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>
52 static int fsync_buffers_list(spinlock_t *lock, struct list_head *list);
53 static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
54 enum rw_hint hint, struct writeback_control *wbc);
56 #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
58 inline void touch_buffer(struct buffer_head *bh)
60 trace_block_touch_buffer(bh);
61 mark_page_accessed(bh->b_page);
63 EXPORT_SYMBOL(touch_buffer);
65 void __lock_buffer(struct buffer_head *bh)
67 wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
69 EXPORT_SYMBOL(__lock_buffer);
71 void unlock_buffer(struct buffer_head *bh)
73 clear_bit_unlock(BH_Lock, &bh->b_state);
74 smp_mb__after_atomic();
75 wake_up_bit(&bh->b_state, BH_Lock);
77 EXPORT_SYMBOL(unlock_buffer);
80 * Returns if the page has dirty or writeback buffers. If all the buffers
81 * are unlocked and clean then the PageDirty information is stale. If
82 * any of the pages are locked, it is assumed they are locked for IO.
84 void buffer_check_dirty_writeback(struct page *page,
85 bool *dirty, bool *writeback)
87 struct buffer_head *head, *bh;
91 BUG_ON(!PageLocked(page));
93 if (!page_has_buffers(page))
96 if (PageWriteback(page))
99 head = page_buffers(page);
102 if (buffer_locked(bh))
105 if (buffer_dirty(bh))
108 bh = bh->b_this_page;
109 } while (bh != head);
111 EXPORT_SYMBOL(buffer_check_dirty_writeback);
114 * Block until a buffer comes unlocked. This doesn't stop it
115 * from becoming locked again - you have to lock it yourself
116 * if you want to preserve its state.
118 void __wait_on_buffer(struct buffer_head * bh)
120 wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
122 EXPORT_SYMBOL(__wait_on_buffer);
125 __clear_page_buffers(struct page *page)
127 ClearPagePrivate(page);
128 set_page_private(page, 0);
132 static void buffer_io_error(struct buffer_head *bh, char *msg)
134 if (!test_bit(BH_Quiet, &bh->b_state))
135 printk_ratelimited(KERN_ERR
136 "Buffer I/O error on dev %pg, logical block %llu%s\n",
137 bh->b_bdev, (unsigned long long)bh->b_blocknr, msg);
141 * End-of-IO handler helper function which does not touch the bh after
143 * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
144 * a race there is benign: unlock_buffer() only use the bh's address for
145 * hashing after unlocking the buffer, so it doesn't actually touch the bh
148 static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate)
151 set_buffer_uptodate(bh);
153 /* This happens, due to failed read-ahead attempts. */
154 clear_buffer_uptodate(bh);
160 * Default synchronous end-of-IO handler.. Just mark it up-to-date and
161 * unlock the buffer. This is what ll_rw_block uses too.
163 void end_buffer_read_sync(struct buffer_head *bh, int uptodate)
165 __end_buffer_read_notouch(bh, uptodate);
168 EXPORT_SYMBOL(end_buffer_read_sync);
170 void end_buffer_write_sync(struct buffer_head *bh, int uptodate)
173 set_buffer_uptodate(bh);
175 buffer_io_error(bh, ", lost sync page write");
176 mark_buffer_write_io_error(bh);
177 clear_buffer_uptodate(bh);
182 EXPORT_SYMBOL(end_buffer_write_sync);
185 * Various filesystems appear to want __find_get_block to be non-blocking.
186 * But it's the page lock which protects the buffers. To get around this,
187 * we get exclusion from try_to_free_buffers with the blockdev mapping's
190 * Hack idea: for the blockdev mapping, private_lock contention
191 * may be quite high. This code could TryLock the page, and if that
192 * succeeds, there is no need to take private_lock.
194 static struct buffer_head *
195 __find_get_block_slow(struct block_device *bdev, sector_t block)
197 struct inode *bd_inode = bdev->bd_inode;
198 struct address_space *bd_mapping = bd_inode->i_mapping;
199 struct buffer_head *ret = NULL;
201 struct buffer_head *bh;
202 struct buffer_head *head;
205 static DEFINE_RATELIMIT_STATE(last_warned, HZ, 1);
207 index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
208 page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED);
212 spin_lock(&bd_mapping->private_lock);
213 if (!page_has_buffers(page))
215 head = page_buffers(page);
218 if (!buffer_mapped(bh))
220 else if (bh->b_blocknr == block) {
225 bh = bh->b_this_page;
226 } while (bh != head);
228 /* we might be here because some of the buffers on this page are
229 * not mapped. This is due to various races between
230 * file io on the block device and getblk. It gets dealt with
231 * elsewhere, don't buffer_error if we had some unmapped buffers
233 ratelimit_set_flags(&last_warned, RATELIMIT_MSG_ON_RELEASE);
234 if (all_mapped && __ratelimit(&last_warned)) {
235 printk("__find_get_block_slow() failed. block=%llu, "
236 "b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, "
237 "device %pg blocksize: %d\n",
238 (unsigned long long)block,
239 (unsigned long long)bh->b_blocknr,
240 bh->b_state, bh->b_size, bdev,
241 1 << bd_inode->i_blkbits);
244 spin_unlock(&bd_mapping->private_lock);
250 static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
253 struct buffer_head *first;
254 struct buffer_head *tmp;
256 int page_uptodate = 1;
258 BUG_ON(!buffer_async_read(bh));
262 set_buffer_uptodate(bh);
264 clear_buffer_uptodate(bh);
265 buffer_io_error(bh, ", async page read");
270 * Be _very_ careful from here on. Bad things can happen if
271 * two buffer heads end IO at almost the same time and both
272 * decide that the page is now completely done.
274 first = page_buffers(page);
275 local_irq_save(flags);
276 bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
277 clear_buffer_async_read(bh);
281 if (!buffer_uptodate(tmp))
283 if (buffer_async_read(tmp)) {
284 BUG_ON(!buffer_locked(tmp));
287 tmp = tmp->b_this_page;
289 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
290 local_irq_restore(flags);
293 * If none of the buffers had errors and they are all
294 * uptodate then we can set the page uptodate.
296 if (page_uptodate && !PageError(page))
297 SetPageUptodate(page);
302 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
303 local_irq_restore(flags);
307 struct decrypt_bh_ctx {
308 struct work_struct work;
309 struct buffer_head *bh;
312 static void decrypt_bh(struct work_struct *work)
314 struct decrypt_bh_ctx *ctx =
315 container_of(work, struct decrypt_bh_ctx, work);
316 struct buffer_head *bh = ctx->bh;
319 err = fscrypt_decrypt_pagecache_blocks(bh->b_page, bh->b_size,
321 end_buffer_async_read(bh, err == 0);
326 * I/O completion handler for block_read_full_page() - pages
327 * which come unlocked at the end of I/O.
329 static void end_buffer_async_read_io(struct buffer_head *bh, int uptodate)
331 /* Decrypt if needed */
332 if (uptodate && IS_ENABLED(CONFIG_FS_ENCRYPTION) &&
333 IS_ENCRYPTED(bh->b_page->mapping->host) &&
334 S_ISREG(bh->b_page->mapping->host->i_mode)) {
335 struct decrypt_bh_ctx *ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
338 INIT_WORK(&ctx->work, decrypt_bh);
340 fscrypt_enqueue_decrypt_work(&ctx->work);
345 end_buffer_async_read(bh, uptodate);
349 * Completion handler for block_write_full_page() - pages which are unlocked
350 * during I/O, and which have PageWriteback cleared upon I/O completion.
352 void end_buffer_async_write(struct buffer_head *bh, int uptodate)
355 struct buffer_head *first;
356 struct buffer_head *tmp;
359 BUG_ON(!buffer_async_write(bh));
363 set_buffer_uptodate(bh);
365 buffer_io_error(bh, ", lost async page write");
366 mark_buffer_write_io_error(bh);
367 clear_buffer_uptodate(bh);
371 first = page_buffers(page);
372 local_irq_save(flags);
373 bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
375 clear_buffer_async_write(bh);
377 tmp = bh->b_this_page;
379 if (buffer_async_write(tmp)) {
380 BUG_ON(!buffer_locked(tmp));
383 tmp = tmp->b_this_page;
385 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
386 local_irq_restore(flags);
387 end_page_writeback(page);
391 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
392 local_irq_restore(flags);
395 EXPORT_SYMBOL(end_buffer_async_write);
398 * If a page's buffers are under async readin (end_buffer_async_read
399 * completion) then there is a possibility that another thread of
400 * control could lock one of the buffers after it has completed
401 * but while some of the other buffers have not completed. This
402 * locked buffer would confuse end_buffer_async_read() into not unlocking
403 * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
404 * that this buffer is not under async I/O.
406 * The page comes unlocked when it has no locked buffer_async buffers
409 * PageLocked prevents anyone starting new async I/O reads any of
412 * PageWriteback is used to prevent simultaneous writeout of the same
415 * PageLocked prevents anyone from starting writeback of a page which is
416 * under read I/O (PageWriteback is only ever set against a locked page).
418 static void mark_buffer_async_read(struct buffer_head *bh)
420 bh->b_end_io = end_buffer_async_read_io;
421 set_buffer_async_read(bh);
424 static void mark_buffer_async_write_endio(struct buffer_head *bh,
425 bh_end_io_t *handler)
427 bh->b_end_io = handler;
428 set_buffer_async_write(bh);
431 void mark_buffer_async_write(struct buffer_head *bh)
433 mark_buffer_async_write_endio(bh, end_buffer_async_write);
435 EXPORT_SYMBOL(mark_buffer_async_write);
439 * fs/buffer.c contains helper functions for buffer-backed address space's
440 * fsync functions. A common requirement for buffer-based filesystems is
441 * that certain data from the backing blockdev needs to be written out for
442 * a successful fsync(). For example, ext2 indirect blocks need to be
443 * written back and waited upon before fsync() returns.
445 * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
446 * inode_has_buffers() and invalidate_inode_buffers() are provided for the
447 * management of a list of dependent buffers at ->i_mapping->private_list.
449 * Locking is a little subtle: try_to_free_buffers() will remove buffers
450 * from their controlling inode's queue when they are being freed. But
451 * try_to_free_buffers() will be operating against the *blockdev* mapping
452 * at the time, not against the S_ISREG file which depends on those buffers.
453 * So the locking for private_list is via the private_lock in the address_space
454 * which backs the buffers. Which is different from the address_space
455 * against which the buffers are listed. So for a particular address_space,
456 * mapping->private_lock does *not* protect mapping->private_list! In fact,
457 * mapping->private_list will always be protected by the backing blockdev's
460 * Which introduces a requirement: all buffers on an address_space's
461 * ->private_list must be from the same address_space: the blockdev's.
463 * address_spaces which do not place buffers at ->private_list via these
464 * utility functions are free to use private_lock and private_list for
465 * whatever they want. The only requirement is that list_empty(private_list)
466 * be true at clear_inode() time.
