projects/linux·blob v5.8-rc3 · refs

2270 lines · 56.7 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/block_dev.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Copyright (C) 2001 Andrea Arcangeli <[email protected]> SuSE
7 */
8
9#include <linux/init.h>
10#include <linux/mm.h>
11#include <linux/fcntl.h>
12#include <linux/slab.h>
13#include <linux/kmod.h>
14#include <linux/major.h>
15#include <linux/device_cgroup.h>
16#include <linux/highmem.h>
17#include <linux/blkdev.h>
18#include <linux/backing-dev.h>
19#include <linux/module.h>
20#include <linux/blkpg.h>
21#include <linux/magic.h>
22#include <linux/buffer_head.h>
23#include <linux/swap.h>
24#include <linux/pagevec.h>
25#include <linux/writeback.h>
26#include <linux/mpage.h>
27#include <linux/mount.h>
28#include <linux/pseudo_fs.h>
29#include <linux/uio.h>
30#include <linux/namei.h>
31#include <linux/log2.h>
32#include <linux/cleancache.h>
33#include <linux/task_io_accounting_ops.h>
34#include <linux/falloc.h>
35#include <linux/uaccess.h>
36#include <linux/suspend.h>
37#include "internal.h"
38
39struct bdev_inode {
40 struct block_device bdev;
41 struct inode vfs_inode;
42};
43
44static const struct address_space_operations def_blk_aops;
45
46static inline struct bdev_inode *BDEV_I(struct inode *inode)
47{
48 return container_of(inode, struct bdev_inode, vfs_inode);
49}
50
51struct block_device *I_BDEV(struct inode *inode)
52{
53 return &BDEV_I(inode)->bdev;
54}
55EXPORT_SYMBOL(I_BDEV);
56
57static void bdev_write_inode(struct block_device *bdev)
58{
59 struct inode *inode = bdev->bd_inode;
60 int ret;
61
62 spin_lock(&inode->i_lock);
63 while (inode->i_state & I_DIRTY) {
64 spin_unlock(&inode->i_lock);
65 ret = write_inode_now(inode, true);
66 if (ret) {
67 char name[BDEVNAME_SIZE];
68 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
69 "for block device %s (err=%d).\n",
70 bdevname(bdev, name), ret);
71 }
72 spin_lock(&inode->i_lock);
73 }
74 spin_unlock(&inode->i_lock);
75}
76
77/* Kill _all_ buffers and pagecache , dirty or not.. */
78static void kill_bdev(struct block_device *bdev)
79{
80 struct address_space *mapping = bdev->bd_inode->i_mapping;
81
82 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
83 return;
84
85 invalidate_bh_lrus();
86 truncate_inode_pages(mapping, 0);
87}
88
89/* Invalidate clean unused buffers and pagecache. */
90void invalidate_bdev(struct block_device *bdev)
91{
92 struct address_space *mapping = bdev->bd_inode->i_mapping;
93
94 if (mapping->nrpages) {
95 invalidate_bh_lrus();
96 lru_add_drain_all(); /* make sure all lru add caches are flushed */
97 invalidate_mapping_pages(mapping, 0, -1);
98 }
99 /* 99% of the time, we don't need to flush the cleancache on the bdev.
100 * But, for the strange corners, lets be cautious
101 */
102 cleancache_invalidate_inode(mapping);
103}
104EXPORT_SYMBOL(invalidate_bdev);
105
106static void set_init_blocksize(struct block_device *bdev)
107{
108 unsigned bsize = bdev_logical_block_size(bdev);
109 loff_t size = i_size_read(bdev->bd_inode);
110
111 while (bsize < PAGE_SIZE) {
112 if (size & bsize)
113 break;
114 bsize <<= 1;
115 }
116 bdev->bd_block_size = bsize;
117 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
118}
119
120int set_blocksize(struct block_device *bdev, int size)
121{
122 /* Size must be a power of two, and between 512 and PAGE_SIZE */
123 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
124 return -EINVAL;
125
126 /* Size cannot be smaller than the size supported by the device */
127 if (size < bdev_logical_block_size(bdev))
128 return -EINVAL;
129
130 /* Don't change the size if it is same as current */
131 if (bdev->bd_block_size != size) {
132 sync_blockdev(bdev);
133 bdev->bd_block_size = size;
134 bdev->bd_inode->i_blkbits = blksize_bits(size);
135 kill_bdev(bdev);
136 }
137 return 0;
138}
139
140EXPORT_SYMBOL(set_blocksize);
141
142int sb_set_blocksize(struct super_block *sb, int size)
143{
144 if (set_blocksize(sb->s_bdev, size))
145 return 0;
146 /* If we get here, we know size is power of two
147 * and it's value is between 512 and PAGE_SIZE */
148 sb->s_blocksize = size;
149 sb->s_blocksize_bits = blksize_bits(size);
150 return sb->s_blocksize;
151}
152
153EXPORT_SYMBOL(sb_set_blocksize);
154
155int sb_min_blocksize(struct super_block *sb, int size)
156{
157 int minsize = bdev_logical_block_size(sb->s_bdev);
158 if (size < minsize)
159 size = minsize;
160 return sb_set_blocksize(sb, size);
161}
162
163EXPORT_SYMBOL(sb_min_blocksize);
164
165static int
166blkdev_get_block(struct inode *inode, sector_t iblock,
167 struct buffer_head *bh, int create)
168{
169 bh->b_bdev = I_BDEV(inode);
170 bh->b_blocknr = iblock;
171 set_buffer_mapped(bh);
172 return 0;
173}
174
175static struct inode *bdev_file_inode(struct file *file)
176{
177 return file->f_mapping->host;
178}
179
180static unsigned int dio_bio_write_op(struct kiocb *iocb)
181{
182 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
183
184 /* avoid the need for a I/O completion work item */
185 if (iocb->ki_flags & IOCB_DSYNC)
186 op |= REQ_FUA;
187 return op;
188}
189
190#define DIO_INLINE_BIO_VECS 4
191
192static void blkdev_bio_end_io_simple(struct bio *bio)
193{
194 struct task_struct *waiter = bio->bi_private;
195
196 WRITE_ONCE(bio->bi_private, NULL);
197 blk_wake_io_task(waiter);
198}
199
200static ssize_t
201__blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
202 int nr_pages)
203{
204 struct file *file = iocb->ki_filp;
205 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
206 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
207 loff_t pos = iocb->ki_pos;
208 bool should_dirty = false;
209 struct bio bio;
210 ssize_t ret;
211 blk_qc_t qc;
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_array(nr_pages, sizeof(struct bio_vec),
221 GFP_KERNEL);
222 if (!vecs)
223 return -ENOMEM;
224 }
225
226 bio_init(&bio, vecs, nr_pages);
227 bio_set_dev(&bio, bdev);
228 bio.bi_iter.bi_sector = pos >> 9;
229 bio.bi_write_hint = iocb->ki_hint;
230 bio.bi_private = current;
231 bio.bi_end_io = blkdev_bio_end_io_simple;
232 bio.bi_ioprio = iocb->ki_ioprio;
233
234 ret = bio_iov_iter_get_pages(&bio, iter);
235 if (unlikely(ret))
236 goto out;
237 ret = bio.bi_iter.bi_size;
238
239 if (iov_iter_rw(iter) == READ) {
240 bio.bi_opf = REQ_OP_READ;
241 if (iter_is_iovec(iter))
242 should_dirty = true;
243 } else {
244 bio.bi_opf = dio_bio_write_op(iocb);
245 task_io_account_write(ret);
246 }
247 if (iocb->ki_flags & IOCB_HIPRI)
248 bio_set_polled(&bio, iocb);
249
250 qc = submit_bio(&bio);
251 for (;;) {
252 set_current_state(TASK_UNINTERRUPTIBLE);
253 if (!READ_ONCE(bio.bi_private))
254 break;
255 if (!(iocb->ki_flags & IOCB_HIPRI) ||
256 !blk_poll(bdev_get_queue(bdev), qc, true))
257 blk_io_schedule();
258 }
259 __set_current_state(TASK_RUNNING);
260
