4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/device_cgroup.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/backing-dev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/magic.h>
21 #include <linux/dax.h>
22 #include <linux/buffer_head.h>
23 #include <linux/swap.h>
24 #include <linux/pagevec.h>
25 #include <linux/writeback.h>
26 #include <linux/mpage.h>
27 #include <linux/mount.h>
28 #include <linux/uio.h>
29 #include <linux/namei.h>
30 #include <linux/log2.h>
31 #include <linux/cleancache.h>
32 #include <linux/dax.h>
33 #include <linux/badblocks.h>
34 #include <linux/task_io_accounting_ops.h>
35 #include <linux/falloc.h>
36 #include <linux/uaccess.h>
40 struct block_device bdev;
41 struct inode vfs_inode;
44 static const struct address_space_operations def_blk_aops;
46 static inline struct bdev_inode *BDEV_I(struct inode *inode)
48 return container_of(inode, struct bdev_inode, vfs_inode);
51 struct block_device *I_BDEV(struct inode *inode)
53 return &BDEV_I(inode)->bdev;
55 EXPORT_SYMBOL(I_BDEV);
57 void __vfs_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
65 printk_ratelimited("%sVFS (%s): %pV\n", prefix, sb->s_id, &vaf);
69 static void bdev_write_inode(struct block_device *bdev)
71 struct inode *inode = bdev->bd_inode;
74 spin_lock(&inode->i_lock);
75 while (inode->i_state & I_DIRTY) {
76 spin_unlock(&inode->i_lock);
77 ret = write_inode_now(inode, true);
79 char name[BDEVNAME_SIZE];
80 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
81 "for block device %s (err=%d).\n",
82 bdevname(bdev, name), ret);
84 spin_lock(&inode->i_lock);
86 spin_unlock(&inode->i_lock);
89 /* Kill _all_ buffers and pagecache , dirty or not.. */
90 void kill_bdev(struct block_device *bdev)
92 struct address_space *mapping = bdev->bd_inode->i_mapping;
94 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
98 truncate_inode_pages(mapping, 0);
100 EXPORT_SYMBOL(kill_bdev);
102 /* Invalidate clean unused buffers and pagecache. */
103 void invalidate_bdev(struct block_device *bdev)
105 struct address_space *mapping = bdev->bd_inode->i_mapping;
107 if (mapping->nrpages) {
108 invalidate_bh_lrus();
109 lru_add_drain_all(); /* make sure all lru add caches are flushed */
110 invalidate_mapping_pages(mapping, 0, -1);
112 /* 99% of the time, we don't need to flush the cleancache on the bdev.
113 * But, for the strange corners, lets be cautious
115 cleancache_invalidate_inode(mapping);
117 EXPORT_SYMBOL(invalidate_bdev);
119 int set_blocksize(struct block_device *bdev, int size)
121 /* Size must be a power of two, and between 512 and PAGE_SIZE */
122 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
125 /* Size cannot be smaller than the size supported by the device */
126 if (size < bdev_logical_block_size(bdev))
129 /* Don't change the size if it is same as current */
130 if (bdev->bd_block_size != size) {
132 bdev->bd_block_size = size;
133 bdev->bd_inode->i_blkbits = blksize_bits(size);
139 EXPORT_SYMBOL(set_blocksize);
141 int sb_set_blocksize(struct super_block *sb, int size)
143 if (set_blocksize(sb->s_bdev, size))
145 /* If we get here, we know size is power of two
146 * and it's value is between 512 and PAGE_SIZE */
147 sb->s_blocksize = size;
148 sb->s_blocksize_bits = blksize_bits(size);
149 return sb->s_blocksize;
152 EXPORT_SYMBOL(sb_set_blocksize);
154 int sb_min_blocksize(struct super_block *sb, int size)
156 int minsize = bdev_logical_block_size(sb->s_bdev);
159 return sb_set_blocksize(sb, size);
162 EXPORT_SYMBOL(sb_min_blocksize);
165 blkdev_get_block(struct inode *inode, sector_t iblock,
166 struct buffer_head *bh, int create)
168 bh->b_bdev = I_BDEV(inode);
169 bh->b_blocknr = iblock;
170 set_buffer_mapped(bh);
174 static struct inode *bdev_file_inode(struct file *file)
176 return file->f_mapping->host;
179 static unsigned int dio_bio_write_op(struct kiocb *iocb)
181 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
183 /* avoid the need for a I/O completion work item */
184 if (iocb->ki_flags & IOCB_DSYNC)
189 #define DIO_INLINE_BIO_VECS 4
191 static void blkdev_bio_end_io_simple(struct bio *bio)
193 struct task_struct *waiter = bio->bi_private;
195 WRITE_ONCE(bio->bi_private, NULL);
196 wake_up_process(waiter);
200 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
203 struct file *file = iocb->ki_filp;
204 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
205 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec;
206 loff_t pos = iocb->ki_pos;
207 bool should_dirty = false;
213 if ((pos | iov_iter_alignment(iter)) &
214 (bdev_logical_block_size(bdev) - 1))
217 if (nr_pages <= DIO_INLINE_BIO_VECS)
220 vecs = kmalloc(nr_pages * sizeof(struct bio_vec), GFP_KERNEL);
225 bio_init(&bio, vecs, nr_pages);
226 bio_set_dev(&bio, bdev);
227 bio.bi_iter.bi_sector = pos >> 9;
228 bio.bi_write_hint = iocb->ki_hint;
229 bio.bi_private = current;
230 bio.bi_end_io = blkdev_bio_end_io_simple;
232 ret = bio_iov_iter_get_pages(&bio, iter);
235 ret = bio.bi_iter.bi_size;
237 if (iov_iter_rw(iter) == READ) {
238 bio.bi_opf = REQ_OP_READ;
239 if (iter_is_iovec(iter))
242 bio.bi_opf = dio_bio_write_op(iocb);
243 task_io_account_write(ret);
246 qc = submit_bio(&bio);
248 set_current_state(TASK_UNINTERRUPTIBLE);
