1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
9 #include <linux/init.h>
11 #include <linux/fcntl.h>
12 #include <linux/slab.h>
13 #include <linux/kmod.h>
14 #include <linux/major.h>
15 #include <linux/device_cgroup.h>
16 #include <linux/highmem.h>
17 #include <linux/blkdev.h>
18 #include <linux/backing-dev.h>
19 #include <linux/module.h>
20 #include <linux/blkpg.h>
21 #include <linux/magic.h>
22 #include <linux/buffer_head.h>
23 #include <linux/swap.h>
24 #include <linux/pagevec.h>
25 #include <linux/writeback.h>
26 #include <linux/mpage.h>
27 #include <linux/mount.h>
28 #include <linux/pseudo_fs.h>
29 #include <linux/uio.h>
30 #include <linux/namei.h>
31 #include <linux/log2.h>
32 #include <linux/cleancache.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/falloc.h>
35 #include <linux/uaccess.h>
36 #include <linux/suspend.h>
40 struct block_device bdev;
41 struct inode vfs_inode;
44 static const struct address_space_operations def_blk_aops;
46 static inline struct bdev_inode *BDEV_I(struct inode *inode)
48 return container_of(inode, struct bdev_inode, vfs_inode);
51 struct block_device *I_BDEV(struct inode *inode)
53 return &BDEV_I(inode)->bdev;
55 EXPORT_SYMBOL(I_BDEV);
57 static void bdev_write_inode(struct block_device *bdev)
59 struct inode *inode = bdev->bd_inode;
62 spin_lock(&inode->i_lock);
63 while (inode->i_state & I_DIRTY) {
64 spin_unlock(&inode->i_lock);
65 ret = write_inode_now(inode, true);
67 char name[BDEVNAME_SIZE];
68 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
69 "for block device %s (err=%d).\n",
70 bdevname(bdev, name), ret);
72 spin_lock(&inode->i_lock);
74 spin_unlock(&inode->i_lock);
77 /* Kill _all_ buffers and pagecache , dirty or not.. */
78 static void kill_bdev(struct block_device *bdev)
80 struct address_space *mapping = bdev->bd_inode->i_mapping;
82 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
86 truncate_inode_pages(mapping, 0);
89 /* Invalidate clean unused buffers and pagecache. */
90 void invalidate_bdev(struct block_device *bdev)
92 struct address_space *mapping = bdev->bd_inode->i_mapping;
94 if (mapping->nrpages) {
96 lru_add_drain_all(); /* make sure all lru add caches are flushed */
97 invalidate_mapping_pages(mapping, 0, -1);
99 /* 99% of the time, we don't need to flush the cleancache on the bdev.
100 * But, for the strange corners, lets be cautious
102 cleancache_invalidate_inode(mapping);
104 EXPORT_SYMBOL(invalidate_bdev);
106 static void set_init_blocksize(struct block_device *bdev)
108 bdev->bd_inode->i_blkbits = blksize_bits(bdev_logical_block_size(bdev));
111 int set_blocksize(struct block_device *bdev, int size)
113 /* Size must be a power of two, and between 512 and PAGE_SIZE */
114 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
117 /* Size cannot be smaller than the size supported by the device */
118 if (size < bdev_logical_block_size(bdev))
121 /* Don't change the size if it is same as current */
122 if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
124 bdev->bd_inode->i_blkbits = blksize_bits(size);
130 EXPORT_SYMBOL(set_blocksize);
132 int sb_set_blocksize(struct super_block *sb, int size)
134 if (set_blocksize(sb->s_bdev, size))
136 /* If we get here, we know size is power of two
137 * and it's value is between 512 and PAGE_SIZE */
138 sb->s_blocksize = size;
139 sb->s_blocksize_bits = blksize_bits(size);
140 return sb->s_blocksize;
143 EXPORT_SYMBOL(sb_set_blocksize);
145 int sb_min_blocksize(struct super_block *sb, int size)
147 int minsize = bdev_logical_block_size(sb->s_bdev);
150 return sb_set_blocksize(sb, size);
153 EXPORT_SYMBOL(sb_min_blocksize);
156 blkdev_get_block(struct inode *inode, sector_t iblock,
157 struct buffer_head *bh, int create)
159 bh->b_bdev = I_BDEV(inode);
160 bh->b_blocknr = iblock;
161 set_buffer_mapped(bh);
165 static struct inode *bdev_file_inode(struct file *file)
167 return file->f_mapping->host;
170 static unsigned int dio_bio_write_op(struct kiocb *iocb)
172 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
174 /* avoid the need for a I/O completion work item */
175 if (iocb->ki_flags & IOCB_DSYNC)
180 #define DIO_INLINE_BIO_VECS 4
182 static void blkdev_bio_end_io_simple(struct bio *bio)
184 struct task_struct *waiter = bio->bi_private;
186 WRITE_ONCE(bio->bi_private, NULL);
187 blk_wake_io_task(waiter);
191 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
194 struct file *file = iocb->ki_filp;
195 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
196 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
197 loff_t pos = iocb->ki_pos;
198 bool should_dirty = false;
203 if ((pos | iov_iter_alignment(iter)) &
204 (bdev_logical_block_size(bdev) - 1))
207 if (nr_pages <= DIO_INLINE_BIO_VECS)
210 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
216 bio_init(&bio, vecs, nr_pages);
217 bio_set_dev(&bio, bdev);
218 bio.bi_iter.bi_sector = pos >> 9;
219 bio.bi_write_hint = iocb->ki_hint;
220 bio.bi_private = current;
221 bio.bi_end_io = blkdev_bio_end_io_simple;
222 bio.bi_ioprio = iocb->ki_ioprio;
224 ret = bio_iov_iter_get_pages(&bio, iter);
227 ret = bio.bi_iter.bi_size;
229 if (iov_iter_rw(iter) == READ) {
230 bio.bi_opf = REQ_OP_READ;
231 if (iter_is_iovec(iter))
234 bio.bi_opf = dio_bio_write_op(iocb);
235 task_io_account_write(ret);
237 if (iocb->ki_flags & IOCB_HIPRI)
238 bio_set_polled(&bio, iocb);
240 qc = submit_bio(&bio);
242 set_current_state(TASK_UNINTERRUPTIBLE);
