projects/linux·blob v4.16-rc6 · refs

2169 lines · 53.7 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
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.. */
78void 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}
88EXPORT_SYMBOL(kill_bdev);
89
90/* Invalidate clean unused buffers and pagecache. */
91void invalidate_bdev(struct block_device *bdev)
92{
93 struct address_space *mapping = bdev->bd_inode->i_mapping;
94
95 if (mapping->nrpages) {
96 invalidate_bh_lrus();
97 lru_add_drain_all(); /* make sure all lru add caches are flushed */
98 invalidate_mapping_pages(mapping, 0, -1);
99 }
100 /* 99% of the time, we don't need to flush the cleancache on the bdev.
101 * But, for the strange corners, lets be cautious
102 */
103 cleancache_invalidate_inode(mapping);
104}
105EXPORT_SYMBOL(invalidate_bdev);
106
107int set_blocksize(struct block_device *bdev, int size)
108{
109 /* Size must be a power of two, and between 512 and PAGE_SIZE */
110 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
111 return -EINVAL;
112
113 /* Size cannot be smaller than the size supported by the device */
114 if (size < bdev_logical_block_size(bdev))
115 return -EINVAL;
116
117 /* Don't change the size if it is same as current */
118 if (bdev->bd_block_size != size) {
119 sync_blockdev(bdev);
120 bdev->bd_block_size = size;
121 bdev->bd_inode->i_blkbits = blksize_bits(size);
122 kill_bdev(bdev);
123 }
124 return 0;
125}
126
127EXPORT_SYMBOL(set_blocksize);
128
129int sb_set_blocksize(struct super_block *sb, int size)
130{
131 if (set_blocksize(sb->s_bdev, size))
132 return 0;
133 /* If we get here, we know size is power of two
134 * and it's value is between 512 and PAGE_SIZE */
135 sb->s_blocksize = size;
136 sb->s_blocksize_bits = blksize_bits(size);
137 return sb->s_blocksize;
138}
139
140EXPORT_SYMBOL(sb_set_blocksize);
141
142int sb_min_blocksize(struct super_block *sb, int size)
143{
144 int minsize = bdev_logical_block_size(sb->s_bdev);
145 if (size < minsize)
146 size = minsize;
147 return sb_set_blocksize(sb, size);
148}
149
150EXPORT_SYMBOL(sb_min_blocksize);
151
152static int
153blkdev_get_block(struct inode *inode, sector_t iblock,
154 struct buffer_head *bh, int create)
155{
156 bh->b_bdev = I_BDEV(inode);
157 bh->b_blocknr = iblock;
158 set_buffer_mapped(bh);
159 return 0;
160}
161
162static struct inode *bdev_file_inode(struct file *file)
163{
164 return file->f_mapping->host;
165}
166
167static unsigned int dio_bio_write_op(struct kiocb *iocb)
168{
169 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
170
171 /* avoid the need for a I/O completion work item */
172 if (iocb->ki_flags & IOCB_DSYNC)
173 op |= REQ_FUA;
174 return op;
175}
176
177#define DIO_INLINE_BIO_VECS 4
178
179static void blkdev_bio_end_io_simple(struct bio *bio)
180{
181 struct task_struct *waiter = bio->bi_private;
182
183 WRITE_ONCE(bio->bi_private, NULL);
184 wake_up_process(waiter);
185}
186
187static ssize_t
188__blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
189 int nr_pages)
190{
191 struct file *file = iocb->ki_filp;
192 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
193 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec;
194 loff_t pos = iocb->ki_pos;
195 bool should_dirty = false;
196 struct bio bio;
197 ssize_t ret;
198 blk_qc_t qc;
199 int i;
200
201 if ((pos | iov_iter_alignment(iter)) &
202 (bdev_logical_block_size(bdev) - 1))
203 return -EINVAL;
204
205 if (nr_pages <= DIO_INLINE_BIO_VECS)
206 vecs = inline_vecs;
207 else {
208 vecs = kmalloc(nr_pages * sizeof(struct bio_vec), GFP_KERNEL);
209 if (!vecs)
210 return -ENOMEM;
211 }
212
213 bio_init(&bio, vecs, nr_pages);
214 bio_set_dev(&bio, bdev);
215 bio.bi_iter.bi_sector = pos >> 9;
216 bio.bi_write_hint = iocb->ki_hint;
217 bio.bi_private = current;
218 bio.bi_end_io = blkdev_bio_end_io_simple;
219
220 ret = bio_iov_iter_get_pages(&bio, iter);
221 if (unlikely(ret))
222 return ret;
223 ret = bio.bi_iter.bi_size;
224
225 if (iov_iter_rw(iter) == READ) {
226 bio.bi_opf = REQ_OP_READ;
227 if (iter_is_iovec(iter))
228 should_dirty = true;
229 } else {
230 bio.bi_opf = dio_bio_write_op(iocb);
231 task_io_account_write(ret);
232 }
233
234 qc = submit_bio(&bio);
235 for (;;) {
236 set_current_state(TASK_UNINTERRUPTIBLE);
237 if (!READ_ONCE(bio.bi_private))
238 break;
239 if (!(iocb->ki_flags & IOCB_HIPRI) ||
240 !blk_poll(bdev_get_queue(bdev), qc))
241 io_schedule();
242 }
243 __set_current_state(TASK_RUNNING);
244
245 bio_for_each_segment_all(bvec, &bio, i) {
246 if (should_dirty && !PageCompound(bvec->bv_page))
247 set_page_dirty_lock(bvec->bv_page);
248 put_page(bvec->bv_page);
249 }
250
251 if (vecs != inline_vecs)
252 kfree(vecs);
253
254 if (unlikely(bio.bi_status))
255 ret = blk_status_to_errno(bio.bi_status);
256
257 bio_uninit(&bio);
258
259 return ret;
260}
261
262struct blkdev_dio {
263 union {
264 struct kiocb *iocb;
265 struct task_struct *waiter;
266 };
267 size_t size;
268 atomic_t ref;
269 bool multi_bio : 1;
270 bool should_dirty : 1;
271 bool is_sync : 1;
272 struct bio bio;
273};
274
275static struct bio_set *blkdev_dio_pool __read_mostly;
276
277static void blkdev_bio_end_io(struct bio *bio)
278{
279 struct blkdev_dio *dio = bio->bi_private;
280 bool should_dirty = dio->should_dirty;
281
282 if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) {
283 if (bio->bi_status && !dio->bio.bi_status)
284 dio->bio.bi_status = bio->bi_status;
285 } else {
286 if (!dio->is_sync) {
287 struct kiocb *iocb = dio->iocb;
288 ssize_t ret;
289
290 if (likely(!dio->bio.bi_status)) {
291 ret = dio->size;
292 iocb->ki_pos += ret;
293 } else {
294 ret = blk_status_to_errno(dio->bio.bi_status);
295 }
296
297 dio->iocb->ki_complete(iocb, ret, 0);
298 bio_put(&dio->bio);
299 } else {
300 struct task_struct *waiter = dio->waiter;
301
302 WRITE_ONCE(dio->waiter, NULL);
303 wake_up_process(waiter);
304 }
305 }
306
307 if (should_dirty) {
308 bio_check_pages_dirty(bio);
309 } else {
310 struct bio_vec *bvec;
311 int i;
312
313 bio_for_each_segment_all(bvec, bio, i)
314 put_page(bvec->bv_page);
315 bio_put(bio);
316 }
317}
318
319static ssize_t
320__blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
321{
322 struct file *file = iocb->ki_filp;
323 struct inode *inode = bdev_file_inode(file);
324 struct block_device *bdev = I_BDEV(inode);
325 struct blk_plug plug;
326 struct blkdev_dio *dio;
327 struct bio *bio;
328 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
329 loff_t pos = iocb->ki_pos;
330 blk_qc_t qc = BLK_QC_T_NONE;
331 int ret = 0;
332
333 if ((pos | iov_iter_alignment(iter)) &
334 (bdev_logical_block_size(bdev) - 1))
335 return -EINVAL;
336
337 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, blkdev_dio_pool);
338 bio_get(bio); /* extra ref for the completion handler */
339
340 dio = container_of(bio, struct blkdev_dio, bio);
341 dio->is_sync = is_sync = is_sync_kiocb(iocb);
342 if (dio->is_sync)
343 dio->waiter = current;
344 else
345 dio->iocb = iocb;
346
347 dio->size = 0;
348 dio->multi_bio = false;
349 dio->should_dirty = is_read && (iter->type == ITER_IOVEC);
350
351 blk_start_plug(&plug);
352 for (;;) {
353 bio_set_dev(bio, bdev);
354 bio->bi_iter.bi_sector = pos >> 9;
355 bio->bi_write_hint = iocb->ki_hint;
