4 * Copyright (C) 1991, 1992, 2002 Linus Torvalds
8 * Start bdflush() with kernel_thread not syscall - Paul Gortmaker, 12/95
10 * Removed a lot of unnecessary code and simplified things now that
11 * the buffer cache isn't our primary cache - Andrew Tridgell 12/96
13 * Speed up hash, lru, and free list operations. Use gfp() for allocating
14 * hash table, use SLAB cache for buffer heads. SMP threading. -DaveM
16 * Added 32k buffer block sizes - these are required older ARM systems. - RMK
18 * async buffer flushing, 1999 Andrea Arcangeli <andrea@suse.de>
21 #include <linux/kernel.h>
22 #include <linux/sched/signal.h>
23 #include <linux/syscalls.h>
25 #include <linux/iomap.h>
27 #include <linux/percpu.h>
28 #include <linux/slab.h>
29 #include <linux/capability.h>
30 #include <linux/blkdev.h>
31 #include <linux/file.h>
32 #include <linux/quotaops.h>
33 #include <linux/highmem.h>
34 #include <linux/export.h>
35 #include <linux/backing-dev.h>
36 #include <linux/writeback.h>
37 #include <linux/hash.h>
38 #include <linux/suspend.h>
39 #include <linux/buffer_head.h>
40 #include <linux/task_io_accounting_ops.h>
41 #include <linux/bio.h>
42 #include <linux/notifier.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 <trace/events/block.h>
50 static int fsync_buffers_list(spinlock_t
*lock
, struct list_head
*list
);
51 static int submit_bh_wbc(int op
, int op_flags
, struct buffer_head
*bh
,
52 enum rw_hint hint
, struct writeback_control
*wbc
);
54 #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
56 void init_buffer(struct buffer_head
*bh
, bh_end_io_t
*handler
, void *private)
58 bh
->b_end_io
= handler
;
59 bh
->b_private
= private;
61 EXPORT_SYMBOL(init_buffer
);
63 inline void touch_buffer(struct buffer_head
*bh
)
65 trace_block_touch_buffer(bh
);
66 mark_page_accessed(bh
->b_page
);
68 EXPORT_SYMBOL(touch_buffer
);
70 void __lock_buffer(struct buffer_head
*bh
)
72 wait_on_bit_lock_io(&bh
->b_state
, BH_Lock
, TASK_UNINTERRUPTIBLE
);
74 EXPORT_SYMBOL(__lock_buffer
);
76 void unlock_buffer(struct buffer_head
*bh
)
78 clear_bit_unlock(BH_Lock
, &bh
->b_state
);
79 smp_mb__after_atomic();
80 wake_up_bit(&bh
->b_state
, BH_Lock
);
82 EXPORT_SYMBOL(unlock_buffer
);
85 * Returns if the page has dirty or writeback buffers. If all the buffers
86 * are unlocked and clean then the PageDirty information is stale. If
87 * any of the pages are locked, it is assumed they are locked for IO.
89 void buffer_check_dirty_writeback(struct page
*page
,
90 bool *dirty
, bool *writeback
)
92 struct buffer_head
*head
, *bh
;
96 BUG_ON(!PageLocked(page
));
98 if (!page_has_buffers(page
))
101 if (PageWriteback(page
))
104 head
= page_buffers(page
);
107 if (buffer_locked(bh
))
110 if (buffer_dirty(bh
))
113 bh
= bh
->b_this_page
;
114 } while (bh
!= head
);
116 EXPORT_SYMBOL(buffer_check_dirty_writeback
);
119 * Block until a buffer comes unlocked. This doesn't stop it
120 * from becoming locked again - you have to lock it yourself
121 * if you want to preserve its state.
123 void __wait_on_buffer(struct buffer_head
* bh
)
125 wait_on_bit_io(&bh
->b_state
, BH_Lock
, TASK_UNINTERRUPTIBLE
);
127 EXPORT_SYMBOL(__wait_on_buffer
);
130 __clear_page_buffers(struct page
*page
)
132 ClearPagePrivate(page
);
133 set_page_private(page
, 0);
137 static void buffer_io_error(struct buffer_head
*bh
, char *msg
)
139 if (!test_bit(BH_Quiet
, &bh
->b_state
))
140 printk_ratelimited(KERN_ERR
141 "Buffer I/O error on dev %pg, logical block %llu%s\n",
142 bh
->b_bdev
, (unsigned long long)bh
->b_blocknr
, msg
);
146 * End-of-IO handler helper function which does not touch the bh after
148 * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
149 * a race there is benign: unlock_buffer() only use the bh's address for
150 * hashing after unlocking the buffer, so it doesn't actually touch the bh
153 static void __end_buffer_read_notouch(struct buffer_head
*bh
, int uptodate
)
156 set_buffer_uptodate(bh
);
158 /* This happens, due to failed read-ahead attempts. */
159 clear_buffer_uptodate(bh
);
165 * Default synchronous end-of-IO handler.. Just mark it up-to-date and
166 * unlock the buffer. This is what ll_rw_block uses too.
168 void end_buffer_read_sync(struct buffer_head
*bh
, int uptodate
)
170 __end_buffer_read_notouch(bh
, uptodate
);
173 EXPORT_SYMBOL(end_buffer_read_sync
);
175 void end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
178 set_buffer_uptodate(bh
);
180 buffer_io_error(bh
, ", lost sync page write");
181 mark_buffer_write_io_error(bh
);
182 clear_buffer_uptodate(bh
);
187 EXPORT_SYMBOL(end_buffer_write_sync
);
190 * Various filesystems appear to want __find_get_block to be non-blocking.
191 * But it's the page lock which protects the buffers. To get around this,
192 * we get exclusion from try_to_free_buffers with the blockdev mapping's
195 * Hack idea: for the blockdev mapping, i_bufferlist_lock contention
196 * may be quite high. This code could TryLock the page, and if that
197 * succeeds, there is no need to take private_lock. (But if
198 * private_lock is contended then so is mapping->tree_lock).
200 static struct buffer_head
*
201 __find_get_block_slow(struct block_device
*bdev
, sector_t block
)
203 struct inode
*bd_inode
= bdev
->bd_inode
;
204 struct address_space
*bd_mapping
= bd_inode
->i_mapping
;
205 struct buffer_head
*ret
= NULL
;
207 struct buffer_head
*bh
;
208 struct buffer_head
*head
;
211 static DEFINE_RATELIMIT_STATE(last_warned
, HZ
, 1);
213 index
= block
>> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
214 page
= find_get_page_flags(bd_mapping
, index
, FGP_ACCESSED
);
218 spin_lock(&bd_mapping
->private_lock
);
219 if (!page_has_buffers(page
))
221 head
= page_buffers(page
);
224 if (!buffer_mapped(bh
))
226 else if (bh
->b_blocknr
== block
) {
231 bh
= bh
->b_this_page
;
232 } while (bh
!= head
);
234 /* we might be here because some of the buffers on this page are
235 * not mapped. This is due to various races between
236 * file io on the block device and getblk. It gets dealt with
237 * elsewhere, don't buffer_error if we had some unmapped buffers
239 ratelimit_set_flags(&last_warned
, RATELIMIT_MSG_ON_RELEASE
);
240 if (all_mapped
&& __ratelimit(&last_warned
)) {
241 printk("__find_get_block_slow() failed. block=%llu, "
242 "b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, "
243 "device %pg blocksize: %d\n",
244 (unsigned long long)block
,
245 (unsigned long long)bh
->b_blocknr
,
246 bh
->b_state
, bh
->b_size
, bdev
,
247 1 << bd_inode
->i_blkbits
);
250 spin_unlock(&bd_mapping
->private_lock
);
257 * I/O completion handler for block_read_full_page() - pages
258 * which come unlocked at the end of I/O.
260 static void end_buffer_async_read(struct buffer_head
*bh
, int uptodate
)
263 struct buffer_head
*first
;
264 struct buffer_head
*tmp
;
266 int page_uptodate
= 1;
268 BUG_ON(!buffer_async_read(bh
));
272 set_buffer_uptodate(bh
);
274 clear_buffer_uptodate(bh
);
275 buffer_io_error(bh
, ", async page read");
280 * Be _very_ careful from here on. Bad things can happen if
281 * two buffer heads end IO at almost the same time and both
282 * decide that the page is now completely done.
284 first
= page_buffers(page
);
285 local_irq_save(flags
);
286 bit_spin_lock(BH_Uptodate_Lock
, &first
->b_state
);
287 clear_buffer_async_read(bh
);
291 if (!buffer_uptodate(tmp
))
293 if (buffer_async_read(tmp
)) {
294 BUG_ON(!buffer_locked(tmp
));
297 tmp
= tmp
->b_this_page
;
299 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
300 local_irq_restore(flags
);
303 * If none of the buffers had errors and they are all
304 * uptodate then we can set the page uptodate.