261 bio_release_pages(&bio, should_dirty);
262 if (unlikely(bio.bi_status))
263 ret = blk_status_to_errno(bio.bi_status);
264
265out:
266 if (vecs != inline_vecs)
267 kfree(vecs);
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;
288
289static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
290{
291 struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
292 struct request_queue *q = bdev_get_queue(bdev);
293
294 return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
295}
296
297static void blkdev_bio_end_io(struct bio *bio)
298{
299 struct blkdev_dio *dio = bio->bi_private;
300 bool should_dirty = dio->should_dirty;
301
302 if (bio->bi_status && !dio->bio.bi_status)
303 dio->bio.bi_status = bio->bi_status;
304
305 if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
306 if (!dio->is_sync) {
307 struct kiocb *iocb = dio->iocb;
308 ssize_t ret;
309
310 if (likely(!dio->bio.bi_status)) {
311 ret = dio->size;
312 iocb->ki_pos += ret;
313 } else {
314 ret = blk_status_to_errno(dio->bio.bi_status);
315 }
316
317 dio->iocb->ki_complete(iocb, ret, 0);
318 if (dio->multi_bio)
319 bio_put(&dio->bio);
320 } else {
321 struct task_struct *waiter = dio->waiter;
322
323 WRITE_ONCE(dio->waiter, NULL);
324 blk_wake_io_task(waiter);
325 }
326 }
327
328 if (should_dirty) {
329 bio_check_pages_dirty(bio);
330 } else {
331 bio_release_pages(bio, false);
332 bio_put(bio);
333 }
334}
335
336static ssize_t
337__blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
338{
339 struct file *file = iocb->ki_filp;
340 struct inode *inode = bdev_file_inode(file);
341 struct block_device *bdev = I_BDEV(inode);
342 struct blk_plug plug;
343 struct blkdev_dio *dio;
344 struct bio *bio;
345 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
346 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
347 loff_t pos = iocb->ki_pos;
348 blk_qc_t qc = BLK_QC_T_NONE;
349 int ret = 0;
350
351 if ((pos | iov_iter_alignment(iter)) &
352 (bdev_logical_block_size(bdev) - 1))
353 return -EINVAL;
354
355 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
356
357 dio = container_of(bio, struct blkdev_dio, bio);
358 dio->is_sync = is_sync = is_sync_kiocb(iocb);
359 if (dio->is_sync) {
360 dio->waiter = current;
361 bio_get(bio);
362 } else {
363 dio->iocb = iocb;
364 }
365
366 dio->size = 0;
367 dio->multi_bio = false;
368 dio->should_dirty = is_read && iter_is_iovec(iter);
369
370 /*
371 * Don't plug for HIPRI/polled IO, as those should go straight
372 * to issue
373 */
374 if (!is_poll)
375 blk_start_plug(&plug);
376
377 for (;;) {
378 bio_set_dev(bio, bdev);
379 bio->bi_iter.bi_sector = pos >> 9;
380 bio->bi_write_hint = iocb->ki_hint;
381 bio->bi_private = dio;
382 bio->bi_end_io = blkdev_bio_end_io;
383 bio->bi_ioprio = iocb->ki_ioprio;
384
385 ret = bio_iov_iter_get_pages(bio, iter);
386 if (unlikely(ret)) {
387 bio->bi_status = BLK_STS_IOERR;
388 bio_endio(bio);
389 break;
390 }
391
392 if (is_read) {
393 bio->bi_opf = REQ_OP_READ;
394 if (dio->should_dirty)
395 bio_set_pages_dirty(bio);
396 } else {
397 bio->bi_opf = dio_bio_write_op(iocb);
398 task_io_account_write(bio->bi_iter.bi_size);
399 }
400
401 dio->size += bio->bi_iter.bi_size;
402 pos += bio->bi_iter.bi_size;
403
404 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
405 if (!nr_pages) {
406 bool polled = false;
407
408 if (iocb->ki_flags & IOCB_HIPRI) {
409 bio_set_polled(bio, iocb);
410 polled = true;
411 }
412
413 qc = submit_bio(bio);
414
415 if (polled)
416 WRITE_ONCE(iocb->ki_cookie, qc);
417 break;
418 }
419
420 if (!dio->multi_bio) {
421 /*
422 * AIO needs an extra reference to ensure the dio
423 * structure which is embedded into the first bio
424 * stays around.
425 */
426 if (!is_sync)
427 bio_get(bio);
428 dio->multi_bio = true;
429 atomic_set(&dio->ref, 2);
430 } else {
431 atomic_inc(&dio->ref);
432 }
433
434 submit_bio(bio);
435 bio = bio_alloc(GFP_KERNEL, nr_pages);
436 }
437
438 if (!is_poll)
439 blk_finish_plug(&plug);
440
441 if (!is_sync)
442 return -EIOCBQUEUED;
443
444 for (;;) {
445 set_current_state(TASK_UNINTERRUPTIBLE);
446 if (!READ_ONCE(dio->waiter))
447 break;
448
449 if (!(iocb->ki_flags & IOCB_HIPRI) ||
450 !blk_poll(bdev_get_queue(bdev), qc, true))
451 blk_io_schedule();
452 }
453 __set_current_state(TASK_RUNNING);
454
455 if (!ret)
456 ret = blk_status_to_errno(dio->bio.bi_status);
457 if (likely(!ret))
458 ret = dio->size;
459
460 bio_put(&dio->bio);
461 return ret;
462}
463
464static ssize_t
465blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
466{
467 int nr_pages;
468
469 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
470 if (!nr_pages)
471 return 0;
472 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
473 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
474
475 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
476}
477
478static __init int blkdev_init(void)
479{
480 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
481}
482module_init(blkdev_init);
483
484int __sync_blockdev(struct block_device *bdev, int wait)
485{
486 if (!bdev)
487 return 0;
488 if (!wait)
489 return filemap_flush(bdev->bd_inode->i_mapping);
490 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
491}
492
493/*
494 * Write out and wait upon all the dirty data associated with a block
495 * device via its mapping. Does not take the superblock lock.
496 */
497int sync_blockdev(struct block_device *bdev)
498{
499 return __sync_blockdev(bdev, 1);
500}
501EXPORT_SYMBOL(sync_blockdev);
502
503/*
504 * Write out and wait upon all dirty data associated with this
505 * device. Filesystem data as well as the underlying block
506 * device. Takes the superblock lock.
507 */
508int fsync_bdev(struct block_device *bdev)
509{
510 struct super_block *sb = get_super(bdev);
511 if (sb) {
512 int res = sync_filesystem(sb);
513 drop_super(sb);
514 return res;
515 }
516 return sync_blockdev(bdev);
517}
518EXPORT_SYMBOL(fsync_bdev);
519
520/**
521 * freeze_bdev -- lock a filesystem and force it into a consistent state
522 * @bdev: blockdevice to lock
523 *
524 * If a superblock is found on this device, we take the s_umount semaphore
525 * on it to make sure nobody unmounts until the snapshot creation is done.
526 * The reference counter (bd_fsfreeze_count) guarantees that only the last
527 * unfreeze process can unfreeze the frozen filesystem actually when multiple
528 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
529 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
530 * actually.
531 */
532struct super_block *freeze_bdev(struct block_device *bdev)
533{
534 struct super_block *sb;
535 int error = 0;
536
537 mutex_lock(&bdev->bd_fsfreeze_mutex);
538 if (++bdev->bd_fsfreeze_count > 1) {
539 /*
540 * We don't even need to grab a reference - the first call
541 * to freeze_bdev grab an active reference and only the last
542 * thaw_bdev drops it.
