projects/linux·blob v5.2 · refs

2221 lines · 54.9 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/dax.h>
23#include <linux/buffer_head.h>
24#include <linux/swap.h>
25#include <linux/pagevec.h>
26#include <linux/writeback.h>
27#include <linux/mpage.h>
28#include <linux/mount.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 "internal.h"
37
38struct bdev_inode {
39 struct block_device bdev;
40 struct inode vfs_inode;
41};
42
43static const struct address_space_operations def_blk_aops;
44
45static inline struct bdev_inode *BDEV_I(struct inode *inode)
46{
47 return container_of(inode, struct bdev_inode, vfs_inode);
48}
49
50struct block_device *I_BDEV(struct inode *inode)
51{
52 return &BDEV_I(inode)->bdev;
53}
54EXPORT_SYMBOL(I_BDEV);
55
56static void bdev_write_inode(struct block_device *bdev)
57{
58 struct inode *inode = bdev->bd_inode;
59 int ret;
60
61 spin_lock(&inode->i_lock);
62 while (inode->i_state & I_DIRTY) {
63 spin_unlock(&inode->i_lock);
64 ret = write_inode_now(inode, true);
65 if (ret) {
66 char name[BDEVNAME_SIZE];
67 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
68 "for block device %s (err=%d).\n",
69 bdevname(bdev, name), ret);
70 }
71 spin_lock(&inode->i_lock);
72 }
73 spin_unlock(&inode->i_lock);
74}
75
76/* Kill _all_ buffers and pagecache , dirty or not.. */
77void kill_bdev(struct block_device *bdev)
78{
79 struct address_space *mapping = bdev->bd_inode->i_mapping;
80
81 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
82 return;
83
84 invalidate_bh_lrus();
85 truncate_inode_pages(mapping, 0);
86}
87EXPORT_SYMBOL(kill_bdev);
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, *bvec;
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 struct bvec_iter_all iter_all;
213
214 if ((pos | iov_iter_alignment(iter)) &
215 (bdev_logical_block_size(bdev) - 1))
216 return -EINVAL;
217
218 if (nr_pages <= DIO_INLINE_BIO_VECS)
219 vecs = inline_vecs;
220 else {
221 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
222 GFP_KERNEL);
223 if (!vecs)
224 return -ENOMEM;
225 }
226
227 bio_init(&bio, vecs, nr_pages);
228 bio_set_dev(&bio, bdev);
229 bio.bi_iter.bi_sector = pos >> 9;
230 bio.bi_write_hint = iocb->ki_hint;
231 bio.bi_private = current;
232 bio.bi_end_io = blkdev_bio_end_io_simple;
233 bio.bi_ioprio = iocb->ki_ioprio;
234
235 ret = bio_iov_iter_get_pages(&bio, iter);
236 if (unlikely(ret))
237 goto out;
238 ret = bio.bi_iter.bi_size;
239
240 if (iov_iter_rw(iter) == READ) {
241 bio.bi_opf = REQ_OP_READ;
242 if (iter_is_iovec(iter))
243 should_dirty = true;
244 } else {
245 bio.bi_opf = dio_bio_write_op(iocb);
246 task_io_account_write(ret);
247 }
248 if (iocb->ki_flags & IOCB_HIPRI)
249 bio_set_polled(&bio, iocb);
250
251 qc = submit_bio(&bio);
252 for (;;) {
253 set_current_state(TASK_UNINTERRUPTIBLE);
254 if (!READ_ONCE(bio.bi_private))
255 break;
256 if (!(iocb->ki_flags & IOCB_HIPRI) ||
257 !blk_poll(bdev_get_queue(bdev), qc, true))
258 io_schedule();
259 }
260 __set_current_state(TASK_RUNNING);
261
262 bio_for_each_segment_all(bvec, &bio, iter_all) {
263 if (should_dirty && !PageCompound(bvec->bv_page))
264 set_page_dirty_lock(bvec->bv_page);
265 if (!bio_flagged(&bio, BIO_NO_PAGE_REF))
266 put_page(bvec->bv_page);
267 }
268
269 if (unlikely(bio.bi_status))
270 ret = blk_status_to_errno(bio.bi_status);
271
272out:
273 if (vecs != inline_vecs)
274 kfree(vecs);
275
276 bio_uninit(&bio);
277
278 return ret;
279}
280
281struct blkdev_dio {
282 union {
283 struct kiocb *iocb;
284 struct task_struct *waiter;
285 };
286 size_t size;
287 atomic_t ref;
288 bool multi_bio : 1;
289 bool should_dirty : 1;
290 bool is_sync : 1;
291 struct bio bio;
292};
293
294static struct bio_set blkdev_dio_pool;
295
296static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
297{
298 struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
299 struct request_queue *q = bdev_get_queue(bdev);
300
301 return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
302}
303
304static void blkdev_bio_end_io(struct bio *bio)
305{
306 struct blkdev_dio *dio = bio->bi_private;
307 bool should_dirty = dio->should_dirty;
308
309 if (bio->bi_status && !dio->bio.bi_status)
310 dio->bio.bi_status = bio->bi_status;
311
312 if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
313 if (!dio->is_sync) {
314 struct kiocb *iocb = dio->iocb;
315 ssize_t ret;
316
317 if (likely(!dio->bio.bi_status)) {
318 ret = dio->size;
319 iocb->ki_pos += ret;
320 } else {
321 ret = blk_status_to_errno(dio->bio.bi_status);
322 }
323
324 dio->iocb->ki_complete(iocb, ret, 0);
325 if (dio->multi_bio)
326 bio_put(&dio->bio);
327 } else {
328 struct task_struct *waiter = dio->waiter;
329
330 WRITE_ONCE(dio->waiter, NULL);
331 blk_wake_io_task(waiter);
332 }
333 }
334
335 if (should_dirty) {
336 bio_check_pages_dirty(bio);
337 } else {
338 if (!bio_flagged(bio, BIO_NO_PAGE_REF)) {
339 struct bvec_iter_all iter_all;
340 struct bio_vec *bvec;
341
342 bio_for_each_segment_all(bvec, bio, iter_all)
343 put_page(bvec->bv_page);
344 }
345 bio_put(bio);
346 }
347}
348
349static ssize_t
350__blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
351{
352 struct file *file = iocb->ki_filp;
353 struct inode *inode = bdev_file_inode(file);
354 struct block_device *bdev = I_BDEV(inode);
355 struct blk_plug plug;
356 struct blkdev_dio *dio;
357 struct bio *bio;
358 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
359 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
360 loff_t pos = iocb->ki_pos;
361 blk_qc_t qc = BLK_QC_T_NONE;
362 int ret = 0;
363
364 if ((pos | iov_iter_alignment(iter)) &
365 (bdev_logical_block_size(bdev) - 1))
366 return -EINVAL;
367
368 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
369
370 dio = container_of(bio, struct blkdev_dio, bio);
371 dio->is_sync = is_sync = is_sync_kiocb(iocb);
372 if (dio->is_sync) {
373 dio->waiter = current;
374 bio_get(bio);
375 } else {
376 dio->iocb = iocb;
377 }
378
379 dio->size = 0;
380 dio->multi_bio = false;
381 dio->should_dirty = is_read && iter_is_iovec(iter);
382
383 /*
384 * Don't plug for HIPRI/polled IO, as those should go straight
385 * to issue
386 */
387 if (!is_poll)
388 blk_start_plug(&plug);
389
390 for (;;) {
391 bio_set_dev(bio, bdev);
392 bio->bi_iter.bi_sector = pos >> 9;
393 bio->bi_write_hint = iocb->ki_hint;
394 bio->bi_private = dio;
395 bio->bi_end_io = blkdev_bio_end_io;
396 bio->bi_ioprio = iocb->ki_ioprio;
397
398 ret = bio_iov_iter_get_pages(bio, iter);
399 if (unlikely(ret)) {
400 bio->bi_status = BLK_STS_IOERR;
401 bio_endio(bio);
402 break;
403 }
404
405 if (is_read) {
406 bio->bi_opf = REQ_OP_READ;
407 if (dio->should_dirty)
408 bio_set_pages_dirty(bio);
409 } else {
410 bio->bi_opf = dio_bio_write_op(iocb);
411 task_io_account_write(bio->bi_iter.bi_size);
412 }
413
414 dio->size += bio->bi_iter.bi_size;
415 pos += bio->bi_iter.bi_size;
416
417 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
418 if (!nr_pages) {
419 bool polled = false;
420
421 if (iocb->ki_flags & IOCB_HIPRI) {
422 bio_set_polled(bio, iocb);
423 polled = true;
424 }
425
426 qc = submit_bio(bio);
427
428 if (polled)
429 WRITE_ONCE(iocb->ki_cookie, qc);
430 break;
431 }
432
433 if (!dio->multi_bio) {
434 /*
435 * AIO needs an extra reference to ensure the dio
436 * structure which is embedded into the first bio
437 * stays around.