306 if (page_uptodate
&& !PageError(page
))
307 SetPageUptodate(page
);
312 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
313 local_irq_restore(flags
);
318 * Completion handler for block_write_full_page() - pages which are unlocked
319 * during I/O, and which have PageWriteback cleared upon I/O completion.
321 void end_buffer_async_write(struct buffer_head
*bh
, int uptodate
)
324 struct buffer_head
*first
;
325 struct buffer_head
*tmp
;
328 BUG_ON(!buffer_async_write(bh
));
332 set_buffer_uptodate(bh
);
334 buffer_io_error(bh
, ", lost async page write");
335 mark_buffer_write_io_error(bh
);
336 clear_buffer_uptodate(bh
);
340 first
= page_buffers(page
);
341 local_irq_save(flags
);
342 bit_spin_lock(BH_Uptodate_Lock
, &first
->b_state
);
344 clear_buffer_async_write(bh
);
346 tmp
= bh
->b_this_page
;
348 if (buffer_async_write(tmp
)) {
349 BUG_ON(!buffer_locked(tmp
));
352 tmp
= tmp
->b_this_page
;
354 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
355 local_irq_restore(flags
);
356 end_page_writeback(page
);
360 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
361 local_irq_restore(flags
);
364 EXPORT_SYMBOL(end_buffer_async_write
);
367 * If a page's buffers are under async readin (end_buffer_async_read
368 * completion) then there is a possibility that another thread of
369 * control could lock one of the buffers after it has completed
370 * but while some of the other buffers have not completed. This
371 * locked buffer would confuse end_buffer_async_read() into not unlocking
372 * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
373 * that this buffer is not under async I/O.
375 * The page comes unlocked when it has no locked buffer_async buffers
378 * PageLocked prevents anyone starting new async I/O reads any of
381 * PageWriteback is used to prevent simultaneous writeout of the same
384 * PageLocked prevents anyone from starting writeback of a page which is
385 * under read I/O (PageWriteback is only ever set against a locked page).
387 static void mark_buffer_async_read(struct buffer_head
*bh
)
389 bh
->b_end_io
= end_buffer_async_read
;
390 set_buffer_async_read(bh
);
393 static void mark_buffer_async_write_endio(struct buffer_head
*bh
,
394 bh_end_io_t
*handler
)
396 bh
->b_end_io
= handler
;
397 set_buffer_async_write(bh
);
400 void mark_buffer_async_write(struct buffer_head
*bh
)
402 mark_buffer_async_write_endio(bh
, end_buffer_async_write
);
404 EXPORT_SYMBOL(mark_buffer_async_write
);
408 * fs/buffer.c contains helper functions for buffer-backed address space's
409 * fsync functions. A common requirement for buffer-based filesystems is
410 * that certain data from the backing blockdev needs to be written out for
411 * a successful fsync(). For example, ext2 indirect blocks need to be
412 * written back and waited upon before fsync() returns.
414 * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
415 * inode_has_buffers() and invalidate_inode_buffers() are provided for the
416 * management of a list of dependent buffers at ->i_mapping->private_list.
418 * Locking is a little subtle: try_to_free_buffers() will remove buffers
419 * from their controlling inode's queue when they are being freed. But
420 * try_to_free_buffers() will be operating against the *blockdev* mapping
421 * at the time, not against the S_ISREG file which depends on those buffers.
422 * So the locking for private_list is via the private_lock in the address_space
423 * which backs the buffers. Which is different from the address_space
424 * against which the buffers are listed. So for a particular address_space,
425 * mapping->private_lock does *not* protect mapping->private_list! In fact,
426 * mapping->private_list will always be protected by the backing blockdev's
429 * Which introduces a requirement: all buffers on an address_space's
430 * ->private_list must be from the same address_space: the blockdev's.
432 * address_spaces which do not place buffers at ->private_list via these
433 * utility functions are free to use private_lock and private_list for
434 * whatever they want. The only requirement is that list_empty(private_list)
435 * be true at clear_inode() time.
437 * FIXME: clear_inode should not call invalidate_inode_buffers(). The
438 * filesystems should do that. invalidate_inode_buffers() should just go
439 * BUG_ON(!list_empty).
441 * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
442 * take an address_space, not an inode. And it should be called
443 * mark_buffer_dirty_fsync() to clearly define why those buffers are being
446 * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
447 * list if it is already on a list. Because if the buffer is on a list,
448 * it *must* already be on the right one. If not, the filesystem is being
449 * silly. This will save a ton of locking. But first we have to ensure
450 * that buffers are taken *off* the old inode's list when they are freed
451 * (presumably in truncate). That requires careful auditing of all
452 * filesystems (do it inside bforget()). It could also be done by bringing
457 * The buffer's backing address_space's private_lock must be held
459 static void __remove_assoc_queue(struct buffer_head
*bh
)
461 list_del_init(&bh
->b_assoc_buffers
);
462 WARN_ON(!bh
->b_assoc_map
);
463 bh
->b_assoc_map
= NULL
;
466 int inode_has_buffers(struct inode
*inode
)
468 return !list_empty(&inode
->i_data
.private_list
);
472 * osync is designed to support O_SYNC io. It waits synchronously for
473 * all already-submitted IO to complete, but does not queue any new
474 * writes to the disk.
476 * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
477 * you dirty the buffers, and then use osync_inode_buffers to wait for
478 * completion. Any other dirty buffers which are not yet queued for
479 * write will not be flushed to disk by the osync.
481 static int osync_buffers_list(spinlock_t
*lock
, struct list_head
*list
)
483 struct buffer_head
*bh
;
489 list_for_each_prev(p
, list
) {
491 if (buffer_locked(bh
)) {
495 if (!buffer_uptodate(bh
))
506 static void do_thaw_one(struct super_block
*sb
, void *unused
)
508 while (sb
->s_bdev
&& !thaw_bdev(sb
->s_bdev
, sb
))
509 printk(KERN_WARNING
"Emergency Thaw on %pg\n", sb
->s_bdev
);
512 static void do_thaw_all(struct work_struct
*work
)
514 iterate_supers(do_thaw_one
, NULL
);
516 printk(KERN_WARNING
"Emergency Thaw complete\n");
520 * emergency_thaw_all -- forcibly thaw every frozen filesystem
522 * Used for emergency unfreeze of all filesystems via SysRq
524 void emergency_thaw_all(void)
526 struct work_struct
*work
;
528 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
530 INIT_WORK(work
, do_thaw_all
);
536 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
537 * @mapping: the mapping which wants those buffers written
539 * Starts I/O against the buffers at mapping->private_list, and waits upon
542 * Basically, this is a convenience function for fsync().
543 * @mapping is a file or directory which needs those buffers to be written for
544 * a successful fsync().
546 int sync_mapping_buffers(struct address_space
*mapping
)
548 struct address_space
*buffer_mapping
= mapping
->private_data
;
550 if (buffer_mapping
== NULL
|| list_empty(&mapping
->private_list
))
553 return fsync_buffers_list(&buffer_mapping
->private_lock
,
554 &mapping
->private_list
);
556 EXPORT_SYMBOL(sync_mapping_buffers
);
559 * Called when we've recently written block `bblock', and it is known that
560 * `bblock' was for a buffer_boundary() buffer. This means that the block at
561 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
562 * dirty, schedule it for IO. So that indirects merge nicely with their data.
564 void write_boundary_block(struct block_device
*bdev
,
565 sector_t bblock
, unsigned blocksize
)
567 struct buffer_head
*bh
= __find_get_block(bdev
, bblock
+ 1, blocksize
);
569 if (buffer_dirty(bh
))
570 ll_rw_block(REQ_OP_WRITE
, 0, 1, &bh
);
575 void mark_buffer_dirty_inode(struct buffer_head
*bh
, struct inode
*inode
)
577 struct address_space
*mapping
= inode
->i_mapping
;
578 struct address_space
*buffer_mapping
= bh
->b_page
->mapping
;
580 mark_buffer_dirty(bh
);
581 if (!mapping
->private_data
) {
582 mapping
->private_data
= buffer_mapping
;
584 BUG_ON(mapping
->private_data
!= buffer_mapping
);
586 if (!bh
->b_assoc_map
) {
587 spin_lock(&buffer_mapping
->private_lock
);
588 list_move_tail(&bh
->b_assoc_buffers
,
589 &mapping
->private_list
);
590 bh
->b_assoc_map
= mapping
;
591 spin_unlock(&buffer_mapping
->private_lock
);
594 EXPORT_SYMBOL(mark_buffer_dirty_inode
);
597 * Mark the page dirty, and set it dirty in the radix tree, and mark the inode
600 * If warn is true, then emit a warning if the page is not uptodate and has
601 * not been truncated.
603 * The caller must hold lock_page_memcg().