543 */
544 sb = get_super(bdev);
545 if (sb)
546 drop_super(sb);
547 mutex_unlock(&bdev->bd_fsfreeze_mutex);
548 return sb;
549 }
550
551 sb = get_active_super(bdev);
552 if (!sb)
553 goto out;
554 if (sb->s_op->freeze_super)
555 error = sb->s_op->freeze_super(sb);
556 else
557 error = freeze_super(sb);
558 if (error) {
559 deactivate_super(sb);
560 bdev->bd_fsfreeze_count--;
561 mutex_unlock(&bdev->bd_fsfreeze_mutex);
562 return ERR_PTR(error);
563 }
564 deactivate_super(sb);
565 out:
566 sync_blockdev(bdev);
567 mutex_unlock(&bdev->bd_fsfreeze_mutex);
568 return sb; /* thaw_bdev releases s->s_umount */
569}
570EXPORT_SYMBOL(freeze_bdev);
571
572/**
573 * thaw_bdev -- unlock filesystem
574 * @bdev: blockdevice to unlock
575 * @sb: associated superblock
576 *
577 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
578 */
579int thaw_bdev(struct block_device *bdev, struct super_block *sb)
580{
581 int error = -EINVAL;
582
583 mutex_lock(&bdev->bd_fsfreeze_mutex);
584 if (!bdev->bd_fsfreeze_count)
585 goto out;
586
587 error = 0;
588 if (--bdev->bd_fsfreeze_count > 0)
589 goto out;
590
591 if (!sb)
592 goto out;
593
594 if (sb->s_op->thaw_super)
595 error = sb->s_op->thaw_super(sb);
596 else
597 error = thaw_super(sb);
598 if (error)
599 bdev->bd_fsfreeze_count++;
600out:
601 mutex_unlock(&bdev->bd_fsfreeze_mutex);
602 return error;
603}
604EXPORT_SYMBOL(thaw_bdev);
605
606static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
607{
608 return block_write_full_page(page, blkdev_get_block, wbc);
609}
610
611static int blkdev_readpage(struct file * file, struct page * page)
612{
613 return block_read_full_page(page, blkdev_get_block);
614}
615
616static void blkdev_readahead(struct readahead_control *rac)
617{
618 mpage_readahead(rac, blkdev_get_block);
619}
620
621static int blkdev_write_begin(struct file *file, struct address_space *mapping,
622 loff_t pos, unsigned len, unsigned flags,
623 struct page **pagep, void **fsdata)
624{
625 return block_write_begin(mapping, pos, len, flags, pagep,
626 blkdev_get_block);
627}
628
629static int blkdev_write_end(struct file *file, struct address_space *mapping,
630 loff_t pos, unsigned len, unsigned copied,
631 struct page *page, void *fsdata)
632{
633 int ret;
634 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
635
636 unlock_page(page);
637 put_page(page);
638
639 return ret;
640}
641
642/*
643 * private llseek:
644 * for a block special file file_inode(file)->i_size is zero
645 * so we compute the size by hand (just as in block_read/write above)
646 */
647static loff_t block_llseek(struct file *file, loff_t offset, int whence)
648{
649 struct inode *bd_inode = bdev_file_inode(file);
650 loff_t retval;
651
652 inode_lock(bd_inode);
653 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
654 inode_unlock(bd_inode);
655 return retval;
656}
657
658int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
659{
660 struct inode *bd_inode = bdev_file_inode(filp);
661 struct block_device *bdev = I_BDEV(bd_inode);
662 int error;
663
664 error = file_write_and_wait_range(filp, start, end);
665 if (error)
666 return error;
667
668 /*
669 * There is no need to serialise calls to blkdev_issue_flush with
670 * i_mutex and doing so causes performance issues with concurrent
671 * O_SYNC writers to a block device.
672 */
673 error = blkdev_issue_flush(bdev, GFP_KERNEL);
674 if (error == -EOPNOTSUPP)
675 error = 0;
676
677 return error;
678}
679EXPORT_SYMBOL(blkdev_fsync);
680
681/**
682 * bdev_read_page() - Start reading a page from a block device
683 * @bdev: The device to read the page from
684 * @sector: The offset on the device to read the page to (need not be aligned)
685 * @page: The page to read
686 *
687 * On entry, the page should be locked. It will be unlocked when the page
688 * has been read. If the block driver implements rw_page synchronously,
689 * that will be true on exit from this function, but it need not be.
690 *
691 * Errors returned by this function are usually "soft", eg out of memory, or
692 * queue full; callers should try a different route to read this page rather
693 * than propagate an error back up the stack.
694 *
695 * Return: negative errno if an error occurs, 0 if submission was successful.
696 */
697int bdev_read_page(struct block_device *bdev, sector_t sector,
698 struct page *page)
699{
700 const struct block_device_operations *ops = bdev->bd_disk->fops;
701 int result = -EOPNOTSUPP;
702
703 if (!ops->rw_page || bdev_get_integrity(bdev))
704 return result;
705
706 result = blk_queue_enter(bdev->bd_queue, 0);
707 if (result)
708 return result;
709 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
710 REQ_OP_READ);
711 blk_queue_exit(bdev->bd_queue);
712 return result;
713}
714
715/**
716 * bdev_write_page() - Start writing a page to a block device
717 * @bdev: The device to write the page to
718 * @sector: The offset on the device to write the page to (need not be aligned)
719 * @page: The page to write
720 * @wbc: The writeback_control for the write
721 *
722 * On entry, the page should be locked and not currently under writeback.
723 * On exit, if the write started successfully, the page will be unlocked and
724 * under writeback. If the write failed already (eg the driver failed to
725 * queue the page to the device), the page will still be locked. If the
726 * caller is a ->writepage implementation, it will need to unlock the page.
727 *
728 * Errors returned by this function are usually "soft", eg out of memory, or
729 * queue full; callers should try a different route to write this page rather
730 * than propagate an error back up the stack.
731 *
732 * Return: negative errno if an error occurs, 0 if submission was successful.
733 */
734int bdev_write_page(struct block_device *bdev, sector_t sector,
735 struct page *page, struct writeback_control *wbc)
736{
737 int result;
738 const struct block_device_operations *ops = bdev->bd_disk->fops;
739
740 if (!ops->rw_page || bdev_get_integrity(bdev))
741 return -EOPNOTSUPP;
742 result = blk_queue_enter(bdev->bd_queue, 0);
743 if (result)
744 return result;
745
746 set_page_writeback(page);
747 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
748 REQ_OP_WRITE);
749 if (result) {
750 end_page_writeback(page);
751 } else {
752 clean_page_buffers(page);
753 unlock_page(page);
754 }
755 blk_queue_exit(bdev->bd_queue);
756 return result;
757}
758
759/*
760 * pseudo-fs
761 */
762
763static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
764static struct kmem_cache * bdev_cachep __read_mostly;
765
766static struct inode *bdev_alloc_inode(struct super_block *sb)
767{
768 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
769 if (!ei)
770 return NULL;
771 return &ei->vfs_inode;
772}
773
774static void bdev_free_inode(struct inode *inode)
775{
776 kmem_cache_free(bdev_cachep, BDEV_I(inode));
777}
778
779static void init_once(void *foo)
780{
781 struct bdev_inode *ei = (struct bdev_inode *) foo;
782 struct block_device *bdev = &ei->bdev;
783
784 memset(bdev, 0, sizeof(*bdev));
785 mutex_init(&bdev->bd_mutex);
786 INIT_LIST_HEAD(&bdev->bd_list);
787#ifdef CONFIG_SYSFS
788 INIT_LIST_HEAD(&bdev->bd_holder_disks);
789#endif
790 bdev->bd_bdi = &noop_backing_dev_info;
791 inode_init_once(&ei->vfs_inode);
792 /* Initialize mutex for freeze. */
793 mutex_init(&bdev->bd_fsfreeze_mutex);
794}
795
796static void bdev_evict_inode(struct inode *inode)
797{
798 struct block_device *bdev = &BDEV_I(inode)->bdev;
799 truncate_inode_pages_final(&inode->i_data);
800 invalidate_inode_buffers(inode); /* is it needed here? */
801 clear_inode(inode);
802 spin_lock(&bdev_lock);
803 list_del_init(&bdev->bd_list);
804 spin_unlock(&bdev_lock);
805 /* Detach inode from wb early as bdi_put() may free bdi->wb */
806 inode_detach_wb(inode);
807 if (bdev->bd_bdi != &noop_backing_dev_info) {
808 bdi_put(bdev->bd_bdi);
809 bdev->bd_bdi = &noop_backing_dev_info;
810 }
811}
812
813static const struct super_operations bdev_sops = {
814 .statfs = simple_statfs,
815 .alloc_inode = bdev_alloc_inode,
816 .free_inode = bdev_free_inode,
817 .drop_inode = generic_delete_inode,
818 .evict_inode = bdev_evict_inode,
819};
820
821static int bd_init_fs_context(struct fs_context *fc)
822{
823 struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
824 if (!ctx)
825 return -ENOMEM;
826 fc->s_iflags |= SB_I_CGROUPWB;
827 ctx->ops = &bdev_sops;
828 return 0;
829}
830
831static struct file_system_type bd_type = {
832 .name = "bdev",
833 .init_fs_context = bd_init_fs_context,
834 .kill_sb = kill_anon_super,
835};
836
837struct super_block *blockdev_superblock __read_mostly;
838EXPORT_SYMBOL_GPL(blockdev_superblock);
839
840void __init bdev_cache_init(void)
841{
842 int err;
843 static struct vfsmount *bd_mnt;
844
845 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
846 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
847 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
848 init_once);
849 err = register_filesystem(&bd_type);
850 if (err)
851 panic("Cannot register bdev pseudo-fs");
852 bd_mnt = kern_mount(&bd_type);
853 if (IS_ERR(bd_mnt))
854 panic("Cannot create bdev pseudo-fs");
855 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
856}
857
858/*
859 * Most likely _very_ bad one - but then it's hardly critical for small
860 * /dev and can be fixed when somebody will need really large one.
861 * Keep in mind that it will be fed through icache hash function too.