438 */
439 if (!is_sync)
440 bio_get(bio);
441 dio->multi_bio = true;
442 atomic_set(&dio->ref, 2);
443 } else {
444 atomic_inc(&dio->ref);
445 }
446
447 submit_bio(bio);
448 bio = bio_alloc(GFP_KERNEL, nr_pages);
449 }
450
451 if (!is_poll)
452 blk_finish_plug(&plug);
453
454 if (!is_sync)
455 return -EIOCBQUEUED;
456
457 for (;;) {
458 set_current_state(TASK_UNINTERRUPTIBLE);
459 if (!READ_ONCE(dio->waiter))
460 break;
461
462 if (!(iocb->ki_flags & IOCB_HIPRI) ||
463 !blk_poll(bdev_get_queue(bdev), qc, true))
464 io_schedule();
465 }
466 __set_current_state(TASK_RUNNING);
467
468 if (!ret)
469 ret = blk_status_to_errno(dio->bio.bi_status);
470 if (likely(!ret))
471 ret = dio->size;
472
473 bio_put(&dio->bio);
474 return ret;
475}
476
477static ssize_t
478blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
479{
480 int nr_pages;
481
482 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
483 if (!nr_pages)
484 return 0;
485 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
486 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
487
488 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
489}
490
491static __init int blkdev_init(void)
492{
493 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
494}
495module_init(blkdev_init);
496
497int __sync_blockdev(struct block_device *bdev, int wait)
498{
499 if (!bdev)
500 return 0;
501 if (!wait)
502 return filemap_flush(bdev->bd_inode->i_mapping);
503 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
504}
505
506/*
507 * Write out and wait upon all the dirty data associated with a block
508 * device via its mapping. Does not take the superblock lock.
509 */
510int sync_blockdev(struct block_device *bdev)
511{
512 return __sync_blockdev(bdev, 1);
513}
514EXPORT_SYMBOL(sync_blockdev);
515
516/*
517 * Write out and wait upon all dirty data associated with this
518 * device. Filesystem data as well as the underlying block
519 * device. Takes the superblock lock.
520 */
521int fsync_bdev(struct block_device *bdev)
522{
523 struct super_block *sb = get_super(bdev);
524 if (sb) {
525 int res = sync_filesystem(sb);
526 drop_super(sb);
527 return res;
528 }
529 return sync_blockdev(bdev);
530}
531EXPORT_SYMBOL(fsync_bdev);
532
533/**
534 * freeze_bdev -- lock a filesystem and force it into a consistent state
535 * @bdev: blockdevice to lock
536 *
537 * If a superblock is found on this device, we take the s_umount semaphore
538 * on it to make sure nobody unmounts until the snapshot creation is done.
539 * The reference counter (bd_fsfreeze_count) guarantees that only the last
540 * unfreeze process can unfreeze the frozen filesystem actually when multiple
541 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
542 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
543 * actually.
544 */
545struct super_block *freeze_bdev(struct block_device *bdev)
546{
547 struct super_block *sb;
548 int error = 0;
549
550 mutex_lock(&bdev->bd_fsfreeze_mutex);
551 if (++bdev->bd_fsfreeze_count > 1) {
552 /*
553 * We don't even need to grab a reference - the first call
554 * to freeze_bdev grab an active reference and only the last
555 * thaw_bdev drops it.
556 */
557 sb = get_super(bdev);
558 if (sb)
559 drop_super(sb);
560 mutex_unlock(&bdev->bd_fsfreeze_mutex);
561 return sb;
562 }
563
564 sb = get_active_super(bdev);
565 if (!sb)
566 goto out;
567 if (sb->s_op->freeze_super)
568 error = sb->s_op->freeze_super(sb);
569 else
570 error = freeze_super(sb);
571 if (error) {
572 deactivate_super(sb);
573 bdev->bd_fsfreeze_count--;
574 mutex_unlock(&bdev->bd_fsfreeze_mutex);
575 return ERR_PTR(error);
576 }
577 deactivate_super(sb);
578 out:
579 sync_blockdev(bdev);
580 mutex_unlock(&bdev->bd_fsfreeze_mutex);
581 return sb; /* thaw_bdev releases s->s_umount */
582}
583EXPORT_SYMBOL(freeze_bdev);
584
585/**
586 * thaw_bdev -- unlock filesystem
587 * @bdev: blockdevice to unlock
588 * @sb: associated superblock
589 *
590 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
591 */
592int thaw_bdev(struct block_device *bdev, struct super_block *sb)
593{
594 int error = -EINVAL;
595
596 mutex_lock(&bdev->bd_fsfreeze_mutex);
597 if (!bdev->bd_fsfreeze_count)
598 goto out;
599
600 error = 0;
601 if (--bdev->bd_fsfreeze_count > 0)
602 goto out;
603
604 if (!sb)
605 goto out;
606
607 if (sb->s_op->thaw_super)
608 error = sb->s_op->thaw_super(sb);
609 else
610 error = thaw_super(sb);
611 if (error)
612 bdev->bd_fsfreeze_count++;
613out:
614 mutex_unlock(&bdev->bd_fsfreeze_mutex);
615 return error;
616}
617EXPORT_SYMBOL(thaw_bdev);
618
619static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
620{
621 return block_write_full_page(page, blkdev_get_block, wbc);
622}
623
624static int blkdev_readpage(struct file * file, struct page * page)
625{
626 return block_read_full_page(page, blkdev_get_block);
627}
628
629static int blkdev_readpages(struct file *file, struct address_space *mapping,
630 struct list_head *pages, unsigned nr_pages)
631{
632 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
633}
634
635static int blkdev_write_begin(struct file *file, struct address_space *mapping,
636 loff_t pos, unsigned len, unsigned flags,
637 struct page **pagep, void **fsdata)
638{
639 return block_write_begin(mapping, pos, len, flags, pagep,
640 blkdev_get_block);
641}
642
643static int blkdev_write_end(struct file *file, struct address_space *mapping,
644 loff_t pos, unsigned len, unsigned copied,
645 struct page *page, void *fsdata)
646{
647 int ret;
648 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
649
650 unlock_page(page);
651 put_page(page);
652
653 return ret;
654}
655
656/*
657 * private llseek:
658 * for a block special file file_inode(file)->i_size is zero
659 * so we compute the size by hand (just as in block_read/write above)
660 */
661static loff_t block_llseek(struct file *file, loff_t offset, int whence)
662{
663 struct inode *bd_inode = bdev_file_inode(file);
664 loff_t retval;
665
666 inode_lock(bd_inode);
667 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
668 inode_unlock(bd_inode);
669 return retval;
670}
671
672int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
673{
674 struct inode *bd_inode = bdev_file_inode(filp);
675 struct block_device *bdev = I_BDEV(bd_inode);
676 int error;
677
678 error = file_write_and_wait_range(filp, start, end);
679 if (error)
680 return error;
681
682 /*
683 * There is no need to serialise calls to blkdev_issue_flush with
684 * i_mutex and doing so causes performance issues with concurrent
685 * O_SYNC writers to a block device.
