Loading...
Note: File does not exist in v6.9.4.
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/block_dev.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
7 */
8
9#include <linux/init.h>
10#include <linux/mm.h>
11#include <linux/fcntl.h>
12#include <linux/slab.h>
13#include <linux/kmod.h>
14#include <linux/major.h>
15#include <linux/device_cgroup.h>
16#include <linux/highmem.h>
17#include <linux/blkdev.h>
18#include <linux/backing-dev.h>
19#include <linux/module.h>
20#include <linux/blkpg.h>
21#include <linux/magic.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/pseudo_fs.h>
29#include <linux/uio.h>
30#include <linux/namei.h>
31#include <linux/log2.h>
32#include <linux/cleancache.h>
33#include <linux/task_io_accounting_ops.h>
34#include <linux/falloc.h>
35#include <linux/uaccess.h>
36#include <linux/suspend.h>
37#include "internal.h"
38
39struct bdev_inode {
40 struct block_device bdev;
41 struct inode vfs_inode;
42};
43
44static const struct address_space_operations def_blk_aops;
45
46static inline struct bdev_inode *BDEV_I(struct inode *inode)
47{
48 return container_of(inode, struct bdev_inode, vfs_inode);
49}
50
51struct block_device *I_BDEV(struct inode *inode)
52{
53 return &BDEV_I(inode)->bdev;
54}
55EXPORT_SYMBOL(I_BDEV);
56
57static void bdev_write_inode(struct block_device *bdev)
58{
59 struct inode *inode = bdev->bd_inode;
60 int ret;
61
62 spin_lock(&inode->i_lock);
63 while (inode->i_state & I_DIRTY) {
64 spin_unlock(&inode->i_lock);
65 ret = write_inode_now(inode, true);
66 if (ret) {
67 char name[BDEVNAME_SIZE];
68 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
69 "for block device %s (err=%d).\n",
70 bdevname(bdev, name), ret);
71 }
72 spin_lock(&inode->i_lock);
73 }
74 spin_unlock(&inode->i_lock);
75}
76
77/* Kill _all_ buffers and pagecache , dirty or not.. */
78static void kill_bdev(struct block_device *bdev)
79{
80 struct address_space *mapping = bdev->bd_inode->i_mapping;
81
82 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
83 return;
84
85 invalidate_bh_lrus();
86 truncate_inode_pages(mapping, 0);
87}
88
89/* Invalidate clean unused buffers and pagecache. */
90void invalidate_bdev(struct block_device *bdev)
91{
92 struct address_space *mapping = bdev->bd_inode->i_mapping;
93
94 if (mapping->nrpages) {
95 invalidate_bh_lrus();
96 lru_add_drain_all(); /* make sure all lru add caches are flushed */
97 invalidate_mapping_pages(mapping, 0, -1);
98 }
99 /* 99% of the time, we don't need to flush the cleancache on the bdev.
100 * But, for the strange corners, lets be cautious
101 */
102 cleancache_invalidate_inode(mapping);
103}
104EXPORT_SYMBOL(invalidate_bdev);
105
106static void set_init_blocksize(struct block_device *bdev)
107{
108 bdev->bd_inode->i_blkbits = blksize_bits(bdev_logical_block_size(bdev));
109}
110
111int set_blocksize(struct block_device *bdev, int size)
112{
113 /* Size must be a power of two, and between 512 and PAGE_SIZE */
114 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
115 return -EINVAL;
116
117 /* Size cannot be smaller than the size supported by the device */
118 if (size < bdev_logical_block_size(bdev))
119 return -EINVAL;
120
121 /* Don't change the size if it is same as current */
122 if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
123 sync_blockdev(bdev);
124 bdev->bd_inode->i_blkbits = blksize_bits(size);
125 kill_bdev(bdev);
126 }
127 return 0;
128}
129
130EXPORT_SYMBOL(set_blocksize);
131
132int sb_set_blocksize(struct super_block *sb, int size)
133{
134 if (set_blocksize(sb->s_bdev, size))
135 return 0;
136 /* If we get here, we know size is power of two
137 * and it's value is between 512 and PAGE_SIZE */
138 sb->s_blocksize = size;
139 sb->s_blocksize_bits = blksize_bits(size);
140 return sb->s_blocksize;
141}
142
143EXPORT_SYMBOL(sb_set_blocksize);
144
145int sb_min_blocksize(struct super_block *sb, int size)
146{
147 int minsize = bdev_logical_block_size(sb->s_bdev);
148 if (size < minsize)
149 size = minsize;
150 return sb_set_blocksize(sb, size);
151}
152
153EXPORT_SYMBOL(sb_min_blocksize);
154
155static int
156blkdev_get_block(struct inode *inode, sector_t iblock,
157 struct buffer_head *bh, int create)
158{
159 bh->b_bdev = I_BDEV(inode);
160 bh->b_blocknr = iblock;
161 set_buffer_mapped(bh);
162 return 0;
163}
164
165static struct inode *bdev_file_inode(struct file *file)
166{
167 return file->f_mapping->host;
168}
169
170static unsigned int dio_bio_write_op(struct kiocb *iocb)
171{
172 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
173
174 /* avoid the need for a I/O completion work item */
175 if (iocb->ki_flags & IOCB_DSYNC)
176 op |= REQ_FUA;
177 return op;
178}
179
180#define DIO_INLINE_BIO_VECS 4
181
182static void blkdev_bio_end_io_simple(struct bio *bio)
183{
184 struct task_struct *waiter = bio->bi_private;
185
186 WRITE_ONCE(bio->bi_private, NULL);
187 blk_wake_io_task(waiter);
188}
189
190static ssize_t
191__blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
192 int nr_pages)
193{
194 struct file *file = iocb->ki_filp;
195 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
196 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
197 loff_t pos = iocb->ki_pos;
198 bool should_dirty = false;
199 struct bio bio;
200 ssize_t ret;
201 blk_qc_t qc;
202
203 if ((pos | iov_iter_alignment(iter)) &
204 (bdev_logical_block_size(bdev) - 1))
205 return -EINVAL;
206
207 if (nr_pages <= DIO_INLINE_BIO_VECS)
208 vecs = inline_vecs;
209 else {
210 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
211 GFP_KERNEL);
212 if (!vecs)
213 return -ENOMEM;
214 }
215
216 bio_init(&bio, vecs, nr_pages);
217 bio_set_dev(&bio, bdev);
218 bio.bi_iter.bi_sector = pos >> 9;
219 bio.bi_write_hint = iocb->ki_hint;
220 bio.bi_private = current;
221 bio.bi_end_io = blkdev_bio_end_io_simple;
222 bio.bi_ioprio = iocb->ki_ioprio;
223
224 ret = bio_iov_iter_get_pages(&bio, iter);
225 if (unlikely(ret))
226 goto out;
227 ret = bio.bi_iter.bi_size;
228
229 if (iov_iter_rw(iter) == READ) {
230 bio.bi_opf = REQ_OP_READ;
231 if (iter_is_iovec(iter))
232 should_dirty = true;
233 } else {
234 bio.bi_opf = dio_bio_write_op(iocb);
235 task_io_account_write(ret);
236 }
237 if (iocb->ki_flags & IOCB_HIPRI)
238 bio_set_polled(&bio, iocb);
239
240 qc = submit_bio(&bio);
241 for (;;) {
242 set_current_state(TASK_UNINTERRUPTIBLE);
243 if (!READ_ONCE(bio.bi_private))
244 break;
245 if (!(iocb->ki_flags & IOCB_HIPRI) ||
246 !blk_poll(bdev_get_queue(bdev), qc, true))
247 blk_io_schedule();
248 }
249 __set_current_state(TASK_RUNNING);
250
251 bio_release_pages(&bio, should_dirty);
252 if (unlikely(bio.bi_status))
253 ret = blk_status_to_errno(bio.bi_status);
254
255out:
256 if (vecs != inline_vecs)
257 kfree(vecs);
258
259 bio_uninit(&bio);
260
261 return ret;
262}
263
264struct blkdev_dio {
265 union {
266 struct kiocb *iocb;
