1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2016 - 2020 Christoph Hellwig
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/device_cgroup.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/backing-dev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/magic.h>
21 #include <linux/buffer_head.h>
22 #include <linux/swap.h>
23 #include <linux/pagevec.h>
24 #include <linux/writeback.h>
25 #include <linux/mpage.h>
26 #include <linux/mount.h>
27 #include <linux/pseudo_fs.h>
28 #include <linux/uio.h>
29 #include <linux/namei.h>
30 #include <linux/log2.h>
31 #include <linux/cleancache.h>
32 #include <linux/task_io_accounting_ops.h>
33 #include <linux/falloc.h>
34 #include <linux/part_stat.h>
35 #include <linux/uaccess.h>
36 #include <linux/suspend.h>
40 struct block_device bdev;
41 struct inode vfs_inode;
44 static const struct address_space_operations def_blk_aops;
46 static inline struct bdev_inode *BDEV_I(struct inode *inode)
48 return container_of(inode, struct bdev_inode, vfs_inode);
51 struct block_device *I_BDEV(struct inode *inode)
53 return &BDEV_I(inode)->bdev;
55 EXPORT_SYMBOL(I_BDEV);
57 static void bdev_write_inode(struct block_device *bdev)
59 struct inode *inode = bdev->bd_inode;
62 spin_lock(&inode->i_lock);
63 while (inode->i_state & I_DIRTY) {
64 spin_unlock(&inode->i_lock);
65 ret = write_inode_now(inode, true);
67 char name[BDEVNAME_SIZE];
68 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
69 "for block device %s (err=%d).\n",
70 bdevname(bdev, name), ret);
72 spin_lock(&inode->i_lock);
74 spin_unlock(&inode->i_lock);
77 /* Kill _all_ buffers and pagecache , dirty or not.. */
78 static void kill_bdev(struct block_device *bdev)
80 struct address_space *mapping = bdev->bd_inode->i_mapping;
82 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
86 truncate_inode_pages(mapping, 0);
89 /* Invalidate clean unused buffers and pagecache. */
90 void invalidate_bdev(struct block_device *bdev)
92 struct address_space *mapping = bdev->bd_inode->i_mapping;
94 if (mapping->nrpages) {
96 lru_add_drain_all(); /* make sure all lru add caches are flushed */
97 invalidate_mapping_pages(mapping, 0, -1);
99 /* 99% of the time, we don't need to flush the cleancache on the bdev.
100 * But, for the strange corners, lets be cautious
102 cleancache_invalidate_inode(mapping);
104 EXPORT_SYMBOL(invalidate_bdev);
107 * Drop all buffers & page cache for given bdev range. This function bails
108 * with error if bdev has other exclusive owner (such as filesystem).
110 int truncate_bdev_range(struct block_device *bdev, fmode_t mode,
111 loff_t lstart, loff_t lend)
114 * If we don't hold exclusive handle for the device, upgrade to it
115 * while we discard the buffer cache to avoid discarding buffers
116 * under live filesystem.
118 if (!(mode & FMODE_EXCL)) {
119 int err = bd_prepare_to_claim(bdev, truncate_bdev_range);
124 truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend);
125 if (!(mode & FMODE_EXCL))
126 bd_abort_claiming(bdev, truncate_bdev_range);
131 * Someone else has handle exclusively open. Try invalidating instead.
132 * The 'end' argument is inclusive so the rounding is safe.
134 return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
135 lstart >> PAGE_SHIFT,
139 static void set_init_blocksize(struct block_device *bdev)
141 unsigned int bsize = bdev_logical_block_size(bdev);
142 loff_t size = i_size_read(bdev->bd_inode);
144 while (bsize < PAGE_SIZE) {
149 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
152 int set_blocksize(struct block_device *bdev, int size)
154 /* Size must be a power of two, and between 512 and PAGE_SIZE */
155 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
158 /* Size cannot be smaller than the size supported by the device */
159 if (size < bdev_logical_block_size(bdev))
162 /* Don't change the size if it is same as current */
163 if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
165 bdev->bd_inode->i_blkbits = blksize_bits(size);
171 EXPORT_SYMBOL(set_blocksize);
173 int sb_set_blocksize(struct super_block *sb, int size)
175 if (set_blocksize(sb->s_bdev, size))
177 /* If we get here, we know size is power of two
178 * and it's value is between 512 and PAGE_SIZE */
179 sb->s_blocksize = size;
180 sb->s_blocksize_bits = blksize_bits(size);
181 return sb->s_blocksize;
184 EXPORT_SYMBOL(sb_set_blocksize);
186 int sb_min_blocksize(struct super_block *sb, int size)
188 int minsize = bdev_logical_block_size(sb->s_bdev);
191 return sb_set_blocksize(sb, size);
194 EXPORT_SYMBOL(sb_min_blocksize);
197 blkdev_get_block(struct inode *inode, sector_t iblock,
198 struct buffer_head *bh, int create)
200 bh->b_bdev = I_BDEV(inode);
201 bh->b_blocknr = iblock;
202 set_buffer_mapped(bh);
206 static struct inode *bdev_file_inode(struct file *file)
208 return file->f_mapping->host;
211 static unsigned int dio_bio_write_op(struct kiocb *iocb)
213 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
215 /* avoid the need for a I/O completion work item */
216 if (iocb->ki_flags & IOCB_DSYNC)
221 #define DIO_INLINE_BIO_VECS 4
223 static void blkdev_bio_end_io_simple(struct bio *bio)
225 struct task_struct *waiter = bio->bi_private;
227 WRITE_ONCE(bio->bi_private, NULL);
228 blk_wake_io_task(waiter);
232 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
233 unsigned int nr_pages)