686 */
687 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
688 if (error == -EOPNOTSUPP)
689 error = 0;
690
691 return error;
692}
693EXPORT_SYMBOL(blkdev_fsync);
694
695/**
696 * bdev_read_page() - Start reading a page from a block device
697 * @bdev: The device to read the page from
698 * @sector: The offset on the device to read the page to (need not be aligned)
699 * @page: The page to read
700 *
701 * On entry, the page should be locked. It will be unlocked when the page
702 * has been read. If the block driver implements rw_page synchronously,
703 * that will be true on exit from this function, but it need not be.
704 *
705 * Errors returned by this function are usually "soft", eg out of memory, or
706 * queue full; callers should try a different route to read this page rather
707 * than propagate an error back up the stack.
708 *
709 * Return: negative errno if an error occurs, 0 if submission was successful.
710 */
711int bdev_read_page(struct block_device *bdev, sector_t sector,
712 struct page *page)
713{
714 const struct block_device_operations *ops = bdev->bd_disk->fops;
715 int result = -EOPNOTSUPP;
716
717 if (!ops->rw_page || bdev_get_integrity(bdev))
718 return result;
719
720 result = blk_queue_enter(bdev->bd_queue, 0);
721 if (result)
722 return result;
723 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
724 REQ_OP_READ);
725 blk_queue_exit(bdev->bd_queue);
726 return result;
727}
728EXPORT_SYMBOL_GPL(bdev_read_page);
729
730/**
731 * bdev_write_page() - Start writing a page to a block device
732 * @bdev: The device to write the page to
733 * @sector: The offset on the device to write the page to (need not be aligned)
734 * @page: The page to write
735 * @wbc: The writeback_control for the write
736 *
737 * On entry, the page should be locked and not currently under writeback.
738 * On exit, if the write started successfully, the page will be unlocked and
739 * under writeback. If the write failed already (eg the driver failed to
740 * queue the page to the device), the page will still be locked. If the
741 * caller is a ->writepage implementation, it will need to unlock the page.
742 *
743 * Errors returned by this function are usually "soft", eg out of memory, or
744 * queue full; callers should try a different route to write this page rather
745 * than propagate an error back up the stack.
746 *
747 * Return: negative errno if an error occurs, 0 if submission was successful.
748 */
749int bdev_write_page(struct block_device *bdev, sector_t sector,
750 struct page *page, struct writeback_control *wbc)
751{
752 int result;
753 const struct block_device_operations *ops = bdev->bd_disk->fops;
754
755 if (!ops->rw_page || bdev_get_integrity(bdev))
756 return -EOPNOTSUPP;
757 result = blk_queue_enter(bdev->bd_queue, 0);
758 if (result)
759 return result;
760
761 set_page_writeback(page);
762 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
763 REQ_OP_WRITE);
764 if (result) {
765 end_page_writeback(page);
766 } else {
767 clean_page_buffers(page);
768 unlock_page(page);
769 }
770 blk_queue_exit(bdev->bd_queue);
771 return result;
772}
773EXPORT_SYMBOL_GPL(bdev_write_page);
774
775/*
776 * pseudo-fs
777 */
778
779static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
780static struct kmem_cache * bdev_cachep __read_mostly;
781
782static struct inode *bdev_alloc_inode(struct super_block *sb)
783{
784 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
785 if (!ei)
786 return NULL;
787 return &ei->vfs_inode;
788}
789
790static void bdev_free_inode(struct inode *inode)
791{
792 kmem_cache_free(bdev_cachep, BDEV_I(inode));
793}
794
795static void init_once(void *foo)
796{
797 struct bdev_inode *ei = (struct bdev_inode *) foo;
798 struct block_device *bdev = &ei->bdev;
799
800 memset(bdev, 0, sizeof(*bdev));
801 mutex_init(&bdev->bd_mutex);
802 INIT_LIST_HEAD(&bdev->bd_list);
803#ifdef CONFIG_SYSFS
804 INIT_LIST_HEAD(&bdev->bd_holder_disks);
805#endif
806 bdev->bd_bdi = &noop_backing_dev_info;
807 inode_init_once(&ei->vfs_inode);
808 /* Initialize mutex for freeze. */
809 mutex_init(&bdev->bd_fsfreeze_mutex);
810}
811
812static void bdev_evict_inode(struct inode *inode)
813{
814 struct block_device *bdev = &BDEV_I(inode)->bdev;
815 truncate_inode_pages_final(&inode->i_data);
816 invalidate_inode_buffers(inode); /* is it needed here? */
817 clear_inode(inode);
818 spin_lock(&bdev_lock);
819 list_del_init(&bdev->bd_list);
820 spin_unlock(&bdev_lock);
821 /* Detach inode from wb early as bdi_put() may free bdi->wb */
822 inode_detach_wb(inode);
823 if (bdev->bd_bdi != &noop_backing_dev_info) {
824 bdi_put(bdev->bd_bdi);
825 bdev->bd_bdi = &noop_backing_dev_info;
826 }
827}
828
829static const struct super_operations bdev_sops = {
830 .statfs = simple_statfs,
831 .alloc_inode = bdev_alloc_inode,
832 .free_inode = bdev_free_inode,
833 .drop_inode = generic_delete_inode,
834 .evict_inode = bdev_evict_inode,
835};
836
837static struct dentry *bd_mount(struct file_system_type *fs_type,
838 int flags, const char *dev_name, void *data)
839{
840 struct dentry *dent;
841 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
842 if (!IS_ERR(dent))
843 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
844 return dent;
845}
846
847static struct file_system_type bd_type = {
848 .name = "bdev",
849 .mount = bd_mount,
850 .kill_sb = kill_anon_super,
851};
852
853struct super_block *blockdev_superblock __read_mostly;
854EXPORT_SYMBOL_GPL(blockdev_superblock);
855
856void __init bdev_cache_init(void)
857{
858 int err;
859 static struct vfsmount *bd_mnt;
860
861 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
862 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
863 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
864 init_once);
865 err = register_filesystem(&bd_type);
866 if (err)
867 panic("Cannot register bdev pseudo-fs");
868 bd_mnt = kern_mount(&bd_type);
869 if (IS_ERR(bd_mnt))
870 panic("Cannot create bdev pseudo-fs");
871 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
872}
873
874/*
875 * Most likely _very_ bad one - but then it's hardly critical for small
876 * /dev and can be fixed when somebody will need really large one.
877 * Keep in mind that it will be fed through icache hash function too.
878 */
879static inline unsigned long hash(dev_t dev)
880{
881 return MAJOR(dev)+MINOR(dev);
882}
883
884static int bdev_test(struct inode *inode, void *data)
885{
886 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
887}
888
889static int bdev_set(struct inode *inode, void *data)
890{
891 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
892 return 0;
893}
894
895static LIST_HEAD(all_bdevs);
896
897/*
898 * If there is a bdev inode for this device, unhash it so that it gets evicted
899 * as soon as last inode reference is dropped.
900 */
901void bdev_unhash_inode(dev_t dev)
902{
903 struct inode *inode;
904
905 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
906 if (inode) {
907 remove_inode_hash(inode);
908 iput(inode);
909 }
910}
911
912struct block_device *bdget(dev_t dev)
913{
914 struct block_device *bdev;
915 struct inode *inode;
916
917 inode = iget5_locked(blockdev_superblock, hash(dev),
918 bdev_test, bdev_set, &dev);
919
920 if (!inode)
921 return NULL;
922
923 bdev = &BDEV_I(inode)->bdev;
924
925 if (inode->i_state & I_NEW) {
926 bdev->bd_contains = NULL;
927 bdev->bd_super = NULL;
928 bdev->bd_inode = inode;
929 bdev->bd_block_size = i_blocksize(inode);
930 bdev->bd_part_count = 0;
931 bdev->bd_invalidated = 0;
932 inode->i_mode = S_IFBLK;
933 inode->i_rdev = dev;
934 inode->i_bdev = bdev;
935 inode->i_data.a_ops = &def_blk_aops;
936 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
937 spin_lock(&bdev_lock);
938 list_add(&bdev->bd_list, &all_bdevs);
939 spin_unlock(&bdev_lock);
940 unlock_new_inode(inode);
941 }
942 return bdev;
943}
944
945EXPORT_SYMBOL(bdget);
946
947/**
948 * bdgrab -- Grab a reference to an already referenced block device
949 * @bdev: Block device to grab a reference to.
