projects/linux·blob v4.14 · refs

2166 lines · 53.4 KiB

1/*
2 * linux/fs/block_dev.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <[email protected]> SuSE
6 */
7
8#include <linux/init.h>
9#include <linux/mm.h>
10#include <linux/fcntl.h>
11#include <linux/slab.h>
12#include <linux/kmod.h>
13#include <linux/major.h>
14#include <linux/device_cgroup.h>
15#include <linux/highmem.h>
16#include <linux/blkdev.h>
17#include <linux/backing-dev.h>
18#include <linux/module.h>
19#include <linux/blkpg.h>
20#include <linux/magic.h>
21#include <linux/dax.h>
22#include <linux/buffer_head.h>
23#include <linux/swap.h>
24#include <linux/pagevec.h>
25#include <linux/writeback.h>
26#include <linux/mpage.h>
27#include <linux/mount.h>
28#include <linux/uio.h>
29#include <linux/namei.h>
30#include <linux/log2.h>
31#include <linux/cleancache.h>
32#include <linux/dax.h>
33#include <linux/badblocks.h>
34#include <linux/task_io_accounting_ops.h>
35#include <linux/falloc.h>
36#include <linux/uaccess.h>
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
57void __vfs_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
58{
59 struct va_format vaf;
60 va_list args;
61
62 va_start(args, fmt);
63 vaf.fmt = fmt;
64 vaf.va = &args;
65 printk_ratelimited("%sVFS (%s): %pV\n", prefix, sb->s_id, &vaf);
66 va_end(args);
67}
68
69static void bdev_write_inode(struct block_device *bdev)
70{
71 struct inode *inode = bdev->bd_inode;
72 int ret;
73
74 spin_lock(&inode->i_lock);
75 while (inode->i_state & I_DIRTY) {
76 spin_unlock(&inode->i_lock);
77 ret = write_inode_now(inode, true);
78 if (ret) {
79 char name[BDEVNAME_SIZE];
80 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
81 "for block device %s (err=%d).\n",
82 bdevname(bdev, name), ret);
83 }
84 spin_lock(&inode->i_lock);
85 }
86 spin_unlock(&inode->i_lock);
87}
88
89/* Kill _all_ buffers and pagecache , dirty or not.. */
90void kill_bdev(struct block_device *bdev)
91{
92 struct address_space *mapping = bdev->bd_inode->i_mapping;
93
94 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
95 return;
96
97 invalidate_bh_lrus();
98 truncate_inode_pages(mapping, 0);
99}
100EXPORT_SYMBOL(kill_bdev);
101
102/* Invalidate clean unused buffers and pagecache. */
103void invalidate_bdev(struct block_device *bdev)
104{
105 struct address_space *mapping = bdev->bd_inode->i_mapping;
106
107 if (mapping->nrpages) {
108 invalidate_bh_lrus();
109 lru_add_drain_all(); /* make sure all lru add caches are flushed */
110 invalidate_mapping_pages(mapping, 0, -1);
111 }
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
114 */
115 cleancache_invalidate_inode(mapping);
116}
117EXPORT_SYMBOL(invalidate_bdev);
118
119int set_blocksize(struct block_device *bdev, int size)
120{
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))
123 return -EINVAL;
124
125 /* Size cannot be smaller than the size supported by the device */
126 if (size < bdev_logical_block_size(bdev))
127 return -EINVAL;
128
129 /* Don't change the size if it is same as current */
130 if (bdev->bd_block_size != size) {
131 sync_blockdev(bdev);
132 bdev->bd_block_size = size;
133 bdev->bd_inode->i_blkbits = blksize_bits(size);
134 kill_bdev(bdev);
135 }
136 return 0;
137}
138
139EXPORT_SYMBOL(set_blocksize);
140
141int sb_set_blocksize(struct super_block *sb, int size)
142{
143 if (set_blocksize(sb->s_bdev, size))
144 return 0;
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;
150}
151
152EXPORT_SYMBOL(sb_set_blocksize);
153
154int sb_min_blocksize(struct super_block *sb, int size)
155{
156 int minsize = bdev_logical_block_size(sb->s_bdev);
157 if (size < minsize)
158 size = minsize;
159 return sb_set_blocksize(sb, size);
160}
161
162EXPORT_SYMBOL(sb_min_blocksize);
163
164static int
165blkdev_get_block(struct inode *inode, sector_t iblock,
166 struct buffer_head *bh, int create)
167{
168 bh->b_bdev = I_BDEV(inode);
169 bh->b_blocknr = iblock;
170 set_buffer_mapped(bh);
171 return 0;
172}
173
174static struct inode *bdev_file_inode(struct file *file)
175{
176 return file->f_mapping->host;
177}
178
179static unsigned int dio_bio_write_op(struct kiocb *iocb)
180{
181 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
182
183 /* avoid the need for a I/O completion work item */
184 if (iocb->ki_flags & IOCB_DSYNC)
185 op |= REQ_FUA;
186 return op;
187}
188
189#define DIO_INLINE_BIO_VECS 4
190
191static void blkdev_bio_end_io_simple(struct bio *bio)
192{
193 struct task_struct *waiter = bio->bi_private;
194
195 WRITE_ONCE(bio->bi_private, NULL);
196 wake_up_process(waiter);
197}
198
199static ssize_t
200__blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
201 int nr_pages)
202{
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;
208 struct bio bio;
209 ssize_t ret;
210 blk_qc_t qc;
211 int i;
212
213 if ((pos | iov_iter_alignment(iter)) &
214 (bdev_logical_block_size(bdev) - 1))
215 return -EINVAL;
216
217 if (nr_pages <= DIO_INLINE_BIO_VECS)
218 vecs = inline_vecs;
219 else {
220 vecs = kmalloc(nr_pages * sizeof(struct bio_vec), GFP_KERNEL);
221 if (!vecs)
222 return -ENOMEM;
223 }
224
225 bio_init(&bio, vecs, nr_pages);
226 bio_set_dev(&bio, bdev);
227 bio.bi_iter.bi_sector = pos >> 9;
228 bio.bi_write_hint = iocb->ki_hint;
229 bio.bi_private = current;
230 bio.bi_end_io = blkdev_bio_end_io_simple;
231
232 ret = bio_iov_iter_get_pages(&bio, iter);
233 if (unlikely(ret))
234 return ret;
235 ret = bio.bi_iter.bi_size;
236
237 if (iov_iter_rw(iter) == READ) {
238 bio.bi_opf = REQ_OP_READ;
239 if (iter_is_iovec(iter))
240 should_dirty = true;
241 } else {
242 bio.bi_opf = dio_bio_write_op(iocb);
243 task_io_account_write(ret);
244 }
245
246 qc = submit_bio(&bio);
247 for (;;) {
248 set_current_state(TASK_UNINTERRUPTIBLE);
249 if (!READ_ONCE(bio.bi_private))
250 break;
251 if (!(iocb->ki_flags & IOCB_HIPRI) ||
252 !blk_mq_poll(bdev_get_queue(bdev), qc))
253 io_schedule();
254 }
255 __set_current_state(TASK_RUNNING);
256
257 bio_for_each_segment_all(bvec, &bio, i) {
258 if (should_dirty && !PageCompound(bvec->bv_page))
259 set_page_dirty_lock(bvec->bv_page);
260 put_page(bvec->bv_page);
261 }
262
263 if (vecs != inline_vecs)
264 kfree(vecs);
265
266 if (unlikely(bio.bi_status))
267 ret = blk_status_to_errno(bio.bi_status);
268
269 bio_uninit(&bio);
270
271 return ret;