267 struct task_struct *waiter;
268 };
269 size_t size;
270 atomic_t ref;
271 bool multi_bio : 1;
272 bool should_dirty : 1;
273 bool is_sync : 1;
274 struct bio bio;
275};
276
277static struct bio_set blkdev_dio_pool;
278
279static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
280{
281 struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
282 struct request_queue *q = bdev_get_queue(bdev);
283
284 return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
285}
286
287static void blkdev_bio_end_io(struct bio *bio)
288{
289 struct blkdev_dio *dio = bio->bi_private;
290 bool should_dirty = dio->should_dirty;
291
292 if (bio->bi_status && !dio->bio.bi_status)
293 dio->bio.bi_status = bio->bi_status;
294
295 if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
296 if (!dio->is_sync) {
297 struct kiocb *iocb = dio->iocb;
298 ssize_t ret;
299
300 if (likely(!dio->bio.bi_status)) {
301 ret = dio->size;
302 iocb->ki_pos += ret;
303 } else {
304 ret = blk_status_to_errno(dio->bio.bi_status);
305 }
306
307 dio->iocb->ki_complete(iocb, ret, 0);
308 if (dio->multi_bio)
309 bio_put(&dio->bio);
310 } else {
311 struct task_struct *waiter = dio->waiter;
312
313 WRITE_ONCE(dio->waiter, NULL);
314 blk_wake_io_task(waiter);
315 }
316 }
317
318 if (should_dirty) {
319 bio_check_pages_dirty(bio);
320 } else {
321 bio_release_pages(bio, false);
322 bio_put(bio);
323 }
324}
325
326static ssize_t
327__blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
328{
329 struct file *file = iocb->ki_filp;
330 struct inode *inode = bdev_file_inode(file);
331 struct block_device *bdev = I_BDEV(inode);
332 struct blk_plug plug;
333 struct blkdev_dio *dio;
334 struct bio *bio;
335 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
336 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
337 loff_t pos = iocb->ki_pos;
338 blk_qc_t qc = BLK_QC_T_NONE;
339 int ret = 0;
340
341 if ((pos | iov_iter_alignment(iter)) &
342 (bdev_logical_block_size(bdev) - 1))
343 return -EINVAL;
344
345 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
346
347 dio = container_of(bio, struct blkdev_dio, bio);
348 dio->is_sync = is_sync = is_sync_kiocb(iocb);
349 if (dio->is_sync) {
350 dio->waiter = current;
351 bio_get(bio);
352 } else {
353 dio->iocb = iocb;
354 }
355
356 dio->size = 0;
357 dio->multi_bio = false;
358 dio->should_dirty = is_read && iter_is_iovec(iter);
359
360 /*
361 * Don't plug for HIPRI/polled IO, as those should go straight
362 * to issue
363 */
364 if (!is_poll)
365 blk_start_plug(&plug);
366
367 for (;;) {
368 bio_set_dev(bio, bdev);
369 bio->bi_iter.bi_sector = pos >> 9;
370 bio->bi_write_hint = iocb->ki_hint;
371 bio->bi_private = dio;
372 bio->bi_end_io = blkdev_bio_end_io;
373 bio->bi_ioprio = iocb->ki_ioprio;
374
375 ret = bio_iov_iter_get_pages(bio, iter);
376 if (unlikely(ret)) {
377 bio->bi_status = BLK_STS_IOERR;
378 bio_endio(bio);
379 break;
380 }
381
382 if (is_read) {
383 bio->bi_opf = REQ_OP_READ;
384 if (dio->should_dirty)
385 bio_set_pages_dirty(bio);
386 } else {
387 bio->bi_opf = dio_bio_write_op(iocb);
388 task_io_account_write(bio->bi_iter.bi_size);
389 }
390
391 dio->size += bio->bi_iter.bi_size;
392 pos += bio->bi_iter.bi_size;
393
394 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
395 if (!nr_pages) {
396 bool polled = false;
397
398 if (iocb->ki_flags & IOCB_HIPRI) {
399 bio_set_polled(bio, iocb);
400 polled = true;
401 }
402
403 qc = submit_bio(bio);
404
405 if (polled)
406 WRITE_ONCE(iocb->ki_cookie, qc);
407 break;
408 }
409
410 if (!dio->multi_bio) {
411 /*
412 * AIO needs an extra reference to ensure the dio
413 * structure which is embedded into the first bio
414 * stays around.
415 */
416 if (!is_sync)
417 bio_get(bio);
418 dio->multi_bio = true;
419 atomic_set(&dio->ref, 2);
420 } else {
421 atomic_inc(&dio->ref);
422 }
423
424 submit_bio(bio);
425 bio = bio_alloc(GFP_KERNEL, nr_pages);
426 }
427
428 if (!is_poll)
429 blk_finish_plug(&plug);
430
431 if (!is_sync)
432 return -EIOCBQUEUED;
433
434 for (;;) {
435 set_current_state(TASK_UNINTERRUPTIBLE);
436 if (!READ_ONCE(dio->waiter))
437 break;
438
439 if (!(iocb->ki_flags & IOCB_HIPRI) ||
440 !blk_poll(bdev_get_queue(bdev), qc, true))
441 blk_io_schedule();
442 }
443 __set_current_state(TASK_RUNNING);
444
445 if (!ret)
446 ret = blk_status_to_errno(dio->bio.bi_status);
447 if (likely(!ret))
448 ret = dio->size;
449
450 bio_put(&dio->bio);
451 return ret;
452}
453
454static ssize_t
455blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
456{
457 int nr_pages;
458
459 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
460 if (!nr_pages)
461 return 0;
462 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
463 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
464
465 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
466}
467
468static __init int blkdev_init(void)
469{
470 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
471}
472module_init(blkdev_init);
473
474int __sync_blockdev(struct block_device *bdev, int wait)
475{
476 if (!bdev)
477 return 0;
478 if (!wait)
479 return filemap_flush(bdev->bd_inode->i_mapping);
480 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
481}
482
483/*
484 * Write out and wait upon all the dirty data associated with a block
485 * device via its mapping. Does not take the superblock lock.
486 */
487int sync_blockdev(struct block_device *bdev)
488{
489 return __sync_blockdev(bdev, 1);
490}
491EXPORT_SYMBOL(sync_blockdev);
492
493/*
494 * Write out and wait upon all dirty data associated with this
495 * device. Filesystem data as well as the underlying block
496 * device. Takes the superblock lock.
497 */
498int fsync_bdev(struct block_device *bdev)
499{
500 struct super_block *sb = get_super(bdev);
501 if (sb) {
502 int res = sync_filesystem(sb);
503 drop_super(sb);
504 return res;
505 }
506 return sync_blockdev(bdev);
507}
508EXPORT_SYMBOL(fsync_bdev);
509
510/**
511 * freeze_bdev -- lock a filesystem and force it into a consistent state
512 * @bdev: blockdevice to lock
513 *
514 * If a superblock is found on this device, we take the s_umount semaphore
515 * on it to make sure nobody unmounts until the snapshot creation is done.
516 * The reference counter (bd_fsfreeze_count) guarantees that only the last
517 * unfreeze process can unfreeze the frozen filesystem actually when multiple
518 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
519 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
520 * actually.
521 */
522struct super_block *freeze_bdev(struct block_device *bdev)
523{
524 struct super_block *sb;
525 int error = 0;
526
527 mutex_lock(&bdev->bd_fsfreeze_mutex);
528 if (++bdev->bd_fsfreeze_count > 1) {
529 /*
530 * We don't even need to grab a reference - the first call
531 * to freeze_bdev grab an active reference and only the last
532 * thaw_bdev drops it.