356 bio->bi_private = dio;
357 bio->bi_end_io = blkdev_bio_end_io;
358
359 ret = bio_iov_iter_get_pages(bio, iter);
360 if (unlikely(ret)) {
361 bio->bi_status = BLK_STS_IOERR;
362 bio_endio(bio);
363 break;
364 }
365
366 if (is_read) {
367 bio->bi_opf = REQ_OP_READ;
368 if (dio->should_dirty)
369 bio_set_pages_dirty(bio);
370 } else {
371 bio->bi_opf = dio_bio_write_op(iocb);
372 task_io_account_write(bio->bi_iter.bi_size);
373 }
374
375 dio->size += bio->bi_iter.bi_size;
376 pos += bio->bi_iter.bi_size;
377
378 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
379 if (!nr_pages) {
380 qc = submit_bio(bio);
381 break;
382 }
383
384 if (!dio->multi_bio) {
385 dio->multi_bio = true;
386 atomic_set(&dio->ref, 2);
387 } else {
388 atomic_inc(&dio->ref);
389 }
390
391 submit_bio(bio);
392 bio = bio_alloc(GFP_KERNEL, nr_pages);
393 }
394 blk_finish_plug(&plug);
395
396 if (!is_sync)
397 return -EIOCBQUEUED;
398
399 for (;;) {
400 set_current_state(TASK_UNINTERRUPTIBLE);
401 if (!READ_ONCE(dio->waiter))
402 break;
403
404 if (!(iocb->ki_flags & IOCB_HIPRI) ||
405 !blk_poll(bdev_get_queue(bdev), qc))
406 io_schedule();
407 }
408 __set_current_state(TASK_RUNNING);
409
410 if (!ret)
411 ret = blk_status_to_errno(dio->bio.bi_status);
412 if (likely(!ret))
413 ret = dio->size;
414
415 bio_put(&dio->bio);
416 return ret;
417}
418
419static ssize_t
420blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
421{
422 int nr_pages;
423
424 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
425 if (!nr_pages)
426 return 0;
427 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
428 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
429
430 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
431}
432
433static __init int blkdev_init(void)
434{
435 blkdev_dio_pool = bioset_create(4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
436 if (!blkdev_dio_pool)
437 return -ENOMEM;
438 return 0;
439}
440module_init(blkdev_init);
441
442int __sync_blockdev(struct block_device *bdev, int wait)
443{
444 if (!bdev)
445 return 0;
446 if (!wait)
447 return filemap_flush(bdev->bd_inode->i_mapping);
448 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
449}
450
451/*
452 * Write out and wait upon all the dirty data associated with a block
453 * device via its mapping. Does not take the superblock lock.
454 */
455int sync_blockdev(struct block_device *bdev)
456{
457 return __sync_blockdev(bdev, 1);
458}
459EXPORT_SYMBOL(sync_blockdev);
460
461/*
462 * Write out and wait upon all dirty data associated with this
463 * device. Filesystem data as well as the underlying block
464 * device. Takes the superblock lock.
465 */
466int fsync_bdev(struct block_device *bdev)
467{
468 struct super_block *sb = get_super(bdev);
469 if (sb) {
470 int res = sync_filesystem(sb);
471 drop_super(sb);
472 return res;
473 }
474 return sync_blockdev(bdev);
475}
476EXPORT_SYMBOL(fsync_bdev);
477
478/**
479 * freeze_bdev -- lock a filesystem and force it into a consistent state
480 * @bdev: blockdevice to lock
481 *
482 * If a superblock is found on this device, we take the s_umount semaphore
483 * on it to make sure nobody unmounts until the snapshot creation is done.
484 * The reference counter (bd_fsfreeze_count) guarantees that only the last
485 * unfreeze process can unfreeze the frozen filesystem actually when multiple
486 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
487 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
488 * actually.
489 */
490struct super_block *freeze_bdev(struct block_device *bdev)
491{
492 struct super_block *sb;
493 int error = 0;
494
495 mutex_lock(&bdev->bd_fsfreeze_mutex);
496 if (++bdev->bd_fsfreeze_count > 1) {
497 /*
498 * We don't even need to grab a reference - the first call
499 * to freeze_bdev grab an active reference and only the last
500 * thaw_bdev drops it.
501 */
502 sb = get_super(bdev);
503 if (sb)
504 drop_super(sb);
505 mutex_unlock(&bdev->bd_fsfreeze_mutex);
506 return sb;
507 }
508
509 sb = get_active_super(bdev);
510 if (!sb)
511 goto out;
512 if (sb->s_op->freeze_super)
513 error = sb->s_op->freeze_super(sb);
514 else
515 error = freeze_super(sb);
516 if (error) {
517 deactivate_super(sb);
518 bdev->bd_fsfreeze_count--;
519 mutex_unlock(&bdev->bd_fsfreeze_mutex);
520 return ERR_PTR(error);
521 }
522 deactivate_super(sb);
523 out:
524 sync_blockdev(bdev);
525 mutex_unlock(&bdev->bd_fsfreeze_mutex);
526 return sb; /* thaw_bdev releases s->s_umount */
527}
528EXPORT_SYMBOL(freeze_bdev);
529
530/**
531 * thaw_bdev -- unlock filesystem
532 * @bdev: blockdevice to unlock
533 * @sb: associated superblock
534 *
535 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
536 */
537int thaw_bdev(struct block_device *bdev, struct super_block *sb)
538{
539 int error = -EINVAL;
540
541 mutex_lock(&bdev->bd_fsfreeze_mutex);
542 if (!bdev->bd_fsfreeze_count)
543 goto out;
544
545 error = 0;
546 if (--bdev->bd_fsfreeze_count > 0)
547 goto out;
548
549 if (!sb)
550 goto out;
551
552 if (sb->s_op->thaw_super)
553 error = sb->s_op->thaw_super(sb);
554 else
555 error = thaw_super(sb);
556 if (error)
557 bdev->bd_fsfreeze_count++;
558out:
559 mutex_unlock(&bdev->bd_fsfreeze_mutex);
560 return error;
561}
562EXPORT_SYMBOL(thaw_bdev);
563
564static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
565{
566 return block_write_full_page(page, blkdev_get_block, wbc);
567}
568
569static int blkdev_readpage(struct file * file, struct page * page)
570{
571 return block_read_full_page(page, blkdev_get_block);
572}
573
574static int blkdev_readpages(struct file *file, struct address_space *mapping,
575 struct list_head *pages, unsigned nr_pages)
576{
577 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
578}
579
580static int blkdev_write_begin(struct file *file, struct address_space *mapping,
581 loff_t pos, unsigned len, unsigned flags,
582 struct page **pagep, void **fsdata)
583{
584 return block_write_begin(mapping, pos, len, flags, pagep,
585 blkdev_get_block);
586}
587
588static int blkdev_write_end(struct file *file, struct address_space *mapping,
589 loff_t pos, unsigned len, unsigned copied,
590 struct page *page, void *fsdata)
591{
592 int ret;
593 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
594
595 unlock_page(page);
596 put_page(page);
597
598 return ret;
599}
600
601/*
602 * private llseek:
603 * for a block special file file_inode(file)->i_size is zero
604 * so we compute the size by hand (just as in block_read/write above)
605 */
606static loff_t block_llseek(struct file *file, loff_t offset, int whence)
607{
608 struct inode *bd_inode = bdev_file_inode(file);
609 loff_t retval;
610
611 inode_lock(bd_inode);
612 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
613 inode_unlock(bd_inode);
614 return retval;
615}
616
617int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
618{
619 struct inode *bd_inode = bdev_file_inode(filp);
620 struct block_device *bdev = I_BDEV(bd_inode);
621 int error;
622
623 error = file_write_and_wait_range(filp, start, end);
624 if (error)
625 return error;
626
627 /*
628 * There is no need to serialise calls to blkdev_issue_flush with
629 * i_mutex and doing so causes performance issues with concurrent
630 * O_SYNC writers to a block device.