468 * FIXME: clear_inode should not call invalidate_inode_buffers(). The
469 * filesystems should do that. invalidate_inode_buffers() should just go
470 * BUG_ON(!list_empty).
472 * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
473 * take an address_space, not an inode. And it should be called
474 * mark_buffer_dirty_fsync() to clearly define why those buffers are being
477 * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
478 * list if it is already on a list. Because if the buffer is on a list,
479 * it *must* already be on the right one. If not, the filesystem is being
480 * silly. This will save a ton of locking. But first we have to ensure
481 * that buffers are taken *off* the old inode's list when they are freed
482 * (presumably in truncate). That requires careful auditing of all
483 * filesystems (do it inside bforget()). It could also be done by bringing
488 * The buffer's backing address_space's private_lock must be held
490 static void __remove_assoc_queue(struct buffer_head *bh)
492 list_del_init(&bh->b_assoc_buffers);
493 WARN_ON(!bh->b_assoc_map);
494 bh->b_assoc_map = NULL;
497 int inode_has_buffers(struct inode *inode)
499 return !list_empty(&inode->i_data.private_list);
503 * osync is designed to support O_SYNC io. It waits synchronously for
504 * all already-submitted IO to complete, but does not queue any new
505 * writes to the disk.
507 * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
508 * you dirty the buffers, and then use osync_inode_buffers to wait for
509 * completion. Any other dirty buffers which are not yet queued for
510 * write will not be flushed to disk by the osync.
512 static int osync_buffers_list(spinlock_t *lock, struct list_head *list)
514 struct buffer_head *bh;
520 list_for_each_prev(p, list) {
522 if (buffer_locked(bh)) {
526 if (!buffer_uptodate(bh))
537 void emergency_thaw_bdev(struct super_block *sb)
539 while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb))
540 printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev);
544 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
545 * @mapping: the mapping which wants those buffers written
547 * Starts I/O against the buffers at mapping->private_list, and waits upon
550 * Basically, this is a convenience function for fsync().
551 * @mapping is a file or directory which needs those buffers to be written for
552 * a successful fsync().
554 int sync_mapping_buffers(struct address_space *mapping)
556 struct address_space *buffer_mapping = mapping->private_data;
558 if (buffer_mapping == NULL || list_empty(&mapping->private_list))
561 return fsync_buffers_list(&buffer_mapping->private_lock,
562 &mapping->private_list);
564 EXPORT_SYMBOL(sync_mapping_buffers);
567 * Called when we've recently written block `bblock', and it is known that
568 * `bblock' was for a buffer_boundary() buffer. This means that the block at
569 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
570 * dirty, schedule it for IO. So that indirects merge nicely with their data.
572 void write_boundary_block(struct block_device *bdev,
573 sector_t bblock, unsigned blocksize)
575 struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize);
577 if (buffer_dirty(bh))
578 ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
583 void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode)
585 struct address_space *mapping = inode->i_mapping;
586 struct address_space *buffer_mapping = bh->b_page->mapping;
588 mark_buffer_dirty(bh);
589 if (!mapping->private_data) {
590 mapping->private_data = buffer_mapping;
592 BUG_ON(mapping->private_data != buffer_mapping);
594 if (!bh->b_assoc_map) {
595 spin_lock(&buffer_mapping->private_lock);
596 list_move_tail(&bh->b_assoc_buffers,
597 &mapping->private_list);
598 bh->b_assoc_map = mapping;
599 spin_unlock(&buffer_mapping->private_lock);
602 EXPORT_SYMBOL(mark_buffer_dirty_inode);
605 * Mark the page dirty, and set it dirty in the page cache, and mark the inode
608 * If warn is true, then emit a warning if the page is not uptodate and has
609 * not been truncated.
611 * The caller must hold lock_page_memcg().
613 void __set_page_dirty(struct page *page, struct address_space *mapping,
618 xa_lock_irqsave(&mapping->i_pages, flags);
619 if (page->mapping) { /* Race with truncate? */
620 WARN_ON_ONCE(warn && !PageUptodate(page));
621 account_page_dirtied(page, mapping);
622 __xa_set_mark(&mapping->i_pages, page_index(page),
623 PAGECACHE_TAG_DIRTY);
625 xa_unlock_irqrestore(&mapping->i_pages, flags);
627 EXPORT_SYMBOL_GPL(__set_page_dirty);
630 * Add a page to the dirty page list.
632 * It is a sad fact of life that this function is called from several places
633 * deeply under spinlocking. It may not sleep.
635 * If the page has buffers, the uptodate buffers are set dirty, to preserve
636 * dirty-state coherency between the page and the buffers. It the page does
637 * not have buffers then when they are later attached they will all be set
640 * The buffers are dirtied before the page is dirtied. There's a small race
641 * window in which a writepage caller may see the page cleanness but not the
642 * buffer dirtiness. That's fine. If this code were to set the page dirty
643 * before the buffers, a concurrent writepage caller could clear the page dirty
644 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
645 * page on the dirty page list.
647 * We use private_lock to lock against try_to_free_buffers while using the
648 * page's buffer list. Also use this to protect against clean buffers being
649 * added to the page after it was set dirty.
651 * FIXME: may need to call ->reservepage here as well. That's rather up to the
652 * address_space though.
654 int __set_page_dirty_buffers(struct page *page)
657 struct address_space *mapping = page_mapping(page);
659 if (unlikely(!mapping))
660 return !TestSetPageDirty(page);
662 spin_lock(&mapping->private_lock);
663 if (page_has_buffers(page)) {
664 struct buffer_head *head = page_buffers(page);
665 struct buffer_head *bh = head;
668 set_buffer_dirty(bh);
669 bh = bh->b_this_page;
670 } while (bh != head);
673 * Lock out page->mem_cgroup migration to keep PageDirty
674 * synchronized with per-memcg dirty page counters.
676 lock_page_memcg(page);
677 newly_dirty = !TestSetPageDirty(page);
678 spin_unlock(&mapping->private_lock);
681 __set_page_dirty(page, mapping, 1);
683 unlock_page_memcg(page);
686 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
690 EXPORT_SYMBOL(__set_page_dirty_buffers);
693 * Write out and wait upon a list of buffers.
695 * We have conflicting pressures: we want to make sure that all
696 * initially dirty buffers get waited on, but that any subsequently
697 * dirtied buffers don't. After all, we don't want fsync to last
698 * forever if somebody is actively writing to the file.
700 * Do this in two main stages: first we copy dirty buffers to a
701 * temporary inode list, queueing the writes as we go. Then we clean
702 * up, waiting for those writes to complete.
704 * During this second stage, any subsequent updates to the file may end
705 * up refiling the buffer on the original inode's dirty list again, so
706 * there is a chance we will end up with a buffer queued for write but
707 * not yet completed on that list. So, as a final cleanup we go through
708 * the osync code to catch these locked, dirty buffers without requeuing
709 * any newly dirty buffers for write.
711 static int fsync_buffers_list(spinlock_t *lock, struct list_head *list)
713 struct buffer_head *bh;
714 struct list_head tmp;
715 struct address_space *mapping;
717 struct blk_plug plug;
719 INIT_LIST_HEAD(&tmp);
720 blk_start_plug(&plug);
723 while (!list_empty(list)) {
724 bh = BH_ENTRY(list->next);
725 mapping = bh->b_assoc_map;
726 __remove_assoc_queue(bh);
727 /* Avoid race with mark_buffer_dirty_inode() which does
728 * a lockless check and we rely on seeing the dirty bit */
730 if (buffer_dirty(bh) || buffer_locked(bh)) {
731 list_add(&bh->b_assoc_buffers, &tmp);
732 bh->b_assoc_map = mapping;
733 if (buffer_dirty(bh)) {
737 * Ensure any pending I/O completes so that
738 * write_dirty_buffer() actually writes the
739 * current contents - it is a noop if I/O is
740 * still in flight on potentially older
743 write_dirty_buffer(bh, REQ_SYNC);
746 * Kick off IO for the previous mapping. Note
747 * that we will not run the very last mapping,
748 * wait_on_buffer() will do that for us
749 * through sync_buffer().
758 blk_finish_plug(&plug);
761 while (!list_empty(&tmp)) {
762 bh = BH_ENTRY(tmp.prev);
764 mapping = bh->b_assoc_map;
765 __remove_assoc_queue(bh);
766 /* Avoid race with mark_buffer_dirty_inode() which does
767 * a lockless check and we rely on seeing the dirty bit */
769 if (buffer_dirty(bh)) {
770 list_add(&bh->b_assoc_buffers,
771 &mapping->private_list);
772 bh->b_assoc_map = mapping;
776 if (!buffer_uptodate(bh))
783 err2 = osync_buffers_list(lock, list);
791 * Invalidate any and all dirty buffers on a given inode. We are
792 * probably unmounting the fs, but that doesn't mean we have already
793 * done a sync(). Just drop the buffers from the inode list.
795 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
796 * assumes that all the buffers are against the blockdev. Not true
799 void invalidate_inode_buffers(struct inode *inode)
801 if (inode_has_buffers(inode)) {
802 struct address_space *mapping = &inode->i_data;
803 struct list_head *list = &mapping->private_list;
804 struct address_space *buffer_mapping = mapping->private_data;
806 spin_lock(&buffer_mapping->private_lock);
807 while (!list_empty(list))
808 __remove_assoc_queue(BH_ENTRY(list->next));
809 spin_unlock(&buffer_mapping->private_lock);
812 EXPORT_SYMBOL(invalidate_inode_buffers);
815 * Remove any clean buffers from the inode's buffer list. This is called
816 * when we're trying to free the inode itself. Those buffers can pin it.
818 * Returns true if all buffers were removed.
820 int remove_inode_buffers(struct inode *inode)
824 if (inode_has_buffers(inode)) {
825 struct address_space *mapping = &inode->i_data;
826 struct list_head *list = &mapping->private_list;
827 struct address_space *buffer_mapping = mapping->private_data;
829 spin_lock(&buffer_mapping->private_lock);
830 while (!list_empty(list)) {
831 struct buffer_head *bh = BH_ENTRY(list->next);
832 if (buffer_dirty(bh)) {
836 __remove_assoc_queue(bh);
838 spin_unlock(&buffer_mapping->private_lock);
844 * Create the appropriate buffers when given a page for data area and
845 * the size of each buffer.. Use the bh->b_this_page linked list to
846 * follow the buffers created. Return NULL if unable to create more
849 * The retry flag is used to differentiate async IO (paging, swapping)
850 * which may not fail from ordinary buffer allocations.
852 struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
855 struct buffer_head *bh, *head;
856 gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT;
858 struct mem_cgroup *memcg;
863 memcg = get_mem_cgroup_from_page(page);
864 memalloc_use_memcg(memcg);
868 while ((offset -= size) >= 0) {
869 bh = alloc_buffer_head(gfp);
873 bh->b_this_page = head;
879 /* Link the buffer to its page */
880 set_bh_page(bh, page, offset);
883 memalloc_unuse_memcg();
884 mem_cgroup_put(memcg);
887 * In case anything failed, we just free everything we got.