605 static void __set_page_dirty(struct page
*page
, struct address_space
*mapping
,
610 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
611 if (page
->mapping
) { /* Race with truncate? */
612 WARN_ON_ONCE(warn
&& !PageUptodate(page
));
613 account_page_dirtied(page
, mapping
);
614 radix_tree_tag_set(&mapping
->page_tree
,
615 page_index(page
), PAGECACHE_TAG_DIRTY
);
617 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
621 * Add a page to the dirty page list.
623 * It is a sad fact of life that this function is called from several places
624 * deeply under spinlocking. It may not sleep.
626 * If the page has buffers, the uptodate buffers are set dirty, to preserve
627 * dirty-state coherency between the page and the buffers. It the page does
628 * not have buffers then when they are later attached they will all be set
631 * The buffers are dirtied before the page is dirtied. There's a small race
632 * window in which a writepage caller may see the page cleanness but not the
633 * buffer dirtiness. That's fine. If this code were to set the page dirty
634 * before the buffers, a concurrent writepage caller could clear the page dirty
635 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
636 * page on the dirty page list.
638 * We use private_lock to lock against try_to_free_buffers while using the
639 * page's buffer list. Also use this to protect against clean buffers being
640 * added to the page after it was set dirty.
642 * FIXME: may need to call ->reservepage here as well. That's rather up to the
643 * address_space though.
645 int __set_page_dirty_buffers(struct page
*page
)
648 struct address_space
*mapping
= page_mapping(page
);
650 if (unlikely(!mapping
))
651 return !TestSetPageDirty(page
);
653 spin_lock(&mapping
->private_lock
);
654 if (page_has_buffers(page
)) {
655 struct buffer_head
*head
= page_buffers(page
);
656 struct buffer_head
*bh
= head
;
659 set_buffer_dirty(bh
);
660 bh
= bh
->b_this_page
;
661 } while (bh
!= head
);
664 * Lock out page->mem_cgroup migration to keep PageDirty
665 * synchronized with per-memcg dirty page counters.
667 lock_page_memcg(page
);
668 newly_dirty
= !TestSetPageDirty(page
);
669 spin_unlock(&mapping
->private_lock
);
672 __set_page_dirty(page
, mapping
, 1);
674 unlock_page_memcg(page
);
677 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
681 EXPORT_SYMBOL(__set_page_dirty_buffers
);
684 * Write out and wait upon a list of buffers.
686 * We have conflicting pressures: we want to make sure that all
687 * initially dirty buffers get waited on, but that any subsequently
688 * dirtied buffers don't. After all, we don't want fsync to last
689 * forever if somebody is actively writing to the file.
691 * Do this in two main stages: first we copy dirty buffers to a
692 * temporary inode list, queueing the writes as we go. Then we clean
693 * up, waiting for those writes to complete.
695 * During this second stage, any subsequent updates to the file may end
696 * up refiling the buffer on the original inode's dirty list again, so
697 * there is a chance we will end up with a buffer queued for write but
698 * not yet completed on that list. So, as a final cleanup we go through
699 * the osync code to catch these locked, dirty buffers without requeuing
700 * any newly dirty buffers for write.
702 static int fsync_buffers_list(spinlock_t
*lock
, struct list_head
*list
)
704 struct buffer_head
*bh
;
705 struct list_head tmp
;
706 struct address_space
*mapping
;
708 struct blk_plug plug
;
710 INIT_LIST_HEAD(&tmp
);
711 blk_start_plug(&plug
);
714 while (!list_empty(list
)) {
715 bh
= BH_ENTRY(list
->next
);
716 mapping
= bh
->b_assoc_map
;
717 __remove_assoc_queue(bh
);
718 /* Avoid race with mark_buffer_dirty_inode() which does
719 * a lockless check and we rely on seeing the dirty bit */
721 if (buffer_dirty(bh
) || buffer_locked(bh
)) {
722 list_add(&bh
->b_assoc_buffers
, &tmp
);
723 bh
->b_assoc_map
= mapping
;
724 if (buffer_dirty(bh
)) {
728 * Ensure any pending I/O completes so that
729 * write_dirty_buffer() actually writes the
730 * current contents - it is a noop if I/O is
731 * still in flight on potentially older
734 write_dirty_buffer(bh
, REQ_SYNC
);
737 * Kick off IO for the previous mapping. Note
738 * that we will not run the very last mapping,
739 * wait_on_buffer() will do that for us
740 * through sync_buffer().
749 blk_finish_plug(&plug
);
752 while (!list_empty(&tmp
)) {
753 bh
= BH_ENTRY(tmp
.prev
);
755 mapping
= bh
->b_assoc_map
;
756 __remove_assoc_queue(bh
);
757 /* Avoid race with mark_buffer_dirty_inode() which does
758 * a lockless check and we rely on seeing the dirty bit */
760 if (buffer_dirty(bh
)) {
761 list_add(&bh
->b_assoc_buffers
,
762 &mapping
->private_list
);
763 bh
->b_assoc_map
= mapping
;
767 if (!buffer_uptodate(bh
))
774 err2
= osync_buffers_list(lock
, list
);
782 * Invalidate any and all dirty buffers on a given inode. We are
783 * probably unmounting the fs, but that doesn't mean we have already
784 * done a sync(). Just drop the buffers from the inode list.
786 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
787 * assumes that all the buffers are against the blockdev. Not true
790 void invalidate_inode_buffers(struct inode
*inode
)
792 if (inode_has_buffers(inode
)) {
793 struct address_space
*mapping
= &inode
->i_data
;
794 struct list_head
*list
= &mapping
->private_list
;
795 struct address_space
*buffer_mapping
= mapping
->private_data
;
797 spin_lock(&buffer_mapping
->private_lock
);
798 while (!list_empty(list
))
799 __remove_assoc_queue(BH_ENTRY(list
->next
));
800 spin_unlock(&buffer_mapping
->private_lock
);
803 EXPORT_SYMBOL(invalidate_inode_buffers
);
806 * Remove any clean buffers from the inode's buffer list. This is called
807 * when we're trying to free the inode itself. Those buffers can pin it.
809 * Returns true if all buffers were removed.
811 int remove_inode_buffers(struct inode
*inode
)
815 if (inode_has_buffers(inode
)) {
816 struct address_space
*mapping
= &inode
->i_data
;
817 struct list_head
*list
= &mapping
->private_list
;
818 struct address_space
*buffer_mapping
= mapping
->private_data
;
820 spin_lock(&buffer_mapping
->private_lock
);
821 while (!list_empty(list
)) {
822 struct buffer_head
*bh
= BH_ENTRY(list
->next
);
823 if (buffer_dirty(bh
)) {
827 __remove_assoc_queue(bh
);
829 spin_unlock(&buffer_mapping
->private_lock
);
835 * Create the appropriate buffers when given a page for data area and
836 * the size of each buffer.. Use the bh->b_this_page linked list to
837 * follow the buffers created. Return NULL if unable to create more
840 * The retry flag is used to differentiate async IO (paging, swapping)
841 * which may not fail from ordinary buffer allocations.
843 struct buffer_head
*alloc_page_buffers(struct page
*page
, unsigned long size
,
846 struct buffer_head
*bh
, *head
;
847 gfp_t gfp
= GFP_NOFS
;
855 while ((offset
-= size
) >= 0) {
856 bh
= alloc_buffer_head(gfp
);
860 bh
->b_this_page
= head
;
866 /* Link the buffer to its page */
867 set_bh_page(bh
, page
, offset
);
871 * In case anything failed, we just free everything we got.
877 head
= head
->b_this_page
;
878 free_buffer_head(bh
);
884 EXPORT_SYMBOL_GPL(alloc_page_buffers
);
887 link_dev_buffers(struct page
*page
, struct buffer_head
*head
)
889 struct buffer_head
*bh
, *tail
;
894 bh
= bh
->b_this_page
;
896 tail
->b_this_page
= head
;
897 attach_page_buffers(page
, head
);
900 static sector_t
blkdev_max_block(struct block_device
*bdev
, unsigned int size
)
902 sector_t retval
= ~((sector_t
)0);
903 loff_t sz
= i_size_read(bdev
->bd_inode
);
906 unsigned int sizebits
= blksize_bits(size
);
907 retval
= (sz
>> sizebits
);
913 * Initialise the state of a blockdev page's buffers.
916 init_page_buffers(struct page
*page
, struct block_device
*bdev
,
917 sector_t block
, int size
)
919 struct buffer_head
*head
= page_buffers(page
);
920 struct buffer_head
*bh
= head
;
921 int uptodate
= PageUptodate(page
);
922 sector_t end_block
= blkdev_max_block(I_BDEV(bdev
->bd_inode
), size
);
925 if (!buffer_mapped(bh
)) {
926 init_buffer(bh
, NULL
, NULL
);
928 bh
->b_blocknr
= block
;
930 set_buffer_uptodate(bh
);
931 if (block
< end_block
)
932 set_buffer_mapped(bh
);
935 bh
= bh
->b_this_page
;
936 } while (bh
!= head
);
939 * Caller needs to validate requested block against end of device.