243 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_release_pages(&bio, should_dirty);
252 if (unlikely(bio.bi_status))
253 ret = blk_status_to_errno(bio.bi_status);
256 if (vecs != inline_vecs)
267 struct task_struct *waiter;
272 bool should_dirty : 1;
277 static struct bio_set blkdev_dio_pool;
279 static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
281 struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
282 struct request_queue *q = bdev_get_queue(bdev);
284 return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
287 static void blkdev_bio_end_io(struct bio *bio)
289 struct blkdev_dio *dio = bio->bi_private;
290 bool should_dirty = dio->should_dirty;
292 if (bio->bi_status && !dio->bio.bi_status)
293 dio->bio.bi_status = bio->bi_status;
295 if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
297 struct kiocb *iocb = dio->iocb;
300 if (likely(!dio->bio.bi_status)) {
304 ret = blk_status_to_errno(dio->bio.bi_status);
307 dio->iocb->ki_complete(iocb, ret, 0);
311 struct task_struct *waiter = dio->waiter;
313 WRITE_ONCE(dio->waiter, NULL);
314 blk_wake_io_task(waiter);
319 bio_check_pages_dirty(bio);
321 bio_release_pages(bio, false);
327 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
329 struct file *file = iocb->ki_filp;
330 struct inode *inode = bdev_file_inode(file);
331 struct block_device *bdev = I_BDEV(inode);
332 struct blk_plug plug;
333 struct blkdev_dio *dio;
335 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
336 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
337 loff_t pos = iocb->ki_pos;
338 blk_qc_t qc = BLK_QC_T_NONE;
341 if ((pos | iov_iter_alignment(iter)) &
342 (bdev_logical_block_size(bdev) - 1))
345 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
347 dio = container_of(bio, struct blkdev_dio, bio);
348 dio->is_sync = is_sync = is_sync_kiocb(iocb);
350 dio->waiter = current;
357 dio->multi_bio = false;
358 dio->should_dirty = is_read && iter_is_iovec(iter);
361 * Don't plug for HIPRI/polled IO, as those should go straight
365 blk_start_plug(&plug);
368 bio_set_dev(bio, bdev);
369 bio->bi_iter.bi_sector = pos >> 9;
370 bio->bi_write_hint = iocb->ki_hint;
371 bio->bi_private = dio;
372 bio->bi_end_io = blkdev_bio_end_io;
373 bio->bi_ioprio = iocb->ki_ioprio;
375 ret = bio_iov_iter_get_pages(bio, iter);
377 bio->bi_status = BLK_STS_IOERR;
383 bio->bi_opf = REQ_OP_READ;
384 if (dio->should_dirty)
385 bio_set_pages_dirty(bio);
387 bio->bi_opf = dio_bio_write_op(iocb);
388 task_io_account_write(bio->bi_iter.bi_size);
391 dio->size += bio->bi_iter.bi_size;
392 pos += bio->bi_iter.bi_size;
394 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
398 if (iocb->ki_flags & IOCB_HIPRI) {
399 bio_set_polled(bio, iocb);
403 qc = submit_bio(bio);
406 WRITE_ONCE(iocb->ki_cookie, qc);
410 if (!dio->multi_bio) {
412 * AIO needs an extra reference to ensure the dio
413 * structure which is embedded into the first bio
418 dio->multi_bio = true;
419 atomic_set(&dio->ref, 2);
421 atomic_inc(&dio->ref);
425 bio = bio_alloc(GFP_KERNEL, nr_pages);
429 blk_finish_plug(&plug);
435 set_current_state(TASK_UNINTERRUPTIBLE);
436 if (!READ_ONCE(dio->waiter))
439 if (!(iocb->ki_flags & IOCB_HIPRI) ||
440 !blk_poll(bdev_get_queue(bdev), qc, true))
443 __set_current_state(TASK_RUNNING);
446 ret = blk_status_to_errno(dio->bio.bi_status);
455 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
459 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
462 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
463 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
465 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
468 static __init int blkdev_init(void)
470 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
472 module_init(blkdev_init);
474 int __sync_blockdev(struct block_device *bdev, int wait)
479 return filemap_flush(bdev->bd_inode->i_mapping);
480 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
484 * Write out and wait upon all the dirty data associated with a block
485 * device via its mapping. Does not take the superblock lock.
487 int sync_blockdev(struct block_device *bdev)
489 return __sync_blockdev(bdev, 1);
491 EXPORT_SYMBOL(sync_blockdev);
494 * Write out and wait upon all dirty data associated with this
495 * device. Filesystem data as well as the underlying block
496 * device. Takes the superblock lock.
498 int fsync_bdev(struct block_device *bdev)
500 struct super_block *sb = get_super(bdev);
502 int res = sync_filesystem(sb);
506 return sync_blockdev(bdev);
508 EXPORT_SYMBOL(fsync_bdev);
511 * freeze_bdev -- lock a filesystem and force it into a consistent state
512 * @bdev: blockdevice to lock
514 * If a superblock is found on this device, we take the s_umount semaphore
515 * on it to make sure nobody unmounts until the snapshot creation is done.
516 * The reference counter (bd_fsfreeze_count) guarantees that only the last
517 * unfreeze process can unfreeze the frozen filesystem actually when multiple
518 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
519 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
522 struct super_block *freeze_bdev(struct block_device *bdev)
524 struct super_block *sb;
527 mutex_lock(&bdev->bd_fsfreeze_mutex);
528 if (++bdev->bd_fsfreeze_count > 1) {
530 * We don't even need to grab a reference - the first call
531 * to freeze_bdev grab an active reference and only the last
532 * thaw_bdev drops it.