533 */
534 sb = get_super(bdev);
535 if (sb)
536 drop_super(sb);
537 mutex_unlock(&bdev->bd_fsfreeze_mutex);
538 return sb;
539 }
540
541 sb = get_active_super(bdev);
542 if (!sb)
543 goto out;
544 if (sb->s_op->freeze_super)
545 error = sb->s_op->freeze_super(sb);
546 else
547 error = freeze_super(sb);
548 if (error) {
549 deactivate_super(sb);
550 bdev->bd_fsfreeze_count--;
551 mutex_unlock(&bdev->bd_fsfreeze_mutex);
552 return ERR_PTR(error);
553 }
554 deactivate_super(sb);
555 out:
556 sync_blockdev(bdev);
557 mutex_unlock(&bdev->bd_fsfreeze_mutex);
558 return sb; /* thaw_bdev releases s->s_umount */
559}
560EXPORT_SYMBOL(freeze_bdev);
561
562/**
563 * thaw_bdev -- unlock filesystem
564 * @bdev: blockdevice to unlock
565 * @sb: associated superblock
566 *
567 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
568 */
569int thaw_bdev(struct block_device *bdev, struct super_block *sb)
570{
571 int error = -EINVAL;
572
573 mutex_lock(&bdev->bd_fsfreeze_mutex);
574 if (!bdev->bd_fsfreeze_count)
575 goto out;
576
577 error = 0;
578 if (--bdev->bd_fsfreeze_count > 0)
579 goto out;
580
581 if (!sb)
582 goto out;
583
584 if (sb->s_op->thaw_super)
585 error = sb->s_op->thaw_super(sb);
586 else
587 error = thaw_super(sb);
588 if (error)
589 bdev->bd_fsfreeze_count++;
590out:
591 mutex_unlock(&bdev->bd_fsfreeze_mutex);
592 return error;
593}
594EXPORT_SYMBOL(thaw_bdev);
595
596static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
597{
598 return block_write_full_page(page, blkdev_get_block, wbc);
599}
600
601static int blkdev_readpage(struct file * file, struct page * page)
602{
603 return block_read_full_page(page, blkdev_get_block);
604}
605
606static void blkdev_readahead(struct readahead_control *rac)
607{
608 mpage_readahead(rac, blkdev_get_block);
609}
610
611static int blkdev_write_begin(struct file *file, struct address_space *mapping,
612 loff_t pos, unsigned len, unsigned flags,
613 struct page **pagep, void **fsdata)
614{
615 return block_write_begin(mapping, pos, len, flags, pagep,
616 blkdev_get_block);
617}
618
619static int blkdev_write_end(struct file *file, struct address_space *mapping,
620 loff_t pos, unsigned len, unsigned copied,
621 struct page *page, void *fsdata)
622{
623 int ret;
624 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
625
626 unlock_page(page);
627 put_page(page);
628
629 return ret;
630}
631
632/*
633 * private llseek:
634 * for a block special file file_inode(file)->i_size is zero
635 * so we compute the size by hand (just as in block_read/write above)
636 */
637static loff_t block_llseek(struct file *file, loff_t offset, int whence)
638{
639 struct inode *bd_inode = bdev_file_inode(file);
640 loff_t retval;
641
642 inode_lock(bd_inode);
643 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
644 inode_unlock(bd_inode);
645 return retval;
646}
647
648int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
649{
650 struct inode *bd_inode = bdev_file_inode(filp);
651 struct block_device *bdev = I_BDEV(bd_inode);
652 int error;
653
654 error = file_write_and_wait_range(filp, start, end);
655 if (error)
656 return error;
657
658 /*
659 * There is no need to serialise calls to blkdev_issue_flush with
660 * i_mutex and doing so causes performance issues with concurrent
661 * O_SYNC writers to a block device.
662 */
663 error = blkdev_issue_flush(bdev, GFP_KERNEL);
664 if (error == -EOPNOTSUPP)
665 error = 0;
666
667 return error;
668}
669EXPORT_SYMBOL(blkdev_fsync);
670
671/**
672 * bdev_read_page() - Start reading a page from a block device
673 * @bdev: The device to read the page from
674 * @sector: The offset on the device to read the page to (need not be aligned)
675 * @page: The page to read
676 *
677 * On entry, the page should be locked. It will be unlocked when the page
678 * has been read. If the block driver implements rw_page synchronously,
679 * that will be true on exit from this function, but it need not be.
680 *
681 * Errors returned by this function are usually "soft", eg out of memory, or
682 * queue full; callers should try a different route to read this page rather
683 * than propagate an error back up the stack.
684 *
685 * Return: negative errno if an error occurs, 0 if submission was successful.
686 */
687int bdev_read_page(struct block_device *bdev, sector_t sector,
688 struct page *page)
689{
690 const struct block_device_operations *ops = bdev->bd_disk->fops;
691 int result = -EOPNOTSUPP;
692
693 if (!ops->rw_page || bdev_get_integrity(bdev))
694 return result;
695
696 result = blk_queue_enter(bdev->bd_disk->queue, 0);
697 if (result)
698 return result;
699 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
700 REQ_OP_READ);
701 blk_queue_exit(bdev->bd_disk->queue);
702 return result;
703}
704
705/**
706 * bdev_write_page() - Start writing a page to a block device
707 * @bdev: The device to write the page to
708 * @sector: The offset on the device to write the page to (need not be aligned)
709 * @page: The page to write
710 * @wbc: The writeback_control for the write
711 *
712 * On entry, the page should be locked and not currently under writeback.
713 * On exit, if the write started successfully, the page will be unlocked and
714 * under writeback. If the write failed already (eg the driver failed to
715 * queue the page to the device), the page will still be locked. If the
716 * caller is a ->writepage implementation, it will need to unlock the page.
717 *
718 * Errors returned by this function are usually "soft", eg out of memory, or
719 * queue full; callers should try a different route to write this page rather
720 * than propagate an error back up the stack.
721 *
722 * Return: negative errno if an error occurs, 0 if submission was successful.
723 */
724int bdev_write_page(struct block_device *bdev, sector_t sector,
725 struct page *page, struct writeback_control *wbc)
726{
727 int result;
728 const struct block_device_operations *ops = bdev->bd_disk->fops;
729
730 if (!ops->rw_page || bdev_get_integrity(bdev))
731 return -EOPNOTSUPP;
732 result = blk_queue_enter(bdev->bd_disk->queue, 0);
733 if (result)
734 return result;
735
736 set_page_writeback(page);
737 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
738 REQ_OP_WRITE);
739 if (result) {
740 end_page_writeback(page);
741 } else {
742 clean_page_buffers(page);
743 unlock_page(page);
744 }
745 blk_queue_exit(bdev->bd_disk->queue);
746 return result;
747}
748
749/*
750 * pseudo-fs
751 */
752
753static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
754static struct kmem_cache * bdev_cachep __read_mostly;
755
756static struct inode *bdev_alloc_inode(struct super_block *sb)
757{
758 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
759 if (!ei)
760 return NULL;
761 return &ei->vfs_inode;
762}
763
764static void bdev_free_inode(struct inode *inode)
765{
766 kmem_cache_free(bdev_cachep, BDEV_I(inode));
767}
768
769static void init_once(void *foo)
770{
771 struct bdev_inode *ei = (struct bdev_inode *) foo;
772 struct block_device *bdev = &ei->bdev;
773
774 memset(bdev, 0, sizeof(*bdev));
775 mutex_init(&bdev->bd_mutex);
776#ifdef CONFIG_SYSFS
777 INIT_LIST_HEAD(&bdev->bd_holder_disks);
778#endif
779 bdev->bd_bdi = &noop_backing_dev_info;
780 inode_init_once(&ei->vfs_inode);
781 /* Initialize mutex for freeze. */
782 mutex_init(&bdev->bd_fsfreeze_mutex);
783}
784
785static void bdev_evict_inode(struct inode *inode)
786{
787 struct block_device *bdev = &BDEV_I(inode)->bdev;
788 truncate_inode_pages_final(&inode->i_data);
789 invalidate_inode_buffers(inode); /* is it needed here? */
790 clear_inode(inode);
791 /* Detach inode from wb early as bdi_put() may free bdi->wb */
792 inode_detach_wb(inode);
793 if (bdev->bd_bdi != &noop_backing_dev_info) {
794 bdi_put(bdev->bd_bdi);
795 bdev->bd_bdi = &noop_backing_dev_info;
796 }
797}
798
799static const struct super_operations bdev_sops = {
800 .statfs = simple_statfs,
801 .alloc_inode = bdev_alloc_inode,
802 .free_inode = bdev_free_inode,
803 .drop_inode = generic_delete_inode,
804 .evict_inode = bdev_evict_inode,
805};
806
807static int bd_init_fs_context(struct fs_context *fc)
808{
809 struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
810 if (!ctx)
811 return -ENOMEM;
812 fc->s_iflags |= SB_I_CGROUPWB;
813 ctx->ops = &bdev_sops;
814 return 0;
815}
816
817static struct file_system_type bd_type = {
818 .name = "bdev",
819 .init_fs_context = bd_init_fs_context,
820 .kill_sb = kill_anon_super,
821};
822
823struct super_block *blockdev_superblock __read_mostly;
824EXPORT_SYMBOL_GPL(blockdev_superblock);
825
826void __init bdev_cache_init(void)
827{
828 int err;
829 static struct vfsmount *bd_mnt;
830
831 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
832 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
833 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
834 init_once);
835 err = register_filesystem(&bd_type);
836 if (err)
837 panic("Cannot register bdev pseudo-fs");
838 bd_mnt = kern_mount(&bd_type);
839 if (IS_ERR(bd_mnt))
840 panic("Cannot create bdev pseudo-fs");
841 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
842}
843
844/*
845 * Most likely _very_ bad one - but then it's hardly critical for small
846 * /dev and can be fixed when somebody will need really large one.