272}
273
274struct blkdev_dio {
275 union {
276 struct kiocb *iocb;
277 struct task_struct *waiter;
278 };
279 size_t size;
280 atomic_t ref;
281 bool multi_bio : 1;
282 bool should_dirty : 1;
283 bool is_sync : 1;
284 struct bio bio;
285};
286
287static struct bio_set *blkdev_dio_pool __read_mostly;
288
289static void blkdev_bio_end_io(struct bio *bio)
290{
291 struct blkdev_dio *dio = bio->bi_private;
292 bool should_dirty = dio->should_dirty;
293
294 if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) {
295 if (bio->bi_status && !dio->bio.bi_status)
296 dio->bio.bi_status = bio->bi_status;
297 } else {
298 if (!dio->is_sync) {
299 struct kiocb *iocb = dio->iocb;
300 ssize_t ret;
301
302 if (likely(!dio->bio.bi_status)) {
303 ret = dio->size;
304 iocb->ki_pos += ret;
305 } else {
306 ret = blk_status_to_errno(dio->bio.bi_status);
307 }
308
309 dio->iocb->ki_complete(iocb, ret, 0);
310 bio_put(&dio->bio);
311 } else {
312 struct task_struct *waiter = dio->waiter;
313
314 WRITE_ONCE(dio->waiter, NULL);
315 wake_up_process(waiter);
316 }
317 }
318
319 if (should_dirty) {
320 bio_check_pages_dirty(bio);
321 } else {
322 struct bio_vec *bvec;
323 int i;
324
325 bio_for_each_segment_all(bvec, bio, i)
326 put_page(bvec->bv_page);
327 bio_put(bio);
328 }
329}
330
331static ssize_t
332__blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
333{
334 struct file *file = iocb->ki_filp;
335 struct inode *inode = bdev_file_inode(file);
336 struct block_device *bdev = I_BDEV(inode);
337 struct blk_plug plug;
338 struct blkdev_dio *dio;
339 struct bio *bio;
340 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
341 loff_t pos = iocb->ki_pos;
342 blk_qc_t qc = BLK_QC_T_NONE;
343 int ret = 0;
344
345 if ((pos | iov_iter_alignment(iter)) &
346 (bdev_logical_block_size(bdev) - 1))
347 return -EINVAL;
348
349 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, blkdev_dio_pool);
350 bio_get(bio); /* extra ref for the completion handler */
351
352 dio = container_of(bio, struct blkdev_dio, bio);
353 dio->is_sync = is_sync = is_sync_kiocb(iocb);
354 if (dio->is_sync)
355 dio->waiter = current;
356 else
357 dio->iocb = iocb;
358
359 dio->size = 0;
360 dio->multi_bio = false;
361 dio->should_dirty = is_read && (iter->type == ITER_IOVEC);
362
363 blk_start_plug(&plug);
364 for (;;) {
365 bio_set_dev(bio, bdev);
366 bio->bi_iter.bi_sector = pos >> 9;
367 bio->bi_write_hint = iocb->ki_hint;
368 bio->bi_private = dio;
369 bio->bi_end_io = blkdev_bio_end_io;
370
371 ret = bio_iov_iter_get_pages(bio, iter);
372 if (unlikely(ret)) {
373 bio->bi_status = BLK_STS_IOERR;
374 bio_endio(bio);
375 break;
376 }
377
378 if (is_read) {
379 bio->bi_opf = REQ_OP_READ;
380 if (dio->should_dirty)
381 bio_set_pages_dirty(bio);
382 } else {
383 bio->bi_opf = dio_bio_write_op(iocb);
384 task_io_account_write(bio->bi_iter.bi_size);
385 }
386
387 dio->size += bio->bi_iter.bi_size;
388 pos += bio->bi_iter.bi_size;
389
390 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
391 if (!nr_pages) {
392 qc = submit_bio(bio);
393 break;
394 }
395
396 if (!dio->multi_bio) {
397 dio->multi_bio = true;
398 atomic_set(&dio->ref, 2);
399 } else {
400 atomic_inc(&dio->ref);
401 }
402
403 submit_bio(bio);
404 bio = bio_alloc(GFP_KERNEL, nr_pages);
405 }
406 blk_finish_plug(&plug);
407
408 if (!is_sync)
409 return -EIOCBQUEUED;
410
411 for (;;) {
412 set_current_state(TASK_UNINTERRUPTIBLE);
413 if (!READ_ONCE(dio->waiter))
414 break;
415
416 if (!(iocb->ki_flags & IOCB_HIPRI) ||
417 !blk_mq_poll(bdev_get_queue(bdev), qc))
418 io_schedule();
419 }
420 __set_current_state(TASK_RUNNING);
421
422 if (!ret)
423 ret = blk_status_to_errno(dio->bio.bi_status);
424 if (likely(!ret))
425 ret = dio->size;
426
427 bio_put(&dio->bio);
428 return ret;
429}
430
431static ssize_t
432blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
433{
434 int nr_pages;
435
436 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
437 if (!nr_pages)
438 return 0;
439 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
440 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
441
442 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
443}
444
445static __init int blkdev_init(void)
446{
447 blkdev_dio_pool = bioset_create(4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
448 if (!blkdev_dio_pool)
449 return -ENOMEM;
450 return 0;
451}
452module_init(blkdev_init);
453
454int __sync_blockdev(struct block_device *bdev, int wait)
455{
456 if (!bdev)
457 return 0;
458 if (!wait)
459 return filemap_flush(bdev->bd_inode->i_mapping);
460 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
461}
462
463/*
464 * Write out and wait upon all the dirty data associated with a block
465 * device via its mapping. Does not take the superblock lock.
466 */
467int sync_blockdev(struct block_device *bdev)
468{
469 return __sync_blockdev(bdev, 1);
470}
471EXPORT_SYMBOL(sync_blockdev);
472
473/*
474 * Write out and wait upon all dirty data associated with this
475 * device. Filesystem data as well as the underlying block
476 * device. Takes the superblock lock.
477 */
478int fsync_bdev(struct block_device *bdev)
479{
480 struct super_block *sb = get_super(bdev);
481 if (sb) {
482 int res = sync_filesystem(sb);
483 drop_super(sb);
484 return res;
485 }
486 return sync_blockdev(bdev);
487}
488EXPORT_SYMBOL(fsync_bdev);
489
490/**
491 * freeze_bdev -- lock a filesystem and force it into a consistent state
492 * @bdev: blockdevice to lock
493 *
494 * If a superblock is found on this device, we take the s_umount semaphore
495 * on it to make sure nobody unmounts until the snapshot creation is done.
496 * The reference counter (bd_fsfreeze_count) guarantees that only the last
497 * unfreeze process can unfreeze the frozen filesystem actually when multiple
498 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
499 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
500 * actually.
501 */
502struct super_block *freeze_bdev(struct block_device *bdev)
503{
504 struct super_block *sb;
505 int error = 0;
506
507 mutex_lock(&bdev->bd_fsfreeze_mutex);
508 if (++bdev->bd_fsfreeze_count > 1) {
509 /*
510 * We don't even need to grab a reference - the first call
511 * to freeze_bdev grab an active reference and only the last
512 * thaw_bdev drops it.