631 */
632 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
633 if (error == -EOPNOTSUPP)
634 error = 0;
635
636 return error;
637}
638EXPORT_SYMBOL(blkdev_fsync);
639
640/**
641 * bdev_read_page() - Start reading a page from a block device
642 * @bdev: The device to read the page from
643 * @sector: The offset on the device to read the page to (need not be aligned)
644 * @page: The page to read
645 *
646 * On entry, the page should be locked. It will be unlocked when the page
647 * has been read. If the block driver implements rw_page synchronously,
648 * that will be true on exit from this function, but it need not be.
649 *
650 * Errors returned by this function are usually "soft", eg out of memory, or
651 * queue full; callers should try a different route to read this page rather
652 * than propagate an error back up the stack.
653 *
654 * Return: negative errno if an error occurs, 0 if submission was successful.
655 */
656int bdev_read_page(struct block_device *bdev, sector_t sector,
657 struct page *page)
658{
659 const struct block_device_operations *ops = bdev->bd_disk->fops;
660 int result = -EOPNOTSUPP;
661
662 if (!ops->rw_page || bdev_get_integrity(bdev))
663 return result;
664
665 result = blk_queue_enter(bdev->bd_queue, 0);
666 if (result)
667 return result;
668 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false);
669 blk_queue_exit(bdev->bd_queue);
670 return result;
671}
672EXPORT_SYMBOL_GPL(bdev_read_page);
673
674/**
675 * bdev_write_page() - Start writing a page to a block device
676 * @bdev: The device to write the page to
677 * @sector: The offset on the device to write the page to (need not be aligned)
678 * @page: The page to write
679 * @wbc: The writeback_control for the write
680 *
681 * On entry, the page should be locked and not currently under writeback.
682 * On exit, if the write started successfully, the page will be unlocked and
683 * under writeback. If the write failed already (eg the driver failed to
684 * queue the page to the device), the page will still be locked. If the
685 * caller is a ->writepage implementation, it will need to unlock the page.
686 *
687 * Errors returned by this function are usually "soft", eg out of memory, or
688 * queue full; callers should try a different route to write this page rather
689 * than propagate an error back up the stack.
690 *
691 * Return: negative errno if an error occurs, 0 if submission was successful.
692 */
693int bdev_write_page(struct block_device *bdev, sector_t sector,
694 struct page *page, struct writeback_control *wbc)
695{
696 int result;
697 const struct block_device_operations *ops = bdev->bd_disk->fops;
698
699 if (!ops->rw_page || bdev_get_integrity(bdev))
700 return -EOPNOTSUPP;
701 result = blk_queue_enter(bdev->bd_queue, 0);
702 if (result)
703 return result;
704
705 set_page_writeback(page);
706 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true);
707 if (result) {
708 end_page_writeback(page);
709 } else {
710 clean_page_buffers(page);
711 unlock_page(page);
712 }
713 blk_queue_exit(bdev->bd_queue);
714 return result;
715}
716EXPORT_SYMBOL_GPL(bdev_write_page);
717
718/*
719 * pseudo-fs
720 */
721
722static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
723static struct kmem_cache * bdev_cachep __read_mostly;
724
725static struct inode *bdev_alloc_inode(struct super_block *sb)
726{
727 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
728 if (!ei)
729 return NULL;
730 return &ei->vfs_inode;
731}
732
733static void bdev_i_callback(struct rcu_head *head)
734{
735 struct inode *inode = container_of(head, struct inode, i_rcu);
736 struct bdev_inode *bdi = BDEV_I(inode);
737
738 kmem_cache_free(bdev_cachep, bdi);
739}
740
741static void bdev_destroy_inode(struct inode *inode)
742{
743 call_rcu(&inode->i_rcu, bdev_i_callback);
744}
745
746static void init_once(void *foo)
747{
748 struct bdev_inode *ei = (struct bdev_inode *) foo;
749 struct block_device *bdev = &ei->bdev;
750
751 memset(bdev, 0, sizeof(*bdev));
752 mutex_init(&bdev->bd_mutex);
753 INIT_LIST_HEAD(&bdev->bd_list);
754#ifdef CONFIG_SYSFS
755 INIT_LIST_HEAD(&bdev->bd_holder_disks);
756#endif
757 bdev->bd_bdi = &noop_backing_dev_info;
758 inode_init_once(&ei->vfs_inode);
759 /* Initialize mutex for freeze. */
760 mutex_init(&bdev->bd_fsfreeze_mutex);
761}
762
763static void bdev_evict_inode(struct inode *inode)
764{
765 struct block_device *bdev = &BDEV_I(inode)->bdev;
766 truncate_inode_pages_final(&inode->i_data);
767 invalidate_inode_buffers(inode); /* is it needed here? */
768 clear_inode(inode);
769 spin_lock(&bdev_lock);
770 list_del_init(&bdev->bd_list);
771 spin_unlock(&bdev_lock);
772 /* Detach inode from wb early as bdi_put() may free bdi->wb */
773 inode_detach_wb(inode);
774 if (bdev->bd_bdi != &noop_backing_dev_info) {
775 bdi_put(bdev->bd_bdi);
776 bdev->bd_bdi = &noop_backing_dev_info;
777 }
778}
779
780static const struct super_operations bdev_sops = {
781 .statfs = simple_statfs,
782 .alloc_inode = bdev_alloc_inode,
783 .destroy_inode = bdev_destroy_inode,
784 .drop_inode = generic_delete_inode,
785 .evict_inode = bdev_evict_inode,
786};
787
788static struct dentry *bd_mount(struct file_system_type *fs_type,
789 int flags, const char *dev_name, void *data)
790{
791 struct dentry *dent;
792 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
793 if (!IS_ERR(dent))
794 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
795 return dent;
796}
797
798static struct file_system_type bd_type = {
799 .name = "bdev",
800 .mount = bd_mount,
801 .kill_sb = kill_anon_super,
802};
803
804struct super_block *blockdev_superblock __read_mostly;
805EXPORT_SYMBOL_GPL(blockdev_superblock);
806
807void __init bdev_cache_init(void)
808{
809 int err;
810 static struct vfsmount *bd_mnt;
811
812 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
813 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
814 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
815 init_once);
816 err = register_filesystem(&bd_type);
817 if (err)
818 panic("Cannot register bdev pseudo-fs");
819 bd_mnt = kern_mount(&bd_type);
820 if (IS_ERR(bd_mnt))
821 panic("Cannot create bdev pseudo-fs");
822 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
823}
824
825/*
826 * Most likely _very_ bad one - but then it's hardly critical for small
827 * /dev and can be fixed when somebody will need really large one.
828 * Keep in mind that it will be fed through icache hash function too.
829 */
830static inline unsigned long hash(dev_t dev)
831{
832 return MAJOR(dev)+MINOR(dev);
833}
834
835static int bdev_test(struct inode *inode, void *data)
836{
837 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
838}
839
840static int bdev_set(struct inode *inode, void *data)
841{
842 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
843 return 0;
844}
845
846static LIST_HEAD(all_bdevs);
847
848/*
849 * If there is a bdev inode for this device, unhash it so that it gets evicted
850 * as soon as last inode reference is dropped.
