4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/device_cgroup.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/backing-dev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/magic.h>
21 #include <linux/buffer_head.h>
22 #include <linux/swap.h>
23 #include <linux/pagevec.h>
24 #include <linux/writeback.h>
25 #include <linux/mpage.h>
26 #include <linux/mount.h>
27 #include <linux/uio.h>
28 #include <linux/namei.h>
29 #include <linux/log2.h>
30 #include <linux/cleancache.h>
31 #include <linux/dax.h>
32 #include <linux/badblocks.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/falloc.h>
35 #include <linux/uaccess.h>
39 struct block_device bdev;
40 struct inode vfs_inode;
43 static const struct address_space_operations def_blk_aops;
45 static inline struct bdev_inode *BDEV_I(struct inode *inode)
47 return container_of(inode, struct bdev_inode, vfs_inode);
50 struct block_device *I_BDEV(struct inode *inode)
52 return &BDEV_I(inode)->bdev;
54 EXPORT_SYMBOL(I_BDEV);
56 void __vfs_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
64 printk_ratelimited("%sVFS (%s): %pV\n", prefix, sb->s_id, &vaf);
68 static void bdev_write_inode(struct block_device *bdev)
70 struct inode *inode = bdev->bd_inode;
73 spin_lock(&inode->i_lock);
74 while (inode->i_state & I_DIRTY) {
75 spin_unlock(&inode->i_lock);
76 ret = write_inode_now(inode, true);
78 char name[BDEVNAME_SIZE];
79 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
80 "for block device %s (err=%d).\n",
81 bdevname(bdev, name), ret);
83 spin_lock(&inode->i_lock);
85 spin_unlock(&inode->i_lock);
88 /* Kill _all_ buffers and pagecache , dirty or not.. */
89 void kill_bdev(struct block_device *bdev)
91 struct address_space *mapping = bdev->bd_inode->i_mapping;
93 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
97 truncate_inode_pages(mapping, 0);
99 EXPORT_SYMBOL(kill_bdev);
101 /* Invalidate clean unused buffers and pagecache. */
102 void invalidate_bdev(struct block_device *bdev)
104 struct address_space *mapping = bdev->bd_inode->i_mapping;
106 if (mapping->nrpages == 0)
109 invalidate_bh_lrus();
110 lru_add_drain_all(); /* make sure all lru add caches are flushed */
111 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_private = current;
229 bio.bi_end_io = blkdev_bio_end_io_simple;
231 ret = bio_iov_iter_get_pages(&bio, iter);
234 ret = bio.bi_iter.bi_size;
236 if (iov_iter_rw(iter) == READ) {
237 bio.bi_opf = REQ_OP_READ;
238 if (iter_is_iovec(iter))
241 bio.bi_opf = dio_bio_write_op(iocb);
242 task_io_account_write(ret);
245 qc = submit_bio(&bio);
247 set_current_state(TASK_UNINTERRUPTIBLE);
248 if (!READ_ONCE(bio.bi_private))
250 if (!(iocb->ki_flags & IOCB_HIPRI) ||
251 !blk_mq_poll(bdev_get_queue(bdev), qc))
254 __set_current_state(TASK_RUNNING);
256 bio_for_each_segment_all(bvec, &bio, i) {
257 if (should_dirty && !PageCompound(bvec->bv_page))
258 set_page_dirty_lock(bvec->bv_page);
259 put_page(bvec->bv_page);
262 if (vecs != inline_vecs)
265 if (unlikely(bio.bi_error))
273 struct task_struct *waiter;
278 bool should_dirty : 1;
283 static struct bio_set *blkdev_dio_pool __read_mostly;
285 static void blkdev_bio_end_io(struct bio *bio)
287 struct blkdev_dio *dio = bio->bi_private;
288 bool should_dirty = dio->should_dirty;
290 if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) {
291 if (bio->bi_error && !dio->bio.bi_error)
292 dio->bio.bi_error = bio->bi_error;
295 struct kiocb *iocb = dio->iocb;
296 ssize_t ret = dio->bio.bi_error;
303 dio->iocb->ki_complete(iocb, ret, 0);
306 struct task_struct *waiter = dio->waiter;
308 WRITE_ONCE(dio->waiter, NULL);
309 wake_up_process(waiter);
314 bio_check_pages_dirty(bio);
316 struct bio_vec *bvec;
319 bio_for_each_segment_all(bvec, bio, i)
320 put_page(bvec->bv_page);
326 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
328 struct file *file = iocb->ki_filp;
329 struct inode *inode = bdev_file_inode(file);
330 struct block_device *bdev = I_BDEV(inode);
331 struct blk_plug plug;
332 struct blkdev_dio *dio;
334 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
335 loff_t pos = iocb->ki_pos;
336 blk_qc_t qc = BLK_QC_T_NONE;
339 if ((pos | iov_iter_alignment(iter)) &
340 (bdev_logical_block_size(bdev) - 1))
343 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, blkdev_dio_pool);
344 bio_get(bio); /* extra ref for the completion handler */
346 dio = container_of(bio, struct blkdev_dio, bio);
347 dio->is_sync = is_sync = is_sync_kiocb(iocb);
349 dio->waiter = current;
354 dio->multi_bio = false;
355 dio->should_dirty = is_read && (iter->type == ITER_IOVEC);
357 blk_start_plug(&plug);
360 bio->bi_iter.bi_sector = pos >> 9;
361 bio->bi_private = dio;
362 bio->bi_end_io = blkdev_bio_end_io;
364 ret = bio_iov_iter_get_pages(bio, iter);
372 bio->bi_opf = REQ_OP_READ;
373 if (dio->should_dirty)
374 bio_set_pages_dirty(bio);
376 bio->bi_opf = dio_bio_write_op(iocb);
377 task_io_account_write(bio->bi_iter.bi_size);
380 dio->size += bio->bi_iter.bi_size;
381 pos += bio->bi_iter.bi_size;
383 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
385 qc = submit_bio(bio);
389 if (!dio->multi_bio) {
390 dio->multi_bio = true;
391 atomic_set(&dio->ref, 2);
393 atomic_inc(&dio->ref);
397 bio = bio_alloc(GFP_KERNEL, nr_pages);
399 blk_finish_plug(&plug);
405 set_current_state(TASK_UNINTERRUPTIBLE);
406 if (!READ_ONCE(dio->waiter))
409 if (!(iocb->ki_flags & IOCB_HIPRI) ||
410 !blk_mq_poll(bdev_get_queue(bdev), qc))
413 __set_current_state(TASK_RUNNING);
415 ret = dio->bio.bi_error;
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 blkdev_dio_pool = bioset_create(4, offsetof(struct blkdev_dio, bio));
440 if (!blkdev_dio_pool)
444 module_init(blkdev_init);
446 int __sync_blockdev(struct block_device *bdev, int wait)
451 return filemap_flush(bdev->bd_inode->i_mapping);
452 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
456 * Write out and wait upon all the dirty data associated with a block
457 * device via its mapping. Does not take the superblock lock.