235 struct file *file = iocb->ki_filp;
236 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
237 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
238 loff_t pos = iocb->ki_pos;
239 bool should_dirty = false;
244 if ((pos | iov_iter_alignment(iter)) &
245 (bdev_logical_block_size(bdev) - 1))
248 if (nr_pages <= DIO_INLINE_BIO_VECS)
251 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
257 bio_init(&bio, vecs, nr_pages);
258 bio_set_dev(&bio, bdev);
259 bio.bi_iter.bi_sector = pos >> 9;
260 bio.bi_write_hint = iocb->ki_hint;
261 bio.bi_private = current;
262 bio.bi_end_io = blkdev_bio_end_io_simple;
263 bio.bi_ioprio = iocb->ki_ioprio;
265 ret = bio_iov_iter_get_pages(&bio, iter);
268 ret = bio.bi_iter.bi_size;
270 if (iov_iter_rw(iter) == READ) {
271 bio.bi_opf = REQ_OP_READ;
272 if (iter_is_iovec(iter))
275 bio.bi_opf = dio_bio_write_op(iocb);
276 task_io_account_write(ret);
278 if (iocb->ki_flags & IOCB_HIPRI)
279 bio_set_polled(&bio, iocb);
281 qc = submit_bio(&bio);
283 set_current_state(TASK_UNINTERRUPTIBLE);
284 if (!READ_ONCE(bio.bi_private))
286 if (!(iocb->ki_flags & IOCB_HIPRI) ||
287 !blk_poll(bdev_get_queue(bdev), qc, true))
290 __set_current_state(TASK_RUNNING);
292 bio_release_pages(&bio, should_dirty);
293 if (unlikely(bio.bi_status))
294 ret = blk_status_to_errno(bio.bi_status);
297 if (vecs != inline_vecs)
308 struct task_struct *waiter;
313 bool should_dirty : 1;
318 static struct bio_set blkdev_dio_pool;
320 static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
322 struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
323 struct request_queue *q = bdev_get_queue(bdev);
325 return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
328 static void blkdev_bio_end_io(struct bio *bio)
330 struct blkdev_dio *dio = bio->bi_private;
331 bool should_dirty = dio->should_dirty;
333 if (bio->bi_status && !dio->bio.bi_status)
334 dio->bio.bi_status = bio->bi_status;
336 if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
338 struct kiocb *iocb = dio->iocb;
341 if (likely(!dio->bio.bi_status)) {
345 ret = blk_status_to_errno(dio->bio.bi_status);
348 dio->iocb->ki_complete(iocb, ret, 0);
352 struct task_struct *waiter = dio->waiter;
354 WRITE_ONCE(dio->waiter, NULL);
355 blk_wake_io_task(waiter);
360 bio_check_pages_dirty(bio);
362 bio_release_pages(bio, false);
367 static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
368 unsigned int nr_pages)
370 struct file *file = iocb->ki_filp;
371 struct inode *inode = bdev_file_inode(file);
372 struct block_device *bdev = I_BDEV(inode);
373 struct blk_plug plug;
374 struct blkdev_dio *dio;
376 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
377 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
378 loff_t pos = iocb->ki_pos;
379 blk_qc_t qc = BLK_QC_T_NONE;
382 if ((pos | iov_iter_alignment(iter)) &
383 (bdev_logical_block_size(bdev) - 1))
386 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
388 dio = container_of(bio, struct blkdev_dio, bio);
389 dio->is_sync = is_sync = is_sync_kiocb(iocb);
391 dio->waiter = current;
398 dio->multi_bio = false;
399 dio->should_dirty = is_read && iter_is_iovec(iter);
402 * Don't plug for HIPRI/polled IO, as those should go straight
406 blk_start_plug(&plug);
409 bio_set_dev(bio, bdev);
410 bio->bi_iter.bi_sector = pos >> 9;
411 bio->bi_write_hint = iocb->ki_hint;
412 bio->bi_private = dio;
413 bio->bi_end_io = blkdev_bio_end_io;
414 bio->bi_ioprio = iocb->ki_ioprio;
416 ret = bio_iov_iter_get_pages(bio, iter);
418 bio->bi_status = BLK_STS_IOERR;
424 bio->bi_opf = REQ_OP_READ;
425 if (dio->should_dirty)
426 bio_set_pages_dirty(bio);
428 bio->bi_opf = dio_bio_write_op(iocb);
429 task_io_account_write(bio->bi_iter.bi_size);
432 dio->size += bio->bi_iter.bi_size;
433 pos += bio->bi_iter.bi_size;
435 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS);
439 if (iocb->ki_flags & IOCB_HIPRI) {
440 bio_set_polled(bio, iocb);
444 qc = submit_bio(bio);
447 WRITE_ONCE(iocb->ki_cookie, qc);
451 if (!dio->multi_bio) {
453 * AIO needs an extra reference to ensure the dio
454 * structure which is embedded into the first bio
459 dio->multi_bio = true;
460 atomic_set(&dio->ref, 2);
462 atomic_inc(&dio->ref);
466 bio = bio_alloc(GFP_KERNEL, nr_pages);
470 blk_finish_plug(&plug);
476 set_current_state(TASK_UNINTERRUPTIBLE);
477 if (!READ_ONCE(dio->waiter))
480 if (!(iocb->ki_flags & IOCB_HIPRI) ||
481 !blk_poll(bdev_get_queue(bdev), qc, true))
484 __set_current_state(TASK_RUNNING);
487 ret = blk_status_to_errno(dio->bio.bi_status);
496 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
498 unsigned int nr_pages;
500 if (!iov_iter_count(iter))
503 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS + 1);
504 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_VECS)
505 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
507 return __blkdev_direct_IO(iocb, iter, bio_max_segs(nr_pages));
510 static __init int blkdev_init(void)
512 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
514 module_init(blkdev_init);
516 int __sync_blockdev(struct block_device *bdev, int wait)
521 return filemap_flush(bdev->bd_inode->i_mapping);
522 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
526 * Write out and wait upon all the dirty data associated with a block
527 * device via its mapping. Does not take the superblock lock.
