1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992, 2002 Linus Torvalds
9 * Start bdflush() with kernel_thread not syscall - Paul Gortmaker, 12/95
11 * Removed a lot of unnecessary code and simplified things now that
12 * the buffer cache isn't our primary cache - Andrew Tridgell 12/96
14 * Speed up hash, lru, and free list operations. Use gfp() for allocating
15 * hash table, use SLAB cache for buffer heads. SMP threading. -DaveM
17 * Added 32k buffer block sizes - these are required older ARM systems. - RMK
19 * async buffer flushing, 1999 Andrea Arcangeli <andrea@suse.de>
22 #include <linux/kernel.h>
23 #include <linux/sched/signal.h>
24 #include <linux/syscalls.h>
26 #include <linux/iomap.h>
28 #include <linux/percpu.h>
29 #include <linux/slab.h>
30 #include <linux/capability.h>
31 #include <linux/blkdev.h>
32 #include <linux/file.h>
33 #include <linux/quotaops.h>
34 #include <linux/highmem.h>
35 #include <linux/export.h>
36 #include <linux/backing-dev.h>
37 #include <linux/writeback.h>
38 #include <linux/hash.h>
39 #include <linux/suspend.h>
40 #include <linux/buffer_head.h>
41 #include <linux/task_io_accounting_ops.h>
42 #include <linux/bio.h>
43 #include <linux/cpu.h>
44 #include <linux/bitops.h>
45 #include <linux/mpage.h>
46 #include <linux/bit_spinlock.h>
47 #include <linux/pagevec.h>
48 #include <linux/sched/mm.h>
49 #include <trace/events/block.h>
50 #include <linux/fscrypt.h>
51 #include <linux/fsverity.h>
55 static int fsync_buffers_list(spinlock_t
*lock
, struct list_head
*list
);
56 static void submit_bh_wbc(blk_opf_t opf
, struct buffer_head
*bh
,
57 struct writeback_control
*wbc
);
59 #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
61 inline void touch_buffer(struct buffer_head
*bh
)
63 trace_block_touch_buffer(bh
);
64 folio_mark_accessed(bh
->b_folio
);
66 EXPORT_SYMBOL(touch_buffer
);
68 void __lock_buffer(struct buffer_head
*bh
)
70 wait_on_bit_lock_io(&bh
->b_state
, BH_Lock
, TASK_UNINTERRUPTIBLE
);
72 EXPORT_SYMBOL(__lock_buffer
);
74 void unlock_buffer(struct buffer_head
*bh
)
76 clear_bit_unlock(BH_Lock
, &bh
->b_state
);
77 smp_mb__after_atomic();
78 wake_up_bit(&bh
->b_state
, BH_Lock
);
80 EXPORT_SYMBOL(unlock_buffer
);
83 * Returns if the folio has dirty or writeback buffers. If all the buffers
84 * are unlocked and clean then the folio_test_dirty information is stale. If
85 * any of the buffers are locked, it is assumed they are locked for IO.
87 void buffer_check_dirty_writeback(struct folio
*folio
,
88 bool *dirty
, bool *writeback
)
90 struct buffer_head
*head
, *bh
;
94 BUG_ON(!folio_test_locked(folio
));
96 head
= folio_buffers(folio
);
100 if (folio_test_writeback(folio
))
105 if (buffer_locked(bh
))
108 if (buffer_dirty(bh
))
111 bh
= bh
->b_this_page
;
112 } while (bh
!= head
);
114 EXPORT_SYMBOL(buffer_check_dirty_writeback
);
117 * Block until a buffer comes unlocked. This doesn't stop it
118 * from becoming locked again - you have to lock it yourself
119 * if you want to preserve its state.
121 void __wait_on_buffer(struct buffer_head
* bh
)
123 wait_on_bit_io(&bh
->b_state
, BH_Lock
, TASK_UNINTERRUPTIBLE
);
125 EXPORT_SYMBOL(__wait_on_buffer
);
127 static void buffer_io_error(struct buffer_head
*bh
, char *msg
)
129 if (!test_bit(BH_Quiet
, &bh
->b_state
))
130 printk_ratelimited(KERN_ERR
131 "Buffer I/O error on dev %pg, logical block %llu%s\n",
132 bh
->b_bdev
, (unsigned long long)bh
->b_blocknr
, msg
);
136 * End-of-IO handler helper function which does not touch the bh after
138 * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
139 * a race there is benign: unlock_buffer() only use the bh's address for
140 * hashing after unlocking the buffer, so it doesn't actually touch the bh
143 static void __end_buffer_read_notouch(struct buffer_head
*bh
, int uptodate
)
146 set_buffer_uptodate(bh
);
148 /* This happens, due to failed read-ahead attempts. */
149 clear_buffer_uptodate(bh
);
155 * Default synchronous end-of-IO handler.. Just mark it up-to-date and
158 void end_buffer_read_sync(struct buffer_head
*bh
, int uptodate
)
160 __end_buffer_read_notouch(bh
, uptodate
);
163 EXPORT_SYMBOL(end_buffer_read_sync
);
165 void end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
168 set_buffer_uptodate(bh
);
170 buffer_io_error(bh
, ", lost sync page write");
171 mark_buffer_write_io_error(bh
);
172 clear_buffer_uptodate(bh
);
177 EXPORT_SYMBOL(end_buffer_write_sync
);
180 * Various filesystems appear to want __find_get_block to be non-blocking.
181 * But it's the page lock which protects the buffers. To get around this,
182 * we get exclusion from try_to_free_buffers with the blockdev mapping's
185 * Hack idea: for the blockdev mapping, private_lock contention
186 * may be quite high. This code could TryLock the page, and if that
187 * succeeds, there is no need to take private_lock.
189 static struct buffer_head
*
190 __find_get_block_slow(struct block_device
*bdev
, sector_t block
)
192 struct inode
*bd_inode
= bdev
->bd_inode
;
193 struct address_space
*bd_mapping
= bd_inode
->i_mapping
;
194 struct buffer_head
*ret
= NULL
;
196 struct buffer_head
*bh
;
197 struct buffer_head
*head
;
200 static DEFINE_RATELIMIT_STATE(last_warned
, HZ
, 1);
202 index
= block
>> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
203 page
= find_get_page_flags(bd_mapping
, index
, FGP_ACCESSED
);
207 spin_lock(&bd_mapping
->private_lock
);
208 if (!page_has_buffers(page
))
210 head
= page_buffers(page
);
213 if (!buffer_mapped(bh
))
215 else if (bh
->b_blocknr
== block
) {
220 bh
= bh
->b_this_page
;
221 } while (bh
!= head
);
223 /* we might be here because some of the buffers on this page are
224 * not mapped. This is due to various races between
225 * file io on the block device and getblk. It gets dealt with
226 * elsewhere, don't buffer_error if we had some unmapped buffers
228 ratelimit_set_flags(&last_warned
, RATELIMIT_MSG_ON_RELEASE
);
229 if (all_mapped
&& __ratelimit(&last_warned
)) {
230 printk("__find_get_block_slow() failed. block=%llu, "
231 "b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, "
232 "device %pg blocksize: %d\n",
233 (unsigned long long)block
,
234 (unsigned long long)bh
->b_blocknr
,
235 bh
->b_state
, bh
->b_size
, bdev
,
236 1 << bd_inode
->i_blkbits
);
239 spin_unlock(&bd_mapping
->private_lock
);
245 static void end_buffer_async_read(struct buffer_head
*bh
, int uptodate
)
248 struct buffer_head
*first
;
249 struct buffer_head
*tmp
;
251 int folio_uptodate
= 1;
253 BUG_ON(!buffer_async_read(bh
));
257 set_buffer_uptodate(bh
);
259 clear_buffer_uptodate(bh
);
260 buffer_io_error(bh
, ", async page read");
261 folio_set_error(folio
);
265 * Be _very_ careful from here on. Bad things can happen if
266 * two buffer heads end IO at almost the same time and both
267 * decide that the page is now completely done.
269 first
= folio_buffers(folio
);
270 spin_lock_irqsave(&first
->b_uptodate_lock
, flags
);
271 clear_buffer_async_read(bh
);
275 if (!buffer_uptodate(tmp
))
277 if (buffer_async_read(tmp
)) {
278 BUG_ON(!buffer_locked(tmp
));
281 tmp
= tmp
->b_this_page
;
283 spin_unlock_irqrestore(&first
->b_uptodate_lock
, flags
);
286 * If all of the buffers are uptodate then we can set the page
290 folio_mark_uptodate(folio
);
295 spin_unlock_irqrestore(&first
->b_uptodate_lock
, flags
);
299 struct postprocess_bh_ctx
{
300 struct work_struct work
;
301 struct buffer_head
*bh
;
304 static void verify_bh(struct work_struct
*work
)
306 struct postprocess_bh_ctx
*ctx
=
307 container_of(work
, struct postprocess_bh_ctx
, work
);
308 struct buffer_head
*bh
= ctx
->bh
;
311 valid
= fsverity_verify_blocks(page_folio(bh
->b_page
), bh
->b_size
,
313 end_buffer_async_read(bh
, valid
);
317 static bool need_fsverity(struct buffer_head
*bh
)
319 struct page
*page
= bh
->b_page
;
320 struct inode
*inode
= page
->mapping
->host
;
322 return fsverity_active(inode
) &&
324 page
->index
< DIV_ROUND_UP(inode
->i_size
, PAGE_SIZE
);
327 static void decrypt_bh(struct work_struct
*work
)
329 struct postprocess_bh_ctx
*ctx
=
330 container_of(work
, struct postprocess_bh_ctx
, work
);
331 struct buffer_head
*bh
= ctx
->bh
;
334 err
= fscrypt_decrypt_pagecache_blocks(page_folio(bh
->b_page
),
335 bh
->b_size
, bh_offset(bh
));
336 if (err
== 0 && need_fsverity(bh
)) {
338 * We use different work queues for decryption and for verity
339 * because verity may require reading metadata pages that need
340 * decryption, and we shouldn't recurse to the same workqueue.
342 INIT_WORK(&ctx
->work
, verify_bh
);
343 fsverity_enqueue_verify_work(&ctx
->work
);
346 end_buffer_async_read(bh
, err
== 0);
351 * I/O completion handler for block_read_full_folio() - pages
352 * which come unlocked at the end of I/O.
354 static void end_buffer_async_read_io(struct buffer_head
*bh
, int uptodate
)
356 struct inode
*inode
= bh
->b_folio
->mapping
->host
;
357 bool decrypt
= fscrypt_inode_uses_fs_layer_crypto(inode
);
358 bool verify
= need_fsverity(bh
);
360 /* Decrypt (with fscrypt) and/or verify (with fsverity) if needed. */
361 if (uptodate
&& (decrypt
|| verify
)) {
362 struct postprocess_bh_ctx
*ctx
=
363 kmalloc(sizeof(*ctx
), GFP_ATOMIC
);
368 INIT_WORK(&ctx
->work
, decrypt_bh
);
369 fscrypt_enqueue_decrypt_work(&ctx
->work
);
371 INIT_WORK(&ctx
->work
, verify_bh
);
372 fsverity_enqueue_verify_work(&ctx
->work
);
378 end_buffer_async_read(bh
, uptodate
);
382 * Completion handler for block_write_full_page() - pages which are unlocked
383 * during I/O, and which have PageWriteback cleared upon I/O completion.
