2196 if (!pathname || !*pathname)
2197 return ERR_PTR(-EINVAL);
2199 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2201 return ERR_PTR(error);
2203 inode = d_backing_inode(path.dentry);
2205 if (!S_ISBLK(inode->i_mode))
2208 if (!may_open_dev(&path))
2211 bdev = bd_acquire(inode);
2218 bdev = ERR_PTR(error);
2221 EXPORT_SYMBOL(lookup_bdev);
2223 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2225 struct super_block *sb = get_super(bdev);
2230 * no need to lock the super, get_super holds the
2231 * read mutex so the filesystem cannot go away
2232 * under us (->put_super runs with the write lock
2235 shrink_dcache_sb(sb);
2236 res = invalidate_inodes(sb, kill_dirty);
2239 invalidate_bdev(bdev);
2242 EXPORT_SYMBOL(__invalidate_device);
2244 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2246 struct inode *inode, *old_inode = NULL;
2248 spin_lock(&blockdev_superblock->s_inode_list_lock);
2249 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2250 struct address_space *mapping = inode->i_mapping;
2251 struct block_device *bdev;
2253 spin_lock(&inode->i_lock);
2254 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2255 mapping->nrpages == 0) {
2256 spin_unlock(&inode->i_lock);
2260 spin_unlock(&inode->i_lock);
2261 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2263 * We hold a reference to 'inode' so it couldn't have been
2264 * removed from s_inodes list while we dropped the
2265 * s_inode_list_lock We cannot iput the inode now as we can
2266 * be holding the last reference and we cannot iput it under
2267 * s_inode_list_lock. So we keep the reference and iput it
2272 bdev = I_BDEV(inode);
2274 mutex_lock(&bdev->bd_mutex);
2275 if (bdev->bd_openers)
2277 mutex_unlock(&bdev->bd_mutex);
2279 spin_lock(&blockdev_superblock->s_inode_list_lock);
2281 spin_unlock(&blockdev_superblock->s_inode_list_lock);