4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/device_cgroup.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/backing-dev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/magic.h>
21 #include <linux/dax.h>
22 #include <linux/buffer_head.h>
23 #include <linux/swap.h>
24 #include <linux/pagevec.h>
25 #include <linux/writeback.h>
26 #include <linux/mpage.h>
27 #include <linux/mount.h>
28 #include <linux/uio.h>
29 #include <linux/namei.h>
30 #include <linux/log2.h>
31 #include <linux/cleancache.h>
32 #include <linux/dax.h>
33 #include <linux/badblocks.h>
34 #include <linux/task_io_accounting_ops.h>
35 #include <linux/falloc.h>
36 #include <linux/uaccess.h>
40 struct block_device bdev;
41 struct inode vfs_inode;
44 static const struct address_space_operations def_blk_aops;
46 static inline struct bdev_inode *BDEV_I(struct inode *inode)
48 return container_of(inode, struct bdev_inode, vfs_inode);
51 struct block_device *I_BDEV(struct inode *inode)
53 return &BDEV_I(inode)->bdev;
55 EXPORT_SYMBOL(I_BDEV);
57 static void bdev_write_inode(struct block_device *bdev)
59 struct inode *inode = bdev->bd_inode;
62 spin_lock(&inode->i_lock);
63 while (inode->i_state & I_DIRTY) {
64 spin_unlock(&inode->i_lock);
65 ret = write_inode_now(inode, true);
67 char name[BDEVNAME_SIZE];
68 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
69 "for block device %s (err=%d).\n",
70 bdevname(bdev, name), ret);
72 spin_lock(&inode->i_lock);
74 spin_unlock(&inode->i_lock);
77 /* Kill _all_ buffers and pagecache , dirty or not.. */
78 void kill_bdev(struct block_device *bdev)
80 struct address_space *mapping = bdev->bd_inode->i_mapping;
82 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
86 truncate_inode_pages(mapping, 0);
88 EXPORT_SYMBOL(kill_bdev);
90 /* Invalidate clean unused buffers and pagecache. */
91 void invalidate_bdev(struct block_device *bdev)
93 struct address_space *mapping = bdev->bd_inode->i_mapping;
95 if (mapping->nrpages) {
97 lru_add_drain_all(); /* make sure all lru add caches are flushed */
98 invalidate_mapping_pages(mapping, 0, -1);
100 /* 99% of the time, we don't need to flush the cleancache on the bdev.
101 * But, for the strange corners, lets be cautious
103 cleancache_invalidate_inode(mapping);
105 EXPORT_SYMBOL(invalidate_bdev);
107 int set_blocksize(struct block_device *bdev, int size)
109 /* Size must be a power of two, and between 512 and PAGE_SIZE */
110 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
113 /* Size cannot be smaller than the size supported by the device */
114 if (size < bdev_logical_block_size(bdev))
117 /* Don't change the size if it is same as current */
118 if (bdev->bd_block_size != size) {
120 bdev->bd_block_size = size;
121 bdev->bd_inode->i_blkbits = blksize_bits(size);
127 EXPORT_SYMBOL(set_blocksize);
129 int sb_set_blocksize(struct super_block *sb, int size)
131 if (set_blocksize(sb->s_bdev, size))
133 /* If we get here, we know size is power of two
134 * and it's value is between 512 and PAGE_SIZE */
135 sb->s_blocksize = size;
136 sb->s_blocksize_bits = blksize_bits(size);
137 return sb->s_blocksize;
140 EXPORT_SYMBOL(sb_set_blocksize);
142 int sb_min_blocksize(struct super_block *sb, int size)
144 int minsize = bdev_logical_block_size(sb->s_bdev);
147 return sb_set_blocksize(sb, size);
150 EXPORT_SYMBOL(sb_min_blocksize);
153 blkdev_get_block(struct inode *inode, sector_t iblock,
154 struct buffer_head *bh, int create)
156 bh->b_bdev = I_BDEV(inode);
157 bh->b_blocknr = iblock;
158 set_buffer_mapped(bh);
162 static struct inode *bdev_file_inode(struct file *file)
164 return file->f_mapping->host;
167 static unsigned int dio_bio_write_op(struct kiocb *iocb)
169 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
171 /* avoid the need for a I/O completion work item */
172 if (iocb->ki_flags & IOCB_DSYNC)
177 #define DIO_INLINE_BIO_VECS 4
179 static void blkdev_bio_end_io_simple(struct bio *bio)
181 struct task_struct *waiter = bio->bi_private;
183 WRITE_ONCE(bio->bi_private, NULL);
184 wake_up_process(waiter);
188 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
191 struct file *file = iocb->ki_filp;
192 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
193 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec;
194 loff_t pos = iocb->ki_pos;
195 bool should_dirty = false;
201 if ((pos | iov_iter_alignment(iter)) &
202 (bdev_logical_block_size(bdev) - 1))
205 if (nr_pages <= DIO_INLINE_BIO_VECS)
208 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
214 bio_init(&bio, vecs, nr_pages);
215 bio_set_dev(&bio, bdev);
216 bio.bi_iter.bi_sector = pos >> 9;
217 bio.bi_write_hint = iocb->ki_hint;
218 bio.bi_private = current;
219 bio.bi_end_io = blkdev_bio_end_io_simple;
220 bio.bi_ioprio = iocb->ki_ioprio;
222 ret = bio_iov_iter_get_pages(&bio, iter);
225 ret = bio.bi_iter.bi_size;
227 if (iov_iter_rw(iter) == READ) {
228 bio.bi_opf = REQ_OP_READ;
229 if (iter_is_iovec(iter))
232 bio.bi_opf = dio_bio_write_op(iocb);
233 task_io_account_write(ret);
236 qc = submit_bio(&bio);
238 set_current_state(TASK_UNINTERRUPTIBLE);
239 if (!READ_ONCE(bio.bi_private))