950 */
951struct block_device *bdgrab(struct block_device *bdev)
952{
953 ihold(bdev->bd_inode);
954 return bdev;
955}
956EXPORT_SYMBOL(bdgrab);
957
958long nr_blockdev_pages(void)
959{
960 struct block_device *bdev;
961 long ret = 0;
962 spin_lock(&bdev_lock);
963 list_for_each_entry(bdev, &all_bdevs, bd_list) {
964 ret += bdev->bd_inode->i_mapping->nrpages;
965 }
966 spin_unlock(&bdev_lock);
967 return ret;
968}
969
970void bdput(struct block_device *bdev)
971{
972 iput(bdev->bd_inode);
973}
974
975EXPORT_SYMBOL(bdput);
976
977static struct block_device *bd_acquire(struct inode *inode)
978{
979 struct block_device *bdev;
980
981 spin_lock(&bdev_lock);
982 bdev = inode->i_bdev;
983 if (bdev && !inode_unhashed(bdev->bd_inode)) {
984 bdgrab(bdev);
985 spin_unlock(&bdev_lock);
986 return bdev;
987 }
988 spin_unlock(&bdev_lock);
989
990 /*
991 * i_bdev references block device inode that was already shut down
992 * (corresponding device got removed). Remove the reference and look
993 * up block device inode again just in case new device got
994 * reestablished under the same device number.
995 */
996 if (bdev)
997 bd_forget(inode);
998
999 bdev = bdget(inode->i_rdev);
1000 if (bdev) {
1001 spin_lock(&bdev_lock);
1002 if (!inode->i_bdev) {
1003 /*
1004 * We take an additional reference to bd_inode,
1005 * and it's released in clear_inode() of inode.
1006 * So, we can access it via ->i_mapping always
1007 * without igrab().
1008 */
1009 bdgrab(bdev);
1010 inode->i_bdev = bdev;
1011 inode->i_mapping = bdev->bd_inode->i_mapping;
1012 }
1013 spin_unlock(&bdev_lock);
1014 }
1015 return bdev;
1016}
1017
1018/* Call when you free inode */
1019
1020void bd_forget(struct inode *inode)
1021{
1022 struct block_device *bdev = NULL;
1023
1024 spin_lock(&bdev_lock);
1025 if (!sb_is_blkdev_sb(inode->i_sb))
1026 bdev = inode->i_bdev;
1027 inode->i_bdev = NULL;
1028 inode->i_mapping = &inode->i_data;
1029 spin_unlock(&bdev_lock);
1030
1031 if (bdev)
1032 bdput(bdev);
1033}
1034
1035/**
1036 * bd_may_claim - test whether a block device can be claimed
1037 * @bdev: block device of interest
1038 * @whole: whole block device containing @bdev, may equal @bdev
1039 * @holder: holder trying to claim @bdev
1040 *
1041 * Test whether @bdev can be claimed by @holder.
1042 *
1043 * CONTEXT:
1044 * spin_lock(&bdev_lock).
1045 *
1046 * RETURNS:
1047 * %true if @bdev can be claimed, %false otherwise.
1048 */
1049static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1050 void *holder)
1051{
1052 if (bdev->bd_holder == holder)
1053 return true; /* already a holder */
1054 else if (bdev->bd_holder != NULL)
1055 return false; /* held by someone else */
1056 else if (whole == bdev)
1057 return true; /* is a whole device which isn't held */
1058
1059 else if (whole->bd_holder == bd_may_claim)
1060 return true; /* is a partition of a device that is being partitioned */
1061 else if (whole->bd_holder != NULL)
1062 return false; /* is a partition of a held device */
1063 else
1064 return true; /* is a partition of an un-held device */
1065}
1066
1067/**
1068 * bd_prepare_to_claim - prepare to claim a block device
1069 * @bdev: block device of interest
1070 * @whole: the whole device containing @bdev, may equal @bdev
1071 * @holder: holder trying to claim @bdev
1072 *
1073 * Prepare to claim @bdev. This function fails if @bdev is already
1074 * claimed by another holder and waits if another claiming is in
1075 * progress. This function doesn't actually claim. On successful
1076 * return, the caller has ownership of bd_claiming and bd_holder[s].
1077 *
1078 * CONTEXT:
1079 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1080 * it multiple times.
1081 *
1082 * RETURNS:
1083 * 0 if @bdev can be claimed, -EBUSY otherwise.
1084 */
1085static int bd_prepare_to_claim(struct block_device *bdev,
1086 struct block_device *whole, void *holder)
1087{
1088retry:
1089 /* if someone else claimed, fail */
1090 if (!bd_may_claim(bdev, whole, holder))
1091 return -EBUSY;
1092
1093 /* if claiming is already in progress, wait for it to finish */
1094 if (whole->bd_claiming) {
1095 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1096 DEFINE_WAIT(wait);
1097
1098 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1099 spin_unlock(&bdev_lock);
1100 schedule();
1101 finish_wait(wq, &wait);
1102 spin_lock(&bdev_lock);
1103 goto retry;
1104 }
1105
1106 /* yay, all mine */
1107 return 0;
1108}
1109
1110static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1111{
1112 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1113
1114 if (!disk)
1115 return NULL;
1116 /*
1117 * Now that we hold gendisk reference we make sure bdev we looked up is
1118 * not stale. If it is, it means device got removed and created before
1119 * we looked up gendisk and we fail open in such case. Associating
1120 * unhashed bdev with newly created gendisk could lead to two bdevs
1121 * (and thus two independent caches) being associated with one device
1122 * which is bad.
1123 */
1124 if (inode_unhashed(bdev->bd_inode)) {
1125 put_disk_and_module(disk);
1126 return NULL;
1127 }
1128 return disk;
1129}
1130
1131/**
1132 * bd_start_claiming - start claiming a block device
1133 * @bdev: block device of interest
1134 * @holder: holder trying to claim @bdev
1135 *
1136 * @bdev is about to be opened exclusively. Check @bdev can be opened
1137 * exclusively and mark that an exclusive open is in progress. Each
1138 * successful call to this function must be matched with a call to
1139 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1140 * fail).
1141 *
1142 * This function is used to gain exclusive access to the block device
1143 * without actually causing other exclusive open attempts to fail. It
1144 * should be used when the open sequence itself requires exclusive
1145 * access but may subsequently fail.
1146 *
1147 * CONTEXT:
1148 * Might sleep.
1149 *
1150 * RETURNS:
1151 * Pointer to the block device containing @bdev on success, ERR_PTR()
1152 * value on failure.
1153 */
1154static struct block_device *bd_start_claiming(struct block_device *bdev,
1155 void *holder)
1156{
1157 struct gendisk *disk;
1158 struct block_device *whole;
1159 int partno, err;
1160
1161 might_sleep();
1162
1163 /*
1164 * @bdev might not have been initialized properly yet, look up
1165 * and grab the outer block device the hard way.
1166 */
1167 disk = bdev_get_gendisk(bdev, &partno);
1168 if (!disk)
1169 return ERR_PTR(-ENXIO);
1170
1171 /*
1172 * Normally, @bdev should equal what's returned from bdget_disk()
1173 * if partno is 0; however, some drivers (floppy) use multiple
1174 * bdev's for the same physical device and @bdev may be one of the
1175 * aliases. Keep @bdev if partno is 0. This means claimer
1176 * tracking is broken for those devices but it has always been that
1177 * way.
