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>
52 #include <linux/sched/isolation.h>
56 static int fsync_buffers_list(spinlock_t *lock, struct list_head *list);
57 static void submit_bh_wbc(blk_opf_t opf, struct buffer_head *bh,
58 struct writeback_control *wbc);
60 #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
62 inline void touch_buffer(struct buffer_head *bh)
64 trace_block_touch_buffer(bh);
65 folio_mark_accessed(bh->b_folio);
67 EXPORT_SYMBOL(touch_buffer);
69 void __lock_buffer(struct buffer_head *bh)
71 wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
73 EXPORT_SYMBOL(__lock_buffer);
75 void unlock_buffer(struct buffer_head *bh)
77 clear_bit_unlock(BH_Lock, &bh->b_state);
78 smp_mb__after_atomic();
79 wake_up_bit(&bh->b_state, BH_Lock);
81 EXPORT_SYMBOL(unlock_buffer);
84 * Returns if the folio has dirty or writeback buffers. If all the buffers
85 * are unlocked and clean then the folio_test_dirty information is stale. If
86 * any of the buffers are locked, it is assumed they are locked for IO.
88 void buffer_check_dirty_writeback(struct folio *folio,
89 bool *dirty, bool *writeback)
91 struct buffer_head *head, *bh;
95 BUG_ON(!folio_test_locked(folio));
97 head = folio_buffers(folio);
101 if (folio_test_writeback(folio))
106 if (buffer_locked(bh))
109 if (buffer_dirty(bh))
112 bh = bh->b_this_page;
113 } while (bh != head);
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 folio = __filemap_get_folio(bd_mapping, index, FGP_ACCESSED, 0);
207 spin_lock(&bd_mapping->private_lock);
208 head = folio_buffers(folio);
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(bh->b_folio, bh->b_size, bh_offset(bh));
312 end_buffer_async_read(bh, valid);
316 static bool need_fsverity(struct buffer_head *bh)
318 struct folio *folio = bh->b_folio;
319 struct inode *inode = folio->mapping->host;
321 return fsverity_active(inode) &&
323 folio->index < DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
326 static void decrypt_bh(struct work_struct *work)
328 struct postprocess_bh_ctx *ctx =
329 container_of(work, struct postprocess_bh_ctx, work);
330 struct buffer_head *bh = ctx->bh;
333 err = fscrypt_decrypt_pagecache_blocks(bh->b_folio, bh->b_size,
335 if (err == 0 && need_fsverity(bh)) {
337 * We use different work queues for decryption and for verity
338 * because verity may require reading metadata pages that need
339 * decryption, and we shouldn't recurse to the same workqueue.
341 INIT_WORK(&ctx->work, verify_bh);
342 fsverity_enqueue_verify_work(&ctx->work);
345 end_buffer_async_read(bh, err == 0);
350 * I/O completion handler for block_read_full_folio() - pages
351 * which come unlocked at the end of I/O.
353 static void end_buffer_async_read_io(struct buffer_head *bh, int uptodate)
355 struct inode *inode = bh->b_folio->mapping->host;
356 bool decrypt = fscrypt_inode_uses_fs_layer_crypto(inode);
357 bool verify = need_fsverity(bh);
359 /* Decrypt (with fscrypt) and/or verify (with fsverity) if needed. */
360 if (uptodate && (decrypt || verify)) {
361 struct postprocess_bh_ctx *ctx =
362 kmalloc(sizeof(*ctx), GFP_ATOMIC);
367 INIT_WORK(&ctx->work, decrypt_bh);
368 fscrypt_enqueue_decrypt_work(&ctx->work);
370 INIT_WORK(&ctx->work, verify_bh);
371 fsverity_enqueue_verify_work(&ctx->work);
377 end_buffer_async_read(bh, uptodate);
381 * Completion handler for block_write_full_page() - pages which are unlocked
382 * during I/O, and which have PageWriteback cleared upon I/O completion.
384 void end_buffer_async_write(struct buffer_head *bh, int uptodate)
387 struct buffer_head *first;
388 struct buffer_head *tmp;
391 BUG_ON(!buffer_async_write(bh));
395 set_buffer_uptodate(bh);
397 buffer_io_error(bh, ", lost async page write");
398 mark_buffer_write_io_error(bh);
399 clear_buffer_uptodate(bh);
400 folio_set_error(folio);
403 first = folio_buffers(folio);
404 spin_lock_irqsave(&first->b_uptodate_lock, flags);
406 clear_buffer_async_write(bh);
408 tmp = bh->b_this_page;
410 if (buffer_async_write(tmp)) {
411 BUG_ON(!buffer_locked(tmp));
414 tmp = tmp->b_this_page;
416 spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
417 folio_end_writeback(folio);
421 spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
424 EXPORT_SYMBOL(end_buffer_async_write);
427 * If a page's buffers are under async readin (end_buffer_async_read
428 * completion) then there is a possibility that another thread of
429 * control could lock one of the buffers after it has completed
430 * but while some of the other buffers have not completed. This
431 * locked buffer would confuse end_buffer_async_read() into not unlocking
432 * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
433 * that this buffer is not under async I/O.
435 * The page comes unlocked when it has no locked buffer_async buffers
438 * PageLocked prevents anyone starting new async I/O reads any of
441 * PageWriteback is used to prevent simultaneous writeout of the same
444 * PageLocked prevents anyone from starting writeback of a page which is
445 * under read I/O (PageWriteback is only ever set against a locked page).
447 static void mark_buffer_async_read(struct buffer_head *bh)
449 bh->b_end_io = end_buffer_async_read_io;
450 set_buffer_async_read(bh);
453 static void mark_buffer_async_write_endio(struct buffer_head *bh,
454 bh_end_io_t *handler)
456 bh->b_end_io = handler;
457 set_buffer_async_write(bh);
460 void mark_buffer_async_write(struct buffer_head *bh)
462 mark_buffer_async_write_endio(bh, end_buffer_async_write);
464 EXPORT_SYMBOL(mark_buffer_async_write);
468 * fs/buffer.c contains helper functions for buffer-backed address space's
469 * fsync functions. A common requirement for buffer-based filesystems is
470 * that certain data from the backing blockdev needs to be written out for
471 * a successful fsync(). For example, ext2 indirect blocks need to be
472 * written back and waited upon before fsync() returns.
474 * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
475 * inode_has_buffers() and invalidate_inode_buffers() are provided for the
476 * management of a list of dependent buffers at ->i_mapping->private_list.
478 * Locking is a little subtle: try_to_free_buffers() will remove buffers
479 * from their controlling inode's queue when they are being freed. But
480 * try_to_free_buffers() will be operating against the *blockdev* mapping
481 * at the time, not against the S_ISREG file which depends on those buffers.
482 * So the locking for private_list is via the private_lock in the address_space
483 * which backs the buffers. Which is different from the address_space
484 * against which the buffers are listed. So for a particular address_space,
485 * mapping->private_lock does *not* protect mapping->private_list! In fact,
486 * mapping->private_list will always be protected by the backing blockdev's
489 * Which introduces a requirement: all buffers on an address_space's
490 * ->private_list must be from the same address_space: the blockdev's.
492 * address_spaces which do not place buffers at ->private_list via these
493 * utility functions are free to use private_lock and private_list for
494 * whatever they want. The only requirement is that list_empty(private_list)
495 * be true at clear_inode() time.
497 * FIXME: clear_inode should not call invalidate_inode_buffers(). The
498 * filesystems should do that. invalidate_inode_buffers() should just go
499 * BUG_ON(!list_empty).
501 * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
502 * take an address_space, not an inode. And it should be called
503 * mark_buffer_dirty_fsync() to clearly define why those buffers are being
506 * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
507 * list if it is already on a list. Because if the buffer is on a list,
508 * it *must* already be on the right one. If not, the filesystem is being
509 * silly. This will save a ton of locking. But first we have to ensure
510 * that buffers are taken *off* the old inode's list when they are freed
511 * (presumably in truncate). That requires careful auditing of all
512 * filesystems (do it inside bforget()). It could also be done by bringing
517 * The buffer's backing address_space's private_lock must be held
519 static void __remove_assoc_queue(struct buffer_head *bh)
521 list_del_init(&bh->b_assoc_buffers);
522 WARN_ON(!bh->b_assoc_map);
523 bh->b_assoc_map = NULL;
526 int inode_has_buffers(struct inode *inode)
528 return !list_empty(&inode->i_data.private_list);
532 * osync is designed to support O_SYNC io. It waits synchronously for
533 * all already-submitted IO to complete, but does not queue any new
534 * writes to the disk.
536 * To do O_SYNC writes, just queue the buffer writes with write_dirty_buffer
537 * as you dirty the buffers, and then use osync_inode_buffers to wait for
538 * completion. Any other dirty buffers which are not yet queued for
539 * write will not be flushed to disk by the osync.
541 static int osync_buffers_list(spinlock_t *lock, struct list_head *list)
543 struct buffer_head *bh;
549 list_for_each_prev(p, list) {
551 if (buffer_locked(bh)) {
555 if (!buffer_uptodate(bh))
566 void emergency_thaw_bdev(struct super_block *sb)
568 while (sb->s_bdev && !thaw_bdev(sb->s_bdev))
569 printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev);
573 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
574 * @mapping: the mapping which wants those buffers written
576 * Starts I/O against the buffers at mapping->private_list, and waits upon
579 * Basically, this is a convenience function for fsync().
580 * @mapping is a file or directory which needs those buffers to be written for
581 * a successful fsync().