851 */
852void bdev_unhash_inode(dev_t dev)
853{
854 struct inode *inode;
855
856 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
857 if (inode) {
858 remove_inode_hash(inode);
859 iput(inode);
860 }
861}
862
863struct block_device *bdget(dev_t dev)
864{
865 struct block_device *bdev;
866 struct inode *inode;
867
868 inode = iget5_locked(blockdev_superblock, hash(dev),
869 bdev_test, bdev_set, &dev);
870
871 if (!inode)
872 return NULL;
873
874 bdev = &BDEV_I(inode)->bdev;
875
876 if (inode->i_state & I_NEW) {
877 bdev->bd_contains = NULL;
878 bdev->bd_super = NULL;
879 bdev->bd_inode = inode;
880 bdev->bd_block_size = i_blocksize(inode);
881 bdev->bd_part_count = 0;
882 bdev->bd_invalidated = 0;
883 inode->i_mode = S_IFBLK;
884 inode->i_rdev = dev;
885 inode->i_bdev = bdev;
886 inode->i_data.a_ops = &def_blk_aops;
887 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
888 spin_lock(&bdev_lock);
889 list_add(&bdev->bd_list, &all_bdevs);
890 spin_unlock(&bdev_lock);
891 unlock_new_inode(inode);
892 }
893 return bdev;
894}
895
896EXPORT_SYMBOL(bdget);
897
898/**
899 * bdgrab -- Grab a reference to an already referenced block device
900 * @bdev: Block device to grab a reference to.
901 */
902struct block_device *bdgrab(struct block_device *bdev)
903{
904 ihold(bdev->bd_inode);
905 return bdev;
906}
907EXPORT_SYMBOL(bdgrab);
908
909long nr_blockdev_pages(void)
910{
911 struct block_device *bdev;
912 long ret = 0;
913 spin_lock(&bdev_lock);
914 list_for_each_entry(bdev, &all_bdevs, bd_list) {
915 ret += bdev->bd_inode->i_mapping->nrpages;
916 }
917 spin_unlock(&bdev_lock);
918 return ret;
919}
920
921void bdput(struct block_device *bdev)
922{
923 iput(bdev->bd_inode);
924}
925
926EXPORT_SYMBOL(bdput);
927
928static struct block_device *bd_acquire(struct inode *inode)
929{
930 struct block_device *bdev;
931
932 spin_lock(&bdev_lock);
933 bdev = inode->i_bdev;
934 if (bdev && !inode_unhashed(bdev->bd_inode)) {
935 bdgrab(bdev);
936 spin_unlock(&bdev_lock);
937 return bdev;
938 }
939 spin_unlock(&bdev_lock);
940
941 /*
942 * i_bdev references block device inode that was already shut down
943 * (corresponding device got removed). Remove the reference and look
944 * up block device inode again just in case new device got
945 * reestablished under the same device number.
946 */
947 if (bdev)
948 bd_forget(inode);
949
950 bdev = bdget(inode->i_rdev);
951 if (bdev) {
952 spin_lock(&bdev_lock);
953 if (!inode->i_bdev) {
954 /*
955 * We take an additional reference to bd_inode,
956 * and it's released in clear_inode() of inode.
957 * So, we can access it via ->i_mapping always
958 * without igrab().
959 */
960 bdgrab(bdev);
961 inode->i_bdev = bdev;
962 inode->i_mapping = bdev->bd_inode->i_mapping;
963 }
964 spin_unlock(&bdev_lock);
965 }
966 return bdev;
967}
968
969/* Call when you free inode */
970
971void bd_forget(struct inode *inode)
972{
973 struct block_device *bdev = NULL;
974
975 spin_lock(&bdev_lock);
976 if (!sb_is_blkdev_sb(inode->i_sb))
977 bdev = inode->i_bdev;
978 inode->i_bdev = NULL;
979 inode->i_mapping = &inode->i_data;
980 spin_unlock(&bdev_lock);
981
982 if (bdev)
983 bdput(bdev);
984}
985
986/**
987 * bd_may_claim - test whether a block device can be claimed
988 * @bdev: block device of interest
989 * @whole: whole block device containing @bdev, may equal @bdev
990 * @holder: holder trying to claim @bdev
991 *
992 * Test whether @bdev can be claimed by @holder.
993 *
994 * CONTEXT:
995 * spin_lock(&bdev_lock).
996 *
997 * RETURNS:
998 * %true if @bdev can be claimed, %false otherwise.
999 */
1000static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1001 void *holder)
1002{
1003 if (bdev->bd_holder == holder)
1004 return true; /* already a holder */
1005 else if (bdev->bd_holder != NULL)
1006 return false; /* held by someone else */
1007 else if (whole == bdev)
1008 return true; /* is a whole device which isn't held */
1009
1010 else if (whole->bd_holder == bd_may_claim)
1011 return true; /* is a partition of a device that is being partitioned */
1012 else if (whole->bd_holder != NULL)
1013 return false; /* is a partition of a held device */
1014 else
1015 return true; /* is a partition of an un-held device */
1016}
1017
1018/**
1019 * bd_prepare_to_claim - prepare to claim a block device
1020 * @bdev: block device of interest
1021 * @whole: the whole device containing @bdev, may equal @bdev
1022 * @holder: holder trying to claim @bdev
1023 *
1024 * Prepare to claim @bdev. This function fails if @bdev is already
1025 * claimed by another holder and waits if another claiming is in
1026 * progress. This function doesn't actually claim. On successful
1027 * return, the caller has ownership of bd_claiming and bd_holder[s].
1028 *
1029 * CONTEXT:
1030 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1031 * it multiple times.
1032 *
1033 * RETURNS:
1034 * 0 if @bdev can be claimed, -EBUSY otherwise.
1035 */
1036static int bd_prepare_to_claim(struct block_device *bdev,
1037 struct block_device *whole, void *holder)
1038{
1039retry:
1040 /* if someone else claimed, fail */
1041 if (!bd_may_claim(bdev, whole, holder))
1042 return -EBUSY;
1043
1044 /* if claiming is already in progress, wait for it to finish */
1045 if (whole->bd_claiming) {
1046 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1047 DEFINE_WAIT(wait);
1048
1049 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1050 spin_unlock(&bdev_lock);
1051 schedule();
1052 finish_wait(wq, &wait);
1053 spin_lock(&bdev_lock);
1054 goto retry;
1055 }
1056
1057 /* yay, all mine */
1058 return 0;
1059}
1060
1061static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1062{
1063 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1064
1065 if (!disk)
1066 return NULL;
1067 /*
1068 * Now that we hold gendisk reference we make sure bdev we looked up is
1069 * not stale. If it is, it means device got removed and created before
1070 * we looked up gendisk and we fail open in such case. Associating
1071 * unhashed bdev with newly created gendisk could lead to two bdevs
1072 * (and thus two independent caches) being associated with one device
1073 * which is bad.
1074 */
1075 if (inode_unhashed(bdev->bd_inode)) {
1076 put_disk_and_module(disk);
1077 return NULL;
1078 }
1079 return disk;
1080}
1081
1082/**
1083 * bd_start_claiming - start claiming a block device
1084 * @bdev: block device of interest
1085 * @holder: holder trying to claim @bdev
1086 *
1087 * @bdev is about to be opened exclusively. Check @bdev can be opened
1088 * exclusively and mark that an exclusive open is in progress. Each
1089 * successful call to this function must be matched with a call to
1090 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1091 * fail).
1092 *
1093 * This function is used to gain exclusive access to the block device
1094 * without actually causing other exclusive open attempts to fail. It
1095 * should be used when the open sequence itself requires exclusive
1096 * access but may subsequently fail.
1097 *
1098 * CONTEXT:
1099 * Might sleep.
1100 *
1101 * RETURNS:
1102 * Pointer to the block device containing @bdev on success, ERR_PTR()
1103 * value on failure.
1104 */
1105static struct block_device *bd_start_claiming(struct block_device *bdev,
1106 void *holder)
1107{
1108 struct gendisk *disk;
1109 struct block_device *whole;
1110 int partno, err;
1111
1112 might_sleep();
1113
1114 /*
1115 * @bdev might not have been initialized properly yet, look up
1116 * and grab the outer block device the hard way.
1117 */
1118 disk = bdev_get_gendisk(bdev, &partno);
1119 if (!disk)
1120 return ERR_PTR(-ENXIO);
1121
1122 /*
1123 * Normally, @bdev should equal what's returned from bdget_disk()
1124 * if partno is 0; however, some drivers (floppy) use multiple
1125 * bdev's for the same physical device and @bdev may be one of the
1126 * aliases. Keep @bdev if partno is 0. This means claimer
1127 * tracking is broken for those devices but it has always been that
1128 * way.
