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);
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 (vecs != inline_vecs)
266 if (unlikely(bio.bi_status))
267 return blk_status_to_errno(bio.bi_status);
274 struct task_struct *waiter;
279 bool should_dirty : 1;
284 static struct bio_set *blkdev_dio_pool __read_mostly;
286 static void blkdev_bio_end_io(struct bio *bio)
288 struct blkdev_dio *dio = bio->bi_private;
289 bool should_dirty = dio->should_dirty;
291 if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) {
292 if (bio->bi_status && !dio->bio.bi_status)
293 dio->bio.bi_status = bio->bi_status;
296 struct kiocb *iocb = dio->iocb;
299 if (likely(!dio->bio.bi_status)) {
303 ret = blk_status_to_errno(dio->bio.bi_status);
306 dio->iocb->ki_complete(iocb, ret, 0);
309 struct task_struct *waiter = dio->waiter;
311 WRITE_ONCE(dio->waiter, NULL);
312 wake_up_process(waiter);
317 bio_check_pages_dirty(bio);
319 struct bio_vec *bvec;
322 bio_for_each_segment_all(bvec, bio, i)
323 put_page(bvec->bv_page);
329 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
331 struct file *file = iocb->ki_filp;
332 struct inode *inode = bdev_file_inode(file);
333 struct block_device *bdev = I_BDEV(inode);
334 struct blk_plug plug;
335 struct blkdev_dio *dio;
337 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
338 loff_t pos = iocb->ki_pos;
339 blk_qc_t qc = BLK_QC_T_NONE;
342 if ((pos | iov_iter_alignment(iter)) &
343 (bdev_logical_block_size(bdev) - 1))
346 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, blkdev_dio_pool);
347 bio_get(bio); /* extra ref for the completion handler */
349 dio = container_of(bio, struct blkdev_dio, bio);
350 dio->is_sync = is_sync = is_sync_kiocb(iocb);
352 dio->waiter = current;
357 dio->multi_bio = false;
358 dio->should_dirty = is_read && (iter->type == ITER_IOVEC);
360 blk_start_plug(&plug);
363 bio->bi_iter.bi_sector = pos >> 9;
364 bio->bi_write_hint = iocb->ki_hint;
365 bio->bi_private = dio;
366 bio->bi_end_io = blkdev_bio_end_io;
368 ret = bio_iov_iter_get_pages(bio, iter);
370 bio->bi_status = BLK_STS_IOERR;
376 bio->bi_opf = REQ_OP_READ;
377 if (dio->should_dirty)
378 bio_set_pages_dirty(bio);
380 bio->bi_opf = dio_bio_write_op(iocb);
381 task_io_account_write(bio->bi_iter.bi_size);
384 dio->size += bio->bi_iter.bi_size;
385 pos += bio->bi_iter.bi_size;
387 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
389 qc = submit_bio(bio);
393 if (!dio->multi_bio) {
394 dio->multi_bio = true;
395 atomic_set(&dio->ref, 2);
397 atomic_inc(&dio->ref);
401 bio = bio_alloc(GFP_KERNEL, nr_pages);
403 blk_finish_plug(&plug);
409 set_current_state(TASK_UNINTERRUPTIBLE);
410 if (!READ_ONCE(dio->waiter))
413 if (!(iocb->ki_flags & IOCB_HIPRI) ||
414 !blk_mq_poll(bdev_get_queue(bdev), qc))
417 __set_current_state(TASK_RUNNING);
420 ret = blk_status_to_errno(dio->bio.bi_status);
429 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
433 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
436 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
437 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
439 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
442 static __init int blkdev_init(void)
444 blkdev_dio_pool = bioset_create(4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
445 if (!blkdev_dio_pool)
449 module_init(blkdev_init);
451 int __sync_blockdev(struct block_device *bdev, int wait)
456 return filemap_flush(bdev->bd_inode->i_mapping);
457 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
461 * Write out and wait upon all the dirty data associated with a block
462 * device via its mapping. Does not take the superblock lock.
464 int sync_blockdev(struct block_device *bdev)
466 return __sync_blockdev(bdev, 1);
468 EXPORT_SYMBOL(sync_blockdev);
471 * Write out and wait upon all dirty data associated with this
472 * device. Filesystem data as well as the underlying block
473 * device. Takes the superblock lock.
475 int fsync_bdev(struct block_device *bdev)
477 struct super_block *sb = get_super(bdev);
479 int res = sync_filesystem(sb);
483 return sync_blockdev(bdev);
485 EXPORT_SYMBOL(fsync_bdev);
488 * freeze_bdev -- lock a filesystem and force it into a consistent state
489 * @bdev: blockdevice to lock
491 * If a superblock is found on this device, we take the s_umount semaphore
492 * on it to make sure nobody unmounts until the snapshot creation is done.
493 * The reference counter (bd_fsfreeze_count) guarantees that only the last
494 * unfreeze process can unfreeze the frozen filesystem actually when multiple
495 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
496 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
499 struct super_block *freeze_bdev(struct block_device *bdev)
501 struct super_block *sb;
504 mutex_lock(&bdev->bd_fsfreeze_mutex);
505 if (++bdev->bd_fsfreeze_count > 1) {
507 * We don't even need to grab a reference - the first call
508 * to freeze_bdev grab an active reference and only the last
509 * thaw_bdev drops it.