529 int sync_blockdev(struct block_device *bdev)
531 return __sync_blockdev(bdev, 1);
533 EXPORT_SYMBOL(sync_blockdev);
536 * Write out and wait upon all dirty data associated with this
537 * device. Filesystem data as well as the underlying block
538 * device. Takes the superblock lock.
540 int fsync_bdev(struct block_device *bdev)
542 struct super_block *sb = get_super(bdev);
544 int res = sync_filesystem(sb);
548 return sync_blockdev(bdev);
550 EXPORT_SYMBOL(fsync_bdev);
553 * freeze_bdev -- lock a filesystem and force it into a consistent state
554 * @bdev: blockdevice to lock
556 * If a superblock is found on this device, we take the s_umount semaphore
557 * on it to make sure nobody unmounts until the snapshot creation is done.
558 * The reference counter (bd_fsfreeze_count) guarantees that only the last
559 * unfreeze process can unfreeze the frozen filesystem actually when multiple
560 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
561 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
564 int freeze_bdev(struct block_device *bdev)
566 struct super_block *sb;
569 mutex_lock(&bdev->bd_fsfreeze_mutex);
570 if (++bdev->bd_fsfreeze_count > 1)
573 sb = get_active_super(bdev);
576 if (sb->s_op->freeze_super)
577 error = sb->s_op->freeze_super(sb);
579 error = freeze_super(sb);
580 deactivate_super(sb);
583 bdev->bd_fsfreeze_count--;
586 bdev->bd_fsfreeze_sb = sb;
591 mutex_unlock(&bdev->bd_fsfreeze_mutex);
594 EXPORT_SYMBOL(freeze_bdev);
597 * thaw_bdev -- unlock filesystem
598 * @bdev: blockdevice to unlock
600 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
602 int thaw_bdev(struct block_device *bdev)
604 struct super_block *sb;
607 mutex_lock(&bdev->bd_fsfreeze_mutex);
608 if (!bdev->bd_fsfreeze_count)
612 if (--bdev->bd_fsfreeze_count > 0)
615 sb = bdev->bd_fsfreeze_sb;
619 if (sb->s_op->thaw_super)
620 error = sb->s_op->thaw_super(sb);
622 error = thaw_super(sb);
624 bdev->bd_fsfreeze_count++;
626 bdev->bd_fsfreeze_sb = NULL;
628 mutex_unlock(&bdev->bd_fsfreeze_mutex);
631 EXPORT_SYMBOL(thaw_bdev);
633 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
635 return block_write_full_page(page, blkdev_get_block, wbc);
638 static int blkdev_readpage(struct file * file, struct page * page)
640 return block_read_full_page(page, blkdev_get_block);
643 static void blkdev_readahead(struct readahead_control *rac)
645 mpage_readahead(rac, blkdev_get_block);
648 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
649 loff_t pos, unsigned len, unsigned flags,
650 struct page **pagep, void **fsdata)
652 return block_write_begin(mapping, pos, len, flags, pagep,
656 static int blkdev_write_end(struct file *file, struct address_space *mapping,
657 loff_t pos, unsigned len, unsigned copied,
658 struct page *page, void *fsdata)
661 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
671 * for a block special file file_inode(file)->i_size is zero
672 * so we compute the size by hand (just as in block_read/write above)
674 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
676 struct inode *bd_inode = bdev_file_inode(file);
679 inode_lock(bd_inode);
680 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
681 inode_unlock(bd_inode);
685 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
687 struct inode *bd_inode = bdev_file_inode(filp);
688 struct block_device *bdev = I_BDEV(bd_inode);
691 error = file_write_and_wait_range(filp, start, end);
696 * There is no need to serialise calls to blkdev_issue_flush with
697 * i_mutex and doing so causes performance issues with concurrent
698 * O_SYNC writers to a block device.
700 error = blkdev_issue_flush(bdev);
701 if (error == -EOPNOTSUPP)
706 EXPORT_SYMBOL(blkdev_fsync);
709 * bdev_read_page() - Start reading a page from a block device
710 * @bdev: The device to read the page from
711 * @sector: The offset on the device to read the page to (need not be aligned)
712 * @page: The page to read
714 * On entry, the page should be locked. It will be unlocked when the page
715 * has been read. If the block driver implements rw_page synchronously,
716 * that will be true on exit from this function, but it need not be.
718 * Errors returned by this function are usually "soft", eg out of memory, or
719 * queue full; callers should try a different route to read 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_read_page(struct block_device *bdev, sector_t sector,
727 const struct block_device_operations *ops = bdev->bd_disk->fops;
728 int result = -EOPNOTSUPP;
730 if (!ops->rw_page || bdev_get_integrity(bdev))
733 result = blk_queue_enter(bdev->bd_disk->queue, 0);
736 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
738 blk_queue_exit(bdev->bd_disk->queue);
743 * bdev_write_page() - Start writing a page to a block device
744 * @bdev: The device to write the page to
745 * @sector: The offset on the device to write the page to (need not be aligned)
746 * @page: The page to write
747 * @wbc: The writeback_control for the write
749 * On entry, the page should be locked and not currently under writeback.
750 * On exit, if the write started successfully, the page will be unlocked and
751 * under writeback. If the write failed already (eg the driver failed to
752 * queue the page to the device), the page will still be locked. If the
753 * caller is a ->writepage implementation, it will need to unlock the page.
755 * Errors returned by this function are usually "soft", eg out of memory, or
756 * queue full; callers should try a different route to write this page rather
757 * than propagate an error back up the stack.
759 * Return: negative errno if an error occurs, 0 if submission was successful.