249 if (!READ_ONCE(bio.bi_private))
251 if (!(iocb->ki_flags & IOCB_HIPRI) ||
252 !blk_mq_poll(bdev_get_queue(bdev), qc))
255 __set_current_state(TASK_RUNNING);
257 bio_for_each_segment_all(bvec, &bio, i) {
258 if (should_dirty && !PageCompound(bvec->bv_page))
259 set_page_dirty_lock(bvec->bv_page);
260 put_page(bvec->bv_page);
263 if (unlikely(bio.bi_status))
264 ret = blk_status_to_errno(bio.bi_status);
267 if (vecs != inline_vecs)
278 struct task_struct *waiter;
283 bool should_dirty : 1;
288 static struct bio_set *blkdev_dio_pool __read_mostly;
290 static void blkdev_bio_end_io(struct bio *bio)
292 struct blkdev_dio *dio = bio->bi_private;
293 bool should_dirty = dio->should_dirty;
295 if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) {
296 if (bio->bi_status && !dio->bio.bi_status)
297 dio->bio.bi_status = bio->bi_status;
300 struct kiocb *iocb = dio->iocb;
303 if (likely(!dio->bio.bi_status)) {
307 ret = blk_status_to_errno(dio->bio.bi_status);
310 dio->iocb->ki_complete(iocb, ret, 0);
313 struct task_struct *waiter = dio->waiter;
315 WRITE_ONCE(dio->waiter, NULL);
316 wake_up_process(waiter);
321 bio_check_pages_dirty(bio);
323 struct bio_vec *bvec;
326 bio_for_each_segment_all(bvec, bio, i)
327 put_page(bvec->bv_page);
333 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
335 struct file *file = iocb->ki_filp;
336 struct inode *inode = bdev_file_inode(file);
337 struct block_device *bdev = I_BDEV(inode);
338 struct blk_plug plug;
339 struct blkdev_dio *dio;
341 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
342 loff_t pos = iocb->ki_pos;
343 blk_qc_t qc = BLK_QC_T_NONE;
346 if ((pos | iov_iter_alignment(iter)) &
347 (bdev_logical_block_size(bdev) - 1))
350 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, blkdev_dio_pool);
351 bio_get(bio); /* extra ref for the completion handler */
353 dio = container_of(bio, struct blkdev_dio, bio);
354 dio->is_sync = is_sync = is_sync_kiocb(iocb);
356 dio->waiter = current;
361 dio->multi_bio = false;
362 dio->should_dirty = is_read && (iter->type == ITER_IOVEC);
364 blk_start_plug(&plug);
366 bio_set_dev(bio, bdev);
367 bio->bi_iter.bi_sector = pos >> 9;
368 bio->bi_write_hint = iocb->ki_hint;
369 bio->bi_private = dio;
370 bio->bi_end_io = blkdev_bio_end_io;
372 ret = bio_iov_iter_get_pages(bio, iter);
374 bio->bi_status = BLK_STS_IOERR;
380 bio->bi_opf = REQ_OP_READ;
381 if (dio->should_dirty)
382 bio_set_pages_dirty(bio);
384 bio->bi_opf = dio_bio_write_op(iocb);
385 task_io_account_write(bio->bi_iter.bi_size);
388 dio->size += bio->bi_iter.bi_size;
389 pos += bio->bi_iter.bi_size;
391 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
393 qc = submit_bio(bio);
397 if (!dio->multi_bio) {
398 dio->multi_bio = true;
399 atomic_set(&dio->ref, 2);
401 atomic_inc(&dio->ref);
405 bio = bio_alloc(GFP_KERNEL, nr_pages);
407 blk_finish_plug(&plug);
413 set_current_state(TASK_UNINTERRUPTIBLE);
414 if (!READ_ONCE(dio->waiter))
417 if (!(iocb->ki_flags & IOCB_HIPRI) ||
418 !blk_mq_poll(bdev_get_queue(bdev), qc))
421 __set_current_state(TASK_RUNNING);
424 ret = blk_status_to_errno(dio->bio.bi_status);
433 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
437 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
440 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
441 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
443 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
446 static __init int blkdev_init(void)
448 blkdev_dio_pool = bioset_create(4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
449 if (!blkdev_dio_pool)
453 module_init(blkdev_init);
455 int __sync_blockdev(struct block_device *bdev, int wait)
460 return filemap_flush(bdev->bd_inode->i_mapping);
461 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
465 * Write out and wait upon all the dirty data associated with a block
466 * device via its mapping. Does not take the superblock lock.
468 int sync_blockdev(struct block_device *bdev)
470 return __sync_blockdev(bdev, 1);
472 EXPORT_SYMBOL(sync_blockdev);
475 * Write out and wait upon all dirty data associated with this
476 * device. Filesystem data as well as the underlying block
477 * device. Takes the superblock lock.
479 int fsync_bdev(struct block_device *bdev)
481 struct super_block *sb = get_super(bdev);
483 int res = sync_filesystem(sb);
487 return sync_blockdev(bdev);
489 EXPORT_SYMBOL(fsync_bdev);
492 * freeze_bdev -- lock a filesystem and force it into a consistent state
493 * @bdev: blockdevice to lock
495 * If a superblock is found on this device, we take the s_umount semaphore
496 * on it to make sure nobody unmounts until the snapshot creation is done.
497 * The reference counter (bd_fsfreeze_count) guarantees that only the last
498 * unfreeze process can unfreeze the frozen filesystem actually when multiple
499 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
500 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
503 struct super_block *freeze_bdev(struct block_device *bdev)
505 struct super_block *sb;
508 mutex_lock(&bdev->bd_fsfreeze_mutex);
509 if (++bdev->bd_fsfreeze_count > 1) {
511 * We don't even need to grab a reference - the first call
512 * to freeze_bdev grab an active reference and only the last
513 * thaw_bdev drops it.