511 sb = get_super(bdev);
514 mutex_unlock(&bdev->bd_fsfreeze_mutex);
518 sb = get_active_super(bdev);
521 if (sb->s_op->freeze_super)
522 error = sb->s_op->freeze_super(sb);
524 error = freeze_super(sb);
526 deactivate_super(sb);
527 bdev->bd_fsfreeze_count--;
528 mutex_unlock(&bdev->bd_fsfreeze_mutex);
529 return ERR_PTR(error);
531 deactivate_super(sb);
534 mutex_unlock(&bdev->bd_fsfreeze_mutex);
535 return sb; /* thaw_bdev releases s->s_umount */
537 EXPORT_SYMBOL(freeze_bdev);
540 * thaw_bdev -- unlock filesystem
541 * @bdev: blockdevice to unlock
542 * @sb: associated superblock
544 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
546 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
550 mutex_lock(&bdev->bd_fsfreeze_mutex);
551 if (!bdev->bd_fsfreeze_count)
555 if (--bdev->bd_fsfreeze_count > 0)
561 if (sb->s_op->thaw_super)
562 error = sb->s_op->thaw_super(sb);
564 error = thaw_super(sb);
566 bdev->bd_fsfreeze_count++;
568 mutex_unlock(&bdev->bd_fsfreeze_mutex);
571 EXPORT_SYMBOL(thaw_bdev);
573 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
575 return block_write_full_page(page, blkdev_get_block, wbc);
578 static int blkdev_readpage(struct file * file, struct page * page)
580 return block_read_full_page(page, blkdev_get_block);
583 static int blkdev_readpages(struct file *file, struct address_space *mapping,
584 struct list_head *pages, unsigned nr_pages)
586 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
589 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
590 loff_t pos, unsigned len, unsigned flags,
591 struct page **pagep, void **fsdata)
593 return block_write_begin(mapping, pos, len, flags, pagep,
597 static int blkdev_write_end(struct file *file, struct address_space *mapping,
598 loff_t pos, unsigned len, unsigned copied,
599 struct page *page, void *fsdata)
602 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
612 * for a block special file file_inode(file)->i_size is zero
613 * so we compute the size by hand (just as in block_read/write above)
615 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
617 struct inode *bd_inode = bdev_file_inode(file);
620 inode_lock(bd_inode);
621 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
622 inode_unlock(bd_inode);
626 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
628 struct inode *bd_inode = bdev_file_inode(filp);
629 struct block_device *bdev = I_BDEV(bd_inode);
632 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
637 * There is no need to serialise calls to blkdev_issue_flush with
638 * i_mutex and doing so causes performance issues with concurrent
639 * O_SYNC writers to a block device.
641 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
642 if (error == -EOPNOTSUPP)
647 EXPORT_SYMBOL(blkdev_fsync);
650 * bdev_read_page() - Start reading a page from a block device
651 * @bdev: The device to read the page from
652 * @sector: The offset on the device to read the page to (need not be aligned)
653 * @page: The page to read
655 * On entry, the page should be locked. It will be unlocked when the page
656 * has been read. If the block driver implements rw_page synchronously,
657 * that will be true on exit from this function, but it need not be.
659 * Errors returned by this function are usually "soft", eg out of memory, or
660 * queue full; callers should try a different route to read this page rather
661 * than propagate an error back up the stack.
663 * Return: negative errno if an error occurs, 0 if submission was successful.
665 int bdev_read_page(struct block_device *bdev, sector_t sector,
668 const struct block_device_operations *ops = bdev->bd_disk->fops;
669 int result = -EOPNOTSUPP;
671 if (!ops->rw_page || bdev_get_integrity(bdev))
674 result = blk_queue_enter(bdev->bd_queue, false);
677 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false);
678 blk_queue_exit(bdev->bd_queue);
681 EXPORT_SYMBOL_GPL(bdev_read_page);
684 * bdev_write_page() - Start writing a page to a block device
685 * @bdev: The device to write the page to
686 * @sector: The offset on the device to write the page to (need not be aligned)
687 * @page: The page to write
688 * @wbc: The writeback_control for the write
690 * On entry, the page should be locked and not currently under writeback.
691 * On exit, if the write started successfully, the page will be unlocked and
692 * under writeback. If the write failed already (eg the driver failed to
693 * queue the page to the device), the page will still be locked. If the
694 * caller is a ->writepage implementation, it will need to unlock the page.
696 * Errors returned by this function are usually "soft", eg out of memory, or
697 * queue full; callers should try a different route to write this page rather
698 * than propagate an error back up the stack.
700 * Return: negative errno if an error occurs, 0 if submission was successful.