862 */
863static inline unsigned long hash(dev_t dev)
864{
865 return MAJOR(dev)+MINOR(dev);
866}
867
868static int bdev_test(struct inode *inode, void *data)
869{
870 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
871}
872
873static int bdev_set(struct inode *inode, void *data)
874{
875 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
876 return 0;
877}
878
879static LIST_HEAD(all_bdevs);
880
881struct block_device *bdget(dev_t dev)
882{
883 struct block_device *bdev;
884 struct inode *inode;
885
886 inode = iget5_locked(blockdev_superblock, hash(dev),
887 bdev_test, bdev_set, &dev);
888
889 if (!inode)
890 return NULL;
891
892 bdev = &BDEV_I(inode)->bdev;
893
894 if (inode->i_state & I_NEW) {
895 bdev->bd_contains = NULL;
896 bdev->bd_super = NULL;
897 bdev->bd_inode = inode;
898 bdev->bd_block_size = i_blocksize(inode);
899 bdev->bd_part_count = 0;
900 bdev->bd_invalidated = 0;
901 inode->i_mode = S_IFBLK;
902 inode->i_rdev = dev;
903 inode->i_bdev = bdev;
904 inode->i_data.a_ops = &def_blk_aops;
905 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
906 spin_lock(&bdev_lock);
907 list_add(&bdev->bd_list, &all_bdevs);
908 spin_unlock(&bdev_lock);
909 unlock_new_inode(inode);
910 }
911 return bdev;
912}
913
914EXPORT_SYMBOL(bdget);
915
916/**
917 * bdgrab -- Grab a reference to an already referenced block device
918 * @bdev: Block device to grab a reference to.
919 */
920struct block_device *bdgrab(struct block_device *bdev)
921{
922 ihold(bdev->bd_inode);
923 return bdev;
924}
925EXPORT_SYMBOL(bdgrab);
926
927long nr_blockdev_pages(void)
928{
929 struct block_device *bdev;
930 long ret = 0;
931 spin_lock(&bdev_lock);
932 list_for_each_entry(bdev, &all_bdevs, bd_list) {
933 ret += bdev->bd_inode->i_mapping->nrpages;
934 }
935 spin_unlock(&bdev_lock);
936 return ret;
937}
938
939void bdput(struct block_device *bdev)
940{
941 iput(bdev->bd_inode);
942}
943
944EXPORT_SYMBOL(bdput);
945
946static struct block_device *bd_acquire(struct inode *inode)
947{
948 struct block_device *bdev;
949
950 spin_lock(&bdev_lock);
951 bdev = inode->i_bdev;
952 if (bdev && !inode_unhashed(bdev->bd_inode)) {
953 bdgrab(bdev);
954 spin_unlock(&bdev_lock);
955 return bdev;
956 }
957 spin_unlock(&bdev_lock);
958
959 /*
960 * i_bdev references block device inode that was already shut down
961 * (corresponding device got removed). Remove the reference and look
962 * up block device inode again just in case new device got
963 * reestablished under the same device number.
964 */
965 if (bdev)
966 bd_forget(inode);
967
968 bdev = bdget(inode->i_rdev);
969 if (bdev) {
970 spin_lock(&bdev_lock);
971 if (!inode->i_bdev) {
972 /*
973 * We take an additional reference to bd_inode,
974 * and it's released in clear_inode() of inode.
975 * So, we can access it via ->i_mapping always
976 * without igrab().
977 */
978 bdgrab(bdev);
979 inode->i_bdev = bdev;
980 inode->i_mapping = bdev->bd_inode->i_mapping;
981 }
982 spin_unlock(&bdev_lock);
983 }
984 return bdev;
985}
986
987/* Call when you free inode */
988
989void bd_forget(struct inode *inode)
990{
991 struct block_device *bdev = NULL;
992
993 spin_lock(&bdev_lock);
994 if (!sb_is_blkdev_sb(inode->i_sb))
995 bdev = inode->i_bdev;
996 inode->i_bdev = NULL;
997 inode->i_mapping = &inode->i_data;
998 spin_unlock(&bdev_lock);
999
1000 if (bdev)
1001 bdput(bdev);
1002}
1003
1004/**
1005 * bd_may_claim - test whether a block device can be claimed
1006 * @bdev: block device of interest
1007 * @whole: whole block device containing @bdev, may equal @bdev
1008 * @holder: holder trying to claim @bdev
1009 *
1010 * Test whether @bdev can be claimed by @holder.
1011 *
1012 * CONTEXT:
1013 * spin_lock(&bdev_lock).
1014 *
1015 * RETURNS:
1016 * %true if @bdev can be claimed, %false otherwise.
1017 */
1018static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1019 void *holder)
1020{
1021 if (bdev->bd_holder == holder)
1022 return true; /* already a holder */
1023 else if (bdev->bd_holder != NULL)
1024 return false; /* held by someone else */
1025 else if (whole == bdev)
1026 return true; /* is a whole device which isn't held */
1027
1028 else if (whole->bd_holder == bd_may_claim)
1029 return true; /* is a partition of a device that is being partitioned */
1030 else if (whole->bd_holder != NULL)
1031 return false; /* is a partition of a held device */
1032 else
1033 return true; /* is a partition of an un-held device */
1034}
1035
1036/**
1037 * bd_prepare_to_claim - prepare to claim a block device
1038 * @bdev: block device of interest
1039 * @whole: the whole device containing @bdev, may equal @bdev
1040 * @holder: holder trying to claim @bdev
1041 *
1042 * Prepare to claim @bdev. This function fails if @bdev is already
1043 * claimed by another holder and waits if another claiming is in
1044 * progress. This function doesn't actually claim. On successful
1045 * return, the caller has ownership of bd_claiming and bd_holder[s].
1046 *
1047 * CONTEXT:
1048 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1049 * it multiple times.
1050 *
1051 * RETURNS:
1052 * 0 if @bdev can be claimed, -EBUSY otherwise.
1053 */
1054static int bd_prepare_to_claim(struct block_device *bdev,
1055 struct block_device *whole, void *holder)
1056{
1057retry:
1058 /* if someone else claimed, fail */
1059 if (!bd_may_claim(bdev, whole, holder))
1060 return -EBUSY;
1061
1062 /* if claiming is already in progress, wait for it to finish */
1063 if (whole->bd_claiming) {
1064 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1065 DEFINE_WAIT(wait);
1066
1067 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1068 spin_unlock(&bdev_lock);
1069 schedule();
1070 finish_wait(wq, &wait);
1071 spin_lock(&bdev_lock);
1072 goto retry;
1073 }
1074
1075 /* yay, all mine */
1076 return 0;
1077}
1078
1079static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1080{
1081 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1082
1083 if (!disk)
1084 return NULL;
1085 /*
1086 * Now that we hold gendisk reference we make sure bdev we looked up is
1087 * not stale. If it is, it means device got removed and created before
1088 * we looked up gendisk and we fail open in such case. Associating
1089 * unhashed bdev with newly created gendisk could lead to two bdevs
1090 * (and thus two independent caches) being associated with one device
1091 * which is bad.
1092 */
1093 if (inode_unhashed(bdev->bd_inode)) {
1094 put_disk_and_module(disk);
1095 return NULL;
1096 }
1097 return disk;
1098}
1099
1100/**
1101 * bd_start_claiming - start claiming a block device
1102 * @bdev: block device of interest
1103 * @holder: holder trying to claim @bdev
1104 *
1105 * @bdev is about to be opened exclusively. Check @bdev can be opened
1106 * exclusively and mark that an exclusive open is in progress. Each
1107 * successful call to this function must be matched with a call to
1108 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1109 * fail).
1110 *
1111 * This function is used to gain exclusive access to the block device
1112 * without actually causing other exclusive open attempts to fail. It
1113 * should be used when the open sequence itself requires exclusive
1114 * access but may subsequently fail.
1115 *
1116 * CONTEXT:
1117 * Might sleep.
1118 *
1119 * RETURNS:
1120 * Pointer to the block device containing @bdev on success, ERR_PTR()
1121 * value on failure.
1122 */
1123struct block_device *bd_start_claiming(struct block_device *bdev, void *holder)
1124{
1125 struct gendisk *disk;
1126 struct block_device *whole;
1127 int partno, err;
1128
1129 might_sleep();
1130
1131 /*
1132 * @bdev might not have been initialized properly yet, look up
1133 * and grab the outer block device the hard way.
1134 */
1135 disk = bdev_get_gendisk(bdev, &partno);
1136 if (!disk)
1137 return ERR_PTR(-ENXIO);
1138
1139 /*
1140 * Normally, @bdev should equal what's returned from bdget_disk()
1141 * if partno is 0; however, some drivers (floppy) use multiple
1142 * bdev's for the same physical device and @bdev may be one of the
1143 * aliases. Keep @bdev if partno is 0. This means claimer
1144 * tracking is broken for those devices but it has always been that
1145 * way.