459 int sync_blockdev(struct block_device *bdev)
461 return __sync_blockdev(bdev, 1);
463 EXPORT_SYMBOL(sync_blockdev);
466 * Write out and wait upon all dirty data associated with this
467 * device. Filesystem data as well as the underlying block
468 * device. Takes the superblock lock.
470 int fsync_bdev(struct block_device *bdev)
472 struct super_block *sb = get_super(bdev);
474 int res = sync_filesystem(sb);
478 return sync_blockdev(bdev);
480 EXPORT_SYMBOL(fsync_bdev);
483 * freeze_bdev -- lock a filesystem and force it into a consistent state
484 * @bdev: blockdevice to lock
486 * If a superblock is found on this device, we take the s_umount semaphore
487 * on it to make sure nobody unmounts until the snapshot creation is done.
488 * The reference counter (bd_fsfreeze_count) guarantees that only the last
489 * unfreeze process can unfreeze the frozen filesystem actually when multiple
490 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
491 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
494 struct super_block *freeze_bdev(struct block_device *bdev)
496 struct super_block *sb;
499 mutex_lock(&bdev->bd_fsfreeze_mutex);
500 if (++bdev->bd_fsfreeze_count > 1) {
502 * We don't even need to grab a reference - the first call
503 * to freeze_bdev grab an active reference and only the last
504 * thaw_bdev drops it.
506 sb = get_super(bdev);
509 mutex_unlock(&bdev->bd_fsfreeze_mutex);
513 sb = get_active_super(bdev);
516 if (sb->s_op->freeze_super)
517 error = sb->s_op->freeze_super(sb);
519 error = freeze_super(sb);
521 deactivate_super(sb);
522 bdev->bd_fsfreeze_count--;
523 mutex_unlock(&bdev->bd_fsfreeze_mutex);
524 return ERR_PTR(error);
526 deactivate_super(sb);
529 mutex_unlock(&bdev->bd_fsfreeze_mutex);
530 return sb; /* thaw_bdev releases s->s_umount */
532 EXPORT_SYMBOL(freeze_bdev);
535 * thaw_bdev -- unlock filesystem
536 * @bdev: blockdevice to unlock
537 * @sb: associated superblock
539 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
541 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
545 mutex_lock(&bdev->bd_fsfreeze_mutex);
546 if (!bdev->bd_fsfreeze_count)
550 if (--bdev->bd_fsfreeze_count > 0)
556 if (sb->s_op->thaw_super)
557 error = sb->s_op->thaw_super(sb);
559 error = thaw_super(sb);
561 bdev->bd_fsfreeze_count++;
563 mutex_unlock(&bdev->bd_fsfreeze_mutex);
566 EXPORT_SYMBOL(thaw_bdev);
568 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
570 return block_write_full_page(page, blkdev_get_block, wbc);
573 static int blkdev_readpage(struct file * file, struct page * page)
575 return block_read_full_page(page, blkdev_get_block);
578 static int blkdev_readpages(struct file *file, struct address_space *mapping,
579 struct list_head *pages, unsigned nr_pages)
581 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
584 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
585 loff_t pos, unsigned len, unsigned flags,
586 struct page **pagep, void **fsdata)
588 return block_write_begin(mapping, pos, len, flags, pagep,
592 static int blkdev_write_end(struct file *file, struct address_space *mapping,
593 loff_t pos, unsigned len, unsigned copied,
594 struct page *page, void *fsdata)
597 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
607 * for a block special file file_inode(file)->i_size is zero
608 * so we compute the size by hand (just as in block_read/write above)
610 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
612 struct inode *bd_inode = bdev_file_inode(file);
615 inode_lock(bd_inode);
616 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
617 inode_unlock(bd_inode);
621 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
623 struct inode *bd_inode = bdev_file_inode(filp);
624 struct block_device *bdev = I_BDEV(bd_inode);
627 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
632 * There is no need to serialise calls to blkdev_issue_flush with
633 * i_mutex and doing so causes performance issues with concurrent
634 * O_SYNC writers to a block device.
636 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
637 if (error == -EOPNOTSUPP)
642 EXPORT_SYMBOL(blkdev_fsync);
645 * bdev_read_page() - Start reading a page from a block device
646 * @bdev: The device to read the page from
647 * @sector: The offset on the device to read the page to (need not be aligned)
648 * @page: The page to read
650 * On entry, the page should be locked. It will be unlocked when the page
651 * has been read. If the block driver implements rw_page synchronously,
652 * that will be true on exit from this function, but it need not be.
654 * Errors returned by this function are usually "soft", eg out of memory, or
655 * queue full; callers should try a different route to read this page rather
656 * than propagate an error back up the stack.
658 * Return: negative errno if an error occurs, 0 if submission was successful.
660 int bdev_read_page(struct block_device *bdev, sector_t sector,
663 const struct block_device_operations *ops = bdev->bd_disk->fops;
664 int result = -EOPNOTSUPP;
666 if (!ops->rw_page || bdev_get_integrity(bdev))
669 result = blk_queue_enter(bdev->bd_queue, false);
672 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false);
673 blk_queue_exit(bdev->bd_queue);
676 EXPORT_SYMBOL_GPL(bdev_read_page);
679 * bdev_write_page() - Start writing a page to a block device
680 * @bdev: The device to write the page to
681 * @sector: The offset on the device to write the page to (need not be aligned)
682 * @page: The page to write
683 * @wbc: The writeback_control for the write
685 * On entry, the page should be locked and not currently under writeback.
686 * On exit, if the write started successfully, the page will be unlocked and
687 * under writeback. If the write failed already (eg the driver failed to
688 * queue the page to the device), the page will still be locked. If the
689 * caller is a ->writepage implementation, it will need to unlock the page.