702 int bdev_write_page(struct block_device *bdev, sector_t sector,
703 struct page *page, struct writeback_control *wbc)
706 const struct block_device_operations *ops = bdev->bd_disk->fops;
708 if (!ops->rw_page || bdev_get_integrity(bdev))
710 result = blk_queue_enter(bdev->bd_queue, false);
714 set_page_writeback(page);
715 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true);
717 end_page_writeback(page);
720 blk_queue_exit(bdev->bd_queue);
723 EXPORT_SYMBOL_GPL(bdev_write_page);
729 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
730 static struct kmem_cache * bdev_cachep __read_mostly;
732 static struct inode *bdev_alloc_inode(struct super_block *sb)
734 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
737 return &ei->vfs_inode;
740 static void bdev_i_callback(struct rcu_head *head)
742 struct inode *inode = container_of(head, struct inode, i_rcu);
743 struct bdev_inode *bdi = BDEV_I(inode);
745 kmem_cache_free(bdev_cachep, bdi);
748 static void bdev_destroy_inode(struct inode *inode)
750 call_rcu(&inode->i_rcu, bdev_i_callback);
753 static void init_once(void *foo)
755 struct bdev_inode *ei = (struct bdev_inode *) foo;
756 struct block_device *bdev = &ei->bdev;
758 memset(bdev, 0, sizeof(*bdev));
759 mutex_init(&bdev->bd_mutex);
760 INIT_LIST_HEAD(&bdev->bd_list);
762 INIT_LIST_HEAD(&bdev->bd_holder_disks);
764 bdev->bd_bdi = &noop_backing_dev_info;
765 inode_init_once(&ei->vfs_inode);
766 /* Initialize mutex for freeze. */
767 mutex_init(&bdev->bd_fsfreeze_mutex);
770 static void bdev_evict_inode(struct inode *inode)
772 struct block_device *bdev = &BDEV_I(inode)->bdev;
773 truncate_inode_pages_final(&inode->i_data);
774 invalidate_inode_buffers(inode); /* is it needed here? */
776 spin_lock(&bdev_lock);
777 list_del_init(&bdev->bd_list);
778 spin_unlock(&bdev_lock);
779 /* Detach inode from wb early as bdi_put() may free bdi->wb */
780 inode_detach_wb(inode);
781 if (bdev->bd_bdi != &noop_backing_dev_info) {
782 bdi_put(bdev->bd_bdi);
783 bdev->bd_bdi = &noop_backing_dev_info;
787 static const struct super_operations bdev_sops = {
788 .statfs = simple_statfs,
789 .alloc_inode = bdev_alloc_inode,
790 .destroy_inode = bdev_destroy_inode,
791 .drop_inode = generic_delete_inode,
792 .evict_inode = bdev_evict_inode,
795 static struct dentry *bd_mount(struct file_system_type *fs_type,
796 int flags, const char *dev_name, void *data)
799 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
801 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
805 static struct file_system_type bd_type = {
808 .kill_sb = kill_anon_super,
811 struct super_block *blockdev_superblock __read_mostly;
812 EXPORT_SYMBOL_GPL(blockdev_superblock);
814 void __init bdev_cache_init(void)
817 static struct vfsmount *bd_mnt;
819 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
820 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
821 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
823 err = register_filesystem(&bd_type);
825 panic("Cannot register bdev pseudo-fs");
826 bd_mnt = kern_mount(&bd_type);
828 panic("Cannot create bdev pseudo-fs");
829 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
833 * Most likely _very_ bad one - but then it's hardly critical for small
834 * /dev and can be fixed when somebody will need really large one.
835 * Keep in mind that it will be fed through icache hash function too.
837 static inline unsigned long hash(dev_t dev)
839 return MAJOR(dev)+MINOR(dev);
842 static int bdev_test(struct inode *inode, void *data)
844 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
847 static int bdev_set(struct inode *inode, void *data)
849 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
853 static LIST_HEAD(all_bdevs);
856 * If there is a bdev inode for this device, unhash it so that it gets evicted
857 * as soon as last inode reference is dropped.
859 void bdev_unhash_inode(dev_t dev)
863 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
865 remove_inode_hash(inode);
870 struct block_device *bdget(dev_t dev)
872 struct block_device *bdev;
875 inode = iget5_locked(blockdev_superblock, hash(dev),
876 bdev_test, bdev_set, &dev);
881 bdev = &BDEV_I(inode)->bdev;
883 if (inode->i_state & I_NEW) {
884 bdev->bd_contains = NULL;
885 bdev->bd_super = NULL;
886 bdev->bd_inode = inode;
887 bdev->bd_block_size = i_blocksize(inode);
888 bdev->bd_part_count = 0;
889 bdev->bd_invalidated = 0;
890 inode->i_mode = S_IFBLK;
892 inode->i_bdev = bdev;
893 inode->i_data.a_ops = &def_blk_aops;
894 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
895 spin_lock(&bdev_lock);
896 list_add(&bdev->bd_list, &all_bdevs);
897 spin_unlock(&bdev_lock);
898 unlock_new_inode(inode);
903 EXPORT_SYMBOL(bdget);
906 * bdgrab -- Grab a reference to an already referenced block device
907 * @bdev: Block device to grab a reference to.
909 struct block_device *bdgrab(struct block_device *bdev)
911 ihold(bdev->bd_inode);
914 EXPORT_SYMBOL(bdgrab);
916 long nr_blockdev_pages(void)
918 struct block_device *bdev;
920 spin_lock(&bdev_lock);
921 list_for_each_entry(bdev, &all_bdevs, bd_list) {
922 ret += bdev->bd_inode->i_mapping->nrpages;
924 spin_unlock(&bdev_lock);
928 void bdput(struct block_device *bdev)
930 iput(bdev->bd_inode);
933 EXPORT_SYMBOL(bdput);
935 static struct block_device *bd_acquire(struct inode *inode)
937 struct block_device *bdev;
939 spin_lock(&bdev_lock);
940 bdev = inode->i_bdev;
941 if (bdev && !inode_unhashed(bdev->bd_inode)) {
943 spin_unlock(&bdev_lock);
946 spin_unlock(&bdev_lock);
949 * i_bdev references block device inode that was already shut down
950 * (corresponding device got removed). Remove the reference and look
951 * up block device inode again just in case new device got
952 * reestablished under the same device number.
957 bdev = bdget(inode->i_rdev);
959 spin_lock(&bdev_lock);
960 if (!inode->i_bdev) {
962 * We take an additional reference to bd_inode,
963 * and it's released in clear_inode() of inode.