583 int sync_mapping_buffers(struct address_space *mapping)
585 struct address_space *buffer_mapping = mapping->private_data;
587 if (buffer_mapping == NULL || list_empty(&mapping->private_list))
590 return fsync_buffers_list(&buffer_mapping->private_lock,
591 &mapping->private_list);
593 EXPORT_SYMBOL(sync_mapping_buffers);
596 * generic_buffers_fsync_noflush - generic buffer fsync implementation
597 * for simple filesystems with no inode lock
599 * @file: file to synchronize
600 * @start: start offset in bytes
601 * @end: end offset in bytes (inclusive)
602 * @datasync: only synchronize essential metadata if true
604 * This is a generic implementation of the fsync method for simple
605 * filesystems which track all non-inode metadata in the buffers list
606 * hanging off the address_space structure.
608 int generic_buffers_fsync_noflush(struct file *file, loff_t start, loff_t end,
611 struct inode *inode = file->f_mapping->host;
615 err = file_write_and_wait_range(file, start, end);
619 ret = sync_mapping_buffers(inode->i_mapping);
620 if (!(inode->i_state & I_DIRTY_ALL))
622 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
625 err = sync_inode_metadata(inode, 1);
630 /* check and advance again to catch errors after syncing out buffers */
631 err = file_check_and_advance_wb_err(file);
636 EXPORT_SYMBOL(generic_buffers_fsync_noflush);
639 * generic_buffers_fsync - generic buffer fsync implementation
640 * for simple filesystems with no inode lock
642 * @file: file to synchronize
643 * @start: start offset in bytes
644 * @end: end offset in bytes (inclusive)
645 * @datasync: only synchronize essential metadata if true
647 * This is a generic implementation of the fsync method for simple
648 * filesystems which track all non-inode metadata in the buffers list
649 * hanging off the address_space structure. This also makes sure that
650 * a device cache flush operation is called at the end.
652 int generic_buffers_fsync(struct file *file, loff_t start, loff_t end,
655 struct inode *inode = file->f_mapping->host;
658 ret = generic_buffers_fsync_noflush(file, start, end, datasync);
660 ret = blkdev_issue_flush(inode->i_sb->s_bdev);
663 EXPORT_SYMBOL(generic_buffers_fsync);
666 * Called when we've recently written block `bblock', and it is known that
667 * `bblock' was for a buffer_boundary() buffer. This means that the block at
668 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
669 * dirty, schedule it for IO. So that indirects merge nicely with their data.
671 void write_boundary_block(struct block_device *bdev,
672 sector_t bblock, unsigned blocksize)
674 struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize);
676 if (buffer_dirty(bh))
677 write_dirty_buffer(bh, 0);
682 void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode)
684 struct address_space *mapping = inode->i_mapping;
685 struct address_space *buffer_mapping = bh->b_folio->mapping;
687 mark_buffer_dirty(bh);
688 if (!mapping->private_data) {
689 mapping->private_data = buffer_mapping;
691 BUG_ON(mapping->private_data != buffer_mapping);
693 if (!bh->b_assoc_map) {
694 spin_lock(&buffer_mapping->private_lock);
695 list_move_tail(&bh->b_assoc_buffers,
696 &mapping->private_list);
697 bh->b_assoc_map = mapping;
698 spin_unlock(&buffer_mapping->private_lock);
701 EXPORT_SYMBOL(mark_buffer_dirty_inode);
704 * Add a page to the dirty page list.
706 * It is a sad fact of life that this function is called from several places
707 * deeply under spinlocking. It may not sleep.
709 * If the page has buffers, the uptodate buffers are set dirty, to preserve
710 * dirty-state coherency between the page and the buffers. It the page does
711 * not have buffers then when they are later attached they will all be set
714 * The buffers are dirtied before the page is dirtied. There's a small race
715 * window in which a writepage caller may see the page cleanness but not the
716 * buffer dirtiness. That's fine. If this code were to set the page dirty
717 * before the buffers, a concurrent writepage caller could clear the page dirty
718 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
719 * page on the dirty page list.
721 * We use private_lock to lock against try_to_free_buffers while using the
722 * page's buffer list. Also use this to protect against clean buffers being
723 * added to the page after it was set dirty.
725 * FIXME: may need to call ->reservepage here as well. That's rather up to the
726 * address_space though.
728 bool block_dirty_folio(struct address_space *mapping, struct folio *folio)
730 struct buffer_head *head;
733 spin_lock(&mapping->private_lock);
734 head = folio_buffers(folio);
736 struct buffer_head *bh = head;
739 set_buffer_dirty(bh);
740 bh = bh->b_this_page;
741 } while (bh != head);
744 * Lock out page's memcg migration to keep PageDirty
745 * synchronized with per-memcg dirty page counters.
747 folio_memcg_lock(folio);
748 newly_dirty = !folio_test_set_dirty(folio);
749 spin_unlock(&mapping->private_lock);
752 __folio_mark_dirty(folio, mapping, 1);
754 folio_memcg_unlock(folio);
757 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
761 EXPORT_SYMBOL(block_dirty_folio);
764 * Write out and wait upon a list of buffers.
766 * We have conflicting pressures: we want to make sure that all
767 * initially dirty buffers get waited on, but that any subsequently
768 * dirtied buffers don't. After all, we don't want fsync to last
769 * forever if somebody is actively writing to the file.
771 * Do this in two main stages: first we copy dirty buffers to a
772 * temporary inode list, queueing the writes as we go. Then we clean
773 * up, waiting for those writes to complete.
775 * During this second stage, any subsequent updates to the file may end
776 * up refiling the buffer on the original inode's dirty list again, so
777 * there is a chance we will end up with a buffer queued for write but
778 * not yet completed on that list. So, as a final cleanup we go through
779 * the osync code to catch these locked, dirty buffers without requeuing
780 * any newly dirty buffers for write.
782 static int fsync_buffers_list(spinlock_t *lock, struct list_head *list)
784 struct buffer_head *bh;
785 struct list_head tmp;
786 struct address_space *mapping;
788 struct blk_plug plug;
790 INIT_LIST_HEAD(&tmp);
791 blk_start_plug(&plug);
794 while (!list_empty(list)) {
795 bh = BH_ENTRY(list->next);
796 mapping = bh->b_assoc_map;
797 __remove_assoc_queue(bh);
798 /* Avoid race with mark_buffer_dirty_inode() which does
799 * a lockless check and we rely on seeing the dirty bit */
801 if (buffer_dirty(bh) || buffer_locked(bh)) {
802 list_add(&bh->b_assoc_buffers, &tmp);
803 bh->b_assoc_map = mapping;
804 if (buffer_dirty(bh)) {
808 * Ensure any pending I/O completes so that
809 * write_dirty_buffer() actually writes the
810 * current contents - it is a noop if I/O is
811 * still in flight on potentially older
814 write_dirty_buffer(bh, REQ_SYNC);
817 * Kick off IO for the previous mapping. Note
818 * that we will not run the very last mapping,
819 * wait_on_buffer() will do that for us
820 * through sync_buffer().
829 blk_finish_plug(&plug);
832 while (!list_empty(&tmp)) {
833 bh = BH_ENTRY(tmp.prev);
835 mapping = bh->b_assoc_map;
836 __remove_assoc_queue(bh);
837 /* Avoid race with mark_buffer_dirty_inode() which does
838 * a lockless check and we rely on seeing the dirty bit */
840 if (buffer_dirty(bh)) {
841 list_add(&bh->b_assoc_buffers,
842 &mapping->private_list);
843 bh->b_assoc_map = mapping;
847 if (!buffer_uptodate(bh))
854 err2 = osync_buffers_list(lock, list);
862 * Invalidate any and all dirty buffers on a given inode. We are
863 * probably unmounting the fs, but that doesn't mean we have already
864 * done a sync(). Just drop the buffers from the inode list.
866 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
867 * assumes that all the buffers are against the blockdev. Not true
870 void invalidate_inode_buffers(struct inode *inode)
872 if (inode_has_buffers(inode)) {
873 struct address_space *mapping = &inode->i_data;
874 struct list_head *list = &mapping->private_list;
875 struct address_space *buffer_mapping = mapping->private_data;
877 spin_lock(&buffer_mapping->private_lock);
878 while (!list_empty(list))
879 __remove_assoc_queue(BH_ENTRY(list->next));
880 spin_unlock(&buffer_mapping->private_lock);
883 EXPORT_SYMBOL(invalidate_inode_buffers);
886 * Remove any clean buffers from the inode's buffer list. This is called
887 * when we're trying to free the inode itself. Those buffers can pin it.
889 * Returns true if all buffers were removed.
891 int remove_inode_buffers(struct inode *inode)
895 if (inode_has_buffers(inode)) {
896 struct address_space *mapping = &inode->i_data;
897 struct list_head *list = &mapping->private_list;
898 struct address_space *buffer_mapping = mapping->private_data;
900 spin_lock(&buffer_mapping->private_lock);
901 while (!list_empty(list)) {
902 struct buffer_head *bh = BH_ENTRY(list->next);
903 if (buffer_dirty(bh)) {
907 __remove_assoc_queue(bh);
909 spin_unlock(&buffer_mapping->private_lock);
915 * Create the appropriate buffers when given a folio for data area and
916 * the size of each buffer.. Use the bh->b_this_page linked list to
917 * follow the buffers created. Return NULL if unable to create more
920 * The retry flag is used to differentiate async IO (paging, swapping)
921 * which may not fail from ordinary buffer allocations.
923 struct buffer_head *folio_alloc_buffers(struct folio *folio, unsigned long size,
926 struct buffer_head *bh, *head;
927 gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT;
929 struct mem_cgroup *memcg, *old_memcg;
934 /* The folio lock pins the memcg */
935 memcg = folio_memcg(folio);
936 old_memcg = set_active_memcg(memcg);
939 offset = folio_size(folio);
940 while ((offset -= size) >= 0) {
941 bh = alloc_buffer_head(gfp);
945 bh->b_this_page = head;
951 /* Link the buffer to its folio */
952 folio_set_bh(bh, folio, offset);
955 set_active_memcg(old_memcg);
958 * In case anything failed, we just free everything we got.