1178 */
1179 if (partno)
1180 whole = bdget_disk(disk, 0);
1181 else
1182 whole = bdgrab(bdev);
1183
1184 put_disk_and_module(disk);
1185 if (!whole)
1186 return ERR_PTR(-ENOMEM);
1187
1188 /* prepare to claim, if successful, mark claiming in progress */
1189 spin_lock(&bdev_lock);
1190
1191 err = bd_prepare_to_claim(bdev, whole, holder);
1192 if (err == 0) {
1193 whole->bd_claiming = holder;
1194 spin_unlock(&bdev_lock);
1195 return whole;
1196 } else {
1197 spin_unlock(&bdev_lock);
1198 bdput(whole);
1199 return ERR_PTR(err);
1200 }
1201}
1202
1203#ifdef CONFIG_SYSFS
1204struct bd_holder_disk {
1205 struct list_head list;
1206 struct gendisk *disk;
1207 int refcnt;
1208};
1209
1210static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1211 struct gendisk *disk)
1212{
1213 struct bd_holder_disk *holder;
1214
1215 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1216 if (holder->disk == disk)
1217 return holder;
1218 return NULL;
1219}
1220
1221static int add_symlink(struct kobject *from, struct kobject *to)
1222{
1223 return sysfs_create_link(from, to, kobject_name(to));
1224}
1225
1226static void del_symlink(struct kobject *from, struct kobject *to)
1227{
1228 sysfs_remove_link(from, kobject_name(to));
1229}
1230
1231/**
1232 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1233 * @bdev: the claimed slave bdev
1234 * @disk: the holding disk
1235 *
1236 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1237 *
1238 * This functions creates the following sysfs symlinks.
1239 *
1240 * - from "slaves" directory of the holder @disk to the claimed @bdev
1241 * - from "holders" directory of the @bdev to the holder @disk
1242 *
1243 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1244 * passed to bd_link_disk_holder(), then:
1245 *
1246 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1247 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1248 *
1249 * The caller must have claimed @bdev before calling this function and
1250 * ensure that both @bdev and @disk are valid during the creation and
1251 * lifetime of these symlinks.
1252 *
1253 * CONTEXT:
1254 * Might sleep.
1255 *
1256 * RETURNS:
1257 * 0 on success, -errno on failure.
1258 */
1259int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1260{
1261 struct bd_holder_disk *holder;
1262 int ret = 0;
1263
1264 mutex_lock(&bdev->bd_mutex);
1265
1266 WARN_ON_ONCE(!bdev->bd_holder);
1267
1268 /* FIXME: remove the following once add_disk() handles errors */
1269 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1270 goto out_unlock;
1271
1272 holder = bd_find_holder_disk(bdev, disk);
1273 if (holder) {
1274 holder->refcnt++;
1275 goto out_unlock;
1276 }
1277
1278 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1279 if (!holder) {
1280 ret = -ENOMEM;
1281 goto out_unlock;
1282 }
1283
1284 INIT_LIST_HEAD(&holder->list);
1285 holder->disk = disk;
1286 holder->refcnt = 1;
1287
1288 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1289 if (ret)
1290 goto out_free;
1291
1292 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1293 if (ret)
1294 goto out_del;
1295 /*
1296 * bdev could be deleted beneath us which would implicitly destroy
1297 * the holder directory. Hold on to it.
1298 */
1299 kobject_get(bdev->bd_part->holder_dir);
1300
1301 list_add(&holder->list, &bdev->bd_holder_disks);
1302 goto out_unlock;
1303
1304out_del:
1305 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1306out_free:
1307 kfree(holder);
1308out_unlock:
1309 mutex_unlock(&bdev->bd_mutex);
1310 return ret;
1311}
1312EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1313
1314/**
1315 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1316 * @bdev: the calimed slave bdev
1317 * @disk: the holding disk
1318 *
1319 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1320 *
1321 * CONTEXT:
1322 * Might sleep.
1323 */
1324void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1325{
1326 struct bd_holder_disk *holder;
1327
1328 mutex_lock(&bdev->bd_mutex);
1329
1330 holder = bd_find_holder_disk(bdev, disk);
1331
1332 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1333 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1334 del_symlink(bdev->bd_part->holder_dir,
1335 &disk_to_dev(disk)->kobj);
1336 kobject_put(bdev->bd_part->holder_dir);
1337 list_del_init(&holder->list);
1338 kfree(holder);
1339 }
1340
1341 mutex_unlock(&bdev->bd_mutex);
1342}
1343EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1344#endif
1345
1346/**
1347 * flush_disk - invalidates all buffer-cache entries on a disk
1348 *
1349 * @bdev: struct block device to be flushed
1350 * @kill_dirty: flag to guide handling of dirty inodes
1351 *
1352 * Invalidates all buffer-cache entries on a disk. It should be called
1353 * when a disk has been changed -- either by a media change or online
1354 * resize.
1355 */
1356static void flush_disk(struct block_device *bdev, bool kill_dirty)
1357{
1358 if (__invalidate_device(bdev, kill_dirty)) {
1359 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1360 "resized disk %s\n",
1361 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1362 }
1363
1364 if (!bdev->bd_disk)
1365 return;
1366 if (disk_part_scan_enabled(bdev->bd_disk))
1367 bdev->bd_invalidated = 1;
1368}
1369
1370/**
1371 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1372 * @disk: struct gendisk to check
1373 * @bdev: struct bdev to adjust.
1374 * @verbose: if %true log a message about a size change if there is any
1375 *
1376 * This routine checks to see if the bdev size does not match the disk size
1377 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1378 * are freed.
1379 */
1380void check_disk_size_change(struct gendisk *disk, struct block_device *bdev,
1381 bool verbose)
1382{
1383 loff_t disk_size, bdev_size;
1384
1385 disk_size = (loff_t)get_capacity(disk) << 9;
1386 bdev_size = i_size_read(bdev->bd_inode);
1387 if (disk_size != bdev_size) {
1388 if (verbose) {
1389 printk(KERN_INFO
1390 "%s: detected capacity change from %lld to %lld\n",
1391 disk->disk_name, bdev_size, disk_size);
1392 }
1393 i_size_write(bdev->bd_inode, disk_size);
1394 if (bdev_size > disk_size)
1395 flush_disk(bdev, false);
1396 }
1397}
1398
1399/**
1400 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1401 * @disk: struct gendisk to be revalidated
1402 *
1403 * This routine is a wrapper for lower-level driver's revalidate_disk
1404 * call-backs. It is used to do common pre and post operations needed
1405 * for all revalidate_disk operations.
1406 */
1407int revalidate_disk(struct gendisk *disk)
1408{
1409 int ret = 0;
1410
1411 if (disk->fops->revalidate_disk)
1412 ret = disk->fops->revalidate_disk(disk);
1413
1414 /*
1415 * Hidden disks don't have associated bdev so there's no point in
1416 * revalidating it.
1417 */
1418 if (!(disk->flags & GENHD_FL_HIDDEN)) {
1419 struct block_device *bdev = bdget_disk(disk, 0);
1420
1421 if (!bdev)
1422 return ret;
1423
1424 mutex_lock(&bdev->bd_mutex);
1425 check_disk_size_change(disk, bdev, ret == 0);
1426 bdev->bd_invalidated = 0;
1427 mutex_unlock(&bdev->bd_mutex);
1428 bdput(bdev);
1429 }
1430 return ret;
1431}
1432EXPORT_SYMBOL(revalidate_disk);
1433
1434/*
1435 * This routine checks whether a removable media has been changed,
1436 * and invalidates all buffer-cache-entries in that case. This
1437 * is a relatively slow routine, so we have to try to minimize using
1438 * it. Thus it is called only upon a 'mount' or 'open'. This
1439 * is the best way of combining speed and utility, I think.