691 * Errors returned by this function are usually "soft", eg out of memory, or
692 * queue full; callers should try a different route to write this page rather
693 * than propagate an error back up the stack.
695 * Return: negative errno if an error occurs, 0 if submission was successful.
697 int bdev_write_page(struct block_device *bdev, sector_t sector,
698 struct page *page, struct writeback_control *wbc)
701 const struct block_device_operations *ops = bdev->bd_disk->fops;
703 if (!ops->rw_page || bdev_get_integrity(bdev))
705 result = blk_queue_enter(bdev->bd_queue, false);
709 set_page_writeback(page);
710 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true);
712 end_page_writeback(page);
715 blk_queue_exit(bdev->bd_queue);
718 EXPORT_SYMBOL_GPL(bdev_write_page);
721 * bdev_direct_access() - Get the address for directly-accessibly memory
722 * @bdev: The device containing the memory
723 * @dax: control and output parameters for ->direct_access
725 * If a block device is made up of directly addressable memory, this function
726 * will tell the caller the PFN and the address of the memory. The address
727 * may be directly dereferenced within the kernel without the need to call
728 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
731 * Return: negative errno if an error occurs, otherwise the number of bytes
732 * accessible at this address.
734 long bdev_direct_access(struct block_device *bdev, struct blk_dax_ctl *dax)
736 sector_t sector = dax->sector;
737 long avail, size = dax->size;
738 const struct block_device_operations *ops = bdev->bd_disk->fops;
741 * The device driver is allowed to sleep, in order to make the
742 * memory directly accessible.
748 if (!blk_queue_dax(bdev_get_queue(bdev)) || !ops->direct_access)
750 if ((sector + DIV_ROUND_UP(size, 512)) >
751 part_nr_sects_read(bdev->bd_part))
753 sector += get_start_sect(bdev);
754 if (sector % (PAGE_SIZE / 512))
756 avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn, size);
759 if (avail > 0 && avail & ~PAGE_MASK)
761 return min(avail, size);
763 EXPORT_SYMBOL_GPL(bdev_direct_access);
766 * bdev_dax_supported() - Check if the device supports dax for filesystem
767 * @sb: The superblock of the device
768 * @blocksize: The block size of the device
770 * This is a library function for filesystems to check if the block device
771 * can be mounted with dax option.
773 * Return: negative errno if unsupported, 0 if supported.
775 int bdev_dax_supported(struct super_block *sb, int blocksize)
777 struct blk_dax_ctl dax = {
783 if (blocksize != PAGE_SIZE) {
784 vfs_msg(sb, KERN_ERR, "error: unsupported blocksize for dax");
788 err = bdev_direct_access(sb->s_bdev, &dax);
792 vfs_msg(sb, KERN_ERR,
793 "error: device does not support dax");
796 vfs_msg(sb, KERN_ERR,
797 "error: unaligned partition for dax");
800 vfs_msg(sb, KERN_ERR,
801 "error: dax access failed (%d)", err);
808 EXPORT_SYMBOL_GPL(bdev_dax_supported);
811 * bdev_dax_capable() - Return if the raw device is capable for dax
812 * @bdev: The device for raw block device access
814 bool bdev_dax_capable(struct block_device *bdev)
816 struct blk_dax_ctl dax = {
820 if (!IS_ENABLED(CONFIG_FS_DAX))
824 if (bdev_direct_access(bdev, &dax) < 0)
827 dax.sector = bdev->bd_part->nr_sects - (PAGE_SIZE / 512);
828 if (bdev_direct_access(bdev, &dax) < 0)
838 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
839 static struct kmem_cache * bdev_cachep __read_mostly;
841 static struct inode *bdev_alloc_inode(struct super_block *sb)
843 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
846 return &ei->vfs_inode;
849 static void bdev_i_callback(struct rcu_head *head)
851 struct inode *inode = container_of(head, struct inode, i_rcu);
852 struct bdev_inode *bdi = BDEV_I(inode);
854 kmem_cache_free(bdev_cachep, bdi);
857 static void bdev_destroy_inode(struct inode *inode)
859 call_rcu(&inode->i_rcu, bdev_i_callback);
862 static void init_once(void *foo)
864 struct bdev_inode *ei = (struct bdev_inode *) foo;
865 struct block_device *bdev = &ei->bdev;
867 memset(bdev, 0, sizeof(*bdev));
868 mutex_init(&bdev->bd_mutex);
869 INIT_LIST_HEAD(&bdev->bd_list);
871 INIT_LIST_HEAD(&bdev->bd_holder_disks);
873 inode_init_once(&ei->vfs_inode);
874 /* Initialize mutex for freeze. */
875 mutex_init(&bdev->bd_fsfreeze_mutex);
878 static void bdev_evict_inode(struct inode *inode)
880 struct block_device *bdev = &BDEV_I(inode)->bdev;
881 truncate_inode_pages_final(&inode->i_data);
882 invalidate_inode_buffers(inode); /* is it needed here? */
884 spin_lock(&bdev_lock);
885 list_del_init(&bdev->bd_list);
886 spin_unlock(&bdev_lock);
887 if (bdev->bd_bdi != &noop_backing_dev_info)
888 bdi_put(bdev->bd_bdi);
891 static const struct super_operations bdev_sops = {
892 .statfs = simple_statfs,
893 .alloc_inode = bdev_alloc_inode,
894 .destroy_inode = bdev_destroy_inode,
895 .drop_inode = generic_delete_inode,
896 .evict_inode = bdev_evict_inode,
899 static struct dentry *bd_mount(struct file_system_type *fs_type,
900 int flags, const char *dev_name, void *data)
903 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
905 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
909 static struct file_system_type bd_type = {
912 .kill_sb = kill_anon_super,
915 struct super_block *blockdev_superblock __read_mostly;
916 EXPORT_SYMBOL_GPL(blockdev_superblock);
918 void __init bdev_cache_init(void)
921 static struct vfsmount *bd_mnt;
923 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
924 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
925 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
927 err = register_filesystem(&bd_type);
929 panic("Cannot register bdev pseudo-fs");
930 bd_mnt = kern_mount(&bd_type);
932 panic("Cannot create bdev pseudo-fs");
933 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
937 * Most likely _very_ bad one - but then it's hardly critical for small
938 * /dev and can be fixed when somebody will need really large one.