964 head = head->b_this_page;
965 free_buffer_head(bh);
971 EXPORT_SYMBOL_GPL(folio_alloc_buffers);
973 struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
976 return folio_alloc_buffers(page_folio(page), size, retry);
978 EXPORT_SYMBOL_GPL(alloc_page_buffers);
980 static inline void link_dev_buffers(struct folio *folio,
981 struct buffer_head *head)
983 struct buffer_head *bh, *tail;
988 bh = bh->b_this_page;
990 tail->b_this_page = head;
991 folio_attach_private(folio, head);
994 static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size)
996 sector_t retval = ~((sector_t)0);
997 loff_t sz = bdev_nr_bytes(bdev);
1000 unsigned int sizebits = blksize_bits(size);
1001 retval = (sz >> sizebits);
1007 * Initialise the state of a blockdev folio's buffers.
1009 static sector_t folio_init_buffers(struct folio *folio,
1010 struct block_device *bdev, sector_t block, int size)
1012 struct buffer_head *head = folio_buffers(folio);
1013 struct buffer_head *bh = head;
1014 bool uptodate = folio_test_uptodate(folio);
1015 sector_t end_block = blkdev_max_block(bdev, size);
1018 if (!buffer_mapped(bh)) {
1019 bh->b_end_io = NULL;
1020 bh->b_private = NULL;
1022 bh->b_blocknr = block;
1024 set_buffer_uptodate(bh);
1025 if (block < end_block)
1026 set_buffer_mapped(bh);
1029 bh = bh->b_this_page;
1030 } while (bh != head);
1033 * Caller needs to validate requested block against end of device.
1039 * Create the page-cache page that contains the requested block.
1041 * This is used purely for blockdev mappings.
1044 grow_dev_page(struct block_device *bdev, sector_t block,
1045 pgoff_t index, int size, int sizebits, gfp_t gfp)
1047 struct inode *inode = bdev->bd_inode;
1048 struct folio *folio;
1049 struct buffer_head *bh;
1054 gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp;
1057 * XXX: __getblk_slow() can not really deal with failure and
1058 * will endlessly loop on improvised global reclaim. Prefer
1059 * looping in the allocator rather than here, at least that
1060 * code knows what it's doing.
1062 gfp_mask |= __GFP_NOFAIL;
1064 folio = __filemap_get_folio(inode->i_mapping, index,
1065 FGP_LOCK | FGP_ACCESSED | FGP_CREAT, gfp_mask);
1067 bh = folio_buffers(folio);
1069 if (bh->b_size == size) {
1070 end_block = folio_init_buffers(folio, bdev,
1071 (sector_t)index << sizebits, size);
1074 if (!try_to_free_buffers(folio))
1078 bh = folio_alloc_buffers(folio, size, true);
1081 * Link the folio to the buffers and initialise them. Take the
1082 * lock to be atomic wrt __find_get_block(), which does not
1083 * run under the folio lock.
1085 spin_lock(&inode->i_mapping->private_lock);
1086 link_dev_buffers(folio, bh);
1087 end_block = folio_init_buffers(folio, bdev,
1088 (sector_t)index << sizebits, size);
1089 spin_unlock(&inode->i_mapping->private_lock);
1091 ret = (block < end_block) ? 1 : -ENXIO;
1093 folio_unlock(folio);
1099 * Create buffers for the specified block device block's page. If
1100 * that page was dirty, the buffers are set dirty also.
1103 grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp)
1108 sizebits = PAGE_SHIFT - __ffs(size);
1109 index = block >> sizebits;
1112 * Check for a block which wants to lie outside our maximum possible
1113 * pagecache index. (this comparison is done using sector_t types).
1115 if (unlikely(index != block >> sizebits)) {
1116 printk(KERN_ERR "%s: requested out-of-range block %llu for "
1118 __func__, (unsigned long long)block,
1123 /* Create a page with the proper size buffers.. */
1124 return grow_dev_page(bdev, block, index, size, sizebits, gfp);
1127 static struct buffer_head *
1128 __getblk_slow(struct block_device *bdev, sector_t block,
1129 unsigned size, gfp_t gfp)
1131 /* Size must be multiple of hard sectorsize */
1132 if (unlikely(size & (bdev_logical_block_size(bdev)-1) ||
1133 (size < 512 || size > PAGE_SIZE))) {
1134 printk(KERN_ERR "getblk(): invalid block size %d requested\n",
1136 printk(KERN_ERR "logical block size: %d\n",
1137 bdev_logical_block_size(bdev));
1144 struct buffer_head *bh;
1147 bh = __find_get_block(bdev, block, size);
1151 ret = grow_buffers(bdev, block, size, gfp);
1158 * The relationship between dirty buffers and dirty pages:
1160 * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1161 * the page is tagged dirty in the page cache.
1163 * At all times, the dirtiness of the buffers represents the dirtiness of
1164 * subsections of the page. If the page has buffers, the page dirty bit is
1165 * merely a hint about the true dirty state.
1167 * When a page is set dirty in its entirety, all its buffers are marked dirty
1168 * (if the page has buffers).
1170 * When a buffer is marked dirty, its page is dirtied, but the page's other
1173 * Also. When blockdev buffers are explicitly read with bread(), they
1174 * individually become uptodate. But their backing page remains not
1175 * uptodate - even if all of its buffers are uptodate. A subsequent
1176 * block_read_full_folio() against that folio will discover all the uptodate
1177 * buffers, will set the folio uptodate and will perform no I/O.
1181 * mark_buffer_dirty - mark a buffer_head as needing writeout
1182 * @bh: the buffer_head to mark dirty
1184 * mark_buffer_dirty() will set the dirty bit against the buffer, then set
1185 * its backing page dirty, then tag the page as dirty in the page cache
1186 * and then attach the address_space's inode to its superblock's dirty
1189 * mark_buffer_dirty() is atomic. It takes bh->b_folio->mapping->private_lock,
1190 * i_pages lock and mapping->host->i_lock.
1192 void mark_buffer_dirty(struct buffer_head *bh)
1194 WARN_ON_ONCE(!buffer_uptodate(bh));
1196 trace_block_dirty_buffer(bh);
1199 * Very *carefully* optimize the it-is-already-dirty case.
1201 * Don't let the final "is it dirty" escape to before we
1202 * perhaps modified the buffer.
1204 if (buffer_dirty(bh)) {
1206 if (buffer_dirty(bh))
1210 if (!test_set_buffer_dirty(bh)) {
1211 struct folio *folio = bh->b_folio;
1212 struct address_space *mapping = NULL;
1214 folio_memcg_lock(folio);
1215 if (!folio_test_set_dirty(folio)) {
1216 mapping = folio->mapping;
1218 __folio_mark_dirty(folio, mapping, 0);
1220 folio_memcg_unlock(folio);
1222 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1225 EXPORT_SYMBOL(mark_buffer_dirty);
1227 void mark_buffer_write_io_error(struct buffer_head *bh)
1229 set_buffer_write_io_error(bh);
1230 /* FIXME: do we need to set this in both places? */
1231 if (bh->b_folio && bh->b_folio->mapping)
1232 mapping_set_error(bh->b_folio->mapping, -EIO);
1233 if (bh->b_assoc_map) {
1234 mapping_set_error(bh->b_assoc_map, -EIO);
1235 errseq_set(&bh->b_assoc_map->host->i_sb->s_wb_err, -EIO);
1238 EXPORT_SYMBOL(mark_buffer_write_io_error);
1241 * Decrement a buffer_head's reference count. If all buffers against a page
1242 * have zero reference count, are clean and unlocked, and if the page is clean
1243 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1244 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1245 * a page but it ends up not being freed, and buffers may later be reattached).
1247 void __brelse(struct buffer_head * buf)
1249 if (atomic_read(&buf->b_count)) {
1253 WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n");
1255 EXPORT_SYMBOL(__brelse);
1258 * bforget() is like brelse(), except it discards any
1259 * potentially dirty data.
1261 void __bforget(struct buffer_head *bh)
1263 clear_buffer_dirty(bh);
1264 if (bh->b_assoc_map) {
1265 struct address_space *buffer_mapping = bh->b_folio->mapping;
1267 spin_lock(&buffer_mapping->private_lock);
1268 list_del_init(&bh->b_assoc_buffers);
1269 bh->b_assoc_map = NULL;
1270 spin_unlock(&buffer_mapping->private_lock);
1274 EXPORT_SYMBOL(__bforget);
1276 static struct buffer_head *__bread_slow(struct buffer_head *bh)
1279 if (buffer_uptodate(bh)) {
1284 bh->b_end_io = end_buffer_read_sync;
1285 submit_bh(REQ_OP_READ, bh);
1287 if (buffer_uptodate(bh))
1295 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1296 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1297 * refcount elevated by one when they're in an LRU. A buffer can only appear
1298 * once in a particular CPU's LRU. A single buffer can be present in multiple
1299 * CPU's LRUs at the same time.
1301 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1302 * sb_find_get_block().
1304 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1305 * a local interrupt disable for that.
1308 #define BH_LRU_SIZE 16
1311 struct buffer_head *bhs[BH_LRU_SIZE];
1314 static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }};
1317 #define bh_lru_lock() local_irq_disable()
1318 #define bh_lru_unlock() local_irq_enable()
1320 #define bh_lru_lock() preempt_disable()
1321 #define bh_lru_unlock() preempt_enable()
1324 static inline void check_irqs_on(void)
1326 #ifdef irqs_disabled
1327 BUG_ON(irqs_disabled());
1332 * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is
1333 * inserted at the front, and the buffer_head at the back if any is evicted.