515 sb = get_super(bdev);
518 mutex_unlock(&bdev->bd_fsfreeze_mutex);
522 sb = get_active_super(bdev);
525 if (sb->s_op->freeze_super)
526 error = sb->s_op->freeze_super(sb);
528 error = freeze_super(sb);
530 deactivate_super(sb);
531 bdev->bd_fsfreeze_count--;
532 mutex_unlock(&bdev->bd_fsfreeze_mutex);
533 return ERR_PTR(error);
535 deactivate_super(sb);
538 mutex_unlock(&bdev->bd_fsfreeze_mutex);
539 return sb; /* thaw_bdev releases s->s_umount */
541 EXPORT_SYMBOL(freeze_bdev);
544 * thaw_bdev -- unlock filesystem
545 * @bdev: blockdevice to unlock
546 * @sb: associated superblock
548 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
550 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
554 mutex_lock(&bdev->bd_fsfreeze_mutex);
555 if (!bdev->bd_fsfreeze_count)
559 if (--bdev->bd_fsfreeze_count > 0)
565 if (sb->s_op->thaw_super)
566 error = sb->s_op->thaw_super(sb);
568 error = thaw_super(sb);
570 bdev->bd_fsfreeze_count++;
572 mutex_unlock(&bdev->bd_fsfreeze_mutex);
575 EXPORT_SYMBOL(thaw_bdev);
577 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
579 return block_write_full_page(page, blkdev_get_block, wbc);
582 static int blkdev_readpage(struct file * file, struct page * page)
584 return block_read_full_page(page, blkdev_get_block);
587 static int blkdev_readpages(struct file *file, struct address_space *mapping,
588 struct list_head *pages, unsigned nr_pages)
590 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
593 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
594 loff_t pos, unsigned len, unsigned flags,
595 struct page **pagep, void **fsdata)
597 return block_write_begin(mapping, pos, len, flags, pagep,
601 static int blkdev_write_end(struct file *file, struct address_space *mapping,
602 loff_t pos, unsigned len, unsigned copied,
603 struct page *page, void *fsdata)
606 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
616 * for a block special file file_inode(file)->i_size is zero
617 * so we compute the size by hand (just as in block_read/write above)
619 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
621 struct inode *bd_inode = bdev_file_inode(file);
624 inode_lock(bd_inode);
625 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
626 inode_unlock(bd_inode);
630 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
632 struct inode *bd_inode = bdev_file_inode(filp);
633 struct block_device *bdev = I_BDEV(bd_inode);
636 error = file_write_and_wait_range(filp, start, end);
641 * There is no need to serialise calls to blkdev_issue_flush with
642 * i_mutex and doing so causes performance issues with concurrent
643 * O_SYNC writers to a block device.
645 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
646 if (error == -EOPNOTSUPP)
651 EXPORT_SYMBOL(blkdev_fsync);
654 * bdev_read_page() - Start reading a page from a block device
655 * @bdev: The device to read the page from
656 * @sector: The offset on the device to read the page to (need not be aligned)
657 * @page: The page to read
659 * On entry, the page should be locked. It will be unlocked when the page
660 * has been read. If the block driver implements rw_page synchronously,
661 * that will be true on exit from this function, but it need not be.
663 * Errors returned by this function are usually "soft", eg out of memory, or
664 * queue full; callers should try a different route to read this page rather
665 * than propagate an error back up the stack.
667 * Return: negative errno if an error occurs, 0 if submission was successful.
669 int bdev_read_page(struct block_device *bdev, sector_t sector,
672 const struct block_device_operations *ops = bdev->bd_disk->fops;
673 int result = -EOPNOTSUPP;
675 if (!ops->rw_page || bdev_get_integrity(bdev))
678 result = blk_queue_enter(bdev->bd_queue, false);
681 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false);
682 blk_queue_exit(bdev->bd_queue);
685 EXPORT_SYMBOL_GPL(bdev_read_page);
688 * bdev_write_page() - Start writing a page to a block device
689 * @bdev: The device to write the page to
690 * @sector: The offset on the device to write the page to (need not be aligned)
691 * @page: The page to write
692 * @wbc: The writeback_control for the write
694 * On entry, the page should be locked and not currently under writeback.
695 * On exit, if the write started successfully, the page will be unlocked and
696 * under writeback. If the write failed already (eg the driver failed to
697 * queue the page to the device), the page will still be locked. If the
698 * caller is a ->writepage implementation, it will need to unlock the page.
700 * Errors returned by this function are usually "soft", eg out of memory, or
701 * queue full; callers should try a different route to write this page rather
702 * than propagate an error back up the stack.
704 * Return: negative errno if an error occurs, 0 if submission was successful.
706 int bdev_write_page(struct block_device *bdev, sector_t sector,
707 struct page *page, struct writeback_control *wbc)
710 const struct block_device_operations *ops = bdev->bd_disk->fops;
712 if (!ops->rw_page || bdev_get_integrity(bdev))
714 result = blk_queue_enter(bdev->bd_queue, false);
718 set_page_writeback(page);
719 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true);
721 end_page_writeback(page);
723 clean_page_buffers(page);
726 blk_queue_exit(bdev->bd_queue);
729 EXPORT_SYMBOL_GPL(bdev_write_page);
735 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
736 static struct kmem_cache * bdev_cachep __read_mostly;
738 static struct inode *bdev_alloc_inode(struct super_block *sb)
740 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
743 return &ei->vfs_inode;
746 static void bdev_i_callback(struct rcu_head *head)
748 struct inode *inode = container_of(head, struct inode, i_rcu);
749 struct bdev_inode *bdi = BDEV_I(inode);
751 kmem_cache_free(bdev_cachep, bdi);
754 static void bdev_destroy_inode(struct inode *inode)
756 call_rcu(&inode->i_rcu, bdev_i_callback);
759 static void init_once(void *foo)
761 struct bdev_inode *ei = (struct bdev_inode *) foo;
762 struct block_device *bdev = &ei->bdev;
764 memset(bdev, 0, sizeof(*bdev));
765 mutex_init(&bdev->bd_mutex);
766 INIT_LIST_HEAD(&bdev->bd_list);
768 INIT_LIST_HEAD(&bdev->bd_holder_disks);
770 bdev->bd_bdi = &noop_backing_dev_info;
771 inode_init_once(&ei->vfs_inode);
772 /* Initialize mutex for freeze. */
773 mutex_init(&bdev->bd_fsfreeze_mutex);
776 static void bdev_evict_inode(struct inode *inode)
778 struct block_device *bdev = &BDEV_I(inode)->bdev;
779 truncate_inode_pages_final(&inode->i_data);
780 invalidate_inode_buffers(inode); /* is it needed here? */
782 spin_lock(&bdev_lock);
783 list_del_init(&bdev->bd_list);
784 spin_unlock(&bdev_lock);
785 /* Detach inode from wb early as bdi_put() may free bdi->wb */
786 inode_detach_wb(inode);
787 if (bdev->bd_bdi != &noop_backing_dev_info) {
788 bdi_put(bdev->bd_bdi);
789 bdev->bd_bdi = &noop_backing_dev_info;
793 static const struct super_operations bdev_sops = {
794 .statfs = simple_statfs,
795 .alloc_inode = bdev_alloc_inode,
796 .destroy_inode = bdev_destroy_inode,
797 .drop_inode = generic_delete_inode,
798 .evict_inode = bdev_evict_inode,
801 static struct dentry *bd_mount(struct file_system_type *fs_type,
802 int flags, const char *dev_name, void *data)
805 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
807 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
811 static struct file_system_type bd_type = {
814 .kill_sb = kill_anon_super,
817 struct super_block *blockdev_superblock __read_mostly;
818 EXPORT_SYMBOL_GPL(blockdev_superblock);
820 void __init bdev_cache_init(void)
823 static struct vfsmount *bd_mnt;
825 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
826 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
827 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
829 err = register_filesystem(&bd_type);
831 panic("Cannot register bdev pseudo-fs");
832 bd_mnt = kern_mount(&bd_type);
834 panic("Cannot create bdev pseudo-fs");
835 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
839 * Most likely _very_ bad one - but then it's hardly critical for small
840 * /dev and can be fixed when somebody will need really large one.