385 void end_buffer_async_write(struct buffer_head
*bh
, int uptodate
)
388 struct buffer_head
*first
;
389 struct buffer_head
*tmp
;
392 BUG_ON(!buffer_async_write(bh
));
396 set_buffer_uptodate(bh
);
398 buffer_io_error(bh
, ", lost async page write");
399 mark_buffer_write_io_error(bh
);
400 clear_buffer_uptodate(bh
);
401 folio_set_error(folio
);
404 first
= folio_buffers(folio
);
405 spin_lock_irqsave(&first
->b_uptodate_lock
, flags
);
407 clear_buffer_async_write(bh
);
409 tmp
= bh
->b_this_page
;
411 if (buffer_async_write(tmp
)) {
412 BUG_ON(!buffer_locked(tmp
));
415 tmp
= tmp
->b_this_page
;
417 spin_unlock_irqrestore(&first
->b_uptodate_lock
, flags
);
418 folio_end_writeback(folio
);
422 spin_unlock_irqrestore(&first
->b_uptodate_lock
, flags
);
425 EXPORT_SYMBOL(end_buffer_async_write
);
428 * If a page's buffers are under async readin (end_buffer_async_read
429 * completion) then there is a possibility that another thread of
430 * control could lock one of the buffers after it has completed
431 * but while some of the other buffers have not completed. This
432 * locked buffer would confuse end_buffer_async_read() into not unlocking
433 * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
434 * that this buffer is not under async I/O.
436 * The page comes unlocked when it has no locked buffer_async buffers
439 * PageLocked prevents anyone starting new async I/O reads any of
442 * PageWriteback is used to prevent simultaneous writeout of the same
445 * PageLocked prevents anyone from starting writeback of a page which is
446 * under read I/O (PageWriteback is only ever set against a locked page).
448 static void mark_buffer_async_read(struct buffer_head
*bh
)
450 bh
->b_end_io
= end_buffer_async_read_io
;
451 set_buffer_async_read(bh
);
454 static void mark_buffer_async_write_endio(struct buffer_head
*bh
,
455 bh_end_io_t
*handler
)
457 bh
->b_end_io
= handler
;
458 set_buffer_async_write(bh
);
461 void mark_buffer_async_write(struct buffer_head
*bh
)
463 mark_buffer_async_write_endio(bh
, end_buffer_async_write
);
465 EXPORT_SYMBOL(mark_buffer_async_write
);
469 * fs/buffer.c contains helper functions for buffer-backed address space's
470 * fsync functions. A common requirement for buffer-based filesystems is
471 * that certain data from the backing blockdev needs to be written out for
472 * a successful fsync(). For example, ext2 indirect blocks need to be
473 * written back and waited upon before fsync() returns.
475 * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
476 * inode_has_buffers() and invalidate_inode_buffers() are provided for the
477 * management of a list of dependent buffers at ->i_mapping->private_list.
479 * Locking is a little subtle: try_to_free_buffers() will remove buffers
480 * from their controlling inode's queue when they are being freed. But
481 * try_to_free_buffers() will be operating against the *blockdev* mapping
482 * at the time, not against the S_ISREG file which depends on those buffers.
483 * So the locking for private_list is via the private_lock in the address_space
484 * which backs the buffers. Which is different from the address_space
485 * against which the buffers are listed. So for a particular address_space,
486 * mapping->private_lock does *not* protect mapping->private_list! In fact,
487 * mapping->private_list will always be protected by the backing blockdev's
490 * Which introduces a requirement: all buffers on an address_space's
491 * ->private_list must be from the same address_space: the blockdev's.
493 * address_spaces which do not place buffers at ->private_list via these
494 * utility functions are free to use private_lock and private_list for
495 * whatever they want. The only requirement is that list_empty(private_list)
496 * be true at clear_inode() time.
498 * FIXME: clear_inode should not call invalidate_inode_buffers(). The
499 * filesystems should do that. invalidate_inode_buffers() should just go
500 * BUG_ON(!list_empty).
502 * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
503 * take an address_space, not an inode. And it should be called
504 * mark_buffer_dirty_fsync() to clearly define why those buffers are being
507 * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
508 * list if it is already on a list. Because if the buffer is on a list,
509 * it *must* already be on the right one. If not, the filesystem is being
510 * silly. This will save a ton of locking. But first we have to ensure
511 * that buffers are taken *off* the old inode's list when they are freed
512 * (presumably in truncate). That requires careful auditing of all
513 * filesystems (do it inside bforget()). It could also be done by bringing
518 * The buffer's backing address_space's private_lock must be held
520 static void __remove_assoc_queue(struct buffer_head
*bh
)
522 list_del_init(&bh
->b_assoc_buffers
);
523 WARN_ON(!bh
->b_assoc_map
);
524 bh
->b_assoc_map
= NULL
;
527 int inode_has_buffers(struct inode
*inode
)
529 return !list_empty(&inode
->i_data
.private_list
);
533 * osync is designed to support O_SYNC io. It waits synchronously for
534 * all already-submitted IO to complete, but does not queue any new
535 * writes to the disk.
537 * To do O_SYNC writes, just queue the buffer writes with write_dirty_buffer
538 * as you dirty the buffers, and then use osync_inode_buffers to wait for
539 * completion. Any other dirty buffers which are not yet queued for
540 * write will not be flushed to disk by the osync.
542 static int osync_buffers_list(spinlock_t
*lock
, struct list_head
*list
)
544 struct buffer_head
*bh
;
550 list_for_each_prev(p
, list
) {
552 if (buffer_locked(bh
)) {
556 if (!buffer_uptodate(bh
))
567 void emergency_thaw_bdev(struct super_block
*sb
)
569 while (sb
->s_bdev
&& !thaw_bdev(sb
->s_bdev
))
570 printk(KERN_WARNING
"Emergency Thaw on %pg\n", sb
->s_bdev
);
574 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
575 * @mapping: the mapping which wants those buffers written
577 * Starts I/O against the buffers at mapping->private_list, and waits upon
580 * Basically, this is a convenience function for fsync().
581 * @mapping is a file or directory which needs those buffers to be written for
582 * a successful fsync().
584 int sync_mapping_buffers(struct address_space
*mapping
)
586 struct address_space
*buffer_mapping
= mapping
->private_data
;
588 if (buffer_mapping
== NULL
|| list_empty(&mapping
->private_list
))
591 return fsync_buffers_list(&buffer_mapping
->private_lock
,
592 &mapping
->private_list
);
594 EXPORT_SYMBOL(sync_mapping_buffers
);
597 * Called when we've recently written block `bblock', and it is known that
598 * `bblock' was for a buffer_boundary() buffer. This means that the block at
599 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
600 * dirty, schedule it for IO. So that indirects merge nicely with their data.
602 void write_boundary_block(struct block_device
*bdev
,
603 sector_t bblock
, unsigned blocksize
)
605 struct buffer_head
*bh
= __find_get_block(bdev
, bblock
+ 1, blocksize
);
607 if (buffer_dirty(bh
))
608 write_dirty_buffer(bh
, 0);
613 void mark_buffer_dirty_inode(struct buffer_head
*bh
, struct inode
*inode
)
615 struct address_space
*mapping
= inode
->i_mapping
;
616 struct address_space
*buffer_mapping
= bh
->b_folio
->mapping
;
618 mark_buffer_dirty(bh
);
619 if (!mapping
->private_data
) {
620 mapping
->private_data
= buffer_mapping
;
622 BUG_ON(mapping
->private_data
!= buffer_mapping
);
624 if (!bh
->b_assoc_map
) {
625 spin_lock(&buffer_mapping
->private_lock
);
626 list_move_tail(&bh
->b_assoc_buffers
,
627 &mapping
->private_list
);
628 bh
->b_assoc_map
= mapping
;
629 spin_unlock(&buffer_mapping
->private_lock
);
632 EXPORT_SYMBOL(mark_buffer_dirty_inode
);
635 * Add a page to the dirty page list.
637 * It is a sad fact of life that this function is called from several places
638 * deeply under spinlocking. It may not sleep.
640 * If the page has buffers, the uptodate buffers are set dirty, to preserve
641 * dirty-state coherency between the page and the buffers. It the page does
642 * not have buffers then when they are later attached they will all be set
645 * The buffers are dirtied before the page is dirtied. There's a small race
646 * window in which a writepage caller may see the page cleanness but not the
647 * buffer dirtiness. That's fine. If this code were to set the page dirty
648 * before the buffers, a concurrent writepage caller could clear the page dirty
649 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
650 * page on the dirty page list.
652 * We use private_lock to lock against try_to_free_buffers while using the
653 * page's buffer list. Also use this to protect against clean buffers being
654 * added to the page after it was set dirty.
656 * FIXME: may need to call ->reservepage here as well. That's rather up to the
657 * address_space though.
659 bool block_dirty_folio(struct address_space
*mapping
, struct folio
*folio
)
661 struct buffer_head
*head
;
664 spin_lock(&mapping
->private_lock
);
665 head
= folio_buffers(folio
);
667 struct buffer_head
*bh
= head
;
670 set_buffer_dirty(bh
);
671 bh
= bh
->b_this_page
;
672 } while (bh
!= head
);
675 * Lock out page's memcg migration to keep PageDirty
676 * synchronized with per-memcg dirty page counters.
678 folio_memcg_lock(folio
);
679 newly_dirty
= !folio_test_set_dirty(folio
);
680 spin_unlock(&mapping
->private_lock
);
683 __folio_mark_dirty(folio
, mapping
, 1);
685 folio_memcg_unlock(folio
);
688 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
692 EXPORT_SYMBOL(block_dirty_folio
);
695 * Write out and wait upon a list of buffers.
697 * We have conflicting pressures: we want to make sure that all
698 * initially dirty buffers get waited on, but that any subsequently
699 * dirtied buffers don't. After all, we don't want fsync to last
700 * forever if somebody is actively writing to the file.
702 * Do this in two main stages: first we copy dirty buffers to a
703 * temporary inode list, queueing the writes as we go. Then we clean
704 * up, waiting for those writes to complete.
706 * During this second stage, any subsequent updates to the file may end
707 * up refiling the buffer on the original inode's dirty list again, so
708 * there is a chance we will end up with a buffer queued for write but
709 * not yet completed on that list. So, as a final cleanup we go through
710 * the osync code to catch these locked, dirty buffers without requeuing
711 * any newly dirty buffers for write.