847 * Keep in mind that it will be fed through icache hash function too.
848 */
849static inline unsigned long hash(dev_t dev)
850{
851 return MAJOR(dev)+MINOR(dev);
852}
853
854static int bdev_test(struct inode *inode, void *data)
855{
856 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
857}
858
859static int bdev_set(struct inode *inode, void *data)
860{
861 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
862 return 0;
863}
864
865struct block_device *bdget(dev_t dev)
866{
867 struct block_device *bdev;
868 struct inode *inode;
869
870 inode = iget5_locked(blockdev_superblock, hash(dev),
871 bdev_test, bdev_set, &dev);
872
873 if (!inode)
874 return NULL;
875
876 bdev = &BDEV_I(inode)->bdev;
877
878 if (inode->i_state & I_NEW) {
879 bdev->bd_contains = NULL;
880 bdev->bd_super = NULL;
881 bdev->bd_inode = inode;
882 bdev->bd_part_count = 0;
883 bdev->bd_invalidated = 0;
884 inode->i_mode = S_IFBLK;
885 inode->i_rdev = dev;
886 inode->i_bdev = bdev;
887 inode->i_data.a_ops = &def_blk_aops;
888 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
889 unlock_new_inode(inode);
890 }
891 return bdev;
892}
893
894EXPORT_SYMBOL(bdget);
895
896/**
897 * bdgrab -- Grab a reference to an already referenced block device
898 * @bdev: Block device to grab a reference to.
899 */
900struct block_device *bdgrab(struct block_device *bdev)
901{
902 ihold(bdev->bd_inode);
903 return bdev;
904}
905EXPORT_SYMBOL(bdgrab);
906
907long nr_blockdev_pages(void)
908{
909 struct inode *inode;
910 long ret = 0;
911
912 spin_lock(&blockdev_superblock->s_inode_list_lock);
913 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
914 ret += inode->i_mapping->nrpages;
915 spin_unlock(&blockdev_superblock->s_inode_list_lock);
916
917 return ret;
918}
919
920void bdput(struct block_device *bdev)
921{
922 iput(bdev->bd_inode);
923}
924
925EXPORT_SYMBOL(bdput);
926
927static struct block_device *bd_acquire(struct inode *inode)
928{
929 struct block_device *bdev;
930
931 spin_lock(&bdev_lock);
932 bdev = inode->i_bdev;
933 if (bdev && !inode_unhashed(bdev->bd_inode)) {
934 bdgrab(bdev);
935 spin_unlock(&bdev_lock);
936 return bdev;
937 }
938 spin_unlock(&bdev_lock);
939
940 /*
941 * i_bdev references block device inode that was already shut down
942 * (corresponding device got removed). Remove the reference and look
943 * up block device inode again just in case new device got
944 * reestablished under the same device number.
945 */
946 if (bdev)
947 bd_forget(inode);
948
949 bdev = bdget(inode->i_rdev);
950 if (bdev) {
951 spin_lock(&bdev_lock);
952 if (!inode->i_bdev) {
953 /*
954 * We take an additional reference to bd_inode,
955 * and it's released in clear_inode() of inode.
956 * So, we can access it via ->i_mapping always
957 * without igrab().
958 */
959 bdgrab(bdev);
960 inode->i_bdev = bdev;
961 inode->i_mapping = bdev->bd_inode->i_mapping;
962 }
963 spin_unlock(&bdev_lock);
964 }
965 return bdev;
966}
967
968/* Call when you free inode */
969
970void bd_forget(struct inode *inode)
971{
972 struct block_device *bdev = NULL;
973
974 spin_lock(&bdev_lock);
975 if (!sb_is_blkdev_sb(inode->i_sb))
976 bdev = inode->i_bdev;
977 inode->i_bdev = NULL;
978 inode->i_mapping = &inode->i_data;
979 spin_unlock(&bdev_lock);
980
981 if (bdev)
982 bdput(bdev);
983}
984
985/**
986 * bd_may_claim - test whether a block device can be claimed
987 * @bdev: block device of interest
988 * @whole: whole block device containing @bdev, may equal @bdev
989 * @holder: holder trying to claim @bdev
990 *
991 * Test whether @bdev can be claimed by @holder.
992 *
993 * CONTEXT:
994 * spin_lock(&bdev_lock).
995 *
996 * RETURNS:
997 * %true if @bdev can be claimed, %false otherwise.
998 */
999static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1000 void *holder)
1001{
1002 if (bdev->bd_holder == holder)
1003 return true; /* already a holder */
1004 else if (bdev->bd_holder != NULL)
1005 return false; /* held by someone else */
1006 else if (whole == bdev)
1007 return true; /* is a whole device which isn't held */
1008
1009 else if (whole->bd_holder == bd_may_claim)
1010 return true; /* is a partition of a device that is being partitioned */
1011 else if (whole->bd_holder != NULL)
1012 return false; /* is a partition of a held device */
1013 else
1014 return true; /* is a partition of an un-held device */
1015}
1016
1017/**
1018 * bd_prepare_to_claim - claim a block device
1019 * @bdev: block device of interest
1020 * @whole: the whole device containing @bdev, may equal @bdev
1021 * @holder: holder trying to claim @bdev
1022 *
1023 * Claim @bdev. This function fails if @bdev is already claimed by another
1024 * holder and waits if another claiming is in progress. return, the caller
1025 * has ownership of bd_claiming and bd_holder[s].
1026 *
1027 * RETURNS:
1028 * 0 if @bdev can be claimed, -EBUSY otherwise.
1029 */
1030int bd_prepare_to_claim(struct block_device *bdev, struct block_device *whole,
1031 void *holder)
1032{
1033retry:
1034 spin_lock(&bdev_lock);
1035 /* if someone else claimed, fail */
1036 if (!bd_may_claim(bdev, whole, holder)) {
1037 spin_unlock(&bdev_lock);
1038 return -EBUSY;
1039 }
1040
1041 /* if claiming is already in progress, wait for it to finish */
1042 if (whole->bd_claiming) {
1043 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1044 DEFINE_WAIT(wait);
1045
1046 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1047 spin_unlock(&bdev_lock);
1048 schedule();
1049 finish_wait(wq, &wait);
1050 goto retry;
1051 }
1052
1053 /* yay, all mine */
1054 whole->bd_claiming = holder;
1055 spin_unlock(&bdev_lock);
1056 return 0;
1057}
1058EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
1059
1060static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1061{
1062 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1063
1064 if (!disk)
1065 return NULL;
1066 /*
1067 * Now that we hold gendisk reference we make sure bdev we looked up is
1068 * not stale. If it is, it means device got removed and created before
1069 * we looked up gendisk and we fail open in such case. Associating
1070 * unhashed bdev with newly created gendisk could lead to two bdevs
1071 * (and thus two independent caches) being associated with one device
1072 * which is bad.
1073 */
1074 if (inode_unhashed(bdev->bd_inode)) {
1075 put_disk_and_module(disk);
1076 return NULL;
1077 }
1078 return disk;
1079}
1080
1081static void bd_clear_claiming(struct block_device *whole, void *holder)
1082{
1083 lockdep_assert_held(&bdev_lock);
1084 /* tell others that we're done */
1085 BUG_ON(whole->bd_claiming != holder);
1086 whole->bd_claiming = NULL;
1087 wake_up_bit(&whole->bd_claiming, 0);
1088}
1089
1090/**
1091 * bd_finish_claiming - finish claiming of a block device
1092 * @bdev: block device of interest
1093 * @whole: whole block device
1094 * @holder: holder that has claimed @bdev
1095 *
1096 * Finish exclusive open of a block device. Mark the device as exlusively
1097 * open by the holder and wake up all waiters for exclusive open to finish.