841 * Keep in mind that it will be fed through icache hash function too.
843 static inline unsigned long hash(dev_t dev)
845 return MAJOR(dev)+MINOR(dev);
848 static int bdev_test(struct inode *inode, void *data)
850 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
853 static int bdev_set(struct inode *inode, void *data)
855 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
859 static LIST_HEAD(all_bdevs);
862 * If there is a bdev inode for this device, unhash it so that it gets evicted
863 * as soon as last inode reference is dropped.
865 void bdev_unhash_inode(dev_t dev)
869 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
871 remove_inode_hash(inode);
876 struct block_device *bdget(dev_t dev)
878 struct block_device *bdev;
881 inode = iget5_locked(blockdev_superblock, hash(dev),
882 bdev_test, bdev_set, &dev);
887 bdev = &BDEV_I(inode)->bdev;
889 if (inode->i_state & I_NEW) {
890 bdev->bd_contains = NULL;
891 bdev->bd_super = NULL;
892 bdev->bd_inode = inode;
893 bdev->bd_block_size = i_blocksize(inode);
894 bdev->bd_part_count = 0;
895 bdev->bd_invalidated = 0;
896 inode->i_mode = S_IFBLK;
898 inode->i_bdev = bdev;
899 inode->i_data.a_ops = &def_blk_aops;
900 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
901 spin_lock(&bdev_lock);
902 list_add(&bdev->bd_list, &all_bdevs);
903 spin_unlock(&bdev_lock);
904 unlock_new_inode(inode);
909 EXPORT_SYMBOL(bdget);
912 * bdgrab -- Grab a reference to an already referenced block device
913 * @bdev: Block device to grab a reference to.
915 struct block_device *bdgrab(struct block_device *bdev)
917 ihold(bdev->bd_inode);
920 EXPORT_SYMBOL(bdgrab);
922 long nr_blockdev_pages(void)
924 struct block_device *bdev;
926 spin_lock(&bdev_lock);
927 list_for_each_entry(bdev, &all_bdevs, bd_list) {
928 ret += bdev->bd_inode->i_mapping->nrpages;
930 spin_unlock(&bdev_lock);
934 void bdput(struct block_device *bdev)
936 iput(bdev->bd_inode);
939 EXPORT_SYMBOL(bdput);
941 static struct block_device *bd_acquire(struct inode *inode)
943 struct block_device *bdev;
945 spin_lock(&bdev_lock);
946 bdev = inode->i_bdev;
947 if (bdev && !inode_unhashed(bdev->bd_inode)) {
949 spin_unlock(&bdev_lock);
952 spin_unlock(&bdev_lock);
955 * i_bdev references block device inode that was already shut down
956 * (corresponding device got removed). Remove the reference and look
957 * up block device inode again just in case new device got
958 * reestablished under the same device number.
963 bdev = bdget(inode->i_rdev);
965 spin_lock(&bdev_lock);
966 if (!inode->i_bdev) {
968 * We take an additional reference to bd_inode,
969 * and it's released in clear_inode() of inode.
970 * So, we can access it via ->i_mapping always
974 inode->i_bdev = bdev;
975 inode->i_mapping = bdev->bd_inode->i_mapping;
977 spin_unlock(&bdev_lock);
982 /* Call when you free inode */
984 void bd_forget(struct inode *inode)
986 struct block_device *bdev = NULL;
988 spin_lock(&bdev_lock);
989 if (!sb_is_blkdev_sb(inode->i_sb))
990 bdev = inode->i_bdev;
991 inode->i_bdev = NULL;
992 inode->i_mapping = &inode->i_data;
993 spin_unlock(&bdev_lock);
1000 * bd_may_claim - test whether a block device can be claimed
1001 * @bdev: block device of interest
1002 * @whole: whole block device containing @bdev, may equal @bdev
1003 * @holder: holder trying to claim @bdev
1005 * Test whether @bdev can be claimed by @holder.
1008 * spin_lock(&bdev_lock).
1011 * %true if @bdev can be claimed, %false otherwise.
1013 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1016 if (bdev->bd_holder == holder)
1017 return true; /* already a holder */
1018 else if (bdev->bd_holder != NULL)
1019 return false; /* held by someone else */
1020 else if (whole == bdev)
1021 return true; /* is a whole device which isn't held */
1023 else if (whole->bd_holder == bd_may_claim)
1024 return true; /* is a partition of a device that is being partitioned */
1025 else if (whole->bd_holder != NULL)
1026 return false; /* is a partition of a held device */
1028 return true; /* is a partition of an un-held device */
1032 * bd_prepare_to_claim - prepare to claim a block device
1033 * @bdev: block device of interest
1034 * @whole: the whole device containing @bdev, may equal @bdev
1035 * @holder: holder trying to claim @bdev
1037 * Prepare to claim @bdev. This function fails if @bdev is already
1038 * claimed by another holder and waits if another claiming is in
1039 * progress. This function doesn't actually claim. On successful
1040 * return, the caller has ownership of bd_claiming and bd_holder[s].
1043 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1044 * it multiple times.
1047 * 0 if @bdev can be claimed, -EBUSY otherwise.