713 static int fsync_buffers_list(spinlock_t
*lock
, struct list_head
*list
)
715 struct buffer_head
*bh
;
716 struct list_head tmp
;
717 struct address_space
*mapping
;
719 struct blk_plug plug
;
721 INIT_LIST_HEAD(&tmp
);
722 blk_start_plug(&plug
);
725 while (!list_empty(list
)) {
726 bh
= BH_ENTRY(list
->next
);
727 mapping
= bh
->b_assoc_map
;
728 __remove_assoc_queue(bh
);
729 /* Avoid race with mark_buffer_dirty_inode() which does
730 * a lockless check and we rely on seeing the dirty bit */
732 if (buffer_dirty(bh
) || buffer_locked(bh
)) {
733 list_add(&bh
->b_assoc_buffers
, &tmp
);
734 bh
->b_assoc_map
= mapping
;
735 if (buffer_dirty(bh
)) {
739 * Ensure any pending I/O completes so that
740 * write_dirty_buffer() actually writes the
741 * current contents - it is a noop if I/O is
742 * still in flight on potentially older
745 write_dirty_buffer(bh
, REQ_SYNC
);
748 * Kick off IO for the previous mapping. Note
749 * that we will not run the very last mapping,
750 * wait_on_buffer() will do that for us
751 * through sync_buffer().
760 blk_finish_plug(&plug
);
763 while (!list_empty(&tmp
)) {
764 bh
= BH_ENTRY(tmp
.prev
);
766 mapping
= bh
->b_assoc_map
;
767 __remove_assoc_queue(bh
);
768 /* Avoid race with mark_buffer_dirty_inode() which does
769 * a lockless check and we rely on seeing the dirty bit */
771 if (buffer_dirty(bh
)) {
772 list_add(&bh
->b_assoc_buffers
,
773 &mapping
->private_list
);
774 bh
->b_assoc_map
= mapping
;
778 if (!buffer_uptodate(bh
))
785 err2
= osync_buffers_list(lock
, list
);
793 * Invalidate any and all dirty buffers on a given inode. We are
794 * probably unmounting the fs, but that doesn't mean we have already
795 * done a sync(). Just drop the buffers from the inode list.
797 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
798 * assumes that all the buffers are against the blockdev. Not true
801 void invalidate_inode_buffers(struct inode
*inode
)
803 if (inode_has_buffers(inode
)) {
804 struct address_space
*mapping
= &inode
->i_data
;
805 struct list_head
*list
= &mapping
->private_list
;
806 struct address_space
*buffer_mapping
= mapping
->private_data
;
808 spin_lock(&buffer_mapping
->private_lock
);
809 while (!list_empty(list
))
810 __remove_assoc_queue(BH_ENTRY(list
->next
));
811 spin_unlock(&buffer_mapping
->private_lock
);
814 EXPORT_SYMBOL(invalidate_inode_buffers
);
817 * Remove any clean buffers from the inode's buffer list. This is called
818 * when we're trying to free the inode itself. Those buffers can pin it.
820 * Returns true if all buffers were removed.
822 int remove_inode_buffers(struct inode
*inode
)
826 if (inode_has_buffers(inode
)) {
827 struct address_space
*mapping
= &inode
->i_data
;
828 struct list_head
*list
= &mapping
->private_list
;
829 struct address_space
*buffer_mapping
= mapping
->private_data
;
831 spin_lock(&buffer_mapping
->private_lock
);
832 while (!list_empty(list
)) {
833 struct buffer_head
*bh
= BH_ENTRY(list
->next
);
834 if (buffer_dirty(bh
)) {
838 __remove_assoc_queue(bh
);
840 spin_unlock(&buffer_mapping
->private_lock
);
846 * Create the appropriate buffers when given a page for data area and
847 * the size of each buffer.. Use the bh->b_this_page linked list to
848 * follow the buffers created. Return NULL if unable to create more
851 * The retry flag is used to differentiate async IO (paging, swapping)
852 * which may not fail from ordinary buffer allocations.
854 struct buffer_head
*alloc_page_buffers(struct page
*page
, unsigned long size
,
857 struct buffer_head
*bh
, *head
;
858 gfp_t gfp
= GFP_NOFS
| __GFP_ACCOUNT
;
860 struct mem_cgroup
*memcg
, *old_memcg
;
865 /* The page lock pins the memcg */
866 memcg
= page_memcg(page
);
867 old_memcg
= set_active_memcg(memcg
);
871 while ((offset
-= size
) >= 0) {
872 bh
= alloc_buffer_head(gfp
);
876 bh
->b_this_page
= head
;
882 /* Link the buffer to its page */
883 set_bh_page(bh
, page
, offset
);
886 set_active_memcg(old_memcg
);
889 * In case anything failed, we just free everything we got.
895 head
= head
->b_this_page
;
896 free_buffer_head(bh
);
902 EXPORT_SYMBOL_GPL(alloc_page_buffers
);
905 link_dev_buffers(struct page
*page
, struct buffer_head
*head
)
907 struct buffer_head
*bh
, *tail
;
912 bh
= bh
->b_this_page
;
914 tail
->b_this_page
= head
;
915 attach_page_private(page
, head
);
918 static sector_t
blkdev_max_block(struct block_device
*bdev
, unsigned int size
)
920 sector_t retval
= ~((sector_t
)0);
921 loff_t sz
= bdev_nr_bytes(bdev
);
924 unsigned int sizebits
= blksize_bits(size
);
925 retval
= (sz
>> sizebits
);
931 * Initialise the state of a blockdev page's buffers.
934 init_page_buffers(struct page
*page
, struct block_device
*bdev
,
935 sector_t block
, int size
)
937 struct buffer_head
*head
= page_buffers(page
);
938 struct buffer_head
*bh
= head
;
939 int uptodate
= PageUptodate(page
);
940 sector_t end_block
= blkdev_max_block(bdev
, size
);
943 if (!buffer_mapped(bh
)) {
945 bh
->b_private
= NULL
;
947 bh
->b_blocknr
= block
;
949 set_buffer_uptodate(bh
);
950 if (block
< end_block
)
951 set_buffer_mapped(bh
);
954 bh
= bh
->b_this_page
;
955 } while (bh
!= head
);
958 * Caller needs to validate requested block against end of device.
964 * Create the page-cache page that contains the requested block.
966 * This is used purely for blockdev mappings.
969 grow_dev_page(struct block_device
*bdev
, sector_t block
,
970 pgoff_t index
, int size
, int sizebits
, gfp_t gfp
)
972 struct inode
*inode
= bdev
->bd_inode
;
974 struct buffer_head
*bh
;
979 gfp_mask
= mapping_gfp_constraint(inode
->i_mapping
, ~__GFP_FS
) | gfp
;
982 * XXX: __getblk_slow() can not really deal with failure and
983 * will endlessly loop on improvised global reclaim. Prefer
984 * looping in the allocator rather than here, at least that
985 * code knows what it's doing.
987 gfp_mask
|= __GFP_NOFAIL
;
989 page
= find_or_create_page(inode
->i_mapping
, index
, gfp_mask
);
991 BUG_ON(!PageLocked(page
));
993 if (page_has_buffers(page
)) {
994 bh
= page_buffers(page
);
995 if (bh
->b_size
== size
) {
996 end_block
= init_page_buffers(page
, bdev
,
997 (sector_t
)index
<< sizebits
,
1001 if (!try_to_free_buffers(page_folio(page
)))
1006 * Allocate some buffers for this page
1008 bh
= alloc_page_buffers(page
, size
, true);
1011 * Link the page to the buffers and initialise them. Take the
1012 * lock to be atomic wrt __find_get_block(), which does not
1013 * run under the page lock.
1015 spin_lock(&inode
->i_mapping
->private_lock
);
1016 link_dev_buffers(page
, bh
);
1017 end_block
= init_page_buffers(page
, bdev
, (sector_t
)index
<< sizebits
,
1019 spin_unlock(&inode
->i_mapping
->private_lock
);
1021 ret
= (block
< end_block
) ? 1 : -ENXIO
;
1029 * Create buffers for the specified block device block's page. If
1030 * that page was dirty, the buffers are set dirty also.
1033 grow_buffers(struct block_device
*bdev
, sector_t block
, int size
, gfp_t gfp
)
1038 sizebits
= PAGE_SHIFT
- __ffs(size
);
1039 index
= block
>> sizebits
;
1042 * Check for a block which wants to lie outside our maximum possible
1043 * pagecache index. (this comparison is done using sector_t types).
1045 if (unlikely(index
!= block
>> sizebits
)) {
1046 printk(KERN_ERR
"%s: requested out-of-range block %llu for "
1048 __func__
, (unsigned long long)block
,
1053 /* Create a page with the proper size buffers.. */
1054 return grow_dev_page(bdev
, block
, index
, size
, sizebits
, gfp
);
1057 static struct buffer_head
*
1058 __getblk_slow(struct block_device
*bdev
, sector_t block
,
1059 unsigned size
, gfp_t gfp
)
1061 /* Size must be multiple of hard sectorsize */
1062 if (unlikely(size
& (bdev_logical_block_size(bdev
)-1) ||
1063 (size
< 512 || size
> PAGE_SIZE
))) {
1064 printk(KERN_ERR
"getblk(): invalid block size %d requested\n",
1066 printk(KERN_ERR
"logical block size: %d\n",
1067 bdev_logical_block_size(bdev
));
1074 struct buffer_head
*bh
;
1077 bh
= __find_get_block(bdev
, block
, size
);
1081 ret
= grow_buffers(bdev
, block
, size
, gfp
);
1088 * The relationship between dirty buffers and dirty pages:
1090 * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1091 * the page is tagged dirty in the page cache.
1093 * At all times, the dirtiness of the buffers represents the dirtiness of
1094 * subsections of the page. If the page has buffers, the page dirty bit is
1095 * merely a hint about the true dirty state.
1097 * When a page is set dirty in its entirety, all its buffers are marked dirty
1098 * (if the page has buffers).
1100 * When a buffer is marked dirty, its page is dirtied, but the page's other
1103 * Also. When blockdev buffers are explicitly read with bread(), they
1104 * individually become uptodate. But their backing page remains not
1105 * uptodate - even if all of its buffers are uptodate. A subsequent
1106 * block_read_full_folio() against that folio will discover all the uptodate
1107 * buffers, will set the folio uptodate and will perform no I/O.
1111 * mark_buffer_dirty - mark a buffer_head as needing writeout
1112 * @bh: the buffer_head to mark dirty
1114 * mark_buffer_dirty() will set the dirty bit against the buffer, then set
1115 * its backing page dirty, then tag the page as dirty in the page cache
1116 * and then attach the address_space's inode to its superblock's dirty
1119 * mark_buffer_dirty() is atomic. It takes bh->b_folio->mapping->private_lock,
1120 * i_pages lock and mapping->host->i_lock.
1122 void mark_buffer_dirty(struct buffer_head
*bh
)
1124 WARN_ON_ONCE(!buffer_uptodate(bh
));
1126 trace_block_dirty_buffer(bh
);
1129 * Very *carefully* optimize the it-is-already-dirty case.
1131 * Don't let the final "is it dirty" escape to before we
1132 * perhaps modified the buffer.