241 if (!(iocb->ki_flags & IOCB_HIPRI) ||
242 !blk_poll(bdev_get_queue(bdev), qc))
245 __set_current_state(TASK_RUNNING);
247 bio_for_each_segment_all(bvec, &bio, i) {
248 if (should_dirty && !PageCompound(bvec->bv_page))
249 set_page_dirty_lock(bvec->bv_page);
250 put_page(bvec->bv_page);
253 if (unlikely(bio.bi_status))
254 ret = blk_status_to_errno(bio.bi_status);
257 if (vecs != inline_vecs)
268 struct task_struct *waiter;
273 bool should_dirty : 1;
278 static struct bio_set blkdev_dio_pool;
280 static void blkdev_bio_end_io(struct bio *bio)
282 struct blkdev_dio *dio = bio->bi_private;
283 bool should_dirty = dio->should_dirty;
285 if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) {
286 if (bio->bi_status && !dio->bio.bi_status)
287 dio->bio.bi_status = bio->bi_status;
290 struct kiocb *iocb = dio->iocb;
293 if (likely(!dio->bio.bi_status)) {
297 ret = blk_status_to_errno(dio->bio.bi_status);
300 dio->iocb->ki_complete(iocb, ret, 0);
303 struct task_struct *waiter = dio->waiter;
305 WRITE_ONCE(dio->waiter, NULL);
306 wake_up_process(waiter);
311 bio_check_pages_dirty(bio);
313 struct bio_vec *bvec;
316 bio_for_each_segment_all(bvec, bio, i)
317 put_page(bvec->bv_page);
323 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
325 struct file *file = iocb->ki_filp;
326 struct inode *inode = bdev_file_inode(file);
327 struct block_device *bdev = I_BDEV(inode);
328 struct blk_plug plug;
329 struct blkdev_dio *dio;
331 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
332 loff_t pos = iocb->ki_pos;
333 blk_qc_t qc = BLK_QC_T_NONE;
336 if ((pos | iov_iter_alignment(iter)) &
337 (bdev_logical_block_size(bdev) - 1))
340 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
341 bio_get(bio); /* extra ref for the completion handler */
343 dio = container_of(bio, struct blkdev_dio, bio);
344 dio->is_sync = is_sync = is_sync_kiocb(iocb);
346 dio->waiter = current;
351 dio->multi_bio = false;
352 dio->should_dirty = is_read && (iter->type == ITER_IOVEC);
354 blk_start_plug(&plug);
356 bio_set_dev(bio, bdev);
357 bio->bi_iter.bi_sector = pos >> 9;
358 bio->bi_write_hint = iocb->ki_hint;
359 bio->bi_private = dio;
360 bio->bi_end_io = blkdev_bio_end_io;
361 bio->bi_ioprio = iocb->ki_ioprio;
363 ret = bio_iov_iter_get_pages(bio, iter);
365 bio->bi_status = BLK_STS_IOERR;
371 bio->bi_opf = REQ_OP_READ;
372 if (dio->should_dirty)
373 bio_set_pages_dirty(bio);
375 bio->bi_opf = dio_bio_write_op(iocb);
376 task_io_account_write(bio->bi_iter.bi_size);
379 dio->size += bio->bi_iter.bi_size;
380 pos += bio->bi_iter.bi_size;
382 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
384 qc = submit_bio(bio);
388 if (!dio->multi_bio) {
389 dio->multi_bio = true;
390 atomic_set(&dio->ref, 2);
392 atomic_inc(&dio->ref);
396 bio = bio_alloc(GFP_KERNEL, nr_pages);
398 blk_finish_plug(&plug);
404 set_current_state(TASK_UNINTERRUPTIBLE);
405 if (!READ_ONCE(dio->waiter))
408 if (!(iocb->ki_flags & IOCB_HIPRI) ||
409 !blk_poll(bdev_get_queue(bdev), qc))
412 __set_current_state(TASK_RUNNING);
415 ret = blk_status_to_errno(dio->bio.bi_status);
424 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
428 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
431 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
432 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
434 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
437 static __init int blkdev_init(void)
439 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
441 module_init(blkdev_init);
443 int __sync_blockdev(struct block_device *bdev, int wait)
448 return filemap_flush(bdev->bd_inode->i_mapping);
449 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
453 * Write out and wait upon all the dirty data associated with a block
454 * device via its mapping. Does not take the superblock lock.
456 int sync_blockdev(struct block_device *bdev)
458 return __sync_blockdev(bdev, 1);
460 EXPORT_SYMBOL(sync_blockdev);
463 * Write out and wait upon all dirty data associated with this
464 * device. Filesystem data as well as the underlying block
465 * device. Takes the superblock lock.
467 int fsync_bdev(struct block_device *bdev)
469 struct super_block *sb = get_super(bdev);
471 int res = sync_filesystem(sb);
475 return sync_blockdev(bdev);
477 EXPORT_SYMBOL(fsync_bdev);
480 * freeze_bdev -- lock a filesystem and force it into a consistent state
481 * @bdev: blockdevice to lock
483 * If a superblock is found on this device, we take the s_umount semaphore
484 * on it to make sure nobody unmounts until the snapshot creation is done.
485 * The reference counter (bd_fsfreeze_count) guarantees that only the last
486 * unfreeze process can unfreeze the frozen filesystem actually when multiple
487 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
488 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
491 struct super_block *freeze_bdev(struct block_device *bdev)
493 struct super_block *sb;
496 mutex_lock(&bdev->bd_fsfreeze_mutex);
497 if (++bdev->bd_fsfreeze_count > 1) {
499 * We don't even need to grab a reference - the first call
500 * to freeze_bdev grab an active reference and only the last
501 * thaw_bdev drops it.