1049 static int bd_prepare_to_claim(struct block_device *bdev,
1050 struct block_device *whole, void *holder)
1053 /* if someone else claimed, fail */
1054 if (!bd_may_claim(bdev, whole, holder))
1057 /* if claiming is already in progress, wait for it to finish */
1058 if (whole->bd_claiming) {
1059 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1062 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1063 spin_unlock(&bdev_lock);
1065 finish_wait(wq, &wait);
1066 spin_lock(&bdev_lock);
1075 * bd_start_claiming - start claiming a block device
1076 * @bdev: block device of interest
1077 * @holder: holder trying to claim @bdev
1079 * @bdev is about to be opened exclusively. Check @bdev can be opened
1080 * exclusively and mark that an exclusive open is in progress. Each
1081 * successful call to this function must be matched with a call to
1082 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1085 * This function is used to gain exclusive access to the block device
1086 * without actually causing other exclusive open attempts to fail. It
1087 * should be used when the open sequence itself requires exclusive
1088 * access but may subsequently fail.
1094 * Pointer to the block device containing @bdev on success, ERR_PTR()
1097 static struct block_device *bd_start_claiming(struct block_device *bdev,
1100 struct gendisk *disk;
1101 struct block_device *whole;
1107 * @bdev might not have been initialized properly yet, look up
1108 * and grab the outer block device the hard way.
1110 disk = get_gendisk(bdev->bd_dev, &partno);
1112 return ERR_PTR(-ENXIO);
1115 * Normally, @bdev should equal what's returned from bdget_disk()
1116 * if partno is 0; however, some drivers (floppy) use multiple
1117 * bdev's for the same physical device and @bdev may be one of the
1118 * aliases. Keep @bdev if partno is 0. This means claimer
1119 * tracking is broken for those devices but it has always been that
1123 whole = bdget_disk(disk, 0);
1125 whole = bdgrab(bdev);
1127 module_put(disk->fops->owner);
1130 return ERR_PTR(-ENOMEM);
1132 /* prepare to claim, if successful, mark claiming in progress */
1133 spin_lock(&bdev_lock);
1135 err = bd_prepare_to_claim(bdev, whole, holder);
1137 whole->bd_claiming = holder;
1138 spin_unlock(&bdev_lock);
1141 spin_unlock(&bdev_lock);
1143 return ERR_PTR(err);
1148 struct bd_holder_disk {
1149 struct list_head list;
1150 struct gendisk *disk;
1154 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1155 struct gendisk *disk)
1157 struct bd_holder_disk *holder;
1159 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1160 if (holder->disk == disk)
1165 static int add_symlink(struct kobject *from, struct kobject *to)
1167 return sysfs_create_link(from, to, kobject_name(to));
1170 static void del_symlink(struct kobject *from, struct kobject *to)
1172 sysfs_remove_link(from, kobject_name(to));
1176 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1177 * @bdev: the claimed slave bdev
1178 * @disk: the holding disk
1180 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1182 * This functions creates the following sysfs symlinks.
1184 * - from "slaves" directory of the holder @disk to the claimed @bdev
1185 * - from "holders" directory of the @bdev to the holder @disk
1187 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1188 * passed to bd_link_disk_holder(), then:
1190 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1191 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1193 * The caller must have claimed @bdev before calling this function and
1194 * ensure that both @bdev and @disk are valid during the creation and
1195 * lifetime of these symlinks.
1201 * 0 on success, -errno on failure.
1203 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1205 struct bd_holder_disk *holder;
1208 mutex_lock(&bdev->bd_mutex);
1210 WARN_ON_ONCE(!bdev->bd_holder);
1212 /* FIXME: remove the following once add_disk() handles errors */
1213 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1216 holder = bd_find_holder_disk(bdev, disk);
1222 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1228 INIT_LIST_HEAD(&holder->list);
1229 holder->disk = disk;
1232 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1236 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1240 * bdev could be deleted beneath us which would implicitly destroy
1241 * the holder directory. Hold on to it.
1243 kobject_get(bdev->bd_part->holder_dir);
1245 list_add(&holder->list, &bdev->bd_holder_disks);
1249 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1253 mutex_unlock(&bdev->bd_mutex);
1256 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1259 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1260 * @bdev: the calimed slave bdev
1261 * @disk: the holding disk
1263 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1268 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1270 struct bd_holder_disk *holder;
1272 mutex_lock(&bdev->bd_mutex);
1274 holder = bd_find_holder_disk(bdev, disk);
1276 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1277 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1278 del_symlink(bdev->bd_part->holder_dir,
1279 &disk_to_dev(disk)->kobj);
1280 kobject_put(bdev->bd_part->holder_dir);
1281 list_del_init(&holder->list);
1285 mutex_unlock(&bdev->bd_mutex);
1287 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1291 * flush_disk - invalidates all buffer-cache entries on a disk
1293 * @bdev: struct block device to be flushed
1294 * @kill_dirty: flag to guide handling of dirty inodes
1296 * Invalidates all buffer-cache entries on a disk. It should be called
1297 * when a disk has been changed -- either by a media change or online
1300 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1302 if (__invalidate_device(bdev, kill_dirty)) {
1303 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1304 "resized disk %s\n",
1305 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1310 if (disk_part_scan_enabled(bdev->bd_disk))
1311 bdev->bd_invalidated = 1;
1315 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1316 * @disk: struct gendisk to check
1317 * @bdev: struct bdev to adjust.
1319 * This routine checks to see if the bdev size does not match the disk size
1320 * and adjusts it if it differs.
1322 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1324 loff_t disk_size, bdev_size;
1326 disk_size = (loff_t)get_capacity(disk) << 9;
1327 bdev_size = i_size_read(bdev->bd_inode);
1328 if (disk_size != bdev_size) {
1330 "%s: detected capacity change from %lld to %lld\n",
1331 disk->disk_name, bdev_size, disk_size);
1332 i_size_write(bdev->bd_inode, disk_size);
1333 flush_disk(bdev, false);
1336 EXPORT_SYMBOL(check_disk_size_change);
1339 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1340 * @disk: struct gendisk to be revalidated
1342 * This routine is a wrapper for lower-level driver's revalidate_disk
1343 * call-backs. It is used to do common pre and post operations needed
1344 * for all revalidate_disk operations.