1129 */
1130 if (partno)
1131 whole = bdget_disk(disk, 0);
1132 else
1133 whole = bdgrab(bdev);
1134
1135 put_disk_and_module(disk);
1136 if (!whole)
1137 return ERR_PTR(-ENOMEM);
1138
1139 /* prepare to claim, if successful, mark claiming in progress */
1140 spin_lock(&bdev_lock);
1141
1142 err = bd_prepare_to_claim(bdev, whole, holder);
1143 if (err == 0) {
1144 whole->bd_claiming = holder;
1145 spin_unlock(&bdev_lock);
1146 return whole;
1147 } else {
1148 spin_unlock(&bdev_lock);
1149 bdput(whole);
1150 return ERR_PTR(err);
1151 }
1152}
1153
1154#ifdef CONFIG_SYSFS
1155struct bd_holder_disk {
1156 struct list_head list;
1157 struct gendisk *disk;
1158 int refcnt;
1159};
1160
1161static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1162 struct gendisk *disk)
1163{
1164 struct bd_holder_disk *holder;
1165
1166 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1167 if (holder->disk == disk)
1168 return holder;
1169 return NULL;
1170}
1171
1172static int add_symlink(struct kobject *from, struct kobject *to)
1173{
1174 return sysfs_create_link(from, to, kobject_name(to));
1175}
1176
1177static void del_symlink(struct kobject *from, struct kobject *to)
1178{
1179 sysfs_remove_link(from, kobject_name(to));
1180}
1181
1182/**
1183 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1184 * @bdev: the claimed slave bdev
1185 * @disk: the holding disk
1186 *
1187 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1188 *
1189 * This functions creates the following sysfs symlinks.
1190 *
1191 * - from "slaves" directory of the holder @disk to the claimed @bdev
1192 * - from "holders" directory of the @bdev to the holder @disk
1193 *
1194 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1195 * passed to bd_link_disk_holder(), then:
1196 *
1197 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1198 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1199 *
1200 * The caller must have claimed @bdev before calling this function and
1201 * ensure that both @bdev and @disk are valid during the creation and
1202 * lifetime of these symlinks.
1203 *
1204 * CONTEXT:
1205 * Might sleep.
1206 *
1207 * RETURNS:
1208 * 0 on success, -errno on failure.
1209 */
1210int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1211{
1212 struct bd_holder_disk *holder;
1213 int ret = 0;
1214
1215 mutex_lock(&bdev->bd_mutex);
1216
1217 WARN_ON_ONCE(!bdev->bd_holder);
1218
1219 /* FIXME: remove the following once add_disk() handles errors */
1220 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1221 goto out_unlock;
1222
1223 holder = bd_find_holder_disk(bdev, disk);
1224 if (holder) {
1225 holder->refcnt++;
1226 goto out_unlock;
1227 }
1228
1229 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1230 if (!holder) {
1231 ret = -ENOMEM;
1232 goto out_unlock;
1233 }
1234
1235 INIT_LIST_HEAD(&holder->list);
1236 holder->disk = disk;
1237 holder->refcnt = 1;
1238
1239 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1240 if (ret)
1241 goto out_free;
1242
1243 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1244 if (ret)
1245 goto out_del;
1246 /*
1247 * bdev could be deleted beneath us which would implicitly destroy
1248 * the holder directory. Hold on to it.
1249 */
1250 kobject_get(bdev->bd_part->holder_dir);
1251
1252 list_add(&holder->list, &bdev->bd_holder_disks);
1253 goto out_unlock;
1254
1255out_del:
1256 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1257out_free:
1258 kfree(holder);
1259out_unlock:
1260 mutex_unlock(&bdev->bd_mutex);
1261 return ret;
1262}
1263EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1264
1265/**
1266 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1267 * @bdev: the calimed slave bdev
1268 * @disk: the holding disk
1269 *
1270 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1271 *
1272 * CONTEXT:
1273 * Might sleep.
1274 */
1275void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1276{
1277 struct bd_holder_disk *holder;
1278
1279 mutex_lock(&bdev->bd_mutex);
1280
1281 holder = bd_find_holder_disk(bdev, disk);
1282
1283 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1284 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1285 del_symlink(bdev->bd_part->holder_dir,
1286 &disk_to_dev(disk)->kobj);
1287 kobject_put(bdev->bd_part->holder_dir);
1288 list_del_init(&holder->list);
1289 kfree(holder);
1290 }
1291
1292 mutex_unlock(&bdev->bd_mutex);
1293}
1294EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1295#endif
1296
1297/**
1298 * flush_disk - invalidates all buffer-cache entries on a disk
1299 *
1300 * @bdev: struct block device to be flushed
1301 * @kill_dirty: flag to guide handling of dirty inodes
1302 *
1303 * Invalidates all buffer-cache entries on a disk. It should be called
1304 * when a disk has been changed -- either by a media change or online
1305 * resize.
1306 */
1307static void flush_disk(struct block_device *bdev, bool kill_dirty)
1308{
1309 if (__invalidate_device(bdev, kill_dirty)) {
1310 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1311 "resized disk %s\n",
1312 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1313 }
1314
1315 if (!bdev->bd_disk)
1316 return;
1317 if (disk_part_scan_enabled(bdev->bd_disk))
1318 bdev->bd_invalidated = 1;
1319}
1320
1321/**
1322 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1323 * @disk: struct gendisk to check
1324 * @bdev: struct bdev to adjust.
1325 *
1326 * This routine checks to see if the bdev size does not match the disk size
1327 * and adjusts it if it differs.
1328 */
1329void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1330{
1331 loff_t disk_size, bdev_size;
1332
1333 disk_size = (loff_t)get_capacity(disk) << 9;
1334 bdev_size = i_size_read(bdev->bd_inode);
1335 if (disk_size != bdev_size) {
1336 printk(KERN_INFO
1337 "%s: detected capacity change from %lld to %lld\n",
1338 disk->disk_name, bdev_size, disk_size);
1339 i_size_write(bdev->bd_inode, disk_size);
1340 flush_disk(bdev, false);
1341 }
1342}
1343EXPORT_SYMBOL(check_disk_size_change);
1344
1345/**
1346 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1347 * @disk: struct gendisk to be revalidated
1348 *
1349 * This routine is a wrapper for lower-level driver's revalidate_disk
1350 * call-backs. It is used to do common pre and post operations needed
1351 * for all revalidate_disk operations.
1352 */
1353int revalidate_disk(struct gendisk *disk)
1354{
1355 struct block_device *bdev;
1356 int ret = 0;
1357
1358 if (disk->fops->revalidate_disk)
1359 ret = disk->fops->revalidate_disk(disk);
1360 bdev = bdget_disk(disk, 0);
1361 if (!bdev)
1362 return ret;
1363
1364 mutex_lock(&bdev->bd_mutex);
1365 check_disk_size_change(disk, bdev);
1366 bdev->bd_invalidated = 0;
1367 mutex_unlock(&bdev->bd_mutex);
1368 bdput(bdev);
1369 return ret;
1370}
1371EXPORT_SYMBOL(revalidate_disk);
1372
1373/*
1374 * This routine checks whether a removable media has been changed,
1375 * and invalidates all buffer-cache-entries in that case. This
1376 * is a relatively slow routine, so we have to try to minimize using
1377 * it. Thus it is called only upon a 'mount' or 'open'. This
1378 * is the best way of combining speed and utility, I think.