534 sb = get_super(bdev);
537 mutex_unlock(&bdev->bd_fsfreeze_mutex);
541 sb = get_active_super(bdev);
544 if (sb->s_op->freeze_super)
545 error = sb->s_op->freeze_super(sb);
547 error = freeze_super(sb);
549 deactivate_super(sb);
550 bdev->bd_fsfreeze_count--;
551 mutex_unlock(&bdev->bd_fsfreeze_mutex);
552 return ERR_PTR(error);
554 deactivate_super(sb);
557 mutex_unlock(&bdev->bd_fsfreeze_mutex);
558 return sb; /* thaw_bdev releases s->s_umount */
560 EXPORT_SYMBOL(freeze_bdev);
563 * thaw_bdev -- unlock filesystem
564 * @bdev: blockdevice to unlock
565 * @sb: associated superblock
567 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
569 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
573 mutex_lock(&bdev->bd_fsfreeze_mutex);
574 if (!bdev->bd_fsfreeze_count)
578 if (--bdev->bd_fsfreeze_count > 0)
584 if (sb->s_op->thaw_super)
585 error = sb->s_op->thaw_super(sb);
587 error = thaw_super(sb);
589 bdev->bd_fsfreeze_count++;
591 mutex_unlock(&bdev->bd_fsfreeze_mutex);
594 EXPORT_SYMBOL(thaw_bdev);
596 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
598 return block_write_full_page(page, blkdev_get_block, wbc);
601 static int blkdev_readpage(struct file * file, struct page * page)
603 return block_read_full_page(page, blkdev_get_block);
606 static void blkdev_readahead(struct readahead_control *rac)
608 mpage_readahead(rac, blkdev_get_block);
611 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
612 loff_t pos, unsigned len, unsigned flags,
613 struct page **pagep, void **fsdata)
615 return block_write_begin(mapping, pos, len, flags, pagep,
619 static int blkdev_write_end(struct file *file, struct address_space *mapping,
620 loff_t pos, unsigned len, unsigned copied,
621 struct page *page, void *fsdata)
624 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
634 * for a block special file file_inode(file)->i_size is zero
635 * so we compute the size by hand (just as in block_read/write above)
637 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
639 struct inode *bd_inode = bdev_file_inode(file);
642 inode_lock(bd_inode);
643 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
644 inode_unlock(bd_inode);
648 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
650 struct inode *bd_inode = bdev_file_inode(filp);
651 struct block_device *bdev = I_BDEV(bd_inode);
654 error = file_write_and_wait_range(filp, start, end);
659 * There is no need to serialise calls to blkdev_issue_flush with
660 * i_mutex and doing so causes performance issues with concurrent
661 * O_SYNC writers to a block device.
663 error = blkdev_issue_flush(bdev, GFP_KERNEL);
664 if (error == -EOPNOTSUPP)
669 EXPORT_SYMBOL(blkdev_fsync);
672 * bdev_read_page() - Start reading a page from a block device
673 * @bdev: The device to read the page from
674 * @sector: The offset on the device to read the page to (need not be aligned)
675 * @page: The page to read
677 * On entry, the page should be locked. It will be unlocked when the page
678 * has been read. If the block driver implements rw_page synchronously,
679 * that will be true on exit from this function, but it need not be.
681 * Errors returned by this function are usually "soft", eg out of memory, or
682 * queue full; callers should try a different route to read this page rather
683 * than propagate an error back up the stack.
685 * Return: negative errno if an error occurs, 0 if submission was successful.
687 int bdev_read_page(struct block_device *bdev, sector_t sector,
690 const struct block_device_operations *ops = bdev->bd_disk->fops;
691 int result = -EOPNOTSUPP;
693 if (!ops->rw_page || bdev_get_integrity(bdev))
696 result = blk_queue_enter(bdev->bd_disk->queue, 0);
699 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
701 blk_queue_exit(bdev->bd_disk->queue);
706 * bdev_write_page() - Start writing a page to a block device
707 * @bdev: The device to write the page to
708 * @sector: The offset on the device to write the page to (need not be aligned)
709 * @page: The page to write
710 * @wbc: The writeback_control for the write
712 * On entry, the page should be locked and not currently under writeback.
713 * On exit, if the write started successfully, the page will be unlocked and
714 * under writeback. If the write failed already (eg the driver failed to
715 * queue the page to the device), the page will still be locked. If the
716 * caller is a ->writepage implementation, it will need to unlock the page.
718 * Errors returned by this function are usually "soft", eg out of memory, or
719 * queue full; callers should try a different route to write this page rather
720 * than propagate an error back up the stack.
722 * Return: negative errno if an error occurs, 0 if submission was successful.
724 int bdev_write_page(struct block_device *bdev, sector_t sector,
725 struct page *page, struct writeback_control *wbc)
728 const struct block_device_operations *ops = bdev->bd_disk->fops;
730 if (!ops->rw_page || bdev_get_integrity(bdev))
732 result = blk_queue_enter(bdev->bd_disk->queue, 0);
736 set_page_writeback(page);
737 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
740 end_page_writeback(page);
742 clean_page_buffers(page);
745 blk_queue_exit(bdev->bd_disk->queue);
753 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
754 static struct kmem_cache * bdev_cachep __read_mostly;
756 static struct inode *bdev_alloc_inode(struct super_block *sb)
758 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
761 return &ei->vfs_inode;
764 static void bdev_free_inode(struct inode *inode)
766 kmem_cache_free(bdev_cachep, BDEV_I(inode));
769 static void init_once(void *foo)
771 struct bdev_inode *ei = (struct bdev_inode *) foo;
772 struct block_device *bdev = &ei->bdev;
774 memset(bdev, 0, sizeof(*bdev));
775 mutex_init(&bdev->bd_mutex);
776 INIT_LIST_HEAD(&bdev->bd_list);
778 INIT_LIST_HEAD(&bdev->bd_holder_disks);
780 bdev->bd_bdi = &noop_backing_dev_info;
781 inode_init_once(&ei->vfs_inode);
782 /* Initialize mutex for freeze. */
783 mutex_init(&bdev->bd_fsfreeze_mutex);
786 static void bdev_evict_inode(struct inode *inode)
788 struct block_device *bdev = &BDEV_I(inode)->bdev;
789 truncate_inode_pages_final(&inode->i_data);
790 invalidate_inode_buffers(inode); /* is it needed here? */
792 spin_lock(&bdev_lock);
793 list_del_init(&bdev->bd_list);
794 spin_unlock(&bdev_lock);
795 /* Detach inode from wb early as bdi_put() may free bdi->wb */
796 inode_detach_wb(inode);
797 if (bdev->bd_bdi != &noop_backing_dev_info) {
798 bdi_put(bdev->bd_bdi);
799 bdev->bd_bdi = &noop_backing_dev_info;
803 static const struct super_operations bdev_sops = {
804 .statfs = simple_statfs,
805 .alloc_inode = bdev_alloc_inode,
806 .free_inode = bdev_free_inode,
807 .drop_inode = generic_delete_inode,
808 .evict_inode = bdev_evict_inode,
811 static int bd_init_fs_context(struct fs_context *fc)
813 struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
816 fc->s_iflags |= SB_I_CGROUPWB;
817 ctx->ops = &bdev_sops;
821 static struct file_system_type bd_type = {
823 .init_fs_context = bd_init_fs_context,
824 .kill_sb = kill_anon_super,
827 struct super_block *blockdev_superblock __read_mostly;
828 EXPORT_SYMBOL_GPL(blockdev_superblock);
830 void __init bdev_cache_init(void)
833 static struct vfsmount *bd_mnt;
835 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
836 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
837 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
839 err = register_filesystem(&bd_type);
841 panic("Cannot register bdev pseudo-fs");
842 bd_mnt = kern_mount(&bd_type);
844 panic("Cannot create bdev pseudo-fs");
845 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
849 * Most likely _very_ bad one - but then it's hardly critical for small
850 * /dev and can be fixed when somebody will need really large one.