1146 */
1147 if (partno)
1148 whole = bdget_disk(disk, 0);
1149 else
1150 whole = bdgrab(bdev);
1151
1152 put_disk_and_module(disk);
1153 if (!whole)
1154 return ERR_PTR(-ENOMEM);
1155
1156 /* prepare to claim, if successful, mark claiming in progress */
1157 spin_lock(&bdev_lock);
1158
1159 err = bd_prepare_to_claim(bdev, whole, holder);
1160 if (err == 0) {
1161 whole->bd_claiming = holder;
1162 spin_unlock(&bdev_lock);
1163 return whole;
1164 } else {
1165 spin_unlock(&bdev_lock);
1166 bdput(whole);
1167 return ERR_PTR(err);
1168 }
1169}
1170EXPORT_SYMBOL(bd_start_claiming);
1171
1172static void bd_clear_claiming(struct block_device *whole, void *holder)
1173{
1174 lockdep_assert_held(&bdev_lock);
1175 /* tell others that we're done */
1176 BUG_ON(whole->bd_claiming != holder);
1177 whole->bd_claiming = NULL;
1178 wake_up_bit(&whole->bd_claiming, 0);
1179}
1180
1181/**
1182 * bd_finish_claiming - finish claiming of a block device
1183 * @bdev: block device of interest
1184 * @whole: whole block device (returned from bd_start_claiming())
1185 * @holder: holder that has claimed @bdev
1186 *
1187 * Finish exclusive open of a block device. Mark the device as exlusively
1188 * open by the holder and wake up all waiters for exclusive open to finish.
1189 */
1190void bd_finish_claiming(struct block_device *bdev, struct block_device *whole,
1191 void *holder)
1192{
1193 spin_lock(&bdev_lock);
1194 BUG_ON(!bd_may_claim(bdev, whole, holder));
1195 /*
1196 * Note that for a whole device bd_holders will be incremented twice,
1197 * and bd_holder will be set to bd_may_claim before being set to holder
1198 */
1199 whole->bd_holders++;
1200 whole->bd_holder = bd_may_claim;
1201 bdev->bd_holders++;
1202 bdev->bd_holder = holder;
1203 bd_clear_claiming(whole, holder);
1204 spin_unlock(&bdev_lock);
1205}
1206EXPORT_SYMBOL(bd_finish_claiming);
1207
1208/**
1209 * bd_abort_claiming - abort claiming of a block device
1210 * @bdev: block device of interest
1211 * @whole: whole block device (returned from bd_start_claiming())
1212 * @holder: holder that has claimed @bdev
1213 *
1214 * Abort claiming of a block device when the exclusive open failed. This can be
1215 * also used when exclusive open is not actually desired and we just needed
1216 * to block other exclusive openers for a while.
1217 */
1218void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
1219 void *holder)
1220{
1221 spin_lock(&bdev_lock);
1222 bd_clear_claiming(whole, holder);
1223 spin_unlock(&bdev_lock);
1224}
1225EXPORT_SYMBOL(bd_abort_claiming);
1226
1227#ifdef CONFIG_SYSFS
1228struct bd_holder_disk {
1229 struct list_head list;
1230 struct gendisk *disk;
1231 int refcnt;
1232};
1233
1234static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1235 struct gendisk *disk)
1236{
1237 struct bd_holder_disk *holder;
1238
1239 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1240 if (holder->disk == disk)
1241 return holder;
1242 return NULL;
1243}
1244
1245static int add_symlink(struct kobject *from, struct kobject *to)
1246{
1247 return sysfs_create_link(from, to, kobject_name(to));
1248}
1249
1250static void del_symlink(struct kobject *from, struct kobject *to)
1251{
1252 sysfs_remove_link(from, kobject_name(to));
1253}
1254
1255/**
1256 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1257 * @bdev: the claimed slave bdev
1258 * @disk: the holding disk
1259 *
1260 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1261 *
1262 * This functions creates the following sysfs symlinks.
1263 *
1264 * - from "slaves" directory of the holder @disk to the claimed @bdev
1265 * - from "holders" directory of the @bdev to the holder @disk
1266 *
1267 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1268 * passed to bd_link_disk_holder(), then:
1269 *
1270 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1271 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1272 *
1273 * The caller must have claimed @bdev before calling this function and
1274 * ensure that both @bdev and @disk are valid during the creation and
1275 * lifetime of these symlinks.
1276 *
1277 * CONTEXT:
1278 * Might sleep.
1279 *
1280 * RETURNS:
1281 * 0 on success, -errno on failure.
1282 */
1283int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1284{
1285 struct bd_holder_disk *holder;
1286 int ret = 0;
1287
1288 mutex_lock(&bdev->bd_mutex);
1289
1290 WARN_ON_ONCE(!bdev->bd_holder);
1291
1292 /* FIXME: remove the following once add_disk() handles errors */
1293 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1294 goto out_unlock;
1295
1296 holder = bd_find_holder_disk(bdev, disk);
1297 if (holder) {
1298 holder->refcnt++;
1299 goto out_unlock;
1300 }
1301
1302 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1303 if (!holder) {
1304 ret = -ENOMEM;
1305 goto out_unlock;
1306 }
1307
1308 INIT_LIST_HEAD(&holder->list);
1309 holder->disk = disk;
1310 holder->refcnt = 1;
1311
1312 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1313 if (ret)
1314 goto out_free;
1315
1316 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1317 if (ret)
1318 goto out_del;
1319 /*
1320 * bdev could be deleted beneath us which would implicitly destroy
1321 * the holder directory. Hold on to it.
1322 */
1323 kobject_get(bdev->bd_part->holder_dir);
1324
1325 list_add(&holder->list, &bdev->bd_holder_disks);
1326 goto out_unlock;
1327
1328out_del:
1329 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1330out_free:
1331 kfree(holder);
1332out_unlock:
1333 mutex_unlock(&bdev->bd_mutex);
1334 return ret;
1335}
1336EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1337
1338/**
1339 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1340 * @bdev: the calimed slave bdev
1341 * @disk: the holding disk
1342 *
1343 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1344 *
1345 * CONTEXT:
1346 * Might sleep.
1347 */
1348void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1349{
1350 struct bd_holder_disk *holder;
1351
1352 mutex_lock(&bdev->bd_mutex);
1353
1354 holder = bd_find_holder_disk(bdev, disk);
1355
1356 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1357 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1358 del_symlink(bdev->bd_part->holder_dir,
1359 &disk_to_dev(disk)->kobj);
1360 kobject_put(bdev->bd_part->holder_dir);
1361 list_del_init(&holder->list);
1362 kfree(holder);
1363 }
1364
1365 mutex_unlock(&bdev->bd_mutex);
1366}
1367EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1368#endif
1369
1370/**
1371 * flush_disk - invalidates all buffer-cache entries on a disk
1372 *
1373 * @bdev: struct block device to be flushed
1374 * @kill_dirty: flag to guide handling of dirty inodes
1375 *
1376 * Invalidates all buffer-cache entries on a disk. It should be called
1377 * when a disk has been changed -- either by a media change or online
1378 * resize.
1379 */
1380static void flush_disk(struct block_device *bdev, bool kill_dirty)
1381{
1382 if (__invalidate_device(bdev, kill_dirty)) {
1383 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1384 "resized disk %s\n",
1385 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1386 }
1387 bdev->bd_invalidated = 1;
1388}
1389
1390/**
1391 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1392 * @disk: struct gendisk to check
1393 * @bdev: struct bdev to adjust.
1394 * @verbose: if %true log a message about a size change if there is any
1395 *
1396 * This routine checks to see if the bdev size does not match the disk size
1397 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1398 * are freed.
1399 */
1400static void check_disk_size_change(struct gendisk *disk,
1401 struct block_device *bdev, bool verbose)
1402{
1403 loff_t disk_size, bdev_size;
1404
1405 disk_size = (loff_t)get_capacity(disk) << 9;
1406 bdev_size = i_size_read(bdev->bd_inode);
1407 if (disk_size != bdev_size) {
1408 if (verbose) {
1409 printk(KERN_INFO
1410 "%s: detected capacity change from %lld to %lld\n",
1411 disk->disk_name, bdev_size, disk_size);
1412 }
1413 i_size_write(bdev->bd_inode, disk_size);
1414 if (bdev_size > disk_size)
1415 flush_disk(bdev, false);
1416 }
1417 bdev->bd_invalidated = 0;
1418}
1419
1420/**
1421 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1422 * @disk: struct gendisk to be revalidated
1423 *
1424 * This routine is a wrapper for lower-level driver's revalidate_disk
1425 * call-backs. It is used to do common pre and post operations needed
1426 * for all revalidate_disk operations.
1427 */
1428int revalidate_disk(struct gendisk *disk)
1429{
1430 int ret = 0;
1431
1432 if (disk->fops->revalidate_disk)
1433 ret = disk->fops->revalidate_disk(disk);
1434
1435 /*
1436 * Hidden disks don't have associated bdev so there's no point in
1437 * revalidating it.