1440 * People changing diskettes in the middle of an operation deserve
1441 * to lose :-)
1442 */
1443int check_disk_change(struct block_device *bdev)
1444{
1445 struct gendisk *disk = bdev->bd_disk;
1446 const struct block_device_operations *bdops = disk->fops;
1447 unsigned int events;
1448
1449 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1450 DISK_EVENT_EJECT_REQUEST);
1451 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1452 return 0;
1453
1454 flush_disk(bdev, true);
1455 if (bdops->revalidate_disk)
1456 bdops->revalidate_disk(bdev->bd_disk);
1457 return 1;
1458}
1459
1460EXPORT_SYMBOL(check_disk_change);
1461
1462void bd_set_size(struct block_device *bdev, loff_t size)
1463{
1464 inode_lock(bdev->bd_inode);
1465 i_size_write(bdev->bd_inode, size);
1466 inode_unlock(bdev->bd_inode);
1467}
1468EXPORT_SYMBOL(bd_set_size);
1469
1470static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1471
1472/*
1473 * bd_mutex locking:
1474 *
1475 * mutex_lock(part->bd_mutex)
1476 * mutex_lock_nested(whole->bd_mutex, 1)
1477 */
1478
1479static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1480{
1481 struct gendisk *disk;
1482 int ret;
1483 int partno;
1484 int perm = 0;
1485 bool first_open = false;
1486
1487 if (mode & FMODE_READ)
1488 perm |= MAY_READ;
1489 if (mode & FMODE_WRITE)
1490 perm |= MAY_WRITE;
1491 /*
1492 * hooks: /n/, see "layering violations".
1493 */
1494 if (!for_part) {
1495 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1496 if (ret != 0) {
1497 bdput(bdev);
1498 return ret;
1499 }
1500 }
1501
1502 restart:
1503
1504 ret = -ENXIO;
1505 disk = bdev_get_gendisk(bdev, &partno);
1506 if (!disk)
1507 goto out;
1508
1509 disk_block_events(disk);
1510 mutex_lock_nested(&bdev->bd_mutex, for_part);
1511 if (!bdev->bd_openers) {
1512 first_open = true;
1513 bdev->bd_disk = disk;
1514 bdev->bd_queue = disk->queue;
1515 bdev->bd_contains = bdev;
1516 bdev->bd_partno = partno;
1517
1518 if (!partno) {
1519 ret = -ENXIO;
1520 bdev->bd_part = disk_get_part(disk, partno);
1521 if (!bdev->bd_part)
1522 goto out_clear;
1523
1524 ret = 0;
1525 if (disk->fops->open) {
1526 ret = disk->fops->open(bdev, mode);
1527 if (ret == -ERESTARTSYS) {
1528 /* Lost a race with 'disk' being
1529 * deleted, try again.
1530 * See md.c
1531 */
1532 disk_put_part(bdev->bd_part);
1533 bdev->bd_part = NULL;
1534 bdev->bd_disk = NULL;
1535 bdev->bd_queue = NULL;
1536 mutex_unlock(&bdev->bd_mutex);
1537 disk_unblock_events(disk);
1538 put_disk_and_module(disk);
1539 goto restart;
1540 }
1541 }
1542
1543 if (!ret) {
1544 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1545 set_init_blocksize(bdev);
1546 }
1547
1548 /*
1549 * If the device is invalidated, rescan partition
1550 * if open succeeded or failed with -ENOMEDIUM.
1551 * The latter is necessary to prevent ghost
1552 * partitions on a removed medium.
1553 */
1554 if (bdev->bd_invalidated) {
1555 if (!ret)
1556 rescan_partitions(disk, bdev);
1557 else if (ret == -ENOMEDIUM)
1558 invalidate_partitions(disk, bdev);
1559 }
1560
1561 if (ret)
1562 goto out_clear;
1563 } else {
1564 struct block_device *whole;
1565 whole = bdget_disk(disk, 0);
1566 ret = -ENOMEM;
1567 if (!whole)
1568 goto out_clear;
1569 BUG_ON(for_part);
1570 ret = __blkdev_get(whole, mode, 1);
1571 if (ret)
1572 goto out_clear;
1573 bdev->bd_contains = whole;
1574 bdev->bd_part = disk_get_part(disk, partno);
1575 if (!(disk->flags & GENHD_FL_UP) ||
1576 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1577 ret = -ENXIO;
1578 goto out_clear;
1579 }
1580 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1581 set_init_blocksize(bdev);
1582 }
1583
1584 if (bdev->bd_bdi == &noop_backing_dev_info)
1585 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1586 } else {
1587 if (bdev->bd_contains == bdev) {
1588 ret = 0;
1589 if (bdev->bd_disk->fops->open)
1590 ret = bdev->bd_disk->fops->open(bdev, mode);
1591 /* the same as first opener case, read comment there */
1592 if (bdev->bd_invalidated) {
1593 if (!ret)
1594 rescan_partitions(bdev->bd_disk, bdev);
1595 else if (ret == -ENOMEDIUM)
1596 invalidate_partitions(bdev->bd_disk, bdev);
1597 }
1598 if (ret)
1599 goto out_unlock_bdev;
1600 }
1601 }
1602 bdev->bd_openers++;
1603 if (for_part)
1604 bdev->bd_part_count++;
1605 mutex_unlock(&bdev->bd_mutex);
1606 disk_unblock_events(disk);
1607 /* only one opener holds refs to the module and disk */
1608 if (!first_open)
1609 put_disk_and_module(disk);
1610 return 0;
1611
1612 out_clear:
1613 disk_put_part(bdev->bd_part);
1614 bdev->bd_disk = NULL;
1615 bdev->bd_part = NULL;
1616 bdev->bd_queue = NULL;
1617 if (bdev != bdev->bd_contains)
1618 __blkdev_put(bdev->bd_contains, mode, 1);
1619 bdev->bd_contains = NULL;
1620 out_unlock_bdev:
1621 mutex_unlock(&bdev->bd_mutex);
1622 disk_unblock_events(disk);
1623 put_disk_and_module(disk);
1624 out:
1625 bdput(bdev);
1626
1627 return ret;
1628}
1629
1630/**
1631 * blkdev_get - open a block device
1632 * @bdev: block_device to open
1633 * @mode: FMODE_* mask
1634 * @holder: exclusive holder identifier
1635 *
1636 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1637 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1638 * @holder is invalid. Exclusive opens may nest for the same @holder.
1639 *
1640 * On success, the reference count of @bdev is unchanged. On failure,
1641 * @bdev is put.
1642 *
1643 * CONTEXT:
1644 * Might sleep.
1645 *
1646 * RETURNS:
1647 * 0 on success, -errno on failure.
1648 */
1649int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1650{
1651 struct block_device *whole = NULL;
1652 int res;
1653
1654 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1655
1656 if ((mode & FMODE_EXCL) && holder) {
1657 whole = bd_start_claiming(bdev, holder);
1658 if (IS_ERR(whole)) {
1659 bdput(bdev);
1660 return PTR_ERR(whole);
1661 }
1662 }
1663
1664 res = __blkdev_get(bdev, mode, 0);
1665
1666 if (whole) {
1667 struct gendisk *disk = whole->bd_disk;
1668
1669 /* finish claiming */
1670 mutex_lock(&bdev->bd_mutex);
1671 spin_lock(&bdev_lock);
1672
1673 if (!res) {
1674 BUG_ON(!bd_may_claim(bdev, whole, holder));
1675 /*
1676 * Note that for a whole device bd_holders
1677 * will be incremented twice, and bd_holder
1678 * will be set to bd_may_claim before being
1679 * set to holder
1680 */
1681 whole->bd_holders++;
1682 whole->bd_holder = bd_may_claim;
1683 bdev->bd_holders++;
1684 bdev->bd_holder = holder;
1685 }
1686
1687 /* tell others that we're done */
1688 BUG_ON(whole->bd_claiming != holder);
1689 whole->bd_claiming = NULL;
1690 wake_up_bit(&whole->bd_claiming, 0);
1691
1692 spin_unlock(&bdev_lock);
1693
1694 /*
1695 * Block event polling for write claims if requested. Any
1696 * write holder makes the write_holder state stick until
1697 * all are released. This is good enough and tracking
1698 * individual writeable reference is too fragile given the
1699 * way @mode is used in blkdev_get/put().