851 * Keep in mind that it will be fed through icache hash function too.
853 static inline unsigned long hash(dev_t dev)
855 return MAJOR(dev)+MINOR(dev);
858 static int bdev_test(struct inode *inode, void *data)
860 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
863 static int bdev_set(struct inode *inode, void *data)
865 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
869 static LIST_HEAD(all_bdevs);
871 struct block_device *bdget(dev_t dev)
873 struct block_device *bdev;
876 inode = iget5_locked(blockdev_superblock, hash(dev),
877 bdev_test, bdev_set, &dev);
882 bdev = &BDEV_I(inode)->bdev;
884 if (inode->i_state & I_NEW) {
885 bdev->bd_contains = NULL;
886 bdev->bd_super = NULL;
887 bdev->bd_inode = inode;
888 bdev->bd_part_count = 0;
889 bdev->bd_invalidated = 0;
890 inode->i_mode = S_IFBLK;
892 inode->i_bdev = bdev;
893 inode->i_data.a_ops = &def_blk_aops;
894 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
895 spin_lock(&bdev_lock);
896 list_add(&bdev->bd_list, &all_bdevs);
897 spin_unlock(&bdev_lock);
898 unlock_new_inode(inode);
903 EXPORT_SYMBOL(bdget);
906 * bdgrab -- Grab a reference to an already referenced block device
907 * @bdev: Block device to grab a reference to.
909 struct block_device *bdgrab(struct block_device *bdev)
911 ihold(bdev->bd_inode);
914 EXPORT_SYMBOL(bdgrab);
916 long nr_blockdev_pages(void)
918 struct block_device *bdev;
920 spin_lock(&bdev_lock);
921 list_for_each_entry(bdev, &all_bdevs, bd_list) {
922 ret += bdev->bd_inode->i_mapping->nrpages;
924 spin_unlock(&bdev_lock);
928 void bdput(struct block_device *bdev)
930 iput(bdev->bd_inode);
933 EXPORT_SYMBOL(bdput);
935 static struct block_device *bd_acquire(struct inode *inode)
937 struct block_device *bdev;
939 spin_lock(&bdev_lock);
940 bdev = inode->i_bdev;
941 if (bdev && !inode_unhashed(bdev->bd_inode)) {
943 spin_unlock(&bdev_lock);
946 spin_unlock(&bdev_lock);
949 * i_bdev references block device inode that was already shut down
950 * (corresponding device got removed). Remove the reference and look
951 * up block device inode again just in case new device got
952 * reestablished under the same device number.
957 bdev = bdget(inode->i_rdev);
959 spin_lock(&bdev_lock);
960 if (!inode->i_bdev) {
962 * We take an additional reference to bd_inode,
963 * and it's released in clear_inode() of inode.
964 * So, we can access it via ->i_mapping always
968 inode->i_bdev = bdev;
969 inode->i_mapping = bdev->bd_inode->i_mapping;
971 spin_unlock(&bdev_lock);
976 /* Call when you free inode */
978 void bd_forget(struct inode *inode)
980 struct block_device *bdev = NULL;
982 spin_lock(&bdev_lock);
983 if (!sb_is_blkdev_sb(inode->i_sb))
984 bdev = inode->i_bdev;
985 inode->i_bdev = NULL;
986 inode->i_mapping = &inode->i_data;
987 spin_unlock(&bdev_lock);
994 * bd_may_claim - test whether a block device can be claimed
995 * @bdev: block device of interest
996 * @whole: whole block device containing @bdev, may equal @bdev
997 * @holder: holder trying to claim @bdev
999 * Test whether @bdev can be claimed by @holder.
1002 * spin_lock(&bdev_lock).
1005 * %true if @bdev can be claimed, %false otherwise.
1007 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1010 if (bdev->bd_holder == holder)
1011 return true; /* already a holder */
1012 else if (bdev->bd_holder != NULL)
1013 return false; /* held by someone else */
1014 else if (whole == bdev)
1015 return true; /* is a whole device which isn't held */
1017 else if (whole->bd_holder == bd_may_claim)
1018 return true; /* is a partition of a device that is being partitioned */
1019 else if (whole->bd_holder != NULL)
1020 return false; /* is a partition of a held device */
1022 return true; /* is a partition of an un-held device */
1026 * bd_prepare_to_claim - prepare to claim a block device
1027 * @bdev: block device of interest
1028 * @whole: the whole device containing @bdev, may equal @bdev
1029 * @holder: holder trying to claim @bdev
1031 * Prepare to claim @bdev. This function fails if @bdev is already
1032 * claimed by another holder and waits if another claiming is in
1033 * progress. This function doesn't actually claim. On successful
1034 * return, the caller has ownership of bd_claiming and bd_holder[s].
1037 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1038 * it multiple times.
1041 * 0 if @bdev can be claimed, -EBUSY otherwise.
1043 static int bd_prepare_to_claim(struct block_device *bdev,
1044 struct block_device *whole, void *holder)
1047 /* if someone else claimed, fail */
1048 if (!bd_may_claim(bdev, whole, holder))
1051 /* if claiming is already in progress, wait for it to finish */
1052 if (whole->bd_claiming) {
1053 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1056 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1057 spin_unlock(&bdev_lock);
1059 finish_wait(wq, &wait);
1060 spin_lock(&bdev_lock);
1068 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1070 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1075 * Now that we hold gendisk reference we make sure bdev we looked up is
1076 * not stale. If it is, it means device got removed and created before
1077 * we looked up gendisk and we fail open in such case. Associating
1078 * unhashed bdev with newly created gendisk could lead to two bdevs
1079 * (and thus two independent caches) being associated with one device
1082 if (inode_unhashed(bdev->bd_inode)) {
1083 put_disk_and_module(disk);
1090 * bd_start_claiming - start claiming a block device
1091 * @bdev: block device of interest
1092 * @holder: holder trying to claim @bdev
1094 * @bdev is about to be opened exclusively. Check @bdev can be opened
1095 * exclusively and mark that an exclusive open is in progress. Each
1096 * successful call to this function must be matched with a call to
1097 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1100 * This function is used to gain exclusive access to the block device
1101 * without actually causing other exclusive open attempts to fail. It
1102 * should be used when the open sequence itself requires exclusive
1103 * access but may subsequently fail.