939 * Keep in mind that it will be fed through icache hash function too.
941 static inline unsigned long hash(dev_t dev)
943 return MAJOR(dev)+MINOR(dev);
946 static int bdev_test(struct inode *inode, void *data)
948 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
951 static int bdev_set(struct inode *inode, void *data)
953 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
957 static LIST_HEAD(all_bdevs);
960 * If there is a bdev inode for this device, unhash it so that it gets evicted
961 * as soon as last inode reference is dropped.
963 void bdev_unhash_inode(dev_t dev)
967 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
969 remove_inode_hash(inode);
974 struct block_device *bdget(dev_t dev)
976 struct block_device *bdev;
979 inode = iget5_locked(blockdev_superblock, hash(dev),
980 bdev_test, bdev_set, &dev);
985 bdev = &BDEV_I(inode)->bdev;
987 if (inode->i_state & I_NEW) {
988 bdev->bd_contains = NULL;
989 bdev->bd_super = NULL;
990 bdev->bd_inode = inode;
991 bdev->bd_bdi = &noop_backing_dev_info;
992 bdev->bd_block_size = i_blocksize(inode);
993 bdev->bd_part_count = 0;
994 bdev->bd_invalidated = 0;
995 inode->i_mode = S_IFBLK;
997 inode->i_bdev = bdev;
998 inode->i_data.a_ops = &def_blk_aops;
999 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
1000 spin_lock(&bdev_lock);
1001 list_add(&bdev->bd_list, &all_bdevs);
1002 spin_unlock(&bdev_lock);
1003 unlock_new_inode(inode);
1008 EXPORT_SYMBOL(bdget);
1011 * bdgrab -- Grab a reference to an already referenced block device
1012 * @bdev: Block device to grab a reference to.
1014 struct block_device *bdgrab(struct block_device *bdev)
1016 ihold(bdev->bd_inode);
1019 EXPORT_SYMBOL(bdgrab);
1021 long nr_blockdev_pages(void)
1023 struct block_device *bdev;
1025 spin_lock(&bdev_lock);
1026 list_for_each_entry(bdev, &all_bdevs, bd_list) {
1027 ret += bdev->bd_inode->i_mapping->nrpages;
1029 spin_unlock(&bdev_lock);
1033 void bdput(struct block_device *bdev)
1035 iput(bdev->bd_inode);
1038 EXPORT_SYMBOL(bdput);
1040 static struct block_device *bd_acquire(struct inode *inode)
1042 struct block_device *bdev;
1044 spin_lock(&bdev_lock);
1045 bdev = inode->i_bdev;
1046 if (bdev && !inode_unhashed(bdev->bd_inode)) {
1048 spin_unlock(&bdev_lock);
1051 spin_unlock(&bdev_lock);
1054 * i_bdev references block device inode that was already shut down
1055 * (corresponding device got removed). Remove the reference and look
1056 * up block device inode again just in case new device got
1057 * reestablished under the same device number.
1062 bdev = bdget(inode->i_rdev);
1064 spin_lock(&bdev_lock);
1065 if (!inode->i_bdev) {
1067 * We take an additional reference to bd_inode,
1068 * and it's released in clear_inode() of inode.
1069 * So, we can access it via ->i_mapping always
1073 inode->i_bdev = bdev;
1074 inode->i_mapping = bdev->bd_inode->i_mapping;
1076 spin_unlock(&bdev_lock);
1081 /* Call when you free inode */
1083 void bd_forget(struct inode *inode)
1085 struct block_device *bdev = NULL;
1087 spin_lock(&bdev_lock);
1088 if (!sb_is_blkdev_sb(inode->i_sb))
1089 bdev = inode->i_bdev;
1090 inode->i_bdev = NULL;
1091 inode->i_mapping = &inode->i_data;
1092 spin_unlock(&bdev_lock);
1099 * bd_may_claim - test whether a block device can be claimed
1100 * @bdev: block device of interest
1101 * @whole: whole block device containing @bdev, may equal @bdev
1102 * @holder: holder trying to claim @bdev
1104 * Test whether @bdev can be claimed by @holder.
1107 * spin_lock(&bdev_lock).
1110 * %true if @bdev can be claimed, %false otherwise.
1112 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1115 if (bdev->bd_holder == holder)
1116 return true; /* already a holder */
1117 else if (bdev->bd_holder != NULL)
1118 return false; /* held by someone else */
1119 else if (whole == bdev)
1120 return true; /* is a whole device which isn't held */
1122 else if (whole->bd_holder == bd_may_claim)
1123 return true; /* is a partition of a device that is being partitioned */
1124 else if (whole->bd_holder != NULL)
1125 return false; /* is a partition of a held device */
1127 return true; /* is a partition of an un-held device */
1131 * bd_prepare_to_claim - prepare to claim a block device
1132 * @bdev: block device of interest
1133 * @whole: the whole device containing @bdev, may equal @bdev
1134 * @holder: holder trying to claim @bdev
1136 * Prepare to claim @bdev. This function fails if @bdev is already
1137 * claimed by another holder and waits if another claiming is in
1138 * progress. This function doesn't actually claim. On successful
1139 * return, the caller has ownership of bd_claiming and bd_holder[s].
1142 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1143 * it multiple times.
1146 * 0 if @bdev can be claimed, -EBUSY otherwise.
1148 static int bd_prepare_to_claim(struct block_device *bdev,
1149 struct block_device *whole, void *holder)
1152 /* if someone else claimed, fail */
1153 if (!bd_may_claim(bdev, whole, holder))
1156 /* if claiming is already in progress, wait for it to finish */
1157 if (whole->bd_claiming) {
1158 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1161 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1162 spin_unlock(&bdev_lock);
1164 finish_wait(wq, &wait);
1165 spin_lock(&bdev_lock);
1174 * bd_start_claiming - start claiming a block device
1175 * @bdev: block device of interest
1176 * @holder: holder trying to claim @bdev
1178 * @bdev is about to be opened exclusively. Check @bdev can be opened
1179 * exclusively and mark that an exclusive open is in progress. Each
1180 * successful call to this function must be matched with a call to
1181 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1184 * This function is used to gain exclusive access to the block device
1185 * without actually causing other exclusive open attempts to fail. It
1186 * should be used when the open sequence itself requires exclusive
1187 * access but may subsequently fail.