1334 * Or, if already in the LRU it is moved to the front.
1336 static void bh_lru_install(struct buffer_head *bh)
1338 struct buffer_head *evictee = bh;
1346 * the refcount of buffer_head in bh_lru prevents dropping the
1347 * attached page(i.e., try_to_free_buffers) so it could cause
1348 * failing page migration.
1349 * Skip putting upcoming bh into bh_lru until migration is done.
1351 if (lru_cache_disabled() || cpu_is_isolated(smp_processor_id())) {
1356 b = this_cpu_ptr(&bh_lrus);
1357 for (i = 0; i < BH_LRU_SIZE; i++) {
1358 swap(evictee, b->bhs[i]);
1359 if (evictee == bh) {
1371 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1373 static struct buffer_head *
1374 lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
1376 struct buffer_head *ret = NULL;
1381 if (cpu_is_isolated(smp_processor_id())) {
1385 for (i = 0; i < BH_LRU_SIZE; i++) {
1386 struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]);
1388 if (bh && bh->b_blocknr == block && bh->b_bdev == bdev &&
1389 bh->b_size == size) {
1392 __this_cpu_write(bh_lrus.bhs[i],
1393 __this_cpu_read(bh_lrus.bhs[i - 1]));
1396 __this_cpu_write(bh_lrus.bhs[0], bh);
1408 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1409 * it in the LRU and mark it as accessed. If it is not present then return
1412 struct buffer_head *
1413 __find_get_block(struct block_device *bdev, sector_t block, unsigned size)
1415 struct buffer_head *bh = lookup_bh_lru(bdev, block, size);
1418 /* __find_get_block_slow will mark the page accessed */
1419 bh = __find_get_block_slow(bdev, block);
1427 EXPORT_SYMBOL(__find_get_block);
1430 * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
1431 * which corresponds to the passed block_device, block and size. The
1432 * returned buffer has its reference count incremented.
1434 * __getblk_gfp() will lock up the machine if grow_dev_page's
1435 * try_to_free_buffers() attempt is failing. FIXME, perhaps?
1437 struct buffer_head *
1438 __getblk_gfp(struct block_device *bdev, sector_t block,
1439 unsigned size, gfp_t gfp)
1441 struct buffer_head *bh = __find_get_block(bdev, block, size);
1445 bh = __getblk_slow(bdev, block, size, gfp);
1448 EXPORT_SYMBOL(__getblk_gfp);
1451 * Do async read-ahead on a buffer..
1453 void __breadahead(struct block_device *bdev, sector_t block, unsigned size)
1455 struct buffer_head *bh = __getblk(bdev, block, size);
1457 bh_readahead(bh, REQ_RAHEAD);
1461 EXPORT_SYMBOL(__breadahead);
1464 * __bread_gfp() - reads a specified block and returns the bh
1465 * @bdev: the block_device to read from
1466 * @block: number of block
1467 * @size: size (in bytes) to read
1468 * @gfp: page allocation flag
1470 * Reads a specified block, and returns buffer head that contains it.
1471 * The page cache can be allocated from non-movable area
1472 * not to prevent page migration if you set gfp to zero.
1473 * It returns NULL if the block was unreadable.
1475 struct buffer_head *
1476 __bread_gfp(struct block_device *bdev, sector_t block,
1477 unsigned size, gfp_t gfp)
1479 struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
1481 if (likely(bh) && !buffer_uptodate(bh))
1482 bh = __bread_slow(bh);
1485 EXPORT_SYMBOL(__bread_gfp);
1487 static void __invalidate_bh_lrus(struct bh_lru *b)
1491 for (i = 0; i < BH_LRU_SIZE; i++) {
1497 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1498 * This doesn't race because it runs in each cpu either in irq
1499 * or with preempt disabled.
1501 static void invalidate_bh_lru(void *arg)
1503 struct bh_lru *b = &get_cpu_var(bh_lrus);
1505 __invalidate_bh_lrus(b);
1506 put_cpu_var(bh_lrus);
1509 bool has_bh_in_lru(int cpu, void *dummy)
1511 struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu);
1514 for (i = 0; i < BH_LRU_SIZE; i++) {
1522 void invalidate_bh_lrus(void)
1524 on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1);
1526 EXPORT_SYMBOL_GPL(invalidate_bh_lrus);
1529 * It's called from workqueue context so we need a bh_lru_lock to close
1530 * the race with preemption/irq.
1532 void invalidate_bh_lrus_cpu(void)
1537 b = this_cpu_ptr(&bh_lrus);
1538 __invalidate_bh_lrus(b);
1542 void set_bh_page(struct buffer_head *bh,
1543 struct page *page, unsigned long offset)
1546 BUG_ON(offset >= PAGE_SIZE);
1547 if (PageHighMem(page))
1549 * This catches illegal uses and preserves the offset:
1551 bh->b_data = (char *)(0 + offset);
1553 bh->b_data = page_address(page) + offset;
1555 EXPORT_SYMBOL(set_bh_page);
1557 void folio_set_bh(struct buffer_head *bh, struct folio *folio,
1558 unsigned long offset)
1560 bh->b_folio = folio;
1561 BUG_ON(offset >= folio_size(folio));
1562 if (folio_test_highmem(folio))
1564 * This catches illegal uses and preserves the offset:
1566 bh->b_data = (char *)(0 + offset);
1568 bh->b_data = folio_address(folio) + offset;
1570 EXPORT_SYMBOL(folio_set_bh);
1573 * Called when truncating a buffer on a page completely.
1576 /* Bits that are cleared during an invalidate */
1577 #define BUFFER_FLAGS_DISCARD \
1578 (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1579 1 << BH_Delay | 1 << BH_Unwritten)
1581 static void discard_buffer(struct buffer_head * bh)
1583 unsigned long b_state;
1586 clear_buffer_dirty(bh);
1588 b_state = READ_ONCE(bh->b_state);
1590 } while (!try_cmpxchg(&bh->b_state, &b_state,
1591 b_state & ~BUFFER_FLAGS_DISCARD));
1596 * block_invalidate_folio - Invalidate part or all of a buffer-backed folio.
1597 * @folio: The folio which is affected.
1598 * @offset: start of the range to invalidate
1599 * @length: length of the range to invalidate
1601 * block_invalidate_folio() is called when all or part of the folio has been
1602 * invalidated by a truncate operation.
1604 * block_invalidate_folio() does not have to release all buffers, but it must
1605 * ensure that no dirty buffer is left outside @offset and that no I/O
1606 * is underway against any of the blocks which are outside the truncation
1607 * point. Because the caller is about to free (and possibly reuse) those
1610 void block_invalidate_folio(struct folio *folio, size_t offset, size_t length)
1612 struct buffer_head *head, *bh, *next;
1613 size_t curr_off = 0;
1614 size_t stop = length + offset;
1616 BUG_ON(!folio_test_locked(folio));
1619 * Check for overflow
1621 BUG_ON(stop > folio_size(folio) || stop < length);
1623 head = folio_buffers(folio);
1629 size_t next_off = curr_off + bh->b_size;
1630 next = bh->b_this_page;
1633 * Are we still fully in range ?
1635 if (next_off > stop)
1639 * is this block fully invalidated?
1641 if (offset <= curr_off)
1643 curr_off = next_off;
1645 } while (bh != head);
1648 * We release buffers only if the entire folio is being invalidated.
1649 * The get_block cached value has been unconditionally invalidated,
1650 * so real IO is not possible anymore.
1652 if (length == folio_size(folio))
1653 filemap_release_folio(folio, 0);
1657 EXPORT_SYMBOL(block_invalidate_folio);
1660 * We attach and possibly dirty the buffers atomically wrt
1661 * block_dirty_folio() via private_lock. try_to_free_buffers
1662 * is already excluded via the folio lock.
1664 void folio_create_empty_buffers(struct folio *folio, unsigned long blocksize,
1665 unsigned long b_state)
1667 struct buffer_head *bh, *head, *tail;
1669 head = folio_alloc_buffers(folio, blocksize, true);
1672 bh->b_state |= b_state;
1674 bh = bh->b_this_page;
1676 tail->b_this_page = head;
1678 spin_lock(&folio->mapping->private_lock);
1679 if (folio_test_uptodate(folio) || folio_test_dirty(folio)) {
1682 if (folio_test_dirty(folio))
1683 set_buffer_dirty(bh);
1684 if (folio_test_uptodate(folio))
1685 set_buffer_uptodate(bh);
1686 bh = bh->b_this_page;
1687 } while (bh != head);
1689 folio_attach_private(folio, head);
1690 spin_unlock(&folio->mapping->private_lock);
1692 EXPORT_SYMBOL(folio_create_empty_buffers);
1694 void create_empty_buffers(struct page *page,
1695 unsigned long blocksize, unsigned long b_state)
1697 folio_create_empty_buffers(page_folio(page), blocksize, b_state);
1699 EXPORT_SYMBOL(create_empty_buffers);
1702 * clean_bdev_aliases: clean a range of buffers in block device
1703 * @bdev: Block device to clean buffers in
1704 * @block: Start of a range of blocks to clean
1705 * @len: Number of blocks to clean
1707 * We are taking a range of blocks for data and we don't want writeback of any
1708 * buffer-cache aliases starting from return from this function and until the
1709 * moment when something will explicitly mark the buffer dirty (hopefully that
1710 * will not happen until we will free that block ;-) We don't even need to mark
1711 * it not-uptodate - nobody can expect anything from a newly allocated buffer
1712 * anyway. We used to use unmap_buffer() for such invalidation, but that was
1713 * wrong. We definitely don't want to mark the alias unmapped, for example - it
1714 * would confuse anyone who might pick it with bread() afterwards...