1438 */
1439 if (!(disk->flags & GENHD_FL_HIDDEN)) {
1440 struct block_device *bdev = bdget_disk(disk, 0);
1441
1442 if (!bdev)
1443 return ret;
1444
1445 mutex_lock(&bdev->bd_mutex);
1446 check_disk_size_change(disk, bdev, ret == 0);
1447 mutex_unlock(&bdev->bd_mutex);
1448 bdput(bdev);
1449 }
1450 return ret;
1451}
1452EXPORT_SYMBOL(revalidate_disk);
1453
1454/*
1455 * This routine checks whether a removable media has been changed,
1456 * and invalidates all buffer-cache-entries in that case. This
1457 * is a relatively slow routine, so we have to try to minimize using
1458 * it. Thus it is called only upon a 'mount' or 'open'. This
1459 * is the best way of combining speed and utility, I think.
1460 * People changing diskettes in the middle of an operation deserve
1461 * to lose :-)
1462 */
1463int check_disk_change(struct block_device *bdev)
1464{
1465 struct gendisk *disk = bdev->bd_disk;
1466 const struct block_device_operations *bdops = disk->fops;
1467 unsigned int events;
1468
1469 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1470 DISK_EVENT_EJECT_REQUEST);
1471 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1472 return 0;
1473
1474 flush_disk(bdev, true);
1475 if (bdops->revalidate_disk)
1476 bdops->revalidate_disk(bdev->bd_disk);
1477 return 1;
1478}
1479
1480EXPORT_SYMBOL(check_disk_change);
1481
1482void bd_set_size(struct block_device *bdev, loff_t size)
1483{
1484 inode_lock(bdev->bd_inode);
1485 i_size_write(bdev->bd_inode, size);
1486 inode_unlock(bdev->bd_inode);
1487}
1488EXPORT_SYMBOL(bd_set_size);
1489
1490static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1491
1492int bdev_disk_changed(struct block_device *bdev, bool invalidate)
1493{
1494 struct gendisk *disk = bdev->bd_disk;
1495 int ret;
1496
1497 lockdep_assert_held(&bdev->bd_mutex);
1498
1499rescan:
1500 ret = blk_drop_partitions(bdev);
1501 if (ret)
1502 return ret;
1503
1504 /*
1505 * Historically we only set the capacity to zero for devices that
1506 * support partitions (independ of actually having partitions created).
1507 * Doing that is rather inconsistent, but changing it broke legacy
1508 * udisks polling for legacy ide-cdrom devices. Use the crude check
1509 * below to get the sane behavior for most device while not breaking
1510 * userspace for this particular setup.
1511 */
1512 if (invalidate) {
1513 if (disk_part_scan_enabled(disk) ||
1514 !(disk->flags & GENHD_FL_REMOVABLE))
1515 set_capacity(disk, 0);
1516 } else {
1517 if (disk->fops->revalidate_disk)
1518 disk->fops->revalidate_disk(disk);
1519 }
1520
1521 check_disk_size_change(disk, bdev, !invalidate);
1522
1523 if (get_capacity(disk)) {
1524 ret = blk_add_partitions(disk, bdev);
1525 if (ret == -EAGAIN)
1526 goto rescan;
1527 } else if (invalidate) {
1528 /*
1529 * Tell userspace that the media / partition table may have
1530 * changed.
1531 */
1532 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
1533 }
1534
1535 return ret;
1536}
1537/*
1538 * Only exported for for loop and dasd for historic reasons. Don't use in new
1539 * code!
1540 */
1541EXPORT_SYMBOL_GPL(bdev_disk_changed);
1542
1543/*
1544 * bd_mutex locking:
1545 *
1546 * mutex_lock(part->bd_mutex)
1547 * mutex_lock_nested(whole->bd_mutex, 1)
1548 */
1549
1550static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1551{
1552 struct gendisk *disk;
1553 int ret;
1554 int partno;
1555 int perm = 0;
1556 bool first_open = false;
1557
1558 if (mode & FMODE_READ)
1559 perm |= MAY_READ;
1560 if (mode & FMODE_WRITE)
1561 perm |= MAY_WRITE;
1562 /*
1563 * hooks: /n/, see "layering violations".
1564 */
1565 if (!for_part) {
1566 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1567 if (ret != 0)
1568 return ret;
1569 }
1570
1571 restart:
1572
1573 ret = -ENXIO;
1574 disk = bdev_get_gendisk(bdev, &partno);
1575 if (!disk)
1576 goto out;
1577
1578 disk_block_events(disk);
1579 mutex_lock_nested(&bdev->bd_mutex, for_part);
1580 if (!bdev->bd_openers) {
1581 first_open = true;
1582 bdev->bd_disk = disk;
1583 bdev->bd_queue = disk->queue;
1584 bdev->bd_contains = bdev;
1585 bdev->bd_partno = partno;
1586
1587 if (!partno) {
1588 ret = -ENXIO;
1589 bdev->bd_part = disk_get_part(disk, partno);
1590 if (!bdev->bd_part)
1591 goto out_clear;
1592
1593 ret = 0;
1594 if (disk->fops->open) {
1595 ret = disk->fops->open(bdev, mode);
1596 if (ret == -ERESTARTSYS) {
1597 /* Lost a race with 'disk' being
1598 * deleted, try again.
1599 * See md.c
1600 */
1601 disk_put_part(bdev->bd_part);
1602 bdev->bd_part = NULL;
1603 bdev->bd_disk = NULL;
1604 bdev->bd_queue = NULL;
1605 mutex_unlock(&bdev->bd_mutex);
1606 disk_unblock_events(disk);
1607 put_disk_and_module(disk);
1608 goto restart;
1609 }
1610 }
1611
1612 if (!ret) {
1613 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1614 set_init_blocksize(bdev);
1615 }
1616
1617 /*
1618 * If the device is invalidated, rescan partition
1619 * if open succeeded or failed with -ENOMEDIUM.
1620 * The latter is necessary to prevent ghost
1621 * partitions on a removed medium.
1622 */
1623 if (bdev->bd_invalidated &&
1624 (!ret || ret == -ENOMEDIUM))
1625 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1626
1627 if (ret)
1628 goto out_clear;
1629 } else {
1630 struct block_device *whole;
1631 whole = bdget_disk(disk, 0);
1632 ret = -ENOMEM;
1633 if (!whole)
1634 goto out_clear;
1635 BUG_ON(for_part);
1636 ret = __blkdev_get(whole, mode, 1);
1637 if (ret) {
1638 bdput(whole);
1639 goto out_clear;
1640 }
1641 bdev->bd_contains = whole;
1642 bdev->bd_part = disk_get_part(disk, partno);
1643 if (!(disk->flags & GENHD_FL_UP) ||
1644 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1645 ret = -ENXIO;
1646 goto out_clear;
1647 }
1648 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1649 set_init_blocksize(bdev);
1650 }
1651
1652 if (bdev->bd_bdi == &noop_backing_dev_info)
1653 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1654 } else {
1655 if (bdev->bd_contains == bdev) {
1656 ret = 0;
1657 if (bdev->bd_disk->fops->open)
1658 ret = bdev->bd_disk->fops->open(bdev, mode);
1659 /* the same as first opener case, read comment there */
1660 if (bdev->bd_invalidated &&
1661 (!ret || ret == -ENOMEDIUM))
1662 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1663 if (ret)
1664 goto out_unlock_bdev;
1665 }
1666 }
1667 bdev->bd_openers++;
1668 if (for_part)
1669 bdev->bd_part_count++;
1670 mutex_unlock(&bdev->bd_mutex);
1671 disk_unblock_events(disk);
1672 /* only one opener holds refs to the module and disk */
1673 if (!first_open)
1674 put_disk_and_module(disk);
1675 return 0;
1676
1677 out_clear:
1678 disk_put_part(bdev->bd_part);
1679 bdev->bd_disk = NULL;
1680 bdev->bd_part = NULL;
1681 bdev->bd_queue = NULL;
1682 if (bdev != bdev->bd_contains)
1683 __blkdev_put(bdev->bd_contains, mode, 1);
1684 bdev->bd_contains = NULL;
1685 out_unlock_bdev:
1686 mutex_unlock(&bdev->bd_mutex);
1687 disk_unblock_events(disk);
1688 put_disk_and_module(disk);
1689 out:
1690
1691 return ret;
1692}
1693
1694/**
1695 * blkdev_get - open a block device
1696 * @bdev: block_device to open
1697 * @mode: FMODE_* mask
1698 * @holder: exclusive holder identifier
1699 *
1700 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1701 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1702 * @holder is invalid. Exclusive opens may nest for the same @holder.
1703 *
1704 * On success, the reference count of @bdev is unchanged. On failure,
1705 * @bdev is put.