1379 * People changing diskettes in the middle of an operation deserve
1380 * to lose :-)
1381 */
1382int check_disk_change(struct block_device *bdev)
1383{
1384 struct gendisk *disk = bdev->bd_disk;
1385 const struct block_device_operations *bdops = disk->fops;
1386 unsigned int events;
1387
1388 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1389 DISK_EVENT_EJECT_REQUEST);
1390 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1391 return 0;
1392
1393 flush_disk(bdev, true);
1394 if (bdops->revalidate_disk)
1395 bdops->revalidate_disk(bdev->bd_disk);
1396 return 1;
1397}
1398
1399EXPORT_SYMBOL(check_disk_change);
1400
1401void bd_set_size(struct block_device *bdev, loff_t size)
1402{
1403 unsigned bsize = bdev_logical_block_size(bdev);
1404
1405 inode_lock(bdev->bd_inode);
1406 i_size_write(bdev->bd_inode, size);
1407 inode_unlock(bdev->bd_inode);
1408 while (bsize < PAGE_SIZE) {
1409 if (size & bsize)
1410 break;
1411 bsize <<= 1;
1412 }
1413 bdev->bd_block_size = bsize;
1414 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1415}
1416EXPORT_SYMBOL(bd_set_size);
1417
1418static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1419
1420/*
1421 * bd_mutex locking:
1422 *
1423 * mutex_lock(part->bd_mutex)
1424 * mutex_lock_nested(whole->bd_mutex, 1)
1425 */
1426
1427static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1428{
1429 struct gendisk *disk;
1430 int ret;
1431 int partno;
1432 int perm = 0;
1433 bool first_open = false;
1434
1435 if (mode & FMODE_READ)
1436 perm |= MAY_READ;
1437 if (mode & FMODE_WRITE)
1438 perm |= MAY_WRITE;
1439 /*
1440 * hooks: /n/, see "layering violations".
1441 */
1442 if (!for_part) {
1443 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1444 if (ret != 0) {
1445 bdput(bdev);
1446 return ret;
1447 }
1448 }
1449
1450 restart:
1451
1452 ret = -ENXIO;
1453 disk = bdev_get_gendisk(bdev, &partno);
1454 if (!disk)
1455 goto out;
1456
1457 disk_block_events(disk);
1458 mutex_lock_nested(&bdev->bd_mutex, for_part);
1459 if (!bdev->bd_openers) {
1460 first_open = true;
1461 bdev->bd_disk = disk;
1462 bdev->bd_queue = disk->queue;
1463 bdev->bd_contains = bdev;
1464 bdev->bd_partno = partno;
1465
1466 if (!partno) {
1467 ret = -ENXIO;
1468 bdev->bd_part = disk_get_part(disk, partno);
1469 if (!bdev->bd_part)
1470 goto out_clear;
1471
1472 ret = 0;
1473 if (disk->fops->open) {
1474 ret = disk->fops->open(bdev, mode);
1475 if (ret == -ERESTARTSYS) {
1476 /* Lost a race with 'disk' being
1477 * deleted, try again.
1478 * See md.c
1479 */
1480 disk_put_part(bdev->bd_part);
1481 bdev->bd_part = NULL;
1482 bdev->bd_disk = NULL;
1483 bdev->bd_queue = NULL;
1484 mutex_unlock(&bdev->bd_mutex);
1485 disk_unblock_events(disk);
1486 put_disk_and_module(disk);
1487 goto restart;
1488 }
1489 }
1490
1491 if (!ret)
1492 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1493
1494 /*
1495 * If the device is invalidated, rescan partition
1496 * if open succeeded or failed with -ENOMEDIUM.
1497 * The latter is necessary to prevent ghost
1498 * partitions on a removed medium.
1499 */
1500 if (bdev->bd_invalidated) {
1501 if (!ret)
1502 rescan_partitions(disk, bdev);
1503 else if (ret == -ENOMEDIUM)
1504 invalidate_partitions(disk, bdev);
1505 }
1506
1507 if (ret)
1508 goto out_clear;
1509 } else {
1510 struct block_device *whole;
1511 whole = bdget_disk(disk, 0);
1512 ret = -ENOMEM;
1513 if (!whole)
1514 goto out_clear;
1515 BUG_ON(for_part);
1516 ret = __blkdev_get(whole, mode, 1);
1517 if (ret)
1518 goto out_clear;
1519 bdev->bd_contains = whole;
1520 bdev->bd_part = disk_get_part(disk, partno);
1521 if (!(disk->flags & GENHD_FL_UP) ||
1522 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1523 ret = -ENXIO;
1524 goto out_clear;
1525 }
1526 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1527 }
1528
1529 if (bdev->bd_bdi == &noop_backing_dev_info)
1530 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1531 } else {
1532 if (bdev->bd_contains == bdev) {
1533 ret = 0;
1534 if (bdev->bd_disk->fops->open)
1535 ret = bdev->bd_disk->fops->open(bdev, mode);
1536 /* the same as first opener case, read comment there */
1537 if (bdev->bd_invalidated) {
1538 if (!ret)
1539 rescan_partitions(bdev->bd_disk, bdev);
1540 else if (ret == -ENOMEDIUM)
1541 invalidate_partitions(bdev->bd_disk, bdev);
1542 }
1543 if (ret)
1544 goto out_unlock_bdev;
1545 }
1546 }
1547 bdev->bd_openers++;
1548 if (for_part)
1549 bdev->bd_part_count++;
1550 mutex_unlock(&bdev->bd_mutex);
1551 disk_unblock_events(disk);
1552 /* only one opener holds refs to the module and disk */
1553 if (!first_open)
1554 put_disk_and_module(disk);
1555 return 0;
1556
1557 out_clear:
1558 disk_put_part(bdev->bd_part);
1559 bdev->bd_disk = NULL;
1560 bdev->bd_part = NULL;
1561 bdev->bd_queue = NULL;
1562 if (bdev != bdev->bd_contains)
1563 __blkdev_put(bdev->bd_contains, mode, 1);
1564 bdev->bd_contains = NULL;
1565 out_unlock_bdev:
1566 mutex_unlock(&bdev->bd_mutex);
1567 disk_unblock_events(disk);
1568 put_disk_and_module(disk);
1569 out:
1570 bdput(bdev);
1571
1572 return ret;
1573}
1574
1575/**
1576 * blkdev_get - open a block device
1577 * @bdev: block_device to open
1578 * @mode: FMODE_* mask
1579 * @holder: exclusive holder identifier
1580 *
1581 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1582 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1583 * @holder is invalid. Exclusive opens may nest for the same @holder.
1584 *
1585 * On success, the reference count of @bdev is unchanged. On failure,
1586 * @bdev is put.
1587 *
1588 * CONTEXT:
1589 * Might sleep.
1590 *
1591 * RETURNS:
1592 * 0 on success, -errno on failure.
1593 */
1594int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1595{
1596 struct block_device *whole = NULL;
1597 int res;
1598
1599 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1600
1601 if ((mode & FMODE_EXCL) && holder) {
1602 whole = bd_start_claiming(bdev, holder);
1603 if (IS_ERR(whole)) {
1604 bdput(bdev);
1605 return PTR_ERR(whole);
1606 }
1607 }
1608
1609 res = __blkdev_get(bdev, mode, 0);
1610
1611 if (whole) {
1612 struct gendisk *disk = whole->bd_disk;
1613
1614 /* finish claiming */
1615 mutex_lock(&bdev->bd_mutex);
1616 spin_lock(&bdev_lock);
1617
1618 if (!res) {
1619 BUG_ON(!bd_may_claim(bdev, whole, holder));
1620 /*
1621 * Note that for a whole device bd_holders
1622 * will be incremented twice, and bd_holder
1623 * will be set to bd_may_claim before being
1624 * set to holder
1625 */
1626 whole->bd_holders++;
1627 whole->bd_holder = bd_may_claim;
1628 bdev->bd_holders++;
1629 bdev->bd_holder = holder;
1630 }
1631
1632 /* tell others that we're done */
1633 BUG_ON(whole->bd_claiming != holder);
1634 whole->bd_claiming = NULL;
1635 wake_up_bit(&whole->bd_claiming, 0);
1636
1637 spin_unlock(&bdev_lock);
1638
1639 /*
1640 * Block event polling for write claims if requested. Any
1641 * write holder makes the write_holder state stick until
1642 * all are released. This is good enough and tracking
1643 * individual writeable reference is too fragile given the
1644 * way @mode is used in blkdev_get/put().