503 sb = get_super(bdev);
506 mutex_unlock(&bdev->bd_fsfreeze_mutex);
510 sb = get_active_super(bdev);
513 if (sb->s_op->freeze_super)
514 error = sb->s_op->freeze_super(sb);
516 error = freeze_super(sb);
518 deactivate_super(sb);
519 bdev->bd_fsfreeze_count--;
520 mutex_unlock(&bdev->bd_fsfreeze_mutex);
521 return ERR_PTR(error);
523 deactivate_super(sb);
526 mutex_unlock(&bdev->bd_fsfreeze_mutex);
527 return sb; /* thaw_bdev releases s->s_umount */
529 EXPORT_SYMBOL(freeze_bdev);
532 * thaw_bdev -- unlock filesystem
533 * @bdev: blockdevice to unlock
534 * @sb: associated superblock
536 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
538 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
542 mutex_lock(&bdev->bd_fsfreeze_mutex);
543 if (!bdev->bd_fsfreeze_count)
547 if (--bdev->bd_fsfreeze_count > 0)
553 if (sb->s_op->thaw_super)
554 error = sb->s_op->thaw_super(sb);
556 error = thaw_super(sb);
558 bdev->bd_fsfreeze_count++;
560 mutex_unlock(&bdev->bd_fsfreeze_mutex);
563 EXPORT_SYMBOL(thaw_bdev);
565 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
567 return block_write_full_page(page, blkdev_get_block, wbc);
570 static int blkdev_readpage(struct file * file, struct page * page)
572 return block_read_full_page(page, blkdev_get_block);
575 static int blkdev_readpages(struct file *file, struct address_space *mapping,
576 struct list_head *pages, unsigned nr_pages)
578 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
581 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
582 loff_t pos, unsigned len, unsigned flags,
583 struct page **pagep, void **fsdata)
585 return block_write_begin(mapping, pos, len, flags, pagep,
589 static int blkdev_write_end(struct file *file, struct address_space *mapping,
590 loff_t pos, unsigned len, unsigned copied,
591 struct page *page, void *fsdata)
594 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
604 * for a block special file file_inode(file)->i_size is zero
605 * so we compute the size by hand (just as in block_read/write above)
607 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
609 struct inode *bd_inode = bdev_file_inode(file);
612 inode_lock(bd_inode);
613 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
614 inode_unlock(bd_inode);
618 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
620 struct inode *bd_inode = bdev_file_inode(filp);
621 struct block_device *bdev = I_BDEV(bd_inode);
624 error = file_write_and_wait_range(filp, start, end);
629 * There is no need to serialise calls to blkdev_issue_flush with
630 * i_mutex and doing so causes performance issues with concurrent
631 * O_SYNC writers to a block device.
633 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
634 if (error == -EOPNOTSUPP)
639 EXPORT_SYMBOL(blkdev_fsync);
642 * bdev_read_page() - Start reading a page from a block device
643 * @bdev: The device to read the page from
644 * @sector: The offset on the device to read the page to (need not be aligned)
645 * @page: The page to read
647 * On entry, the page should be locked. It will be unlocked when the page
648 * has been read. If the block driver implements rw_page synchronously,
649 * that will be true on exit from this function, but it need not be.
651 * Errors returned by this function are usually "soft", eg out of memory, or
652 * queue full; callers should try a different route to read this page rather
653 * than propagate an error back up the stack.
655 * Return: negative errno if an error occurs, 0 if submission was successful.
657 int bdev_read_page(struct block_device *bdev, sector_t sector,
660 const struct block_device_operations *ops = bdev->bd_disk->fops;
661 int result = -EOPNOTSUPP;
663 if (!ops->rw_page || bdev_get_integrity(bdev))
666 result = blk_queue_enter(bdev->bd_queue, 0);
669 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
671 blk_queue_exit(bdev->bd_queue);
674 EXPORT_SYMBOL_GPL(bdev_read_page);
677 * bdev_write_page() - Start writing a page to a block device
678 * @bdev: The device to write the page to
679 * @sector: The offset on the device to write the page to (need not be aligned)
680 * @page: The page to write
681 * @wbc: The writeback_control for the write
683 * On entry, the page should be locked and not currently under writeback.
684 * On exit, if the write started successfully, the page will be unlocked and
685 * under writeback. If the write failed already (eg the driver failed to
686 * queue the page to the device), the page will still be locked. If the
687 * caller is a ->writepage implementation, it will need to unlock the page.
689 * Errors returned by this function are usually "soft", eg out of memory, or
690 * queue full; callers should try a different route to write this page rather
691 * than propagate an error back up the stack.
693 * Return: negative errno if an error occurs, 0 if submission was successful.
695 int bdev_write_page(struct block_device *bdev, sector_t sector,
696 struct page *page, struct writeback_control *wbc)
699 const struct block_device_operations *ops = bdev->bd_disk->fops;
701 if (!ops->rw_page || bdev_get_integrity(bdev))
703 result = blk_queue_enter(bdev->bd_queue, 0);
707 set_page_writeback(page);
708 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
711 end_page_writeback(page);
713 clean_page_buffers(page);
716 blk_queue_exit(bdev->bd_queue);
719 EXPORT_SYMBOL_GPL(bdev_write_page);
725 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
726 static struct kmem_cache * bdev_cachep __read_mostly;
728 static struct inode *bdev_alloc_inode(struct super_block *sb)
730 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
733 return &ei->vfs_inode;
736 static void bdev_i_callback(struct rcu_head *head)
738 struct inode *inode = container_of(head, struct inode, i_rcu);
739 struct bdev_inode *bdi = BDEV_I(inode);
741 kmem_cache_free(bdev_cachep, bdi);
744 static void bdev_destroy_inode(struct inode *inode)
746 call_rcu(&inode->i_rcu, bdev_i_callback);
749 static void init_once(void *foo)
751 struct bdev_inode *ei = (struct bdev_inode *) foo;
752 struct block_device *bdev = &ei->bdev;
754 memset(bdev, 0, sizeof(*bdev));
755 mutex_init(&bdev->bd_mutex);
756 INIT_LIST_HEAD(&bdev->bd_list);
758 INIT_LIST_HEAD(&bdev->bd_holder_disks);
760 bdev->bd_bdi = &noop_backing_dev_info;
761 inode_init_once(&ei->vfs_inode);
762 /* Initialize mutex for freeze. */
763 mutex_init(&bdev->bd_fsfreeze_mutex);
766 static void bdev_evict_inode(struct inode *inode)
768 struct block_device *bdev = &BDEV_I(inode)->bdev;
769 truncate_inode_pages_final(&inode->i_data);
770 invalidate_inode_buffers(inode); /* is it needed here? */
772 spin_lock(&bdev_lock);
773 list_del_init(&bdev->bd_list);
774 spin_unlock(&bdev_lock);
775 /* Detach inode from wb early as bdi_put() may free bdi->wb */
776 inode_detach_wb(inode);
777 if (bdev->bd_bdi != &noop_backing_dev_info) {
778 bdi_put(bdev->bd_bdi);
779 bdev->bd_bdi = &noop_backing_dev_info;
783 static const struct super_operations bdev_sops = {
784 .statfs = simple_statfs,
785 .alloc_inode = bdev_alloc_inode,
786 .destroy_inode = bdev_destroy_inode,
787 .drop_inode = generic_delete_inode,
788 .evict_inode = bdev_evict_inode,
791 static struct dentry *bd_mount(struct file_system_type *fs_type,
792 int flags, const char *dev_name, void *data)
795 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
797 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
801 static struct file_system_type bd_type = {
804 .kill_sb = kill_anon_super,
807 struct super_block *blockdev_superblock __read_mostly;
808 EXPORT_SYMBOL_GPL(blockdev_superblock);
810 void __init bdev_cache_init(void)
813 static struct vfsmount *bd_mnt;
815 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
816 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
817 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
819 err = register_filesystem(&bd_type);
821 panic("Cannot register bdev pseudo-fs");
822 bd_mnt = kern_mount(&bd_type);
824 panic("Cannot create bdev pseudo-fs");
825 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
829 * Most likely _very_ bad one - but then it's hardly critical for small
830 * /dev and can be fixed when somebody will need really large one.