513 */
514 sb = get_super(bdev);
515 if (sb)
516 drop_super(sb);
517 mutex_unlock(&bdev->bd_fsfreeze_mutex);
518 return sb;
519 }
520
521 sb = get_active_super(bdev);
522 if (!sb)
523 goto out;
524 if (sb->s_op->freeze_super)
525 error = sb->s_op->freeze_super(sb);
526 else
527 error = freeze_super(sb);
528 if (error) {
529 deactivate_super(sb);
530 bdev->bd_fsfreeze_count--;
531 mutex_unlock(&bdev->bd_fsfreeze_mutex);
532 return ERR_PTR(error);
533 }
534 deactivate_super(sb);
535 out:
536 sync_blockdev(bdev);
537 mutex_unlock(&bdev->bd_fsfreeze_mutex);
538 return sb; /* thaw_bdev releases s->s_umount */
539}
540EXPORT_SYMBOL(freeze_bdev);
541
542/**
543 * thaw_bdev -- unlock filesystem
544 * @bdev: blockdevice to unlock
545 * @sb: associated superblock
546 *
547 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
548 */
549int thaw_bdev(struct block_device *bdev, struct super_block *sb)
550{
551 int error = -EINVAL;
552
553 mutex_lock(&bdev->bd_fsfreeze_mutex);
554 if (!bdev->bd_fsfreeze_count)
555 goto out;
556
557 error = 0;
558 if (--bdev->bd_fsfreeze_count > 0)
559 goto out;
560
561 if (!sb)
562 goto out;
563
564 if (sb->s_op->thaw_super)
565 error = sb->s_op->thaw_super(sb);
566 else
567 error = thaw_super(sb);
568 if (error)
569 bdev->bd_fsfreeze_count++;
570out:
571 mutex_unlock(&bdev->bd_fsfreeze_mutex);
572 return error;
573}
574EXPORT_SYMBOL(thaw_bdev);
575
576static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
577{
578 return block_write_full_page(page, blkdev_get_block, wbc);
579}
580
581static int blkdev_readpage(struct file * file, struct page * page)
582{
583 return block_read_full_page(page, blkdev_get_block);
584}
585
586static int blkdev_readpages(struct file *file, struct address_space *mapping,
587 struct list_head *pages, unsigned nr_pages)
588{
589 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
590}
591
592static int blkdev_write_begin(struct file *file, struct address_space *mapping,
593 loff_t pos, unsigned len, unsigned flags,
594 struct page **pagep, void **fsdata)
595{
596 return block_write_begin(mapping, pos, len, flags, pagep,
597 blkdev_get_block);
598}
599
600static int blkdev_write_end(struct file *file, struct address_space *mapping,
601 loff_t pos, unsigned len, unsigned copied,
602 struct page *page, void *fsdata)
603{
604 int ret;
605 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
606
607 unlock_page(page);
608 put_page(page);
609
610 return ret;
611}
612
613/*
614 * private llseek:
615 * for a block special file file_inode(file)->i_size is zero
616 * so we compute the size by hand (just as in block_read/write above)
617 */
618static loff_t block_llseek(struct file *file, loff_t offset, int whence)
619{
620 struct inode *bd_inode = bdev_file_inode(file);
621 loff_t retval;
622
623 inode_lock(bd_inode);
624 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
625 inode_unlock(bd_inode);
626 return retval;
627}
628
629int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
630{
631 struct inode *bd_inode = bdev_file_inode(filp);
632 struct block_device *bdev = I_BDEV(bd_inode);
633 int error;
634
635 error = file_write_and_wait_range(filp, start, end);
636 if (error)
637 return error;
638
639 /*
640 * There is no need to serialise calls to blkdev_issue_flush with
641 * i_mutex and doing so causes performance issues with concurrent
642 * O_SYNC writers to a block device.
643 */
644 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
645 if (error == -EOPNOTSUPP)
646 error = 0;
647
648 return error;
649}
650EXPORT_SYMBOL(blkdev_fsync);
651
652/**
653 * bdev_read_page() - Start reading a page from a block device
654 * @bdev: The device to read the page from
655 * @sector: The offset on the device to read the page to (need not be aligned)
656 * @page: The page to read
657 *
658 * On entry, the page should be locked. It will be unlocked when the page
659 * has been read. If the block driver implements rw_page synchronously,
660 * that will be true on exit from this function, but it need not be.
661 *
662 * Errors returned by this function are usually "soft", eg out of memory, or
663 * queue full; callers should try a different route to read this page rather
664 * than propagate an error back up the stack.
665 *
666 * Return: negative errno if an error occurs, 0 if submission was successful.
667 */
668int bdev_read_page(struct block_device *bdev, sector_t sector,
669 struct page *page)
670{
671 const struct block_device_operations *ops = bdev->bd_disk->fops;
672 int result = -EOPNOTSUPP;
673
674 if (!ops->rw_page || bdev_get_integrity(bdev))
675 return result;
676
677 result = blk_queue_enter(bdev->bd_queue, false);
678 if (result)
679 return result;
680 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false);
681 blk_queue_exit(bdev->bd_queue);
682 return result;
683}
684EXPORT_SYMBOL_GPL(bdev_read_page);
685
686/**
687 * bdev_write_page() - Start writing a page to a block device
688 * @bdev: The device to write the page to
689 * @sector: The offset on the device to write the page to (need not be aligned)
690 * @page: The page to write
691 * @wbc: The writeback_control for the write
692 *
693 * On entry, the page should be locked and not currently under writeback.
694 * On exit, if the write started successfully, the page will be unlocked and
695 * under writeback. If the write failed already (eg the driver failed to
696 * queue the page to the device), the page will still be locked. If the
697 * caller is a ->writepage implementation, it will need to unlock the page.
698 *
699 * Errors returned by this function are usually "soft", eg out of memory, or
700 * queue full; callers should try a different route to write this page rather
701 * than propagate an error back up the stack.
702 *
703 * Return: negative errno if an error occurs, 0 if submission was successful.
704 */
705int bdev_write_page(struct block_device *bdev, sector_t sector,
706 struct page *page, struct writeback_control *wbc)
707{
708 int result;
709 const struct block_device_operations *ops = bdev->bd_disk->fops;
710
711 if (!ops->rw_page || bdev_get_integrity(bdev))
712 return -EOPNOTSUPP;
713 result = blk_queue_enter(bdev->bd_queue, false);
714 if (result)
715 return result;
716
717 set_page_writeback(page);
718 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true);
719 if (result) {
720 end_page_writeback(page);
721 } else {
722 clean_page_buffers(page);
723 unlock_page(page);
724 }
725 blk_queue_exit(bdev->bd_queue);
726 return result;
727}
728EXPORT_SYMBOL_GPL(bdev_write_page);
729
730/*
731 * pseudo-fs
732 */
733
734static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
735static struct kmem_cache * bdev_cachep __read_mostly;
736
737static struct inode *bdev_alloc_inode(struct super_block *sb)
738{
739 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
740 if (!ei)
741 return NULL;
742 return &ei->vfs_inode;
743}
744
745static void bdev_i_callback(struct rcu_head *head)
746{
747 struct inode *inode = container_of(head, struct inode, i_rcu);
748 struct bdev_inode *bdi = BDEV_I(inode);
749
750 kmem_cache_free(bdev_cachep, bdi);
751}
752
753static void bdev_destroy_inode(struct inode *inode)
754{
755 call_rcu(&inode->i_rcu, bdev_i_callback);
756}
757
758static void init_once(void *foo)
759{
760 struct bdev_inode *ei = (struct bdev_inode *) foo;
761 struct block_device *bdev = &ei->bdev;
762
763 memset(bdev, 0, sizeof(*bdev));
764 mutex_init(&bdev->bd_mutex);
765 INIT_LIST_HEAD(&bdev->bd_list);
766#ifdef CONFIG_SYSFS
767 INIT_LIST_HEAD(&bdev->bd_holder_disks);
768#endif
769 bdev->bd_bdi = &noop_backing_dev_info;
770 inode_init_once(&ei->vfs_inode);
771 /* Initialize mutex for freeze. */
772 mutex_init(&bdev->bd_fsfreeze_mutex);
773}
774
775static void bdev_evict_inode(struct inode *inode)
776{
777 struct block_device *bdev = &BDEV_I(inode)->bdev;
778 truncate_inode_pages_final(&inode->i_data);
779 invalidate_inode_buffers(inode); /* is it needed here? */
780 clear_inode(inode);
781 spin_lock(&bdev_lock);
782 list_del_init(&bdev->bd_list);
783 spin_unlock(&bdev_lock);
784 /* Detach inode from wb early as bdi_put() may free bdi->wb */
785 inode_detach_wb(inode);
786 if (bdev->bd_bdi != &noop_backing_dev_info) {
787 bdi_put(bdev->bd_bdi);
788 bdev->bd_bdi = &noop_backing_dev_info;
789 }
790}
791
792static const struct super_operations bdev_sops = {
793 .statfs = simple_statfs,
794 .alloc_inode = bdev_alloc_inode,
795 .destroy_inode = bdev_destroy_inode,
796 .drop_inode = generic_delete_inode,
797 .evict_inode = bdev_evict_inode,
798};
799
800static struct dentry *bd_mount(struct file_system_type *fs_type,
801 int flags, const char *dev_name, void *data)
802{
803 struct dentry *dent;
804 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
805 if (!IS_ERR(dent))
806 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
807 return dent;
808}
809
810static struct file_system_type bd_type = {
811 .name = "bdev",
812 .mount = bd_mount,
813 .kill_sb = kill_anon_super,
814};
815
816struct super_block *blockdev_superblock __read_mostly;
817EXPORT_SYMBOL_GPL(blockdev_superblock);
818
819void __init bdev_cache_init(void)
820{
821 int err;
822 static struct vfsmount *bd_mnt;
823
824 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
825 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
826 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
827 init_once);
828 err = register_filesystem(&bd_type);
829 if (err)
830 panic("Cannot register bdev pseudo-fs");
831 bd_mnt = kern_mount(&bd_type);
832 if (IS_ERR(bd_mnt))
833 panic("Cannot create bdev pseudo-fs");
834 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
835}
836
837/*
838 * Most likely _very_ bad one - but then it's hardly critical for small
839 * /dev and can be fixed when somebody will need really large one.