893 head = head->b_this_page;
894 free_buffer_head(bh);
900 EXPORT_SYMBOL_GPL(alloc_page_buffers);
903 link_dev_buffers(struct page *page, struct buffer_head *head)
905 struct buffer_head *bh, *tail;
910 bh = bh->b_this_page;
912 tail->b_this_page = head;
913 attach_page_buffers(page, head);
916 static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size)
918 sector_t retval = ~((sector_t)0);
919 loff_t sz = i_size_read(bdev->bd_inode);
922 unsigned int sizebits = blksize_bits(size);
923 retval = (sz >> sizebits);
929 * Initialise the state of a blockdev page's buffers.
932 init_page_buffers(struct page *page, struct block_device *bdev,
933 sector_t block, int size)
935 struct buffer_head *head = page_buffers(page);
936 struct buffer_head *bh = head;
937 int uptodate = PageUptodate(page);
938 sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size);
941 if (!buffer_mapped(bh)) {
943 bh->b_private = NULL;
945 bh->b_blocknr = block;
947 set_buffer_uptodate(bh);
948 if (block < end_block)
949 set_buffer_mapped(bh);
952 bh = bh->b_this_page;
953 } while (bh != head);
956 * Caller needs to validate requested block against end of device.
962 * Create the page-cache page that contains the requested block.
964 * This is used purely for blockdev mappings.
967 grow_dev_page(struct block_device *bdev, sector_t block,
968 pgoff_t index, int size, int sizebits, gfp_t gfp)
970 struct inode *inode = bdev->bd_inode;
972 struct buffer_head *bh;
974 int ret = 0; /* Will call free_more_memory() */
977 gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp;
980 * XXX: __getblk_slow() can not really deal with failure and
981 * will endlessly loop on improvised global reclaim. Prefer
982 * looping in the allocator rather than here, at least that
983 * code knows what it's doing.
985 gfp_mask |= __GFP_NOFAIL;
987 page = find_or_create_page(inode->i_mapping, index, gfp_mask);
989 BUG_ON(!PageLocked(page));
991 if (page_has_buffers(page)) {
992 bh = page_buffers(page);
993 if (bh->b_size == size) {
994 end_block = init_page_buffers(page, bdev,
995 (sector_t)index << sizebits,
999 if (!try_to_free_buffers(page))
1004 * Allocate some buffers for this page
1006 bh = alloc_page_buffers(page, size, true);
1009 * Link the page to the buffers and initialise them. Take the
1010 * lock to be atomic wrt __find_get_block(), which does not
1011 * run under the page lock.
1013 spin_lock(&inode->i_mapping->private_lock);
1014 link_dev_buffers(page, bh);
1015 end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits,
1017 spin_unlock(&inode->i_mapping->private_lock);
1019 ret = (block < end_block) ? 1 : -ENXIO;
1027 * Create buffers for the specified block device block's page. If
1028 * that page was dirty, the buffers are set dirty also.
1031 grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp)
1039 } while ((size << sizebits) < PAGE_SIZE);
1041 index = block >> sizebits;
1044 * Check for a block which wants to lie outside our maximum possible
1045 * pagecache index. (this comparison is done using sector_t types).
1047 if (unlikely(index != block >> sizebits)) {
1048 printk(KERN_ERR "%s: requested out-of-range block %llu for "
1050 __func__, (unsigned long long)block,
1055 /* Create a page with the proper size buffers.. */
1056 return grow_dev_page(bdev, block, index, size, sizebits, gfp);
1059 static struct buffer_head *
1060 __getblk_slow(struct block_device *bdev, sector_t block,
1061 unsigned size, gfp_t gfp)
1063 /* Size must be multiple of hard sectorsize */
1064 if (unlikely(size & (bdev_logical_block_size(bdev)-1) ||
1065 (size < 512 || size > PAGE_SIZE))) {
1066 printk(KERN_ERR "getblk(): invalid block size %d requested\n",
1068 printk(KERN_ERR "logical block size: %d\n",
1069 bdev_logical_block_size(bdev));
1076 struct buffer_head *bh;
1079 bh = __find_get_block(bdev, block, size);
1083 ret = grow_buffers(bdev, block, size, gfp);
1090 * The relationship between dirty buffers and dirty pages:
1092 * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1093 * the page is tagged dirty in the page cache.
1095 * At all times, the dirtiness of the buffers represents the dirtiness of
1096 * subsections of the page. If the page has buffers, the page dirty bit is
1097 * merely a hint about the true dirty state.
1099 * When a page is set dirty in its entirety, all its buffers are marked dirty
1100 * (if the page has buffers).
1102 * When a buffer is marked dirty, its page is dirtied, but the page's other
1105 * Also. When blockdev buffers are explicitly read with bread(), they
1106 * individually become uptodate. But their backing page remains not
1107 * uptodate - even if all of its buffers are uptodate. A subsequent
1108 * block_read_full_page() against that page will discover all the uptodate
1109 * buffers, will set the page uptodate and will perform no I/O.
1113 * mark_buffer_dirty - mark a buffer_head as needing writeout
1114 * @bh: the buffer_head to mark dirty
1116 * mark_buffer_dirty() will set the dirty bit against the buffer, then set
1117 * its backing page dirty, then tag the page as dirty in the page cache
1118 * and then attach the address_space's inode to its superblock's dirty
1121 * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
1122 * i_pages lock and mapping->host->i_lock.
1124 void mark_buffer_dirty(struct buffer_head *bh)
1126 WARN_ON_ONCE(!buffer_uptodate(bh));
1128 trace_block_dirty_buffer(bh);
1131 * Very *carefully* optimize the it-is-already-dirty case.
1133 * Don't let the final "is it dirty" escape to before we
1134 * perhaps modified the buffer.
1136 if (buffer_dirty(bh)) {
1138 if (buffer_dirty(bh))
1142 if (!test_set_buffer_dirty(bh)) {
1143 struct page *page = bh->b_page;
1144 struct address_space *mapping = NULL;
1146 lock_page_memcg(page);
1147 if (!TestSetPageDirty(page)) {
1148 mapping = page_mapping(page);
1150 __set_page_dirty(page, mapping, 0);
1152 unlock_page_memcg(page);
1154 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1157 EXPORT_SYMBOL(mark_buffer_dirty);
1159 void mark_buffer_write_io_error(struct buffer_head *bh)
1161 set_buffer_write_io_error(bh);
1162 /* FIXME: do we need to set this in both places? */
1163 if (bh->b_page && bh->b_page->mapping)
1164 mapping_set_error(bh->b_page->mapping, -EIO);
1165 if (bh->b_assoc_map)
1166 mapping_set_error(bh->b_assoc_map, -EIO);
1168 EXPORT_SYMBOL(mark_buffer_write_io_error);
1171 * Decrement a buffer_head's reference count. If all buffers against a page
1172 * have zero reference count, are clean and unlocked, and if the page is clean
1173 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1174 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1175 * a page but it ends up not being freed, and buffers may later be reattached).
1177 void __brelse(struct buffer_head * buf)
1179 if (atomic_read(&buf->b_count)) {
1183 WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n");
1185 EXPORT_SYMBOL(__brelse);
1188 * bforget() is like brelse(), except it discards any
1189 * potentially dirty data.
1191 void __bforget(struct buffer_head *bh)
1193 clear_buffer_dirty(bh);
1194 if (bh->b_assoc_map) {
1195 struct address_space *buffer_mapping = bh->b_page->mapping;
1197 spin_lock(&buffer_mapping->private_lock);
1198 list_del_init(&bh->b_assoc_buffers);
1199 bh->b_assoc_map = NULL;
1200 spin_unlock(&buffer_mapping->private_lock);
1204 EXPORT_SYMBOL(__bforget);
1206 static struct buffer_head *__bread_slow(struct buffer_head *bh)
1209 if (buffer_uptodate(bh)) {
1214 bh->b_end_io = end_buffer_read_sync;
1215 submit_bh(REQ_OP_READ, 0, bh);
1217 if (buffer_uptodate(bh))
1225 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1226 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1227 * refcount elevated by one when they're in an LRU. A buffer can only appear
1228 * once in a particular CPU's LRU. A single buffer can be present in multiple
1229 * CPU's LRUs at the same time.
1231 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1232 * sb_find_get_block().
1234 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1235 * a local interrupt disable for that.
1238 #define BH_LRU_SIZE 16
1241 struct buffer_head *bhs[BH_LRU_SIZE];
1244 static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }};
1247 #define bh_lru_lock() local_irq_disable()
1248 #define bh_lru_unlock() local_irq_enable()
1250 #define bh_lru_lock() preempt_disable()
1251 #define bh_lru_unlock() preempt_enable()
1254 static inline void check_irqs_on(void)
1256 #ifdef irqs_disabled
1257 BUG_ON(irqs_disabled());
1262 * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is
1263 * inserted at the front, and the buffer_head at the back if any is evicted.
1264 * Or, if already in the LRU it is moved to the front.
1266 static void bh_lru_install(struct buffer_head *bh)
1268 struct buffer_head *evictee = bh;
1275 b = this_cpu_ptr(&bh_lrus);
1276 for (i = 0; i < BH_LRU_SIZE; i++) {
1277 swap(evictee, b->bhs[i]);
1278 if (evictee == bh) {
1290 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1292 static struct buffer_head *
1293 lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
1295 struct buffer_head *ret = NULL;
1300 for (i = 0; i < BH_LRU_SIZE; i++) {
1301 struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]);
1303 if (bh && bh->b_blocknr == block && bh->b_bdev == bdev &&
1304 bh->b_size == size) {
1307 __this_cpu_write(bh_lrus.bhs[i],
1308 __this_cpu_read(bh_lrus.bhs[i - 1]));
1311 __this_cpu_write(bh_lrus.bhs[0], bh);
1323 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1324 * it in the LRU and mark it as accessed. If it is not present then return
1327 struct buffer_head *
1328 __find_get_block(struct block_device *bdev, sector_t block, unsigned size)
1330 struct buffer_head *bh = lookup_bh_lru(bdev, block, size);
1333 /* __find_get_block_slow will mark the page accessed */
1334 bh = __find_get_block_slow(bdev, block);
1342 EXPORT_SYMBOL(__find_get_block);
1345 * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
1346 * which corresponds to the passed block_device, block and size. The
1347 * returned buffer has its reference count incremented.
1349 * __getblk_gfp() will lock up the machine if grow_dev_page's
1350 * try_to_free_buffers() attempt is failing. FIXME, perhaps?
1352 struct buffer_head *
1353 __getblk_gfp(struct block_device *bdev, sector_t block,
1354 unsigned size, gfp_t gfp)
1356 struct buffer_head *bh = __find_get_block(bdev, block, size);
1360 bh = __getblk_slow(bdev, block, size, gfp);
1363 EXPORT_SYMBOL(__getblk_gfp);
1366 * Do async read-ahead on a buffer..