1716 * Also.. Note that bforget() doesn't lock the buffer. So there can be
1717 * writeout I/O going on against recently-freed buffers. We don't wait on that
1718 * I/O in bforget() - it's more efficient to wait on the I/O only if we really
1719 * need to. That happens here.
1721 void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len)
1723 struct inode *bd_inode = bdev->bd_inode;
1724 struct address_space *bd_mapping = bd_inode->i_mapping;
1725 struct folio_batch fbatch;
1726 pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
1729 struct buffer_head *bh;
1730 struct buffer_head *head;
1732 end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits);
1733 folio_batch_init(&fbatch);
1734 while (filemap_get_folios(bd_mapping, &index, end, &fbatch)) {
1735 count = folio_batch_count(&fbatch);
1736 for (i = 0; i < count; i++) {
1737 struct folio *folio = fbatch.folios[i];
1739 if (!folio_buffers(folio))
1742 * We use folio lock instead of bd_mapping->private_lock
1743 * to pin buffers here since we can afford to sleep and
1744 * it scales better than a global spinlock lock.
1747 /* Recheck when the folio is locked which pins bhs */
1748 head = folio_buffers(folio);
1753 if (!buffer_mapped(bh) || (bh->b_blocknr < block))
1755 if (bh->b_blocknr >= block + len)
1757 clear_buffer_dirty(bh);
1759 clear_buffer_req(bh);
1761 bh = bh->b_this_page;
1762 } while (bh != head);
1764 folio_unlock(folio);
1766 folio_batch_release(&fbatch);
1768 /* End of range already reached? */
1769 if (index > end || !index)
1773 EXPORT_SYMBOL(clean_bdev_aliases);
1776 * Size is a power-of-two in the range 512..PAGE_SIZE,
1777 * and the case we care about most is PAGE_SIZE.
1779 * So this *could* possibly be written with those
1780 * constraints in mind (relevant mostly if some
1781 * architecture has a slow bit-scan instruction)
1783 static inline int block_size_bits(unsigned int blocksize)
1785 return ilog2(blocksize);
1788 static struct buffer_head *folio_create_buffers(struct folio *folio,
1789 struct inode *inode,
1790 unsigned int b_state)
1792 BUG_ON(!folio_test_locked(folio));
1794 if (!folio_buffers(folio))
1795 folio_create_empty_buffers(folio,
1796 1 << READ_ONCE(inode->i_blkbits),
1798 return folio_buffers(folio);
1802 * NOTE! All mapped/uptodate combinations are valid:
1804 * Mapped Uptodate Meaning
1806 * No No "unknown" - must do get_block()
1807 * No Yes "hole" - zero-filled
1808 * Yes No "allocated" - allocated on disk, not read in
1809 * Yes Yes "valid" - allocated and up-to-date in memory.
1811 * "Dirty" is valid only with the last case (mapped+uptodate).
1815 * While block_write_full_page is writing back the dirty buffers under
1816 * the page lock, whoever dirtied the buffers may decide to clean them
1817 * again at any time. We handle that by only looking at the buffer
1818 * state inside lock_buffer().
1820 * If block_write_full_page() is called for regular writeback
1821 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1822 * locked buffer. This only can happen if someone has written the buffer
1823 * directly, with submit_bh(). At the address_space level PageWriteback
1824 * prevents this contention from occurring.
1826 * If block_write_full_page() is called with wbc->sync_mode ==
1827 * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this
1828 * causes the writes to be flagged as synchronous writes.
1830 int __block_write_full_folio(struct inode *inode, struct folio *folio,
1831 get_block_t *get_block, struct writeback_control *wbc,
1832 bh_end_io_t *handler)
1836 sector_t last_block;
1837 struct buffer_head *bh, *head;
1838 unsigned int blocksize, bbits;
1839 int nr_underway = 0;
1840 blk_opf_t write_flags = wbc_to_write_flags(wbc);
1842 head = folio_create_buffers(folio, inode,
1843 (1 << BH_Dirty) | (1 << BH_Uptodate));
1846 * Be very careful. We have no exclusion from block_dirty_folio
1847 * here, and the (potentially unmapped) buffers may become dirty at
1848 * any time. If a buffer becomes dirty here after we've inspected it
1849 * then we just miss that fact, and the folio stays dirty.
1851 * Buffers outside i_size may be dirtied by block_dirty_folio;
1852 * handle that here by just cleaning them.
1856 blocksize = bh->b_size;
1857 bbits = block_size_bits(blocksize);
1859 block = (sector_t)folio->index << (PAGE_SHIFT - bbits);
1860 last_block = (i_size_read(inode) - 1) >> bbits;
1863 * Get all the dirty buffers mapped to disk addresses and
1864 * handle any aliases from the underlying blockdev's mapping.
1867 if (block > last_block) {
1869 * mapped buffers outside i_size will occur, because
1870 * this folio can be outside i_size when there is a
1871 * truncate in progress.
1874 * The buffer was zeroed by block_write_full_page()
1876 clear_buffer_dirty(bh);
1877 set_buffer_uptodate(bh);
1878 } else if ((!buffer_mapped(bh) || buffer_delay(bh)) &&
1880 WARN_ON(bh->b_size != blocksize);
1881 err = get_block(inode, block, bh, 1);
1884 clear_buffer_delay(bh);
1885 if (buffer_new(bh)) {
1886 /* blockdev mappings never come here */
1887 clear_buffer_new(bh);
1888 clean_bdev_bh_alias(bh);
1891 bh = bh->b_this_page;
1893 } while (bh != head);
1896 if (!buffer_mapped(bh))
1899 * If it's a fully non-blocking write attempt and we cannot
1900 * lock the buffer then redirty the folio. Note that this can
1901 * potentially cause a busy-wait loop from writeback threads
1902 * and kswapd activity, but those code paths have their own
1903 * higher-level throttling.
1905 if (wbc->sync_mode != WB_SYNC_NONE) {
1907 } else if (!trylock_buffer(bh)) {
1908 folio_redirty_for_writepage(wbc, folio);
1911 if (test_clear_buffer_dirty(bh)) {
1912 mark_buffer_async_write_endio(bh, handler);
1916 } while ((bh = bh->b_this_page) != head);
1919 * The folio and its buffers are protected by the writeback flag,
1920 * so we can drop the bh refcounts early.
1922 BUG_ON(folio_test_writeback(folio));
1923 folio_start_writeback(folio);
1926 struct buffer_head *next = bh->b_this_page;
1927 if (buffer_async_write(bh)) {
1928 submit_bh_wbc(REQ_OP_WRITE | write_flags, bh, wbc);
1932 } while (bh != head);
1933 folio_unlock(folio);
1937 if (nr_underway == 0) {
1939 * The folio was marked dirty, but the buffers were
1940 * clean. Someone wrote them back by hand with
1941 * write_dirty_buffer/submit_bh. A rare case.
1943 folio_end_writeback(folio);
1946 * The folio and buffer_heads can be released at any time from
1954 * ENOSPC, or some other error. We may already have added some
1955 * blocks to the file, so we need to write these out to avoid
1956 * exposing stale data.
1957 * The folio is currently locked and not marked for writeback
1960 /* Recovery: lock and submit the mapped buffers */
1962 if (buffer_mapped(bh) && buffer_dirty(bh) &&
1963 !buffer_delay(bh)) {
1965 mark_buffer_async_write_endio(bh, handler);
1968 * The buffer may have been set dirty during
1969 * attachment to a dirty folio.
1971 clear_buffer_dirty(bh);
1973 } while ((bh = bh->b_this_page) != head);
1974 folio_set_error(folio);
1975 BUG_ON(folio_test_writeback(folio));
1976 mapping_set_error(folio->mapping, err);
1977 folio_start_writeback(folio);
1979 struct buffer_head *next = bh->b_this_page;
1980 if (buffer_async_write(bh)) {
1981 clear_buffer_dirty(bh);
1982 submit_bh_wbc(REQ_OP_WRITE | write_flags, bh, wbc);
1986 } while (bh != head);
1987 folio_unlock(folio);
1990 EXPORT_SYMBOL(__block_write_full_folio);
1993 * If a folio has any new buffers, zero them out here, and mark them uptodate
1994 * and dirty so they'll be written out (in order to prevent uninitialised
1995 * block data from leaking). And clear the new bit.
1997 void folio_zero_new_buffers(struct folio *folio, size_t from, size_t to)
1999 size_t block_start, block_end;
2000 struct buffer_head *head, *bh;
2002 BUG_ON(!folio_test_locked(folio));
2003 head = folio_buffers(folio);
2010 block_end = block_start + bh->b_size;
2012 if (buffer_new(bh)) {
2013 if (block_end > from && block_start < to) {
2014 if (!folio_test_uptodate(folio)) {
2017 start = max(from, block_start);
2018 xend = min(to, block_end);
2020 folio_zero_segment(folio, start, xend);
2021 set_buffer_uptodate(bh);
2024 clear_buffer_new(bh);
2025 mark_buffer_dirty(bh);
2029 block_start = block_end;
2030 bh = bh->b_this_page;
2031 } while (bh != head);
2033 EXPORT_SYMBOL(folio_zero_new_buffers);
2036 iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh,
2037 const struct iomap *iomap)
2039 loff_t offset = block << inode->i_blkbits;
2041 bh->b_bdev = iomap->bdev;
2044 * Block points to offset in file we need to map, iomap contains
2045 * the offset at which the map starts. If the map ends before the
2046 * current block, then do not map the buffer and let the caller
2049 BUG_ON(offset >= iomap->offset + iomap->length);
2051 switch (iomap->type) {
2054 * If the buffer is not up to date or beyond the current EOF,
2055 * we need to mark it as new to ensure sub-block zeroing is
2056 * executed if necessary.