1193 * Pointer to the block device containing @bdev on success, ERR_PTR()
1196 static struct block_device *bd_start_claiming(struct block_device *bdev,
1199 struct gendisk *disk;
1200 struct block_device *whole;
1206 * @bdev might not have been initialized properly yet, look up
1207 * and grab the outer block device the hard way.
1209 disk = get_gendisk(bdev->bd_dev, &partno);
1211 return ERR_PTR(-ENXIO);
1214 * Normally, @bdev should equal what's returned from bdget_disk()
1215 * if partno is 0; however, some drivers (floppy) use multiple
1216 * bdev's for the same physical device and @bdev may be one of the
1217 * aliases. Keep @bdev if partno is 0. This means claimer
1218 * tracking is broken for those devices but it has always been that
1222 whole = bdget_disk(disk, 0);
1224 whole = bdgrab(bdev);
1226 module_put(disk->fops->owner);
1229 return ERR_PTR(-ENOMEM);
1231 /* prepare to claim, if successful, mark claiming in progress */
1232 spin_lock(&bdev_lock);
1234 err = bd_prepare_to_claim(bdev, whole, holder);
1236 whole->bd_claiming = holder;
1237 spin_unlock(&bdev_lock);
1240 spin_unlock(&bdev_lock);
1242 return ERR_PTR(err);
1247 struct bd_holder_disk {
1248 struct list_head list;
1249 struct gendisk *disk;
1253 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1254 struct gendisk *disk)
1256 struct bd_holder_disk *holder;
1258 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1259 if (holder->disk == disk)
1264 static int add_symlink(struct kobject *from, struct kobject *to)
1266 return sysfs_create_link(from, to, kobject_name(to));
1269 static void del_symlink(struct kobject *from, struct kobject *to)
1271 sysfs_remove_link(from, kobject_name(to));
1275 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1276 * @bdev: the claimed slave bdev
1277 * @disk: the holding disk
1279 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1281 * This functions creates the following sysfs symlinks.
1283 * - from "slaves" directory of the holder @disk to the claimed @bdev
1284 * - from "holders" directory of the @bdev to the holder @disk
1286 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1287 * passed to bd_link_disk_holder(), then:
1289 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1290 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1292 * The caller must have claimed @bdev before calling this function and
1293 * ensure that both @bdev and @disk are valid during the creation and
1294 * lifetime of these symlinks.
1300 * 0 on success, -errno on failure.
1302 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1304 struct bd_holder_disk *holder;
1307 mutex_lock(&bdev->bd_mutex);
1309 WARN_ON_ONCE(!bdev->bd_holder);
1311 /* FIXME: remove the following once add_disk() handles errors */
1312 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1315 holder = bd_find_holder_disk(bdev, disk);
1321 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1327 INIT_LIST_HEAD(&holder->list);
1328 holder->disk = disk;
1331 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1335 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1339 * bdev could be deleted beneath us which would implicitly destroy
1340 * the holder directory. Hold on to it.
1342 kobject_get(bdev->bd_part->holder_dir);
1344 list_add(&holder->list, &bdev->bd_holder_disks);
1348 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1352 mutex_unlock(&bdev->bd_mutex);
1355 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1358 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1359 * @bdev: the calimed slave bdev
1360 * @disk: the holding disk
1362 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1367 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1369 struct bd_holder_disk *holder;
1371 mutex_lock(&bdev->bd_mutex);
1373 holder = bd_find_holder_disk(bdev, disk);
1375 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1376 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1377 del_symlink(bdev->bd_part->holder_dir,
1378 &disk_to_dev(disk)->kobj);
1379 kobject_put(bdev->bd_part->holder_dir);
1380 list_del_init(&holder->list);
1384 mutex_unlock(&bdev->bd_mutex);
1386 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1390 * flush_disk - invalidates all buffer-cache entries on a disk
1392 * @bdev: struct block device to be flushed
1393 * @kill_dirty: flag to guide handling of dirty inodes
1395 * Invalidates all buffer-cache entries on a disk. It should be called
1396 * when a disk has been changed -- either by a media change or online
1399 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1401 if (__invalidate_device(bdev, kill_dirty)) {
1402 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1403 "resized disk %s\n",
1404 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1409 if (disk_part_scan_enabled(bdev->bd_disk))
1410 bdev->bd_invalidated = 1;
1414 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1415 * @disk: struct gendisk to check
1416 * @bdev: struct bdev to adjust.
1418 * This routine checks to see if the bdev size does not match the disk size
1419 * and adjusts it if it differs.
1421 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1423 loff_t disk_size, bdev_size;
1425 disk_size = (loff_t)get_capacity(disk) << 9;
1426 bdev_size = i_size_read(bdev->bd_inode);
1427 if (disk_size != bdev_size) {
1429 "%s: detected capacity change from %lld to %lld\n",
1430 disk->disk_name, bdev_size, disk_size);
1431 i_size_write(bdev->bd_inode, disk_size);
1432 flush_disk(bdev, false);
1435 EXPORT_SYMBOL(check_disk_size_change);
1438 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1439 * @disk: struct gendisk to be revalidated
1441 * This routine is a wrapper for lower-level driver's revalidate_disk
1442 * call-backs. It is used to do common pre and post operations needed
1443 * for all revalidate_disk operations.
1445 int revalidate_disk(struct gendisk *disk)
1447 struct block_device *bdev;
1450 if (disk->fops->revalidate_disk)
1451 ret = disk->fops->revalidate_disk(disk);
1452 blk_integrity_revalidate(disk);
1453 bdev = bdget_disk(disk, 0);
1457 mutex_lock(&bdev->bd_mutex);
1458 check_disk_size_change(disk, bdev);
1459 bdev->bd_invalidated = 0;
1460 mutex_unlock(&bdev->bd_mutex);
1464 EXPORT_SYMBOL(revalidate_disk);
1467 * This routine checks whether a removable media has been changed,
1468 * and invalidates all buffer-cache-entries in that case. This
1469 * is a relatively slow routine, so we have to try to minimize using
1470 * it. Thus it is called only upon a 'mount' or 'open'. This
1471 * is the best way of combining speed and utility, I think.