1346 int revalidate_disk(struct gendisk *disk)
1348 struct block_device *bdev;
1351 if (disk->fops->revalidate_disk)
1352 ret = disk->fops->revalidate_disk(disk);
1353 bdev = bdget_disk(disk, 0);
1357 mutex_lock(&bdev->bd_mutex);
1358 check_disk_size_change(disk, bdev);
1359 bdev->bd_invalidated = 0;
1360 mutex_unlock(&bdev->bd_mutex);
1364 EXPORT_SYMBOL(revalidate_disk);
1367 * This routine checks whether a removable media has been changed,
1368 * and invalidates all buffer-cache-entries in that case. This
1369 * is a relatively slow routine, so we have to try to minimize using
1370 * it. Thus it is called only upon a 'mount' or 'open'. This
1371 * is the best way of combining speed and utility, I think.
1372 * People changing diskettes in the middle of an operation deserve
1375 int check_disk_change(struct block_device *bdev)
1377 struct gendisk *disk = bdev->bd_disk;
1378 const struct block_device_operations *bdops = disk->fops;
1379 unsigned int events;
1381 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1382 DISK_EVENT_EJECT_REQUEST);
1383 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1386 flush_disk(bdev, true);
1387 if (bdops->revalidate_disk)
1388 bdops->revalidate_disk(bdev->bd_disk);
1392 EXPORT_SYMBOL(check_disk_change);
1394 void bd_set_size(struct block_device *bdev, loff_t size)
1396 unsigned bsize = bdev_logical_block_size(bdev);
1398 inode_lock(bdev->bd_inode);
1399 i_size_write(bdev->bd_inode, size);
1400 inode_unlock(bdev->bd_inode);
1401 while (bsize < PAGE_SIZE) {
1406 bdev->bd_block_size = bsize;
1407 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1409 EXPORT_SYMBOL(bd_set_size);
1411 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1416 * mutex_lock(part->bd_mutex)
1417 * mutex_lock_nested(whole->bd_mutex, 1)
1420 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1422 struct gendisk *disk;
1423 struct module *owner;
1428 if (mode & FMODE_READ)
1430 if (mode & FMODE_WRITE)
1433 * hooks: /n/, see "layering violations".
1436 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1446 disk = get_gendisk(bdev->bd_dev, &partno);
1449 owner = disk->fops->owner;
1451 disk_block_events(disk);
1452 mutex_lock_nested(&bdev->bd_mutex, for_part);
1453 if (!bdev->bd_openers) {
1454 bdev->bd_disk = disk;
1455 bdev->bd_queue = disk->queue;
1456 bdev->bd_contains = bdev;
1457 bdev->bd_partno = partno;
1461 bdev->bd_part = disk_get_part(disk, partno);
1466 if (disk->fops->open) {
1467 ret = disk->fops->open(bdev, mode);
1468 if (ret == -ERESTARTSYS) {
1469 /* Lost a race with 'disk' being
1470 * deleted, try again.
1473 disk_put_part(bdev->bd_part);
1474 bdev->bd_part = NULL;
1475 bdev->bd_disk = NULL;
1476 bdev->bd_queue = NULL;
1477 mutex_unlock(&bdev->bd_mutex);
1478 disk_unblock_events(disk);
1486 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1489 * If the device is invalidated, rescan partition
1490 * if open succeeded or failed with -ENOMEDIUM.
1491 * The latter is necessary to prevent ghost
1492 * partitions on a removed medium.
1494 if (bdev->bd_invalidated) {
1496 rescan_partitions(disk, bdev);
1497 else if (ret == -ENOMEDIUM)
1498 invalidate_partitions(disk, bdev);
1504 struct block_device *whole;
1505 whole = bdget_disk(disk, 0);
1510 ret = __blkdev_get(whole, mode, 1);
1513 bdev->bd_contains = whole;
1514 bdev->bd_part = disk_get_part(disk, partno);
1515 if (!(disk->flags & GENHD_FL_UP) ||
1516 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1520 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1523 if (bdev->bd_bdi == &noop_backing_dev_info)
1524 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1526 if (bdev->bd_contains == bdev) {
1528 if (bdev->bd_disk->fops->open)
1529 ret = bdev->bd_disk->fops->open(bdev, mode);
1530 /* the same as first opener case, read comment there */
1531 if (bdev->bd_invalidated) {
1533 rescan_partitions(bdev->bd_disk, bdev);
1534 else if (ret == -ENOMEDIUM)
1535 invalidate_partitions(bdev->bd_disk, bdev);
1538 goto out_unlock_bdev;
1540 /* only one opener holds refs to the module and disk */
1546 bdev->bd_part_count++;
1547 mutex_unlock(&bdev->bd_mutex);
1548 disk_unblock_events(disk);
1552 disk_put_part(bdev->bd_part);
1553 bdev->bd_disk = NULL;
1554 bdev->bd_part = NULL;
1555 bdev->bd_queue = NULL;
1556 if (bdev != bdev->bd_contains)
1557 __blkdev_put(bdev->bd_contains, mode, 1);
1558 bdev->bd_contains = NULL;
1560 mutex_unlock(&bdev->bd_mutex);
1561 disk_unblock_events(disk);
1571 * blkdev_get - open a block device
1572 * @bdev: block_device to open
1573 * @mode: FMODE_* mask
1574 * @holder: exclusive holder identifier
1576 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1577 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1578 * @holder is invalid. Exclusive opens may nest for the same @holder.
1580 * On success, the reference count of @bdev is unchanged. On failure,
1587 * 0 on success, -errno on failure.
1589 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1591 struct block_device *whole = NULL;
1594 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1596 if ((mode & FMODE_EXCL) && holder) {
1597 whole = bd_start_claiming(bdev, holder);
1598 if (IS_ERR(whole)) {
1600 return PTR_ERR(whole);
1604 res = __blkdev_get(bdev, mode, 0);
1607 struct gendisk *disk = whole->bd_disk;
1609 /* finish claiming */
1610 mutex_lock(&bdev->bd_mutex);
1611 spin_lock(&bdev_lock);
1614 BUG_ON(!bd_may_claim(bdev, whole, holder));
1616 * Note that for a whole device bd_holders
1617 * will be incremented twice, and bd_holder
1618 * will be set to bd_may_claim before being
1621 whole->bd_holders++;
1622 whole->bd_holder = bd_may_claim;
1624 bdev->bd_holder = holder;
1627 /* tell others that we're done */
1628 BUG_ON(whole->bd_claiming != holder);
1629 whole->bd_claiming = NULL;
1630 wake_up_bit(&whole->bd_claiming, 0);
1632 spin_unlock(&bdev_lock);
1635 * Block event polling for write claims if requested. Any
1636 * write holder makes the write_holder state stick until
1637 * all are released. This is good enough and tracking
1638 * individual writeable reference is too fragile given the
1639 * way @mode is used in blkdev_get/put().