945 * Create the page-cache page that contains the requested block.
947 * This is used purely for blockdev mappings.
950 grow_dev_page(struct block_device
*bdev
, sector_t block
,
951 pgoff_t index
, int size
, int sizebits
, gfp_t gfp
)
953 struct inode
*inode
= bdev
->bd_inode
;
955 struct buffer_head
*bh
;
957 int ret
= 0; /* Will call free_more_memory() */
960 gfp_mask
= mapping_gfp_constraint(inode
->i_mapping
, ~__GFP_FS
) | gfp
;
963 * XXX: __getblk_slow() can not really deal with failure and
964 * will endlessly loop on improvised global reclaim. Prefer
965 * looping in the allocator rather than here, at least that
966 * code knows what it's doing.
968 gfp_mask
|= __GFP_NOFAIL
;
970 page
= find_or_create_page(inode
->i_mapping
, index
, gfp_mask
);
972 BUG_ON(!PageLocked(page
));
974 if (page_has_buffers(page
)) {
975 bh
= page_buffers(page
);
976 if (bh
->b_size
== size
) {
977 end_block
= init_page_buffers(page
, bdev
,
978 (sector_t
)index
<< sizebits
,
982 if (!try_to_free_buffers(page
))
987 * Allocate some buffers for this page
989 bh
= alloc_page_buffers(page
, size
, true);
992 * Link the page to the buffers and initialise them. Take the
993 * lock to be atomic wrt __find_get_block(), which does not
994 * run under the page lock.
996 spin_lock(&inode
->i_mapping
->private_lock
);
997 link_dev_buffers(page
, bh
);
998 end_block
= init_page_buffers(page
, bdev
, (sector_t
)index
<< sizebits
,
1000 spin_unlock(&inode
->i_mapping
->private_lock
);
1002 ret
= (block
< end_block
) ? 1 : -ENXIO
;
1010 * Create buffers for the specified block device block's page. If
1011 * that page was dirty, the buffers are set dirty also.
1014 grow_buffers(struct block_device
*bdev
, sector_t block
, int size
, gfp_t gfp
)
1022 } while ((size
<< sizebits
) < PAGE_SIZE
);
1024 index
= block
>> sizebits
;
1027 * Check for a block which wants to lie outside our maximum possible
1028 * pagecache index. (this comparison is done using sector_t types).
1030 if (unlikely(index
!= block
>> sizebits
)) {
1031 printk(KERN_ERR
"%s: requested out-of-range block %llu for "
1033 __func__
, (unsigned long long)block
,
1038 /* Create a page with the proper size buffers.. */
1039 return grow_dev_page(bdev
, block
, index
, size
, sizebits
, gfp
);
1042 static struct buffer_head
*
1043 __getblk_slow(struct block_device
*bdev
, sector_t block
,
1044 unsigned size
, gfp_t gfp
)
1046 /* Size must be multiple of hard sectorsize */
1047 if (unlikely(size
& (bdev_logical_block_size(bdev
)-1) ||
1048 (size
< 512 || size
> PAGE_SIZE
))) {
1049 printk(KERN_ERR
"getblk(): invalid block size %d requested\n",
1051 printk(KERN_ERR
"logical block size: %d\n",
1052 bdev_logical_block_size(bdev
));
1059 struct buffer_head
*bh
;
1062 bh
= __find_get_block(bdev
, block
, size
);
1066 ret
= grow_buffers(bdev
, block
, size
, gfp
);
1073 * The relationship between dirty buffers and dirty pages:
1075 * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1076 * the page is tagged dirty in its radix tree.
1078 * At all times, the dirtiness of the buffers represents the dirtiness of
1079 * subsections of the page. If the page has buffers, the page dirty bit is
1080 * merely a hint about the true dirty state.
1082 * When a page is set dirty in its entirety, all its buffers are marked dirty
1083 * (if the page has buffers).
1085 * When a buffer is marked dirty, its page is dirtied, but the page's other
1088 * Also. When blockdev buffers are explicitly read with bread(), they
1089 * individually become uptodate. But their backing page remains not
1090 * uptodate - even if all of its buffers are uptodate. A subsequent
1091 * block_read_full_page() against that page will discover all the uptodate
1092 * buffers, will set the page uptodate and will perform no I/O.
1096 * mark_buffer_dirty - mark a buffer_head as needing writeout
1097 * @bh: the buffer_head to mark dirty
1099 * mark_buffer_dirty() will set the dirty bit against the buffer, then set its
1100 * backing page dirty, then tag the page as dirty in its address_space's radix
1101 * tree and then attach the address_space's inode to its superblock's dirty
1104 * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
1105 * mapping->tree_lock and mapping->host->i_lock.
1107 void mark_buffer_dirty(struct buffer_head
*bh
)
1109 WARN_ON_ONCE(!buffer_uptodate(bh
));
1111 trace_block_dirty_buffer(bh
);
1114 * Very *carefully* optimize the it-is-already-dirty case.
1116 * Don't let the final "is it dirty" escape to before we
1117 * perhaps modified the buffer.
1119 if (buffer_dirty(bh
)) {
1121 if (buffer_dirty(bh
))
1125 if (!test_set_buffer_dirty(bh
)) {
1126 struct page
*page
= bh
->b_page
;
1127 struct address_space
*mapping
= NULL
;
1129 lock_page_memcg(page
);
1130 if (!TestSetPageDirty(page
)) {
1131 mapping
= page_mapping(page
);
1133 __set_page_dirty(page
, mapping
, 0);
1135 unlock_page_memcg(page
);
1137 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
1140 EXPORT_SYMBOL(mark_buffer_dirty
);
1142 void mark_buffer_write_io_error(struct buffer_head
*bh
)
1144 set_buffer_write_io_error(bh
);
1145 /* FIXME: do we need to set this in both places? */
1146 if (bh
->b_page
&& bh
->b_page
->mapping
)
1147 mapping_set_error(bh
->b_page
->mapping
, -EIO
);
1148 if (bh
->b_assoc_map
)
1149 mapping_set_error(bh
->b_assoc_map
, -EIO
);
1151 EXPORT_SYMBOL(mark_buffer_write_io_error
);
1154 * Decrement a buffer_head's reference count. If all buffers against a page
1155 * have zero reference count, are clean and unlocked, and if the page is clean
1156 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1157 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1158 * a page but it ends up not being freed, and buffers may later be reattached).
1160 void __brelse(struct buffer_head
* buf
)
1162 if (atomic_read(&buf
->b_count
)) {
1166 WARN(1, KERN_ERR
"VFS: brelse: Trying to free free buffer\n");
1168 EXPORT_SYMBOL(__brelse
);
1171 * bforget() is like brelse(), except it discards any
1172 * potentially dirty data.
1174 void __bforget(struct buffer_head
*bh
)
1176 clear_buffer_dirty(bh
);
1177 if (bh
->b_assoc_map
) {
1178 struct address_space
*buffer_mapping
= bh
->b_page
->mapping
;
1180 spin_lock(&buffer_mapping
->private_lock
);
1181 list_del_init(&bh
->b_assoc_buffers
);
1182 bh
->b_assoc_map
= NULL
;
1183 spin_unlock(&buffer_mapping
->private_lock
);
1187 EXPORT_SYMBOL(__bforget
);
1189 static struct buffer_head
*__bread_slow(struct buffer_head
*bh
)
1192 if (buffer_uptodate(bh
)) {
1197 bh
->b_end_io
= end_buffer_read_sync
;
1198 submit_bh(REQ_OP_READ
, 0, bh
);
1200 if (buffer_uptodate(bh
))
1208 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1209 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1210 * refcount elevated by one when they're in an LRU. A buffer can only appear
1211 * once in a particular CPU's LRU. A single buffer can be present in multiple
1212 * CPU's LRUs at the same time.
1214 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1215 * sb_find_get_block().
1217 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1218 * a local interrupt disable for that.
1221 #define BH_LRU_SIZE 16
1224 struct buffer_head
*bhs
[BH_LRU_SIZE
];
1227 static DEFINE_PER_CPU(struct bh_lru
, bh_lrus
) = {{ NULL
}};
1230 #define bh_lru_lock() local_irq_disable()
1231 #define bh_lru_unlock() local_irq_enable()
1233 #define bh_lru_lock() preempt_disable()
1234 #define bh_lru_unlock() preempt_enable()
1237 static inline void check_irqs_on(void)
1239 #ifdef irqs_disabled
1240 BUG_ON(irqs_disabled());
1245 * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is
1246 * inserted at the front, and the buffer_head at the back if any is evicted.
1247 * Or, if already in the LRU it is moved to the front.