1472 * People changing diskettes in the middle of an operation deserve
1475 int check_disk_change(struct block_device *bdev)
1477 struct gendisk *disk = bdev->bd_disk;
1478 const struct block_device_operations *bdops = disk->fops;
1479 unsigned int events;
1481 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1482 DISK_EVENT_EJECT_REQUEST);
1483 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1486 flush_disk(bdev, true);
1487 if (bdops->revalidate_disk)
1488 bdops->revalidate_disk(bdev->bd_disk);
1492 EXPORT_SYMBOL(check_disk_change);
1494 void bd_set_size(struct block_device *bdev, loff_t size)
1496 unsigned bsize = bdev_logical_block_size(bdev);
1498 inode_lock(bdev->bd_inode);
1499 i_size_write(bdev->bd_inode, size);
1500 inode_unlock(bdev->bd_inode);
1501 while (bsize < PAGE_SIZE) {
1506 bdev->bd_block_size = bsize;
1507 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1509 EXPORT_SYMBOL(bd_set_size);
1511 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1516 * mutex_lock(part->bd_mutex)
1517 * mutex_lock_nested(whole->bd_mutex, 1)
1520 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1522 struct gendisk *disk;
1523 struct module *owner;
1528 if (mode & FMODE_READ)
1530 if (mode & FMODE_WRITE)
1533 * hooks: /n/, see "layering violations".
1536 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1546 disk = get_gendisk(bdev->bd_dev, &partno);
1549 owner = disk->fops->owner;
1551 disk_block_events(disk);
1552 mutex_lock_nested(&bdev->bd_mutex, for_part);
1553 if (!bdev->bd_openers) {
1554 bdev->bd_disk = disk;
1555 bdev->bd_queue = disk->queue;
1556 bdev->bd_contains = bdev;
1557 if (bdev->bd_bdi == &noop_backing_dev_info)
1558 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1562 bdev->bd_part = disk_get_part(disk, partno);
1567 if (disk->fops->open) {
1568 ret = disk->fops->open(bdev, mode);
1569 if (ret == -ERESTARTSYS) {
1570 /* Lost a race with 'disk' being
1571 * deleted, try again.
1574 disk_put_part(bdev->bd_part);
1575 bdev->bd_part = NULL;
1576 bdev->bd_disk = NULL;
1577 bdev->bd_queue = NULL;
1578 mutex_unlock(&bdev->bd_mutex);
1579 disk_unblock_events(disk);
1587 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1590 * If the device is invalidated, rescan partition
1591 * if open succeeded or failed with -ENOMEDIUM.
1592 * The latter is necessary to prevent ghost
1593 * partitions on a removed medium.
1595 if (bdev->bd_invalidated) {
1597 rescan_partitions(disk, bdev);
1598 else if (ret == -ENOMEDIUM)
1599 invalidate_partitions(disk, bdev);
1605 struct block_device *whole;
1606 whole = bdget_disk(disk, 0);
1611 ret = __blkdev_get(whole, mode, 1);
1614 bdev->bd_contains = whole;
1615 bdev->bd_part = disk_get_part(disk, partno);
1616 if (!(disk->flags & GENHD_FL_UP) ||
1617 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1621 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1624 if (bdev->bd_contains == bdev) {
1626 if (bdev->bd_disk->fops->open)
1627 ret = bdev->bd_disk->fops->open(bdev, mode);
1628 /* the same as first opener case, read comment there */
1629 if (bdev->bd_invalidated) {
1631 rescan_partitions(bdev->bd_disk, bdev);
1632 else if (ret == -ENOMEDIUM)
1633 invalidate_partitions(bdev->bd_disk, bdev);
1636 goto out_unlock_bdev;
1638 /* only one opener holds refs to the module and disk */
1644 bdev->bd_part_count++;
1645 mutex_unlock(&bdev->bd_mutex);
1646 disk_unblock_events(disk);
1650 disk_put_part(bdev->bd_part);
1651 bdev->bd_disk = NULL;
1652 bdev->bd_part = NULL;
1653 bdev->bd_queue = NULL;
1654 bdi_put(bdev->bd_bdi);
1655 bdev->bd_bdi = &noop_backing_dev_info;
1656 if (bdev != bdev->bd_contains)
1657 __blkdev_put(bdev->bd_contains, mode, 1);
1658 bdev->bd_contains = NULL;
1660 mutex_unlock(&bdev->bd_mutex);
1661 disk_unblock_events(disk);
1671 * blkdev_get - open a block device
1672 * @bdev: block_device to open
1673 * @mode: FMODE_* mask
1674 * @holder: exclusive holder identifier
1676 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1677 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1678 * @holder is invalid. Exclusive opens may nest for the same @holder.
1680 * On success, the reference count of @bdev is unchanged. On failure,
1687 * 0 on success, -errno on failure.
1689 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1691 struct block_device *whole = NULL;
1694 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1696 if ((mode & FMODE_EXCL) && holder) {
1697 whole = bd_start_claiming(bdev, holder);
1698 if (IS_ERR(whole)) {
1700 return PTR_ERR(whole);
1704 res = __blkdev_get(bdev, mode, 0);
1707 struct gendisk *disk = whole->bd_disk;
1709 /* finish claiming */
1710 mutex_lock(&bdev->bd_mutex);
1711 spin_lock(&bdev_lock);
1714 BUG_ON(!bd_may_claim(bdev, whole, holder));
1716 * Note that for a whole device bd_holders
1717 * will be incremented twice, and bd_holder
1718 * will be set to bd_may_claim before being
1721 whole->bd_holders++;
1722 whole->bd_holder = bd_may_claim;
1724 bdev->bd_holder = holder;
1727 /* tell others that we're done */
1728 BUG_ON(whole->bd_claiming != holder);
1729 whole->bd_claiming = NULL;
1730 wake_up_bit(&whole->bd_claiming, 0);
1732 spin_unlock(&bdev_lock);
1735 * Block event polling for write claims if requested. Any
1736 * write holder makes the write_holder state stick until
1737 * all are released. This is good enough and tracking
1738 * individual writeable reference is too fragile given the
1739 * way @mode is used in blkdev_get/put().
1741 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1742 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1743 bdev->bd_write_holder = true;
1744 disk_block_events(disk);
1747 mutex_unlock(&bdev->bd_mutex);
1753 EXPORT_SYMBOL(blkdev_get);
1756 * blkdev_get_by_path - open a block device by name
1757 * @path: path to the block device to open
1758 * @mode: FMODE_* mask
1759 * @holder: exclusive holder identifier
1761 * Open the blockdevice described by the device file at @path. @mode
1762 * and @holder are identical to blkdev_get().