1109 * Pointer to the block device containing @bdev on success, ERR_PTR()
1112 struct block_device *bd_start_claiming(struct block_device *bdev, void *holder)
1114 struct gendisk *disk;
1115 struct block_device *whole;
1121 * @bdev might not have been initialized properly yet, look up
1122 * and grab the outer block device the hard way.
1124 disk = bdev_get_gendisk(bdev, &partno);
1126 return ERR_PTR(-ENXIO);
1129 * Normally, @bdev should equal what's returned from bdget_disk()
1130 * if partno is 0; however, some drivers (floppy) use multiple
1131 * bdev's for the same physical device and @bdev may be one of the
1132 * aliases. Keep @bdev if partno is 0. This means claimer
1133 * tracking is broken for those devices but it has always been that
1137 whole = bdget_disk(disk, 0);
1139 whole = bdgrab(bdev);
1141 put_disk_and_module(disk);
1143 return ERR_PTR(-ENOMEM);
1145 /* prepare to claim, if successful, mark claiming in progress */
1146 spin_lock(&bdev_lock);
1148 err = bd_prepare_to_claim(bdev, whole, holder);
1150 whole->bd_claiming = holder;
1151 spin_unlock(&bdev_lock);
1154 spin_unlock(&bdev_lock);
1156 return ERR_PTR(err);
1159 EXPORT_SYMBOL(bd_start_claiming);
1161 static void bd_clear_claiming(struct block_device *whole, void *holder)
1163 lockdep_assert_held(&bdev_lock);
1164 /* tell others that we're done */
1165 BUG_ON(whole->bd_claiming != holder);
1166 whole->bd_claiming = NULL;
1167 wake_up_bit(&whole->bd_claiming, 0);
1171 * bd_finish_claiming - finish claiming of a block device
1172 * @bdev: block device of interest
1173 * @whole: whole block device (returned from bd_start_claiming())
1174 * @holder: holder that has claimed @bdev
1176 * Finish exclusive open of a block device. Mark the device as exlusively
1177 * open by the holder and wake up all waiters for exclusive open to finish.
1179 static void bd_finish_claiming(struct block_device *bdev,
1180 struct block_device *whole, void *holder)
1182 spin_lock(&bdev_lock);
1183 BUG_ON(!bd_may_claim(bdev, whole, holder));
1185 * Note that for a whole device bd_holders will be incremented twice,
1186 * and bd_holder will be set to bd_may_claim before being set to holder
1188 whole->bd_holders++;
1189 whole->bd_holder = bd_may_claim;
1191 bdev->bd_holder = holder;
1192 bd_clear_claiming(whole, holder);
1193 spin_unlock(&bdev_lock);
1197 * bd_abort_claiming - abort claiming of a block device
1198 * @bdev: block device of interest
1199 * @whole: whole block device (returned from bd_start_claiming())
1200 * @holder: holder that has claimed @bdev
1202 * Abort claiming of a block device when the exclusive open failed. This can be
1203 * also used when exclusive open is not actually desired and we just needed
1204 * to block other exclusive openers for a while.
1206 void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
1209 spin_lock(&bdev_lock);
1210 bd_clear_claiming(whole, holder);
1211 spin_unlock(&bdev_lock);
1213 EXPORT_SYMBOL(bd_abort_claiming);
1216 struct bd_holder_disk {
1217 struct list_head list;
1218 struct gendisk *disk;
1222 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1223 struct gendisk *disk)
1225 struct bd_holder_disk *holder;
1227 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1228 if (holder->disk == disk)
1233 static int add_symlink(struct kobject *from, struct kobject *to)
1235 return sysfs_create_link(from, to, kobject_name(to));
1238 static void del_symlink(struct kobject *from, struct kobject *to)
1240 sysfs_remove_link(from, kobject_name(to));
1244 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1245 * @bdev: the claimed slave bdev
1246 * @disk: the holding disk
1248 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1250 * This functions creates the following sysfs symlinks.
1252 * - from "slaves" directory of the holder @disk to the claimed @bdev
1253 * - from "holders" directory of the @bdev to the holder @disk
1255 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1256 * passed to bd_link_disk_holder(), then:
1258 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1259 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1261 * The caller must have claimed @bdev before calling this function and
1262 * ensure that both @bdev and @disk are valid during the creation and
1263 * lifetime of these symlinks.
1269 * 0 on success, -errno on failure.
1271 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1273 struct bd_holder_disk *holder;
1276 mutex_lock(&bdev->bd_mutex);
1278 WARN_ON_ONCE(!bdev->bd_holder);
1280 /* FIXME: remove the following once add_disk() handles errors */
1281 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1284 holder = bd_find_holder_disk(bdev, disk);
1290 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1296 INIT_LIST_HEAD(&holder->list);
1297 holder->disk = disk;
1300 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1304 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1308 * bdev could be deleted beneath us which would implicitly destroy
1309 * the holder directory. Hold on to it.
1311 kobject_get(bdev->bd_part->holder_dir);
1313 list_add(&holder->list, &bdev->bd_holder_disks);
1317 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1321 mutex_unlock(&bdev->bd_mutex);
1324 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1327 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1328 * @bdev: the calimed slave bdev
1329 * @disk: the holding disk
1331 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1336 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1338 struct bd_holder_disk *holder;
1340 mutex_lock(&bdev->bd_mutex);
1342 holder = bd_find_holder_disk(bdev, disk);
1344 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1345 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1346 del_symlink(bdev->bd_part->holder_dir,
1347 &disk_to_dev(disk)->kobj);
1348 kobject_put(bdev->bd_part->holder_dir);
1349 list_del_init(&holder->list);
1353 mutex_unlock(&bdev->bd_mutex);
1355 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1359 * flush_disk - invalidates all buffer-cache entries on a disk
1361 * @bdev: struct block device to be flushed
1362 * @kill_dirty: flag to guide handling of dirty inodes
1364 * Invalidates all buffer-cache entries on a disk. It should be called
1365 * when a disk has been changed -- either by a media change or online
1368 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1370 if (__invalidate_device(bdev, kill_dirty)) {
1371 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1372 "resized disk %s\n",
1373 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1375 bdev->bd_invalidated = 1;
1379 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1380 * @disk: struct gendisk to check
1381 * @bdev: struct bdev to adjust.