840 * Keep in mind that it will be fed through icache hash function too.
841 */
842static inline unsigned long hash(dev_t dev)
843{
844 return MAJOR(dev)+MINOR(dev);
845}
846
847static int bdev_test(struct inode *inode, void *data)
848{
849 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
850}
851
852static int bdev_set(struct inode *inode, void *data)
853{
854 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
855 return 0;
856}
857
858static LIST_HEAD(all_bdevs);
859
860/*
861 * If there is a bdev inode for this device, unhash it so that it gets evicted
862 * as soon as last inode reference is dropped.
863 */
864void bdev_unhash_inode(dev_t dev)
865{
866 struct inode *inode;
867
868 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
869 if (inode) {
870 remove_inode_hash(inode);
871 iput(inode);
872 }
873}
874
875struct block_device *bdget(dev_t dev)
876{
877 struct block_device *bdev;
878 struct inode *inode;
879
880 inode = iget5_locked(blockdev_superblock, hash(dev),
881 bdev_test, bdev_set, &dev);
882
883 if (!inode)
884 return NULL;
885
886 bdev = &BDEV_I(inode)->bdev;
887
888 if (inode->i_state & I_NEW) {
889 bdev->bd_contains = NULL;
890 bdev->bd_super = NULL;
891 bdev->bd_inode = inode;
892 bdev->bd_block_size = i_blocksize(inode);
893 bdev->bd_part_count = 0;
894 bdev->bd_invalidated = 0;
895 inode->i_mode = S_IFBLK;
896 inode->i_rdev = dev;
897 inode->i_bdev = bdev;
898 inode->i_data.a_ops = &def_blk_aops;
899 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
900 spin_lock(&bdev_lock);
901 list_add(&bdev->bd_list, &all_bdevs);
902 spin_unlock(&bdev_lock);
903 unlock_new_inode(inode);
904 }
905 return bdev;
906}
907
908EXPORT_SYMBOL(bdget);
909
910/**
911 * bdgrab -- Grab a reference to an already referenced block device
912 * @bdev: Block device to grab a reference to.
913 */
914struct block_device *bdgrab(struct block_device *bdev)
915{
916 ihold(bdev->bd_inode);
917 return bdev;
918}
919EXPORT_SYMBOL(bdgrab);
920
921long nr_blockdev_pages(void)
922{
923 struct block_device *bdev;
924 long ret = 0;
925 spin_lock(&bdev_lock);
926 list_for_each_entry(bdev, &all_bdevs, bd_list) {
927 ret += bdev->bd_inode->i_mapping->nrpages;
928 }
929 spin_unlock(&bdev_lock);
930 return ret;
931}
932
933void bdput(struct block_device *bdev)
934{
935 iput(bdev->bd_inode);
936}
937
938EXPORT_SYMBOL(bdput);
939
940static struct block_device *bd_acquire(struct inode *inode)
941{
942 struct block_device *bdev;
943
944 spin_lock(&bdev_lock);
945 bdev = inode->i_bdev;
946 if (bdev && !inode_unhashed(bdev->bd_inode)) {
947 bdgrab(bdev);
948 spin_unlock(&bdev_lock);
949 return bdev;
950 }
951 spin_unlock(&bdev_lock);
952
953 /*
954 * i_bdev references block device inode that was already shut down
955 * (corresponding device got removed). Remove the reference and look
956 * up block device inode again just in case new device got
957 * reestablished under the same device number.
958 */
959 if (bdev)
960 bd_forget(inode);
961
962 bdev = bdget(inode->i_rdev);
963 if (bdev) {
964 spin_lock(&bdev_lock);
965 if (!inode->i_bdev) {
966 /*
967 * We take an additional reference to bd_inode,
968 * and it's released in clear_inode() of inode.
969 * So, we can access it via ->i_mapping always
970 * without igrab().
971 */
972 bdgrab(bdev);
973 inode->i_bdev = bdev;
974 inode->i_mapping = bdev->bd_inode->i_mapping;
975 }
976 spin_unlock(&bdev_lock);
977 }
978 return bdev;
979}
980
981/* Call when you free inode */
982
983void bd_forget(struct inode *inode)
984{
985 struct block_device *bdev = NULL;
986
987 spin_lock(&bdev_lock);
988 if (!sb_is_blkdev_sb(inode->i_sb))
989 bdev = inode->i_bdev;
990 inode->i_bdev = NULL;
991 inode->i_mapping = &inode->i_data;
992 spin_unlock(&bdev_lock);
993
994 if (bdev)
995 bdput(bdev);
996}
997
998/**
999 * bd_may_claim - test whether a block device can be claimed
1000 * @bdev: block device of interest
1001 * @whole: whole block device containing @bdev, may equal @bdev
1002 * @holder: holder trying to claim @bdev
1003 *
1004 * Test whether @bdev can be claimed by @holder.
1005 *
1006 * CONTEXT:
1007 * spin_lock(&bdev_lock).
1008 *
1009 * RETURNS:
1010 * %true if @bdev can be claimed, %false otherwise.
1011 */
1012static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1013 void *holder)
1014{
1015 if (bdev->bd_holder == holder)
1016 return true; /* already a holder */
1017 else if (bdev->bd_holder != NULL)
1018 return false; /* held by someone else */
1019 else if (whole == bdev)
1020 return true; /* is a whole device which isn't held */
1021
1022 else if (whole->bd_holder == bd_may_claim)
1023 return true; /* is a partition of a device that is being partitioned */
1024 else if (whole->bd_holder != NULL)
1025 return false; /* is a partition of a held device */
1026 else
1027 return true; /* is a partition of an un-held device */
1028}
1029
1030/**
1031 * bd_prepare_to_claim - prepare to claim a block device
1032 * @bdev: block device of interest
1033 * @whole: the whole device containing @bdev, may equal @bdev
1034 * @holder: holder trying to claim @bdev
1035 *
1036 * Prepare to claim @bdev. This function fails if @bdev is already
1037 * claimed by another holder and waits if another claiming is in
1038 * progress. This function doesn't actually claim. On successful
1039 * return, the caller has ownership of bd_claiming and bd_holder[s].
1040 *
1041 * CONTEXT:
1042 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1043 * it multiple times.
1044 *
1045 * RETURNS:
1046 * 0 if @bdev can be claimed, -EBUSY otherwise.
1047 */
1048static int bd_prepare_to_claim(struct block_device *bdev,
1049 struct block_device *whole, void *holder)
1050{
1051retry:
1052 /* if someone else claimed, fail */
1053 if (!bd_may_claim(bdev, whole, holder))
1054 return -EBUSY;
1055
1056 /* if claiming is already in progress, wait for it to finish */
1057 if (whole->bd_claiming) {
1058 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1059 DEFINE_WAIT(wait);
1060
1061 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1062 spin_unlock(&bdev_lock);
1063 schedule();
1064 finish_wait(wq, &wait);
1065 spin_lock(&bdev_lock);
1066 goto retry;
1067 }
1068
1069 /* yay, all mine */
1070 return 0;
1071}
1072
1073/**
1074 * bd_start_claiming - start claiming a block device
1075 * @bdev: block device of interest
1076 * @holder: holder trying to claim @bdev
1077 *
1078 * @bdev is about to be opened exclusively. Check @bdev can be opened
1079 * exclusively and mark that an exclusive open is in progress. Each
1080 * successful call to this function must be matched with a call to
1081 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1082 * fail).
1083 *
1084 * This function is used to gain exclusive access to the block device
1085 * without actually causing other exclusive open attempts to fail. It
1086 * should be used when the open sequence itself requires exclusive
1087 * access but may subsequently fail.
1088 *
1089 * CONTEXT:
1090 * Might sleep.
1091 *
1092 * RETURNS:
1093 * Pointer to the block device containing @bdev on success, ERR_PTR()
1094 * value on failure.