1700 */
1701 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1702 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1703 bdev->bd_write_holder = true;
1704 disk_block_events(disk);
1705 }
1706
1707 mutex_unlock(&bdev->bd_mutex);
1708 bdput(whole);
1709 }
1710
1711 return res;
1712}
1713EXPORT_SYMBOL(blkdev_get);
1714
1715/**
1716 * blkdev_get_by_path - open a block device by name
1717 * @path: path to the block device to open
1718 * @mode: FMODE_* mask
1719 * @holder: exclusive holder identifier
1720 *
1721 * Open the blockdevice described by the device file at @path. @mode
1722 * and @holder are identical to blkdev_get().
1723 *
1724 * On success, the returned block_device has reference count of one.
1725 *
1726 * CONTEXT:
1727 * Might sleep.
1728 *
1729 * RETURNS:
1730 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1731 */
1732struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1733 void *holder)
1734{
1735 struct block_device *bdev;
1736 int err;
1737
1738 bdev = lookup_bdev(path);
1739 if (IS_ERR(bdev))
1740 return bdev;
1741
1742 err = blkdev_get(bdev, mode, holder);
1743 if (err)
1744 return ERR_PTR(err);
1745
1746 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1747 blkdev_put(bdev, mode);
1748 return ERR_PTR(-EACCES);
1749 }
1750
1751 return bdev;
1752}
1753EXPORT_SYMBOL(blkdev_get_by_path);
1754
1755/**
1756 * blkdev_get_by_dev - open a block device by device number
1757 * @dev: device number of block device to open
1758 * @mode: FMODE_* mask
1759 * @holder: exclusive holder identifier
1760 *
1761 * Open the blockdevice described by device number @dev. @mode and
1762 * @holder are identical to blkdev_get().
1763 *
1764 * Use it ONLY if you really do not have anything better - i.e. when
1765 * you are behind a truly sucky interface and all you are given is a
1766 * device number. _Never_ to be used for internal purposes. If you
1767 * ever need it - reconsider your API.
1768 *
1769 * On success, the returned block_device has reference count of one.
1770 *
1771 * CONTEXT:
1772 * Might sleep.
1773 *
1774 * RETURNS:
1775 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1776 */
1777struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1778{
1779 struct block_device *bdev;
1780 int err;
1781
1782 bdev = bdget(dev);
1783 if (!bdev)
1784 return ERR_PTR(-ENOMEM);
1785
1786 err = blkdev_get(bdev, mode, holder);
1787 if (err)
1788 return ERR_PTR(err);
1789
1790 return bdev;
1791}
1792EXPORT_SYMBOL(blkdev_get_by_dev);
1793
1794static int blkdev_open(struct inode * inode, struct file * filp)
1795{
1796 struct block_device *bdev;
1797
1798 /*
1799 * Preserve backwards compatibility and allow large file access
1800 * even if userspace doesn't ask for it explicitly. Some mkfs
1801 * binary needs it. We might want to drop this workaround
1802 * during an unstable branch.
1803 */
1804 filp->f_flags |= O_LARGEFILE;
1805
1806 filp->f_mode |= FMODE_NOWAIT;
1807
1808 if (filp->f_flags & O_NDELAY)
1809 filp->f_mode |= FMODE_NDELAY;
1810 if (filp->f_flags & O_EXCL)
1811 filp->f_mode |= FMODE_EXCL;
1812 if ((filp->f_flags & O_ACCMODE) == 3)
1813 filp->f_mode |= FMODE_WRITE_IOCTL;
1814
1815 bdev = bd_acquire(inode);
1816 if (bdev == NULL)
1817 return -ENOMEM;
1818
1819 filp->f_mapping = bdev->bd_inode->i_mapping;
1820 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1821
1822 return blkdev_get(bdev, filp->f_mode, filp);
1823}
1824
1825static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1826{
1827 struct gendisk *disk = bdev->bd_disk;
1828 struct block_device *victim = NULL;
1829
1830 mutex_lock_nested(&bdev->bd_mutex, for_part);
1831 if (for_part)
1832 bdev->bd_part_count--;
1833
1834 if (!--bdev->bd_openers) {
1835 WARN_ON_ONCE(bdev->bd_holders);
1836 sync_blockdev(bdev);
1837 kill_bdev(bdev);
1838
1839 bdev_write_inode(bdev);
1840 }
1841 if (bdev->bd_contains == bdev) {
1842 if (disk->fops->release)
1843 disk->fops->release(disk, mode);
1844 }
1845 if (!bdev->bd_openers) {
1846 disk_put_part(bdev->bd_part);
1847 bdev->bd_part = NULL;
1848 bdev->bd_disk = NULL;
1849 if (bdev != bdev->bd_contains)
1850 victim = bdev->bd_contains;
1851 bdev->bd_contains = NULL;
1852
1853 put_disk_and_module(disk);
1854 }
1855 mutex_unlock(&bdev->bd_mutex);
1856 bdput(bdev);
1857 if (victim)
1858 __blkdev_put(victim, mode, 1);
1859}
1860
1861void blkdev_put(struct block_device *bdev, fmode_t mode)
1862{
1863 mutex_lock(&bdev->bd_mutex);
1864
1865 if (mode & FMODE_EXCL) {
1866 bool bdev_free;
1867
1868 /*
1869 * Release a claim on the device. The holder fields
1870 * are protected with bdev_lock. bd_mutex is to
1871 * synchronize disk_holder unlinking.
1872 */
1873 spin_lock(&bdev_lock);
1874
1875 WARN_ON_ONCE(--bdev->bd_holders < 0);
1876 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1877
1878 /* bd_contains might point to self, check in a separate step */
1879 if ((bdev_free = !bdev->bd_holders))
1880 bdev->bd_holder = NULL;
1881 if (!bdev->bd_contains->bd_holders)
1882 bdev->bd_contains->bd_holder = NULL;
1883
1884 spin_unlock(&bdev_lock);
1885
1886 /*
1887 * If this was the last claim, remove holder link and
1888 * unblock evpoll if it was a write holder.
1889 */
1890 if (bdev_free && bdev->bd_write_holder) {
1891 disk_unblock_events(bdev->bd_disk);
1892 bdev->bd_write_holder = false;
1893 }
1894 }
1895
1896 /*
1897 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1898 * event. This is to ensure detection of media removal commanded
1899 * from userland - e.g. eject(1).
1900 */
1901 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1902
1903 mutex_unlock(&bdev->bd_mutex);
1904
1905 __blkdev_put(bdev, mode, 0);
1906}
1907EXPORT_SYMBOL(blkdev_put);
1908
1909static int blkdev_close(struct inode * inode, struct file * filp)
1910{
1911 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1912 blkdev_put(bdev, filp->f_mode);
1913 return 0;
1914}
1915
1916static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1917{
1918 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1919 fmode_t mode = file->f_mode;
1920
1921 /*
1922 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1923 * to updated it before every ioctl.
1924 */
1925 if (file->f_flags & O_NDELAY)
1926 mode |= FMODE_NDELAY;
1927 else
1928 mode &= ~FMODE_NDELAY;
1929
1930 return blkdev_ioctl(bdev, mode, cmd, arg);
1931}
1932
1933/*
1934 * Write data to the block device. Only intended for the block device itself
1935 * and the raw driver which basically is a fake block device.
1936 *
1937 * Does not take i_mutex for the write and thus is not for general purpose
1938 * use.