1368 void __breadahead(struct block_device *bdev, sector_t block, unsigned size)
1370 struct buffer_head *bh = __getblk(bdev, block, size);
1372 ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh);
1376 EXPORT_SYMBOL(__breadahead);
1379 * __bread_gfp() - reads a specified block and returns the bh
1380 * @bdev: the block_device to read from
1381 * @block: number of block
1382 * @size: size (in bytes) to read
1383 * @gfp: page allocation flag
1385 * Reads a specified block, and returns buffer head that contains it.
1386 * The page cache can be allocated from non-movable area
1387 * not to prevent page migration if you set gfp to zero.
1388 * It returns NULL if the block was unreadable.
1390 struct buffer_head *
1391 __bread_gfp(struct block_device *bdev, sector_t block,
1392 unsigned size, gfp_t gfp)
1394 struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
1396 if (likely(bh) && !buffer_uptodate(bh))
1397 bh = __bread_slow(bh);
1400 EXPORT_SYMBOL(__bread_gfp);
1403 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1404 * This doesn't race because it runs in each cpu either in irq
1405 * or with preempt disabled.
1407 static void invalidate_bh_lru(void *arg)
1409 struct bh_lru *b = &get_cpu_var(bh_lrus);
1412 for (i = 0; i < BH_LRU_SIZE; i++) {
1416 put_cpu_var(bh_lrus);
1419 static bool has_bh_in_lru(int cpu, void *dummy)
1421 struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu);
1424 for (i = 0; i < BH_LRU_SIZE; i++) {
1432 void invalidate_bh_lrus(void)
1434 on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL);
1436 EXPORT_SYMBOL_GPL(invalidate_bh_lrus);
1438 void set_bh_page(struct buffer_head *bh,
1439 struct page *page, unsigned long offset)
1442 BUG_ON(offset >= PAGE_SIZE);
1443 if (PageHighMem(page))
1445 * This catches illegal uses and preserves the offset:
1447 bh->b_data = (char *)(0 + offset);
1449 bh->b_data = page_address(page) + offset;
1451 EXPORT_SYMBOL(set_bh_page);
1454 * Called when truncating a buffer on a page completely.
1457 /* Bits that are cleared during an invalidate */
1458 #define BUFFER_FLAGS_DISCARD \
1459 (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1460 1 << BH_Delay | 1 << BH_Unwritten)
1462 static void discard_buffer(struct buffer_head * bh)
1464 unsigned long b_state, b_state_old;
1467 clear_buffer_dirty(bh);
1469 b_state = bh->b_state;
1471 b_state_old = cmpxchg(&bh->b_state, b_state,
1472 (b_state & ~BUFFER_FLAGS_DISCARD));
1473 if (b_state_old == b_state)
1475 b_state = b_state_old;
1481 * block_invalidatepage - invalidate part or all of a buffer-backed page
1483 * @page: the page which is affected
1484 * @offset: start of the range to invalidate
1485 * @length: length of the range to invalidate
1487 * block_invalidatepage() is called when all or part of the page has become
1488 * invalidated by a truncate operation.
1490 * block_invalidatepage() does not have to release all buffers, but it must
1491 * ensure that no dirty buffer is left outside @offset and that no I/O
1492 * is underway against any of the blocks which are outside the truncation
1493 * point. Because the caller is about to free (and possibly reuse) those
1496 void block_invalidatepage(struct page *page, unsigned int offset,
1497 unsigned int length)
1499 struct buffer_head *head, *bh, *next;
1500 unsigned int curr_off = 0;
1501 unsigned int stop = length + offset;
1503 BUG_ON(!PageLocked(page));
1504 if (!page_has_buffers(page))
1508 * Check for overflow
1510 BUG_ON(stop > PAGE_SIZE || stop < length);
1512 head = page_buffers(page);
1515 unsigned int next_off = curr_off + bh->b_size;
1516 next = bh->b_this_page;
1519 * Are we still fully in range ?
1521 if (next_off > stop)
1525 * is this block fully invalidated?
1527 if (offset <= curr_off)
1529 curr_off = next_off;
1531 } while (bh != head);
1534 * We release buffers only if the entire page is being invalidated.
1535 * The get_block cached value has been unconditionally invalidated,
1536 * so real IO is not possible anymore.
1538 if (length == PAGE_SIZE)
1539 try_to_release_page(page, 0);
1543 EXPORT_SYMBOL(block_invalidatepage);
1547 * We attach and possibly dirty the buffers atomically wrt
1548 * __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1549 * is already excluded via the page lock.
1551 void create_empty_buffers(struct page *page,
1552 unsigned long blocksize, unsigned long b_state)
1554 struct buffer_head *bh, *head, *tail;
1556 head = alloc_page_buffers(page, blocksize, true);
1559 bh->b_state |= b_state;
1561 bh = bh->b_this_page;
1563 tail->b_this_page = head;
1565 spin_lock(&page->mapping->private_lock);
1566 if (PageUptodate(page) || PageDirty(page)) {
1569 if (PageDirty(page))
1570 set_buffer_dirty(bh);
1571 if (PageUptodate(page))
1572 set_buffer_uptodate(bh);
1573 bh = bh->b_this_page;
1574 } while (bh != head);
1576 attach_page_buffers(page, head);
1577 spin_unlock(&page->mapping->private_lock);
1579 EXPORT_SYMBOL(create_empty_buffers);
1582 * clean_bdev_aliases: clean a range of buffers in block device
1583 * @bdev: Block device to clean buffers in
1584 * @block: Start of a range of blocks to clean
1585 * @len: Number of blocks to clean
1587 * We are taking a range of blocks for data and we don't want writeback of any
1588 * buffer-cache aliases starting from return from this function and until the
1589 * moment when something will explicitly mark the buffer dirty (hopefully that
1590 * will not happen until we will free that block ;-) We don't even need to mark
1591 * it not-uptodate - nobody can expect anything from a newly allocated buffer
1592 * anyway. We used to use unmap_buffer() for such invalidation, but that was
1593 * wrong. We definitely don't want to mark the alias unmapped, for example - it
1594 * would confuse anyone who might pick it with bread() afterwards...
1596 * Also.. Note that bforget() doesn't lock the buffer. So there can be
1597 * writeout I/O going on against recently-freed buffers. We don't wait on that
1598 * I/O in bforget() - it's more efficient to wait on the I/O only if we really
1599 * need to. That happens here.
1601 void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len)
1603 struct inode *bd_inode = bdev->bd_inode;
1604 struct address_space *bd_mapping = bd_inode->i_mapping;
1605 struct pagevec pvec;
1606 pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
1609 struct buffer_head *bh;
1610 struct buffer_head *head;
1612 end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits);
1613 pagevec_init(&pvec);
1614 while (pagevec_lookup_range(&pvec, bd_mapping, &index, end)) {
1615 count = pagevec_count(&pvec);
1616 for (i = 0; i < count; i++) {
1617 struct page *page = pvec.pages[i];
1619 if (!page_has_buffers(page))
1622 * We use page lock instead of bd_mapping->private_lock
1623 * to pin buffers here since we can afford to sleep and
1624 * it scales better than a global spinlock lock.
1627 /* Recheck when the page is locked which pins bhs */
1628 if (!page_has_buffers(page))
1630 head = page_buffers(page);
1633 if (!buffer_mapped(bh) || (bh->b_blocknr < block))
1635 if (bh->b_blocknr >= block + len)
1637 clear_buffer_dirty(bh);
1639 clear_buffer_req(bh);
1641 bh = bh->b_this_page;
1642 } while (bh != head);
1646 pagevec_release(&pvec);
1648 /* End of range already reached? */
1649 if (index > end || !index)
1653 EXPORT_SYMBOL(clean_bdev_aliases);
1656 * Size is a power-of-two in the range 512..PAGE_SIZE,
1657 * and the case we care about most is PAGE_SIZE.
1659 * So this *could* possibly be written with those
1660 * constraints in mind (relevant mostly if some
1661 * architecture has a slow bit-scan instruction)
1663 static inline int block_size_bits(unsigned int blocksize)
1665 return ilog2(blocksize);
1668 static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state)
1670 BUG_ON(!PageLocked(page));
1672 if (!page_has_buffers(page))
1673 create_empty_buffers(page, 1 << READ_ONCE(inode->i_blkbits),
1675 return page_buffers(page);
1679 * NOTE! All mapped/uptodate combinations are valid:
1681 * Mapped Uptodate Meaning
1683 * No No "unknown" - must do get_block()
1684 * No Yes "hole" - zero-filled
1685 * Yes No "allocated" - allocated on disk, not read in
1686 * Yes Yes "valid" - allocated and up-to-date in memory.
1688 * "Dirty" is valid only with the last case (mapped+uptodate).
1692 * While block_write_full_page is writing back the dirty buffers under
1693 * the page lock, whoever dirtied the buffers may decide to clean them
1694 * again at any time. We handle that by only looking at the buffer
1695 * state inside lock_buffer().
1697 * If block_write_full_page() is called for regular writeback
1698 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1699 * locked buffer. This only can happen if someone has written the buffer
1700 * directly, with submit_bh(). At the address_space level PageWriteback
1701 * prevents this contention from occurring.
1703 * If block_write_full_page() is called with wbc->sync_mode ==
1704 * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this
1705 * causes the writes to be flagged as synchronous writes.
1707 int __block_write_full_page(struct inode *inode, struct page *page,
1708 get_block_t *get_block, struct writeback_control *wbc,
1709 bh_end_io_t *handler)
1713 sector_t last_block;
1714 struct buffer_head *bh, *head;
1715 unsigned int blocksize, bbits;
1716 int nr_underway = 0;
1717 int write_flags = wbc_to_write_flags(wbc);
1719 head = create_page_buffers(page, inode,
1720 (1 << BH_Dirty)|(1 << BH_Uptodate));
1723 * Be very careful. We have no exclusion from __set_page_dirty_buffers
1724 * here, and the (potentially unmapped) buffers may become dirty at
1725 * any time. If a buffer becomes dirty here after we've inspected it
1726 * then we just miss that fact, and the page stays dirty.
1728 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1729 * handle that here by just cleaning them.
1733 blocksize = bh->b_size;
1734 bbits = block_size_bits(blocksize);
1736 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1737 last_block = (i_size_read(inode) - 1) >> bbits;
1740 * Get all the dirty buffers mapped to disk addresses and
1741 * handle any aliases from the underlying blockdev's mapping.
1744 if (block > last_block) {
1746 * mapped buffers outside i_size will occur, because
1747 * this page can be outside i_size when there is a
1748 * truncate in progress.