761 int bdev_write_page(struct block_device *bdev, sector_t sector,
762 struct page *page, struct writeback_control *wbc)
765 const struct block_device_operations *ops = bdev->bd_disk->fops;
767 if (!ops->rw_page || bdev_get_integrity(bdev))
769 result = blk_queue_enter(bdev->bd_disk->queue, 0);
773 set_page_writeback(page);
774 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
777 end_page_writeback(page);
779 clean_page_buffers(page);
782 blk_queue_exit(bdev->bd_disk->queue);
790 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
791 static struct kmem_cache * bdev_cachep __read_mostly;
793 static struct inode *bdev_alloc_inode(struct super_block *sb)
795 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
799 memset(&ei->bdev, 0, sizeof(ei->bdev));
800 ei->bdev.bd_bdi = &noop_backing_dev_info;
801 return &ei->vfs_inode;
804 static void bdev_free_inode(struct inode *inode)
806 struct block_device *bdev = I_BDEV(inode);
808 free_percpu(bdev->bd_stats);
809 kfree(bdev->bd_meta_info);
811 kmem_cache_free(bdev_cachep, BDEV_I(inode));
814 static void init_once(void *data)
816 struct bdev_inode *ei = data;
818 inode_init_once(&ei->vfs_inode);
821 static void bdev_evict_inode(struct inode *inode)
823 struct block_device *bdev = &BDEV_I(inode)->bdev;
824 truncate_inode_pages_final(&inode->i_data);
825 invalidate_inode_buffers(inode); /* is it needed here? */
827 /* Detach inode from wb early as bdi_put() may free bdi->wb */
828 inode_detach_wb(inode);
829 if (bdev->bd_bdi != &noop_backing_dev_info) {
830 bdi_put(bdev->bd_bdi);
831 bdev->bd_bdi = &noop_backing_dev_info;
835 static const struct super_operations bdev_sops = {
836 .statfs = simple_statfs,
837 .alloc_inode = bdev_alloc_inode,
838 .free_inode = bdev_free_inode,
839 .drop_inode = generic_delete_inode,
840 .evict_inode = bdev_evict_inode,
843 static int bd_init_fs_context(struct fs_context *fc)
845 struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
848 fc->s_iflags |= SB_I_CGROUPWB;
849 ctx->ops = &bdev_sops;
853 static struct file_system_type bd_type = {
855 .init_fs_context = bd_init_fs_context,
856 .kill_sb = kill_anon_super,
859 struct super_block *blockdev_superblock __read_mostly;
860 EXPORT_SYMBOL_GPL(blockdev_superblock);
862 void __init bdev_cache_init(void)
865 static struct vfsmount *bd_mnt;
867 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
868 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
869 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
871 err = register_filesystem(&bd_type);
873 panic("Cannot register bdev pseudo-fs");
874 bd_mnt = kern_mount(&bd_type);
876 panic("Cannot create bdev pseudo-fs");
877 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
880 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
882 struct block_device *bdev;
885 inode = new_inode(blockdev_superblock);
888 inode->i_mode = S_IFBLK;
890 inode->i_data.a_ops = &def_blk_aops;
891 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
893 bdev = I_BDEV(inode);
894 mutex_init(&bdev->bd_mutex);
895 mutex_init(&bdev->bd_fsfreeze_mutex);
896 spin_lock_init(&bdev->bd_size_lock);
897 bdev->bd_disk = disk;
898 bdev->bd_partno = partno;
899 bdev->bd_inode = inode;
901 INIT_LIST_HEAD(&bdev->bd_holder_disks);
903 bdev->bd_stats = alloc_percpu(struct disk_stats);
904 if (!bdev->bd_stats) {
911 void bdev_add(struct block_device *bdev, dev_t dev)
914 bdev->bd_inode->i_rdev = dev;
915 bdev->bd_inode->i_ino = dev;
916 insert_inode_hash(bdev->bd_inode);
919 static struct block_device *bdget(dev_t dev)
923 inode = ilookup(blockdev_superblock, dev);
926 return &BDEV_I(inode)->bdev;
930 * bdgrab -- Grab a reference to an already referenced block device
931 * @bdev: Block device to grab a reference to.
933 * Returns the block_device with an additional reference when successful,
934 * or NULL if the inode is already beeing freed.
936 struct block_device *bdgrab(struct block_device *bdev)
938 if (!igrab(bdev->bd_inode))
942 EXPORT_SYMBOL(bdgrab);
944 long nr_blockdev_pages(void)
949 spin_lock(&blockdev_superblock->s_inode_list_lock);
950 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
951 ret += inode->i_mapping->nrpages;
952 spin_unlock(&blockdev_superblock->s_inode_list_lock);
957 void bdput(struct block_device *bdev)
959 iput(bdev->bd_inode);
961 EXPORT_SYMBOL(bdput);
964 * bd_may_claim - test whether a block device can be claimed
965 * @bdev: block device of interest
966 * @whole: whole block device containing @bdev, may equal @bdev
967 * @holder: holder trying to claim @bdev
969 * Test whether @bdev can be claimed by @holder.
972 * spin_lock(&bdev_lock).
975 * %true if @bdev can be claimed, %false otherwise.
977 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
980 if (bdev->bd_holder == holder)
981 return true; /* already a holder */
982 else if (bdev->bd_holder != NULL)
983 return false; /* held by someone else */
984 else if (whole == bdev)
985 return true; /* is a whole device which isn't held */
987 else if (whole->bd_holder == bd_may_claim)
988 return true; /* is a partition of a device that is being partitioned */
989 else if (whole->bd_holder != NULL)
990 return false; /* is a partition of a held device */
992 return true; /* is a partition of an un-held device */
996 * bd_prepare_to_claim - claim a block device
997 * @bdev: block device of interest
998 * @holder: holder trying to claim @bdev
1000 * Claim @bdev. This function fails if @bdev is already claimed by another
1001 * holder and waits if another claiming is in progress. return, the caller
1002 * has ownership of bd_claiming and bd_holder[s].