964 * So, we can access it via ->i_mapping always
968 inode->i_bdev = bdev;
969 inode->i_mapping = bdev->bd_inode->i_mapping;
971 spin_unlock(&bdev_lock);
976 /* Call when you free inode */
978 void bd_forget(struct inode *inode)
980 struct block_device *bdev = NULL;
982 spin_lock(&bdev_lock);
983 if (!sb_is_blkdev_sb(inode->i_sb))
984 bdev = inode->i_bdev;
985 inode->i_bdev = NULL;
986 inode->i_mapping = &inode->i_data;
987 spin_unlock(&bdev_lock);
994 * bd_may_claim - test whether a block device can be claimed
995 * @bdev: block device of interest
996 * @whole: whole block device containing @bdev, may equal @bdev
997 * @holder: holder trying to claim @bdev
999 * Test whether @bdev can be claimed by @holder.
1002 * spin_lock(&bdev_lock).
1005 * %true if @bdev can be claimed, %false otherwise.
1007 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1010 if (bdev->bd_holder == holder)
1011 return true; /* already a holder */
1012 else if (bdev->bd_holder != NULL)
1013 return false; /* held by someone else */
1014 else if (whole == bdev)
1015 return true; /* is a whole device which isn't held */
1017 else if (whole->bd_holder == bd_may_claim)
1018 return true; /* is a partition of a device that is being partitioned */
1019 else if (whole->bd_holder != NULL)
1020 return false; /* is a partition of a held device */
1022 return true; /* is a partition of an un-held device */
1026 * bd_prepare_to_claim - prepare to claim a block device
1027 * @bdev: block device of interest
1028 * @whole: the whole device containing @bdev, may equal @bdev
1029 * @holder: holder trying to claim @bdev
1031 * Prepare to claim @bdev. This function fails if @bdev is already
1032 * claimed by another holder and waits if another claiming is in
1033 * progress. This function doesn't actually claim. On successful
1034 * return, the caller has ownership of bd_claiming and bd_holder[s].
1037 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1038 * it multiple times.
1041 * 0 if @bdev can be claimed, -EBUSY otherwise.
1043 static int bd_prepare_to_claim(struct block_device *bdev,
1044 struct block_device *whole, void *holder)
1047 /* if someone else claimed, fail */
1048 if (!bd_may_claim(bdev, whole, holder))
1051 /* if claiming is already in progress, wait for it to finish */
1052 if (whole->bd_claiming) {
1053 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1056 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1057 spin_unlock(&bdev_lock);
1059 finish_wait(wq, &wait);
1060 spin_lock(&bdev_lock);
1069 * bd_start_claiming - start claiming a block device
1070 * @bdev: block device of interest
1071 * @holder: holder trying to claim @bdev
1073 * @bdev is about to be opened exclusively. Check @bdev can be opened
1074 * exclusively and mark that an exclusive open is in progress. Each
1075 * successful call to this function must be matched with a call to
1076 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1079 * This function is used to gain exclusive access to the block device
1080 * without actually causing other exclusive open attempts to fail. It
1081 * should be used when the open sequence itself requires exclusive
1082 * access but may subsequently fail.
1088 * Pointer to the block device containing @bdev on success, ERR_PTR()
1091 static struct block_device *bd_start_claiming(struct block_device *bdev,
1094 struct gendisk *disk;
1095 struct block_device *whole;
1101 * @bdev might not have been initialized properly yet, look up
1102 * and grab the outer block device the hard way.
1104 disk = get_gendisk(bdev->bd_dev, &partno);
1106 return ERR_PTR(-ENXIO);
1109 * Normally, @bdev should equal what's returned from bdget_disk()
1110 * if partno is 0; however, some drivers (floppy) use multiple
1111 * bdev's for the same physical device and @bdev may be one of the
1112 * aliases. Keep @bdev if partno is 0. This means claimer
1113 * tracking is broken for those devices but it has always been that
1117 whole = bdget_disk(disk, 0);
1119 whole = bdgrab(bdev);
1121 module_put(disk->fops->owner);
1124 return ERR_PTR(-ENOMEM);
1126 /* prepare to claim, if successful, mark claiming in progress */
1127 spin_lock(&bdev_lock);
1129 err = bd_prepare_to_claim(bdev, whole, holder);
1131 whole->bd_claiming = holder;
1132 spin_unlock(&bdev_lock);
1135 spin_unlock(&bdev_lock);
1137 return ERR_PTR(err);
1142 struct bd_holder_disk {
1143 struct list_head list;
1144 struct gendisk *disk;
1148 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1149 struct gendisk *disk)
1151 struct bd_holder_disk *holder;
1153 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1154 if (holder->disk == disk)
1159 static int add_symlink(struct kobject *from, struct kobject *to)
1161 return sysfs_create_link(from, to, kobject_name(to));
1164 static void del_symlink(struct kobject *from, struct kobject *to)
1166 sysfs_remove_link(from, kobject_name(to));
1170 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1171 * @bdev: the claimed slave bdev
1172 * @disk: the holding disk
1174 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1176 * This functions creates the following sysfs symlinks.
1178 * - from "slaves" directory of the holder @disk to the claimed @bdev
1179 * - from "holders" directory of the @bdev to the holder @disk
1181 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1182 * passed to bd_link_disk_holder(), then:
1184 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1185 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1187 * The caller must have claimed @bdev before calling this function and
1188 * ensure that both @bdev and @disk are valid during the creation and
1189 * lifetime of these symlinks.
1195 * 0 on success, -errno on failure.