1751 * The buffer was zeroed by block_write_full_page()
1753 clear_buffer_dirty(bh);
1754 set_buffer_uptodate(bh);
1755 } else if ((!buffer_mapped(bh) || buffer_delay(bh)) &&
1757 WARN_ON(bh->b_size != blocksize);
1758 err = get_block(inode, block, bh, 1);
1761 clear_buffer_delay(bh);
1762 if (buffer_new(bh)) {
1763 /* blockdev mappings never come here */
1764 clear_buffer_new(bh);
1765 clean_bdev_bh_alias(bh);
1768 bh = bh->b_this_page;
1770 } while (bh != head);
1773 if (!buffer_mapped(bh))
1776 * If it's a fully non-blocking write attempt and we cannot
1777 * lock the buffer then redirty the page. Note that this can
1778 * potentially cause a busy-wait loop from writeback threads
1779 * and kswapd activity, but those code paths have their own
1780 * higher-level throttling.
1782 if (wbc->sync_mode != WB_SYNC_NONE) {
1784 } else if (!trylock_buffer(bh)) {
1785 redirty_page_for_writepage(wbc, page);
1788 if (test_clear_buffer_dirty(bh)) {
1789 mark_buffer_async_write_endio(bh, handler);
1793 } while ((bh = bh->b_this_page) != head);
1796 * The page and its buffers are protected by PageWriteback(), so we can
1797 * drop the bh refcounts early.
1799 BUG_ON(PageWriteback(page));
1800 set_page_writeback(page);
1803 struct buffer_head *next = bh->b_this_page;
1804 if (buffer_async_write(bh)) {
1805 submit_bh_wbc(REQ_OP_WRITE, write_flags, bh,
1806 inode->i_write_hint, wbc);
1810 } while (bh != head);
1815 if (nr_underway == 0) {
1817 * The page was marked dirty, but the buffers were
1818 * clean. Someone wrote them back by hand with
1819 * ll_rw_block/submit_bh. A rare case.
1821 end_page_writeback(page);
1824 * The page and buffer_heads can be released at any time from
1832 * ENOSPC, or some other error. We may already have added some
1833 * blocks to the file, so we need to write these out to avoid
1834 * exposing stale data.
1835 * The page is currently locked and not marked for writeback
1838 /* Recovery: lock and submit the mapped buffers */
1840 if (buffer_mapped(bh) && buffer_dirty(bh) &&
1841 !buffer_delay(bh)) {
1843 mark_buffer_async_write_endio(bh, handler);
1846 * The buffer may have been set dirty during
1847 * attachment to a dirty page.
1849 clear_buffer_dirty(bh);
1851 } while ((bh = bh->b_this_page) != head);
1853 BUG_ON(PageWriteback(page));
1854 mapping_set_error(page->mapping, err);
1855 set_page_writeback(page);
1857 struct buffer_head *next = bh->b_this_page;
1858 if (buffer_async_write(bh)) {
1859 clear_buffer_dirty(bh);
1860 submit_bh_wbc(REQ_OP_WRITE, write_flags, bh,
1861 inode->i_write_hint, wbc);
1865 } while (bh != head);
1869 EXPORT_SYMBOL(__block_write_full_page);
1872 * If a page has any new buffers, zero them out here, and mark them uptodate
1873 * and dirty so they'll be written out (in order to prevent uninitialised
1874 * block data from leaking). And clear the new bit.
1876 void page_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1878 unsigned int block_start, block_end;
1879 struct buffer_head *head, *bh;
1881 BUG_ON(!PageLocked(page));
1882 if (!page_has_buffers(page))
1885 bh = head = page_buffers(page);
1888 block_end = block_start + bh->b_size;
1890 if (buffer_new(bh)) {
1891 if (block_end > from && block_start < to) {
1892 if (!PageUptodate(page)) {
1893 unsigned start, size;
1895 start = max(from, block_start);
1896 size = min(to, block_end) - start;
1898 zero_user(page, start, size);
1899 set_buffer_uptodate(bh);
1902 clear_buffer_new(bh);
1903 mark_buffer_dirty(bh);
1907 block_start = block_end;
1908 bh = bh->b_this_page;
1909 } while (bh != head);
1911 EXPORT_SYMBOL(page_zero_new_buffers);
1914 iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh,
1915 struct iomap *iomap)
1917 loff_t offset = block << inode->i_blkbits;
1919 bh->b_bdev = iomap->bdev;
1922 * Block points to offset in file we need to map, iomap contains
1923 * the offset at which the map starts. If the map ends before the
1924 * current block, then do not map the buffer and let the caller
1927 BUG_ON(offset >= iomap->offset + iomap->length);
1929 switch (iomap->type) {
1932 * If the buffer is not up to date or beyond the current EOF,
1933 * we need to mark it as new to ensure sub-block zeroing is
1934 * executed if necessary.
1936 if (!buffer_uptodate(bh) ||
1937 (offset >= i_size_read(inode)))
1940 case IOMAP_DELALLOC:
1941 if (!buffer_uptodate(bh) ||
1942 (offset >= i_size_read(inode)))
1944 set_buffer_uptodate(bh);
1945 set_buffer_mapped(bh);
1946 set_buffer_delay(bh);
1948 case IOMAP_UNWRITTEN:
1950 * For unwritten regions, we always need to ensure that regions
1951 * in the block we are not writing to are zeroed. Mark the
1952 * buffer as new to ensure this.
1955 set_buffer_unwritten(bh);
1958 if ((iomap->flags & IOMAP_F_NEW) ||
1959 offset >= i_size_read(inode))
1961 bh->b_blocknr = (iomap->addr + offset - iomap->offset) >>
1963 set_buffer_mapped(bh);
1968 int __block_write_begin_int(struct page *page, loff_t pos, unsigned len,
1969 get_block_t *get_block, struct iomap *iomap)
1971 unsigned from = pos & (PAGE_SIZE - 1);
1972 unsigned to = from + len;
1973 struct inode *inode = page->mapping->host;
1974 unsigned block_start, block_end;
1977 unsigned blocksize, bbits;
1978 struct buffer_head *bh, *head, *wait[2], **wait_bh=wait;
1980 BUG_ON(!PageLocked(page));
1981 BUG_ON(from > PAGE_SIZE);
1982 BUG_ON(to > PAGE_SIZE);
1985 head = create_page_buffers(page, inode, 0);
1986 blocksize = head->b_size;
1987 bbits = block_size_bits(blocksize);
1989 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1991 for(bh = head, block_start = 0; bh != head || !block_start;
1992 block++, block_start=block_end, bh = bh->b_this_page) {
1993 block_end = block_start + blocksize;
1994 if (block_end <= from || block_start >= to) {
1995 if (PageUptodate(page)) {
1996 if (!buffer_uptodate(bh))
1997 set_buffer_uptodate(bh);
2002 clear_buffer_new(bh);
2003 if (!buffer_mapped(bh)) {
2004 WARN_ON(bh->b_size != blocksize);
2006 err = get_block(inode, block, bh, 1);
2010 iomap_to_bh(inode, block, bh, iomap);
2013 if (buffer_new(bh)) {
2014 clean_bdev_bh_alias(bh);
2015 if (PageUptodate(page)) {
2016 clear_buffer_new(bh);
2017 set_buffer_uptodate(bh);
2018 mark_buffer_dirty(bh);
2021 if (block_end > to || block_start < from)
2022 zero_user_segments(page,
2028 if (PageUptodate(page)) {
2029 if (!buffer_uptodate(bh))
2030 set_buffer_uptodate(bh);
2033 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
2034 !buffer_unwritten(bh) &&
2035 (block_start < from || block_end > to)) {
2036 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
2041 * If we issued read requests - let them complete.
2043 while(wait_bh > wait) {
2044 wait_on_buffer(*--wait_bh);
2045 if (!buffer_uptodate(*wait_bh))
2049 page_zero_new_buffers(page, from, to);
2053 int __block_write_begin(struct page *page, loff_t pos, unsigned len,
2054 get_block_t *get_block)
2056 return __block_write_begin_int(page, pos, len, get_block, NULL);
2058 EXPORT_SYMBOL(__block_write_begin);
2060 static int __block_commit_write(struct inode *inode, struct page *page,
2061 unsigned from, unsigned to)
2063 unsigned block_start, block_end;
2066 struct buffer_head *bh, *head;
2068 bh = head = page_buffers(page);
2069 blocksize = bh->b_size;
2073 block_end = block_start + blocksize;
2074 if (block_end <= from || block_start >= to) {
2075 if (!buffer_uptodate(bh))
2078 set_buffer_uptodate(bh);
2079 mark_buffer_dirty(bh);
2081 clear_buffer_new(bh);
2083 block_start = block_end;
2084 bh = bh->b_this_page;
2085 } while (bh != head);
2088 * If this is a partial write which happened to make all buffers
2089 * uptodate then we can optimize away a bogus readpage() for
2090 * the next read(). Here we 'discover' whether the page went
2091 * uptodate as a result of this (potentially partial) write.
2094 SetPageUptodate(page);
2099 * block_write_begin takes care of the basic task of block allocation and
2100 * bringing partial write blocks uptodate first.
2102 * The filesystem needs to handle block truncation upon failure.
2104 int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
2105 unsigned flags, struct page **pagep, get_block_t *get_block)
2107 pgoff_t index = pos >> PAGE_SHIFT;
2111 page = grab_cache_page_write_begin(mapping, index, flags);
2115 status = __block_write_begin(page, pos, len, get_block);
2116 if (unlikely(status)) {
2125 EXPORT_SYMBOL(block_write_begin);
2127 int block_write_end(struct file *file, struct address_space *mapping,
2128 loff_t pos, unsigned len, unsigned copied,
2129 struct page *page, void *fsdata)
2131 struct inode *inode = mapping->host;
2134 start = pos & (PAGE_SIZE - 1);
2136 if (unlikely(copied < len)) {
2138 * The buffers that were written will now be uptodate, so we
2139 * don't have to worry about a readpage reading them and
2140 * overwriting a partial write. However if we have encountered
2141 * a short write and only partially written into a buffer, it
2142 * will not be marked uptodate, so a readpage might come in and
2143 * destroy our partial write.
2145 * Do the simplest thing, and just treat any short write to a
2146 * non uptodate page as a zero-length write, and force the
2147 * caller to redo the whole thing.
2149 if (!PageUptodate(page))
2152 page_zero_new_buffers(page, start+copied, start+len);
2154 flush_dcache_page(page);
2156 /* This could be a short (even 0-length) commit */
2157 __block_commit_write(inode, page, start, start+copied);
2161 EXPORT_SYMBOL(block_write_end);
2163 int generic_write_end(struct file *file, struct address_space *mapping,
2164 loff_t pos, unsigned len, unsigned copied,
2165 struct page *page, void *fsdata)
2167 struct inode *inode = mapping->host;
2168 loff_t old_size = inode->i_size;
2169 bool i_size_changed = false;
2171 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
2174 * No need to use i_size_read() here, the i_size cannot change under us
2175 * because we hold i_rwsem.
2177 * But it's important to update i_size while still holding page lock:
2178 * page writeout could otherwise come in and zero beyond i_size.