1764 * On success, the returned block_device has reference count of one.
1770 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1772 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1775 struct block_device *bdev;
1778 bdev = lookup_bdev(path);
1782 err = blkdev_get(bdev, mode, holder);
1784 return ERR_PTR(err);
1786 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1787 blkdev_put(bdev, mode);
1788 return ERR_PTR(-EACCES);
1793 EXPORT_SYMBOL(blkdev_get_by_path);
1796 * blkdev_get_by_dev - open a block device by device number
1797 * @dev: device number of block device to open
1798 * @mode: FMODE_* mask
1799 * @holder: exclusive holder identifier
1801 * Open the blockdevice described by device number @dev. @mode and
1802 * @holder are identical to blkdev_get().
1804 * Use it ONLY if you really do not have anything better - i.e. when
1805 * you are behind a truly sucky interface and all you are given is a
1806 * device number. _Never_ to be used for internal purposes. If you
1807 * ever need it - reconsider your API.
1809 * On success, the returned block_device has reference count of one.
1815 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1817 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1819 struct block_device *bdev;
1824 return ERR_PTR(-ENOMEM);
1826 err = blkdev_get(bdev, mode, holder);
1828 return ERR_PTR(err);
1832 EXPORT_SYMBOL(blkdev_get_by_dev);
1834 static int blkdev_open(struct inode * inode, struct file * filp)
1836 struct block_device *bdev;
1839 * Preserve backwards compatibility and allow large file access
1840 * even if userspace doesn't ask for it explicitly. Some mkfs
1841 * binary needs it. We might want to drop this workaround
1842 * during an unstable branch.
1844 filp->f_flags |= O_LARGEFILE;
1846 if (filp->f_flags & O_NDELAY)
1847 filp->f_mode |= FMODE_NDELAY;
1848 if (filp->f_flags & O_EXCL)
1849 filp->f_mode |= FMODE_EXCL;
1850 if ((filp->f_flags & O_ACCMODE) == 3)
1851 filp->f_mode |= FMODE_WRITE_IOCTL;
1853 bdev = bd_acquire(inode);
1857 filp->f_mapping = bdev->bd_inode->i_mapping;
1859 return blkdev_get(bdev, filp->f_mode, filp);
1862 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1864 struct gendisk *disk = bdev->bd_disk;
1865 struct block_device *victim = NULL;
1867 mutex_lock_nested(&bdev->bd_mutex, for_part);
1869 bdev->bd_part_count--;
1871 if (!--bdev->bd_openers) {
1872 WARN_ON_ONCE(bdev->bd_holders);
1873 sync_blockdev(bdev);
1876 bdev_write_inode(bdev);
1878 * Detaching bdev inode from its wb in __destroy_inode()
1879 * is too late: the queue which embeds its bdi (along with
1880 * root wb) can be gone as soon as we put_disk() below.
1882 inode_detach_wb(bdev->bd_inode);
1884 if (bdev->bd_contains == bdev) {
1885 if (disk->fops->release)
1886 disk->fops->release(disk, mode);
1888 if (!bdev->bd_openers) {
1889 struct module *owner = disk->fops->owner;
1891 disk_put_part(bdev->bd_part);
1892 bdev->bd_part = NULL;
1893 bdev->bd_disk = NULL;
1894 if (bdev != bdev->bd_contains)
1895 victim = bdev->bd_contains;
1896 bdev->bd_contains = NULL;
1901 mutex_unlock(&bdev->bd_mutex);
1904 __blkdev_put(victim, mode, 1);
1907 void blkdev_put(struct block_device *bdev, fmode_t mode)
1909 mutex_lock(&bdev->bd_mutex);
1911 if (mode & FMODE_EXCL) {
1915 * Release a claim on the device. The holder fields
1916 * are protected with bdev_lock. bd_mutex is to
1917 * synchronize disk_holder unlinking.
1919 spin_lock(&bdev_lock);
1921 WARN_ON_ONCE(--bdev->bd_holders < 0);
1922 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1924 /* bd_contains might point to self, check in a separate step */
1925 if ((bdev_free = !bdev->bd_holders))
1926 bdev->bd_holder = NULL;
1927 if (!bdev->bd_contains->bd_holders)
1928 bdev->bd_contains->bd_holder = NULL;
1930 spin_unlock(&bdev_lock);
1933 * If this was the last claim, remove holder link and
1934 * unblock evpoll if it was a write holder.
1936 if (bdev_free && bdev->bd_write_holder) {
1937 disk_unblock_events(bdev->bd_disk);
1938 bdev->bd_write_holder = false;
1943 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1944 * event. This is to ensure detection of media removal commanded
1945 * from userland - e.g. eject(1).
1947 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1949 mutex_unlock(&bdev->bd_mutex);
1951 __blkdev_put(bdev, mode, 0);
1953 EXPORT_SYMBOL(blkdev_put);
1955 static int blkdev_close(struct inode * inode, struct file * filp)
1957 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1958 blkdev_put(bdev, filp->f_mode);
1962 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1964 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1965 fmode_t mode = file->f_mode;
1968 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1969 * to updated it before every ioctl.
1971 if (file->f_flags & O_NDELAY)
1972 mode |= FMODE_NDELAY;
1974 mode &= ~FMODE_NDELAY;
1976 return blkdev_ioctl(bdev, mode, cmd, arg);
1980 * Write data to the block device. Only intended for the block device itself
1981 * and the raw driver which basically is a fake block device.
1983 * Does not take i_mutex for the write and thus is not for general purpose
1986 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1988 struct file *file = iocb->ki_filp;
1989 struct inode *bd_inode = bdev_file_inode(file);
1990 loff_t size = i_size_read(bd_inode);
1991 struct blk_plug plug;
1994 if (bdev_read_only(I_BDEV(bd_inode)))
1997 if (!iov_iter_count(from))
2000 if (iocb->ki_pos >= size)
2003 iov_iter_truncate(from, size - iocb->ki_pos);
2005 blk_start_plug(&plug);
2006 ret = __generic_file_write_iter(iocb, from);
2008 ret = generic_write_sync(iocb, ret);
2009 blk_finish_plug(&plug);
2012 EXPORT_SYMBOL_GPL(blkdev_write_iter);
2014 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
2016 struct file *file = iocb->ki_filp;
2017 struct inode *bd_inode = bdev_file_inode(file);
2018 loff_t size = i_size_read(bd_inode);
2019 loff_t pos = iocb->ki_pos;
2025 iov_iter_truncate(to, size);
2026 return generic_file_read_iter(iocb, to);
2028 EXPORT_SYMBOL_GPL(blkdev_read_iter);
2031 * Try to release a page associated with block device when the system
2032 * is under memory pressure.