831 * Keep in mind that it will be fed through icache hash function too.
833 static inline unsigned long hash(dev_t dev)
835 return MAJOR(dev)+MINOR(dev);
838 static int bdev_test(struct inode *inode, void *data)
840 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
843 static int bdev_set(struct inode *inode, void *data)
845 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
849 static LIST_HEAD(all_bdevs);
852 * If there is a bdev inode for this device, unhash it so that it gets evicted
853 * as soon as last inode reference is dropped.
855 void bdev_unhash_inode(dev_t dev)
859 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
861 remove_inode_hash(inode);
866 struct block_device *bdget(dev_t dev)
868 struct block_device *bdev;
871 inode = iget5_locked(blockdev_superblock, hash(dev),
872 bdev_test, bdev_set, &dev);
877 bdev = &BDEV_I(inode)->bdev;
879 if (inode->i_state & I_NEW) {
880 bdev->bd_contains = NULL;
881 bdev->bd_super = NULL;
882 bdev->bd_inode = inode;
883 bdev->bd_block_size = i_blocksize(inode);
884 bdev->bd_part_count = 0;
885 bdev->bd_invalidated = 0;
886 inode->i_mode = S_IFBLK;
888 inode->i_bdev = bdev;
889 inode->i_data.a_ops = &def_blk_aops;
890 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
891 spin_lock(&bdev_lock);
892 list_add(&bdev->bd_list, &all_bdevs);
893 spin_unlock(&bdev_lock);
894 unlock_new_inode(inode);
899 EXPORT_SYMBOL(bdget);
902 * bdgrab -- Grab a reference to an already referenced block device
903 * @bdev: Block device to grab a reference to.
905 struct block_device *bdgrab(struct block_device *bdev)
907 ihold(bdev->bd_inode);
910 EXPORT_SYMBOL(bdgrab);
912 long nr_blockdev_pages(void)
914 struct block_device *bdev;
916 spin_lock(&bdev_lock);
917 list_for_each_entry(bdev, &all_bdevs, bd_list) {
918 ret += bdev->bd_inode->i_mapping->nrpages;
920 spin_unlock(&bdev_lock);
924 void bdput(struct block_device *bdev)
926 iput(bdev->bd_inode);
929 EXPORT_SYMBOL(bdput);
931 static struct block_device *bd_acquire(struct inode *inode)
933 struct block_device *bdev;
935 spin_lock(&bdev_lock);
936 bdev = inode->i_bdev;
937 if (bdev && !inode_unhashed(bdev->bd_inode)) {
939 spin_unlock(&bdev_lock);
942 spin_unlock(&bdev_lock);
945 * i_bdev references block device inode that was already shut down
946 * (corresponding device got removed). Remove the reference and look
947 * up block device inode again just in case new device got
948 * reestablished under the same device number.
953 bdev = bdget(inode->i_rdev);
955 spin_lock(&bdev_lock);
956 if (!inode->i_bdev) {
958 * We take an additional reference to bd_inode,
959 * and it's released in clear_inode() of inode.
960 * So, we can access it via ->i_mapping always
964 inode->i_bdev = bdev;
965 inode->i_mapping = bdev->bd_inode->i_mapping;
967 spin_unlock(&bdev_lock);
972 /* Call when you free inode */
974 void bd_forget(struct inode *inode)
976 struct block_device *bdev = NULL;
978 spin_lock(&bdev_lock);
979 if (!sb_is_blkdev_sb(inode->i_sb))
980 bdev = inode->i_bdev;
981 inode->i_bdev = NULL;
982 inode->i_mapping = &inode->i_data;
983 spin_unlock(&bdev_lock);
990 * bd_may_claim - test whether a block device can be claimed
991 * @bdev: block device of interest
992 * @whole: whole block device containing @bdev, may equal @bdev
993 * @holder: holder trying to claim @bdev
995 * Test whether @bdev can be claimed by @holder.
998 * spin_lock(&bdev_lock).
1001 * %true if @bdev can be claimed, %false otherwise.
1003 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1006 if (bdev->bd_holder == holder)
1007 return true; /* already a holder */
1008 else if (bdev->bd_holder != NULL)
1009 return false; /* held by someone else */
1010 else if (whole == bdev)
1011 return true; /* is a whole device which isn't held */
1013 else if (whole->bd_holder == bd_may_claim)
1014 return true; /* is a partition of a device that is being partitioned */
1015 else if (whole->bd_holder != NULL)
1016 return false; /* is a partition of a held device */
1018 return true; /* is a partition of an un-held device */
1022 * bd_prepare_to_claim - prepare to claim a block device
1023 * @bdev: block device of interest
1024 * @whole: the whole device containing @bdev, may equal @bdev
1025 * @holder: holder trying to claim @bdev
1027 * Prepare to claim @bdev. This function fails if @bdev is already
1028 * claimed by another holder and waits if another claiming is in
1029 * progress. This function doesn't actually claim. On successful
1030 * return, the caller has ownership of bd_claiming and bd_holder[s].
1033 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1034 * it multiple times.
1037 * 0 if @bdev can be claimed, -EBUSY otherwise.