2180 if (pos + copied > inode->i_size) {
2181 i_size_write(inode, pos + copied);
2182 i_size_changed = true;
2189 pagecache_isize_extended(inode, old_size, pos);
2191 * Don't mark the inode dirty under page lock. First, it unnecessarily
2192 * makes the holding time of page lock longer. Second, it forces lock
2193 * ordering of page lock and transaction start for journaling
2197 mark_inode_dirty(inode);
2200 EXPORT_SYMBOL(generic_write_end);
2203 * block_is_partially_uptodate checks whether buffers within a page are
2206 * Returns true if all buffers which correspond to a file portion
2207 * we want to read are uptodate.
2209 int block_is_partially_uptodate(struct page *page, unsigned long from,
2210 unsigned long count)
2212 unsigned block_start, block_end, blocksize;
2214 struct buffer_head *bh, *head;
2217 if (!page_has_buffers(page))
2220 head = page_buffers(page);
2221 blocksize = head->b_size;
2222 to = min_t(unsigned, PAGE_SIZE - from, count);
2224 if (from < blocksize && to > PAGE_SIZE - blocksize)
2230 block_end = block_start + blocksize;
2231 if (block_end > from && block_start < to) {
2232 if (!buffer_uptodate(bh)) {
2236 if (block_end >= to)
2239 block_start = block_end;
2240 bh = bh->b_this_page;
2241 } while (bh != head);
2245 EXPORT_SYMBOL(block_is_partially_uptodate);
2248 * Generic "read page" function for block devices that have the normal
2249 * get_block functionality. This is most of the block device filesystems.
2250 * Reads the page asynchronously --- the unlock_buffer() and
2251 * set/clear_buffer_uptodate() functions propagate buffer state into the
2252 * page struct once IO has completed.
2254 int block_read_full_page(struct page *page, get_block_t *get_block)
2256 struct inode *inode = page->mapping->host;
2257 sector_t iblock, lblock;
2258 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
2259 unsigned int blocksize, bbits;
2261 int fully_mapped = 1;
2263 head = create_page_buffers(page, inode, 0);
2264 blocksize = head->b_size;
2265 bbits = block_size_bits(blocksize);
2267 iblock = (sector_t)page->index << (PAGE_SHIFT - bbits);
2268 lblock = (i_size_read(inode)+blocksize-1) >> bbits;
2274 if (buffer_uptodate(bh))
2277 if (!buffer_mapped(bh)) {
2281 if (iblock < lblock) {
2282 WARN_ON(bh->b_size != blocksize);
2283 err = get_block(inode, iblock, bh, 0);
2287 if (!buffer_mapped(bh)) {
2288 zero_user(page, i * blocksize, blocksize);
2290 set_buffer_uptodate(bh);
2294 * get_block() might have updated the buffer
2297 if (buffer_uptodate(bh))
2301 } while (i++, iblock++, (bh = bh->b_this_page) != head);
2304 SetPageMappedToDisk(page);
2308 * All buffers are uptodate - we can set the page uptodate
2309 * as well. But not if get_block() returned an error.
2311 if (!PageError(page))
2312 SetPageUptodate(page);
2317 /* Stage two: lock the buffers */
2318 for (i = 0; i < nr; i++) {
2321 mark_buffer_async_read(bh);
2325 * Stage 3: start the IO. Check for uptodateness
2326 * inside the buffer lock in case another process reading
2327 * the underlying blockdev brought it uptodate (the sct fix).
2329 for (i = 0; i < nr; i++) {
2331 if (buffer_uptodate(bh))
2332 end_buffer_async_read(bh, 1);
2334 submit_bh(REQ_OP_READ, 0, bh);
2338 EXPORT_SYMBOL(block_read_full_page);
2340 /* utility function for filesystems that need to do work on expanding
2341 * truncates. Uses filesystem pagecache writes to allow the filesystem to
2342 * deal with the hole.
2344 int generic_cont_expand_simple(struct inode *inode, loff_t size)
2346 struct address_space *mapping = inode->i_mapping;
2351 err = inode_newsize_ok(inode, size);
2355 err = pagecache_write_begin(NULL, mapping, size, 0,
2356 AOP_FLAG_CONT_EXPAND, &page, &fsdata);
2360 err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata);
2366 EXPORT_SYMBOL(generic_cont_expand_simple);
2368 static int cont_expand_zero(struct file *file, struct address_space *mapping,
2369 loff_t pos, loff_t *bytes)
2371 struct inode *inode = mapping->host;
2372 unsigned int blocksize = i_blocksize(inode);
2375 pgoff_t index, curidx;
2377 unsigned zerofrom, offset, len;
2380 index = pos >> PAGE_SHIFT;
2381 offset = pos & ~PAGE_MASK;
2383 while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) {
2384 zerofrom = curpos & ~PAGE_MASK;
2385 if (zerofrom & (blocksize-1)) {
2386 *bytes |= (blocksize-1);
2389 len = PAGE_SIZE - zerofrom;
2391 err = pagecache_write_begin(file, mapping, curpos, len, 0,
2395 zero_user(page, zerofrom, len);
2396 err = pagecache_write_end(file, mapping, curpos, len, len,
2403 balance_dirty_pages_ratelimited(mapping);
2405 if (fatal_signal_pending(current)) {
2411 /* page covers the boundary, find the boundary offset */
2412 if (index == curidx) {
2413 zerofrom = curpos & ~PAGE_MASK;
2414 /* if we will expand the thing last block will be filled */
2415 if (offset <= zerofrom) {
2418 if (zerofrom & (blocksize-1)) {
2419 *bytes |= (blocksize-1);
2422 len = offset - zerofrom;
2424 err = pagecache_write_begin(file, mapping, curpos, len, 0,
2428 zero_user(page, zerofrom, len);
2429 err = pagecache_write_end(file, mapping, curpos, len, len,
2441 * For moronic filesystems that do not allow holes in file.
2442 * We may have to extend the file.
2444 int cont_write_begin(struct file *file, struct address_space *mapping,
2445 loff_t pos, unsigned len, unsigned flags,
2446 struct page **pagep, void **fsdata,
2447 get_block_t *get_block, loff_t *bytes)
2449 struct inode *inode = mapping->host;
2450 unsigned int blocksize = i_blocksize(inode);
2451 unsigned int zerofrom;
2454 err = cont_expand_zero(file, mapping, pos, bytes);
2458 zerofrom = *bytes & ~PAGE_MASK;
2459 if (pos+len > *bytes && zerofrom & (blocksize-1)) {
2460 *bytes |= (blocksize-1);
2464 return block_write_begin(mapping, pos, len, flags, pagep, get_block);
2466 EXPORT_SYMBOL(cont_write_begin);
2468 int block_commit_write(struct page *page, unsigned from, unsigned to)
2470 struct inode *inode = page->mapping->host;
2471 __block_commit_write(inode,page,from,to);
2474 EXPORT_SYMBOL(block_commit_write);
2477 * block_page_mkwrite() is not allowed to change the file size as it gets
2478 * called from a page fault handler when a page is first dirtied. Hence we must
2479 * be careful to check for EOF conditions here. We set the page up correctly
2480 * for a written page which means we get ENOSPC checking when writing into
2481 * holes and correct delalloc and unwritten extent mapping on filesystems that
2482 * support these features.
2484 * We are not allowed to take the i_mutex here so we have to play games to
2485 * protect against truncate races as the page could now be beyond EOF. Because
2486 * truncate writes the inode size before removing pages, once we have the
2487 * page lock we can determine safely if the page is beyond EOF. If it is not
2488 * beyond EOF, then the page is guaranteed safe against truncation until we
2491 * Direct callers of this function should protect against filesystem freezing
2492 * using sb_start_pagefault() - sb_end_pagefault() functions.
2494 int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
2495 get_block_t get_block)
2497 struct page *page = vmf->page;
2498 struct inode *inode = file_inode(vma->vm_file);
2504 size = i_size_read(inode);
2505 if ((page->mapping != inode->i_mapping) ||
2506 (page_offset(page) > size)) {
2507 /* We overload EFAULT to mean page got truncated */
2512 /* page is wholly or partially inside EOF */
2513 if (((page->index + 1) << PAGE_SHIFT) > size)
2514 end = size & ~PAGE_MASK;
2518 ret = __block_write_begin(page, 0, end, get_block);
2520 ret = block_commit_write(page, 0, end);
2522 if (unlikely(ret < 0))
2524 set_page_dirty(page);
2525 wait_for_stable_page(page);
2531 EXPORT_SYMBOL(block_page_mkwrite);
2534 * nobh_write_begin()'s prereads are special: the buffer_heads are freed
2535 * immediately, while under the page lock. So it needs a special end_io
2536 * handler which does not touch the bh after unlocking it.
2538 static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate)
2540 __end_buffer_read_notouch(bh, uptodate);
2544 * Attach the singly-linked list of buffers created by nobh_write_begin, to
2545 * the page (converting it to circular linked list and taking care of page
2548 static void attach_nobh_buffers(struct page *page, struct buffer_head *head)
2550 struct buffer_head *bh;
2552 BUG_ON(!PageLocked(page));
2554 spin_lock(&page->mapping->private_lock);
2557 if (PageDirty(page))
2558 set_buffer_dirty(bh);
2559 if (!bh->b_this_page)
2560 bh->b_this_page = head;
2561 bh = bh->b_this_page;
2562 } while (bh != head);
2563 attach_page_buffers(page, head);
2564 spin_unlock(&page->mapping->private_lock);
2568 * On entry, the page is fully not uptodate.
2569 * On exit the page is fully uptodate in the areas outside (from,to)
2570 * The filesystem needs to handle block truncation upon failure.
2572 int nobh_write_begin(struct address_space *mapping,
2573 loff_t pos, unsigned len, unsigned flags,
2574 struct page **pagep, void **fsdata,
2575 get_block_t *get_block)
2577 struct inode *inode = mapping->host;
2578 const unsigned blkbits = inode->i_blkbits;
2579 const unsigned blocksize = 1 << blkbits;
2580 struct buffer_head *head, *bh;
2584 unsigned block_in_page;
2585 unsigned block_start, block_end;
2586 sector_t block_in_file;
2589 int is_mapped_to_disk = 1;
2591 index = pos >> PAGE_SHIFT;
2592 from = pos & (PAGE_SIZE - 1);
2595 page = grab_cache_page_write_begin(mapping, index, flags);
2601 if (page_has_buffers(page)) {
2602 ret = __block_write_begin(page, pos, len, get_block);
2608 if (PageMappedToDisk(page))
2612 * Allocate buffers so that we can keep track of state, and potentially
2613 * attach them to the page if an error occurs. In the common case of
2614 * no error, they will just be freed again without ever being attached
2615 * to the page (which is all OK, because we're under the page lock).
2617 * Be careful: the buffer linked list is a NULL terminated one, rather
2618 * than the circular one we're used to.
2620 head = alloc_page_buffers(page, blocksize, false);
2626 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
2629 * We loop across all blocks in the page, whether or not they are
2630 * part of the affected region. This is so we can discover if the
2631 * page is fully mapped-to-disk.