1098 */
1099static void bd_finish_claiming(struct block_device *bdev,
1100 struct block_device *whole, void *holder)
1101{
1102 spin_lock(&bdev_lock);
1103 BUG_ON(!bd_may_claim(bdev, whole, holder));
1104 /*
1105 * Note that for a whole device bd_holders will be incremented twice,
1106 * and bd_holder will be set to bd_may_claim before being set to holder
1107 */
1108 whole->bd_holders++;
1109 whole->bd_holder = bd_may_claim;
1110 bdev->bd_holders++;
1111 bdev->bd_holder = holder;
1112 bd_clear_claiming(whole, holder);
1113 spin_unlock(&bdev_lock);
1114}
1115
1116/**
1117 * bd_abort_claiming - abort claiming of a block device
1118 * @bdev: block device of interest
1119 * @whole: whole block device
1120 * @holder: holder that has claimed @bdev
1121 *
1122 * Abort claiming of a block device when the exclusive open failed. This can be
1123 * also used when exclusive open is not actually desired and we just needed
1124 * to block other exclusive openers for a while.
1125 */
1126void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
1127 void *holder)
1128{
1129 spin_lock(&bdev_lock);
1130 bd_clear_claiming(whole, holder);
1131 spin_unlock(&bdev_lock);
1132}
1133EXPORT_SYMBOL(bd_abort_claiming);
1134
1135#ifdef CONFIG_SYSFS
1136struct bd_holder_disk {
1137 struct list_head list;
1138 struct gendisk *disk;
1139 int refcnt;
1140};
1141
1142static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1143 struct gendisk *disk)
1144{
1145 struct bd_holder_disk *holder;
1146
1147 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1148 if (holder->disk == disk)
1149 return holder;
1150 return NULL;
1151}
1152
1153static int add_symlink(struct kobject *from, struct kobject *to)
1154{
1155 return sysfs_create_link(from, to, kobject_name(to));
1156}
1157
1158static void del_symlink(struct kobject *from, struct kobject *to)
1159{
1160 sysfs_remove_link(from, kobject_name(to));
1161}
1162
1163/**
1164 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1165 * @bdev: the claimed slave bdev
1166 * @disk: the holding disk
1167 *
1168 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1169 *
1170 * This functions creates the following sysfs symlinks.
1171 *
1172 * - from "slaves" directory of the holder @disk to the claimed @bdev
1173 * - from "holders" directory of the @bdev to the holder @disk
1174 *
1175 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1176 * passed to bd_link_disk_holder(), then:
1177 *
1178 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1179 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1180 *
1181 * The caller must have claimed @bdev before calling this function and
1182 * ensure that both @bdev and @disk are valid during the creation and
1183 * lifetime of these symlinks.
1184 *
1185 * CONTEXT:
1186 * Might sleep.
1187 *
1188 * RETURNS:
1189 * 0 on success, -errno on failure.
1190 */
1191int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1192{
1193 struct bd_holder_disk *holder;
1194 int ret = 0;
1195
1196 mutex_lock(&bdev->bd_mutex);
1197
1198 WARN_ON_ONCE(!bdev->bd_holder);
1199
1200 /* FIXME: remove the following once add_disk() handles errors */
1201 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1202 goto out_unlock;
1203
1204 holder = bd_find_holder_disk(bdev, disk);
1205 if (holder) {
1206 holder->refcnt++;
1207 goto out_unlock;
1208 }
1209
1210 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1211 if (!holder) {
1212 ret = -ENOMEM;
1213 goto out_unlock;
1214 }
1215
1216 INIT_LIST_HEAD(&holder->list);
1217 holder->disk = disk;
1218 holder->refcnt = 1;
1219
1220 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1221 if (ret)
1222 goto out_free;
1223
1224 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1225 if (ret)
1226 goto out_del;
1227 /*
1228 * bdev could be deleted beneath us which would implicitly destroy
1229 * the holder directory. Hold on to it.
1230 */
1231 kobject_get(bdev->bd_part->holder_dir);
1232
1233 list_add(&holder->list, &bdev->bd_holder_disks);
1234 goto out_unlock;
1235
1236out_del:
1237 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1238out_free:
1239 kfree(holder);
1240out_unlock:
1241 mutex_unlock(&bdev->bd_mutex);
1242 return ret;
1243}
1244EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1245
1246/**
1247 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1248 * @bdev: the calimed slave bdev
1249 * @disk: the holding disk
1250 *
1251 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1252 *
1253 * CONTEXT:
1254 * Might sleep.
1255 */
1256void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1257{
1258 struct bd_holder_disk *holder;
1259
1260 mutex_lock(&bdev->bd_mutex);
1261
1262 holder = bd_find_holder_disk(bdev, disk);
1263
1264 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1265 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1266 del_symlink(bdev->bd_part->holder_dir,
1267 &disk_to_dev(disk)->kobj);
1268 kobject_put(bdev->bd_part->holder_dir);
1269 list_del_init(&holder->list);
1270 kfree(holder);
1271 }
1272
1273 mutex_unlock(&bdev->bd_mutex);
1274}
1275EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1276#endif
1277
1278/**
1279 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1280 * @disk: struct gendisk to check
1281 * @bdev: struct bdev to adjust.
1282 * @verbose: if %true log a message about a size change if there is any
1283 *
1284 * This routine checks to see if the bdev size does not match the disk size
1285 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1286 * are freed.
1287 */
1288static void check_disk_size_change(struct gendisk *disk,
1289 struct block_device *bdev, bool verbose)
1290{
1291 loff_t disk_size, bdev_size;
1292
1293 disk_size = (loff_t)get_capacity(disk) << 9;
1294 bdev_size = i_size_read(bdev->bd_inode);
1295 if (disk_size != bdev_size) {
1296 if (verbose) {
1297 printk(KERN_INFO
1298 "%s: detected capacity change from %lld to %lld\n",
1299 disk->disk_name, bdev_size, disk_size);
1300 }
1301 i_size_write(bdev->bd_inode, disk_size);
1302 if (bdev_size > disk_size && __invalidate_device(bdev, false))
1303 pr_warn("VFS: busy inodes on resized disk %s\n",
1304 disk->disk_name);
1305 }
1306 bdev->bd_invalidated = 0;
1307}
1308
1309/**
1310 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1311 * @disk: struct gendisk to be revalidated
1312 *
1313 * This routine is a wrapper for lower-level driver's revalidate_disk
1314 * call-backs. It is used to do common pre and post operations needed
1315 * for all revalidate_disk operations.
1316 */
1317int revalidate_disk(struct gendisk *disk)
1318{
1319 int ret = 0;
1320
1321 if (disk->fops->revalidate_disk)
1322 ret = disk->fops->revalidate_disk(disk);
1323
1324 /*
1325 * Hidden disks don't have associated bdev so there's no point in
1326 * revalidating it.
1327 */
1328 if (!(disk->flags & GENHD_FL_HIDDEN)) {
1329 struct block_device *bdev = bdget_disk(disk, 0);
1330
1331 if (!bdev)
1332 return ret;
1333
1334 mutex_lock(&bdev->bd_mutex);
1335 check_disk_size_change(disk, bdev, ret == 0);
1336 mutex_unlock(&bdev->bd_mutex);
1337 bdput(bdev);
1338 }
1339 return ret;
1340}
1341EXPORT_SYMBOL(revalidate_disk);
1342
1343/*
1344 * This routine checks whether a removable media has been changed,
1345 * and invalidates all buffer-cache-entries in that case. This
1346 * is a relatively slow routine, so we have to try to minimize using
1347 * it. Thus it is called only upon a 'mount' or 'open'. This
1348 * is the best way of combining speed and utility, I think.