1039 static int bd_prepare_to_claim(struct block_device *bdev,
1040 struct block_device *whole, void *holder)
1043 /* if someone else claimed, fail */
1044 if (!bd_may_claim(bdev, whole, holder))
1047 /* if claiming is already in progress, wait for it to finish */
1048 if (whole->bd_claiming) {
1049 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1052 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1053 spin_unlock(&bdev_lock);
1055 finish_wait(wq, &wait);
1056 spin_lock(&bdev_lock);
1064 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1066 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1071 * Now that we hold gendisk reference we make sure bdev we looked up is
1072 * not stale. If it is, it means device got removed and created before
1073 * we looked up gendisk and we fail open in such case. Associating
1074 * unhashed bdev with newly created gendisk could lead to two bdevs
1075 * (and thus two independent caches) being associated with one device
1078 if (inode_unhashed(bdev->bd_inode)) {
1079 put_disk_and_module(disk);
1086 * bd_start_claiming - start claiming a block device
1087 * @bdev: block device of interest
1088 * @holder: holder trying to claim @bdev
1090 * @bdev is about to be opened exclusively. Check @bdev can be opened
1091 * exclusively and mark that an exclusive open is in progress. Each
1092 * successful call to this function must be matched with a call to
1093 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1096 * This function is used to gain exclusive access to the block device
1097 * without actually causing other exclusive open attempts to fail. It
1098 * should be used when the open sequence itself requires exclusive
1099 * access but may subsequently fail.
1105 * Pointer to the block device containing @bdev on success, ERR_PTR()
1108 static struct block_device *bd_start_claiming(struct block_device *bdev,
1111 struct gendisk *disk;
1112 struct block_device *whole;
1118 * @bdev might not have been initialized properly yet, look up
1119 * and grab the outer block device the hard way.
1121 disk = bdev_get_gendisk(bdev, &partno);
1123 return ERR_PTR(-ENXIO);
1126 * Normally, @bdev should equal what's returned from bdget_disk()
1127 * if partno is 0; however, some drivers (floppy) use multiple
1128 * bdev's for the same physical device and @bdev may be one of the
1129 * aliases. Keep @bdev if partno is 0. This means claimer
1130 * tracking is broken for those devices but it has always been that
1134 whole = bdget_disk(disk, 0);
1136 whole = bdgrab(bdev);
1138 put_disk_and_module(disk);
1140 return ERR_PTR(-ENOMEM);
1142 /* prepare to claim, if successful, mark claiming in progress */
1143 spin_lock(&bdev_lock);
1145 err = bd_prepare_to_claim(bdev, whole, holder);
1147 whole->bd_claiming = holder;
1148 spin_unlock(&bdev_lock);
1151 spin_unlock(&bdev_lock);
1153 return ERR_PTR(err);
1158 struct bd_holder_disk {
1159 struct list_head list;
1160 struct gendisk *disk;
1164 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1165 struct gendisk *disk)
1167 struct bd_holder_disk *holder;
1169 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1170 if (holder->disk == disk)
1175 static int add_symlink(struct kobject *from, struct kobject *to)
1177 return sysfs_create_link(from, to, kobject_name(to));
1180 static void del_symlink(struct kobject *from, struct kobject *to)
1182 sysfs_remove_link(from, kobject_name(to));
1186 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1187 * @bdev: the claimed slave bdev
1188 * @disk: the holding disk
1190 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1192 * This functions creates the following sysfs symlinks.
1194 * - from "slaves" directory of the holder @disk to the claimed @bdev
1195 * - from "holders" directory of the @bdev to the holder @disk
1197 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1198 * passed to bd_link_disk_holder(), then:
1200 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1201 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1203 * The caller must have claimed @bdev before calling this function and
1204 * ensure that both @bdev and @disk are valid during the creation and
1205 * lifetime of these symlinks.
1211 * 0 on success, -errno on failure.
1213 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1215 struct bd_holder_disk *holder;
1218 mutex_lock(&bdev->bd_mutex);
1220 WARN_ON_ONCE(!bdev->bd_holder);
1222 /* FIXME: remove the following once add_disk() handles errors */
1223 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1226 holder = bd_find_holder_disk(bdev, disk);
1232 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1238 INIT_LIST_HEAD(&holder->list);
1239 holder->disk = disk;
1242 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1246 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1250 * bdev could be deleted beneath us which would implicitly destroy
1251 * the holder directory. Hold on to it.
1253 kobject_get(bdev->bd_part->holder_dir);
1255 list_add(&holder->list, &bdev->bd_holder_disks);
1259 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1263 mutex_unlock(&bdev->bd_mutex);
1266 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1269 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1270 * @bdev: the calimed slave bdev
1271 * @disk: the holding disk
1273 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1278 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1280 struct bd_holder_disk *holder;
1282 mutex_lock(&bdev->bd_mutex);
1284 holder = bd_find_holder_disk(bdev, disk);
1286 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1287 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1288 del_symlink(bdev->bd_part->holder_dir,
1289 &disk_to_dev(disk)->kobj);
1290 kobject_put(bdev->bd_part->holder_dir);
1291 list_del_init(&holder->list);
1295 mutex_unlock(&bdev->bd_mutex);
1297 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1301 * flush_disk - invalidates all buffer-cache entries on a disk
1303 * @bdev: struct block device to be flushed
1304 * @kill_dirty: flag to guide handling of dirty inodes
1306 * Invalidates all buffer-cache entries on a disk. It should be called
1307 * when a disk has been changed -- either by a media change or online
1310 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1312 if (__invalidate_device(bdev, kill_dirty)) {
1313 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1314 "resized disk %s\n",
1315 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1320 if (disk_part_scan_enabled(bdev->bd_disk))
1321 bdev->bd_invalidated = 1;
1325 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1326 * @disk: struct gendisk to check
1327 * @bdev: struct bdev to adjust.
1328 * @verbose: if %true log a message about a size change if there is any
1330 * This routine checks to see if the bdev size does not match the disk size
1331 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1334 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev,
1337 loff_t disk_size, bdev_size;
1339 disk_size = (loff_t)get_capacity(disk) << 9;
1340 bdev_size = i_size_read(bdev->bd_inode);
1341 if (disk_size != bdev_size) {
1344 "%s: detected capacity change from %lld to %lld\n",
1345 disk->disk_name, bdev_size, disk_size);
1347 i_size_write(bdev->bd_inode, disk_size);
1348 if (bdev_size > disk_size)
1349 flush_disk(bdev, false);
1354 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1355 * @disk: struct gendisk to be revalidated
1357 * This routine is a wrapper for lower-level driver's revalidate_disk
1358 * call-backs. It is used to do common pre and post operations needed
1359 * for all revalidate_disk operations.