2633 for (block_start = 0, block_in_page = 0, bh = head;
2634 block_start < PAGE_SIZE;
2635 block_in_page++, block_start += blocksize, bh = bh->b_this_page) {
2638 block_end = block_start + blocksize;
2641 if (block_start >= to)
2643 ret = get_block(inode, block_in_file + block_in_page,
2647 if (!buffer_mapped(bh))
2648 is_mapped_to_disk = 0;
2650 clean_bdev_bh_alias(bh);
2651 if (PageUptodate(page)) {
2652 set_buffer_uptodate(bh);
2655 if (buffer_new(bh) || !buffer_mapped(bh)) {
2656 zero_user_segments(page, block_start, from,
2660 if (buffer_uptodate(bh))
2661 continue; /* reiserfs does this */
2662 if (block_start < from || block_end > to) {
2664 bh->b_end_io = end_buffer_read_nobh;
2665 submit_bh(REQ_OP_READ, 0, bh);
2672 * The page is locked, so these buffers are protected from
2673 * any VM or truncate activity. Hence we don't need to care
2674 * for the buffer_head refcounts.
2676 for (bh = head; bh; bh = bh->b_this_page) {
2678 if (!buffer_uptodate(bh))
2685 if (is_mapped_to_disk)
2686 SetPageMappedToDisk(page);
2688 *fsdata = head; /* to be released by nobh_write_end */
2695 * Error recovery is a bit difficult. We need to zero out blocks that
2696 * were newly allocated, and dirty them to ensure they get written out.
2697 * Buffers need to be attached to the page at this point, otherwise
2698 * the handling of potential IO errors during writeout would be hard
2699 * (could try doing synchronous writeout, but what if that fails too?)
2701 attach_nobh_buffers(page, head);
2702 page_zero_new_buffers(page, from, to);
2711 EXPORT_SYMBOL(nobh_write_begin);
2713 int nobh_write_end(struct file *file, struct address_space *mapping,
2714 loff_t pos, unsigned len, unsigned copied,
2715 struct page *page, void *fsdata)
2717 struct inode *inode = page->mapping->host;
2718 struct buffer_head *head = fsdata;
2719 struct buffer_head *bh;
2720 BUG_ON(fsdata != NULL && page_has_buffers(page));
2722 if (unlikely(copied < len) && head)
2723 attach_nobh_buffers(page, head);
2724 if (page_has_buffers(page))
2725 return generic_write_end(file, mapping, pos, len,
2726 copied, page, fsdata);
2728 SetPageUptodate(page);
2729 set_page_dirty(page);
2730 if (pos+copied > inode->i_size) {
2731 i_size_write(inode, pos+copied);
2732 mark_inode_dirty(inode);
2740 head = head->b_this_page;
2741 free_buffer_head(bh);
2746 EXPORT_SYMBOL(nobh_write_end);
2749 * nobh_writepage() - based on block_full_write_page() except
2750 * that it tries to operate without attaching bufferheads to
2753 int nobh_writepage(struct page *page, get_block_t *get_block,
2754 struct writeback_control *wbc)
2756 struct inode * const inode = page->mapping->host;
2757 loff_t i_size = i_size_read(inode);
2758 const pgoff_t end_index = i_size >> PAGE_SHIFT;
2762 /* Is the page fully inside i_size? */
2763 if (page->index < end_index)
2766 /* Is the page fully outside i_size? (truncate in progress) */
2767 offset = i_size & (PAGE_SIZE-1);
2768 if (page->index >= end_index+1 || !offset) {
2770 * The page may have dirty, unmapped buffers. For example,
2771 * they may have been added in ext3_writepage(). Make them
2772 * freeable here, so the page does not leak.
2775 /* Not really sure about this - do we need this ? */
2776 if (page->mapping->a_ops->invalidatepage)
2777 page->mapping->a_ops->invalidatepage(page, offset);
2780 return 0; /* don't care */
2784 * The page straddles i_size. It must be zeroed out on each and every
2785 * writepage invocation because it may be mmapped. "A file is mapped
2786 * in multiples of the page size. For a file that is not a multiple of
2787 * the page size, the remaining memory is zeroed when mapped, and
2788 * writes to that region are not written out to the file."
2790 zero_user_segment(page, offset, PAGE_SIZE);
2792 ret = mpage_writepage(page, get_block, wbc);
2794 ret = __block_write_full_page(inode, page, get_block, wbc,
2795 end_buffer_async_write);
2798 EXPORT_SYMBOL(nobh_writepage);
2800 int nobh_truncate_page(struct address_space *mapping,
2801 loff_t from, get_block_t *get_block)
2803 pgoff_t index = from >> PAGE_SHIFT;
2804 unsigned offset = from & (PAGE_SIZE-1);
2807 unsigned length, pos;
2808 struct inode *inode = mapping->host;
2810 struct buffer_head map_bh;
2813 blocksize = i_blocksize(inode);
2814 length = offset & (blocksize - 1);
2816 /* Block boundary? Nothing to do */
2820 length = blocksize - length;
2821 iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
2823 page = grab_cache_page(mapping, index);
2828 if (page_has_buffers(page)) {
2832 return block_truncate_page(mapping, from, get_block);
2835 /* Find the buffer that contains "offset" */
2837 while (offset >= pos) {
2842 map_bh.b_size = blocksize;
2844 err = get_block(inode, iblock, &map_bh, 0);
2847 /* unmapped? It's a hole - nothing to do */
2848 if (!buffer_mapped(&map_bh))
2851 /* Ok, it's mapped. Make sure it's up-to-date */
2852 if (!PageUptodate(page)) {
2853 err = mapping->a_ops->readpage(NULL, page);
2859 if (!PageUptodate(page)) {
2863 if (page_has_buffers(page))
2866 zero_user(page, offset, length);
2867 set_page_dirty(page);
2876 EXPORT_SYMBOL(nobh_truncate_page);
2878 int block_truncate_page(struct address_space *mapping,
2879 loff_t from, get_block_t *get_block)
2881 pgoff_t index = from >> PAGE_SHIFT;
2882 unsigned offset = from & (PAGE_SIZE-1);
2885 unsigned length, pos;
2886 struct inode *inode = mapping->host;
2888 struct buffer_head *bh;
2891 blocksize = i_blocksize(inode);
2892 length = offset & (blocksize - 1);
2894 /* Block boundary? Nothing to do */
2898 length = blocksize - length;
2899 iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
2901 page = grab_cache_page(mapping, index);
2906 if (!page_has_buffers(page))
2907 create_empty_buffers(page, blocksize, 0);
2909 /* Find the buffer that contains "offset" */
2910 bh = page_buffers(page);
2912 while (offset >= pos) {
2913 bh = bh->b_this_page;
2919 if (!buffer_mapped(bh)) {
2920 WARN_ON(bh->b_size != blocksize);
2921 err = get_block(inode, iblock, bh, 0);
2924 /* unmapped? It's a hole - nothing to do */
2925 if (!buffer_mapped(bh))
2929 /* Ok, it's mapped. Make sure it's up-to-date */
2930 if (PageUptodate(page))
2931 set_buffer_uptodate(bh);
2933 if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
2935 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
2937 /* Uhhuh. Read error. Complain and punt. */
2938 if (!buffer_uptodate(bh))
2942 zero_user(page, offset, length);
2943 mark_buffer_dirty(bh);
2952 EXPORT_SYMBOL(block_truncate_page);
2955 * The generic ->writepage function for buffer-backed address_spaces
2957 int block_write_full_page(struct page *page, get_block_t *get_block,
2958 struct writeback_control *wbc)
2960 struct inode * const inode = page->mapping->host;
2961 loff_t i_size = i_size_read(inode);
2962 const pgoff_t end_index = i_size >> PAGE_SHIFT;
2965 /* Is the page fully inside i_size? */
2966 if (page->index < end_index)
2967 return __block_write_full_page(inode, page, get_block, wbc,
2968 end_buffer_async_write);
2970 /* Is the page fully outside i_size? (truncate in progress) */
2971 offset = i_size & (PAGE_SIZE-1);
2972 if (page->index >= end_index+1 || !offset) {
2974 * The page may have dirty, unmapped buffers. For example,
2975 * they may have been added in ext3_writepage(). Make them
2976 * freeable here, so the page does not leak.
2978 do_invalidatepage(page, 0, PAGE_SIZE);
2980 return 0; /* don't care */
2984 * The page straddles i_size. It must be zeroed out on each and every
2985 * writepage invocation because it may be mmapped. "A file is mapped
2986 * in multiples of the page size. For a file that is not a multiple of
2987 * the page size, the remaining memory is zeroed when mapped, and
2988 * writes to that region are not written out to the file."
2990 zero_user_segment(page, offset, PAGE_SIZE);
2991 return __block_write_full_page(inode, page, get_block, wbc,
2992 end_buffer_async_write);
2994 EXPORT_SYMBOL(block_write_full_page);
2996 sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
2997 get_block_t *get_block)
2999 struct inode *inode = mapping->host;
3000 struct buffer_head tmp = {
3001 .b_size = i_blocksize(inode),
3004 get_block(inode, block, &tmp, 0);
3005 return tmp.b_blocknr;
3007 EXPORT_SYMBOL(generic_block_bmap);
3009 static void end_bio_bh_io_sync(struct bio *bio)
3011 struct buffer_head *bh = bio->bi_private;
3013 if (unlikely(bio_flagged(bio, BIO_QUIET)))
3014 set_bit(BH_Quiet, &bh->b_state);
3016 bh->b_end_io(bh, !bio->bi_status);
3021 * This allows us to do IO even on the odd last sectors
3022 * of a device, even if the block size is some multiple
3023 * of the physical sector size.
3025 * We'll just truncate the bio to the size of the device,
3026 * and clear the end of the buffer head manually.
3028 * Truly out-of-range accesses will turn into actual IO
3029 * errors, this only handles the "we need to be able to
3030 * do IO at the final sector" case.