2058 if (!buffer_uptodate(bh) ||
2059 (offset >= i_size_read(inode)))
2062 case IOMAP_DELALLOC:
2063 if (!buffer_uptodate(bh) ||
2064 (offset >= i_size_read(inode)))
2066 set_buffer_uptodate(bh);
2067 set_buffer_mapped(bh);
2068 set_buffer_delay(bh);
2070 case IOMAP_UNWRITTEN:
2072 * For unwritten regions, we always need to ensure that regions
2073 * in the block we are not writing to are zeroed. Mark the
2074 * buffer as new to ensure this.
2077 set_buffer_unwritten(bh);
2080 if ((iomap->flags & IOMAP_F_NEW) ||
2081 offset >= i_size_read(inode))
2083 bh->b_blocknr = (iomap->addr + offset - iomap->offset) >>
2085 set_buffer_mapped(bh);
2090 int __block_write_begin_int(struct folio *folio, loff_t pos, unsigned len,
2091 get_block_t *get_block, const struct iomap *iomap)
2093 unsigned from = pos & (PAGE_SIZE - 1);
2094 unsigned to = from + len;
2095 struct inode *inode = folio->mapping->host;
2096 unsigned block_start, block_end;
2099 unsigned blocksize, bbits;
2100 struct buffer_head *bh, *head, *wait[2], **wait_bh=wait;
2102 BUG_ON(!folio_test_locked(folio));
2103 BUG_ON(from > PAGE_SIZE);
2104 BUG_ON(to > PAGE_SIZE);
2107 head = folio_create_buffers(folio, inode, 0);
2108 blocksize = head->b_size;
2109 bbits = block_size_bits(blocksize);
2111 block = (sector_t)folio->index << (PAGE_SHIFT - bbits);
2113 for(bh = head, block_start = 0; bh != head || !block_start;
2114 block++, block_start=block_end, bh = bh->b_this_page) {
2115 block_end = block_start + blocksize;
2116 if (block_end <= from || block_start >= to) {
2117 if (folio_test_uptodate(folio)) {
2118 if (!buffer_uptodate(bh))
2119 set_buffer_uptodate(bh);
2124 clear_buffer_new(bh);
2125 if (!buffer_mapped(bh)) {
2126 WARN_ON(bh->b_size != blocksize);
2128 err = get_block(inode, block, bh, 1);
2132 iomap_to_bh(inode, block, bh, iomap);
2135 if (buffer_new(bh)) {
2136 clean_bdev_bh_alias(bh);
2137 if (folio_test_uptodate(folio)) {
2138 clear_buffer_new(bh);
2139 set_buffer_uptodate(bh);
2140 mark_buffer_dirty(bh);
2143 if (block_end > to || block_start < from)
2144 folio_zero_segments(folio,
2150 if (folio_test_uptodate(folio)) {
2151 if (!buffer_uptodate(bh))
2152 set_buffer_uptodate(bh);
2155 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
2156 !buffer_unwritten(bh) &&
2157 (block_start < from || block_end > to)) {
2158 bh_read_nowait(bh, 0);
2163 * If we issued read requests - let them complete.
2165 while(wait_bh > wait) {
2166 wait_on_buffer(*--wait_bh);
2167 if (!buffer_uptodate(*wait_bh))
2171 folio_zero_new_buffers(folio, from, to);
2175 int __block_write_begin(struct page *page, loff_t pos, unsigned len,
2176 get_block_t *get_block)
2178 return __block_write_begin_int(page_folio(page), pos, len, get_block,
2181 EXPORT_SYMBOL(__block_write_begin);
2183 static int __block_commit_write(struct inode *inode, struct folio *folio,
2184 size_t from, size_t to)
2186 size_t block_start, block_end;
2187 bool partial = false;
2189 struct buffer_head *bh, *head;
2191 bh = head = folio_buffers(folio);
2192 blocksize = bh->b_size;
2196 block_end = block_start + blocksize;
2197 if (block_end <= from || block_start >= to) {
2198 if (!buffer_uptodate(bh))
2201 set_buffer_uptodate(bh);
2202 mark_buffer_dirty(bh);
2205 clear_buffer_new(bh);
2207 block_start = block_end;
2208 bh = bh->b_this_page;
2209 } while (bh != head);
2212 * If this is a partial write which happened to make all buffers
2213 * uptodate then we can optimize away a bogus read_folio() for
2214 * the next read(). Here we 'discover' whether the folio went
2215 * uptodate as a result of this (potentially partial) write.
2218 folio_mark_uptodate(folio);
2223 * block_write_begin takes care of the basic task of block allocation and
2224 * bringing partial write blocks uptodate first.
2226 * The filesystem needs to handle block truncation upon failure.
2228 int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
2229 struct page **pagep, get_block_t *get_block)
2231 pgoff_t index = pos >> PAGE_SHIFT;
2235 page = grab_cache_page_write_begin(mapping, index);
2239 status = __block_write_begin(page, pos, len, get_block);
2240 if (unlikely(status)) {
2249 EXPORT_SYMBOL(block_write_begin);
2251 int block_write_end(struct file *file, struct address_space *mapping,
2252 loff_t pos, unsigned len, unsigned copied,
2253 struct page *page, void *fsdata)
2255 struct folio *folio = page_folio(page);
2256 struct inode *inode = mapping->host;
2257 size_t start = pos - folio_pos(folio);
2259 if (unlikely(copied < len)) {
2261 * The buffers that were written will now be uptodate, so
2262 * we don't have to worry about a read_folio reading them
2263 * and overwriting a partial write. However if we have
2264 * encountered a short write and only partially written
2265 * into a buffer, it will not be marked uptodate, so a
2266 * read_folio might come in and destroy our partial write.
2268 * Do the simplest thing, and just treat any short write to a
2269 * non uptodate folio as a zero-length write, and force the
2270 * caller to redo the whole thing.
2272 if (!folio_test_uptodate(folio))
2275 folio_zero_new_buffers(folio, start+copied, start+len);
2277 flush_dcache_folio(folio);
2279 /* This could be a short (even 0-length) commit */
2280 __block_commit_write(inode, folio, start, start + copied);
2284 EXPORT_SYMBOL(block_write_end);
2286 int generic_write_end(struct file *file, struct address_space *mapping,
2287 loff_t pos, unsigned len, unsigned copied,
2288 struct page *page, void *fsdata)
2290 struct inode *inode = mapping->host;
2291 loff_t old_size = inode->i_size;
2292 bool i_size_changed = false;
2294 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
2297 * No need to use i_size_read() here, the i_size cannot change under us
2298 * because we hold i_rwsem.
2300 * But it's important to update i_size while still holding page lock:
2301 * page writeout could otherwise come in and zero beyond i_size.
2303 if (pos + copied > inode->i_size) {
2304 i_size_write(inode, pos + copied);
2305 i_size_changed = true;
2312 pagecache_isize_extended(inode, old_size, pos);
2314 * Don't mark the inode dirty under page lock. First, it unnecessarily
2315 * makes the holding time of page lock longer. Second, it forces lock
2316 * ordering of page lock and transaction start for journaling
2320 mark_inode_dirty(inode);
2323 EXPORT_SYMBOL(generic_write_end);
2326 * block_is_partially_uptodate checks whether buffers within a folio are
2329 * Returns true if all buffers which correspond to the specified part
2330 * of the folio are uptodate.
2332 bool block_is_partially_uptodate(struct folio *folio, size_t from, size_t count)
2334 unsigned block_start, block_end, blocksize;
2336 struct buffer_head *bh, *head;
2339 head = folio_buffers(folio);
2342 blocksize = head->b_size;
2343 to = min_t(unsigned, folio_size(folio) - from, count);
2345 if (from < blocksize && to > folio_size(folio) - blocksize)
2351 block_end = block_start + blocksize;
2352 if (block_end > from && block_start < to) {
2353 if (!buffer_uptodate(bh)) {
2357 if (block_end >= to)
2360 block_start = block_end;
2361 bh = bh->b_this_page;
2362 } while (bh != head);
2366 EXPORT_SYMBOL(block_is_partially_uptodate);
2369 * Generic "read_folio" function for block devices that have the normal
2370 * get_block functionality. This is most of the block device filesystems.
2371 * Reads the folio asynchronously --- the unlock_buffer() and
2372 * set/clear_buffer_uptodate() functions propagate buffer state into the
2373 * folio once IO has completed.
2375 int block_read_full_folio(struct folio *folio, get_block_t *get_block)
2377 struct inode *inode = folio->mapping->host;
2378 sector_t iblock, lblock;
2379 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
2380 unsigned int blocksize, bbits;
2382 int fully_mapped = 1;
2383 bool page_error = false;
2384 loff_t limit = i_size_read(inode);
2386 /* This is needed for ext4. */
2387 if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode))
2388 limit = inode->i_sb->s_maxbytes;
2390 VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
2392 head = folio_create_buffers(folio, inode, 0);
2393 blocksize = head->b_size;
2394 bbits = block_size_bits(blocksize);
2396 iblock = (sector_t)folio->index << (PAGE_SHIFT - bbits);
2397 lblock = (limit+blocksize-1) >> bbits;
2403 if (buffer_uptodate(bh))
2406 if (!buffer_mapped(bh)) {
2410 if (iblock < lblock) {
2411 WARN_ON(bh->b_size != blocksize);
2412 err = get_block(inode, iblock, bh, 0);
2414 folio_set_error(folio);
2418 if (!buffer_mapped(bh)) {
2419 folio_zero_range(folio, i * blocksize,
2422 set_buffer_uptodate(bh);
2426 * get_block() might have updated the buffer
2429 if (buffer_uptodate(bh))
2433 } while (i++, iblock++, (bh = bh->b_this_page) != head);
2436 folio_set_mappedtodisk(folio);
2440 * All buffers are uptodate - we can set the folio uptodate
2441 * as well. But not if get_block() returned an error.