1645 */
1646 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1647 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1648 bdev->bd_write_holder = true;
1649 disk_block_events(disk);
1650 }
1651
1652 mutex_unlock(&bdev->bd_mutex);
1653 bdput(whole);
1654 }
1655
1656 return res;
1657}
1658EXPORT_SYMBOL(blkdev_get);
1659
1660/**
1661 * blkdev_get_by_path - open a block device by name
1662 * @path: path to the block device to open
1663 * @mode: FMODE_* mask
1664 * @holder: exclusive holder identifier
1665 *
1666 * Open the blockdevice described by the device file at @path. @mode
1667 * and @holder are identical to blkdev_get().
1668 *
1669 * On success, the returned block_device has reference count of one.
1670 *
1671 * CONTEXT:
1672 * Might sleep.
1673 *
1674 * RETURNS:
1675 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1676 */
1677struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1678 void *holder)
1679{
1680 struct block_device *bdev;
1681 int err;
1682
1683 bdev = lookup_bdev(path);
1684 if (IS_ERR(bdev))
1685 return bdev;
1686
1687 err = blkdev_get(bdev, mode, holder);
1688 if (err)
1689 return ERR_PTR(err);
1690
1691 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1692 blkdev_put(bdev, mode);
1693 return ERR_PTR(-EACCES);
1694 }
1695
1696 return bdev;
1697}
1698EXPORT_SYMBOL(blkdev_get_by_path);
1699
1700/**
1701 * blkdev_get_by_dev - open a block device by device number
1702 * @dev: device number of block device to open
1703 * @mode: FMODE_* mask
1704 * @holder: exclusive holder identifier
1705 *
1706 * Open the blockdevice described by device number @dev. @mode and
1707 * @holder are identical to blkdev_get().
1708 *
1709 * Use it ONLY if you really do not have anything better - i.e. when
1710 * you are behind a truly sucky interface and all you are given is a
1711 * device number. _Never_ to be used for internal purposes. If you
1712 * ever need it - reconsider your API.
1713 *
1714 * On success, the returned block_device has reference count of one.
1715 *
1716 * CONTEXT:
1717 * Might sleep.
1718 *
1719 * RETURNS:
1720 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1721 */
1722struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1723{
1724 struct block_device *bdev;
1725 int err;
1726
1727 bdev = bdget(dev);
1728 if (!bdev)
1729 return ERR_PTR(-ENOMEM);
1730
1731 err = blkdev_get(bdev, mode, holder);
1732 if (err)
1733 return ERR_PTR(err);
1734
1735 return bdev;
1736}
1737EXPORT_SYMBOL(blkdev_get_by_dev);
1738
1739static int blkdev_open(struct inode * inode, struct file * filp)
1740{
1741 struct block_device *bdev;
1742
1743 /*
1744 * Preserve backwards compatibility and allow large file access
1745 * even if userspace doesn't ask for it explicitly. Some mkfs
1746 * binary needs it. We might want to drop this workaround
1747 * during an unstable branch.
1748 */
1749 filp->f_flags |= O_LARGEFILE;
1750
1751 filp->f_mode |= FMODE_NOWAIT;
1752
1753 if (filp->f_flags & O_NDELAY)
1754 filp->f_mode |= FMODE_NDELAY;
1755 if (filp->f_flags & O_EXCL)
1756 filp->f_mode |= FMODE_EXCL;
1757 if ((filp->f_flags & O_ACCMODE) == 3)
1758 filp->f_mode |= FMODE_WRITE_IOCTL;
1759
1760 bdev = bd_acquire(inode);
1761 if (bdev == NULL)
1762 return -ENOMEM;
1763
1764 filp->f_mapping = bdev->bd_inode->i_mapping;
1765 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1766
1767 return blkdev_get(bdev, filp->f_mode, filp);
1768}
1769
1770static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1771{
1772 struct gendisk *disk = bdev->bd_disk;
1773 struct block_device *victim = NULL;
1774
1775 mutex_lock_nested(&bdev->bd_mutex, for_part);
1776 if (for_part)
1777 bdev->bd_part_count--;
1778
1779 if (!--bdev->bd_openers) {
1780 WARN_ON_ONCE(bdev->bd_holders);
1781 sync_blockdev(bdev);
1782 kill_bdev(bdev);
1783
1784 bdev_write_inode(bdev);
1785 }
1786 if (bdev->bd_contains == bdev) {
1787 if (disk->fops->release)
1788 disk->fops->release(disk, mode);
1789 }
1790 if (!bdev->bd_openers) {
1791 disk_put_part(bdev->bd_part);
1792 bdev->bd_part = NULL;
1793 bdev->bd_disk = NULL;
1794 if (bdev != bdev->bd_contains)
1795 victim = bdev->bd_contains;
1796 bdev->bd_contains = NULL;
1797
1798 put_disk_and_module(disk);
1799 }
1800 mutex_unlock(&bdev->bd_mutex);
1801 bdput(bdev);
1802 if (victim)
1803 __blkdev_put(victim, mode, 1);
1804}
1805
1806void blkdev_put(struct block_device *bdev, fmode_t mode)
1807{
1808 mutex_lock(&bdev->bd_mutex);
1809
1810 if (mode & FMODE_EXCL) {
1811 bool bdev_free;
1812
1813 /*
1814 * Release a claim on the device. The holder fields
1815 * are protected with bdev_lock. bd_mutex is to
1816 * synchronize disk_holder unlinking.
1817 */
1818 spin_lock(&bdev_lock);
1819
1820 WARN_ON_ONCE(--bdev->bd_holders < 0);
1821 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1822
1823 /* bd_contains might point to self, check in a separate step */
1824 if ((bdev_free = !bdev->bd_holders))
1825 bdev->bd_holder = NULL;
1826 if (!bdev->bd_contains->bd_holders)
1827 bdev->bd_contains->bd_holder = NULL;
1828
1829 spin_unlock(&bdev_lock);
1830
1831 /*
1832 * If this was the last claim, remove holder link and
1833 * unblock evpoll if it was a write holder.
1834 */
1835 if (bdev_free && bdev->bd_write_holder) {
1836 disk_unblock_events(bdev->bd_disk);
1837 bdev->bd_write_holder = false;
1838 }
1839 }
1840
1841 /*
1842 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1843 * event. This is to ensure detection of media removal commanded
1844 * from userland - e.g. eject(1).
1845 */
1846 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1847
1848 mutex_unlock(&bdev->bd_mutex);
1849
1850 __blkdev_put(bdev, mode, 0);
1851}
1852EXPORT_SYMBOL(blkdev_put);
1853
1854static int blkdev_close(struct inode * inode, struct file * filp)
1855{
1856 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1857 blkdev_put(bdev, filp->f_mode);
1858 return 0;
1859}
1860
1861static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1862{
1863 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1864 fmode_t mode = file->f_mode;
1865
1866 /*
1867 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1868 * to updated it before every ioctl.
1869 */
1870 if (file->f_flags & O_NDELAY)
1871 mode |= FMODE_NDELAY;
1872 else
1873 mode &= ~FMODE_NDELAY;
1874
1875 return blkdev_ioctl(bdev, mode, cmd, arg);
1876}
1877
1878/*
1879 * Write data to the block device. Only intended for the block device itself
1880 * and the raw driver which basically is a fake block device.
1881 *
1882 * Does not take i_mutex for the write and thus is not for general purpose
1883 * use.