1134 if (buffer_dirty(bh
)) {
1136 if (buffer_dirty(bh
))
1140 if (!test_set_buffer_dirty(bh
)) {
1141 struct folio
*folio
= bh
->b_folio
;
1142 struct address_space
*mapping
= NULL
;
1144 folio_memcg_lock(folio
);
1145 if (!folio_test_set_dirty(folio
)) {
1146 mapping
= folio
->mapping
;
1148 __folio_mark_dirty(folio
, mapping
, 0);
1150 folio_memcg_unlock(folio
);
1152 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
1155 EXPORT_SYMBOL(mark_buffer_dirty
);
1157 void mark_buffer_write_io_error(struct buffer_head
*bh
)
1159 struct super_block
*sb
;
1161 set_buffer_write_io_error(bh
);
1162 /* FIXME: do we need to set this in both places? */
1163 if (bh
->b_folio
&& bh
->b_folio
->mapping
)
1164 mapping_set_error(bh
->b_folio
->mapping
, -EIO
);
1165 if (bh
->b_assoc_map
)
1166 mapping_set_error(bh
->b_assoc_map
, -EIO
);
1168 sb
= READ_ONCE(bh
->b_bdev
->bd_super
);
1170 errseq_set(&sb
->s_wb_err
, -EIO
);
1173 EXPORT_SYMBOL(mark_buffer_write_io_error
);
1176 * Decrement a buffer_head's reference count. If all buffers against a page
1177 * have zero reference count, are clean and unlocked, and if the page is clean
1178 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1179 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1180 * a page but it ends up not being freed, and buffers may later be reattached).
1182 void __brelse(struct buffer_head
* buf
)
1184 if (atomic_read(&buf
->b_count
)) {
1188 WARN(1, KERN_ERR
"VFS: brelse: Trying to free free buffer\n");
1190 EXPORT_SYMBOL(__brelse
);
1193 * bforget() is like brelse(), except it discards any
1194 * potentially dirty data.
1196 void __bforget(struct buffer_head
*bh
)
1198 clear_buffer_dirty(bh
);
1199 if (bh
->b_assoc_map
) {
1200 struct address_space
*buffer_mapping
= bh
->b_folio
->mapping
;
1202 spin_lock(&buffer_mapping
->private_lock
);
1203 list_del_init(&bh
->b_assoc_buffers
);
1204 bh
->b_assoc_map
= NULL
;
1205 spin_unlock(&buffer_mapping
->private_lock
);
1209 EXPORT_SYMBOL(__bforget
);
1211 static struct buffer_head
*__bread_slow(struct buffer_head
*bh
)
1214 if (buffer_uptodate(bh
)) {
1219 bh
->b_end_io
= end_buffer_read_sync
;
1220 submit_bh(REQ_OP_READ
, bh
);
1222 if (buffer_uptodate(bh
))
1230 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1231 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1232 * refcount elevated by one when they're in an LRU. A buffer can only appear
1233 * once in a particular CPU's LRU. A single buffer can be present in multiple
1234 * CPU's LRUs at the same time.
1236 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1237 * sb_find_get_block().
1239 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1240 * a local interrupt disable for that.
1243 #define BH_LRU_SIZE 16
1246 struct buffer_head
*bhs
[BH_LRU_SIZE
];
1249 static DEFINE_PER_CPU(struct bh_lru
, bh_lrus
) = {{ NULL
}};
1252 #define bh_lru_lock() local_irq_disable()
1253 #define bh_lru_unlock() local_irq_enable()
1255 #define bh_lru_lock() preempt_disable()
1256 #define bh_lru_unlock() preempt_enable()
1259 static inline void check_irqs_on(void)
1261 #ifdef irqs_disabled
1262 BUG_ON(irqs_disabled());
1267 * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is
1268 * inserted at the front, and the buffer_head at the back if any is evicted.
1269 * Or, if already in the LRU it is moved to the front.
1271 static void bh_lru_install(struct buffer_head
*bh
)
1273 struct buffer_head
*evictee
= bh
;
1281 * the refcount of buffer_head in bh_lru prevents dropping the
1282 * attached page(i.e., try_to_free_buffers) so it could cause
1283 * failing page migration.
1284 * Skip putting upcoming bh into bh_lru until migration is done.
1286 if (lru_cache_disabled()) {
1291 b
= this_cpu_ptr(&bh_lrus
);
1292 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1293 swap(evictee
, b
->bhs
[i
]);
1294 if (evictee
== bh
) {
1306 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1308 static struct buffer_head
*
1309 lookup_bh_lru(struct block_device
*bdev
, sector_t block
, unsigned size
)
1311 struct buffer_head
*ret
= NULL
;
1316 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1317 struct buffer_head
*bh
= __this_cpu_read(bh_lrus
.bhs
[i
]);
1319 if (bh
&& bh
->b_blocknr
== block
&& bh
->b_bdev
== bdev
&&
1320 bh
->b_size
== size
) {
1323 __this_cpu_write(bh_lrus
.bhs
[i
],
1324 __this_cpu_read(bh_lrus
.bhs
[i
- 1]));
1327 __this_cpu_write(bh_lrus
.bhs
[0], bh
);
1339 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1340 * it in the LRU and mark it as accessed. If it is not present then return
1343 struct buffer_head
*
1344 __find_get_block(struct block_device
*bdev
, sector_t block
, unsigned size
)
1346 struct buffer_head
*bh
= lookup_bh_lru(bdev
, block
, size
);
1349 /* __find_get_block_slow will mark the page accessed */
1350 bh
= __find_get_block_slow(bdev
, block
);
1358 EXPORT_SYMBOL(__find_get_block
);
1361 * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
1362 * which corresponds to the passed block_device, block and size. The
1363 * returned buffer has its reference count incremented.
1365 * __getblk_gfp() will lock up the machine if grow_dev_page's
1366 * try_to_free_buffers() attempt is failing. FIXME, perhaps?
1368 struct buffer_head
*
1369 __getblk_gfp(struct block_device
*bdev
, sector_t block
,
1370 unsigned size
, gfp_t gfp
)
1372 struct buffer_head
*bh
= __find_get_block(bdev
, block
, size
);
1376 bh
= __getblk_slow(bdev
, block
, size
, gfp
);
1379 EXPORT_SYMBOL(__getblk_gfp
);
1382 * Do async read-ahead on a buffer..
1384 void __breadahead(struct block_device
*bdev
, sector_t block
, unsigned size
)
1386 struct buffer_head
*bh
= __getblk(bdev
, block
, size
);
1388 bh_readahead(bh
, REQ_RAHEAD
);
1392 EXPORT_SYMBOL(__breadahead
);
1395 * __bread_gfp() - reads a specified block and returns the bh
1396 * @bdev: the block_device to read from
1397 * @block: number of block
1398 * @size: size (in bytes) to read
1399 * @gfp: page allocation flag
1401 * Reads a specified block, and returns buffer head that contains it.
1402 * The page cache can be allocated from non-movable area
1403 * not to prevent page migration if you set gfp to zero.
1404 * It returns NULL if the block was unreadable.
1406 struct buffer_head
*
1407 __bread_gfp(struct block_device
*bdev
, sector_t block
,
1408 unsigned size
, gfp_t gfp
)
1410 struct buffer_head
*bh
= __getblk_gfp(bdev
, block
, size
, gfp
);
1412 if (likely(bh
) && !buffer_uptodate(bh
))
1413 bh
= __bread_slow(bh
);
1416 EXPORT_SYMBOL(__bread_gfp
);
1418 static void __invalidate_bh_lrus(struct bh_lru
*b
)
1422 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1428 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1429 * This doesn't race because it runs in each cpu either in irq
1430 * or with preempt disabled.
1432 static void invalidate_bh_lru(void *arg
)
1434 struct bh_lru
*b
= &get_cpu_var(bh_lrus
);
1436 __invalidate_bh_lrus(b
);
1437 put_cpu_var(bh_lrus
);
1440 bool has_bh_in_lru(int cpu
, void *dummy
)
1442 struct bh_lru
*b
= per_cpu_ptr(&bh_lrus
, cpu
);
1445 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1453 void invalidate_bh_lrus(void)
1455 on_each_cpu_cond(has_bh_in_lru
, invalidate_bh_lru
, NULL
, 1);
1457 EXPORT_SYMBOL_GPL(invalidate_bh_lrus
);
1460 * It's called from workqueue context so we need a bh_lru_lock to close
1461 * the race with preemption/irq.
1463 void invalidate_bh_lrus_cpu(void)
1468 b
= this_cpu_ptr(&bh_lrus
);
1469 __invalidate_bh_lrus(b
);
1473 void set_bh_page(struct buffer_head
*bh
,
1474 struct page
*page
, unsigned long offset
)
1477 BUG_ON(offset
>= PAGE_SIZE
);
1478 if (PageHighMem(page
))
1480 * This catches illegal uses and preserves the offset:
1482 bh
->b_data
= (char *)(0 + offset
);
1484 bh
->b_data
= page_address(page
) + offset
;
1486 EXPORT_SYMBOL(set_bh_page
);
1489 * Called when truncating a buffer on a page completely.
1492 /* Bits that are cleared during an invalidate */
1493 #define BUFFER_FLAGS_DISCARD \
1494 (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1495 1 << BH_Delay | 1 << BH_Unwritten)
1497 static void discard_buffer(struct buffer_head
* bh
)
1499 unsigned long b_state
;
1502 clear_buffer_dirty(bh
);
1504 b_state
= READ_ONCE(bh
->b_state
);
1506 } while (!try_cmpxchg(&bh
->b_state
, &b_state
,
1507 b_state
& ~BUFFER_FLAGS_DISCARD
));
1512 * block_invalidate_folio - Invalidate part or all of a buffer-backed folio.
1513 * @folio: The folio which is affected.
1514 * @offset: start of the range to invalidate
1515 * @length: length of the range to invalidate
1517 * block_invalidate_folio() is called when all or part of the folio has been
1518 * invalidated by a truncate operation.
1520 * block_invalidate_folio() does not have to release all buffers, but it must
1521 * ensure that no dirty buffer is left outside @offset and that no I/O
1522 * is underway against any of the blocks which are outside the truncation
1523 * point. Because the caller is about to free (and possibly reuse) those
1526 void block_invalidate_folio(struct folio
*folio
, size_t offset
, size_t length
)
1528 struct buffer_head
*head
, *bh
, *next
;
1529 size_t curr_off
= 0;
1530 size_t stop
= length
+ offset
;
1532 BUG_ON(!folio_test_locked(folio
));
1535 * Check for overflow
1537 BUG_ON(stop
> folio_size(folio
) || stop
< length
);
1539 head
= folio_buffers(folio
);
1545 size_t next_off
= curr_off
+ bh
->b_size
;
1546 next
= bh
->b_this_page
;
1549 * Are we still fully in range ?
1551 if (next_off
> stop
)
1555 * is this block fully invalidated?
1557 if (offset
<= curr_off
)
1559 curr_off
= next_off
;
1561 } while (bh
!= head
);
1564 * We release buffers only if the entire folio is being invalidated.
1565 * The get_block cached value has been unconditionally invalidated,
1566 * so real IO is not possible anymore.