2444 folio_mark_uptodate(folio);
2445 folio_unlock(folio);
2449 /* Stage two: lock the buffers */
2450 for (i = 0; i < nr; i++) {
2453 mark_buffer_async_read(bh);
2457 * Stage 3: start the IO. Check for uptodateness
2458 * inside the buffer lock in case another process reading
2459 * the underlying blockdev brought it uptodate (the sct fix).
2461 for (i = 0; i < nr; i++) {
2463 if (buffer_uptodate(bh))
2464 end_buffer_async_read(bh, 1);
2466 submit_bh(REQ_OP_READ, bh);
2470 EXPORT_SYMBOL(block_read_full_folio);
2472 /* utility function for filesystems that need to do work on expanding
2473 * truncates. Uses filesystem pagecache writes to allow the filesystem to
2474 * deal with the hole.
2476 int generic_cont_expand_simple(struct inode *inode, loff_t size)
2478 struct address_space *mapping = inode->i_mapping;
2479 const struct address_space_operations *aops = mapping->a_ops;
2481 void *fsdata = NULL;
2484 err = inode_newsize_ok(inode, size);
2488 err = aops->write_begin(NULL, mapping, size, 0, &page, &fsdata);
2492 err = aops->write_end(NULL, mapping, size, 0, 0, page, fsdata);
2498 EXPORT_SYMBOL(generic_cont_expand_simple);
2500 static int cont_expand_zero(struct file *file, struct address_space *mapping,
2501 loff_t pos, loff_t *bytes)
2503 struct inode *inode = mapping->host;
2504 const struct address_space_operations *aops = mapping->a_ops;
2505 unsigned int blocksize = i_blocksize(inode);
2507 void *fsdata = NULL;
2508 pgoff_t index, curidx;
2510 unsigned zerofrom, offset, len;
2513 index = pos >> PAGE_SHIFT;
2514 offset = pos & ~PAGE_MASK;
2516 while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) {
2517 zerofrom = curpos & ~PAGE_MASK;
2518 if (zerofrom & (blocksize-1)) {
2519 *bytes |= (blocksize-1);
2522 len = PAGE_SIZE - zerofrom;
2524 err = aops->write_begin(file, mapping, curpos, len,
2528 zero_user(page, zerofrom, len);
2529 err = aops->write_end(file, mapping, curpos, len, len,
2536 balance_dirty_pages_ratelimited(mapping);
2538 if (fatal_signal_pending(current)) {
2544 /* page covers the boundary, find the boundary offset */
2545 if (index == curidx) {
2546 zerofrom = curpos & ~PAGE_MASK;
2547 /* if we will expand the thing last block will be filled */
2548 if (offset <= zerofrom) {
2551 if (zerofrom & (blocksize-1)) {
2552 *bytes |= (blocksize-1);
2555 len = offset - zerofrom;
2557 err = aops->write_begin(file, mapping, curpos, len,
2561 zero_user(page, zerofrom, len);
2562 err = aops->write_end(file, mapping, curpos, len, len,
2574 * For moronic filesystems that do not allow holes in file.
2575 * We may have to extend the file.
2577 int cont_write_begin(struct file *file, struct address_space *mapping,
2578 loff_t pos, unsigned len,
2579 struct page **pagep, void **fsdata,
2580 get_block_t *get_block, loff_t *bytes)
2582 struct inode *inode = mapping->host;
2583 unsigned int blocksize = i_blocksize(inode);
2584 unsigned int zerofrom;
2587 err = cont_expand_zero(file, mapping, pos, bytes);
2591 zerofrom = *bytes & ~PAGE_MASK;
2592 if (pos+len > *bytes && zerofrom & (blocksize-1)) {
2593 *bytes |= (blocksize-1);
2597 return block_write_begin(mapping, pos, len, pagep, get_block);
2599 EXPORT_SYMBOL(cont_write_begin);
2601 int block_commit_write(struct page *page, unsigned from, unsigned to)
2603 struct folio *folio = page_folio(page);
2604 struct inode *inode = folio->mapping->host;
2605 __block_commit_write(inode, folio, from, to);
2608 EXPORT_SYMBOL(block_commit_write);
2611 * block_page_mkwrite() is not allowed to change the file size as it gets
2612 * called from a page fault handler when a page is first dirtied. Hence we must
2613 * be careful to check for EOF conditions here. We set the page up correctly
2614 * for a written page which means we get ENOSPC checking when writing into
2615 * holes and correct delalloc and unwritten extent mapping on filesystems that
2616 * support these features.
2618 * We are not allowed to take the i_mutex here so we have to play games to
2619 * protect against truncate races as the page could now be beyond EOF. Because
2620 * truncate writes the inode size before removing pages, once we have the
2621 * page lock we can determine safely if the page is beyond EOF. If it is not
2622 * beyond EOF, then the page is guaranteed safe against truncation until we
2625 * Direct callers of this function should protect against filesystem freezing
2626 * using sb_start_pagefault() - sb_end_pagefault() functions.
2628 int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
2629 get_block_t get_block)
2631 struct folio *folio = page_folio(vmf->page);
2632 struct inode *inode = file_inode(vma->vm_file);
2638 size = i_size_read(inode);
2639 if ((folio->mapping != inode->i_mapping) ||
2640 (folio_pos(folio) >= size)) {
2641 /* We overload EFAULT to mean page got truncated */
2646 end = folio_size(folio);
2647 /* folio is wholly or partially inside EOF */
2648 if (folio_pos(folio) + end > size)
2649 end = size - folio_pos(folio);
2651 ret = __block_write_begin_int(folio, 0, end, get_block, NULL);
2653 ret = __block_commit_write(inode, folio, 0, end);
2655 if (unlikely(ret < 0))
2657 folio_mark_dirty(folio);
2658 folio_wait_stable(folio);
2661 folio_unlock(folio);
2664 EXPORT_SYMBOL(block_page_mkwrite);
2666 int block_truncate_page(struct address_space *mapping,
2667 loff_t from, get_block_t *get_block)
2669 pgoff_t index = from >> PAGE_SHIFT;
2672 size_t offset, length, pos;
2673 struct inode *inode = mapping->host;
2674 struct folio *folio;
2675 struct buffer_head *bh;
2678 blocksize = i_blocksize(inode);
2679 length = from & (blocksize - 1);
2681 /* Block boundary? Nothing to do */
2685 length = blocksize - length;
2686 iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
2688 folio = filemap_grab_folio(mapping, index);
2690 return PTR_ERR(folio);
2692 bh = folio_buffers(folio);
2694 folio_create_empty_buffers(folio, blocksize, 0);
2695 bh = folio_buffers(folio);
2698 /* Find the buffer that contains "offset" */
2699 offset = offset_in_folio(folio, from);
2701 while (offset >= pos) {
2702 bh = bh->b_this_page;
2707 if (!buffer_mapped(bh)) {
2708 WARN_ON(bh->b_size != blocksize);
2709 err = get_block(inode, iblock, bh, 0);
2712 /* unmapped? It's a hole - nothing to do */
2713 if (!buffer_mapped(bh))
2717 /* Ok, it's mapped. Make sure it's up-to-date */
2718 if (folio_test_uptodate(folio))
2719 set_buffer_uptodate(bh);
2721 if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
2722 err = bh_read(bh, 0);
2723 /* Uhhuh. Read error. Complain and punt. */
2728 folio_zero_range(folio, offset, length);
2729 mark_buffer_dirty(bh);
2732 folio_unlock(folio);
2737 EXPORT_SYMBOL(block_truncate_page);
2740 * The generic ->writepage function for buffer-backed address_spaces
2742 int block_write_full_page(struct page *page, get_block_t *get_block,
2743 struct writeback_control *wbc)
2745 struct folio *folio = page_folio(page);
2746 struct inode * const inode = folio->mapping->host;
2747 loff_t i_size = i_size_read(inode);
2749 /* Is the folio fully inside i_size? */
2750 if (folio_pos(folio) + folio_size(folio) <= i_size)
2751 return __block_write_full_folio(inode, folio, get_block, wbc,
2752 end_buffer_async_write);
2754 /* Is the folio fully outside i_size? (truncate in progress) */
2755 if (folio_pos(folio) >= i_size) {
2756 folio_unlock(folio);
2757 return 0; /* don't care */
2761 * The folio straddles i_size. It must be zeroed out on each and every
2762 * writepage invocation because it may be mmapped. "A file is mapped
2763 * in multiples of the page size. For a file that is not a multiple of
2764 * the page size, the remaining memory is zeroed when mapped, and
2765 * writes to that region are not written out to the file."