1361 int revalidate_disk(struct gendisk *disk)
1363 struct block_device *bdev;
1366 if (disk->fops->revalidate_disk)
1367 ret = disk->fops->revalidate_disk(disk);
1368 bdev = bdget_disk(disk, 0);
1372 mutex_lock(&bdev->bd_mutex);
1373 check_disk_size_change(disk, bdev, ret == 0);
1374 bdev->bd_invalidated = 0;
1375 mutex_unlock(&bdev->bd_mutex);
1379 EXPORT_SYMBOL(revalidate_disk);
1382 * This routine checks whether a removable media has been changed,
1383 * and invalidates all buffer-cache-entries in that case. This
1384 * is a relatively slow routine, so we have to try to minimize using
1385 * it. Thus it is called only upon a 'mount' or 'open'. This
1386 * is the best way of combining speed and utility, I think.
1387 * People changing diskettes in the middle of an operation deserve
1390 int check_disk_change(struct block_device *bdev)
1392 struct gendisk *disk = bdev->bd_disk;
1393 const struct block_device_operations *bdops = disk->fops;
1394 unsigned int events;
1396 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1397 DISK_EVENT_EJECT_REQUEST);
1398 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1401 flush_disk(bdev, true);
1402 if (bdops->revalidate_disk)
1403 bdops->revalidate_disk(bdev->bd_disk);
1407 EXPORT_SYMBOL(check_disk_change);
1409 void bd_set_size(struct block_device *bdev, loff_t size)
1411 unsigned bsize = bdev_logical_block_size(bdev);
1413 inode_lock(bdev->bd_inode);
1414 i_size_write(bdev->bd_inode, size);
1415 inode_unlock(bdev->bd_inode);
1416 while (bsize < PAGE_SIZE) {
1421 bdev->bd_block_size = bsize;
1422 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1424 EXPORT_SYMBOL(bd_set_size);
1426 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1431 * mutex_lock(part->bd_mutex)
1432 * mutex_lock_nested(whole->bd_mutex, 1)
1435 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1437 struct gendisk *disk;
1441 bool first_open = false;
1443 if (mode & FMODE_READ)
1445 if (mode & FMODE_WRITE)
1448 * hooks: /n/, see "layering violations".
1451 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1461 disk = bdev_get_gendisk(bdev, &partno);
1465 disk_block_events(disk);
1466 mutex_lock_nested(&bdev->bd_mutex, for_part);
1467 if (!bdev->bd_openers) {
1469 bdev->bd_disk = disk;
1470 bdev->bd_queue = disk->queue;
1471 bdev->bd_contains = bdev;
1472 bdev->bd_partno = partno;
1476 bdev->bd_part = disk_get_part(disk, partno);
1481 if (disk->fops->open) {
1482 ret = disk->fops->open(bdev, mode);
1483 if (ret == -ERESTARTSYS) {
1484 /* Lost a race with 'disk' being
1485 * deleted, try again.
1488 disk_put_part(bdev->bd_part);
1489 bdev->bd_part = NULL;
1490 bdev->bd_disk = NULL;
1491 bdev->bd_queue = NULL;
1492 mutex_unlock(&bdev->bd_mutex);
1493 disk_unblock_events(disk);
1494 put_disk_and_module(disk);
1500 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1503 * If the device is invalidated, rescan partition
1504 * if open succeeded or failed with -ENOMEDIUM.
1505 * The latter is necessary to prevent ghost
1506 * partitions on a removed medium.
1508 if (bdev->bd_invalidated) {
1510 rescan_partitions(disk, bdev);
1511 else if (ret == -ENOMEDIUM)
1512 invalidate_partitions(disk, bdev);
1518 struct block_device *whole;
1519 whole = bdget_disk(disk, 0);
1524 ret = __blkdev_get(whole, mode, 1);
1527 bdev->bd_contains = whole;
1528 bdev->bd_part = disk_get_part(disk, partno);
1529 if (!(disk->flags & GENHD_FL_UP) ||
1530 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1534 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1537 if (bdev->bd_bdi == &noop_backing_dev_info)
1538 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1540 if (bdev->bd_contains == bdev) {
1542 if (bdev->bd_disk->fops->open)
1543 ret = bdev->bd_disk->fops->open(bdev, mode);
1544 /* the same as first opener case, read comment there */
1545 if (bdev->bd_invalidated) {
1547 rescan_partitions(bdev->bd_disk, bdev);
1548 else if (ret == -ENOMEDIUM)
1549 invalidate_partitions(bdev->bd_disk, bdev);
1552 goto out_unlock_bdev;
1557 bdev->bd_part_count++;
1558 mutex_unlock(&bdev->bd_mutex);
1559 disk_unblock_events(disk);
1560 /* only one opener holds refs to the module and disk */
1562 put_disk_and_module(disk);
1566 disk_put_part(bdev->bd_part);
1567 bdev->bd_disk = NULL;
1568 bdev->bd_part = NULL;
1569 bdev->bd_queue = NULL;
1570 if (bdev != bdev->bd_contains)
1571 __blkdev_put(bdev->bd_contains, mode, 1);
1572 bdev->bd_contains = NULL;
1574 mutex_unlock(&bdev->bd_mutex);
1575 disk_unblock_events(disk);
1576 put_disk_and_module(disk);
1584 * blkdev_get - open a block device
1585 * @bdev: block_device to open
1586 * @mode: FMODE_* mask
1587 * @holder: exclusive holder identifier
1589 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1590 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1591 * @holder is invalid. Exclusive opens may nest for the same @holder.
1593 * On success, the reference count of @bdev is unchanged. On failure,
1600 * 0 on success, -errno on failure.
1602 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1604 struct block_device *whole = NULL;
1607 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1609 if ((mode & FMODE_EXCL) && holder) {
1610 whole = bd_start_claiming(bdev, holder);
1611 if (IS_ERR(whole)) {
1613 return PTR_ERR(whole);
1617 res = __blkdev_get(bdev, mode, 0);
1620 struct gendisk *disk = whole->bd_disk;
1622 /* finish claiming */
1623 mutex_lock(&bdev->bd_mutex);
1624 spin_lock(&bdev_lock);
1627 BUG_ON(!bd_may_claim(bdev, whole, holder));
1629 * Note that for a whole device bd_holders
1630 * will be incremented twice, and bd_holder
1631 * will be set to bd_may_claim before being
1634 whole->bd_holders++;
1635 whole->bd_holder = bd_may_claim;
1637 bdev->bd_holder = holder;
1640 /* tell others that we're done */
1641 BUG_ON(whole->bd_claiming != holder);
1642 whole->bd_claiming = NULL;
1643 wake_up_bit(&whole->bd_claiming, 0);
1645 spin_unlock(&bdev_lock);
1648 * Block event polling for write claims if requested. Any
1649 * write holder makes the write_holder state stick until
1650 * all are released. This is good enough and tracking
1651 * individual writeable reference is too fragile given the
1652 * way @mode is used in blkdev_get/put().