1641 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1642 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1643 bdev->bd_write_holder = true;
1644 disk_block_events(disk);
1647 mutex_unlock(&bdev->bd_mutex);
1653 EXPORT_SYMBOL(blkdev_get);
1656 * blkdev_get_by_path - open a block device by name
1657 * @path: path to the block device to open
1658 * @mode: FMODE_* mask
1659 * @holder: exclusive holder identifier
1661 * Open the blockdevice described by the device file at @path. @mode
1662 * and @holder are identical to blkdev_get().
1664 * On success, the returned block_device has reference count of one.
1670 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1672 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1675 struct block_device *bdev;
1678 bdev = lookup_bdev(path);
1682 err = blkdev_get(bdev, mode, holder);
1684 return ERR_PTR(err);
1686 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1687 blkdev_put(bdev, mode);
1688 return ERR_PTR(-EACCES);
1693 EXPORT_SYMBOL(blkdev_get_by_path);
1696 * blkdev_get_by_dev - open a block device by device number
1697 * @dev: device number of block device to open
1698 * @mode: FMODE_* mask
1699 * @holder: exclusive holder identifier
1701 * Open the blockdevice described by device number @dev. @mode and
1702 * @holder are identical to blkdev_get().
1704 * Use it ONLY if you really do not have anything better - i.e. when
1705 * you are behind a truly sucky interface and all you are given is a
1706 * device number. _Never_ to be used for internal purposes. If you
1707 * ever need it - reconsider your API.
1709 * On success, the returned block_device has reference count of one.
1715 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1717 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1719 struct block_device *bdev;
1724 return ERR_PTR(-ENOMEM);
1726 err = blkdev_get(bdev, mode, holder);
1728 return ERR_PTR(err);
1732 EXPORT_SYMBOL(blkdev_get_by_dev);
1734 static int blkdev_open(struct inode * inode, struct file * filp)
1736 struct block_device *bdev;
1739 * Preserve backwards compatibility and allow large file access
1740 * even if userspace doesn't ask for it explicitly. Some mkfs
1741 * binary needs it. We might want to drop this workaround
1742 * during an unstable branch.
1744 filp->f_flags |= O_LARGEFILE;
1746 filp->f_mode |= FMODE_NOWAIT;
1748 if (filp->f_flags & O_NDELAY)
1749 filp->f_mode |= FMODE_NDELAY;
1750 if (filp->f_flags & O_EXCL)
1751 filp->f_mode |= FMODE_EXCL;
1752 if ((filp->f_flags & O_ACCMODE) == 3)
1753 filp->f_mode |= FMODE_WRITE_IOCTL;
1755 bdev = bd_acquire(inode);
1759 filp->f_mapping = bdev->bd_inode->i_mapping;
1760 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1762 return blkdev_get(bdev, filp->f_mode, filp);
1765 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1767 struct gendisk *disk = bdev->bd_disk;
1768 struct block_device *victim = NULL;
1770 mutex_lock_nested(&bdev->bd_mutex, for_part);
1772 bdev->bd_part_count--;
1774 if (!--bdev->bd_openers) {
1775 WARN_ON_ONCE(bdev->bd_holders);
1776 sync_blockdev(bdev);
1779 bdev_write_inode(bdev);
1781 if (bdev->bd_contains == bdev) {
1782 if (disk->fops->release)
1783 disk->fops->release(disk, mode);
1785 if (!bdev->bd_openers) {
1786 struct module *owner = disk->fops->owner;
1788 disk_put_part(bdev->bd_part);
1789 bdev->bd_part = NULL;
1790 bdev->bd_disk = NULL;
1791 if (bdev != bdev->bd_contains)
1792 victim = bdev->bd_contains;
1793 bdev->bd_contains = NULL;
1798 mutex_unlock(&bdev->bd_mutex);
1801 __blkdev_put(victim, mode, 1);
1804 void blkdev_put(struct block_device *bdev, fmode_t mode)
1806 mutex_lock(&bdev->bd_mutex);
1808 if (mode & FMODE_EXCL) {
1812 * Release a claim on the device. The holder fields
1813 * are protected with bdev_lock. bd_mutex is to
1814 * synchronize disk_holder unlinking.
1816 spin_lock(&bdev_lock);
1818 WARN_ON_ONCE(--bdev->bd_holders < 0);
1819 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1821 /* bd_contains might point to self, check in a separate step */
1822 if ((bdev_free = !bdev->bd_holders))
1823 bdev->bd_holder = NULL;
1824 if (!bdev->bd_contains->bd_holders)
1825 bdev->bd_contains->bd_holder = NULL;
1827 spin_unlock(&bdev_lock);
1830 * If this was the last claim, remove holder link and
1831 * unblock evpoll if it was a write holder.
1833 if (bdev_free && bdev->bd_write_holder) {
1834 disk_unblock_events(bdev->bd_disk);
1835 bdev->bd_write_holder = false;
1840 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1841 * event. This is to ensure detection of media removal commanded
1842 * from userland - e.g. eject(1).
1844 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1846 mutex_unlock(&bdev->bd_mutex);
1848 __blkdev_put(bdev, mode, 0);
1850 EXPORT_SYMBOL(blkdev_put);
1852 static int blkdev_close(struct inode * inode, struct file * filp)
1854 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1855 blkdev_put(bdev, filp->f_mode);
1859 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1861 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1862 fmode_t mode = file->f_mode;
1865 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1866 * to updated it before every ioctl.
1868 if (file->f_flags & O_NDELAY)
1869 mode |= FMODE_NDELAY;
1871 mode &= ~FMODE_NDELAY;
1873 return blkdev_ioctl(bdev, mode, cmd, arg);
1877 * Write data to the block device. Only intended for the block device itself
1878 * and the raw driver which basically is a fake block device.