1095 */
1096static struct block_device *bd_start_claiming(struct block_device *bdev,
1097 void *holder)
1098{
1099 struct gendisk *disk;
1100 struct block_device *whole;
1101 int partno, err;
1102
1103 might_sleep();
1104
1105 /*
1106 * @bdev might not have been initialized properly yet, look up
1107 * and grab the outer block device the hard way.
1108 */
1109 disk = get_gendisk(bdev->bd_dev, &partno);
1110 if (!disk)
1111 return ERR_PTR(-ENXIO);
1112
1113 /*
1114 * Normally, @bdev should equal what's returned from bdget_disk()
1115 * if partno is 0; however, some drivers (floppy) use multiple
1116 * bdev's for the same physical device and @bdev may be one of the
1117 * aliases. Keep @bdev if partno is 0. This means claimer
1118 * tracking is broken for those devices but it has always been that
1119 * way.
1120 */
1121 if (partno)
1122 whole = bdget_disk(disk, 0);
1123 else
1124 whole = bdgrab(bdev);
1125
1126 module_put(disk->fops->owner);
1127 put_disk(disk);
1128 if (!whole)
1129 return ERR_PTR(-ENOMEM);
1130
1131 /* prepare to claim, if successful, mark claiming in progress */
1132 spin_lock(&bdev_lock);
1133
1134 err = bd_prepare_to_claim(bdev, whole, holder);
1135 if (err == 0) {
1136 whole->bd_claiming = holder;
1137 spin_unlock(&bdev_lock);
1138 return whole;
1139 } else {
1140 spin_unlock(&bdev_lock);
1141 bdput(whole);
1142 return ERR_PTR(err);
1143 }
1144}
1145
1146#ifdef CONFIG_SYSFS
1147struct bd_holder_disk {
1148 struct list_head list;
1149 struct gendisk *disk;
1150 int refcnt;
1151};
1152
1153static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1154 struct gendisk *disk)
1155{
1156 struct bd_holder_disk *holder;
1157
1158 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1159 if (holder->disk == disk)
1160 return holder;
1161 return NULL;
1162}
1163
1164static int add_symlink(struct kobject *from, struct kobject *to)
1165{
1166 return sysfs_create_link(from, to, kobject_name(to));
1167}
1168
1169static void del_symlink(struct kobject *from, struct kobject *to)
1170{
1171 sysfs_remove_link(from, kobject_name(to));
1172}
1173
1174/**
1175 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1176 * @bdev: the claimed slave bdev
1177 * @disk: the holding disk
1178 *
1179 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1180 *
1181 * This functions creates the following sysfs symlinks.
1182 *
1183 * - from "slaves" directory of the holder @disk to the claimed @bdev
1184 * - from "holders" directory of the @bdev to the holder @disk
1185 *
1186 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1187 * passed to bd_link_disk_holder(), then:
1188 *
1189 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1190 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1191 *
1192 * The caller must have claimed @bdev before calling this function and
1193 * ensure that both @bdev and @disk are valid during the creation and
1194 * lifetime of these symlinks.
1195 *
1196 * CONTEXT:
1197 * Might sleep.
1198 *
1199 * RETURNS:
1200 * 0 on success, -errno on failure.
1201 */
1202int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1203{
1204 struct bd_holder_disk *holder;
1205 int ret = 0;
1206
1207 mutex_lock(&bdev->bd_mutex);
1208
1209 WARN_ON_ONCE(!bdev->bd_holder);
1210
1211 /* FIXME: remove the following once add_disk() handles errors */
1212 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1213 goto out_unlock;
1214
1215 holder = bd_find_holder_disk(bdev, disk);
1216 if (holder) {
1217 holder->refcnt++;
1218 goto out_unlock;
1219 }
1220
1221 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1222 if (!holder) {
1223 ret = -ENOMEM;
1224 goto out_unlock;
1225 }
1226
1227 INIT_LIST_HEAD(&holder->list);
1228 holder->disk = disk;
1229 holder->refcnt = 1;
1230
1231 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1232 if (ret)
1233 goto out_free;
1234
1235 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1236 if (ret)
1237 goto out_del;
1238 /*
1239 * bdev could be deleted beneath us which would implicitly destroy
1240 * the holder directory. Hold on to it.
1241 */
1242 kobject_get(bdev->bd_part->holder_dir);
1243
1244 list_add(&holder->list, &bdev->bd_holder_disks);
1245 goto out_unlock;
1246
1247out_del:
1248 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1249out_free:
1250 kfree(holder);
1251out_unlock:
1252 mutex_unlock(&bdev->bd_mutex);
1253 return ret;
1254}
1255EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1256
1257/**
1258 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1259 * @bdev: the calimed slave bdev
1260 * @disk: the holding disk
1261 *
1262 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1263 *
1264 * CONTEXT:
1265 * Might sleep.
1266 */
1267void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1268{
1269 struct bd_holder_disk *holder;
1270
1271 mutex_lock(&bdev->bd_mutex);
1272
1273 holder = bd_find_holder_disk(bdev, disk);
1274
1275 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1276 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1277 del_symlink(bdev->bd_part->holder_dir,
1278 &disk_to_dev(disk)->kobj);
1279 kobject_put(bdev->bd_part->holder_dir);
1280 list_del_init(&holder->list);
1281 kfree(holder);
1282 }
1283
1284 mutex_unlock(&bdev->bd_mutex);
1285}
1286EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1287#endif
1288
1289/**
1290 * flush_disk - invalidates all buffer-cache entries on a disk
1291 *
1292 * @bdev: struct block device to be flushed
1293 * @kill_dirty: flag to guide handling of dirty inodes
1294 *
1295 * Invalidates all buffer-cache entries on a disk. It should be called
1296 * when a disk has been changed -- either by a media change or online
1297 * resize.
1298 */
1299static void flush_disk(struct block_device *bdev, bool kill_dirty)
1300{
1301 if (__invalidate_device(bdev, kill_dirty)) {
1302 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1303 "resized disk %s\n",
1304 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1305 }
1306
1307 if (!bdev->bd_disk)
1308 return;
1309 if (disk_part_scan_enabled(bdev->bd_disk))
1310 bdev->bd_invalidated = 1;
1311}
1312
1313/**
1314 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1315 * @disk: struct gendisk to check
1316 * @bdev: struct bdev to adjust.
1317 *
1318 * This routine checks to see if the bdev size does not match the disk size
1319 * and adjusts it if it differs.
1320 */
1321void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1322{
1323 loff_t disk_size, bdev_size;
1324
1325 disk_size = (loff_t)get_capacity(disk) << 9;
1326 bdev_size = i_size_read(bdev->bd_inode);
1327 if (disk_size != bdev_size) {
1328 printk(KERN_INFO
1329 "%s: detected capacity change from %lld to %lld\n",
1330 disk->disk_name, bdev_size, disk_size);
1331 i_size_write(bdev->bd_inode, disk_size);
1332 flush_disk(bdev, false);
1333 }
1334}
1335EXPORT_SYMBOL(check_disk_size_change);
1336
1337/**
1338 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1339 * @disk: struct gendisk to be revalidated
1340 *
1341 * This routine is a wrapper for lower-level driver's revalidate_disk
1342 * call-backs. It is used to do common pre and post operations needed
1343 * for all revalidate_disk operations.
1344 */
1345int revalidate_disk(struct gendisk *disk)
1346{
1347 struct block_device *bdev;
1348 int ret = 0;
1349
1350 if (disk->fops->revalidate_disk)
1351 ret = disk->fops->revalidate_disk(disk);
1352 bdev = bdget_disk(disk, 0);
1353 if (!bdev)
1354 return ret;
1355
1356 mutex_lock(&bdev->bd_mutex);
1357 check_disk_size_change(disk, bdev);
1358 bdev->bd_invalidated = 0;
1359 mutex_unlock(&bdev->bd_mutex);
1360 bdput(bdev);
1361 return ret;
1362}
1363EXPORT_SYMBOL(revalidate_disk);
1364
1365/*
1366 * This routine checks whether a removable media has been changed,
1367 * and invalidates all buffer-cache-entries in that case. This
1368 * is a relatively slow routine, so we have to try to minimize using
1369 * it. Thus it is called only upon a 'mount' or 'open'. This
1370 * is the best way of combining speed and utility, I think.