1249 static void bh_lru_install(struct buffer_head
*bh
)
1251 struct buffer_head
*evictee
= bh
;
1258 b
= this_cpu_ptr(&bh_lrus
);
1259 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1260 swap(evictee
, b
->bhs
[i
]);
1261 if (evictee
== bh
) {
1273 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1275 static struct buffer_head
*
1276 lookup_bh_lru(struct block_device
*bdev
, sector_t block
, unsigned size
)
1278 struct buffer_head
*ret
= NULL
;
1283 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1284 struct buffer_head
*bh
= __this_cpu_read(bh_lrus
.bhs
[i
]);
1286 if (bh
&& bh
->b_blocknr
== block
&& bh
->b_bdev
== bdev
&&
1287 bh
->b_size
== size
) {
1290 __this_cpu_write(bh_lrus
.bhs
[i
],
1291 __this_cpu_read(bh_lrus
.bhs
[i
- 1]));
1294 __this_cpu_write(bh_lrus
.bhs
[0], bh
);
1306 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1307 * it in the LRU and mark it as accessed. If it is not present then return
1310 struct buffer_head
*
1311 __find_get_block(struct block_device
*bdev
, sector_t block
, unsigned size
)
1313 struct buffer_head
*bh
= lookup_bh_lru(bdev
, block
, size
);
1316 /* __find_get_block_slow will mark the page accessed */
1317 bh
= __find_get_block_slow(bdev
, block
);
1325 EXPORT_SYMBOL(__find_get_block
);
1328 * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
1329 * which corresponds to the passed block_device, block and size. The
1330 * returned buffer has its reference count incremented.
1332 * __getblk_gfp() will lock up the machine if grow_dev_page's
1333 * try_to_free_buffers() attempt is failing. FIXME, perhaps?
1335 struct buffer_head
*
1336 __getblk_gfp(struct block_device
*bdev
, sector_t block
,
1337 unsigned size
, gfp_t gfp
)
1339 struct buffer_head
*bh
= __find_get_block(bdev
, block
, size
);
1343 bh
= __getblk_slow(bdev
, block
, size
, gfp
);
1346 EXPORT_SYMBOL(__getblk_gfp
);
1349 * Do async read-ahead on a buffer..
1351 void __breadahead(struct block_device
*bdev
, sector_t block
, unsigned size
)
1353 struct buffer_head
*bh
= __getblk(bdev
, block
, size
);
1355 ll_rw_block(REQ_OP_READ
, REQ_RAHEAD
, 1, &bh
);
1359 EXPORT_SYMBOL(__breadahead
);
1362 * __bread_gfp() - reads a specified block and returns the bh
1363 * @bdev: the block_device to read from
1364 * @block: number of block
1365 * @size: size (in bytes) to read
1366 * @gfp: page allocation flag
1368 * Reads a specified block, and returns buffer head that contains it.
1369 * The page cache can be allocated from non-movable area
1370 * not to prevent page migration if you set gfp to zero.
1371 * It returns NULL if the block was unreadable.
1373 struct buffer_head
*
1374 __bread_gfp(struct block_device
*bdev
, sector_t block
,
1375 unsigned size
, gfp_t gfp
)
1377 struct buffer_head
*bh
= __getblk_gfp(bdev
, block
, size
, gfp
);
1379 if (likely(bh
) && !buffer_uptodate(bh
))
1380 bh
= __bread_slow(bh
);
1383 EXPORT_SYMBOL(__bread_gfp
);
1386 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1387 * This doesn't race because it runs in each cpu either in irq
1388 * or with preempt disabled.
1390 static void invalidate_bh_lru(void *arg
)
1392 struct bh_lru
*b
= &get_cpu_var(bh_lrus
);
1395 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1399 put_cpu_var(bh_lrus
);
1402 static bool has_bh_in_lru(int cpu
, void *dummy
)
1404 struct bh_lru
*b
= per_cpu_ptr(&bh_lrus
, cpu
);
1407 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1415 void invalidate_bh_lrus(void)
1417 on_each_cpu_cond(has_bh_in_lru
, invalidate_bh_lru
, NULL
, 1, GFP_KERNEL
);
1419 EXPORT_SYMBOL_GPL(invalidate_bh_lrus
);
1421 void set_bh_page(struct buffer_head
*bh
,
1422 struct page
*page
, unsigned long offset
)
1425 BUG_ON(offset
>= PAGE_SIZE
);
1426 if (PageHighMem(page
))
1428 * This catches illegal uses and preserves the offset:
1430 bh
->b_data
= (char *)(0 + offset
);
1432 bh
->b_data
= page_address(page
) + offset
;
1434 EXPORT_SYMBOL(set_bh_page
);
1437 * Called when truncating a buffer on a page completely.
1440 /* Bits that are cleared during an invalidate */
1441 #define BUFFER_FLAGS_DISCARD \
1442 (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1443 1 << BH_Delay | 1 << BH_Unwritten)
1445 static void discard_buffer(struct buffer_head
* bh
)
1447 unsigned long b_state
, b_state_old
;
1450 clear_buffer_dirty(bh
);
1452 b_state
= bh
->b_state
;
1454 b_state_old
= cmpxchg(&bh
->b_state
, b_state
,
1455 (b_state
& ~BUFFER_FLAGS_DISCARD
));
1456 if (b_state_old
== b_state
)
1458 b_state
= b_state_old
;
1464 * block_invalidatepage - invalidate part or all of a buffer-backed page
1466 * @page: the page which is affected
1467 * @offset: start of the range to invalidate
1468 * @length: length of the range to invalidate
1470 * block_invalidatepage() is called when all or part of the page has become
1471 * invalidated by a truncate operation.
1473 * block_invalidatepage() does not have to release all buffers, but it must
1474 * ensure that no dirty buffer is left outside @offset and that no I/O
1475 * is underway against any of the blocks which are outside the truncation
1476 * point. Because the caller is about to free (and possibly reuse) those
1479 void block_invalidatepage(struct page
*page
, unsigned int offset
,
1480 unsigned int length
)
1482 struct buffer_head
*head
, *bh
, *next
;
1483 unsigned int curr_off
= 0;
1484 unsigned int stop
= length
+ offset
;
1486 BUG_ON(!PageLocked(page
));
1487 if (!page_has_buffers(page
))
1491 * Check for overflow
1493 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1495 head
= page_buffers(page
);
1498 unsigned int next_off
= curr_off
+ bh
->b_size
;
1499 next
= bh
->b_this_page
;
1502 * Are we still fully in range ?
1504 if (next_off
> stop
)
1508 * is this block fully invalidated?
1510 if (offset
<= curr_off
)
1512 curr_off
= next_off
;
1514 } while (bh
!= head
);
1517 * We release buffers only if the entire page is being invalidated.
1518 * The get_block cached value has been unconditionally invalidated,
1519 * so real IO is not possible anymore.
1522 try_to_release_page(page
, 0);
1526 EXPORT_SYMBOL(block_invalidatepage
);
1530 * We attach and possibly dirty the buffers atomically wrt
1531 * __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1532 * is already excluded via the page lock.
1534 void create_empty_buffers(struct page
*page
,
1535 unsigned long blocksize
, unsigned long b_state
)
1537 struct buffer_head
*bh
, *head
, *tail
;
1539 head
= alloc_page_buffers(page
, blocksize
, true);
1542 bh
->b_state
|= b_state
;
1544 bh
= bh
->b_this_page
;
1546 tail
->b_this_page
= head
;
1548 spin_lock(&page
->mapping
->private_lock
);
1549 if (PageUptodate(page
) || PageDirty(page
)) {
1552 if (PageDirty(page
))
1553 set_buffer_dirty(bh
);
1554 if (PageUptodate(page
))
1555 set_buffer_uptodate(bh
);
1556 bh
= bh
->b_this_page
;
1557 } while (bh
!= head
);
1559 attach_page_buffers(page
, head
);
1560 spin_unlock(&page
->mapping
->private_lock
);
1562 EXPORT_SYMBOL(create_empty_buffers
);
1565 * clean_bdev_aliases: clean a range of buffers in block device
1566 * @bdev: Block device to clean buffers in
1567 * @block: Start of a range of blocks to clean
1568 * @len: Number of blocks to clean
1570 * We are taking a range of blocks for data and we don't want writeback of any
1571 * buffer-cache aliases starting from return from this function and until the
1572 * moment when something will explicitly mark the buffer dirty (hopefully that
1573 * will not happen until we will free that block ;-) We don't even need to mark
1574 * it not-uptodate - nobody can expect anything from a newly allocated buffer
1575 * anyway. We used to use unmap_buffer() for such invalidation, but that was
1576 * wrong. We definitely don't want to mark the alias unmapped, for example - it
1577 * would confuse anyone who might pick it with bread() afterwards...