1005 * 0 if @bdev can be claimed, -EBUSY otherwise.
1007 int bd_prepare_to_claim(struct block_device *bdev, void *holder)
1009 struct block_device *whole = bdev_whole(bdev);
1011 if (WARN_ON_ONCE(!holder))
1014 spin_lock(&bdev_lock);
1015 /* if someone else claimed, fail */
1016 if (!bd_may_claim(bdev, whole, holder)) {
1017 spin_unlock(&bdev_lock);
1021 /* if claiming is already in progress, wait for it to finish */
1022 if (whole->bd_claiming) {
1023 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1026 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1027 spin_unlock(&bdev_lock);
1029 finish_wait(wq, &wait);
1034 whole->bd_claiming = holder;
1035 spin_unlock(&bdev_lock);
1038 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
1040 static void bd_clear_claiming(struct block_device *whole, void *holder)
1042 lockdep_assert_held(&bdev_lock);
1043 /* tell others that we're done */
1044 BUG_ON(whole->bd_claiming != holder);
1045 whole->bd_claiming = NULL;
1046 wake_up_bit(&whole->bd_claiming, 0);
1050 * bd_finish_claiming - finish claiming of a block device
1051 * @bdev: block device of interest
1052 * @holder: holder that has claimed @bdev
1054 * Finish exclusive open of a block device. Mark the device as exlusively
1055 * open by the holder and wake up all waiters for exclusive open to finish.
1057 static void bd_finish_claiming(struct block_device *bdev, void *holder)
1059 struct block_device *whole = bdev_whole(bdev);
1061 spin_lock(&bdev_lock);
1062 BUG_ON(!bd_may_claim(bdev, whole, holder));
1064 * Note that for a whole device bd_holders will be incremented twice,
1065 * and bd_holder will be set to bd_may_claim before being set to holder
1067 whole->bd_holders++;
1068 whole->bd_holder = bd_may_claim;
1070 bdev->bd_holder = holder;
1071 bd_clear_claiming(whole, holder);
1072 spin_unlock(&bdev_lock);
1076 * bd_abort_claiming - abort claiming of a block device
1077 * @bdev: block device of interest
1078 * @holder: holder that has claimed @bdev
1080 * Abort claiming of a block device when the exclusive open failed. This can be
1081 * also used when exclusive open is not actually desired and we just needed
1082 * to block other exclusive openers for a while.
1084 void bd_abort_claiming(struct block_device *bdev, void *holder)
1086 spin_lock(&bdev_lock);
1087 bd_clear_claiming(bdev_whole(bdev), holder);
1088 spin_unlock(&bdev_lock);
1090 EXPORT_SYMBOL(bd_abort_claiming);
1093 struct bd_holder_disk {
1094 struct list_head list;
1095 struct gendisk *disk;
1099 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1100 struct gendisk *disk)
1102 struct bd_holder_disk *holder;
1104 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1105 if (holder->disk == disk)
1110 static int add_symlink(struct kobject *from, struct kobject *to)
1112 return sysfs_create_link(from, to, kobject_name(to));
1115 static void del_symlink(struct kobject *from, struct kobject *to)
1117 sysfs_remove_link(from, kobject_name(to));
1121 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1122 * @bdev: the claimed slave bdev
1123 * @disk: the holding disk
1125 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1127 * This functions creates the following sysfs symlinks.
1129 * - from "slaves" directory of the holder @disk to the claimed @bdev
1130 * - from "holders" directory of the @bdev to the holder @disk
1132 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1133 * passed to bd_link_disk_holder(), then:
1135 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1136 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1138 * The caller must have claimed @bdev before calling this function and
1139 * ensure that both @bdev and @disk are valid during the creation and
1140 * lifetime of these symlinks.
1146 * 0 on success, -errno on failure.
1148 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1150 struct bd_holder_disk *holder;
1153 mutex_lock(&bdev->bd_mutex);
1155 WARN_ON_ONCE(!bdev->bd_holder);
1157 /* FIXME: remove the following once add_disk() handles errors */
1158 if (WARN_ON(!disk->slave_dir || !bdev->bd_holder_dir))
1161 holder = bd_find_holder_disk(bdev, disk);
1167 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1173 INIT_LIST_HEAD(&holder->list);
1174 holder->disk = disk;
1177 ret = add_symlink(disk->slave_dir, bdev_kobj(bdev));
1181 ret = add_symlink(bdev->bd_holder_dir, &disk_to_dev(disk)->kobj);
1185 * bdev could be deleted beneath us which would implicitly destroy
1186 * the holder directory. Hold on to it.
1188 kobject_get(bdev->bd_holder_dir);
1190 list_add(&holder->list, &bdev->bd_holder_disks);
1194 del_symlink(disk->slave_dir, bdev_kobj(bdev));
1198 mutex_unlock(&bdev->bd_mutex);
1201 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1204 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1205 * @bdev: the calimed slave bdev
1206 * @disk: the holding disk
1208 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1213 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1215 struct bd_holder_disk *holder;
1217 mutex_lock(&bdev->bd_mutex);
1219 holder = bd_find_holder_disk(bdev, disk);
1221 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1222 del_symlink(disk->slave_dir, bdev_kobj(bdev));
1223 del_symlink(bdev->bd_holder_dir, &disk_to_dev(disk)->kobj);
1224 kobject_put(bdev->bd_holder_dir);
1225 list_del_init(&holder->list);
1229 mutex_unlock(&bdev->bd_mutex);
1231 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1234 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1236 int bdev_disk_changed(struct block_device *bdev, bool invalidate)
1238 struct gendisk *disk = bdev->bd_disk;
1241 lockdep_assert_held(&bdev->bd_mutex);
1243 clear_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1246 ret = blk_drop_partitions(bdev);
1251 * Historically we only set the capacity to zero for devices that
1252 * support partitions (independ of actually having partitions created).