1371 * People changing diskettes in the middle of an operation deserve
1372 * to lose :-)
1373 */
1374int check_disk_change(struct block_device *bdev)
1375{
1376 struct gendisk *disk = bdev->bd_disk;
1377 const struct block_device_operations *bdops = disk->fops;
1378 unsigned int events;
1379
1380 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1381 DISK_EVENT_EJECT_REQUEST);
1382 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1383 return 0;
1384
1385 flush_disk(bdev, true);
1386 if (bdops->revalidate_disk)
1387 bdops->revalidate_disk(bdev->bd_disk);
1388 return 1;
1389}
1390
1391EXPORT_SYMBOL(check_disk_change);
1392
1393void bd_set_size(struct block_device *bdev, loff_t size)
1394{
1395 unsigned bsize = bdev_logical_block_size(bdev);
1396
1397 inode_lock(bdev->bd_inode);
1398 i_size_write(bdev->bd_inode, size);
1399 inode_unlock(bdev->bd_inode);
1400 while (bsize < PAGE_SIZE) {
1401 if (size & bsize)
1402 break;
1403 bsize <<= 1;
1404 }
1405 bdev->bd_block_size = bsize;
1406 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1407}
1408EXPORT_SYMBOL(bd_set_size);
1409
1410static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1411
1412/*
1413 * bd_mutex locking:
1414 *
1415 * mutex_lock(part->bd_mutex)
1416 * mutex_lock_nested(whole->bd_mutex, 1)
1417 */
1418
1419static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1420{
1421 struct gendisk *disk;
1422 struct module *owner;
1423 int ret;
1424 int partno;
1425 int perm = 0;
1426
1427 if (mode & FMODE_READ)
1428 perm |= MAY_READ;
1429 if (mode & FMODE_WRITE)
1430 perm |= MAY_WRITE;
1431 /*
1432 * hooks: /n/, see "layering violations".
1433 */
1434 if (!for_part) {
1435 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1436 if (ret != 0) {
1437 bdput(bdev);
1438 return ret;
1439 }
1440 }
1441
1442 restart:
1443
1444 ret = -ENXIO;
1445 disk = get_gendisk(bdev->bd_dev, &partno);
1446 if (!disk)
1447 goto out;
1448 owner = disk->fops->owner;
1449
1450 disk_block_events(disk);
1451 mutex_lock_nested(&bdev->bd_mutex, for_part);
1452 if (!bdev->bd_openers) {
1453 bdev->bd_disk = disk;
1454 bdev->bd_queue = disk->queue;
1455 bdev->bd_contains = bdev;
1456 bdev->bd_partno = partno;
1457
1458 if (!partno) {
1459 ret = -ENXIO;
1460 bdev->bd_part = disk_get_part(disk, partno);
1461 if (!bdev->bd_part)
1462 goto out_clear;
1463
1464 ret = 0;
1465 if (disk->fops->open) {
1466 ret = disk->fops->open(bdev, mode);
1467 if (ret == -ERESTARTSYS) {
1468 /* Lost a race with 'disk' being
1469 * deleted, try again.
1470 * See md.c
1471 */
1472 disk_put_part(bdev->bd_part);
1473 bdev->bd_part = NULL;
1474 bdev->bd_disk = NULL;
1475 bdev->bd_queue = NULL;
1476 mutex_unlock(&bdev->bd_mutex);
1477 disk_unblock_events(disk);
1478 put_disk(disk);
1479 module_put(owner);
1480 goto restart;
1481 }
1482 }
1483
1484 if (!ret)
1485 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1486
1487 /*
1488 * If the device is invalidated, rescan partition
1489 * if open succeeded or failed with -ENOMEDIUM.
1490 * The latter is necessary to prevent ghost
1491 * partitions on a removed medium.
1492 */
1493 if (bdev->bd_invalidated) {
1494 if (!ret)
1495 rescan_partitions(disk, bdev);
1496 else if (ret == -ENOMEDIUM)
1497 invalidate_partitions(disk, bdev);
1498 }
1499
1500 if (ret)
1501 goto out_clear;
1502 } else {
1503 struct block_device *whole;
1504 whole = bdget_disk(disk, 0);
1505 ret = -ENOMEM;
1506 if (!whole)
1507 goto out_clear;
1508 BUG_ON(for_part);
1509 ret = __blkdev_get(whole, mode, 1);
1510 if (ret)
1511 goto out_clear;
1512 bdev->bd_contains = whole;
1513 bdev->bd_part = disk_get_part(disk, partno);
1514 if (!(disk->flags & GENHD_FL_UP) ||
1515 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1516 ret = -ENXIO;
1517 goto out_clear;
1518 }
1519 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1520 }
1521
1522 if (bdev->bd_bdi == &noop_backing_dev_info)
1523 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1524 } else {
1525 if (bdev->bd_contains == bdev) {
1526 ret = 0;
1527 if (bdev->bd_disk->fops->open)
1528 ret = bdev->bd_disk->fops->open(bdev, mode);
1529 /* the same as first opener case, read comment there */
1530 if (bdev->bd_invalidated) {
1531 if (!ret)
1532 rescan_partitions(bdev->bd_disk, bdev);
1533 else if (ret == -ENOMEDIUM)
1534 invalidate_partitions(bdev->bd_disk, bdev);
1535 }
1536 if (ret)
1537 goto out_unlock_bdev;
1538 }
1539 /* only one opener holds refs to the module and disk */
1540 put_disk(disk);
1541 module_put(owner);
1542 }
1543 bdev->bd_openers++;
1544 if (for_part)
1545 bdev->bd_part_count++;
1546 mutex_unlock(&bdev->bd_mutex);
1547 disk_unblock_events(disk);
1548 return 0;
1549
1550 out_clear:
1551 disk_put_part(bdev->bd_part);
1552 bdev->bd_disk = NULL;
1553 bdev->bd_part = NULL;
1554 bdev->bd_queue = NULL;
1555 if (bdev != bdev->bd_contains)
1556 __blkdev_put(bdev->bd_contains, mode, 1);
1557 bdev->bd_contains = NULL;
1558 out_unlock_bdev:
1559 mutex_unlock(&bdev->bd_mutex);
1560 disk_unblock_events(disk);
1561 put_disk(disk);
1562 module_put(owner);
1563 out:
1564 bdput(bdev);
1565
1566 return ret;
1567}
1568
1569/**
1570 * blkdev_get - open a block device
1571 * @bdev: block_device to open
1572 * @mode: FMODE_* mask
1573 * @holder: exclusive holder identifier
1574 *
1575 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1576 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1577 * @holder is invalid. Exclusive opens may nest for the same @holder.
1578 *
1579 * On success, the reference count of @bdev is unchanged. On failure,
1580 * @bdev is put.
1581 *
1582 * CONTEXT:
1583 * Might sleep.
1584 *
1585 * RETURNS:
1586 * 0 on success, -errno on failure.
1587 */
1588int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1589{
1590 struct block_device *whole = NULL;
1591 int res;
1592
1593 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1594
1595 if ((mode & FMODE_EXCL) && holder) {
1596 whole = bd_start_claiming(bdev, holder);
1597 if (IS_ERR(whole)) {
1598 bdput(bdev);
1599 return PTR_ERR(whole);
1600 }
1601 }
1602
1603 res = __blkdev_get(bdev, mode, 0);
1604
1605 if (whole) {
1606 struct gendisk *disk = whole->bd_disk;
1607
1608 /* finish claiming */
1609 mutex_lock(&bdev->bd_mutex);
1610 spin_lock(&bdev_lock);
1611
1612 if (!res) {
1613 BUG_ON(!bd_may_claim(bdev, whole, holder));
1614 /*
1615 * Note that for a whole device bd_holders
1616 * will be incremented twice, and bd_holder
1617 * will be set to bd_may_claim before being
1618 * set to holder
1619 */
1620 whole->bd_holders++;
1621 whole->bd_holder = bd_may_claim;
1622 bdev->bd_holders++;
1623 bdev->bd_holder = holder;
1624 }
1625
1626 /* tell others that we're done */
1627 BUG_ON(whole->bd_claiming != holder);
1628 whole->bd_claiming = NULL;
1629 wake_up_bit(&whole->bd_claiming, 0);
1630
1631 spin_unlock(&bdev_lock);
1632
1633 /*
1634 * Block event polling for write claims if requested. Any
1635 * write holder makes the write_holder state stick until
1636 * all are released. This is good enough and tracking
1637 * individual writeable reference is too fragile given the
1638 * way @mode is used in blkdev_get/put().