2767 folio_zero_segment(folio, offset_in_folio(folio, i_size),
2769 return __block_write_full_folio(inode, folio, get_block, wbc,
2770 end_buffer_async_write);
2772 EXPORT_SYMBOL(block_write_full_page);
2774 sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
2775 get_block_t *get_block)
2777 struct inode *inode = mapping->host;
2778 struct buffer_head tmp = {
2779 .b_size = i_blocksize(inode),
2782 get_block(inode, block, &tmp, 0);
2783 return tmp.b_blocknr;
2785 EXPORT_SYMBOL(generic_block_bmap);
2787 static void end_bio_bh_io_sync(struct bio *bio)
2789 struct buffer_head *bh = bio->bi_private;
2791 if (unlikely(bio_flagged(bio, BIO_QUIET)))
2792 set_bit(BH_Quiet, &bh->b_state);
2794 bh->b_end_io(bh, !bio->bi_status);
2798 static void submit_bh_wbc(blk_opf_t opf, struct buffer_head *bh,
2799 struct writeback_control *wbc)
2801 const enum req_op op = opf & REQ_OP_MASK;
2804 BUG_ON(!buffer_locked(bh));
2805 BUG_ON(!buffer_mapped(bh));
2806 BUG_ON(!bh->b_end_io);
2807 BUG_ON(buffer_delay(bh));
2808 BUG_ON(buffer_unwritten(bh));
2811 * Only clear out a write error when rewriting
2813 if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE))
2814 clear_buffer_write_io_error(bh);
2816 if (buffer_meta(bh))
2818 if (buffer_prio(bh))
2821 bio = bio_alloc(bh->b_bdev, 1, opf, GFP_NOIO);
2823 fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO);
2825 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
2827 __bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
2829 bio->bi_end_io = end_bio_bh_io_sync;
2830 bio->bi_private = bh;
2832 /* Take care of bh's that straddle the end of the device */
2836 wbc_init_bio(wbc, bio);
2837 wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size);
2843 void submit_bh(blk_opf_t opf, struct buffer_head *bh)
2845 submit_bh_wbc(opf, bh, NULL);
2847 EXPORT_SYMBOL(submit_bh);
2849 void write_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags)
2852 if (!test_clear_buffer_dirty(bh)) {
2856 bh->b_end_io = end_buffer_write_sync;
2858 submit_bh(REQ_OP_WRITE | op_flags, bh);
2860 EXPORT_SYMBOL(write_dirty_buffer);
2863 * For a data-integrity writeout, we need to wait upon any in-progress I/O
2864 * and then start new I/O and then wait upon it. The caller must have a ref on
2867 int __sync_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags)
2869 WARN_ON(atomic_read(&bh->b_count) < 1);
2871 if (test_clear_buffer_dirty(bh)) {
2873 * The bh should be mapped, but it might not be if the
2874 * device was hot-removed. Not much we can do but fail the I/O.
2876 if (!buffer_mapped(bh)) {
2882 bh->b_end_io = end_buffer_write_sync;
2883 submit_bh(REQ_OP_WRITE | op_flags, bh);
2885 if (!buffer_uptodate(bh))
2892 EXPORT_SYMBOL(__sync_dirty_buffer);
2894 int sync_dirty_buffer(struct buffer_head *bh)
2896 return __sync_dirty_buffer(bh, REQ_SYNC);
2898 EXPORT_SYMBOL(sync_dirty_buffer);
2901 * try_to_free_buffers() checks if all the buffers on this particular folio
2902 * are unused, and releases them if so.
2904 * Exclusion against try_to_free_buffers may be obtained by either
2905 * locking the folio or by holding its mapping's private_lock.
2907 * If the folio is dirty but all the buffers are clean then we need to
2908 * be sure to mark the folio clean as well. This is because the folio
2909 * may be against a block device, and a later reattachment of buffers
2910 * to a dirty folio will set *all* buffers dirty. Which would corrupt
2911 * filesystem data on the same device.
2913 * The same applies to regular filesystem folios: if all the buffers are
2914 * clean then we set the folio clean and proceed. To do that, we require
2915 * total exclusion from block_dirty_folio(). That is obtained with
2918 * try_to_free_buffers() is non-blocking.
2920 static inline int buffer_busy(struct buffer_head *bh)
2922 return atomic_read(&bh->b_count) |
2923 (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock)));
2927 drop_buffers(struct folio *folio, struct buffer_head **buffers_to_free)
2929 struct buffer_head *head = folio_buffers(folio);
2930 struct buffer_head *bh;
2934 if (buffer_busy(bh))
2936 bh = bh->b_this_page;
2937 } while (bh != head);
2940 struct buffer_head *next = bh->b_this_page;
2942 if (bh->b_assoc_map)
2943 __remove_assoc_queue(bh);
2945 } while (bh != head);
2946 *buffers_to_free = head;
2947 folio_detach_private(folio);
2953 bool try_to_free_buffers(struct folio *folio)
2955 struct address_space * const mapping = folio->mapping;
2956 struct buffer_head *buffers_to_free = NULL;
2959 BUG_ON(!folio_test_locked(folio));
2960 if (folio_test_writeback(folio))
2963 if (mapping == NULL) { /* can this still happen? */
2964 ret = drop_buffers(folio, &buffers_to_free);
2968 spin_lock(&mapping->private_lock);
2969 ret = drop_buffers(folio, &buffers_to_free);
2972 * If the filesystem writes its buffers by hand (eg ext3)
2973 * then we can have clean buffers against a dirty folio. We
2974 * clean the folio here; otherwise the VM will never notice
2975 * that the filesystem did any IO at all.
2977 * Also, during truncate, discard_buffer will have marked all
2978 * the folio's buffers clean. We discover that here and clean
2981 * private_lock must be held over this entire operation in order
2982 * to synchronise against block_dirty_folio and prevent the
2983 * dirty bit from being lost.
2986 folio_cancel_dirty(folio);
2987 spin_unlock(&mapping->private_lock);
2989 if (buffers_to_free) {
2990 struct buffer_head *bh = buffers_to_free;
2993 struct buffer_head *next = bh->b_this_page;
2994 free_buffer_head(bh);
2996 } while (bh != buffers_to_free);
3000 EXPORT_SYMBOL(try_to_free_buffers);
3003 * Buffer-head allocation
3005 static struct kmem_cache *bh_cachep __read_mostly;
3008 * Once the number of bh's in the machine exceeds this level, we start
3009 * stripping them in writeback.
3011 static unsigned long max_buffer_heads;
3013 int buffer_heads_over_limit;
3015 struct bh_accounting {
3016 int nr; /* Number of live bh's */
3017 int ratelimit; /* Limit cacheline bouncing */
3020 static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0};
3022 static void recalc_bh_state(void)
3027 if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096)
3029 __this_cpu_write(bh_accounting.ratelimit, 0);
3030 for_each_online_cpu(i)
3031 tot += per_cpu(bh_accounting, i).nr;
3032 buffer_heads_over_limit = (tot > max_buffer_heads);
3035 struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
3037 struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags);
3039 INIT_LIST_HEAD(&ret->b_assoc_buffers);
3040 spin_lock_init(&ret->b_uptodate_lock);
3042 __this_cpu_inc(bh_accounting.nr);
3048 EXPORT_SYMBOL(alloc_buffer_head);
3050 void free_buffer_head(struct buffer_head *bh)
3052 BUG_ON(!list_empty(&bh->b_assoc_buffers));
3053 kmem_cache_free(bh_cachep, bh);
3055 __this_cpu_dec(bh_accounting.nr);
3059 EXPORT_SYMBOL(free_buffer_head);
3061 static int buffer_exit_cpu_dead(unsigned int cpu)
3064 struct bh_lru *b = &per_cpu(bh_lrus, cpu);
3066 for (i = 0; i < BH_LRU_SIZE; i++) {
3070 this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr);
3071 per_cpu(bh_accounting, cpu).nr = 0;
3076 * bh_uptodate_or_lock - Test whether the buffer is uptodate
3077 * @bh: struct buffer_head
3079 * Return true if the buffer is up-to-date and false,
3080 * with the buffer locked, if not.
3082 int bh_uptodate_or_lock(struct buffer_head *bh)
3084 if (!buffer_uptodate(bh)) {
3086 if (!buffer_uptodate(bh))
3092 EXPORT_SYMBOL(bh_uptodate_or_lock);
3095 * __bh_read - Submit read for a locked buffer
3096 * @bh: struct buffer_head
3097 * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ
3098 * @wait: wait until reading finish
3100 * Returns zero on success or don't wait, and -EIO on error.
3102 int __bh_read(struct buffer_head *bh, blk_opf_t op_flags, bool wait)
3106 BUG_ON(!buffer_locked(bh));
3109 bh->b_end_io = end_buffer_read_sync;
3110 submit_bh(REQ_OP_READ | op_flags, bh);
3113 if (!buffer_uptodate(bh))
3118 EXPORT_SYMBOL(__bh_read);
3121 * __bh_read_batch - Submit read for a batch of unlocked buffers
3122 * @nr: entry number of the buffer batch
3123 * @bhs: a batch of struct buffer_head
3124 * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ
3125 * @force_lock: force to get a lock on the buffer if set, otherwise drops any
3126 * buffer that cannot lock.
3128 * Returns zero on success or don't wait, and -EIO on error.
3130 void __bh_read_batch(int nr, struct buffer_head *bhs[],
3131 blk_opf_t op_flags, bool force_lock)
3135 for (i = 0; i < nr; i++) {
3136 struct buffer_head *bh = bhs[i];
3138 if (buffer_uptodate(bh))
3144 if (!trylock_buffer(bh))
3147 if (buffer_uptodate(bh)) {
3152 bh->b_end_io = end_buffer_read_sync;
3154 submit_bh(REQ_OP_READ | op_flags, bh);
3157 EXPORT_SYMBOL(__bh_read_batch);
3159 void __init buffer_init(void)
3161 unsigned long nrpages;
3164 bh_cachep = kmem_cache_create("buffer_head",
3165 sizeof(struct buffer_head), 0,
3166 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
3171 * Limit the bh occupancy to 10% of ZONE_NORMAL
3173 nrpages = (nr_free_buffer_pages() * 10) / 100;
3174 max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head));
3175 ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead",
3176 NULL, buffer_exit_cpu_dead);