1654 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1655 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1656 bdev->bd_write_holder = true;
1657 disk_block_events(disk);
1660 mutex_unlock(&bdev->bd_mutex);
1666 EXPORT_SYMBOL(blkdev_get);
1669 * blkdev_get_by_path - open a block device by name
1670 * @path: path to the block device to open
1671 * @mode: FMODE_* mask
1672 * @holder: exclusive holder identifier
1674 * Open the blockdevice described by the device file at @path. @mode
1675 * and @holder are identical to blkdev_get().
1677 * On success, the returned block_device has reference count of one.
1683 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1685 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1688 struct block_device *bdev;
1691 bdev = lookup_bdev(path);
1695 err = blkdev_get(bdev, mode, holder);
1697 return ERR_PTR(err);
1699 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1700 blkdev_put(bdev, mode);
1701 return ERR_PTR(-EACCES);
1706 EXPORT_SYMBOL(blkdev_get_by_path);
1709 * blkdev_get_by_dev - open a block device by device number
1710 * @dev: device number of block device to open
1711 * @mode: FMODE_* mask
1712 * @holder: exclusive holder identifier
1714 * Open the blockdevice described by device number @dev. @mode and
1715 * @holder are identical to blkdev_get().
1717 * Use it ONLY if you really do not have anything better - i.e. when
1718 * you are behind a truly sucky interface and all you are given is a
1719 * device number. _Never_ to be used for internal purposes. If you
1720 * ever need it - reconsider your API.
1722 * On success, the returned block_device has reference count of one.
1728 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1730 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1732 struct block_device *bdev;
1737 return ERR_PTR(-ENOMEM);
1739 err = blkdev_get(bdev, mode, holder);
1741 return ERR_PTR(err);
1745 EXPORT_SYMBOL(blkdev_get_by_dev);
1747 static int blkdev_open(struct inode * inode, struct file * filp)
1749 struct block_device *bdev;
1752 * Preserve backwards compatibility and allow large file access
1753 * even if userspace doesn't ask for it explicitly. Some mkfs
1754 * binary needs it. We might want to drop this workaround
1755 * during an unstable branch.
1757 filp->f_flags |= O_LARGEFILE;
1759 filp->f_mode |= FMODE_NOWAIT;
1761 if (filp->f_flags & O_NDELAY)
1762 filp->f_mode |= FMODE_NDELAY;
1763 if (filp->f_flags & O_EXCL)
1764 filp->f_mode |= FMODE_EXCL;
1765 if ((filp->f_flags & O_ACCMODE) == 3)
1766 filp->f_mode |= FMODE_WRITE_IOCTL;
1768 bdev = bd_acquire(inode);
1772 filp->f_mapping = bdev->bd_inode->i_mapping;
1773 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1775 return blkdev_get(bdev, filp->f_mode, filp);
1778 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1780 struct gendisk *disk = bdev->bd_disk;
1781 struct block_device *victim = NULL;
1783 mutex_lock_nested(&bdev->bd_mutex, for_part);
1785 bdev->bd_part_count--;
1787 if (!--bdev->bd_openers) {
1788 WARN_ON_ONCE(bdev->bd_holders);
1789 sync_blockdev(bdev);
1792 bdev_write_inode(bdev);
1794 if (bdev->bd_contains == bdev) {
1795 if (disk->fops->release)
1796 disk->fops->release(disk, mode);
1798 if (!bdev->bd_openers) {
1799 disk_put_part(bdev->bd_part);
1800 bdev->bd_part = NULL;
1801 bdev->bd_disk = NULL;
1802 if (bdev != bdev->bd_contains)
1803 victim = bdev->bd_contains;
1804 bdev->bd_contains = NULL;
1806 put_disk_and_module(disk);
1808 mutex_unlock(&bdev->bd_mutex);
1811 __blkdev_put(victim, mode, 1);
1814 void blkdev_put(struct block_device *bdev, fmode_t mode)
1816 mutex_lock(&bdev->bd_mutex);
1818 if (mode & FMODE_EXCL) {
1822 * Release a claim on the device. The holder fields
1823 * are protected with bdev_lock. bd_mutex is to
1824 * synchronize disk_holder unlinking.
1826 spin_lock(&bdev_lock);
1828 WARN_ON_ONCE(--bdev->bd_holders < 0);
1829 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1831 /* bd_contains might point to self, check in a separate step */
1832 if ((bdev_free = !bdev->bd_holders))
1833 bdev->bd_holder = NULL;
1834 if (!bdev->bd_contains->bd_holders)
1835 bdev->bd_contains->bd_holder = NULL;
1837 spin_unlock(&bdev_lock);
1840 * If this was the last claim, remove holder link and
1841 * unblock evpoll if it was a write holder.
1843 if (bdev_free && bdev->bd_write_holder) {
1844 disk_unblock_events(bdev->bd_disk);
1845 bdev->bd_write_holder = false;
1850 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1851 * event. This is to ensure detection of media removal commanded
1852 * from userland - e.g. eject(1).
1854 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1856 mutex_unlock(&bdev->bd_mutex);
1858 __blkdev_put(bdev, mode, 0);
1860 EXPORT_SYMBOL(blkdev_put);
1862 static int blkdev_close(struct inode * inode, struct file * filp)
1864 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1865 blkdev_put(bdev, filp->f_mode);
1869 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1871 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1872 fmode_t mode = file->f_mode;
1875 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1876 * to updated it before every ioctl.
1878 if (file->f_flags & O_NDELAY)
1879 mode |= FMODE_NDELAY;
1881 mode &= ~FMODE_NDELAY;
1883 return blkdev_ioctl(bdev, mode, cmd, arg);
1887 * Write data to the block device. Only intended for the block device itself
1888 * and the raw driver which basically is a fake block device.