1639 */
1640 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1641 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1642 bdev->bd_write_holder = true;
1643 disk_block_events(disk);
1644 }
1645
1646 mutex_unlock(&bdev->bd_mutex);
1647 bdput(whole);
1648 }
1649
1650 return res;
1651}
1652EXPORT_SYMBOL(blkdev_get);
1653
1654/**
1655 * blkdev_get_by_path - open a block device by name
1656 * @path: path to the block device to open
1657 * @mode: FMODE_* mask
1658 * @holder: exclusive holder identifier
1659 *
1660 * Open the blockdevice described by the device file at @path. @mode
1661 * and @holder are identical to blkdev_get().
1662 *
1663 * On success, the returned block_device has reference count of one.
1664 *
1665 * CONTEXT:
1666 * Might sleep.
1667 *
1668 * RETURNS:
1669 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1670 */
1671struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1672 void *holder)
1673{
1674 struct block_device *bdev;
1675 int err;
1676
1677 bdev = lookup_bdev(path);
1678 if (IS_ERR(bdev))
1679 return bdev;
1680
1681 err = blkdev_get(bdev, mode, holder);
1682 if (err)
1683 return ERR_PTR(err);
1684
1685 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1686 blkdev_put(bdev, mode);
1687 return ERR_PTR(-EACCES);
1688 }
1689
1690 return bdev;
1691}
1692EXPORT_SYMBOL(blkdev_get_by_path);
1693
1694/**
1695 * blkdev_get_by_dev - open a block device by device number
1696 * @dev: device number of block device to open
1697 * @mode: FMODE_* mask
1698 * @holder: exclusive holder identifier
1699 *
1700 * Open the blockdevice described by device number @dev. @mode and
1701 * @holder are identical to blkdev_get().
1702 *
1703 * Use it ONLY if you really do not have anything better - i.e. when
1704 * you are behind a truly sucky interface and all you are given is a
1705 * device number. _Never_ to be used for internal purposes. If you
1706 * ever need it - reconsider your API.
1707 *
1708 * On success, the returned block_device has reference count of one.
1709 *
1710 * CONTEXT:
1711 * Might sleep.
1712 *
1713 * RETURNS:
1714 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1715 */
1716struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1717{
1718 struct block_device *bdev;
1719 int err;
1720
1721 bdev = bdget(dev);
1722 if (!bdev)
1723 return ERR_PTR(-ENOMEM);
1724
1725 err = blkdev_get(bdev, mode, holder);
1726 if (err)
1727 return ERR_PTR(err);
1728
1729 return bdev;
1730}
1731EXPORT_SYMBOL(blkdev_get_by_dev);
1732
1733static int blkdev_open(struct inode * inode, struct file * filp)
1734{
1735 struct block_device *bdev;
1736
1737 /*
1738 * Preserve backwards compatibility and allow large file access
1739 * even if userspace doesn't ask for it explicitly. Some mkfs
1740 * binary needs it. We might want to drop this workaround
1741 * during an unstable branch.
1742 */
1743 filp->f_flags |= O_LARGEFILE;
1744
1745 filp->f_mode |= FMODE_NOWAIT;
1746
1747 if (filp->f_flags & O_NDELAY)
1748 filp->f_mode |= FMODE_NDELAY;
1749 if (filp->f_flags & O_EXCL)
1750 filp->f_mode |= FMODE_EXCL;
1751 if ((filp->f_flags & O_ACCMODE) == 3)
1752 filp->f_mode |= FMODE_WRITE_IOCTL;
1753
1754 bdev = bd_acquire(inode);
1755 if (bdev == NULL)
1756 return -ENOMEM;
1757
1758 filp->f_mapping = bdev->bd_inode->i_mapping;
1759 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1760
1761 return blkdev_get(bdev, filp->f_mode, filp);
1762}
1763
1764static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1765{
1766 struct gendisk *disk = bdev->bd_disk;
1767 struct block_device *victim = NULL;
1768
1769 mutex_lock_nested(&bdev->bd_mutex, for_part);
1770 if (for_part)
1771 bdev->bd_part_count--;
1772
1773 if (!--bdev->bd_openers) {
1774 WARN_ON_ONCE(bdev->bd_holders);
1775 sync_blockdev(bdev);
1776 kill_bdev(bdev);
1777
1778 bdev_write_inode(bdev);
1779 }
1780 if (bdev->bd_contains == bdev) {
1781 if (disk->fops->release)
1782 disk->fops->release(disk, mode);
1783 }
1784 if (!bdev->bd_openers) {
1785 struct module *owner = disk->fops->owner;
1786
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(disk);
1795 module_put(owner);
1796 }
1797 mutex_unlock(&bdev->bd_mutex);
1798 bdput(bdev);
1799 if (victim)
1800 __blkdev_put(victim, mode, 1);
1801}
1802
1803void blkdev_put(struct block_device *bdev, fmode_t mode)
1804{
1805 mutex_lock(&bdev->bd_mutex);
1806
1807 if (mode & FMODE_EXCL) {
1808 bool bdev_free;
1809
1810 /*
1811 * Release a claim on the device. The holder fields
1812 * are protected with bdev_lock. bd_mutex is to
1813 * synchronize disk_holder unlinking.
1814 */
1815 spin_lock(&bdev_lock);
1816
1817 WARN_ON_ONCE(--bdev->bd_holders < 0);
1818 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1819
1820 /* bd_contains might point to self, check in a separate step */
1821 if ((bdev_free = !bdev->bd_holders))
1822 bdev->bd_holder = NULL;
1823 if (!bdev->bd_contains->bd_holders)
1824 bdev->bd_contains->bd_holder = NULL;
1825
1826 spin_unlock(&bdev_lock);
1827
1828 /*
1829 * If this was the last claim, remove holder link and
1830 * unblock evpoll if it was a write holder.
1831 */
1832 if (bdev_free && bdev->bd_write_holder) {
1833 disk_unblock_events(bdev->bd_disk);
1834 bdev->bd_write_holder = false;
1835 }
1836 }
1837
1838 /*
1839 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1840 * event. This is to ensure detection of media removal commanded
1841 * from userland - e.g. eject(1).
1842 */
1843 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1844
1845 mutex_unlock(&bdev->bd_mutex);
1846
1847 __blkdev_put(bdev, mode, 0);
1848}
1849EXPORT_SYMBOL(blkdev_put);
1850
1851static int blkdev_close(struct inode * inode, struct file * filp)
1852{
1853 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1854 blkdev_put(bdev, filp->f_mode);
1855 return 0;
1856}
1857
1858static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1859{
1860 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1861 fmode_t mode = file->f_mode;
1862
1863 /*
1864 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1865 * to updated it before every ioctl.
1866 */
1867 if (file->f_flags & O_NDELAY)
1868 mode |= FMODE_NDELAY;
1869 else
1870 mode &= ~FMODE_NDELAY;
1871
1872 return blkdev_ioctl(bdev, mode, cmd, arg);
1873}
1874
1875/*
1876 * Write data to the block device. Only intended for the block device itself
1877 * and the raw driver which basically is a fake block device.
1878 *
1879 * Does not take i_mutex for the write and thus is not for general purpose
1880 * use.