1579 * Also.. Note that bforget() doesn't lock the buffer. So there can be
1580 * writeout I/O going on against recently-freed buffers. We don't wait on that
1581 * I/O in bforget() - it's more efficient to wait on the I/O only if we really
1582 * need to. That happens here.
1584 void clean_bdev_aliases(struct block_device
*bdev
, sector_t block
, sector_t len
)
1586 struct inode
*bd_inode
= bdev
->bd_inode
;
1587 struct address_space
*bd_mapping
= bd_inode
->i_mapping
;
1588 struct pagevec pvec
;
1589 pgoff_t index
= block
>> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
1592 struct buffer_head
*bh
;
1593 struct buffer_head
*head
;
1595 end
= (block
+ len
- 1) >> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
1596 pagevec_init(&pvec
);
1597 while (pagevec_lookup_range(&pvec
, bd_mapping
, &index
, end
)) {
1598 count
= pagevec_count(&pvec
);
1599 for (i
= 0; i
< count
; i
++) {
1600 struct page
*page
= pvec
.pages
[i
];
1602 if (!page_has_buffers(page
))
1605 * We use page lock instead of bd_mapping->private_lock
1606 * to pin buffers here since we can afford to sleep and
1607 * it scales better than a global spinlock lock.
1610 /* Recheck when the page is locked which pins bhs */
1611 if (!page_has_buffers(page
))
1613 head
= page_buffers(page
);
1616 if (!buffer_mapped(bh
) || (bh
->b_blocknr
< block
))
1618 if (bh
->b_blocknr
>= block
+ len
)
1620 clear_buffer_dirty(bh
);
1622 clear_buffer_req(bh
);
1624 bh
= bh
->b_this_page
;
1625 } while (bh
!= head
);
1629 pagevec_release(&pvec
);
1631 /* End of range already reached? */
1632 if (index
> end
|| !index
)
1636 EXPORT_SYMBOL(clean_bdev_aliases
);
1639 * Size is a power-of-two in the range 512..PAGE_SIZE,
1640 * and the case we care about most is PAGE_SIZE.
1642 * So this *could* possibly be written with those
1643 * constraints in mind (relevant mostly if some
1644 * architecture has a slow bit-scan instruction)
1646 static inline int block_size_bits(unsigned int blocksize
)
1648 return ilog2(blocksize
);
1651 static struct buffer_head
*create_page_buffers(struct page
*page
, struct inode
*inode
, unsigned int b_state
)
1653 BUG_ON(!PageLocked(page
));
1655 if (!page_has_buffers(page
))
1656 create_empty_buffers(page
, 1 << READ_ONCE(inode
->i_blkbits
),
1658 return page_buffers(page
);
1662 * NOTE! All mapped/uptodate combinations are valid:
1664 * Mapped Uptodate Meaning
1666 * No No "unknown" - must do get_block()
1667 * No Yes "hole" - zero-filled
1668 * Yes No "allocated" - allocated on disk, not read in
1669 * Yes Yes "valid" - allocated and up-to-date in memory.
1671 * "Dirty" is valid only with the last case (mapped+uptodate).
1675 * While block_write_full_page is writing back the dirty buffers under
1676 * the page lock, whoever dirtied the buffers may decide to clean them
1677 * again at any time. We handle that by only looking at the buffer
1678 * state inside lock_buffer().
1680 * If block_write_full_page() is called for regular writeback
1681 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1682 * locked buffer. This only can happen if someone has written the buffer
1683 * directly, with submit_bh(). At the address_space level PageWriteback
1684 * prevents this contention from occurring.
1686 * If block_write_full_page() is called with wbc->sync_mode ==
1687 * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this
1688 * causes the writes to be flagged as synchronous writes.
1690 int __block_write_full_page(struct inode
*inode
, struct page
*page
,
1691 get_block_t
*get_block
, struct writeback_control
*wbc
,
1692 bh_end_io_t
*handler
)
1696 sector_t last_block
;
1697 struct buffer_head
*bh
, *head
;
1698 unsigned int blocksize
, bbits
;
1699 int nr_underway
= 0;
1700 int write_flags
= wbc_to_write_flags(wbc
);
1702 head
= create_page_buffers(page
, inode
,
1703 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
1706 * Be very careful. We have no exclusion from __set_page_dirty_buffers
1707 * here, and the (potentially unmapped) buffers may become dirty at
1708 * any time. If a buffer becomes dirty here after we've inspected it
1709 * then we just miss that fact, and the page stays dirty.
1711 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1712 * handle that here by just cleaning them.
1716 blocksize
= bh
->b_size
;
1717 bbits
= block_size_bits(blocksize
);
1719 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1720 last_block
= (i_size_read(inode
) - 1) >> bbits
;
1723 * Get all the dirty buffers mapped to disk addresses and
1724 * handle any aliases from the underlying blockdev's mapping.
1727 if (block
> last_block
) {
1729 * mapped buffers outside i_size will occur, because
1730 * this page can be outside i_size when there is a
1731 * truncate in progress.
1734 * The buffer was zeroed by block_write_full_page()
1736 clear_buffer_dirty(bh
);
1737 set_buffer_uptodate(bh
);
1738 } else if ((!buffer_mapped(bh
) || buffer_delay(bh
)) &&
1740 WARN_ON(bh
->b_size
!= blocksize
);
1741 err
= get_block(inode
, block
, bh
, 1);
1744 clear_buffer_delay(bh
);
1745 if (buffer_new(bh
)) {
1746 /* blockdev mappings never come here */
1747 clear_buffer_new(bh
);
1748 clean_bdev_bh_alias(bh
);
1751 bh
= bh
->b_this_page
;
1753 } while (bh
!= head
);
1756 if (!buffer_mapped(bh
))
1759 * If it's a fully non-blocking write attempt and we cannot
1760 * lock the buffer then redirty the page. Note that this can
1761 * potentially cause a busy-wait loop from writeback threads
1762 * and kswapd activity, but those code paths have their own
1763 * higher-level throttling.
1765 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
1767 } else if (!trylock_buffer(bh
)) {
1768 redirty_page_for_writepage(wbc
, page
);
1771 if (test_clear_buffer_dirty(bh
)) {
1772 mark_buffer_async_write_endio(bh
, handler
);
1776 } while ((bh
= bh
->b_this_page
) != head
);
1779 * The page and its buffers are protected by PageWriteback(), so we can
1780 * drop the bh refcounts early.
1782 BUG_ON(PageWriteback(page
));
1783 set_page_writeback(page
);
1786 struct buffer_head
*next
= bh
->b_this_page
;
1787 if (buffer_async_write(bh
)) {
1788 submit_bh_wbc(REQ_OP_WRITE
, write_flags
, bh
,
1789 inode
->i_write_hint
, wbc
);
1793 } while (bh
!= head
);
1798 if (nr_underway
== 0) {
1800 * The page was marked dirty, but the buffers were
1801 * clean. Someone wrote them back by hand with
1802 * ll_rw_block/submit_bh. A rare case.
1804 end_page_writeback(page
);
1807 * The page and buffer_heads can be released at any time from
1815 * ENOSPC, or some other error. We may already have added some
1816 * blocks to the file, so we need to write these out to avoid
1817 * exposing stale data.
1818 * The page is currently locked and not marked for writeback
1821 /* Recovery: lock and submit the mapped buffers */
1823 if (buffer_mapped(bh
) && buffer_dirty(bh
) &&
1824 !buffer_delay(bh
)) {
1826 mark_buffer_async_write_endio(bh
, handler
);
1829 * The buffer may have been set dirty during
1830 * attachment to a dirty page.
1832 clear_buffer_dirty(bh
);
1834 } while ((bh
= bh
->b_this_page
) != head
);
1836 BUG_ON(PageWriteback(page
));
1837 mapping_set_error(page
->mapping
, err
);
1838 set_page_writeback(page
);
1840 struct buffer_head
*next
= bh
->b_this_page
;
1841 if (buffer_async_write(bh
)) {
1842 clear_buffer_dirty(bh
);
1843 submit_bh_wbc(REQ_OP_WRITE
, write_flags
, bh
,
1844 inode
->i_write_hint
, wbc
);
1848 } while (bh
!= head
);
1852 EXPORT_SYMBOL(__block_write_full_page
);
1855 * If a page has any new buffers, zero them out here, and mark them uptodate
1856 * and dirty so they'll be written out (in order to prevent uninitialised
1857 * block data from leaking). And clear the new bit.