1382 * @verbose: if %true log a message about a size change if there is any
1384 * This routine checks to see if the bdev size does not match the disk size
1385 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1388 static void check_disk_size_change(struct gendisk *disk,
1389 struct block_device *bdev, bool verbose)
1391 loff_t disk_size, bdev_size;
1393 disk_size = (loff_t)get_capacity(disk) << 9;
1394 bdev_size = i_size_read(bdev->bd_inode);
1395 if (disk_size != bdev_size) {
1398 "%s: detected capacity change from %lld to %lld\n",
1399 disk->disk_name, bdev_size, disk_size);
1401 i_size_write(bdev->bd_inode, disk_size);
1402 if (bdev_size > disk_size)
1403 flush_disk(bdev, false);
1405 bdev->bd_invalidated = 0;
1409 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1410 * @disk: struct gendisk to be revalidated
1412 * This routine is a wrapper for lower-level driver's revalidate_disk
1413 * call-backs. It is used to do common pre and post operations needed
1414 * for all revalidate_disk operations.
1416 int revalidate_disk(struct gendisk *disk)
1420 if (disk->fops->revalidate_disk)
1421 ret = disk->fops->revalidate_disk(disk);
1424 * Hidden disks don't have associated bdev so there's no point in
1427 if (!(disk->flags & GENHD_FL_HIDDEN)) {
1428 struct block_device *bdev = bdget_disk(disk, 0);
1433 mutex_lock(&bdev->bd_mutex);
1434 check_disk_size_change(disk, bdev, ret == 0);
1435 mutex_unlock(&bdev->bd_mutex);
1440 EXPORT_SYMBOL(revalidate_disk);
1443 * This routine checks whether a removable media has been changed,
1444 * and invalidates all buffer-cache-entries in that case. This
1445 * is a relatively slow routine, so we have to try to minimize using
1446 * it. Thus it is called only upon a 'mount' or 'open'. This
1447 * is the best way of combining speed and utility, I think.
1448 * People changing diskettes in the middle of an operation deserve
1451 int check_disk_change(struct block_device *bdev)
1453 struct gendisk *disk = bdev->bd_disk;
1454 const struct block_device_operations *bdops = disk->fops;
1455 unsigned int events;
1457 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1458 DISK_EVENT_EJECT_REQUEST);
1459 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1462 flush_disk(bdev, true);
1463 if (bdops->revalidate_disk)
1464 bdops->revalidate_disk(bdev->bd_disk);
1468 EXPORT_SYMBOL(check_disk_change);
1470 void bd_set_size(struct block_device *bdev, loff_t size)
1472 inode_lock(bdev->bd_inode);
1473 i_size_write(bdev->bd_inode, size);
1474 inode_unlock(bdev->bd_inode);
1476 EXPORT_SYMBOL(bd_set_size);
1478 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1480 int bdev_disk_changed(struct block_device *bdev, bool invalidate)
1482 struct gendisk *disk = bdev->bd_disk;
1485 lockdep_assert_held(&bdev->bd_mutex);
1488 ret = blk_drop_partitions(bdev);
1493 * Historically we only set the capacity to zero for devices that
1494 * support partitions (independ of actually having partitions created).
1495 * Doing that is rather inconsistent, but changing it broke legacy
1496 * udisks polling for legacy ide-cdrom devices. Use the crude check
1497 * below to get the sane behavior for most device while not breaking
1498 * userspace for this particular setup.
1501 if (disk_part_scan_enabled(disk) ||
1502 !(disk->flags & GENHD_FL_REMOVABLE))
1503 set_capacity(disk, 0);
1505 if (disk->fops->revalidate_disk)
1506 disk->fops->revalidate_disk(disk);
1509 check_disk_size_change(disk, bdev, !invalidate);
1511 if (get_capacity(disk)) {
1512 ret = blk_add_partitions(disk, bdev);
1515 } else if (invalidate) {
1517 * Tell userspace that the media / partition table may have
1520 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
1526 * Only exported for for loop and dasd for historic reasons. Don't use in new
1529 EXPORT_SYMBOL_GPL(bdev_disk_changed);
1534 * mutex_lock(part->bd_mutex)
1535 * mutex_lock_nested(whole->bd_mutex, 1)
1538 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1540 struct gendisk *disk;
1544 bool first_open = false;
1546 if (mode & FMODE_READ)
1548 if (mode & FMODE_WRITE)
1551 * hooks: /n/, see "layering violations".
1554 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1562 disk = bdev_get_gendisk(bdev, &partno);
1566 disk_block_events(disk);
1567 mutex_lock_nested(&bdev->bd_mutex, for_part);
1568 if (!bdev->bd_openers) {
1570 bdev->bd_disk = disk;
1571 bdev->bd_contains = bdev;
1572 bdev->bd_partno = partno;
1576 bdev->bd_part = disk_get_part(disk, partno);
1581 if (disk->fops->open) {
1582 ret = disk->fops->open(bdev, mode);
1583 if (ret == -ERESTARTSYS) {
1584 /* Lost a race with 'disk' being
1585 * deleted, try again.
1588 disk_put_part(bdev->bd_part);
1589 bdev->bd_part = NULL;
1590 bdev->bd_disk = 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 if (bdev != bdev->bd_contains)
1668 __blkdev_put(bdev->bd_contains, mode, 1);
1669 bdev->bd_contains = NULL;
1671 mutex_unlock(&bdev->bd_mutex);
1672 disk_unblock_events(disk);
1673 put_disk_and_module(disk);
1680 * blkdev_get - open a block device
1681 * @bdev: block_device to open
1682 * @mode: FMODE_* mask
1683 * @holder: exclusive holder identifier
1685 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1686 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1687 * @holder is invalid. Exclusive opens may nest for the same @holder.
1689 * On success, the reference count of @bdev is unchanged. On failure,
1696 * 0 on success, -errno on failure.
1698 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1700 struct block_device *whole = NULL;
1703 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1705 if ((mode & FMODE_EXCL) && holder) {
1706 whole = bd_start_claiming(bdev, holder);
1707 if (IS_ERR(whole)) {
1709 return PTR_ERR(whole);
1713 res = __blkdev_get(bdev, mode, 0);
1716 struct gendisk *disk = whole->bd_disk;
1718 /* finish claiming */
1719 mutex_lock(&bdev->bd_mutex);
1721 bd_finish_claiming(bdev, whole, holder);
1723 bd_abort_claiming(bdev, whole, holder);
1725 * Block event polling for write claims if requested. Any
1726 * write holder makes the write_holder state stick until
1727 * all are released. This is good enough and tracking
1728 * individual writeable reference is too fragile given the
1729 * way @mode is used in blkdev_get/put().