1890 * Does not take i_mutex for the write and thus is not for general purpose
1893 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1895 struct file *file = iocb->ki_filp;
1896 struct inode *bd_inode = bdev_file_inode(file);
1897 loff_t size = i_size_read(bd_inode);
1898 struct blk_plug plug;
1901 if (bdev_read_only(I_BDEV(bd_inode)))
1904 if (!iov_iter_count(from))
1907 if (iocb->ki_pos >= size)
1910 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1913 iov_iter_truncate(from, size - iocb->ki_pos);
1915 blk_start_plug(&plug);
1916 ret = __generic_file_write_iter(iocb, from);
1918 ret = generic_write_sync(iocb, ret);
1919 blk_finish_plug(&plug);
1922 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1924 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1926 struct file *file = iocb->ki_filp;
1927 struct inode *bd_inode = bdev_file_inode(file);
1928 loff_t size = i_size_read(bd_inode);
1929 loff_t pos = iocb->ki_pos;
1935 iov_iter_truncate(to, size);
1936 return generic_file_read_iter(iocb, to);
1938 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1941 * Try to release a page associated with block device when the system
1942 * is under memory pressure.
1944 static int blkdev_releasepage(struct page *page, gfp_t wait)
1946 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1948 if (super && super->s_op->bdev_try_to_free_page)
1949 return super->s_op->bdev_try_to_free_page(super, page, wait);
1951 return try_to_free_buffers(page);
1954 static int blkdev_writepages(struct address_space *mapping,
1955 struct writeback_control *wbc)
1957 return generic_writepages(mapping, wbc);
1960 static const struct address_space_operations def_blk_aops = {
1961 .readpage = blkdev_readpage,
1962 .readpages = blkdev_readpages,
1963 .writepage = blkdev_writepage,
1964 .write_begin = blkdev_write_begin,
1965 .write_end = blkdev_write_end,
1966 .writepages = blkdev_writepages,
1967 .releasepage = blkdev_releasepage,
1968 .direct_IO = blkdev_direct_IO,
1969 .is_dirty_writeback = buffer_check_dirty_writeback,
1972 #define BLKDEV_FALLOC_FL_SUPPORTED \
1973 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1974 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1976 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1979 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1980 struct address_space *mapping;
1981 loff_t end = start + len - 1;
1985 /* Fail if we don't recognize the flags. */
1986 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1989 /* Don't go off the end of the device. */
1990 isize = i_size_read(bdev->bd_inode);
1994 if (mode & FALLOC_FL_KEEP_SIZE) {
1995 len = isize - start;
1996 end = start + len - 1;
2002 * Don't allow IO that isn't aligned to logical block size.
2004 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2007 /* Invalidate the page cache, including dirty pages. */
2008 mapping = bdev->bd_inode->i_mapping;
2009 truncate_inode_pages_range(mapping, start, end);
2012 case FALLOC_FL_ZERO_RANGE:
2013 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2014 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2015 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2017 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2018 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2019 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2021 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2022 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2032 * Invalidate again; if someone wandered in and dirtied a page,
2033 * the caller will be given -EBUSY. The third argument is
2034 * inclusive, so the rounding here is safe.
2036 return invalidate_inode_pages2_range(mapping,
2037 start >> PAGE_SHIFT,
2041 const struct file_operations def_blk_fops = {
2042 .open = blkdev_open,
2043 .release = blkdev_close,
2044 .llseek = block_llseek,
2045 .read_iter = blkdev_read_iter,
2046 .write_iter = blkdev_write_iter,
2047 .mmap = generic_file_mmap,
2048 .fsync = blkdev_fsync,
2049 .unlocked_ioctl = block_ioctl,
2050 #ifdef CONFIG_COMPAT
2051 .compat_ioctl = compat_blkdev_ioctl,
2053 .splice_read = generic_file_splice_read,
2054 .splice_write = iter_file_splice_write,
2055 .fallocate = blkdev_fallocate,
2058 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2061 mm_segment_t old_fs = get_fs();
2063 res = blkdev_ioctl(bdev, 0, cmd, arg);
2068 EXPORT_SYMBOL(ioctl_by_bdev);
2071 * lookup_bdev - lookup a struct block_device by name
2072 * @pathname: special file representing the block device
2074 * Get a reference to the blockdevice at @pathname in the current
2075 * namespace if possible and return it. Return ERR_PTR(error)
2078 struct block_device *lookup_bdev(const char *pathname)
2080 struct block_device *bdev;
2081 struct inode *inode;
2085 if (!pathname || !*pathname)
2086 return ERR_PTR(-EINVAL);
2088 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2090 return ERR_PTR(error);
2092 inode = d_backing_inode(path.dentry);
2094 if (!S_ISBLK(inode->i_mode))
2097 if (!may_open_dev(&path))
2100 bdev = bd_acquire(inode);
2107 bdev = ERR_PTR(error);
2110 EXPORT_SYMBOL(lookup_bdev);
2112 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2114 struct super_block *sb = get_super(bdev);
2119 * no need to lock the super, get_super holds the
2120 * read mutex so the filesystem cannot go away
2121 * under us (->put_super runs with the write lock
2124 shrink_dcache_sb(sb);
2125 res = invalidate_inodes(sb, kill_dirty);
2128 invalidate_bdev(bdev);
2131 EXPORT_SYMBOL(__invalidate_device);
2133 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2135 struct inode *inode, *old_inode = NULL;
2137 spin_lock(&blockdev_superblock->s_inode_list_lock);
2138 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2139 struct address_space *mapping = inode->i_mapping;
2140 struct block_device *bdev;
2142 spin_lock(&inode->i_lock);
2143 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2144 mapping->nrpages == 0) {
2145 spin_unlock(&inode->i_lock);
2149 spin_unlock(&inode->i_lock);
2150 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2152 * We hold a reference to 'inode' so it couldn't have been
2153 * removed from s_inodes list while we dropped the
2154 * s_inode_list_lock We cannot iput the inode now as we can
2155 * be holding the last reference and we cannot iput it under
2156 * s_inode_list_lock. So we keep the reference and iput it
2161 bdev = I_BDEV(inode);
2163 mutex_lock(&bdev->bd_mutex);
2164 if (bdev->bd_openers)
2166 mutex_unlock(&bdev->bd_mutex);
2168 spin_lock(&blockdev_superblock->s_inode_list_lock);
2170 spin_unlock(&blockdev_superblock->s_inode_list_lock);