1197 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1199 struct bd_holder_disk *holder;
1202 mutex_lock(&bdev->bd_mutex);
1204 WARN_ON_ONCE(!bdev->bd_holder);
1206 /* FIXME: remove the following once add_disk() handles errors */
1207 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1210 holder = bd_find_holder_disk(bdev, disk);
1216 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1222 INIT_LIST_HEAD(&holder->list);
1223 holder->disk = disk;
1226 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1230 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1234 * bdev could be deleted beneath us which would implicitly destroy
1235 * the holder directory. Hold on to it.
1237 kobject_get(bdev->bd_part->holder_dir);
1239 list_add(&holder->list, &bdev->bd_holder_disks);
1243 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1247 mutex_unlock(&bdev->bd_mutex);
1250 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1253 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1254 * @bdev: the calimed slave bdev
1255 * @disk: the holding disk
1257 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1262 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1264 struct bd_holder_disk *holder;
1266 mutex_lock(&bdev->bd_mutex);
1268 holder = bd_find_holder_disk(bdev, disk);
1270 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1271 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1272 del_symlink(bdev->bd_part->holder_dir,
1273 &disk_to_dev(disk)->kobj);
1274 kobject_put(bdev->bd_part->holder_dir);
1275 list_del_init(&holder->list);
1279 mutex_unlock(&bdev->bd_mutex);
1281 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1285 * flush_disk - invalidates all buffer-cache entries on a disk
1287 * @bdev: struct block device to be flushed
1288 * @kill_dirty: flag to guide handling of dirty inodes
1290 * Invalidates all buffer-cache entries on a disk. It should be called
1291 * when a disk has been changed -- either by a media change or online
1294 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1296 if (__invalidate_device(bdev, kill_dirty)) {
1297 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1298 "resized disk %s\n",
1299 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1304 if (disk_part_scan_enabled(bdev->bd_disk))
1305 bdev->bd_invalidated = 1;
1309 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1310 * @disk: struct gendisk to check
1311 * @bdev: struct bdev to adjust.
1313 * This routine checks to see if the bdev size does not match the disk size
1314 * and adjusts it if it differs.
1316 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1318 loff_t disk_size, bdev_size;
1320 disk_size = (loff_t)get_capacity(disk) << 9;
1321 bdev_size = i_size_read(bdev->bd_inode);
1322 if (disk_size != bdev_size) {
1324 "%s: detected capacity change from %lld to %lld\n",
1325 disk->disk_name, bdev_size, disk_size);
1326 i_size_write(bdev->bd_inode, disk_size);
1327 flush_disk(bdev, false);
1330 EXPORT_SYMBOL(check_disk_size_change);
1333 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1334 * @disk: struct gendisk to be revalidated
1336 * This routine is a wrapper for lower-level driver's revalidate_disk
1337 * call-backs. It is used to do common pre and post operations needed
1338 * for all revalidate_disk operations.
1340 int revalidate_disk(struct gendisk *disk)
1342 struct block_device *bdev;
1345 if (disk->fops->revalidate_disk)
1346 ret = disk->fops->revalidate_disk(disk);
1347 bdev = bdget_disk(disk, 0);
1351 mutex_lock(&bdev->bd_mutex);
1352 check_disk_size_change(disk, bdev);
1353 bdev->bd_invalidated = 0;
1354 mutex_unlock(&bdev->bd_mutex);
1358 EXPORT_SYMBOL(revalidate_disk);
1361 * This routine checks whether a removable media has been changed,
1362 * and invalidates all buffer-cache-entries in that case. This
1363 * is a relatively slow routine, so we have to try to minimize using
1364 * it. Thus it is called only upon a 'mount' or 'open'. This
1365 * is the best way of combining speed and utility, I think.
1366 * People changing diskettes in the middle of an operation deserve
1369 int check_disk_change(struct block_device *bdev)
1371 struct gendisk *disk = bdev->bd_disk;
1372 const struct block_device_operations *bdops = disk->fops;
1373 unsigned int events;
1375 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1376 DISK_EVENT_EJECT_REQUEST);
1377 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1380 flush_disk(bdev, true);
1381 if (bdops->revalidate_disk)
1382 bdops->revalidate_disk(bdev->bd_disk);
1386 EXPORT_SYMBOL(check_disk_change);
1388 void bd_set_size(struct block_device *bdev, loff_t size)
1390 unsigned bsize = bdev_logical_block_size(bdev);
1392 inode_lock(bdev->bd_inode);
1393 i_size_write(bdev->bd_inode, size);
1394 inode_unlock(bdev->bd_inode);
1395 while (bsize < PAGE_SIZE) {
1400 bdev->bd_block_size = bsize;
1401 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1403 EXPORT_SYMBOL(bd_set_size);
1405 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1410 * mutex_lock(part->bd_mutex)
1411 * mutex_lock_nested(whole->bd_mutex, 1)
1414 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1416 struct gendisk *disk;
1417 struct module *owner;
1422 if (mode & FMODE_READ)
1424 if (mode & FMODE_WRITE)
1427 * hooks: /n/, see "layering violations".
1430 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1440 disk = get_gendisk(bdev->bd_dev, &partno);
1443 owner = disk->fops->owner;
1445 disk_block_events(disk);
1446 mutex_lock_nested(&bdev->bd_mutex, for_part);
1447 if (!bdev->bd_openers) {
1448 bdev->bd_disk = disk;
1449 bdev->bd_queue = disk->queue;
1450 bdev->bd_contains = bdev;
1454 bdev->bd_part = disk_get_part(disk, partno);
1459 if (disk->fops->open) {
1460 ret = disk->fops->open(bdev, mode);
1461 if (ret == -ERESTARTSYS) {
1462 /* Lost a race with 'disk' being
1463 * deleted, try again.