1706 *
1707 * CONTEXT:
1708 * Might sleep.
1709 *
1710 * RETURNS:
1711 * 0 on success, -errno on failure.
1712 */
1713int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1714{
1715 struct block_device *whole = NULL;
1716 int res;
1717
1718 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1719
1720 if ((mode & FMODE_EXCL) && holder) {
1721 whole = bd_start_claiming(bdev, holder);
1722 if (IS_ERR(whole)) {
1723 bdput(bdev);
1724 return PTR_ERR(whole);
1725 }
1726 }
1727
1728 res = __blkdev_get(bdev, mode, 0);
1729
1730 if (whole) {
1731 struct gendisk *disk = whole->bd_disk;
1732
1733 /* finish claiming */
1734 mutex_lock(&bdev->bd_mutex);
1735 if (!res)
1736 bd_finish_claiming(bdev, whole, holder);
1737 else
1738 bd_abort_claiming(bdev, whole, holder);
1739 /*
1740 * Block event polling for write claims if requested. Any
1741 * write holder makes the write_holder state stick until
1742 * all are released. This is good enough and tracking
1743 * individual writeable reference is too fragile given the
1744 * way @mode is used in blkdev_get/put().
1745 */
1746 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1747 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1748 bdev->bd_write_holder = true;
1749 disk_block_events(disk);
1750 }
1751
1752 mutex_unlock(&bdev->bd_mutex);
1753 bdput(whole);
1754 }
1755
1756 if (res)
1757 bdput(bdev);
1758
1759 return res;
1760}
1761EXPORT_SYMBOL(blkdev_get);
1762
1763/**
1764 * blkdev_get_by_path - open a block device by name
1765 * @path: path to the block device to open
1766 * @mode: FMODE_* mask
1767 * @holder: exclusive holder identifier
1768 *
1769 * Open the blockdevice described by the device file at @path. @mode
1770 * and @holder are identical to blkdev_get().
1771 *
1772 * On success, the returned block_device has reference count of one.
1773 *
1774 * CONTEXT:
1775 * Might sleep.
1776 *
1777 * RETURNS:
1778 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1779 */
1780struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1781 void *holder)
1782{
1783 struct block_device *bdev;
1784 int err;
1785
1786 bdev = lookup_bdev(path);
1787 if (IS_ERR(bdev))
1788 return bdev;
1789
1790 err = blkdev_get(bdev, mode, holder);
1791 if (err)
1792 return ERR_PTR(err);
1793
1794 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1795 blkdev_put(bdev, mode);
1796 return ERR_PTR(-EACCES);
1797 }
1798
1799 return bdev;
1800}
1801EXPORT_SYMBOL(blkdev_get_by_path);
1802
1803/**
1804 * blkdev_get_by_dev - open a block device by device number
1805 * @dev: device number of block device to open
1806 * @mode: FMODE_* mask
1807 * @holder: exclusive holder identifier
1808 *
1809 * Open the blockdevice described by device number @dev. @mode and
1810 * @holder are identical to blkdev_get().
1811 *
1812 * Use it ONLY if you really do not have anything better - i.e. when
1813 * you are behind a truly sucky interface and all you are given is a
1814 * device number. _Never_ to be used for internal purposes. If you
1815 * ever need it - reconsider your API.
1816 *
1817 * On success, the returned block_device has reference count of one.
1818 *
1819 * CONTEXT:
1820 * Might sleep.
1821 *
1822 * RETURNS:
1823 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1824 */
1825struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1826{
1827 struct block_device *bdev;
1828 int err;
1829
1830 bdev = bdget(dev);
1831 if (!bdev)
1832 return ERR_PTR(-ENOMEM);
1833
1834 err = blkdev_get(bdev, mode, holder);
1835 if (err)
1836 return ERR_PTR(err);
1837
1838 return bdev;
1839}
1840EXPORT_SYMBOL(blkdev_get_by_dev);
1841
1842static int blkdev_open(struct inode * inode, struct file * filp)
1843{
1844 struct block_device *bdev;
1845
1846 /*
1847 * Preserve backwards compatibility and allow large file access
1848 * even if userspace doesn't ask for it explicitly. Some mkfs
1849 * binary needs it. We might want to drop this workaround
1850 * during an unstable branch.
1851 */
1852 filp->f_flags |= O_LARGEFILE;
1853
1854 filp->f_mode |= FMODE_NOWAIT;
1855
1856 if (filp->f_flags & O_NDELAY)
1857 filp->f_mode |= FMODE_NDELAY;
1858 if (filp->f_flags & O_EXCL)
1859 filp->f_mode |= FMODE_EXCL;
1860 if ((filp->f_flags & O_ACCMODE) == 3)
1861 filp->f_mode |= FMODE_WRITE_IOCTL;
1862
1863 bdev = bd_acquire(inode);
1864 if (bdev == NULL)
1865 return -ENOMEM;
1866
1867 filp->f_mapping = bdev->bd_inode->i_mapping;
1868 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1869
1870 return blkdev_get(bdev, filp->f_mode, filp);
1871}
1872
1873static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1874{
1875 struct gendisk *disk = bdev->bd_disk;
1876 struct block_device *victim = NULL;
1877
1878 /*
1879 * Sync early if it looks like we're the last one. If someone else
1880 * opens the block device between now and the decrement of bd_openers
1881 * then we did a sync that we didn't need to, but that's not the end
1882 * of the world and we want to avoid long (could be several minute)
1883 * syncs while holding the mutex.
1884 */
1885 if (bdev->bd_openers == 1)
1886 sync_blockdev(bdev);
1887
1888 mutex_lock_nested(&bdev->bd_mutex, for_part);
1889 if (for_part)
1890 bdev->bd_part_count--;
1891
1892 if (!--bdev->bd_openers) {
1893 WARN_ON_ONCE(bdev->bd_holders);
1894 sync_blockdev(bdev);
1895 kill_bdev(bdev);
1896
1897 bdev_write_inode(bdev);
1898 }
1899 if (bdev->bd_contains == bdev) {
1900 if (disk->fops->release)
1901 disk->fops->release(disk, mode);
1902 }
1903 if (!bdev->bd_openers) {
1904 disk_put_part(bdev->bd_part);
1905 bdev->bd_part = NULL;
1906 bdev->bd_disk = NULL;
1907 if (bdev != bdev->bd_contains)
1908 victim = bdev->bd_contains;
1909 bdev->bd_contains = NULL;
1910
1911 put_disk_and_module(disk);
1912 }
1913 mutex_unlock(&bdev->bd_mutex);
1914 bdput(bdev);
1915 if (victim)
1916 __blkdev_put(victim, mode, 1);
1917}
1918
1919void blkdev_put(struct block_device *bdev, fmode_t mode)
1920{
1921 mutex_lock(&bdev->bd_mutex);
1922
1923 if (mode & FMODE_EXCL) {
1924 bool bdev_free;
1925
1926 /*
1927 * Release a claim on the device. The holder fields
1928 * are protected with bdev_lock. bd_mutex is to
1929 * synchronize disk_holder unlinking.
1930 */
1931 spin_lock(&bdev_lock);
1932
1933 WARN_ON_ONCE(--bdev->bd_holders < 0);
1934 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1935
1936 /* bd_contains might point to self, check in a separate step */
1937 if ((bdev_free = !bdev->bd_holders))
1938 bdev->bd_holder = NULL;
1939 if (!bdev->bd_contains->bd_holders)
1940 bdev->bd_contains->bd_holder = NULL;
1941
1942 spin_unlock(&bdev_lock);
1943
1944 /*
1945 * If this was the last claim, remove holder link and
1946 * unblock evpoll if it was a write holder.
1947 */
1948 if (bdev_free && bdev->bd_write_holder) {
1949 disk_unblock_events(bdev->bd_disk);
1950 bdev->bd_write_holder = false;
1951 }
1952 }
1953
1954 /*
1955 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1956 * event. This is to ensure detection of media removal commanded
1957 * from userland - e.g. eject(1).
1958 */
1959 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1960
1961 mutex_unlock(&bdev->bd_mutex);
1962
1963 __blkdev_put(bdev, mode, 0);
1964}
1965EXPORT_SYMBOL(blkdev_put);
1966
1967static int blkdev_close(struct inode * inode, struct file * filp)
1968{
1969 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1970 blkdev_put(bdev, filp->f_mode);
1971 return 0;
1972}
1973
1974static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1975{
1976 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1977 fmode_t mode = file->f_mode;
1978
1979 /*
1980 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1981 * to updated it before every ioctl.
1982 */
1983 if (file->f_flags & O_NDELAY)
1984 mode |= FMODE_NDELAY;
1985 else
1986 mode &= ~FMODE_NDELAY;
1987
1988 return blkdev_ioctl(bdev, mode, cmd, arg);
1989}
1990
1991/*
1992 * Write data to the block device. Only intended for the block device itself
1993 * and the raw driver which basically is a fake block device.