1568 if (length
== folio_size(folio
))
1569 filemap_release_folio(folio
, 0);
1573 EXPORT_SYMBOL(block_invalidate_folio
);
1577 * We attach and possibly dirty the buffers atomically wrt
1578 * block_dirty_folio() via private_lock. try_to_free_buffers
1579 * is already excluded via the page lock.
1581 void create_empty_buffers(struct page
*page
,
1582 unsigned long blocksize
, unsigned long b_state
)
1584 struct buffer_head
*bh
, *head
, *tail
;
1586 head
= alloc_page_buffers(page
, blocksize
, true);
1589 bh
->b_state
|= b_state
;
1591 bh
= bh
->b_this_page
;
1593 tail
->b_this_page
= head
;
1595 spin_lock(&page
->mapping
->private_lock
);
1596 if (PageUptodate(page
) || PageDirty(page
)) {
1599 if (PageDirty(page
))
1600 set_buffer_dirty(bh
);
1601 if (PageUptodate(page
))
1602 set_buffer_uptodate(bh
);
1603 bh
= bh
->b_this_page
;
1604 } while (bh
!= head
);
1606 attach_page_private(page
, head
);
1607 spin_unlock(&page
->mapping
->private_lock
);
1609 EXPORT_SYMBOL(create_empty_buffers
);
1612 * clean_bdev_aliases: clean a range of buffers in block device
1613 * @bdev: Block device to clean buffers in
1614 * @block: Start of a range of blocks to clean
1615 * @len: Number of blocks to clean
1617 * We are taking a range of blocks for data and we don't want writeback of any
1618 * buffer-cache aliases starting from return from this function and until the
1619 * moment when something will explicitly mark the buffer dirty (hopefully that
1620 * will not happen until we will free that block ;-) We don't even need to mark
1621 * it not-uptodate - nobody can expect anything from a newly allocated buffer
1622 * anyway. We used to use unmap_buffer() for such invalidation, but that was
1623 * wrong. We definitely don't want to mark the alias unmapped, for example - it
1624 * would confuse anyone who might pick it with bread() afterwards...
1626 * Also.. Note that bforget() doesn't lock the buffer. So there can be
1627 * writeout I/O going on against recently-freed buffers. We don't wait on that
1628 * I/O in bforget() - it's more efficient to wait on the I/O only if we really
1629 * need to. That happens here.
1631 void clean_bdev_aliases(struct block_device
*bdev
, sector_t block
, sector_t len
)
1633 struct inode
*bd_inode
= bdev
->bd_inode
;
1634 struct address_space
*bd_mapping
= bd_inode
->i_mapping
;
1635 struct folio_batch fbatch
;
1636 pgoff_t index
= block
>> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
1639 struct buffer_head
*bh
;
1640 struct buffer_head
*head
;
1642 end
= (block
+ len
- 1) >> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
1643 folio_batch_init(&fbatch
);
1644 while (filemap_get_folios(bd_mapping
, &index
, end
, &fbatch
)) {
1645 count
= folio_batch_count(&fbatch
);
1646 for (i
= 0; i
< count
; i
++) {
1647 struct folio
*folio
= fbatch
.folios
[i
];
1649 if (!folio_buffers(folio
))
1652 * We use folio lock instead of bd_mapping->private_lock
1653 * to pin buffers here since we can afford to sleep and
1654 * it scales better than a global spinlock lock.
1657 /* Recheck when the folio is locked which pins bhs */
1658 head
= folio_buffers(folio
);
1663 if (!buffer_mapped(bh
) || (bh
->b_blocknr
< block
))
1665 if (bh
->b_blocknr
>= block
+ len
)
1667 clear_buffer_dirty(bh
);
1669 clear_buffer_req(bh
);
1671 bh
= bh
->b_this_page
;
1672 } while (bh
!= head
);
1674 folio_unlock(folio
);
1676 folio_batch_release(&fbatch
);
1678 /* End of range already reached? */
1679 if (index
> end
|| !index
)
1683 EXPORT_SYMBOL(clean_bdev_aliases
);
1686 * Size is a power-of-two in the range 512..PAGE_SIZE,
1687 * and the case we care about most is PAGE_SIZE.
1689 * So this *could* possibly be written with those
1690 * constraints in mind (relevant mostly if some
1691 * architecture has a slow bit-scan instruction)
1693 static inline int block_size_bits(unsigned int blocksize
)
1695 return ilog2(blocksize
);
1698 static struct buffer_head
*create_page_buffers(struct page
*page
, struct inode
*inode
, unsigned int b_state
)
1700 BUG_ON(!PageLocked(page
));
1702 if (!page_has_buffers(page
))
1703 create_empty_buffers(page
, 1 << READ_ONCE(inode
->i_blkbits
),
1705 return page_buffers(page
);
1709 * NOTE! All mapped/uptodate combinations are valid:
1711 * Mapped Uptodate Meaning
1713 * No No "unknown" - must do get_block()
1714 * No Yes "hole" - zero-filled
1715 * Yes No "allocated" - allocated on disk, not read in
1716 * Yes Yes "valid" - allocated and up-to-date in memory.
1718 * "Dirty" is valid only with the last case (mapped+uptodate).
1722 * While block_write_full_page is writing back the dirty buffers under
1723 * the page lock, whoever dirtied the buffers may decide to clean them
1724 * again at any time. We handle that by only looking at the buffer
1725 * state inside lock_buffer().
1727 * If block_write_full_page() is called for regular writeback
1728 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1729 * locked buffer. This only can happen if someone has written the buffer
1730 * directly, with submit_bh(). At the address_space level PageWriteback
1731 * prevents this contention from occurring.
1733 * If block_write_full_page() is called with wbc->sync_mode ==
1734 * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this
1735 * causes the writes to be flagged as synchronous writes.
1737 int __block_write_full_page(struct inode
*inode
, struct page
*page
,
1738 get_block_t
*get_block
, struct writeback_control
*wbc
,
1739 bh_end_io_t
*handler
)
1743 sector_t last_block
;
1744 struct buffer_head
*bh
, *head
;
1745 unsigned int blocksize
, bbits
;
1746 int nr_underway
= 0;
1747 blk_opf_t write_flags
= wbc_to_write_flags(wbc
);
1749 head
= create_page_buffers(page
, inode
,
1750 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
1753 * Be very careful. We have no exclusion from block_dirty_folio
1754 * here, and the (potentially unmapped) buffers may become dirty at
1755 * any time. If a buffer becomes dirty here after we've inspected it
1756 * then we just miss that fact, and the page stays dirty.
1758 * Buffers outside i_size may be dirtied by block_dirty_folio;
1759 * handle that here by just cleaning them.
1763 blocksize
= bh
->b_size
;
1764 bbits
= block_size_bits(blocksize
);
1766 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1767 last_block
= (i_size_read(inode
) - 1) >> bbits
;
1770 * Get all the dirty buffers mapped to disk addresses and
1771 * handle any aliases from the underlying blockdev's mapping.
1774 if (block
> last_block
) {
1776 * mapped buffers outside i_size will occur, because
1777 * this page can be outside i_size when there is a
1778 * truncate in progress.
1781 * The buffer was zeroed by block_write_full_page()
1783 clear_buffer_dirty(bh
);
1784 set_buffer_uptodate(bh
);
1785 } else if ((!buffer_mapped(bh
) || buffer_delay(bh
)) &&
1787 WARN_ON(bh
->b_size
!= blocksize
);
1788 err
= get_block(inode
, block
, bh
, 1);
1791 clear_buffer_delay(bh
);
1792 if (buffer_new(bh
)) {
1793 /* blockdev mappings never come here */
1794 clear_buffer_new(bh
);
1795 clean_bdev_bh_alias(bh
);
1798 bh
= bh
->b_this_page
;
1800 } while (bh
!= head
);
1803 if (!buffer_mapped(bh
))
1806 * If it's a fully non-blocking write attempt and we cannot
1807 * lock the buffer then redirty the page. Note that this can
1808 * potentially cause a busy-wait loop from writeback threads
1809 * and kswapd activity, but those code paths have their own
1810 * higher-level throttling.
1812 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
1814 } else if (!trylock_buffer(bh
)) {
1815 redirty_page_for_writepage(wbc
, page
);
1818 if (test_clear_buffer_dirty(bh
)) {
1819 mark_buffer_async_write_endio(bh
, handler
);
1823 } while ((bh
= bh
->b_this_page
) != head
);
1826 * The page and its buffers are protected by PageWriteback(), so we can
1827 * drop the bh refcounts early.
1829 BUG_ON(PageWriteback(page
));
1830 set_page_writeback(page
);
1833 struct buffer_head
*next
= bh
->b_this_page
;
1834 if (buffer_async_write(bh
)) {
1835 submit_bh_wbc(REQ_OP_WRITE
| write_flags
, bh
, wbc
);
1839 } while (bh
!= head
);
1844 if (nr_underway
== 0) {
1846 * The page was marked dirty, but the buffers were
1847 * clean. Someone wrote them back by hand with
1848 * write_dirty_buffer/submit_bh. A rare case.
1850 end_page_writeback(page
);
1853 * The page and buffer_heads can be released at any time from
1861 * ENOSPC, or some other error. We may already have added some
1862 * blocks to the file, so we need to write these out to avoid
1863 * exposing stale data.
1864 * The page is currently locked and not marked for writeback
1867 /* Recovery: lock and submit the mapped buffers */
1869 if (buffer_mapped(bh
) && buffer_dirty(bh
) &&
1870 !buffer_delay(bh
)) {
1872 mark_buffer_async_write_endio(bh
, handler
);
1875 * The buffer may have been set dirty during
1876 * attachment to a dirty page.
1878 clear_buffer_dirty(bh
);
1880 } while ((bh
= bh
->b_this_page
) != head
);
1882 BUG_ON(PageWriteback(page
));
1883 mapping_set_error(page
->mapping
, err
);
1884 set_page_writeback(page
);
1886 struct buffer_head
*next
= bh
->b_this_page
;
1887 if (buffer_async_write(bh
)) {
1888 clear_buffer_dirty(bh
);
1889 submit_bh_wbc(REQ_OP_WRITE
| write_flags
, bh
, wbc
);
1893 } while (bh
!= head
);
1897 EXPORT_SYMBOL(__block_write_full_page
);
1900 * If a page has any new buffers, zero them out here, and mark them uptodate
1901 * and dirty so they'll be written out (in order to prevent uninitialised
1902 * block data from leaking). And clear the new bit.