1466 disk_put_part(bdev->bd_part);
1467 bdev->bd_part = NULL;
1468 bdev->bd_disk = NULL;
1469 bdev->bd_queue = NULL;
1470 mutex_unlock(&bdev->bd_mutex);
1471 disk_unblock_events(disk);
1479 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1482 * If the device is invalidated, rescan partition
1483 * if open succeeded or failed with -ENOMEDIUM.
1484 * The latter is necessary to prevent ghost
1485 * partitions on a removed medium.
1487 if (bdev->bd_invalidated) {
1489 rescan_partitions(disk, bdev);
1490 else if (ret == -ENOMEDIUM)
1491 invalidate_partitions(disk, bdev);
1497 struct block_device *whole;
1498 whole = bdget_disk(disk, 0);
1503 ret = __blkdev_get(whole, mode, 1);
1506 bdev->bd_contains = whole;
1507 bdev->bd_part = disk_get_part(disk, partno);
1508 if (!(disk->flags & GENHD_FL_UP) ||
1509 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1513 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1516 if (bdev->bd_bdi == &noop_backing_dev_info)
1517 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1519 if (bdev->bd_contains == bdev) {
1521 if (bdev->bd_disk->fops->open)
1522 ret = bdev->bd_disk->fops->open(bdev, mode);
1523 /* the same as first opener case, read comment there */
1524 if (bdev->bd_invalidated) {
1526 rescan_partitions(bdev->bd_disk, bdev);
1527 else if (ret == -ENOMEDIUM)
1528 invalidate_partitions(bdev->bd_disk, bdev);
1531 goto out_unlock_bdev;
1533 /* only one opener holds refs to the module and disk */
1539 bdev->bd_part_count++;
1540 mutex_unlock(&bdev->bd_mutex);
1541 disk_unblock_events(disk);
1545 disk_put_part(bdev->bd_part);
1546 bdev->bd_disk = NULL;
1547 bdev->bd_part = NULL;
1548 bdev->bd_queue = NULL;
1549 if (bdev != bdev->bd_contains)
1550 __blkdev_put(bdev->bd_contains, mode, 1);
1551 bdev->bd_contains = NULL;
1553 mutex_unlock(&bdev->bd_mutex);
1554 disk_unblock_events(disk);
1564 * blkdev_get - open a block device
1565 * @bdev: block_device to open
1566 * @mode: FMODE_* mask
1567 * @holder: exclusive holder identifier
1569 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1570 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1571 * @holder is invalid. Exclusive opens may nest for the same @holder.
1573 * On success, the reference count of @bdev is unchanged. On failure,
1580 * 0 on success, -errno on failure.
1582 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1584 struct block_device *whole = NULL;
1587 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1589 if ((mode & FMODE_EXCL) && holder) {
1590 whole = bd_start_claiming(bdev, holder);
1591 if (IS_ERR(whole)) {
1593 return PTR_ERR(whole);
1597 res = __blkdev_get(bdev, mode, 0);
1600 struct gendisk *disk = whole->bd_disk;
1602 /* finish claiming */
1603 mutex_lock(&bdev->bd_mutex);
1604 spin_lock(&bdev_lock);
1607 BUG_ON(!bd_may_claim(bdev, whole, holder));
1609 * Note that for a whole device bd_holders
1610 * will be incremented twice, and bd_holder
1611 * will be set to bd_may_claim before being
1614 whole->bd_holders++;
1615 whole->bd_holder = bd_may_claim;
1617 bdev->bd_holder = holder;
1620 /* tell others that we're done */
1621 BUG_ON(whole->bd_claiming != holder);
1622 whole->bd_claiming = NULL;
1623 wake_up_bit(&whole->bd_claiming, 0);
1625 spin_unlock(&bdev_lock);
1628 * Block event polling for write claims if requested. Any
1629 * write holder makes the write_holder state stick until
1630 * all are released. This is good enough and tracking
1631 * individual writeable reference is too fragile given the
1632 * way @mode is used in blkdev_get/put().
1634 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1635 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1636 bdev->bd_write_holder = true;
1637 disk_block_events(disk);
1640 mutex_unlock(&bdev->bd_mutex);
1646 EXPORT_SYMBOL(blkdev_get);
1649 * blkdev_get_by_path - open a block device by name
1650 * @path: path to the block device to open
1651 * @mode: FMODE_* mask
1652 * @holder: exclusive holder identifier
1654 * Open the blockdevice described by the device file at @path. @mode
1655 * and @holder are identical to blkdev_get().
1657 * On success, the returned block_device has reference count of one.
1663 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1665 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1668 struct block_device *bdev;
1671 bdev = lookup_bdev(path);
1675 err = blkdev_get(bdev, mode, holder);
1677 return ERR_PTR(err);
1679 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1680 blkdev_put(bdev, mode);
1681 return ERR_PTR(-EACCES);
1686 EXPORT_SYMBOL(blkdev_get_by_path);
1689 * blkdev_get_by_dev - open a block device by device number
1690 * @dev: device number of block device to open
1691 * @mode: FMODE_* mask
1692 * @holder: exclusive holder identifier
1694 * Open the blockdevice described by device number @dev. @mode and
1695 * @holder are identical to blkdev_get().