1349 * People changing diskettes in the middle of an operation deserve
1350 * to lose :-)
1351 */
1352int check_disk_change(struct block_device *bdev)
1353{
1354 struct gendisk *disk = bdev->bd_disk;
1355 const struct block_device_operations *bdops = disk->fops;
1356 unsigned int events;
1357
1358 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1359 DISK_EVENT_EJECT_REQUEST);
1360 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1361 return 0;
1362
1363 if (__invalidate_device(bdev, true))
1364 pr_warn("VFS: busy inodes on changed media %s\n",
1365 disk->disk_name);
1366 bdev->bd_invalidated = 1;
1367 if (bdops->revalidate_disk)
1368 bdops->revalidate_disk(bdev->bd_disk);
1369 return 1;
1370}
1371
1372EXPORT_SYMBOL(check_disk_change);
1373
1374void bd_set_size(struct block_device *bdev, loff_t size)
1375{
1376 inode_lock(bdev->bd_inode);
1377 i_size_write(bdev->bd_inode, size);
1378 inode_unlock(bdev->bd_inode);
1379}
1380EXPORT_SYMBOL(bd_set_size);
1381
1382static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1383
1384int bdev_disk_changed(struct block_device *bdev, bool invalidate)
1385{
1386 struct gendisk *disk = bdev->bd_disk;
1387 int ret;
1388
1389 lockdep_assert_held(&bdev->bd_mutex);
1390
1391rescan:
1392 ret = blk_drop_partitions(bdev);
1393 if (ret)
1394 return ret;
1395
1396 /*
1397 * Historically we only set the capacity to zero for devices that
1398 * support partitions (independ of actually having partitions created).
1399 * Doing that is rather inconsistent, but changing it broke legacy
1400 * udisks polling for legacy ide-cdrom devices. Use the crude check
1401 * below to get the sane behavior for most device while not breaking
1402 * userspace for this particular setup.
1403 */
1404 if (invalidate) {
1405 if (disk_part_scan_enabled(disk) ||
1406 !(disk->flags & GENHD_FL_REMOVABLE))
1407 set_capacity(disk, 0);
1408 } else {
1409 if (disk->fops->revalidate_disk)
1410 disk->fops->revalidate_disk(disk);
1411 }
1412
1413 check_disk_size_change(disk, bdev, !invalidate);
1414
1415 if (get_capacity(disk)) {
1416 ret = blk_add_partitions(disk, bdev);
1417 if (ret == -EAGAIN)
1418 goto rescan;
1419 } else if (invalidate) {
1420 /*
1421 * Tell userspace that the media / partition table may have
1422 * changed.
1423 */
1424 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
1425 }
1426
1427 return ret;
1428}
1429/*
1430 * Only exported for for loop and dasd for historic reasons. Don't use in new
1431 * code!
1432 */
1433EXPORT_SYMBOL_GPL(bdev_disk_changed);
1434
1435/*
1436 * bd_mutex locking:
1437 *
1438 * mutex_lock(part->bd_mutex)
1439 * mutex_lock_nested(whole->bd_mutex, 1)
1440 */
1441
1442static int __blkdev_get(struct block_device *bdev, fmode_t mode, void *holder,
1443 int for_part)
1444{
1445 struct block_device *whole = NULL, *claiming = NULL;
1446 struct gendisk *disk;
1447 int ret;
1448 int partno;
1449 int perm = 0;
1450 bool first_open = false, unblock_events = true, need_restart;
1451
1452 if (mode & FMODE_READ)
1453 perm |= MAY_READ;
1454 if (mode & FMODE_WRITE)
1455 perm |= MAY_WRITE;
1456 /*
1457 * hooks: /n/, see "layering violations".
1458 */
1459 if (!for_part) {
1460 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1461 if (ret != 0)
1462 return ret;
1463 }
1464
1465 restart:
1466 need_restart = false;
1467 ret = -ENXIO;
1468 disk = bdev_get_gendisk(bdev, &partno);
1469 if (!disk)
1470 goto out;
1471
1472 if (partno) {
1473 whole = bdget_disk(disk, 0);
1474 if (!whole) {
1475 ret = -ENOMEM;
1476 goto out_put_disk;
1477 }
1478 }
1479
1480 if (!for_part && (mode & FMODE_EXCL)) {
1481 WARN_ON_ONCE(!holder);
1482 if (whole)
1483 claiming = whole;
1484 else
1485 claiming = bdev;
1486 ret = bd_prepare_to_claim(bdev, claiming, holder);
1487 if (ret)
1488 goto out_put_whole;
1489 }
1490
1491 disk_block_events(disk);
1492 mutex_lock_nested(&bdev->bd_mutex, for_part);
1493 if (!bdev->bd_openers) {
1494 first_open = true;
1495 bdev->bd_disk = disk;
1496 bdev->bd_contains = bdev;
1497 bdev->bd_partno = partno;
1498
1499 if (!partno) {
1500 ret = -ENXIO;
1501 bdev->bd_part = disk_get_part(disk, partno);
1502 if (!bdev->bd_part)
1503 goto out_clear;
1504
1505 ret = 0;
1506 if (disk->fops->open) {
1507 ret = disk->fops->open(bdev, mode);
1508 /*
1509 * If we lost a race with 'disk' being deleted,
1510 * try again. See md.c
1511 */
1512 if (ret == -ERESTARTSYS)
1513 need_restart = true;
1514 }
1515
1516 if (!ret) {
1517 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1518 set_init_blocksize(bdev);
1519 }
1520
1521 /*
1522 * If the device is invalidated, rescan partition
1523 * if open succeeded or failed with -ENOMEDIUM.
1524 * The latter is necessary to prevent ghost
1525 * partitions on a removed medium.
1526 */
1527 if (bdev->bd_invalidated &&
1528 (!ret || ret == -ENOMEDIUM))
1529 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1530
1531 if (ret)
1532 goto out_clear;
1533 } else {
1534 BUG_ON(for_part);
1535 ret = __blkdev_get(whole, mode, NULL, 1);
1536 if (ret)
1537 goto out_clear;
1538 bdev->bd_contains = bdgrab(whole);
1539 bdev->bd_part = disk_get_part(disk, partno);
1540 if (!(disk->flags & GENHD_FL_UP) ||
1541 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1542 ret = -ENXIO;
1543 goto out_clear;
1544 }
1545 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1546 set_init_blocksize(bdev);
1547 }
1548
1549 if (bdev->bd_bdi == &noop_backing_dev_info)
1550 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1551 } else {
1552 if (bdev->bd_contains == bdev) {
1553 ret = 0;
1554 if (bdev->bd_disk->fops->open)
1555 ret = bdev->bd_disk->fops->open(bdev, mode);
1556 /* the same as first opener case, read comment there */
1557 if (bdev->bd_invalidated &&
1558 (!ret || ret == -ENOMEDIUM))
1559 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1560 if (ret)
1561 goto out_unlock_bdev;
1562 }
1563 }
1564 bdev->bd_openers++;
1565 if (for_part)
1566 bdev->bd_part_count++;
1567 if (claiming)
1568 bd_finish_claiming(bdev, claiming, holder);
1569
1570 /*
1571 * Block event polling for write claims if requested. Any write holder
1572 * makes the write_holder state stick until all are released. This is
1573 * good enough and tracking individual writeable reference is too
1574 * fragile given the way @mode is used in blkdev_get/put().
1575 */
1576 if (claiming && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1577 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1578 bdev->bd_write_holder = true;
1579 unblock_events = false;
1580 }
1581 mutex_unlock(&bdev->bd_mutex);
1582
1583 if (unblock_events)
1584 disk_unblock_events(disk);
1585
1586 /* only one opener holds refs to the module and disk */
1587 if (!first_open)
1588 put_disk_and_module(disk);
1589 if (whole)
1590 bdput(whole);
1591 return 0;
1592
1593 out_clear:
1594 disk_put_part(bdev->bd_part);
1595 bdev->bd_disk = NULL;
1596 bdev->bd_part = NULL;
1597 if (bdev != bdev->bd_contains)
1598 __blkdev_put(bdev->bd_contains, mode, 1);
1599 bdev->bd_contains = NULL;
1600 out_unlock_bdev:
1601 if (claiming)
1602 bd_abort_claiming(bdev, claiming, holder);
1603 mutex_unlock(&bdev->bd_mutex);
1604 disk_unblock_events(disk);
1605 out_put_whole:
1606 if (whole)
1607 bdput(whole);
1608 out_put_disk:
1609 put_disk_and_module(disk);
1610 if (need_restart)
1611 goto restart;
1612 out:
1613 return ret;
1614}
1615
1616/**
1617 * blkdev_get - open a block device
1618 * @bdev: block_device to open
1619 * @mode: FMODE_* mask
1620 * @holder: exclusive holder identifier
1621 *
1622 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1623 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1624 * @holder is invalid. Exclusive opens may nest for the same @holder.