1253 * Doing that is rather inconsistent, but changing it broke legacy
1254 * udisks polling for legacy ide-cdrom devices. Use the crude check
1255 * below to get the sane behavior for most device while not breaking
1256 * userspace for this particular setup.
1259 if (disk_part_scan_enabled(disk) ||
1260 !(disk->flags & GENHD_FL_REMOVABLE))
1261 set_capacity(disk, 0);
1263 if (disk->fops->revalidate_disk)
1264 disk->fops->revalidate_disk(disk);
1267 if (get_capacity(disk)) {
1268 ret = blk_add_partitions(disk, bdev);
1271 } else if (invalidate) {
1273 * Tell userspace that the media / partition table may have
1276 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
1282 * Only exported for loop and dasd for historic reasons. Don't use in new
1285 EXPORT_SYMBOL_GPL(bdev_disk_changed);
1290 * mutex_lock(part->bd_mutex)
1291 * mutex_lock_nested(whole->bd_mutex, 1)
1293 static int __blkdev_get(struct block_device *bdev, fmode_t mode)
1295 struct gendisk *disk = bdev->bd_disk;
1298 if (!bdev->bd_openers) {
1299 if (!bdev_is_partition(bdev)) {
1301 if (disk->fops->open)
1302 ret = disk->fops->open(bdev, mode);
1305 set_init_blocksize(bdev);
1308 * If the device is invalidated, rescan partition
1309 * if open succeeded or failed with -ENOMEDIUM.
1310 * The latter is necessary to prevent ghost
1311 * partitions on a removed medium.
1313 if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
1314 (!ret || ret == -ENOMEDIUM))
1315 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1320 struct block_device *whole = bdgrab(disk->part0);
1322 mutex_lock_nested(&whole->bd_mutex, 1);
1323 ret = __blkdev_get(whole, mode);
1325 mutex_unlock(&whole->bd_mutex);
1329 whole->bd_part_count++;
1330 mutex_unlock(&whole->bd_mutex);
1332 if (!(disk->flags & GENHD_FL_UP) ||
1333 !bdev_nr_sectors(bdev)) {
1334 __blkdev_put(whole, mode, 1);
1338 set_init_blocksize(bdev);
1341 if (bdev->bd_bdi == &noop_backing_dev_info)
1342 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1344 if (!bdev_is_partition(bdev)) {
1345 if (bdev->bd_disk->fops->open)
1346 ret = bdev->bd_disk->fops->open(bdev, mode);
1347 /* the same as first opener case, read comment there */
1348 if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
1349 (!ret || ret == -ENOMEDIUM))
1350 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1359 struct block_device *blkdev_get_no_open(dev_t dev)
1361 struct block_device *bdev;
1362 struct gendisk *disk;
1364 down_read(&bdev_lookup_sem);
1367 up_read(&bdev_lookup_sem);
1368 blk_request_module(dev);
1369 down_read(&bdev_lookup_sem);
1376 disk = bdev->bd_disk;
1377 if (!kobject_get_unless_zero(&disk_to_dev(disk)->kobj))
1379 if ((disk->flags & (GENHD_FL_UP | GENHD_FL_HIDDEN)) != GENHD_FL_UP)
1381 if (!try_module_get(bdev->bd_disk->fops->owner))
1383 up_read(&bdev_lookup_sem);
1390 up_read(&bdev_lookup_sem);
1394 void blkdev_put_no_open(struct block_device *bdev)
1396 module_put(bdev->bd_disk->fops->owner);
1397 put_disk(bdev->bd_disk);
1402 * blkdev_get_by_dev - open a block device by device number
1403 * @dev: device number of block device to open
1404 * @mode: FMODE_* mask
1405 * @holder: exclusive holder identifier
1407 * Open the block device described by device number @dev. If @mode includes
1408 * %FMODE_EXCL, the block device is opened with exclusive access. Specifying
1409 * %FMODE_EXCL with a %NULL @holder is invalid. Exclusive opens may nest for
1412 * Use this interface ONLY if you really do not have anything better - i.e. when
1413 * you are behind a truly sucky interface and all you are given is a device
1414 * number. Everything else should use blkdev_get_by_path().
1420 * Reference to the block_device on success, ERR_PTR(-errno) on failure.
1422 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1424 bool unblock_events = true;
1425 struct block_device *bdev;
1426 struct gendisk *disk;
1429 ret = devcgroup_check_permission(DEVCG_DEV_BLOCK,
1430 MAJOR(dev), MINOR(dev),
1431 ((mode & FMODE_READ) ? DEVCG_ACC_READ : 0) |
1432 ((mode & FMODE_WRITE) ? DEVCG_ACC_WRITE : 0));
1434 return ERR_PTR(ret);
1437 * If we lost a race with 'disk' being deleted, try again. See md.c.
1440 bdev = blkdev_get_no_open(dev);
1442 return ERR_PTR(-ENXIO);
1443 disk = bdev->bd_disk;
1445 if (mode & FMODE_EXCL) {
1446 ret = bd_prepare_to_claim(bdev, holder);
1451 disk_block_events(disk);
1453 mutex_lock(&bdev->bd_mutex);
1454 ret =__blkdev_get(bdev, mode);
1456 goto abort_claiming;
1457 if (mode & FMODE_EXCL) {
1458 bd_finish_claiming(bdev, holder);
1461 * Block event polling for write claims if requested. Any write
1462 * holder makes the write_holder state stick until all are
1463 * released. This is good enough and tracking individual
1464 * writeable reference is too fragile given the way @mode is
1465 * used in blkdev_get/put().