1697 * Use it ONLY if you really do not have anything better - i.e. when
1698 * you are behind a truly sucky interface and all you are given is a
1699 * device number. _Never_ to be used for internal purposes. If you
1700 * ever need it - reconsider your API.
1702 * On success, the returned block_device has reference count of one.
1708 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1710 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1712 struct block_device *bdev;
1717 return ERR_PTR(-ENOMEM);
1719 err = blkdev_get(bdev, mode, holder);
1721 return ERR_PTR(err);
1725 EXPORT_SYMBOL(blkdev_get_by_dev);
1727 static int blkdev_open(struct inode * inode, struct file * filp)
1729 struct block_device *bdev;
1732 * Preserve backwards compatibility and allow large file access
1733 * even if userspace doesn't ask for it explicitly. Some mkfs
1734 * binary needs it. We might want to drop this workaround
1735 * during an unstable branch.
1737 filp->f_flags |= O_LARGEFILE;
1739 if (filp->f_flags & O_NDELAY)
1740 filp->f_mode |= FMODE_NDELAY;
1741 if (filp->f_flags & O_EXCL)
1742 filp->f_mode |= FMODE_EXCL;
1743 if ((filp->f_flags & O_ACCMODE) == 3)
1744 filp->f_mode |= FMODE_WRITE_IOCTL;
1746 bdev = bd_acquire(inode);
1750 filp->f_mapping = bdev->bd_inode->i_mapping;
1752 return blkdev_get(bdev, filp->f_mode, filp);
1755 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1757 struct gendisk *disk = bdev->bd_disk;
1758 struct block_device *victim = NULL;
1760 mutex_lock_nested(&bdev->bd_mutex, for_part);
1762 bdev->bd_part_count--;
1764 if (!--bdev->bd_openers) {
1765 WARN_ON_ONCE(bdev->bd_holders);
1766 sync_blockdev(bdev);
1769 bdev_write_inode(bdev);
1771 if (bdev->bd_contains == bdev) {
1772 if (disk->fops->release)
1773 disk->fops->release(disk, mode);
1775 if (!bdev->bd_openers) {
1776 struct module *owner = disk->fops->owner;
1778 disk_put_part(bdev->bd_part);
1779 bdev->bd_part = NULL;
1780 bdev->bd_disk = NULL;
1781 if (bdev != bdev->bd_contains)
1782 victim = bdev->bd_contains;
1783 bdev->bd_contains = NULL;
1788 mutex_unlock(&bdev->bd_mutex);
1791 __blkdev_put(victim, mode, 1);
1794 void blkdev_put(struct block_device *bdev, fmode_t mode)
1796 mutex_lock(&bdev->bd_mutex);
1798 if (mode & FMODE_EXCL) {
1802 * Release a claim on the device. The holder fields
1803 * are protected with bdev_lock. bd_mutex is to
1804 * synchronize disk_holder unlinking.
1806 spin_lock(&bdev_lock);
1808 WARN_ON_ONCE(--bdev->bd_holders < 0);
1809 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1811 /* bd_contains might point to self, check in a separate step */
1812 if ((bdev_free = !bdev->bd_holders))
1813 bdev->bd_holder = NULL;
1814 if (!bdev->bd_contains->bd_holders)
1815 bdev->bd_contains->bd_holder = NULL;
1817 spin_unlock(&bdev_lock);
1820 * If this was the last claim, remove holder link and
1821 * unblock evpoll if it was a write holder.
1823 if (bdev_free && bdev->bd_write_holder) {
1824 disk_unblock_events(bdev->bd_disk);
1825 bdev->bd_write_holder = false;
1830 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1831 * event. This is to ensure detection of media removal commanded
1832 * from userland - e.g. eject(1).
1834 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1836 mutex_unlock(&bdev->bd_mutex);
1838 __blkdev_put(bdev, mode, 0);
1840 EXPORT_SYMBOL(blkdev_put);
1842 static int blkdev_close(struct inode * inode, struct file * filp)
1844 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1845 blkdev_put(bdev, filp->f_mode);
1849 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1851 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1852 fmode_t mode = file->f_mode;
1855 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1856 * to updated it before every ioctl.
1858 if (file->f_flags & O_NDELAY)
1859 mode |= FMODE_NDELAY;
1861 mode &= ~FMODE_NDELAY;
1863 return blkdev_ioctl(bdev, mode, cmd, arg);
1867 * Write data to the block device. Only intended for the block device itself
1868 * and the raw driver which basically is a fake block device.
1870 * Does not take i_mutex for the write and thus is not for general purpose
1873 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1875 struct file *file = iocb->ki_filp;
1876 struct inode *bd_inode = bdev_file_inode(file);
1877 loff_t size = i_size_read(bd_inode);
1878 struct blk_plug plug;
1881 if (bdev_read_only(I_BDEV(bd_inode)))
1884 if (!iov_iter_count(from))
1887 if (iocb->ki_pos >= size)
1890 iov_iter_truncate(from, size - iocb->ki_pos);
1892 blk_start_plug(&plug);
1893 ret = __generic_file_write_iter(iocb, from);
1895 ret = generic_write_sync(iocb, ret);
1896 blk_finish_plug(&plug);
1899 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1901 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1903 struct file *file = iocb->ki_filp;
1904 struct inode *bd_inode = bdev_file_inode(file);
1905 loff_t size = i_size_read(bd_inode);
1906 loff_t pos = iocb->ki_pos;
1912 iov_iter_truncate(to, size);
1913 return generic_file_read_iter(iocb, to);
1915 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1918 * Try to release a page associated with block device when the system
1919 * is under memory pressure.