1859 void page_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1861 unsigned int block_start
, block_end
;
1862 struct buffer_head
*head
, *bh
;
1864 BUG_ON(!PageLocked(page
));
1865 if (!page_has_buffers(page
))
1868 bh
= head
= page_buffers(page
);
1871 block_end
= block_start
+ bh
->b_size
;
1873 if (buffer_new(bh
)) {
1874 if (block_end
> from
&& block_start
< to
) {
1875 if (!PageUptodate(page
)) {
1876 unsigned start
, size
;
1878 start
= max(from
, block_start
);
1879 size
= min(to
, block_end
) - start
;
1881 zero_user(page
, start
, size
);
1882 set_buffer_uptodate(bh
);
1885 clear_buffer_new(bh
);
1886 mark_buffer_dirty(bh
);
1890 block_start
= block_end
;
1891 bh
= bh
->b_this_page
;
1892 } while (bh
!= head
);
1894 EXPORT_SYMBOL(page_zero_new_buffers
);
1897 iomap_to_bh(struct inode
*inode
, sector_t block
, struct buffer_head
*bh
,
1898 struct iomap
*iomap
)
1900 loff_t offset
= block
<< inode
->i_blkbits
;
1902 bh
->b_bdev
= iomap
->bdev
;
1905 * Block points to offset in file we need to map, iomap contains
1906 * the offset at which the map starts. If the map ends before the
1907 * current block, then do not map the buffer and let the caller
1910 BUG_ON(offset
>= iomap
->offset
+ iomap
->length
);
1912 switch (iomap
->type
) {
1915 * If the buffer is not up to date or beyond the current EOF,
1916 * we need to mark it as new to ensure sub-block zeroing is
1917 * executed if necessary.
1919 if (!buffer_uptodate(bh
) ||
1920 (offset
>= i_size_read(inode
)))
1923 case IOMAP_DELALLOC
:
1924 if (!buffer_uptodate(bh
) ||
1925 (offset
>= i_size_read(inode
)))
1927 set_buffer_uptodate(bh
);
1928 set_buffer_mapped(bh
);
1929 set_buffer_delay(bh
);
1931 case IOMAP_UNWRITTEN
:
1933 * For unwritten regions, we always need to ensure that
1934 * sub-block writes cause the regions in the block we are not
1935 * writing to are zeroed. Set the buffer as new to ensure this.
1938 set_buffer_unwritten(bh
);
1941 if (offset
>= i_size_read(inode
))
1943 bh
->b_blocknr
= (iomap
->addr
+ offset
- iomap
->offset
) >>
1945 set_buffer_mapped(bh
);
1950 int __block_write_begin_int(struct page
*page
, loff_t pos
, unsigned len
,
1951 get_block_t
*get_block
, struct iomap
*iomap
)
1953 unsigned from
= pos
& (PAGE_SIZE
- 1);
1954 unsigned to
= from
+ len
;
1955 struct inode
*inode
= page
->mapping
->host
;
1956 unsigned block_start
, block_end
;
1959 unsigned blocksize
, bbits
;
1960 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
=wait
;
1962 BUG_ON(!PageLocked(page
));
1963 BUG_ON(from
> PAGE_SIZE
);
1964 BUG_ON(to
> PAGE_SIZE
);
1967 head
= create_page_buffers(page
, inode
, 0);
1968 blocksize
= head
->b_size
;
1969 bbits
= block_size_bits(blocksize
);
1971 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1973 for(bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1974 block
++, block_start
=block_end
, bh
= bh
->b_this_page
) {
1975 block_end
= block_start
+ blocksize
;
1976 if (block_end
<= from
|| block_start
>= to
) {
1977 if (PageUptodate(page
)) {
1978 if (!buffer_uptodate(bh
))
1979 set_buffer_uptodate(bh
);
1984 clear_buffer_new(bh
);
1985 if (!buffer_mapped(bh
)) {
1986 WARN_ON(bh
->b_size
!= blocksize
);
1988 err
= get_block(inode
, block
, bh
, 1);
1992 iomap_to_bh(inode
, block
, bh
, iomap
);
1995 if (buffer_new(bh
)) {
1996 clean_bdev_bh_alias(bh
);
1997 if (PageUptodate(page
)) {
1998 clear_buffer_new(bh
);
1999 set_buffer_uptodate(bh
);
2000 mark_buffer_dirty(bh
);
2003 if (block_end
> to
|| block_start
< from
)
2004 zero_user_segments(page
,
2010 if (PageUptodate(page
)) {
2011 if (!buffer_uptodate(bh
))
2012 set_buffer_uptodate(bh
);
2015 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
2016 !buffer_unwritten(bh
) &&
2017 (block_start
< from
|| block_end
> to
)) {
2018 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
2023 * If we issued read requests - let them complete.
2025 while(wait_bh
> wait
) {
2026 wait_on_buffer(*--wait_bh
);
2027 if (!buffer_uptodate(*wait_bh
))
2031 page_zero_new_buffers(page
, from
, to
);
2035 int __block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
2036 get_block_t
*get_block
)
2038 return __block_write_begin_int(page
, pos
, len
, get_block
, NULL
);
2040 EXPORT_SYMBOL(__block_write_begin
);
2042 static int __block_commit_write(struct inode
*inode
, struct page
*page
,
2043 unsigned from
, unsigned to
)
2045 unsigned block_start
, block_end
;
2048 struct buffer_head
*bh
, *head
;
2050 bh
= head
= page_buffers(page
);
2051 blocksize
= bh
->b_size
;
2055 block_end
= block_start
+ blocksize
;
2056 if (block_end
<= from
|| block_start
>= to
) {
2057 if (!buffer_uptodate(bh
))
2060 set_buffer_uptodate(bh
);
2061 mark_buffer_dirty(bh
);
2063 clear_buffer_new(bh
);
2065 block_start
= block_end
;
2066 bh
= bh
->b_this_page
;
2067 } while (bh
!= head
);
2070 * If this is a partial write which happened to make all buffers
2071 * uptodate then we can optimize away a bogus readpage() for
2072 * the next read(). Here we 'discover' whether the page went
2073 * uptodate as a result of this (potentially partial) write.
2076 SetPageUptodate(page
);
2081 * block_write_begin takes care of the basic task of block allocation and
2082 * bringing partial write blocks uptodate first.
2084 * The filesystem needs to handle block truncation upon failure.
2086 int block_write_begin(struct address_space
*mapping
, loff_t pos
, unsigned len
,
2087 unsigned flags
, struct page
**pagep
, get_block_t
*get_block
)
2089 pgoff_t index
= pos
>> PAGE_SHIFT
;
2093 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2097 status
= __block_write_begin(page
, pos
, len
, get_block
);
2098 if (unlikely(status
)) {
2107 EXPORT_SYMBOL(block_write_begin
);
2109 int block_write_end(struct file
*file
, struct address_space
*mapping
,
2110 loff_t pos
, unsigned len
, unsigned copied
,
2111 struct page
*page
, void *fsdata
)
2113 struct inode
*inode
= mapping
->host
;
2116 start
= pos
& (PAGE_SIZE
- 1);
2118 if (unlikely(copied
< len
)) {
2120 * The buffers that were written will now be uptodate, so we
2121 * don't have to worry about a readpage reading them and
2122 * overwriting a partial write. However if we have encountered
2123 * a short write and only partially written into a buffer, it
2124 * will not be marked uptodate, so a readpage might come in and
2125 * destroy our partial write.
2127 * Do the simplest thing, and just treat any short write to a
2128 * non uptodate page as a zero-length write, and force the
2129 * caller to redo the whole thing.
2131 if (!PageUptodate(page
))
2134 page_zero_new_buffers(page
, start
+copied
, start
+len
);
2136 flush_dcache_page(page
);
2138 /* This could be a short (even 0-length) commit */
2139 __block_commit_write(inode
, page
, start
, start
+copied
);
2143 EXPORT_SYMBOL(block_write_end
);
2145 int generic_write_end(struct file
*file
, struct address_space
*mapping
,
2146 loff_t pos
, unsigned len
, unsigned copied
,
2147 struct page
*page
, void *fsdata
)
2149 struct inode
*inode
= mapping
->host
;
2150 loff_t old_size
= inode
->i_size
;
2151 int i_size_changed
= 0;
2153 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2156 * No need to use i_size_read() here, the i_size
2157 * cannot change under us because we hold i_mutex.
2159 * But it's important to update i_size while still holding page lock:
2160 * page writeout could otherwise come in and zero beyond i_size.
2162 if (pos
+copied
> inode
->i_size
) {
2163 i_size_write(inode
, pos
+copied
);
2171 pagecache_isize_extended(inode
, old_size
, pos
);
2173 * Don't mark the inode dirty under page lock. First, it unnecessarily
2174 * makes the holding time of page lock longer. Second, it forces lock
2175 * ordering of page lock and transaction start for journaling
2179 mark_inode_dirty(inode
);
2183 EXPORT_SYMBOL(generic_write_end
);
2186 * block_is_partially_uptodate checks whether buffers within a page are
2189 * Returns true if all buffers which correspond to a file portion
2190 * we want to read are uptodate.