1884 */
1885ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1886{
1887 struct file *file = iocb->ki_filp;
1888 struct inode *bd_inode = bdev_file_inode(file);
1889 loff_t size = i_size_read(bd_inode);
1890 struct blk_plug plug;
1891 ssize_t ret;
1892
1893 if (bdev_read_only(I_BDEV(bd_inode)))
1894 return -EPERM;
1895
1896 if (!iov_iter_count(from))
1897 return 0;
1898
1899 if (iocb->ki_pos >= size)
1900 return -ENOSPC;
1901
1902 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1903 return -EOPNOTSUPP;
1904
1905 iov_iter_truncate(from, size - iocb->ki_pos);
1906
1907 blk_start_plug(&plug);
1908 ret = __generic_file_write_iter(iocb, from);
1909 if (ret > 0)
1910 ret = generic_write_sync(iocb, ret);
1911 blk_finish_plug(&plug);
1912 return ret;
1913}
1914EXPORT_SYMBOL_GPL(blkdev_write_iter);
1915
1916ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1917{
1918 struct file *file = iocb->ki_filp;
1919 struct inode *bd_inode = bdev_file_inode(file);
1920 loff_t size = i_size_read(bd_inode);
1921 loff_t pos = iocb->ki_pos;
1922
1923 if (pos >= size)
1924 return 0;
1925
1926 size -= pos;
1927 iov_iter_truncate(to, size);
1928 return generic_file_read_iter(iocb, to);
1929}
1930EXPORT_SYMBOL_GPL(blkdev_read_iter);
1931
1932/*
1933 * Try to release a page associated with block device when the system
1934 * is under memory pressure.
1935 */
1936static int blkdev_releasepage(struct page *page, gfp_t wait)
1937{
1938 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1939
1940 if (super && super->s_op->bdev_try_to_free_page)
1941 return super->s_op->bdev_try_to_free_page(super, page, wait);
1942
1943 return try_to_free_buffers(page);
1944}
1945
1946static int blkdev_writepages(struct address_space *mapping,
1947 struct writeback_control *wbc)
1948{
1949 if (dax_mapping(mapping)) {
1950 struct block_device *bdev = I_BDEV(mapping->host);
1951
1952 return dax_writeback_mapping_range(mapping, bdev, wbc);
1953 }
1954 return generic_writepages(mapping, wbc);
1955}
1956
1957static const struct address_space_operations def_blk_aops = {
1958 .readpage = blkdev_readpage,
1959 .readpages = blkdev_readpages,
1960 .writepage = blkdev_writepage,
1961 .write_begin = blkdev_write_begin,
1962 .write_end = blkdev_write_end,
1963 .writepages = blkdev_writepages,
1964 .releasepage = blkdev_releasepage,
1965 .direct_IO = blkdev_direct_IO,
1966 .is_dirty_writeback = buffer_check_dirty_writeback,
1967};
1968
1969#define BLKDEV_FALLOC_FL_SUPPORTED \
1970 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1971 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1972
1973static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1974 loff_t len)
1975{
1976 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1977 struct address_space *mapping;
1978 loff_t end = start + len - 1;
1979 loff_t isize;
1980 int error;
1981
1982 /* Fail if we don't recognize the flags. */
1983 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1984 return -EOPNOTSUPP;
1985
1986 /* Don't go off the end of the device. */
1987 isize = i_size_read(bdev->bd_inode);
1988 if (start >= isize)
1989 return -EINVAL;
1990 if (end >= isize) {
1991 if (mode & FALLOC_FL_KEEP_SIZE) {
1992 len = isize - start;
1993 end = start + len - 1;
1994 } else
1995 return -EINVAL;
1996 }
1997
1998 /*
1999 * Don't allow IO that isn't aligned to logical block size.
2000 */
2001 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2002 return -EINVAL;
2003
2004 /* Invalidate the page cache, including dirty pages. */
2005 mapping = bdev->bd_inode->i_mapping;
2006 truncate_inode_pages_range(mapping, start, end);
2007
2008 switch (mode) {
2009 case FALLOC_FL_ZERO_RANGE:
2010 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2011 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2012 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2013 break;
2014 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2015 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2016 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2017 break;
2018 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2019 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2020 GFP_KERNEL, 0);
2021 break;
2022 default:
2023 return -EOPNOTSUPP;
2024 }
2025 if (error)
2026 return error;
2027
2028 /*
2029 * Invalidate again; if someone wandered in and dirtied a page,
2030 * the caller will be given -EBUSY. The third argument is
2031 * inclusive, so the rounding here is safe.
2032 */
2033 return invalidate_inode_pages2_range(mapping,
2034 start >> PAGE_SHIFT,
2035 end >> PAGE_SHIFT);
2036}
2037
2038const struct file_operations def_blk_fops = {
2039 .open = blkdev_open,
2040 .release = blkdev_close,
2041 .llseek = block_llseek,
2042 .read_iter = blkdev_read_iter,
2043 .write_iter = blkdev_write_iter,
2044 .mmap = generic_file_mmap,
2045 .fsync = blkdev_fsync,
2046 .unlocked_ioctl = block_ioctl,
2047#ifdef CONFIG_COMPAT
2048 .compat_ioctl = compat_blkdev_ioctl,
2049#endif
2050 .splice_read = generic_file_splice_read,
2051 .splice_write = iter_file_splice_write,
2052 .fallocate = blkdev_fallocate,
2053};
2054
2055int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2056{
2057 int res;
2058 mm_segment_t old_fs = get_fs();
2059 set_fs(KERNEL_DS);
2060 res = blkdev_ioctl(bdev, 0, cmd, arg);
2061 set_fs(old_fs);
2062 return res;
2063}
2064
2065EXPORT_SYMBOL(ioctl_by_bdev);
2066
2067/**
2068 * lookup_bdev - lookup a struct block_device by name
2069 * @pathname: special file representing the block device
2070 *
2071 * Get a reference to the blockdevice at @pathname in the current
2072 * namespace if possible and return it. Return ERR_PTR(error)
2073 * otherwise.
2074 */
2075struct block_device *lookup_bdev(const char *pathname)
2076{
2077 struct block_device *bdev;
2078 struct inode *inode;
2079 struct path path;
2080 int error;
2081
2082 if (!pathname || !*pathname)
2083 return ERR_PTR(-EINVAL);
2084
2085 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2086 if (error)
2087 return ERR_PTR(error);
2088
2089 inode = d_backing_inode(path.dentry);
2090 error = -ENOTBLK;
2091 if (!S_ISBLK(inode->i_mode))
2092 goto fail;
2093 error = -EACCES;
2094 if (!may_open_dev(&path))
2095 goto fail;
2096 error = -ENOMEM;
2097 bdev = bd_acquire(inode);
2098 if (!bdev)
2099 goto fail;
2100out:
2101 path_put(&path);
2102 return bdev;
2103fail:
2104 bdev = ERR_PTR(error);
2105 goto out;
2106}
2107EXPORT_SYMBOL(lookup_bdev);
2108
2109int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2110{
2111 struct super_block *sb = get_super(bdev);
2112 int res = 0;
2113
2114 if (sb) {
2115 /*
2116 * no need to lock the super, get_super holds the
2117 * read mutex so the filesystem cannot go away
2118 * under us (->put_super runs with the write lock
2119 * hold).
2120 */
2121 shrink_dcache_sb(sb);
2122 res = invalidate_inodes(sb, kill_dirty);
2123 drop_super(sb);
2124 }
2125 invalidate_bdev(bdev);
2126 return res;
2127}
2128EXPORT_SYMBOL(__invalidate_device);
2129
2130void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2131{
2132 struct inode *inode, *old_inode = NULL;
2133
2134 spin_lock(&blockdev_superblock->s_inode_list_lock);
2135 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2136 struct address_space *mapping = inode->i_mapping;
2137 struct block_device *bdev;
2138
2139 spin_lock(&inode->i_lock);
2140 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2141 mapping->nrpages == 0) {
2142 spin_unlock(&inode->i_lock);
2143 continue;
2144 }
2145 __iget(inode);
2146 spin_unlock(&inode->i_lock);
2147 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2148 /*
2149 * We hold a reference to 'inode' so it couldn't have been
2150 * removed from s_inodes list while we dropped the
2151 * s_inode_list_lock We cannot iput the inode now as we can
2152 * be holding the last reference and we cannot iput it under
2153 * s_inode_list_lock. So we keep the reference and iput it
2154 * later.
2155 */
2156 iput(old_inode);
2157 old_inode = inode;
2158 bdev = I_BDEV(inode);
2159
2160 mutex_lock(&bdev->bd_mutex);
2161 if (bdev->bd_openers)
2162 func(bdev, arg);
2163 mutex_unlock(&bdev->bd_mutex);
2164
2165 spin_lock(&blockdev_superblock->s_inode_list_lock);
2166 }
2167 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2168 iput(old_inode);
2169}