1881 */
1882ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1883{
1884 struct file *file = iocb->ki_filp;
1885 struct inode *bd_inode = bdev_file_inode(file);
1886 loff_t size = i_size_read(bd_inode);
1887 struct blk_plug plug;
1888 ssize_t ret;
1889
1890 if (bdev_read_only(I_BDEV(bd_inode)))
1891 return -EPERM;
1892
1893 if (!iov_iter_count(from))
1894 return 0;
1895
1896 if (iocb->ki_pos >= size)
1897 return -ENOSPC;
1898
1899 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1900 return -EOPNOTSUPP;
1901
1902 iov_iter_truncate(from, size - iocb->ki_pos);
1903
1904 blk_start_plug(&plug);
1905 ret = __generic_file_write_iter(iocb, from);
1906 if (ret > 0)
1907 ret = generic_write_sync(iocb, ret);
1908 blk_finish_plug(&plug);
1909 return ret;
1910}
1911EXPORT_SYMBOL_GPL(blkdev_write_iter);
1912
1913ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1914{
1915 struct file *file = iocb->ki_filp;
1916 struct inode *bd_inode = bdev_file_inode(file);
1917 loff_t size = i_size_read(bd_inode);
1918 loff_t pos = iocb->ki_pos;
1919
1920 if (pos >= size)
1921 return 0;
1922
1923 size -= pos;
1924 iov_iter_truncate(to, size);
1925 return generic_file_read_iter(iocb, to);
1926}
1927EXPORT_SYMBOL_GPL(blkdev_read_iter);
1928
1929/*
1930 * Try to release a page associated with block device when the system
1931 * is under memory pressure.
1932 */
1933static int blkdev_releasepage(struct page *page, gfp_t wait)
1934{
1935 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1936
1937 if (super && super->s_op->bdev_try_to_free_page)
1938 return super->s_op->bdev_try_to_free_page(super, page, wait);
1939
1940 return try_to_free_buffers(page);
1941}
1942
1943static int blkdev_writepages(struct address_space *mapping,
1944 struct writeback_control *wbc)
1945{
1946 if (dax_mapping(mapping)) {
1947 struct block_device *bdev = I_BDEV(mapping->host);
1948
1949 return dax_writeback_mapping_range(mapping, bdev, wbc);
1950 }
1951 return generic_writepages(mapping, wbc);
1952}
1953
1954static const struct address_space_operations def_blk_aops = {
1955 .readpage = blkdev_readpage,
1956 .readpages = blkdev_readpages,
1957 .writepage = blkdev_writepage,
1958 .write_begin = blkdev_write_begin,
1959 .write_end = blkdev_write_end,
1960 .writepages = blkdev_writepages,
1961 .releasepage = blkdev_releasepage,
1962 .direct_IO = blkdev_direct_IO,
1963 .is_dirty_writeback = buffer_check_dirty_writeback,
1964};
1965
1966#define BLKDEV_FALLOC_FL_SUPPORTED \
1967 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1968 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1969
1970static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1971 loff_t len)
1972{
1973 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1974 struct address_space *mapping;
1975 loff_t end = start + len - 1;
1976 loff_t isize;
1977 int error;
1978
1979 /* Fail if we don't recognize the flags. */
1980 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1981 return -EOPNOTSUPP;
1982
1983 /* Don't go off the end of the device. */
1984 isize = i_size_read(bdev->bd_inode);
1985 if (start >= isize)
1986 return -EINVAL;
1987 if (end >= isize) {
1988 if (mode & FALLOC_FL_KEEP_SIZE) {
1989 len = isize - start;
1990 end = start + len - 1;
1991 } else
1992 return -EINVAL;
1993 }
1994
1995 /*
1996 * Don't allow IO that isn't aligned to logical block size.
1997 */
1998 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1999 return -EINVAL;
2000
2001 /* Invalidate the page cache, including dirty pages. */
2002 mapping = bdev->bd_inode->i_mapping;
2003 truncate_inode_pages_range(mapping, start, end);
2004
2005 switch (mode) {
2006 case FALLOC_FL_ZERO_RANGE:
2007 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2008 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2009 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2010 break;
2011 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2012 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2013 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2014 break;
2015 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2016 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2017 GFP_KERNEL, 0);
2018 break;
2019 default:
2020 return -EOPNOTSUPP;
2021 }
2022 if (error)
2023 return error;
2024
2025 /*
2026 * Invalidate again; if someone wandered in and dirtied a page,
2027 * the caller will be given -EBUSY. The third argument is
2028 * inclusive, so the rounding here is safe.
2029 */
2030 return invalidate_inode_pages2_range(mapping,
2031 start >> PAGE_SHIFT,
2032 end >> PAGE_SHIFT);
2033}
2034
2035const struct file_operations def_blk_fops = {
2036 .open = blkdev_open,
2037 .release = blkdev_close,
2038 .llseek = block_llseek,
2039 .read_iter = blkdev_read_iter,
2040 .write_iter = blkdev_write_iter,
2041 .mmap = generic_file_mmap,
2042 .fsync = blkdev_fsync,
2043 .unlocked_ioctl = block_ioctl,
2044#ifdef CONFIG_COMPAT
2045 .compat_ioctl = compat_blkdev_ioctl,
2046#endif
2047 .splice_read = generic_file_splice_read,
2048 .splice_write = iter_file_splice_write,
2049 .fallocate = blkdev_fallocate,
2050};
2051
2052int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2053{
2054 int res;
2055 mm_segment_t old_fs = get_fs();
2056 set_fs(KERNEL_DS);
2057 res = blkdev_ioctl(bdev, 0, cmd, arg);
2058 set_fs(old_fs);
2059 return res;
2060}
2061
2062EXPORT_SYMBOL(ioctl_by_bdev);
2063
2064/**
2065 * lookup_bdev - lookup a struct block_device by name
2066 * @pathname: special file representing the block device
2067 *
2068 * Get a reference to the blockdevice at @pathname in the current
2069 * namespace if possible and return it. Return ERR_PTR(error)
2070 * otherwise.
2071 */
2072struct block_device *lookup_bdev(const char *pathname)
2073{
2074 struct block_device *bdev;
2075 struct inode *inode;
2076 struct path path;
2077 int error;
2078
2079 if (!pathname || !*pathname)
2080 return ERR_PTR(-EINVAL);
2081
2082 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2083 if (error)
2084 return ERR_PTR(error);
2085
2086 inode = d_backing_inode(path.dentry);
2087 error = -ENOTBLK;
2088 if (!S_ISBLK(inode->i_mode))
2089 goto fail;
2090 error = -EACCES;
2091 if (!may_open_dev(&path))
2092 goto fail;
2093 error = -ENOMEM;
2094 bdev = bd_acquire(inode);
2095 if (!bdev)
2096 goto fail;
2097out:
2098 path_put(&path);
2099 return bdev;
2100fail:
2101 bdev = ERR_PTR(error);
2102 goto out;
2103}
2104EXPORT_SYMBOL(lookup_bdev);
2105
2106int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2107{
2108 struct super_block *sb = get_super(bdev);
2109 int res = 0;
2110
2111 if (sb) {
2112 /*
2113 * no need to lock the super, get_super holds the
2114 * read mutex so the filesystem cannot go away
2115 * under us (->put_super runs with the write lock
2116 * hold).
2117 */
2118 shrink_dcache_sb(sb);
2119 res = invalidate_inodes(sb, kill_dirty);
2120 drop_super(sb);
2121 }
2122 invalidate_bdev(bdev);
2123 return res;
2124}
2125EXPORT_SYMBOL(__invalidate_device);
2126
2127void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2128{
2129 struct inode *inode, *old_inode = NULL;
2130
2131 spin_lock(&blockdev_superblock->s_inode_list_lock);
2132 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2133 struct address_space *mapping = inode->i_mapping;
2134 struct block_device *bdev;
2135
2136 spin_lock(&inode->i_lock);
2137 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2138 mapping->nrpages == 0) {
2139 spin_unlock(&inode->i_lock);
2140 continue;
2141 }
2142 __iget(inode);
2143 spin_unlock(&inode->i_lock);
2144 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2145 /*
2146 * We hold a reference to 'inode' so it couldn't have been
2147 * removed from s_inodes list while we dropped the
2148 * s_inode_list_lock We cannot iput the inode now as we can
2149 * be holding the last reference and we cannot iput it under
2150 * s_inode_list_lock. So we keep the reference and iput it
2151 * later.
2152 */
2153 iput(old_inode);
2154 old_inode = inode;
2155 bdev = I_BDEV(inode);
2156
2157 mutex_lock(&bdev->bd_mutex);
2158 if (bdev->bd_openers)
2159 func(bdev, arg);
2160 mutex_unlock(&bdev->bd_mutex);
2161
2162 spin_lock(&blockdev_superblock->s_inode_list_lock);
2163 }
2164 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2165 iput(old_inode);
2166}