1625 *
1626 * On success, the reference count of @bdev is unchanged. On failure,
1627 * @bdev is put.
1628 *
1629 * CONTEXT:
1630 * Might sleep.
1631 *
1632 * RETURNS:
1633 * 0 on success, -errno on failure.
1634 */
1635int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1636{
1637 int res;
1638
1639 res =__blkdev_get(bdev, mode, holder, 0);
1640 if (res)
1641 bdput(bdev);
1642 return res;
1643}
1644EXPORT_SYMBOL(blkdev_get);
1645
1646/**
1647 * blkdev_get_by_path - open a block device by name
1648 * @path: path to the block device to open
1649 * @mode: FMODE_* mask
1650 * @holder: exclusive holder identifier
1651 *
1652 * Open the blockdevice described by the device file at @path. @mode
1653 * and @holder are identical to blkdev_get().
1654 *
1655 * On success, the returned block_device has reference count of one.
1656 *
1657 * CONTEXT:
1658 * Might sleep.
1659 *
1660 * RETURNS:
1661 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1662 */
1663struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1664 void *holder)
1665{
1666 struct block_device *bdev;
1667 int err;
1668
1669 bdev = lookup_bdev(path);
1670 if (IS_ERR(bdev))
1671 return bdev;
1672
1673 err = blkdev_get(bdev, mode, holder);
1674 if (err)
1675 return ERR_PTR(err);
1676
1677 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1678 blkdev_put(bdev, mode);
1679 return ERR_PTR(-EACCES);
1680 }
1681
1682 return bdev;
1683}
1684EXPORT_SYMBOL(blkdev_get_by_path);
1685
1686/**
1687 * blkdev_get_by_dev - open a block device by device number
1688 * @dev: device number of block device to open
1689 * @mode: FMODE_* mask
1690 * @holder: exclusive holder identifier
1691 *
1692 * Open the blockdevice described by device number @dev. @mode and
1693 * @holder are identical to blkdev_get().
1694 *
1695 * Use it ONLY if you really do not have anything better - i.e. when
1696 * you are behind a truly sucky interface and all you are given is a
1697 * device number. _Never_ to be used for internal purposes. If you
1698 * ever need it - reconsider your API.
1699 *
1700 * On success, the returned block_device has reference count of one.
1701 *
1702 * CONTEXT:
1703 * Might sleep.
1704 *
1705 * RETURNS:
1706 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1707 */
1708struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1709{
1710 struct block_device *bdev;
1711 int err;
1712
1713 bdev = bdget(dev);
1714 if (!bdev)
1715 return ERR_PTR(-ENOMEM);
1716
1717 err = blkdev_get(bdev, mode, holder);
1718 if (err)
1719 return ERR_PTR(err);
1720
1721 return bdev;
1722}
1723EXPORT_SYMBOL(blkdev_get_by_dev);
1724
1725static int blkdev_open(struct inode * inode, struct file * filp)
1726{
1727 struct block_device *bdev;
1728
1729 /*
1730 * Preserve backwards compatibility and allow large file access
1731 * even if userspace doesn't ask for it explicitly. Some mkfs
1732 * binary needs it. We might want to drop this workaround
1733 * during an unstable branch.
1734 */
1735 filp->f_flags |= O_LARGEFILE;
1736
1737 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
1738
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;
1745
1746 bdev = bd_acquire(inode);
1747 if (bdev == NULL)
1748 return -ENOMEM;
1749
1750 filp->f_mapping = bdev->bd_inode->i_mapping;
1751 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1752
1753 return blkdev_get(bdev, filp->f_mode, filp);
1754}
1755
1756static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1757{
1758 struct gendisk *disk = bdev->bd_disk;
1759 struct block_device *victim = NULL;
1760
1761 /*
1762 * Sync early if it looks like we're the last one. If someone else
1763 * opens the block device between now and the decrement of bd_openers
1764 * then we did a sync that we didn't need to, but that's not the end
1765 * of the world and we want to avoid long (could be several minute)
1766 * syncs while holding the mutex.
1767 */
1768 if (bdev->bd_openers == 1)
1769 sync_blockdev(bdev);
1770
1771 mutex_lock_nested(&bdev->bd_mutex, for_part);
1772 if (for_part)
1773 bdev->bd_part_count--;
1774
1775 if (!--bdev->bd_openers) {
1776 WARN_ON_ONCE(bdev->bd_holders);
1777 sync_blockdev(bdev);
1778 kill_bdev(bdev);
1779
1780 bdev_write_inode(bdev);
1781 }
1782 if (bdev->bd_contains == bdev) {
1783 if (disk->fops->release)
1784 disk->fops->release(disk, mode);
1785 }
1786 if (!bdev->bd_openers) {
1787 disk_put_part(bdev->bd_part);
1788 bdev->bd_part = NULL;
1789 bdev->bd_disk = NULL;
1790 if (bdev != bdev->bd_contains)
1791 victim = bdev->bd_contains;
1792 bdev->bd_contains = NULL;
1793
1794 put_disk_and_module(disk);
1795 }
1796 mutex_unlock(&bdev->bd_mutex);
1797 bdput(bdev);
1798 if (victim)
1799 __blkdev_put(victim, mode, 1);
1800}
1801
1802void blkdev_put(struct block_device *bdev, fmode_t mode)
1803{
1804 mutex_lock(&bdev->bd_mutex);
1805
1806 if (mode & FMODE_EXCL) {
1807 bool bdev_free;
1808
1809 /*
1810 * Release a claim on the device. The holder fields
1811 * are protected with bdev_lock. bd_mutex is to
1812 * synchronize disk_holder unlinking.
1813 */
1814 spin_lock(&bdev_lock);
1815
1816 WARN_ON_ONCE(--bdev->bd_holders < 0);
1817 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1818
1819 /* bd_contains might point to self, check in a separate step */
1820 if ((bdev_free = !bdev->bd_holders))
1821 bdev->bd_holder = NULL;
1822 if (!bdev->bd_contains->bd_holders)
1823 bdev->bd_contains->bd_holder = NULL;
1824
1825 spin_unlock(&bdev_lock);
1826
1827 /*
1828 * If this was the last claim, remove holder link and
1829 * unblock evpoll if it was a write holder.
1830 */
1831 if (bdev_free && bdev->bd_write_holder) {
1832 disk_unblock_events(bdev->bd_disk);
1833 bdev->bd_write_holder = false;
1834 }
1835 }
1836
1837 /*
1838 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1839 * event. This is to ensure detection of media removal commanded
1840 * from userland - e.g. eject(1).
1841 */
1842 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1843
1844 mutex_unlock(&bdev->bd_mutex);
1845
1846 __blkdev_put(bdev, mode, 0);
1847}
1848EXPORT_SYMBOL(blkdev_put);
1849
1850static int blkdev_close(struct inode * inode, struct file * filp)
1851{
1852 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1853 blkdev_put(bdev, filp->f_mode);
1854 return 0;
1855}
1856
1857static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1858{
1859 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1860 fmode_t mode = file->f_mode;
1861
1862 /*
1863 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1864 * to updated it before every ioctl.
1865 */
1866 if (file->f_flags & O_NDELAY)
1867 mode |= FMODE_NDELAY;
1868 else
1869 mode &= ~FMODE_NDELAY;
1870
1871 return blkdev_ioctl(bdev, mode, cmd, arg);
1872}
1873
1874/*
1875 * Write data to the block device. Only intended for the block device itself
1876 * and the raw driver which basically is a fake block device.
1877 *
1878 * Does not take i_mutex for the write and thus is not for general purpose
1879 * use.