1904 void page_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1906 unsigned int block_start
, block_end
;
1907 struct buffer_head
*head
, *bh
;
1909 BUG_ON(!PageLocked(page
));
1910 if (!page_has_buffers(page
))
1913 bh
= head
= page_buffers(page
);
1916 block_end
= block_start
+ bh
->b_size
;
1918 if (buffer_new(bh
)) {
1919 if (block_end
> from
&& block_start
< to
) {
1920 if (!PageUptodate(page
)) {
1921 unsigned start
, size
;
1923 start
= max(from
, block_start
);
1924 size
= min(to
, block_end
) - start
;
1926 zero_user(page
, start
, size
);
1927 set_buffer_uptodate(bh
);
1930 clear_buffer_new(bh
);
1931 mark_buffer_dirty(bh
);
1935 block_start
= block_end
;
1936 bh
= bh
->b_this_page
;
1937 } while (bh
!= head
);
1939 EXPORT_SYMBOL(page_zero_new_buffers
);
1942 iomap_to_bh(struct inode
*inode
, sector_t block
, struct buffer_head
*bh
,
1943 const struct iomap
*iomap
)
1945 loff_t offset
= block
<< inode
->i_blkbits
;
1947 bh
->b_bdev
= iomap
->bdev
;
1950 * Block points to offset in file we need to map, iomap contains
1951 * the offset at which the map starts. If the map ends before the
1952 * current block, then do not map the buffer and let the caller
1955 BUG_ON(offset
>= iomap
->offset
+ iomap
->length
);
1957 switch (iomap
->type
) {
1960 * If the buffer is not up to date or beyond the current EOF,
1961 * we need to mark it as new to ensure sub-block zeroing is
1962 * executed if necessary.
1964 if (!buffer_uptodate(bh
) ||
1965 (offset
>= i_size_read(inode
)))
1968 case IOMAP_DELALLOC
:
1969 if (!buffer_uptodate(bh
) ||
1970 (offset
>= i_size_read(inode
)))
1972 set_buffer_uptodate(bh
);
1973 set_buffer_mapped(bh
);
1974 set_buffer_delay(bh
);
1976 case IOMAP_UNWRITTEN
:
1978 * For unwritten regions, we always need to ensure that regions
1979 * in the block we are not writing to are zeroed. Mark the
1980 * buffer as new to ensure this.
1983 set_buffer_unwritten(bh
);
1986 if ((iomap
->flags
& IOMAP_F_NEW
) ||
1987 offset
>= i_size_read(inode
))
1989 bh
->b_blocknr
= (iomap
->addr
+ offset
- iomap
->offset
) >>
1991 set_buffer_mapped(bh
);
1996 int __block_write_begin_int(struct folio
*folio
, loff_t pos
, unsigned len
,
1997 get_block_t
*get_block
, const struct iomap
*iomap
)
1999 unsigned from
= pos
& (PAGE_SIZE
- 1);
2000 unsigned to
= from
+ len
;
2001 struct inode
*inode
= folio
->mapping
->host
;
2002 unsigned block_start
, block_end
;
2005 unsigned blocksize
, bbits
;
2006 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
=wait
;
2008 BUG_ON(!folio_test_locked(folio
));
2009 BUG_ON(from
> PAGE_SIZE
);
2010 BUG_ON(to
> PAGE_SIZE
);
2013 head
= create_page_buffers(&folio
->page
, inode
, 0);
2014 blocksize
= head
->b_size
;
2015 bbits
= block_size_bits(blocksize
);
2017 block
= (sector_t
)folio
->index
<< (PAGE_SHIFT
- bbits
);
2019 for(bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
2020 block
++, block_start
=block_end
, bh
= bh
->b_this_page
) {
2021 block_end
= block_start
+ blocksize
;
2022 if (block_end
<= from
|| block_start
>= to
) {
2023 if (folio_test_uptodate(folio
)) {
2024 if (!buffer_uptodate(bh
))
2025 set_buffer_uptodate(bh
);
2030 clear_buffer_new(bh
);
2031 if (!buffer_mapped(bh
)) {
2032 WARN_ON(bh
->b_size
!= blocksize
);
2034 err
= get_block(inode
, block
, bh
, 1);
2038 iomap_to_bh(inode
, block
, bh
, iomap
);
2041 if (buffer_new(bh
)) {
2042 clean_bdev_bh_alias(bh
);
2043 if (folio_test_uptodate(folio
)) {
2044 clear_buffer_new(bh
);
2045 set_buffer_uptodate(bh
);
2046 mark_buffer_dirty(bh
);
2049 if (block_end
> to
|| block_start
< from
)
2050 folio_zero_segments(folio
,
2056 if (folio_test_uptodate(folio
)) {
2057 if (!buffer_uptodate(bh
))
2058 set_buffer_uptodate(bh
);
2061 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
2062 !buffer_unwritten(bh
) &&
2063 (block_start
< from
|| block_end
> to
)) {
2064 bh_read_nowait(bh
, 0);
2069 * If we issued read requests - let them complete.
2071 while(wait_bh
> wait
) {
2072 wait_on_buffer(*--wait_bh
);
2073 if (!buffer_uptodate(*wait_bh
))
2077 page_zero_new_buffers(&folio
->page
, from
, to
);
2081 int __block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
2082 get_block_t
*get_block
)
2084 return __block_write_begin_int(page_folio(page
), pos
, len
, get_block
,
2087 EXPORT_SYMBOL(__block_write_begin
);
2089 static int __block_commit_write(struct inode
*inode
, struct page
*page
,
2090 unsigned from
, unsigned to
)
2092 unsigned block_start
, block_end
;
2095 struct buffer_head
*bh
, *head
;
2097 bh
= head
= page_buffers(page
);
2098 blocksize
= bh
->b_size
;
2102 block_end
= block_start
+ blocksize
;
2103 if (block_end
<= from
|| block_start
>= to
) {
2104 if (!buffer_uptodate(bh
))
2107 set_buffer_uptodate(bh
);
2108 mark_buffer_dirty(bh
);
2111 clear_buffer_new(bh
);
2113 block_start
= block_end
;
2114 bh
= bh
->b_this_page
;
2115 } while (bh
!= head
);
2118 * If this is a partial write which happened to make all buffers
2119 * uptodate then we can optimize away a bogus read_folio() for
2120 * the next read(). Here we 'discover' whether the page went
2121 * uptodate as a result of this (potentially partial) write.
2124 SetPageUptodate(page
);
2129 * block_write_begin takes care of the basic task of block allocation and
2130 * bringing partial write blocks uptodate first.
2132 * The filesystem needs to handle block truncation upon failure.
2134 int block_write_begin(struct address_space
*mapping
, loff_t pos
, unsigned len
,
2135 struct page
**pagep
, get_block_t
*get_block
)
2137 pgoff_t index
= pos
>> PAGE_SHIFT
;
2141 page
= grab_cache_page_write_begin(mapping
, index
);
2145 status
= __block_write_begin(page
, pos
, len
, get_block
);
2146 if (unlikely(status
)) {
2155 EXPORT_SYMBOL(block_write_begin
);
2157 int block_write_end(struct file
*file
, struct address_space
*mapping
,
2158 loff_t pos
, unsigned len
, unsigned copied
,
2159 struct page
*page
, void *fsdata
)
2161 struct inode
*inode
= mapping
->host
;
2164 start
= pos
& (PAGE_SIZE
- 1);
2166 if (unlikely(copied
< len
)) {
2168 * The buffers that were written will now be uptodate, so
2169 * we don't have to worry about a read_folio reading them
2170 * and overwriting a partial write. However if we have
2171 * encountered a short write and only partially written
2172 * into a buffer, it will not be marked uptodate, so a
2173 * read_folio might come in and destroy our partial write.
2175 * Do the simplest thing, and just treat any short write to a
2176 * non uptodate page as a zero-length write, and force the
2177 * caller to redo the whole thing.
2179 if (!PageUptodate(page
))
2182 page_zero_new_buffers(page
, start
+copied
, start
+len
);
2184 flush_dcache_page(page
);
2186 /* This could be a short (even 0-length) commit */
2187 __block_commit_write(inode
, page
, start
, start
+copied
);
2191 EXPORT_SYMBOL(block_write_end
);
2193 int generic_write_end(struct file
*file
, struct address_space
*mapping
,
2194 loff_t pos
, unsigned len
, unsigned copied
,
2195 struct page
*page
, void *fsdata
)
2197 struct inode
*inode
= mapping
->host
;
2198 loff_t old_size
= inode
->i_size
;
2199 bool i_size_changed
= false;
2201 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2204 * No need to use i_size_read() here, the i_size cannot change under us
2205 * because we hold i_rwsem.
2207 * But it's important to update i_size while still holding page lock:
2208 * page writeout could otherwise come in and zero beyond i_size.
2210 if (pos
+ copied
> inode
->i_size
) {
2211 i_size_write(inode
, pos
+ copied
);
2212 i_size_changed
= true;
2219 pagecache_isize_extended(inode
, old_size
, pos
);
2221 * Don't mark the inode dirty under page lock. First, it unnecessarily
2222 * makes the holding time of page lock longer. Second, it forces lock
2223 * ordering of page lock and transaction start for journaling
2227 mark_inode_dirty(inode
);
2230 EXPORT_SYMBOL(generic_write_end
);
2233 * block_is_partially_uptodate checks whether buffers within a folio are
2236 * Returns true if all buffers which correspond to the specified part
2237 * of the folio are uptodate.
2239 bool block_is_partially_uptodate(struct folio
*folio
, size_t from
, size_t count
)
2241 unsigned block_start
, block_end
, blocksize
;
2243 struct buffer_head
*bh
, *head
;
2246 head
= folio_buffers(folio
);
2249 blocksize
= head
->b_size
;
2250 to
= min_t(unsigned, folio_size(folio
) - from
, count
);
2252 if (from
< blocksize
&& to
> folio_size(folio
) - blocksize
)
2258 block_end
= block_start
+ blocksize
;
2259 if (block_end
> from
&& block_start
< to
) {
2260 if (!buffer_uptodate(bh
)) {
2264 if (block_end
>= to
)
2267 block_start
= block_end
;
2268 bh
= bh
->b_this_page
;
2269 } while (bh
!= head
);
2273 EXPORT_SYMBOL(block_is_partially_uptodate
);
2276 * Generic "read_folio" function for block devices that have the normal
2277 * get_block functionality. This is most of the block device filesystems.
2278 * Reads the folio asynchronously --- the unlock_buffer() and
2279 * set/clear_buffer_uptodate() functions propagate buffer state into the
2280 * folio once IO has completed.
2282 int block_read_full_folio(struct folio
*folio
, get_block_t
*get_block
)
2284 struct inode
*inode
= folio
->mapping
->host
;
2285 sector_t iblock
, lblock
;
2286 struct buffer_head
*bh
, *head
, *arr
[MAX_BUF_PER_PAGE
];
2287 unsigned int blocksize
, bbits
;
2289 int fully_mapped
= 1;
2290 bool page_error
= false;
2291 loff_t limit
= i_size_read(inode
);
2293 /* This is needed for ext4. */
2294 if (IS_ENABLED(CONFIG_FS_VERITY
) && IS_VERITY(inode
))
2295 limit
= inode
->i_sb
->s_maxbytes
;
2297 VM_BUG_ON_FOLIO(folio_test_large(folio
), folio
);
2299 head
= create_page_buffers(&folio
->page
, inode
, 0);
2300 blocksize
= head
->b_size
;
2301 bbits
= block_size_bits(blocksize
);
2303 iblock
= (sector_t
)folio
->index
<< (PAGE_SHIFT
- bbits
);
2304 lblock
= (limit
+blocksize
-1) >> bbits
;
2310 if (buffer_uptodate(bh
))
2313 if (!buffer_mapped(bh
)) {
2317 if (iblock
< lblock
) {
2318 WARN_ON(bh
->b_size
!= blocksize
);
2319 err
= get_block(inode
, iblock
, bh
, 0);
2321 folio_set_error(folio
);
2325 if (!buffer_mapped(bh
)) {
2326 folio_zero_range(folio
, i
* blocksize
,
2329 set_buffer_uptodate(bh
);
2333 * get_block() might have updated the buffer
2336 if (buffer_uptodate(bh
))
2340 } while (i
++, iblock
++, (bh
= bh
->b_this_page
) != head
);
2343 folio_set_mappedtodisk(folio
);
2347 * All buffers are uptodate - we can set the folio uptodate
2348 * as well. But not if get_block() returned an error.