1939 */
1940ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1941{
1942 struct file *file = iocb->ki_filp;
1943 struct inode *bd_inode = bdev_file_inode(file);
1944 loff_t size = i_size_read(bd_inode);
1945 struct blk_plug plug;
1946 ssize_t ret;
1947
1948 if (bdev_read_only(I_BDEV(bd_inode)))
1949 return -EPERM;
1950
1951 if (!iov_iter_count(from))
1952 return 0;
1953
1954 if (iocb->ki_pos >= size)
1955 return -ENOSPC;
1956
1957 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1958 return -EOPNOTSUPP;
1959
1960 iov_iter_truncate(from, size - iocb->ki_pos);
1961
1962 blk_start_plug(&plug);
1963 ret = __generic_file_write_iter(iocb, from);
1964 if (ret > 0)
1965 ret = generic_write_sync(iocb, ret);
1966 blk_finish_plug(&plug);
1967 return ret;
1968}
1969EXPORT_SYMBOL_GPL(blkdev_write_iter);
1970
1971ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1972{
1973 struct file *file = iocb->ki_filp;
1974 struct inode *bd_inode = bdev_file_inode(file);
1975 loff_t size = i_size_read(bd_inode);
1976 loff_t pos = iocb->ki_pos;
1977
1978 if (pos >= size)
1979 return 0;
1980
1981 size -= pos;
1982 iov_iter_truncate(to, size);
1983 return generic_file_read_iter(iocb, to);
1984}
1985EXPORT_SYMBOL_GPL(blkdev_read_iter);
1986
1987/*
1988 * Try to release a page associated with block device when the system
1989 * is under memory pressure.
1990 */
1991static int blkdev_releasepage(struct page *page, gfp_t wait)
1992{
1993 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1994
1995 if (super && super->s_op->bdev_try_to_free_page)
1996 return super->s_op->bdev_try_to_free_page(super, page, wait);
1997
1998 return try_to_free_buffers(page);
1999}
2000
2001static int blkdev_writepages(struct address_space *mapping,
2002 struct writeback_control *wbc)
2003{
2004 return generic_writepages(mapping, wbc);
2005}
2006
2007static const struct address_space_operations def_blk_aops = {
2008 .readpage = blkdev_readpage,
2009 .readpages = blkdev_readpages,
2010 .writepage = blkdev_writepage,
2011 .write_begin = blkdev_write_begin,
2012 .write_end = blkdev_write_end,
2013 .writepages = blkdev_writepages,
2014 .releasepage = blkdev_releasepage,
2015 .direct_IO = blkdev_direct_IO,
2016 .migratepage = buffer_migrate_page_norefs,
2017 .is_dirty_writeback = buffer_check_dirty_writeback,
2018};
2019
2020#define BLKDEV_FALLOC_FL_SUPPORTED \
2021 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2022 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2023
2024static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2025 loff_t len)
2026{
2027 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2028 struct address_space *mapping;
2029 loff_t end = start + len - 1;
2030 loff_t isize;
2031 int error;
2032
2033 /* Fail if we don't recognize the flags. */
2034 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2035 return -EOPNOTSUPP;
2036
2037 /* Don't go off the end of the device. */
2038 isize = i_size_read(bdev->bd_inode);
2039 if (start >= isize)
2040 return -EINVAL;
2041 if (end >= isize) {
2042 if (mode & FALLOC_FL_KEEP_SIZE) {
2043 len = isize - start;
2044 end = start + len - 1;
2045 } else
2046 return -EINVAL;
2047 }
2048
2049 /*
2050 * Don't allow IO that isn't aligned to logical block size.
2051 */
2052 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2053 return -EINVAL;
2054
2055 /* Invalidate the page cache, including dirty pages. */
2056 mapping = bdev->bd_inode->i_mapping;
2057 truncate_inode_pages_range(mapping, start, end);
2058
2059 switch (mode) {
2060 case FALLOC_FL_ZERO_RANGE:
2061 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2062 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2063 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2064 break;
2065 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2066 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2067 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2068 break;
2069 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2070 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2071 GFP_KERNEL, 0);
2072 break;
2073 default:
2074 return -EOPNOTSUPP;
2075 }
2076 if (error)
2077 return error;
2078
2079 /*
2080 * Invalidate again; if someone wandered in and dirtied a page,
2081 * the caller will be given -EBUSY. The third argument is
2082 * inclusive, so the rounding here is safe.
2083 */
2084 return invalidate_inode_pages2_range(mapping,
2085 start >> PAGE_SHIFT,
2086 end >> PAGE_SHIFT);
2087}
2088
2089const struct file_operations def_blk_fops = {
2090 .open = blkdev_open,
2091 .release = blkdev_close,
2092 .llseek = block_llseek,
2093 .read_iter = blkdev_read_iter,
2094 .write_iter = blkdev_write_iter,
2095 .iopoll = blkdev_iopoll,
2096 .mmap = generic_file_mmap,
2097 .fsync = blkdev_fsync,
2098 .unlocked_ioctl = block_ioctl,
2099#ifdef CONFIG_COMPAT
2100 .compat_ioctl = compat_blkdev_ioctl,
2101#endif
2102 .splice_read = generic_file_splice_read,
2103 .splice_write = iter_file_splice_write,
2104 .fallocate = blkdev_fallocate,
2105};
2106
2107int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2108{
2109 int res;
2110 mm_segment_t old_fs = get_fs();
2111 set_fs(KERNEL_DS);
2112 res = blkdev_ioctl(bdev, 0, cmd, arg);
2113 set_fs(old_fs);
2114 return res;
2115}
2116
2117EXPORT_SYMBOL(ioctl_by_bdev);
2118
2119/**
2120 * lookup_bdev - lookup a struct block_device by name
2121 * @pathname: special file representing the block device
2122 *
2123 * Get a reference to the blockdevice at @pathname in the current
2124 * namespace if possible and return it. Return ERR_PTR(error)
2125 * otherwise.
2126 */
2127struct block_device *lookup_bdev(const char *pathname)
2128{
2129 struct block_device *bdev;
2130 struct inode *inode;
2131 struct path path;
2132 int error;
2133
2134 if (!pathname || !*pathname)
2135 return ERR_PTR(-EINVAL);
2136
2137 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2138 if (error)
2139 return ERR_PTR(error);
2140
2141 inode = d_backing_inode(path.dentry);
2142 error = -ENOTBLK;
2143 if (!S_ISBLK(inode->i_mode))
2144 goto fail;
2145 error = -EACCES;
2146 if (!may_open_dev(&path))
2147 goto fail;
2148 error = -ENOMEM;
2149 bdev = bd_acquire(inode);
2150 if (!bdev)
2151 goto fail;
2152out:
2153 path_put(&path);
2154 return bdev;
2155fail:
2156 bdev = ERR_PTR(error);
2157 goto out;
2158}
2159EXPORT_SYMBOL(lookup_bdev);
2160
2161int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2162{
2163 struct super_block *sb = get_super(bdev);
2164 int res = 0;
2165
2166 if (sb) {
2167 /*
2168 * no need to lock the super, get_super holds the
2169 * read mutex so the filesystem cannot go away
2170 * under us (->put_super runs with the write lock
2171 * hold).
2172 */
2173 shrink_dcache_sb(sb);
2174 res = invalidate_inodes(sb, kill_dirty);
2175 drop_super(sb);
2176 }
2177 invalidate_bdev(bdev);
2178 return res;
2179}
2180EXPORT_SYMBOL(__invalidate_device);
2181
2182void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2183{
2184 struct inode *inode, *old_inode = NULL;
2185
2186 spin_lock(&blockdev_superblock->s_inode_list_lock);
2187 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2188 struct address_space *mapping = inode->i_mapping;
2189 struct block_device *bdev;
2190
2191 spin_lock(&inode->i_lock);
2192 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2193 mapping->nrpages == 0) {
2194 spin_unlock(&inode->i_lock);
2195 continue;
2196 }
2197 __iget(inode);
2198 spin_unlock(&inode->i_lock);
2199 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2200 /*
2201 * We hold a reference to 'inode' so it couldn't have been
2202 * removed from s_inodes list while we dropped the
2203 * s_inode_list_lock We cannot iput the inode now as we can
2204 * be holding the last reference and we cannot iput it under
2205 * s_inode_list_lock. So we keep the reference and iput it
2206 * later.
2207 */
2208 iput(old_inode);
2209 old_inode = inode;
2210 bdev = I_BDEV(inode);
2211
2212 mutex_lock(&bdev->bd_mutex);
2213 if (bdev->bd_openers)
2214 func(bdev, arg);
2215 mutex_unlock(&bdev->bd_mutex);
2216
2217 spin_lock(&blockdev_superblock->s_inode_list_lock);
2218 }
2219 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2220 iput(old_inode);
2221}