1880 */
1881ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1882{
1883 struct file *file = iocb->ki_filp;
1884 struct inode *bd_inode = bdev_file_inode(file);
1885 loff_t size = i_size_read(bd_inode);
1886 struct blk_plug plug;
1887 ssize_t ret;
1888
1889 if (bdev_read_only(I_BDEV(bd_inode)))
1890 return -EPERM;
1891
1892 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode))
1893 return -ETXTBSY;
1894
1895 if (!iov_iter_count(from))
1896 return 0;
1897
1898 if (iocb->ki_pos >= size)
1899 return -ENOSPC;
1900
1901 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1902 return -EOPNOTSUPP;
1903
1904 iov_iter_truncate(from, size - iocb->ki_pos);
1905
1906 blk_start_plug(&plug);
1907 ret = __generic_file_write_iter(iocb, from);
1908 if (ret > 0)
1909 ret = generic_write_sync(iocb, ret);
1910 blk_finish_plug(&plug);
1911 return ret;
1912}
1913EXPORT_SYMBOL_GPL(blkdev_write_iter);
1914
1915ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1916{
1917 struct file *file = iocb->ki_filp;
1918 struct inode *bd_inode = bdev_file_inode(file);
1919 loff_t size = i_size_read(bd_inode);
1920 loff_t pos = iocb->ki_pos;
1921
1922 if (pos >= size)
1923 return 0;
1924
1925 size -= pos;
1926 iov_iter_truncate(to, size);
1927 return generic_file_read_iter(iocb, to);
1928}
1929EXPORT_SYMBOL_GPL(blkdev_read_iter);
1930
1931/*
1932 * Try to release a page associated with block device when the system
1933 * is under memory pressure.
1934 */
1935static int blkdev_releasepage(struct page *page, gfp_t wait)
1936{
1937 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1938
1939 if (super && super->s_op->bdev_try_to_free_page)
1940 return super->s_op->bdev_try_to_free_page(super, page, wait);
1941
1942 return try_to_free_buffers(page);
1943}
1944
1945static int blkdev_writepages(struct address_space *mapping,
1946 struct writeback_control *wbc)
1947{
1948 return generic_writepages(mapping, wbc);
1949}
1950
1951static const struct address_space_operations def_blk_aops = {
1952 .readpage = blkdev_readpage,
1953 .readahead = blkdev_readahead,
1954 .writepage = blkdev_writepage,
1955 .write_begin = blkdev_write_begin,
1956 .write_end = blkdev_write_end,
1957 .writepages = blkdev_writepages,
1958 .releasepage = blkdev_releasepage,
1959 .direct_IO = blkdev_direct_IO,
1960 .migratepage = buffer_migrate_page_norefs,
1961 .is_dirty_writeback = buffer_check_dirty_writeback,
1962};
1963
1964#define BLKDEV_FALLOC_FL_SUPPORTED \
1965 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1966 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1967
1968static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1969 loff_t len)
1970{
1971 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1972 struct address_space *mapping;
1973 loff_t end = start + len - 1;
1974 loff_t isize;
1975 int error;
1976
1977 /* Fail if we don't recognize the flags. */
1978 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1979 return -EOPNOTSUPP;
1980
1981 /* Don't go off the end of the device. */
1982 isize = i_size_read(bdev->bd_inode);
1983 if (start >= isize)
1984 return -EINVAL;
1985 if (end >= isize) {
1986 if (mode & FALLOC_FL_KEEP_SIZE) {
1987 len = isize - start;
1988 end = start + len - 1;
1989 } else
1990 return -EINVAL;
1991 }
1992
1993 /*
1994 * Don't allow IO that isn't aligned to logical block size.
1995 */
1996 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1997 return -EINVAL;
1998
1999 /* Invalidate the page cache, including dirty pages. */
2000 mapping = bdev->bd_inode->i_mapping;
2001 truncate_inode_pages_range(mapping, start, end);
2002
2003 switch (mode) {
2004 case FALLOC_FL_ZERO_RANGE:
2005 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2006 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2007 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2008 break;
2009 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2010 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2011 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2012 break;
2013 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2014 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2015 GFP_KERNEL, 0);
2016 break;
2017 default:
2018 return -EOPNOTSUPP;
2019 }
2020 if (error)
2021 return error;
2022
2023 /*
2024 * Invalidate again; if someone wandered in and dirtied a page,
2025 * the caller will be given -EBUSY. The third argument is
2026 * inclusive, so the rounding here is safe.
2027 */
2028 return invalidate_inode_pages2_range(mapping,
2029 start >> PAGE_SHIFT,
2030 end >> PAGE_SHIFT);
2031}
2032
2033const struct file_operations def_blk_fops = {
2034 .open = blkdev_open,
2035 .release = blkdev_close,
2036 .llseek = block_llseek,
2037 .read_iter = blkdev_read_iter,
2038 .write_iter = blkdev_write_iter,
2039 .iopoll = blkdev_iopoll,
2040 .mmap = generic_file_mmap,
2041 .fsync = blkdev_fsync,
2042 .unlocked_ioctl = block_ioctl,
2043#ifdef CONFIG_COMPAT
2044 .compat_ioctl = compat_blkdev_ioctl,
2045#endif
2046 .splice_read = generic_file_splice_read,
2047 .splice_write = iter_file_splice_write,
2048 .fallocate = blkdev_fallocate,
2049};
2050
2051/**
2052 * lookup_bdev - lookup a struct block_device by name
2053 * @pathname: special file representing the block device
2054 *
2055 * Get a reference to the blockdevice at @pathname in the current
2056 * namespace if possible and return it. Return ERR_PTR(error)
2057 * otherwise.
2058 */
2059struct block_device *lookup_bdev(const char *pathname)
2060{
2061 struct block_device *bdev;
2062 struct inode *inode;
2063 struct path path;
2064 int error;
2065
2066 if (!pathname || !*pathname)
2067 return ERR_PTR(-EINVAL);
2068
2069 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2070 if (error)
2071 return ERR_PTR(error);
2072
2073 inode = d_backing_inode(path.dentry);
2074 error = -ENOTBLK;
2075 if (!S_ISBLK(inode->i_mode))
2076 goto fail;
2077 error = -EACCES;
2078 if (!may_open_dev(&path))
2079 goto fail;
2080 error = -ENOMEM;
2081 bdev = bd_acquire(inode);
2082 if (!bdev)
2083 goto fail;
2084out:
2085 path_put(&path);
2086 return bdev;
2087fail:
2088 bdev = ERR_PTR(error);
2089 goto out;
2090}
2091EXPORT_SYMBOL(lookup_bdev);
2092
2093int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2094{
2095 struct super_block *sb = get_super(bdev);
2096 int res = 0;
2097
2098 if (sb) {
2099 /*
2100 * no need to lock the super, get_super holds the
2101 * read mutex so the filesystem cannot go away
2102 * under us (->put_super runs with the write lock
2103 * hold).
2104 */
2105 shrink_dcache_sb(sb);
2106 res = invalidate_inodes(sb, kill_dirty);
2107 drop_super(sb);
2108 }
2109 invalidate_bdev(bdev);
2110 return res;
2111}
2112EXPORT_SYMBOL(__invalidate_device);
2113
2114void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2115{
2116 struct inode *inode, *old_inode = NULL;
2117
2118 spin_lock(&blockdev_superblock->s_inode_list_lock);
2119 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2120 struct address_space *mapping = inode->i_mapping;
2121 struct block_device *bdev;
2122
2123 spin_lock(&inode->i_lock);
2124 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2125 mapping->nrpages == 0) {
2126 spin_unlock(&inode->i_lock);
2127 continue;
2128 }
2129 __iget(inode);
2130 spin_unlock(&inode->i_lock);
2131 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2132 /*
2133 * We hold a reference to 'inode' so it couldn't have been
2134 * removed from s_inodes list while we dropped the
2135 * s_inode_list_lock We cannot iput the inode now as we can
2136 * be holding the last reference and we cannot iput it under
2137 * s_inode_list_lock. So we keep the reference and iput it
2138 * later.
2139 */
2140 iput(old_inode);
2141 old_inode = inode;
2142 bdev = I_BDEV(inode);
2143
2144 mutex_lock(&bdev->bd_mutex);
2145 if (bdev->bd_openers)
2146 func(bdev, arg);
2147 mutex_unlock(&bdev->bd_mutex);
2148
2149 spin_lock(&blockdev_superblock->s_inode_list_lock);
2150 }
2151 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2152 iput(old_inode);
2153}