1731 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1732 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1733 bdev->bd_write_holder = true;
1734 disk_block_events(disk);
1737 mutex_unlock(&bdev->bd_mutex);
1746 EXPORT_SYMBOL(blkdev_get);
1749 * blkdev_get_by_path - open a block device by name
1750 * @path: path to the block device to open
1751 * @mode: FMODE_* mask
1752 * @holder: exclusive holder identifier
1754 * Open the blockdevice described by the device file at @path. @mode
1755 * and @holder are identical to blkdev_get().
1757 * On success, the returned block_device has reference count of one.
1763 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1765 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1768 struct block_device *bdev;
1771 bdev = lookup_bdev(path);
1775 err = blkdev_get(bdev, mode, holder);
1777 return ERR_PTR(err);
1779 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1780 blkdev_put(bdev, mode);
1781 return ERR_PTR(-EACCES);
1786 EXPORT_SYMBOL(blkdev_get_by_path);
1789 * blkdev_get_by_dev - open a block device by device number
1790 * @dev: device number of block device to open
1791 * @mode: FMODE_* mask
1792 * @holder: exclusive holder identifier
1794 * Open the blockdevice described by device number @dev. @mode and
1795 * @holder are identical to blkdev_get().
1797 * Use it ONLY if you really do not have anything better - i.e. when
1798 * you are behind a truly sucky interface and all you are given is a
1799 * device number. _Never_ to be used for internal purposes. If you
1800 * ever need it - reconsider your API.
1802 * On success, the returned block_device has reference count of one.
1808 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1810 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1812 struct block_device *bdev;
1817 return ERR_PTR(-ENOMEM);
1819 err = blkdev_get(bdev, mode, holder);
1821 return ERR_PTR(err);
1825 EXPORT_SYMBOL(blkdev_get_by_dev);
1827 static int blkdev_open(struct inode * inode, struct file * filp)
1829 struct block_device *bdev;
1832 * Preserve backwards compatibility and allow large file access
1833 * even if userspace doesn't ask for it explicitly. Some mkfs
1834 * binary needs it. We might want to drop this workaround
1835 * during an unstable branch.
1837 filp->f_flags |= O_LARGEFILE;
1839 filp->f_mode |= FMODE_NOWAIT;
1841 if (filp->f_flags & O_NDELAY)
1842 filp->f_mode |= FMODE_NDELAY;
1843 if (filp->f_flags & O_EXCL)
1844 filp->f_mode |= FMODE_EXCL;
1845 if ((filp->f_flags & O_ACCMODE) == 3)
1846 filp->f_mode |= FMODE_WRITE_IOCTL;
1848 bdev = bd_acquire(inode);
1852 filp->f_mapping = bdev->bd_inode->i_mapping;
1853 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1855 return blkdev_get(bdev, filp->f_mode, filp);
1858 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1860 struct gendisk *disk = bdev->bd_disk;
1861 struct block_device *victim = NULL;
1864 * Sync early if it looks like we're the last one. If someone else
1865 * opens the block device between now and the decrement of bd_openers
1866 * then we did a sync that we didn't need to, but that's not the end
1867 * of the world and we want to avoid long (could be several minute)
1868 * syncs while holding the mutex.
1870 if (bdev->bd_openers == 1)
1871 sync_blockdev(bdev);
1873 mutex_lock_nested(&bdev->bd_mutex, for_part);
1875 bdev->bd_part_count--;
1877 if (!--bdev->bd_openers) {
1878 WARN_ON_ONCE(bdev->bd_holders);
1879 sync_blockdev(bdev);
1882 bdev_write_inode(bdev);
1884 if (bdev->bd_contains == bdev) {
1885 if (disk->fops->release)
1886 disk->fops->release(disk, mode);
1888 if (!bdev->bd_openers) {
1889 disk_put_part(bdev->bd_part);
1890 bdev->bd_part = NULL;
1891 bdev->bd_disk = NULL;
1892 if (bdev != bdev->bd_contains)
1893 victim = bdev->bd_contains;
1894 bdev->bd_contains = NULL;
1896 put_disk_and_module(disk);
1898 mutex_unlock(&bdev->bd_mutex);
1901 __blkdev_put(victim, mode, 1);
1904 void blkdev_put(struct block_device *bdev, fmode_t mode)
1906 mutex_lock(&bdev->bd_mutex);
1908 if (mode & FMODE_EXCL) {
1912 * Release a claim on the device. The holder fields
1913 * are protected with bdev_lock. bd_mutex is to
1914 * synchronize disk_holder unlinking.
1916 spin_lock(&bdev_lock);
1918 WARN_ON_ONCE(--bdev->bd_holders < 0);
1919 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1921 /* bd_contains might point to self, check in a separate step */
1922 if ((bdev_free = !bdev->bd_holders))
1923 bdev->bd_holder = NULL;
1924 if (!bdev->bd_contains->bd_holders)
1925 bdev->bd_contains->bd_holder = NULL;
1927 spin_unlock(&bdev_lock);
1930 * If this was the last claim, remove holder link and
1931 * unblock evpoll if it was a write holder.
1933 if (bdev_free && bdev->bd_write_holder) {
1934 disk_unblock_events(bdev->bd_disk);
1935 bdev->bd_write_holder = false;
1940 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1941 * event. This is to ensure detection of media removal commanded
1942 * from userland - e.g. eject(1).
1944 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1946 mutex_unlock(&bdev->bd_mutex);
1948 __blkdev_put(bdev, mode, 0);
1950 EXPORT_SYMBOL(blkdev_put);
1952 static int blkdev_close(struct inode * inode, struct file * filp)
1954 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1955 blkdev_put(bdev, filp->f_mode);
1959 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1961 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1962 fmode_t mode = file->f_mode;
1965 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1966 * to updated it before every ioctl.
1968 if (file->f_flags & O_NDELAY)
1969 mode |= FMODE_NDELAY;
1971 mode &= ~FMODE_NDELAY;
1973 return blkdev_ioctl(bdev, mode, cmd, arg);
1977 * Write data to the block device. Only intended for the block device itself
1978 * and the raw driver which basically is a fake block device.