2351 folio_mark_uptodate(folio
);
2352 folio_unlock(folio
);
2356 /* Stage two: lock the buffers */
2357 for (i
= 0; i
< nr
; i
++) {
2360 mark_buffer_async_read(bh
);
2364 * Stage 3: start the IO. Check for uptodateness
2365 * inside the buffer lock in case another process reading
2366 * the underlying blockdev brought it uptodate (the sct fix).
2368 for (i
= 0; i
< nr
; i
++) {
2370 if (buffer_uptodate(bh
))
2371 end_buffer_async_read(bh
, 1);
2373 submit_bh(REQ_OP_READ
, bh
);
2377 EXPORT_SYMBOL(block_read_full_folio
);
2379 /* utility function for filesystems that need to do work on expanding
2380 * truncates. Uses filesystem pagecache writes to allow the filesystem to
2381 * deal with the hole.
2383 int generic_cont_expand_simple(struct inode
*inode
, loff_t size
)
2385 struct address_space
*mapping
= inode
->i_mapping
;
2386 const struct address_space_operations
*aops
= mapping
->a_ops
;
2388 void *fsdata
= NULL
;
2391 err
= inode_newsize_ok(inode
, size
);
2395 err
= aops
->write_begin(NULL
, mapping
, size
, 0, &page
, &fsdata
);
2399 err
= aops
->write_end(NULL
, mapping
, size
, 0, 0, page
, fsdata
);
2405 EXPORT_SYMBOL(generic_cont_expand_simple
);
2407 static int cont_expand_zero(struct file
*file
, struct address_space
*mapping
,
2408 loff_t pos
, loff_t
*bytes
)
2410 struct inode
*inode
= mapping
->host
;
2411 const struct address_space_operations
*aops
= mapping
->a_ops
;
2412 unsigned int blocksize
= i_blocksize(inode
);
2414 void *fsdata
= NULL
;
2415 pgoff_t index
, curidx
;
2417 unsigned zerofrom
, offset
, len
;
2420 index
= pos
>> PAGE_SHIFT
;
2421 offset
= pos
& ~PAGE_MASK
;
2423 while (index
> (curidx
= (curpos
= *bytes
)>>PAGE_SHIFT
)) {
2424 zerofrom
= curpos
& ~PAGE_MASK
;
2425 if (zerofrom
& (blocksize
-1)) {
2426 *bytes
|= (blocksize
-1);
2429 len
= PAGE_SIZE
- zerofrom
;
2431 err
= aops
->write_begin(file
, mapping
, curpos
, len
,
2435 zero_user(page
, zerofrom
, len
);
2436 err
= aops
->write_end(file
, mapping
, curpos
, len
, len
,
2443 balance_dirty_pages_ratelimited(mapping
);
2445 if (fatal_signal_pending(current
)) {
2451 /* page covers the boundary, find the boundary offset */
2452 if (index
== curidx
) {
2453 zerofrom
= curpos
& ~PAGE_MASK
;
2454 /* if we will expand the thing last block will be filled */
2455 if (offset
<= zerofrom
) {
2458 if (zerofrom
& (blocksize
-1)) {
2459 *bytes
|= (blocksize
-1);
2462 len
= offset
- zerofrom
;
2464 err
= aops
->write_begin(file
, mapping
, curpos
, len
,
2468 zero_user(page
, zerofrom
, len
);
2469 err
= aops
->write_end(file
, mapping
, curpos
, len
, len
,
2481 * For moronic filesystems that do not allow holes in file.
2482 * We may have to extend the file.
2484 int cont_write_begin(struct file
*file
, struct address_space
*mapping
,
2485 loff_t pos
, unsigned len
,
2486 struct page
**pagep
, void **fsdata
,
2487 get_block_t
*get_block
, loff_t
*bytes
)
2489 struct inode
*inode
= mapping
->host
;
2490 unsigned int blocksize
= i_blocksize(inode
);
2491 unsigned int zerofrom
;
2494 err
= cont_expand_zero(file
, mapping
, pos
, bytes
);
2498 zerofrom
= *bytes
& ~PAGE_MASK
;
2499 if (pos
+len
> *bytes
&& zerofrom
& (blocksize
-1)) {
2500 *bytes
|= (blocksize
-1);
2504 return block_write_begin(mapping
, pos
, len
, pagep
, get_block
);
2506 EXPORT_SYMBOL(cont_write_begin
);
2508 int block_commit_write(struct page
*page
, unsigned from
, unsigned to
)
2510 struct inode
*inode
= page
->mapping
->host
;
2511 __block_commit_write(inode
,page
,from
,to
);
2514 EXPORT_SYMBOL(block_commit_write
);
2517 * block_page_mkwrite() is not allowed to change the file size as it gets
2518 * called from a page fault handler when a page is first dirtied. Hence we must
2519 * be careful to check for EOF conditions here. We set the page up correctly
2520 * for a written page which means we get ENOSPC checking when writing into
2521 * holes and correct delalloc and unwritten extent mapping on filesystems that
2522 * support these features.
2524 * We are not allowed to take the i_mutex here so we have to play games to
2525 * protect against truncate races as the page could now be beyond EOF. Because
2526 * truncate writes the inode size before removing pages, once we have the
2527 * page lock we can determine safely if the page is beyond EOF. If it is not
2528 * beyond EOF, then the page is guaranteed safe against truncation until we
2531 * Direct callers of this function should protect against filesystem freezing
2532 * using sb_start_pagefault() - sb_end_pagefault() functions.
2534 int block_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
2535 get_block_t get_block
)
2537 struct page
*page
= vmf
->page
;
2538 struct inode
*inode
= file_inode(vma
->vm_file
);
2544 size
= i_size_read(inode
);
2545 if ((page
->mapping
!= inode
->i_mapping
) ||
2546 (page_offset(page
) > size
)) {
2547 /* We overload EFAULT to mean page got truncated */
2552 /* page is wholly or partially inside EOF */
2553 if (((page
->index
+ 1) << PAGE_SHIFT
) > size
)
2554 end
= size
& ~PAGE_MASK
;
2558 ret
= __block_write_begin(page
, 0, end
, get_block
);
2560 ret
= block_commit_write(page
, 0, end
);
2562 if (unlikely(ret
< 0))
2564 set_page_dirty(page
);
2565 wait_for_stable_page(page
);
2571 EXPORT_SYMBOL(block_page_mkwrite
);
2573 int block_truncate_page(struct address_space
*mapping
,
2574 loff_t from
, get_block_t
*get_block
)
2576 pgoff_t index
= from
>> PAGE_SHIFT
;
2577 unsigned offset
= from
& (PAGE_SIZE
-1);
2580 unsigned length
, pos
;
2581 struct inode
*inode
= mapping
->host
;
2583 struct buffer_head
*bh
;
2586 blocksize
= i_blocksize(inode
);
2587 length
= offset
& (blocksize
- 1);
2589 /* Block boundary? Nothing to do */
2593 length
= blocksize
- length
;
2594 iblock
= (sector_t
)index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
2596 page
= grab_cache_page(mapping
, index
);
2601 if (!page_has_buffers(page
))
2602 create_empty_buffers(page
, blocksize
, 0);
2604 /* Find the buffer that contains "offset" */
2605 bh
= page_buffers(page
);
2607 while (offset
>= pos
) {
2608 bh
= bh
->b_this_page
;
2614 if (!buffer_mapped(bh
)) {
2615 WARN_ON(bh
->b_size
!= blocksize
);
2616 err
= get_block(inode
, iblock
, bh
, 0);
2619 /* unmapped? It's a hole - nothing to do */
2620 if (!buffer_mapped(bh
))
2624 /* Ok, it's mapped. Make sure it's up-to-date */
2625 if (PageUptodate(page
))
2626 set_buffer_uptodate(bh
);
2628 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) && !buffer_unwritten(bh
)) {
2629 err
= bh_read(bh
, 0);
2630 /* Uhhuh. Read error. Complain and punt. */
2635 zero_user(page
, offset
, length
);
2636 mark_buffer_dirty(bh
);
2645 EXPORT_SYMBOL(block_truncate_page
);
2648 * The generic ->writepage function for buffer-backed address_spaces
2650 int block_write_full_page(struct page
*page
, get_block_t
*get_block
,
2651 struct writeback_control
*wbc
)
2653 struct inode
* const inode
= page
->mapping
->host
;
2654 loff_t i_size
= i_size_read(inode
);
2655 const pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2658 /* Is the page fully inside i_size? */
2659 if (page
->index
< end_index
)
2660 return __block_write_full_page(inode
, page
, get_block
, wbc
,
2661 end_buffer_async_write
);
2663 /* Is the page fully outside i_size? (truncate in progress) */
2664 offset
= i_size
& (PAGE_SIZE
-1);
2665 if (page
->index
>= end_index
+1 || !offset
) {
2667 return 0; /* don't care */
2671 * The page straddles i_size. It must be zeroed out on each and every
2672 * writepage invocation because it may be mmapped. "A file is mapped
2673 * in multiples of the page size. For a file that is not a multiple of
2674 * the page size, the remaining memory is zeroed when mapped, and
2675 * writes to that region are not written out to the file."