1467 if ((mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1468 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1469 bdev->bd_write_holder = true;
1470 unblock_events = false;
1473 mutex_unlock(&bdev->bd_mutex);
1476 disk_unblock_events(disk);
1480 if (mode & FMODE_EXCL)
1481 bd_abort_claiming(bdev, holder);
1482 mutex_unlock(&bdev->bd_mutex);
1483 disk_unblock_events(disk);
1485 blkdev_put_no_open(bdev);
1486 if (ret == -ERESTARTSYS)
1488 return ERR_PTR(ret);
1490 EXPORT_SYMBOL(blkdev_get_by_dev);
1493 * blkdev_get_by_path - open a block device by name
1494 * @path: path to the block device to open
1495 * @mode: FMODE_* mask
1496 * @holder: exclusive holder identifier
1498 * Open the block device described by the device file at @path. If @mode
1499 * includes %FMODE_EXCL, the block device is opened with exclusive access.
1500 * Specifying %FMODE_EXCL with a %NULL @holder is invalid. Exclusive opens may
1501 * nest for the same @holder.
1507 * Reference to the block_device on success, ERR_PTR(-errno) on failure.
1509 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1512 struct block_device *bdev;
1516 error = lookup_bdev(path, &dev);
1518 return ERR_PTR(error);
1520 bdev = blkdev_get_by_dev(dev, mode, holder);
1521 if (!IS_ERR(bdev) && (mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1522 blkdev_put(bdev, mode);
1523 return ERR_PTR(-EACCES);
1528 EXPORT_SYMBOL(blkdev_get_by_path);
1530 static int blkdev_open(struct inode * inode, struct file * filp)
1532 struct block_device *bdev;
1535 * Preserve backwards compatibility and allow large file access
1536 * even if userspace doesn't ask for it explicitly. Some mkfs
1537 * binary needs it. We might want to drop this workaround
1538 * during an unstable branch.
1540 filp->f_flags |= O_LARGEFILE;
1542 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
1544 if (filp->f_flags & O_NDELAY)
1545 filp->f_mode |= FMODE_NDELAY;
1546 if (filp->f_flags & O_EXCL)
1547 filp->f_mode |= FMODE_EXCL;
1548 if ((filp->f_flags & O_ACCMODE) == 3)
1549 filp->f_mode |= FMODE_WRITE_IOCTL;
1551 bdev = blkdev_get_by_dev(inode->i_rdev, filp->f_mode, filp);
1553 return PTR_ERR(bdev);
1554 filp->f_mapping = bdev->bd_inode->i_mapping;
1555 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1559 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1561 struct gendisk *disk = bdev->bd_disk;
1562 struct block_device *victim = NULL;
1565 * Sync early if it looks like we're the last one. If someone else
1566 * opens the block device between now and the decrement of bd_openers
1567 * then we did a sync that we didn't need to, but that's not the end
1568 * of the world and we want to avoid long (could be several minute)
1569 * syncs while holding the mutex.
1571 if (bdev->bd_openers == 1)
1572 sync_blockdev(bdev);
1574 mutex_lock_nested(&bdev->bd_mutex, for_part);
1576 bdev->bd_part_count--;
1578 if (!--bdev->bd_openers) {
1579 WARN_ON_ONCE(bdev->bd_holders);
1580 sync_blockdev(bdev);
1582 bdev_write_inode(bdev);
1583 if (bdev_is_partition(bdev))
1584 victim = bdev_whole(bdev);
1587 if (!bdev_is_partition(bdev) && disk->fops->release)
1588 disk->fops->release(disk, mode);
1589 mutex_unlock(&bdev->bd_mutex);
1591 __blkdev_put(victim, mode, 1);
1596 void blkdev_put(struct block_device *bdev, fmode_t mode)
1598 struct gendisk *disk = bdev->bd_disk;
1600 mutex_lock(&bdev->bd_mutex);
1602 if (mode & FMODE_EXCL) {
1603 struct block_device *whole = bdev_whole(bdev);
1607 * Release a claim on the device. The holder fields
1608 * are protected with bdev_lock. bd_mutex is to
1609 * synchronize disk_holder unlinking.
1611 spin_lock(&bdev_lock);
1613 WARN_ON_ONCE(--bdev->bd_holders < 0);
1614 WARN_ON_ONCE(--whole->bd_holders < 0);
1616 if ((bdev_free = !bdev->bd_holders))
1617 bdev->bd_holder = NULL;
1618 if (!whole->bd_holders)
1619 whole->bd_holder = NULL;
1621 spin_unlock(&bdev_lock);
1624 * If this was the last claim, remove holder link and
1625 * unblock evpoll if it was a write holder.
1627 if (bdev_free && bdev->bd_write_holder) {
1628 disk_unblock_events(disk);
1629 bdev->bd_write_holder = false;
1634 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1635 * event. This is to ensure detection of media removal commanded
1636 * from userland - e.g. eject(1).
1638 disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE);
1639 mutex_unlock(&bdev->bd_mutex);
1641 __blkdev_put(bdev, mode, 0);
1642 blkdev_put_no_open(bdev);
1644 EXPORT_SYMBOL(blkdev_put);
1646 static int blkdev_close(struct inode * inode, struct file * filp)
1648 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1649 blkdev_put(bdev, filp->f_mode);
1653 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1655 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1656 fmode_t mode = file->f_mode;
1659 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1660 * to updated it before every ioctl.
1662 if (file->f_flags & O_NDELAY)
1663 mode |= FMODE_NDELAY;
1665 mode &= ~FMODE_NDELAY;
1667 return blkdev_ioctl(bdev, mode, cmd, arg);
1671 * Write data to the block device. Only intended for the block device itself
1672 * and the raw driver which basically is a fake block device.