1994 *
1995 * Does not take i_mutex for the write and thus is not for general purpose
1996 * use.
1997 */
1998ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1999{
2000 struct file *file = iocb->ki_filp;
2001 struct inode *bd_inode = bdev_file_inode(file);
2002 loff_t size = i_size_read(bd_inode);
2003 struct blk_plug plug;
2004 ssize_t ret;
2005
2006 if (bdev_read_only(I_BDEV(bd_inode)))
2007 return -EPERM;
2008
2009 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode))
2010 return -ETXTBSY;
2011
2012 if (!iov_iter_count(from))
2013 return 0;
2014
2015 if (iocb->ki_pos >= size)
2016 return -ENOSPC;
2017
2018 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
2019 return -EOPNOTSUPP;
2020
2021 iov_iter_truncate(from, size - iocb->ki_pos);
2022
2023 blk_start_plug(&plug);
2024 ret = __generic_file_write_iter(iocb, from);
2025 if (ret > 0)
2026 ret = generic_write_sync(iocb, ret);
2027 blk_finish_plug(&plug);
2028 return ret;
2029}
2030EXPORT_SYMBOL_GPL(blkdev_write_iter);
2031
2032ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
2033{
2034 struct file *file = iocb->ki_filp;
2035 struct inode *bd_inode = bdev_file_inode(file);
2036 loff_t size = i_size_read(bd_inode);
2037 loff_t pos = iocb->ki_pos;
2038
2039 if (pos >= size)
2040 return 0;
2041
2042 size -= pos;
2043 iov_iter_truncate(to, size);
2044 return generic_file_read_iter(iocb, to);
2045}
2046EXPORT_SYMBOL_GPL(blkdev_read_iter);
2047
2048/*
2049 * Try to release a page associated with block device when the system
2050 * is under memory pressure.
2051 */
2052static int blkdev_releasepage(struct page *page, gfp_t wait)
2053{
2054 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
2055
2056 if (super && super->s_op->bdev_try_to_free_page)
2057 return super->s_op->bdev_try_to_free_page(super, page, wait);
2058
2059 return try_to_free_buffers(page);
2060}
2061
2062static int blkdev_writepages(struct address_space *mapping,
2063 struct writeback_control *wbc)
2064{
2065 return generic_writepages(mapping, wbc);
2066}
2067
2068static const struct address_space_operations def_blk_aops = {
2069 .readpage = blkdev_readpage,
2070 .readahead = blkdev_readahead,
2071 .writepage = blkdev_writepage,
2072 .write_begin = blkdev_write_begin,
2073 .write_end = blkdev_write_end,
2074 .writepages = blkdev_writepages,
2075 .releasepage = blkdev_releasepage,
2076 .direct_IO = blkdev_direct_IO,
2077 .migratepage = buffer_migrate_page_norefs,
2078 .is_dirty_writeback = buffer_check_dirty_writeback,
2079};
2080
2081#define BLKDEV_FALLOC_FL_SUPPORTED \
2082 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2083 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2084
2085static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2086 loff_t len)
2087{
2088 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2089 struct address_space *mapping;
2090 loff_t end = start + len - 1;
2091 loff_t isize;
2092 int error;
2093
2094 /* Fail if we don't recognize the flags. */
2095 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2096 return -EOPNOTSUPP;
2097
2098 /* Don't go off the end of the device. */
2099 isize = i_size_read(bdev->bd_inode);
2100 if (start >= isize)
2101 return -EINVAL;
2102 if (end >= isize) {
2103 if (mode & FALLOC_FL_KEEP_SIZE) {
2104 len = isize - start;
2105 end = start + len - 1;
2106 } else
2107 return -EINVAL;
2108 }
2109
2110 /*
2111 * Don't allow IO that isn't aligned to logical block size.
2112 */
2113 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2114 return -EINVAL;
2115
2116 /* Invalidate the page cache, including dirty pages. */
2117 mapping = bdev->bd_inode->i_mapping;
2118 truncate_inode_pages_range(mapping, start, end);
2119
2120 switch (mode) {
2121 case FALLOC_FL_ZERO_RANGE:
2122 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2123 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2124 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2125 break;
2126 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2127 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2128 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2129 break;
2130 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2131 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2132 GFP_KERNEL, 0);
2133 break;
2134 default:
2135 return -EOPNOTSUPP;
2136 }
2137 if (error)
2138 return error;
2139
2140 /*
2141 * Invalidate again; if someone wandered in and dirtied a page,
2142 * the caller will be given -EBUSY. The third argument is
2143 * inclusive, so the rounding here is safe.
2144 */
2145 return invalidate_inode_pages2_range(mapping,
2146 start >> PAGE_SHIFT,
2147 end >> PAGE_SHIFT);
2148}
2149
2150const struct file_operations def_blk_fops = {
2151 .open = blkdev_open,
2152 .release = blkdev_close,
2153 .llseek = block_llseek,
2154 .read_iter = blkdev_read_iter,
2155 .write_iter = blkdev_write_iter,
2156 .iopoll = blkdev_iopoll,
2157 .mmap = generic_file_mmap,
2158 .fsync = blkdev_fsync,
2159 .unlocked_ioctl = block_ioctl,
2160#ifdef CONFIG_COMPAT
2161 .compat_ioctl = compat_blkdev_ioctl,
2162#endif
2163 .splice_read = generic_file_splice_read,
2164 .splice_write = iter_file_splice_write,
2165 .fallocate = blkdev_fallocate,
2166};
2167
2168/**
2169 * lookup_bdev - lookup a struct block_device by name
2170 * @pathname: special file representing the block device
2171 *
2172 * Get a reference to the blockdevice at @pathname in the current
2173 * namespace if possible and return it. Return ERR_PTR(error)
2174 * otherwise.
2175 */
2176struct block_device *lookup_bdev(const char *pathname)
2177{
2178 struct block_device *bdev;
2179 struct inode *inode;
2180 struct path path;
2181 int error;
2182
2183 if (!pathname || !*pathname)
2184 return ERR_PTR(-EINVAL);
2185
2186 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2187 if (error)
2188 return ERR_PTR(error);
2189
2190 inode = d_backing_inode(path.dentry);
2191 error = -ENOTBLK;
2192 if (!S_ISBLK(inode->i_mode))
2193 goto fail;
2194 error = -EACCES;
2195 if (!may_open_dev(&path))
2196 goto fail;
2197 error = -ENOMEM;
2198 bdev = bd_acquire(inode);
2199 if (!bdev)
2200 goto fail;
2201out:
2202 path_put(&path);
2203 return bdev;
2204fail:
2205 bdev = ERR_PTR(error);
2206 goto out;
2207}
2208EXPORT_SYMBOL(lookup_bdev);
2209
2210int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2211{
2212 struct super_block *sb = get_super(bdev);
2213 int res = 0;
2214
2215 if (sb) {
2216 /*
2217 * no need to lock the super, get_super holds the
2218 * read mutex so the filesystem cannot go away
2219 * under us (->put_super runs with the write lock
2220 * hold).
2221 */
2222 shrink_dcache_sb(sb);
2223 res = invalidate_inodes(sb, kill_dirty);
2224 drop_super(sb);
2225 }
2226 invalidate_bdev(bdev);
2227 return res;
2228}
2229EXPORT_SYMBOL(__invalidate_device);
2230
2231void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2232{
2233 struct inode *inode, *old_inode = NULL;
2234
2235 spin_lock(&blockdev_superblock->s_inode_list_lock);
2236 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2237 struct address_space *mapping = inode->i_mapping;
2238 struct block_device *bdev;
2239
2240 spin_lock(&inode->i_lock);
2241 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2242 mapping->nrpages == 0) {
2243 spin_unlock(&inode->i_lock);
2244 continue;
2245 }
2246 __iget(inode);
2247 spin_unlock(&inode->i_lock);
2248 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2249 /*
2250 * We hold a reference to 'inode' so it couldn't have been
2251 * removed from s_inodes list while we dropped the
2252 * s_inode_list_lock We cannot iput the inode now as we can
2253 * be holding the last reference and we cannot iput it under
2254 * s_inode_list_lock. So we keep the reference and iput it
2255 * later.
2256 */
2257 iput(old_inode);
2258 old_inode = inode;
2259 bdev = I_BDEV(inode);
2260
2261 mutex_lock(&bdev->bd_mutex);
2262 if (bdev->bd_openers)
2263 func(bdev, arg);
2264 mutex_unlock(&bdev->bd_mutex);
2265
2266 spin_lock(&blockdev_superblock->s_inode_list_lock);
2267 }
2268 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2269 iput(old_inode);
2270}