2034 static int blkdev_releasepage(struct page *page, gfp_t wait)
2036 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
2038 if (super && super->s_op->bdev_try_to_free_page)
2039 return super->s_op->bdev_try_to_free_page(super, page, wait);
2041 return try_to_free_buffers(page);
2044 static int blkdev_writepages(struct address_space *mapping,
2045 struct writeback_control *wbc)
2047 if (dax_mapping(mapping)) {
2048 struct block_device *bdev = I_BDEV(mapping->host);
2050 return dax_writeback_mapping_range(mapping, bdev, wbc);
2052 return generic_writepages(mapping, wbc);
2055 static const struct address_space_operations def_blk_aops = {
2056 .readpage = blkdev_readpage,
2057 .readpages = blkdev_readpages,
2058 .writepage = blkdev_writepage,
2059 .write_begin = blkdev_write_begin,
2060 .write_end = blkdev_write_end,
2061 .writepages = blkdev_writepages,
2062 .releasepage = blkdev_releasepage,
2063 .direct_IO = blkdev_direct_IO,
2064 .is_dirty_writeback = buffer_check_dirty_writeback,
2067 #define BLKDEV_FALLOC_FL_SUPPORTED \
2068 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2069 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2071 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2074 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2075 struct request_queue *q = bdev_get_queue(bdev);
2076 struct address_space *mapping;
2077 loff_t end = start + len - 1;
2081 /* Fail if we don't recognize the flags. */
2082 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2085 /* Don't go off the end of the device. */
2086 isize = i_size_read(bdev->bd_inode);
2090 if (mode & FALLOC_FL_KEEP_SIZE) {
2091 len = isize - start;
2092 end = start + len - 1;
2098 * Don't allow IO that isn't aligned to logical block size.
2100 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2103 /* Invalidate the page cache, including dirty pages. */
2104 mapping = bdev->bd_inode->i_mapping;
2105 truncate_inode_pages_range(mapping, start, end);
2108 case FALLOC_FL_ZERO_RANGE:
2109 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2110 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2113 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2114 /* Only punch if the device can do zeroing discard. */
2115 if (!blk_queue_discard(q) || !q->limits.discard_zeroes_data)
2117 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2120 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2121 if (!blk_queue_discard(q))
2123 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2133 * Invalidate again; if someone wandered in and dirtied a page,
2134 * the caller will be given -EBUSY. The third argument is
2135 * inclusive, so the rounding here is safe.
2137 return invalidate_inode_pages2_range(mapping,
2138 start >> PAGE_SHIFT,
2142 const struct file_operations def_blk_fops = {
2143 .open = blkdev_open,
2144 .release = blkdev_close,
2145 .llseek = block_llseek,
2146 .read_iter = blkdev_read_iter,
2147 .write_iter = blkdev_write_iter,
2148 .mmap = generic_file_mmap,
2149 .fsync = blkdev_fsync,
2150 .unlocked_ioctl = block_ioctl,
2151 #ifdef CONFIG_COMPAT
2152 .compat_ioctl = compat_blkdev_ioctl,
2154 .splice_read = generic_file_splice_read,
2155 .splice_write = iter_file_splice_write,
2156 .fallocate = blkdev_fallocate,
2159 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2162 mm_segment_t old_fs = get_fs();
2164 res = blkdev_ioctl(bdev, 0, cmd, arg);
2169 EXPORT_SYMBOL(ioctl_by_bdev);
2172 * lookup_bdev - lookup a struct block_device by name
2173 * @pathname: special file representing the block device
2175 * Get a reference to the blockdevice at @pathname in the current
2176 * namespace if possible and return it. Return ERR_PTR(error)
2179 struct block_device *lookup_bdev(const char *pathname)
2181 struct block_device *bdev;
2182 struct inode *inode;
2186 if (!pathname || !*pathname)
2187 return ERR_PTR(-EINVAL);
2189 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2191 return ERR_PTR(error);
2193 inode = d_backing_inode(path.dentry);
2195 if (!S_ISBLK(inode->i_mode))
2198 if (!may_open_dev(&path))
2201 bdev = bd_acquire(inode);
2208 bdev = ERR_PTR(error);
2211 EXPORT_SYMBOL(lookup_bdev);
2213 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2215 struct super_block *sb = get_super(bdev);
2220 * no need to lock the super, get_super holds the
2221 * read mutex so the filesystem cannot go away
2222 * under us (->put_super runs with the write lock
2225 shrink_dcache_sb(sb);
2226 res = invalidate_inodes(sb, kill_dirty);
2229 invalidate_bdev(bdev);
2232 EXPORT_SYMBOL(__invalidate_device);
2234 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2236 struct inode *inode, *old_inode = NULL;
2238 spin_lock(&blockdev_superblock->s_inode_list_lock);
2239 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2240 struct address_space *mapping = inode->i_mapping;
2241 struct block_device *bdev;
2243 spin_lock(&inode->i_lock);
2244 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2245 mapping->nrpages == 0) {
2246 spin_unlock(&inode->i_lock);
2250 spin_unlock(&inode->i_lock);
2251 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2253 * We hold a reference to 'inode' so it couldn't have been
2254 * removed from s_inodes list while we dropped the
2255 * s_inode_list_lock We cannot iput the inode now as we can
2256 * be holding the last reference and we cannot iput it under
2257 * s_inode_list_lock. So we keep the reference and iput it
2262 bdev = I_BDEV(inode);
2264 mutex_lock(&bdev->bd_mutex);
2265 if (bdev->bd_openers)
2267 mutex_unlock(&bdev->bd_mutex);
2269 spin_lock(&blockdev_superblock->s_inode_list_lock);
2271 spin_unlock(&blockdev_superblock->s_inode_list_lock);