1980 * Does not take i_mutex for the write and thus is not for general purpose
1983 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1985 struct file *file = iocb->ki_filp;
1986 struct inode *bd_inode = bdev_file_inode(file);
1987 loff_t size = i_size_read(bd_inode);
1988 struct blk_plug plug;
1991 if (bdev_read_only(I_BDEV(bd_inode)))
1994 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode))
1997 if (!iov_iter_count(from))
2000 if (iocb->ki_pos >= size)
2003 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
2006 iov_iter_truncate(from, size - iocb->ki_pos);
2008 blk_start_plug(&plug);
2009 ret = __generic_file_write_iter(iocb, from);
2011 ret = generic_write_sync(iocb, ret);
2012 blk_finish_plug(&plug);
2015 EXPORT_SYMBOL_GPL(blkdev_write_iter);
2017 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
2019 struct file *file = iocb->ki_filp;
2020 struct inode *bd_inode = bdev_file_inode(file);
2021 loff_t size = i_size_read(bd_inode);
2022 loff_t pos = iocb->ki_pos;
2028 iov_iter_truncate(to, size);
2029 return generic_file_read_iter(iocb, to);
2031 EXPORT_SYMBOL_GPL(blkdev_read_iter);
2034 * Try to release a page associated with block device when the system
2035 * is under memory pressure.
2037 static int blkdev_releasepage(struct page *page, gfp_t wait)
2039 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
2041 if (super && super->s_op->bdev_try_to_free_page)
2042 return super->s_op->bdev_try_to_free_page(super, page, wait);
2044 return try_to_free_buffers(page);
2047 static int blkdev_writepages(struct address_space *mapping,
2048 struct writeback_control *wbc)
2050 return generic_writepages(mapping, wbc);
2053 static const struct address_space_operations def_blk_aops = {
2054 .readpage = blkdev_readpage,
2055 .readahead = blkdev_readahead,
2056 .writepage = blkdev_writepage,
2057 .write_begin = blkdev_write_begin,
2058 .write_end = blkdev_write_end,
2059 .writepages = blkdev_writepages,
2060 .releasepage = blkdev_releasepage,
2061 .direct_IO = blkdev_direct_IO,
2062 .migratepage = buffer_migrate_page_norefs,
2063 .is_dirty_writeback = buffer_check_dirty_writeback,
2066 #define BLKDEV_FALLOC_FL_SUPPORTED \
2067 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2068 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2070 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2073 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2074 struct address_space *mapping;
2075 loff_t end = start + len - 1;
2079 /* Fail if we don't recognize the flags. */
2080 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2083 /* Don't go off the end of the device. */
2084 isize = i_size_read(bdev->bd_inode);
2088 if (mode & FALLOC_FL_KEEP_SIZE) {
2089 len = isize - start;
2090 end = start + len - 1;
2096 * Don't allow IO that isn't aligned to logical block size.
2098 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2101 /* Invalidate the page cache, including dirty pages. */
2102 mapping = bdev->bd_inode->i_mapping;
2103 truncate_inode_pages_range(mapping, start, end);
2106 case FALLOC_FL_ZERO_RANGE:
2107 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2108 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2109 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2111 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2112 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2113 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2115 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2116 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2126 * Invalidate again; if someone wandered in and dirtied a page,
2127 * the caller will be given -EBUSY. The third argument is
2128 * inclusive, so the rounding here is safe.
2130 return invalidate_inode_pages2_range(mapping,
2131 start >> PAGE_SHIFT,
2135 const struct file_operations def_blk_fops = {
2136 .open = blkdev_open,
2137 .release = blkdev_close,
2138 .llseek = block_llseek,
2139 .read_iter = blkdev_read_iter,
2140 .write_iter = blkdev_write_iter,
2141 .iopoll = blkdev_iopoll,
2142 .mmap = generic_file_mmap,
2143 .fsync = blkdev_fsync,
2144 .unlocked_ioctl = block_ioctl,
2145 #ifdef CONFIG_COMPAT
2146 .compat_ioctl = compat_blkdev_ioctl,
2148 .splice_read = generic_file_splice_read,
2149 .splice_write = iter_file_splice_write,
2150 .fallocate = blkdev_fallocate,
2154 * lookup_bdev - lookup a struct block_device by name
2155 * @pathname: special file representing the block device
2157 * Get a reference to the blockdevice at @pathname in the current
2158 * namespace if possible and return it. Return ERR_PTR(error)
2161 struct block_device *lookup_bdev(const char *pathname)
2163 struct block_device *bdev;
2164 struct inode *inode;
2168 if (!pathname || !*pathname)
2169 return ERR_PTR(-EINVAL);
2171 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2173 return ERR_PTR(error);
2175 inode = d_backing_inode(path.dentry);
2177 if (!S_ISBLK(inode->i_mode))
2180 if (!may_open_dev(&path))
2183 bdev = bd_acquire(inode);
2190 bdev = ERR_PTR(error);
2193 EXPORT_SYMBOL(lookup_bdev);
2195 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2197 struct super_block *sb = get_super(bdev);
2202 * no need to lock the super, get_super holds the
2203 * read mutex so the filesystem cannot go away
2204 * under us (->put_super runs with the write lock
2207 shrink_dcache_sb(sb);
2208 res = invalidate_inodes(sb, kill_dirty);
2211 invalidate_bdev(bdev);
2214 EXPORT_SYMBOL(__invalidate_device);
2216 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2218 struct inode *inode, *old_inode = NULL;
2220 spin_lock(&blockdev_superblock->s_inode_list_lock);
2221 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2222 struct address_space *mapping = inode->i_mapping;
2223 struct block_device *bdev;
2225 spin_lock(&inode->i_lock);
2226 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2227 mapping->nrpages == 0) {
2228 spin_unlock(&inode->i_lock);
2232 spin_unlock(&inode->i_lock);
2233 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2235 * We hold a reference to 'inode' so it couldn't have been
2236 * removed from s_inodes list while we dropped the
2237 * s_inode_list_lock We cannot iput the inode now as we can
2238 * be holding the last reference and we cannot iput it under
2239 * s_inode_list_lock. So we keep the reference and iput it
2244 bdev = I_BDEV(inode);
2246 mutex_lock(&bdev->bd_mutex);
2247 if (bdev->bd_openers)
2249 mutex_unlock(&bdev->bd_mutex);
2251 spin_lock(&blockdev_superblock->s_inode_list_lock);
2253 spin_unlock(&blockdev_superblock->s_inode_list_lock);