1921 static int blkdev_releasepage(struct page *page, gfp_t wait)
1923 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1925 if (super && super->s_op->bdev_try_to_free_page)
1926 return super->s_op->bdev_try_to_free_page(super, page, wait);
1928 return try_to_free_buffers(page);
1931 static int blkdev_writepages(struct address_space *mapping,
1932 struct writeback_control *wbc)
1934 if (dax_mapping(mapping)) {
1935 struct block_device *bdev = I_BDEV(mapping->host);
1937 return dax_writeback_mapping_range(mapping, bdev, wbc);
1939 return generic_writepages(mapping, wbc);
1942 static const struct address_space_operations def_blk_aops = {
1943 .readpage = blkdev_readpage,
1944 .readpages = blkdev_readpages,
1945 .writepage = blkdev_writepage,
1946 .write_begin = blkdev_write_begin,
1947 .write_end = blkdev_write_end,
1948 .writepages = blkdev_writepages,
1949 .releasepage = blkdev_releasepage,
1950 .direct_IO = blkdev_direct_IO,
1951 .is_dirty_writeback = buffer_check_dirty_writeback,
1954 #define BLKDEV_FALLOC_FL_SUPPORTED \
1955 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1956 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1958 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1961 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1962 struct address_space *mapping;
1963 loff_t end = start + len - 1;
1967 /* Fail if we don't recognize the flags. */
1968 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1971 /* Don't go off the end of the device. */
1972 isize = i_size_read(bdev->bd_inode);
1976 if (mode & FALLOC_FL_KEEP_SIZE) {
1977 len = isize - start;
1978 end = start + len - 1;
1984 * Don't allow IO that isn't aligned to logical block size.
1986 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1989 /* Invalidate the page cache, including dirty pages. */
1990 mapping = bdev->bd_inode->i_mapping;
1991 truncate_inode_pages_range(mapping, start, end);
1994 case FALLOC_FL_ZERO_RANGE:
1995 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
1996 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1997 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
1999 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2000 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2001 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2003 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2004 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2014 * Invalidate again; if someone wandered in and dirtied a page,
2015 * the caller will be given -EBUSY. The third argument is
2016 * inclusive, so the rounding here is safe.
2018 return invalidate_inode_pages2_range(mapping,
2019 start >> PAGE_SHIFT,
2023 const struct file_operations def_blk_fops = {
2024 .open = blkdev_open,
2025 .release = blkdev_close,
2026 .llseek = block_llseek,
2027 .read_iter = blkdev_read_iter,
2028 .write_iter = blkdev_write_iter,
2029 .mmap = generic_file_mmap,
2030 .fsync = blkdev_fsync,
2031 .unlocked_ioctl = block_ioctl,
2032 #ifdef CONFIG_COMPAT
2033 .compat_ioctl = compat_blkdev_ioctl,
2035 .splice_read = generic_file_splice_read,
2036 .splice_write = iter_file_splice_write,
2037 .fallocate = blkdev_fallocate,
2040 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2043 mm_segment_t old_fs = get_fs();
2045 res = blkdev_ioctl(bdev, 0, cmd, arg);
2050 EXPORT_SYMBOL(ioctl_by_bdev);
2053 * lookup_bdev - lookup a struct block_device by name
2054 * @pathname: special file representing the block device
2056 * Get a reference to the blockdevice at @pathname in the current
2057 * namespace if possible and return it. Return ERR_PTR(error)
2060 struct block_device *lookup_bdev(const char *pathname)
2062 struct block_device *bdev;
2063 struct inode *inode;
2067 if (!pathname || !*pathname)
2068 return ERR_PTR(-EINVAL);
2070 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2072 return ERR_PTR(error);
2074 inode = d_backing_inode(path.dentry);
2076 if (!S_ISBLK(inode->i_mode))
2079 if (!may_open_dev(&path))
2082 bdev = bd_acquire(inode);
2089 bdev = ERR_PTR(error);
2092 EXPORT_SYMBOL(lookup_bdev);
2094 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2096 struct super_block *sb = get_super(bdev);
2101 * no need to lock the super, get_super holds the
2102 * read mutex so the filesystem cannot go away
2103 * under us (->put_super runs with the write lock
2106 shrink_dcache_sb(sb);
2107 res = invalidate_inodes(sb, kill_dirty);
2110 invalidate_bdev(bdev);
2113 EXPORT_SYMBOL(__invalidate_device);
2115 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2117 struct inode *inode, *old_inode = NULL;
2119 spin_lock(&blockdev_superblock->s_inode_list_lock);
2120 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2121 struct address_space *mapping = inode->i_mapping;
2122 struct block_device *bdev;
2124 spin_lock(&inode->i_lock);
2125 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2126 mapping->nrpages == 0) {
2127 spin_unlock(&inode->i_lock);
2131 spin_unlock(&inode->i_lock);
2132 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2134 * We hold a reference to 'inode' so it couldn't have been
2135 * removed from s_inodes list while we dropped the
2136 * s_inode_list_lock We cannot iput the inode now as we can
2137 * be holding the last reference and we cannot iput it under
2138 * s_inode_list_lock. So we keep the reference and iput it
2143 bdev = I_BDEV(inode);
2145 mutex_lock(&bdev->bd_mutex);
2146 if (bdev->bd_openers)
2148 mutex_unlock(&bdev->bd_mutex);
2150 spin_lock(&blockdev_superblock->s_inode_list_lock);
2152 spin_unlock(&blockdev_superblock->s_inode_list_lock);