3032 void guard_bio_eod(int op, struct bio *bio)
3035 struct bio_vec *bvec = bio_last_bvec_all(bio);
3036 unsigned truncated_bytes;
3037 struct hd_struct *part;
3040 part = __disk_get_part(bio->bi_disk, bio->bi_partno);
3042 maxsector = part_nr_sects_read(part);
3044 maxsector = get_capacity(bio->bi_disk);
3051 * If the *whole* IO is past the end of the device,
3052 * let it through, and the IO layer will turn it into
3055 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
3058 maxsector -= bio->bi_iter.bi_sector;
3059 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
3062 /* Uhhuh. We've got a bio that straddles the device size! */
3063 truncated_bytes = bio->bi_iter.bi_size - (maxsector << 9);
3066 * The bio contains more than one segment which spans EOD, just return
3067 * and let IO layer turn it into an EIO
3069 if (truncated_bytes > bvec->bv_len)
3072 /* Truncate the bio.. */
3073 bio->bi_iter.bi_size -= truncated_bytes;
3074 bvec->bv_len -= truncated_bytes;
3076 /* ..and clear the end of the buffer for reads */
3077 if (op == REQ_OP_READ) {
3080 mp_bvec_last_segment(bvec, &bv);
3081 zero_user(bv.bv_page, bv.bv_offset + bv.bv_len,
3086 static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
3087 enum rw_hint write_hint, struct writeback_control *wbc)
3091 BUG_ON(!buffer_locked(bh));
3092 BUG_ON(!buffer_mapped(bh));
3093 BUG_ON(!bh->b_end_io);
3094 BUG_ON(buffer_delay(bh));
3095 BUG_ON(buffer_unwritten(bh));
3098 * Only clear out a write error when rewriting
3100 if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE))
3101 clear_buffer_write_io_error(bh);
3104 * from here on down, it's all bio -- do the initial mapping,
3105 * submit_bio -> generic_make_request may further map this bio around
3107 bio = bio_alloc(GFP_NOIO, 1);
3109 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
3110 bio_set_dev(bio, bh->b_bdev);
3111 bio->bi_write_hint = write_hint;
3113 bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
3114 BUG_ON(bio->bi_iter.bi_size != bh->b_size);
3116 bio->bi_end_io = end_bio_bh_io_sync;
3117 bio->bi_private = bh;
3119 /* Take care of bh's that straddle the end of the device */
3120 guard_bio_eod(op, bio);
3122 if (buffer_meta(bh))
3123 op_flags |= REQ_META;
3124 if (buffer_prio(bh))
3125 op_flags |= REQ_PRIO;
3126 bio_set_op_attrs(bio, op, op_flags);
3129 wbc_init_bio(wbc, bio);
3130 wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size);
3137 int submit_bh(int op, int op_flags, struct buffer_head *bh)
3139 return submit_bh_wbc(op, op_flags, bh, 0, NULL);
3141 EXPORT_SYMBOL(submit_bh);
3144 * ll_rw_block: low-level access to block devices (DEPRECATED)
3145 * @op: whether to %READ or %WRITE
3146 * @op_flags: req_flag_bits
3147 * @nr: number of &struct buffer_heads in the array
3148 * @bhs: array of pointers to &struct buffer_head
3150 * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
3151 * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE.
3152 * @op_flags contains flags modifying the detailed I/O behavior, most notably
3155 * This function drops any buffer that it cannot get a lock on (with the
3156 * BH_Lock state bit), any buffer that appears to be clean when doing a write
3157 * request, and any buffer that appears to be up-to-date when doing read
3158 * request. Further it marks as clean buffers that are processed for
3159 * writing (the buffer cache won't assume that they are actually clean
3160 * until the buffer gets unlocked).
3162 * ll_rw_block sets b_end_io to simple completion handler that marks
3163 * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
3166 * All of the buffers must be for the same device, and must also be a
3167 * multiple of the current approved size for the device.
3169 void ll_rw_block(int op, int op_flags, int nr, struct buffer_head *bhs[])
3173 for (i = 0; i < nr; i++) {
3174 struct buffer_head *bh = bhs[i];
3176 if (!trylock_buffer(bh))
3179 if (test_clear_buffer_dirty(bh)) {
3180 bh->b_end_io = end_buffer_write_sync;
3182 submit_bh(op, op_flags, bh);
3186 if (!buffer_uptodate(bh)) {
3187 bh->b_end_io = end_buffer_read_sync;
3189 submit_bh(op, op_flags, bh);
3196 EXPORT_SYMBOL(ll_rw_block);
3198 void write_dirty_buffer(struct buffer_head *bh, int op_flags)
3201 if (!test_clear_buffer_dirty(bh)) {
3205 bh->b_end_io = end_buffer_write_sync;
3207 submit_bh(REQ_OP_WRITE, op_flags, bh);
3209 EXPORT_SYMBOL(write_dirty_buffer);
3212 * For a data-integrity writeout, we need to wait upon any in-progress I/O
3213 * and then start new I/O and then wait upon it. The caller must have a ref on
3216 int __sync_dirty_buffer(struct buffer_head *bh, int op_flags)
3220 WARN_ON(atomic_read(&bh->b_count) < 1);
3222 if (test_clear_buffer_dirty(bh)) {
3224 bh->b_end_io = end_buffer_write_sync;
3225 ret = submit_bh(REQ_OP_WRITE, op_flags, bh);
3227 if (!ret && !buffer_uptodate(bh))
3234 EXPORT_SYMBOL(__sync_dirty_buffer);
3236 int sync_dirty_buffer(struct buffer_head *bh)
3238 return __sync_dirty_buffer(bh, REQ_SYNC);
3240 EXPORT_SYMBOL(sync_dirty_buffer);
3243 * try_to_free_buffers() checks if all the buffers on this particular page
3244 * are unused, and releases them if so.
3246 * Exclusion against try_to_free_buffers may be obtained by either
3247 * locking the page or by holding its mapping's private_lock.
3249 * If the page is dirty but all the buffers are clean then we need to
3250 * be sure to mark the page clean as well. This is because the page
3251 * may be against a block device, and a later reattachment of buffers
3252 * to a dirty page will set *all* buffers dirty. Which would corrupt
3253 * filesystem data on the same device.
3255 * The same applies to regular filesystem pages: if all the buffers are
3256 * clean then we set the page clean and proceed. To do that, we require
3257 * total exclusion from __set_page_dirty_buffers(). That is obtained with
3260 * try_to_free_buffers() is non-blocking.
3262 static inline int buffer_busy(struct buffer_head *bh)
3264 return atomic_read(&bh->b_count) |
3265 (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock)));
3269 drop_buffers(struct page *page, struct buffer_head **buffers_to_free)
3271 struct buffer_head *head = page_buffers(page);
3272 struct buffer_head *bh;
3276 if (buffer_busy(bh))
3278 bh = bh->b_this_page;
3279 } while (bh != head);
3282 struct buffer_head *next = bh->b_this_page;
3284 if (bh->b_assoc_map)
3285 __remove_assoc_queue(bh);
3287 } while (bh != head);
3288 *buffers_to_free = head;
3289 __clear_page_buffers(page);
3295 int try_to_free_buffers(struct page *page)
3297 struct address_space * const mapping = page->mapping;
3298 struct buffer_head *buffers_to_free = NULL;
3301 BUG_ON(!PageLocked(page));
3302 if (PageWriteback(page))
3305 if (mapping == NULL) { /* can this still happen? */
3306 ret = drop_buffers(page, &buffers_to_free);
3310 spin_lock(&mapping->private_lock);
3311 ret = drop_buffers(page, &buffers_to_free);
3314 * If the filesystem writes its buffers by hand (eg ext3)
3315 * then we can have clean buffers against a dirty page. We
3316 * clean the page here; otherwise the VM will never notice
3317 * that the filesystem did any IO at all.
3319 * Also, during truncate, discard_buffer will have marked all
3320 * the page's buffers clean. We discover that here and clean
3323 * private_lock must be held over this entire operation in order
3324 * to synchronise against __set_page_dirty_buffers and prevent the
3325 * dirty bit from being lost.
3328 cancel_dirty_page(page);
3329 spin_unlock(&mapping->private_lock);
3331 if (buffers_to_free) {
3332 struct buffer_head *bh = buffers_to_free;
3335 struct buffer_head *next = bh->b_this_page;
3336 free_buffer_head(bh);
3338 } while (bh != buffers_to_free);
3342 EXPORT_SYMBOL(try_to_free_buffers);
3345 * There are no bdflush tunables left. But distributions are
3346 * still running obsolete flush daemons, so we terminate them here.
3348 * Use of bdflush() is deprecated and will be removed in a future kernel.
3349 * The `flush-X' kernel threads fully replace bdflush daemons and this call.
3351 SYSCALL_DEFINE2(bdflush, int, func, long, data)
3353 static int msg_count;
3355 if (!capable(CAP_SYS_ADMIN))
3358 if (msg_count < 5) {
3361 "warning: process `%s' used the obsolete bdflush"
3362 " system call\n", current->comm);
3363 printk(KERN_INFO "Fix your initscripts?\n");
3372 * Buffer-head allocation
3374 static struct kmem_cache *bh_cachep __read_mostly;
3377 * Once the number of bh's in the machine exceeds this level, we start
3378 * stripping them in writeback.
3380 static unsigned long max_buffer_heads;
3382 int buffer_heads_over_limit;
3384 struct bh_accounting {
3385 int nr; /* Number of live bh's */
3386 int ratelimit; /* Limit cacheline bouncing */
3389 static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0};
3391 static void recalc_bh_state(void)
3396 if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096)
3398 __this_cpu_write(bh_accounting.ratelimit, 0);
3399 for_each_online_cpu(i)
3400 tot += per_cpu(bh_accounting, i).nr;
3401 buffer_heads_over_limit = (tot > max_buffer_heads);
3404 struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
3406 struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags);
3408 INIT_LIST_HEAD(&ret->b_assoc_buffers);
3410 __this_cpu_inc(bh_accounting.nr);
3416 EXPORT_SYMBOL(alloc_buffer_head);
3418 void free_buffer_head(struct buffer_head *bh)
3420 BUG_ON(!list_empty(&bh->b_assoc_buffers));
3421 kmem_cache_free(bh_cachep, bh);
3423 __this_cpu_dec(bh_accounting.nr);
3427 EXPORT_SYMBOL(free_buffer_head);
3429 static int buffer_exit_cpu_dead(unsigned int cpu)
3432 struct bh_lru *b = &per_cpu(bh_lrus, cpu);
3434 for (i = 0; i < BH_LRU_SIZE; i++) {
3438 this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr);
3439 per_cpu(bh_accounting, cpu).nr = 0;
3444 * bh_uptodate_or_lock - Test whether the buffer is uptodate
3445 * @bh: struct buffer_head
3447 * Return true if the buffer is up-to-date and false,
3448 * with the buffer locked, if not.
3450 int bh_uptodate_or_lock(struct buffer_head *bh)
3452 if (!buffer_uptodate(bh)) {
3454 if (!buffer_uptodate(bh))
3460 EXPORT_SYMBOL(bh_uptodate_or_lock);
3463 * bh_submit_read - Submit a locked buffer for reading
3464 * @bh: struct buffer_head
3466 * Returns zero on success and -EIO on error.
3468 int bh_submit_read(struct buffer_head *bh)
3470 BUG_ON(!buffer_locked(bh));
3472 if (buffer_uptodate(bh)) {
3478 bh->b_end_io = end_buffer_read_sync;
3479 submit_bh(REQ_OP_READ, 0, bh);
3481 if (buffer_uptodate(bh))
3485 EXPORT_SYMBOL(bh_submit_read);
3487 void __init buffer_init(void)
3489 unsigned long nrpages;
3492 bh_cachep = kmem_cache_create("buffer_head",
3493 sizeof(struct buffer_head), 0,
3494 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
3499 * Limit the bh occupancy to 10% of ZONE_NORMAL
3501 nrpages = (nr_free_buffer_pages() * 10) / 100;
3502 max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head));
3503 ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead",
3504 NULL, buffer_exit_cpu_dead);