1674 * Does not take i_mutex for the write and thus is not for general purpose
1677 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1679 struct file *file = iocb->ki_filp;
1680 struct inode *bd_inode = bdev_file_inode(file);
1681 loff_t size = i_size_read(bd_inode);
1682 struct blk_plug plug;
1685 if (bdev_read_only(I_BDEV(bd_inode)))
1688 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
1691 if (!iov_iter_count(from))
1694 if (iocb->ki_pos >= size)
1697 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1700 iov_iter_truncate(from, size - iocb->ki_pos);
1702 blk_start_plug(&plug);
1703 ret = __generic_file_write_iter(iocb, from);
1705 ret = generic_write_sync(iocb, ret);
1706 blk_finish_plug(&plug);
1709 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1711 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1713 struct file *file = iocb->ki_filp;
1714 struct inode *bd_inode = bdev_file_inode(file);
1715 loff_t size = i_size_read(bd_inode);
1716 loff_t pos = iocb->ki_pos;
1722 iov_iter_truncate(to, size);
1723 return generic_file_read_iter(iocb, to);
1725 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1728 * Try to release a page associated with block device when the system
1729 * is under memory pressure.
1731 static int blkdev_releasepage(struct page *page, gfp_t wait)
1733 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1735 if (super && super->s_op->bdev_try_to_free_page)
1736 return super->s_op->bdev_try_to_free_page(super, page, wait);
1738 return try_to_free_buffers(page);
1741 static int blkdev_writepages(struct address_space *mapping,
1742 struct writeback_control *wbc)
1744 return generic_writepages(mapping, wbc);
1747 static const struct address_space_operations def_blk_aops = {
1748 .readpage = blkdev_readpage,
1749 .readahead = blkdev_readahead,
1750 .writepage = blkdev_writepage,
1751 .write_begin = blkdev_write_begin,
1752 .write_end = blkdev_write_end,
1753 .writepages = blkdev_writepages,
1754 .releasepage = blkdev_releasepage,
1755 .direct_IO = blkdev_direct_IO,
1756 .migratepage = buffer_migrate_page_norefs,
1757 .is_dirty_writeback = buffer_check_dirty_writeback,
1760 #define BLKDEV_FALLOC_FL_SUPPORTED \
1761 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1762 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1764 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1767 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1768 loff_t end = start + len - 1;
1772 /* Fail if we don't recognize the flags. */
1773 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1776 /* Don't go off the end of the device. */
1777 isize = i_size_read(bdev->bd_inode);
1781 if (mode & FALLOC_FL_KEEP_SIZE) {
1782 len = isize - start;
1783 end = start + len - 1;
1789 * Don't allow IO that isn't aligned to logical block size.
1791 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1794 /* Invalidate the page cache, including dirty pages. */
1795 error = truncate_bdev_range(bdev, file->f_mode, start, end);
1800 case FALLOC_FL_ZERO_RANGE:
1801 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
1802 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1803 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
1805 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
1806 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1807 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
1809 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
1810 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1820 * Invalidate the page cache again; if someone wandered in and dirtied
1821 * a page, we just discard it - userspace has no way of knowing whether
1822 * the write happened before or after discard completing...
1824 return truncate_bdev_range(bdev, file->f_mode, start, end);
1827 const struct file_operations def_blk_fops = {
1828 .open = blkdev_open,
1829 .release = blkdev_close,
1830 .llseek = block_llseek,
1831 .read_iter = blkdev_read_iter,
1832 .write_iter = blkdev_write_iter,
1833 .iopoll = blkdev_iopoll,
1834 .mmap = generic_file_mmap,
1835 .fsync = blkdev_fsync,
1836 .unlocked_ioctl = block_ioctl,
1837 #ifdef CONFIG_COMPAT
1838 .compat_ioctl = compat_blkdev_ioctl,
1840 .splice_read = generic_file_splice_read,
1841 .splice_write = iter_file_splice_write,
1842 .fallocate = blkdev_fallocate,
1846 * lookup_bdev - lookup a struct block_device by name
1847 * @pathname: special file representing the block device
1848 * @dev: return value of the block device's dev_t
1850 * Get a reference to the blockdevice at @pathname in the current
1851 * namespace if possible and return it. Return ERR_PTR(error)
1854 int lookup_bdev(const char *pathname, dev_t *dev)
1856 struct inode *inode;
1860 if (!pathname || !*pathname)
1863 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1867 inode = d_backing_inode(path.dentry);
1869 if (!S_ISBLK(inode->i_mode))
1872 if (!may_open_dev(&path))
1875 *dev = inode->i_rdev;
1881 EXPORT_SYMBOL(lookup_bdev);
1883 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1885 struct super_block *sb = get_super(bdev);
1890 * no need to lock the super, get_super holds the
1891 * read mutex so the filesystem cannot go away
1892 * under us (->put_super runs with the write lock
1895 shrink_dcache_sb(sb);
1896 res = invalidate_inodes(sb, kill_dirty);
1899 invalidate_bdev(bdev);
1902 EXPORT_SYMBOL(__invalidate_device);
1904 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1906 struct inode *inode, *old_inode = NULL;
1908 spin_lock(&blockdev_superblock->s_inode_list_lock);
1909 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1910 struct address_space *mapping = inode->i_mapping;
1911 struct block_device *bdev;
1913 spin_lock(&inode->i_lock);
1914 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1915 mapping->nrpages == 0) {
1916 spin_unlock(&inode->i_lock);
1920 spin_unlock(&inode->i_lock);
1921 spin_unlock(&blockdev_superblock->s_inode_list_lock);
1923 * We hold a reference to 'inode' so it couldn't have been
1924 * removed from s_inodes list while we dropped the
1925 * s_inode_list_lock We cannot iput the inode now as we can
1926 * be holding the last reference and we cannot iput it under
1927 * s_inode_list_lock. So we keep the reference and iput it
1932 bdev = I_BDEV(inode);
1934 mutex_lock(&bdev->bd_mutex);
1935 if (bdev->bd_openers)
1937 mutex_unlock(&bdev->bd_mutex);
1939 spin_lock(&blockdev_superblock->s_inode_list_lock);
1941 spin_unlock(&blockdev_superblock->s_inode_list_lock);