2677 zero_user_segment(page
, offset
, PAGE_SIZE
);
2678 return __block_write_full_page(inode
, page
, get_block
, wbc
,
2679 end_buffer_async_write
);
2681 EXPORT_SYMBOL(block_write_full_page
);
2683 sector_t
generic_block_bmap(struct address_space
*mapping
, sector_t block
,
2684 get_block_t
*get_block
)
2686 struct inode
*inode
= mapping
->host
;
2687 struct buffer_head tmp
= {
2688 .b_size
= i_blocksize(inode
),
2691 get_block(inode
, block
, &tmp
, 0);
2692 return tmp
.b_blocknr
;
2694 EXPORT_SYMBOL(generic_block_bmap
);
2696 static void end_bio_bh_io_sync(struct bio
*bio
)
2698 struct buffer_head
*bh
= bio
->bi_private
;
2700 if (unlikely(bio_flagged(bio
, BIO_QUIET
)))
2701 set_bit(BH_Quiet
, &bh
->b_state
);
2703 bh
->b_end_io(bh
, !bio
->bi_status
);
2707 static void submit_bh_wbc(blk_opf_t opf
, struct buffer_head
*bh
,
2708 struct writeback_control
*wbc
)
2710 const enum req_op op
= opf
& REQ_OP_MASK
;
2713 BUG_ON(!buffer_locked(bh
));
2714 BUG_ON(!buffer_mapped(bh
));
2715 BUG_ON(!bh
->b_end_io
);
2716 BUG_ON(buffer_delay(bh
));
2717 BUG_ON(buffer_unwritten(bh
));
2720 * Only clear out a write error when rewriting
2722 if (test_set_buffer_req(bh
) && (op
== REQ_OP_WRITE
))
2723 clear_buffer_write_io_error(bh
);
2725 if (buffer_meta(bh
))
2727 if (buffer_prio(bh
))
2730 bio
= bio_alloc(bh
->b_bdev
, 1, opf
, GFP_NOIO
);
2732 fscrypt_set_bio_crypt_ctx_bh(bio
, bh
, GFP_NOIO
);
2734 bio
->bi_iter
.bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
2736 bio_add_page(bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
2737 BUG_ON(bio
->bi_iter
.bi_size
!= bh
->b_size
);
2739 bio
->bi_end_io
= end_bio_bh_io_sync
;
2740 bio
->bi_private
= bh
;
2742 /* Take care of bh's that straddle the end of the device */
2746 wbc_init_bio(wbc
, bio
);
2747 wbc_account_cgroup_owner(wbc
, bh
->b_page
, bh
->b_size
);
2753 void submit_bh(blk_opf_t opf
, struct buffer_head
*bh
)
2755 submit_bh_wbc(opf
, bh
, NULL
);
2757 EXPORT_SYMBOL(submit_bh
);
2759 void write_dirty_buffer(struct buffer_head
*bh
, blk_opf_t op_flags
)
2762 if (!test_clear_buffer_dirty(bh
)) {
2766 bh
->b_end_io
= end_buffer_write_sync
;
2768 submit_bh(REQ_OP_WRITE
| op_flags
, bh
);
2770 EXPORT_SYMBOL(write_dirty_buffer
);
2773 * For a data-integrity writeout, we need to wait upon any in-progress I/O
2774 * and then start new I/O and then wait upon it. The caller must have a ref on
2777 int __sync_dirty_buffer(struct buffer_head
*bh
, blk_opf_t op_flags
)
2779 WARN_ON(atomic_read(&bh
->b_count
) < 1);
2781 if (test_clear_buffer_dirty(bh
)) {
2783 * The bh should be mapped, but it might not be if the
2784 * device was hot-removed. Not much we can do but fail the I/O.
2786 if (!buffer_mapped(bh
)) {
2792 bh
->b_end_io
= end_buffer_write_sync
;
2793 submit_bh(REQ_OP_WRITE
| op_flags
, bh
);
2795 if (!buffer_uptodate(bh
))
2802 EXPORT_SYMBOL(__sync_dirty_buffer
);
2804 int sync_dirty_buffer(struct buffer_head
*bh
)
2806 return __sync_dirty_buffer(bh
, REQ_SYNC
);
2808 EXPORT_SYMBOL(sync_dirty_buffer
);
2811 * try_to_free_buffers() checks if all the buffers on this particular folio
2812 * are unused, and releases them if so.
2814 * Exclusion against try_to_free_buffers may be obtained by either
2815 * locking the folio or by holding its mapping's private_lock.
2817 * If the folio is dirty but all the buffers are clean then we need to
2818 * be sure to mark the folio clean as well. This is because the folio
2819 * may be against a block device, and a later reattachment of buffers
2820 * to a dirty folio will set *all* buffers dirty. Which would corrupt
2821 * filesystem data on the same device.
2823 * The same applies to regular filesystem folios: if all the buffers are
2824 * clean then we set the folio clean and proceed. To do that, we require
2825 * total exclusion from block_dirty_folio(). That is obtained with
2828 * try_to_free_buffers() is non-blocking.
2830 static inline int buffer_busy(struct buffer_head
*bh
)
2832 return atomic_read(&bh
->b_count
) |
2833 (bh
->b_state
& ((1 << BH_Dirty
) | (1 << BH_Lock
)));
2837 drop_buffers(struct folio
*folio
, struct buffer_head
**buffers_to_free
)
2839 struct buffer_head
*head
= folio_buffers(folio
);
2840 struct buffer_head
*bh
;
2844 if (buffer_busy(bh
))
2846 bh
= bh
->b_this_page
;
2847 } while (bh
!= head
);
2850 struct buffer_head
*next
= bh
->b_this_page
;
2852 if (bh
->b_assoc_map
)
2853 __remove_assoc_queue(bh
);
2855 } while (bh
!= head
);
2856 *buffers_to_free
= head
;
2857 folio_detach_private(folio
);
2863 bool try_to_free_buffers(struct folio
*folio
)
2865 struct address_space
* const mapping
= folio
->mapping
;
2866 struct buffer_head
*buffers_to_free
= NULL
;
2869 BUG_ON(!folio_test_locked(folio
));
2870 if (folio_test_writeback(folio
))
2873 if (mapping
== NULL
) { /* can this still happen? */
2874 ret
= drop_buffers(folio
, &buffers_to_free
);
2878 spin_lock(&mapping
->private_lock
);
2879 ret
= drop_buffers(folio
, &buffers_to_free
);
2882 * If the filesystem writes its buffers by hand (eg ext3)
2883 * then we can have clean buffers against a dirty folio. We
2884 * clean the folio here; otherwise the VM will never notice
2885 * that the filesystem did any IO at all.
2887 * Also, during truncate, discard_buffer will have marked all
2888 * the folio's buffers clean. We discover that here and clean
2891 * private_lock must be held over this entire operation in order
2892 * to synchronise against block_dirty_folio and prevent the
2893 * dirty bit from being lost.
2896 folio_cancel_dirty(folio
);
2897 spin_unlock(&mapping
->private_lock
);
2899 if (buffers_to_free
) {
2900 struct buffer_head
*bh
= buffers_to_free
;
2903 struct buffer_head
*next
= bh
->b_this_page
;
2904 free_buffer_head(bh
);
2906 } while (bh
!= buffers_to_free
);
2910 EXPORT_SYMBOL(try_to_free_buffers
);
2913 * Buffer-head allocation
2915 static struct kmem_cache
*bh_cachep __read_mostly
;
2918 * Once the number of bh's in the machine exceeds this level, we start
2919 * stripping them in writeback.
2921 static unsigned long max_buffer_heads
;
2923 int buffer_heads_over_limit
;
2925 struct bh_accounting
{
2926 int nr
; /* Number of live bh's */
2927 int ratelimit
; /* Limit cacheline bouncing */
2930 static DEFINE_PER_CPU(struct bh_accounting
, bh_accounting
) = {0, 0};
2932 static void recalc_bh_state(void)
2937 if (__this_cpu_inc_return(bh_accounting
.ratelimit
) - 1 < 4096)
2939 __this_cpu_write(bh_accounting
.ratelimit
, 0);
2940 for_each_online_cpu(i
)
2941 tot
+= per_cpu(bh_accounting
, i
).nr
;
2942 buffer_heads_over_limit
= (tot
> max_buffer_heads
);
2945 struct buffer_head
*alloc_buffer_head(gfp_t gfp_flags
)
2947 struct buffer_head
*ret
= kmem_cache_zalloc(bh_cachep
, gfp_flags
);
2949 INIT_LIST_HEAD(&ret
->b_assoc_buffers
);
2950 spin_lock_init(&ret
->b_uptodate_lock
);
2952 __this_cpu_inc(bh_accounting
.nr
);
2958 EXPORT_SYMBOL(alloc_buffer_head
);
2960 void free_buffer_head(struct buffer_head
*bh
)
2962 BUG_ON(!list_empty(&bh
->b_assoc_buffers
));
2963 kmem_cache_free(bh_cachep
, bh
);
2965 __this_cpu_dec(bh_accounting
.nr
);
2969 EXPORT_SYMBOL(free_buffer_head
);
2971 static int buffer_exit_cpu_dead(unsigned int cpu
)
2974 struct bh_lru
*b
= &per_cpu(bh_lrus
, cpu
);
2976 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
2980 this_cpu_add(bh_accounting
.nr
, per_cpu(bh_accounting
, cpu
).nr
);
2981 per_cpu(bh_accounting
, cpu
).nr
= 0;
2986 * bh_uptodate_or_lock - Test whether the buffer is uptodate
2987 * @bh: struct buffer_head
2989 * Return true if the buffer is up-to-date and false,
2990 * with the buffer locked, if not.
2992 int bh_uptodate_or_lock(struct buffer_head
*bh
)
2994 if (!buffer_uptodate(bh
)) {
2996 if (!buffer_uptodate(bh
))
3002 EXPORT_SYMBOL(bh_uptodate_or_lock
);
3005 * __bh_read - Submit read for a locked buffer
3006 * @bh: struct buffer_head
3007 * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ
3008 * @wait: wait until reading finish
3010 * Returns zero on success or don't wait, and -EIO on error.
3012 int __bh_read(struct buffer_head
*bh
, blk_opf_t op_flags
, bool wait
)
3016 BUG_ON(!buffer_locked(bh
));
3019 bh
->b_end_io
= end_buffer_read_sync
;
3020 submit_bh(REQ_OP_READ
| op_flags
, bh
);
3023 if (!buffer_uptodate(bh
))
3028 EXPORT_SYMBOL(__bh_read
);
3031 * __bh_read_batch - Submit read for a batch of unlocked buffers
3032 * @nr: entry number of the buffer batch
3033 * @bhs: a batch of struct buffer_head
3034 * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ
3035 * @force_lock: force to get a lock on the buffer if set, otherwise drops any
3036 * buffer that cannot lock.
3038 * Returns zero on success or don't wait, and -EIO on error.
3040 void __bh_read_batch(int nr
, struct buffer_head
*bhs
[],
3041 blk_opf_t op_flags
, bool force_lock
)
3045 for (i
= 0; i
< nr
; i
++) {
3046 struct buffer_head
*bh
= bhs
[i
];
3048 if (buffer_uptodate(bh
))
3054 if (!trylock_buffer(bh
))
3057 if (buffer_uptodate(bh
)) {
3062 bh
->b_end_io
= end_buffer_read_sync
;
3064 submit_bh(REQ_OP_READ
| op_flags
, bh
);
3067 EXPORT_SYMBOL(__bh_read_batch
);
3069 void __init
buffer_init(void)
3071 unsigned long nrpages
;
3074 bh_cachep
= kmem_cache_create("buffer_head",
3075 sizeof(struct buffer_head
), 0,
3076 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
3081 * Limit the bh occupancy to 10% of ZONE_NORMAL
3083 nrpages
= (nr_free_buffer_pages() * 10) / 100;
3084 max_buffer_heads
= nrpages
* (PAGE_SIZE
/ sizeof(struct buffer_head
));
3085 ret
= cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD
, "fs/buffer:dead",
3086 NULL
, buffer_exit_cpu_dead
);