1880 * Does not take i_mutex for the write and thus is not for general purpose
1883 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1885 struct file *file = iocb->ki_filp;
1886 struct inode *bd_inode = bdev_file_inode(file);
1887 loff_t size = i_size_read(bd_inode);
1888 struct blk_plug plug;
1891 if (bdev_read_only(I_BDEV(bd_inode)))
1894 if (!iov_iter_count(from))
1897 if (iocb->ki_pos >= size)
1900 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1903 iov_iter_truncate(from, size - iocb->ki_pos);
1905 blk_start_plug(&plug);
1906 ret = __generic_file_write_iter(iocb, from);
1908 ret = generic_write_sync(iocb, ret);
1909 blk_finish_plug(&plug);
1912 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1914 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1916 struct file *file = iocb->ki_filp;
1917 struct inode *bd_inode = bdev_file_inode(file);
1918 loff_t size = i_size_read(bd_inode);
1919 loff_t pos = iocb->ki_pos;
1925 iov_iter_truncate(to, size);
1926 return generic_file_read_iter(iocb, to);
1928 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1931 * Try to release a page associated with block device when the system
1932 * is under memory pressure.
1934 static int blkdev_releasepage(struct page *page, gfp_t wait)
1936 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1938 if (super && super->s_op->bdev_try_to_free_page)
1939 return super->s_op->bdev_try_to_free_page(super, page, wait);
1941 return try_to_free_buffers(page);
1944 static int blkdev_writepages(struct address_space *mapping,
1945 struct writeback_control *wbc)
1947 if (dax_mapping(mapping)) {
1948 struct block_device *bdev = I_BDEV(mapping->host);
1950 return dax_writeback_mapping_range(mapping, bdev, wbc);
1952 return generic_writepages(mapping, wbc);
1955 static const struct address_space_operations def_blk_aops = {
1956 .readpage = blkdev_readpage,
1957 .readpages = blkdev_readpages,
1958 .writepage = blkdev_writepage,
1959 .write_begin = blkdev_write_begin,
1960 .write_end = blkdev_write_end,
1961 .writepages = blkdev_writepages,
1962 .releasepage = blkdev_releasepage,
1963 .direct_IO = blkdev_direct_IO,
1964 .is_dirty_writeback = buffer_check_dirty_writeback,
1967 #define BLKDEV_FALLOC_FL_SUPPORTED \
1968 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1969 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1971 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1974 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1975 struct address_space *mapping;
1976 loff_t end = start + len - 1;
1980 /* Fail if we don't recognize the flags. */
1981 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1984 /* Don't go off the end of the device. */
1985 isize = i_size_read(bdev->bd_inode);
1989 if (mode & FALLOC_FL_KEEP_SIZE) {
1990 len = isize - start;
1991 end = start + len - 1;
1997 * Don't allow IO that isn't aligned to logical block size.
1999 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2002 /* Invalidate the page cache, including dirty pages. */
2003 mapping = bdev->bd_inode->i_mapping;
2004 truncate_inode_pages_range(mapping, start, end);
2007 case FALLOC_FL_ZERO_RANGE:
2008 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2009 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2010 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2012 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2013 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2014 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2016 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2017 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2027 * Invalidate again; if someone wandered in and dirtied a page,
2028 * the caller will be given -EBUSY. The third argument is
2029 * inclusive, so the rounding here is safe.
2031 return invalidate_inode_pages2_range(mapping,
2032 start >> PAGE_SHIFT,
2036 const struct file_operations def_blk_fops = {
2037 .open = blkdev_open,
2038 .release = blkdev_close,
2039 .llseek = block_llseek,
2040 .read_iter = blkdev_read_iter,
2041 .write_iter = blkdev_write_iter,
2042 .mmap = generic_file_mmap,
2043 .fsync = blkdev_fsync,
2044 .unlocked_ioctl = block_ioctl,
2045 #ifdef CONFIG_COMPAT
2046 .compat_ioctl = compat_blkdev_ioctl,
2048 .splice_read = generic_file_splice_read,
2049 .splice_write = iter_file_splice_write,
2050 .fallocate = blkdev_fallocate,
2053 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2056 mm_segment_t old_fs = get_fs();
2058 res = blkdev_ioctl(bdev, 0, cmd, arg);
2063 EXPORT_SYMBOL(ioctl_by_bdev);
2066 * lookup_bdev - lookup a struct block_device by name
2067 * @pathname: special file representing the block device
2069 * Get a reference to the blockdevice at @pathname in the current
2070 * namespace if possible and return it. Return ERR_PTR(error)
2073 struct block_device *lookup_bdev(const char *pathname)
2075 struct block_device *bdev;
2076 struct inode *inode;
2080 if (!pathname || !*pathname)
2081 return ERR_PTR(-EINVAL);
2083 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2085 return ERR_PTR(error);
2087 inode = d_backing_inode(path.dentry);
2089 if (!S_ISBLK(inode->i_mode))
2092 if (!may_open_dev(&path))
2095 bdev = bd_acquire(inode);
2102 bdev = ERR_PTR(error);
2105 EXPORT_SYMBOL(lookup_bdev);
2107 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2109 struct super_block *sb = get_super(bdev);
2114 * no need to lock the super, get_super holds the
2115 * read mutex so the filesystem cannot go away
2116 * under us (->put_super runs with the write lock
2119 shrink_dcache_sb(sb);
2120 res = invalidate_inodes(sb, kill_dirty);
2123 invalidate_bdev(bdev);
2126 EXPORT_SYMBOL(__invalidate_device);
2128 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2130 struct inode *inode, *old_inode = NULL;
2132 spin_lock(&blockdev_superblock->s_inode_list_lock);
2133 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2134 struct address_space *mapping = inode->i_mapping;
2135 struct block_device *bdev;
2137 spin_lock(&inode->i_lock);
2138 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2139 mapping->nrpages == 0) {
2140 spin_unlock(&inode->i_lock);
2144 spin_unlock(&inode->i_lock);
2145 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2147 * We hold a reference to 'inode' so it couldn't have been
2148 * removed from s_inodes list while we dropped the
2149 * s_inode_list_lock We cannot iput the inode now as we can
2150 * be holding the last reference and we cannot iput it under
2151 * s_inode_list_lock. So we keep the reference and iput it
2156 bdev = I_BDEV(inode);
2158 mutex_lock(&bdev->bd_mutex);
2159 if (bdev->bd_openers)
2161 mutex_unlock(&bdev->bd_mutex);
2163 spin_lock(&blockdev_superblock->s_inode_list_lock);
2165 spin_unlock(&blockdev_superblock->s_inode_list_lock);