2192 int block_is_partially_uptodate(struct page
*page
, unsigned long from
,
2193 unsigned long count
)
2195 unsigned block_start
, block_end
, blocksize
;
2197 struct buffer_head
*bh
, *head
;
2200 if (!page_has_buffers(page
))
2203 head
= page_buffers(page
);
2204 blocksize
= head
->b_size
;
2205 to
= min_t(unsigned, PAGE_SIZE
- from
, count
);
2207 if (from
< blocksize
&& to
> PAGE_SIZE
- blocksize
)
2213 block_end
= block_start
+ blocksize
;
2214 if (block_end
> from
&& block_start
< to
) {
2215 if (!buffer_uptodate(bh
)) {
2219 if (block_end
>= to
)
2222 block_start
= block_end
;
2223 bh
= bh
->b_this_page
;
2224 } while (bh
!= head
);
2228 EXPORT_SYMBOL(block_is_partially_uptodate
);
2231 * Generic "read page" function for block devices that have the normal
2232 * get_block functionality. This is most of the block device filesystems.
2233 * Reads the page asynchronously --- the unlock_buffer() and
2234 * set/clear_buffer_uptodate() functions propagate buffer state into the
2235 * page struct once IO has completed.
2237 int block_read_full_page(struct page
*page
, get_block_t
*get_block
)
2239 struct inode
*inode
= page
->mapping
->host
;
2240 sector_t iblock
, lblock
;
2241 struct buffer_head
*bh
, *head
, *arr
[MAX_BUF_PER_PAGE
];
2242 unsigned int blocksize
, bbits
;
2244 int fully_mapped
= 1;
2246 head
= create_page_buffers(page
, inode
, 0);
2247 blocksize
= head
->b_size
;
2248 bbits
= block_size_bits(blocksize
);
2250 iblock
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
2251 lblock
= (i_size_read(inode
)+blocksize
-1) >> bbits
;
2257 if (buffer_uptodate(bh
))
2260 if (!buffer_mapped(bh
)) {
2264 if (iblock
< lblock
) {
2265 WARN_ON(bh
->b_size
!= blocksize
);
2266 err
= get_block(inode
, iblock
, bh
, 0);
2270 if (!buffer_mapped(bh
)) {
2271 zero_user(page
, i
* blocksize
, blocksize
);
2273 set_buffer_uptodate(bh
);
2277 * get_block() might have updated the buffer
2280 if (buffer_uptodate(bh
))
2284 } while (i
++, iblock
++, (bh
= bh
->b_this_page
) != head
);
2287 SetPageMappedToDisk(page
);
2291 * All buffers are uptodate - we can set the page uptodate
2292 * as well. But not if get_block() returned an error.
2294 if (!PageError(page
))
2295 SetPageUptodate(page
);
2300 /* Stage two: lock the buffers */
2301 for (i
= 0; i
< nr
; i
++) {
2304 mark_buffer_async_read(bh
);
2308 * Stage 3: start the IO. Check for uptodateness
2309 * inside the buffer lock in case another process reading
2310 * the underlying blockdev brought it uptodate (the sct fix).
2312 for (i
= 0; i
< nr
; i
++) {
2314 if (buffer_uptodate(bh
))
2315 end_buffer_async_read(bh
, 1);
2317 submit_bh(REQ_OP_READ
, 0, bh
);
2321 EXPORT_SYMBOL(block_read_full_page
);
2323 /* utility function for filesystems that need to do work on expanding
2324 * truncates. Uses filesystem pagecache writes to allow the filesystem to
2325 * deal with the hole.
2327 int generic_cont_expand_simple(struct inode
*inode
, loff_t size
)
2329 struct address_space
*mapping
= inode
->i_mapping
;
2334 err
= inode_newsize_ok(inode
, size
);
2338 err
= pagecache_write_begin(NULL
, mapping
, size
, 0,
2339 AOP_FLAG_CONT_EXPAND
, &page
, &fsdata
);
2343 err
= pagecache_write_end(NULL
, mapping
, size
, 0, 0, page
, fsdata
);
2349 EXPORT_SYMBOL(generic_cont_expand_simple
);
2351 static int cont_expand_zero(struct file
*file
, struct address_space
*mapping
,
2352 loff_t pos
, loff_t
*bytes
)
2354 struct inode
*inode
= mapping
->host
;
2355 unsigned int blocksize
= i_blocksize(inode
);
2358 pgoff_t index
, curidx
;
2360 unsigned zerofrom
, offset
, len
;
2363 index
= pos
>> PAGE_SHIFT
;
2364 offset
= pos
& ~PAGE_MASK
;
2366 while (index
> (curidx
= (curpos
= *bytes
)>>PAGE_SHIFT
)) {
2367 zerofrom
= curpos
& ~PAGE_MASK
;
2368 if (zerofrom
& (blocksize
-1)) {
2369 *bytes
|= (blocksize
-1);
2372 len
= PAGE_SIZE
- zerofrom
;
2374 err
= pagecache_write_begin(file
, mapping
, curpos
, len
, 0,
2378 zero_user(page
, zerofrom
, len
);
2379 err
= pagecache_write_end(file
, mapping
, curpos
, len
, len
,
2386 balance_dirty_pages_ratelimited(mapping
);
2388 if (unlikely(fatal_signal_pending(current
))) {
2394 /* page covers the boundary, find the boundary offset */
2395 if (index
== curidx
) {
2396 zerofrom
= curpos
& ~PAGE_MASK
;
2397 /* if we will expand the thing last block will be filled */
2398 if (offset
<= zerofrom
) {
2401 if (zerofrom
& (blocksize
-1)) {
2402 *bytes
|= (blocksize
-1);
2405 len
= offset
- zerofrom
;
2407 err
= pagecache_write_begin(file
, mapping
, curpos
, len
, 0,
2411 zero_user(page
, zerofrom
, len
);
2412 err
= pagecache_write_end(file
, mapping
, curpos
, len
, len
,
2424 * For moronic filesystems that do not allow holes in file.
2425 * We may have to extend the file.
2427 int cont_write_begin(struct file
*file
, struct address_space
*mapping
,
2428 loff_t pos
, unsigned len
, unsigned flags
,
2429 struct page
**pagep
, void **fsdata
,
2430 get_block_t
*get_block
, loff_t
*bytes
)
2432 struct inode
*inode
= mapping
->host
;
2433 unsigned int blocksize
= i_blocksize(inode
);
2434 unsigned int zerofrom
;
2437 err
= cont_expand_zero(file
, mapping
, pos
, bytes
);
2441 zerofrom
= *bytes
& ~PAGE_MASK
;
2442 if (pos
+len
> *bytes
&& zerofrom
& (blocksize
-1)) {
2443 *bytes
|= (blocksize
-1);
2447 return block_write_begin(mapping
, pos
, len
, flags
, pagep
, get_block
);
2449 EXPORT_SYMBOL(cont_write_begin
);
2451 int block_commit_write(struct page
*page
, unsigned from
, unsigned to
)
2453 struct inode
*inode
= page
->mapping
->host
;
2454 __block_commit_write(inode
,page
,from
,to
);
2457 EXPORT_SYMBOL(block_commit_write
);
2460 * block_page_mkwrite() is not allowed to change the file size as it gets
2461 * called from a page fault handler when a page is first dirtied. Hence we must
2462 * be careful to check for EOF conditions here. We set the page up correctly
2463 * for a written page which means we get ENOSPC checking when writing into
2464 * holes and correct delalloc and unwritten extent mapping on filesystems that
2465 * support these features.
2467 * We are not allowed to take the i_mutex here so we have to play games to
2468 * protect against truncate races as the page could now be beyond EOF. Because
2469 * truncate writes the inode size before removing pages, once we have the
2470 * page lock we can determine safely if the page is beyond EOF. If it is not
2471 * beyond EOF, then the page is guaranteed safe against truncation until we
2474 * Direct callers of this function should protect against filesystem freezing
2475 * using sb_start_pagefault() - sb_end_pagefault() functions.
2477 int block_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
2478 get_block_t get_block
)
2480 struct page
*page
= vmf
->page
;
2481 struct inode
*inode
= file_inode(vma
->vm_file
);
2487 size
= i_size_read(inode
);
2488 if ((page
->mapping
!= inode
->i_mapping
) ||
2489 (page_offset(page
) > size
)) {
2490 /* We overload EFAULT to mean page got truncated */
2495 /* page is wholly or partially inside EOF */
2496 if (((page
->index
+ 1) << PAGE_SHIFT
) > size
)
2497 end
= size
& ~PAGE_MASK
;
2501 ret
= __block_write_begin(page
, 0, end
, get_block
);
2503 ret
= block_commit_write(page
, 0, end
);
2505 if (unlikely(ret
< 0))
2507 set_page_dirty(page
);
2508 wait_for_stable_page(page
);
2514 EXPORT_SYMBOL(block_page_mkwrite
);
2517 * nobh_write_begin()'s prereads are special: the buffer_heads are freed
2518 * immediately, while under the page lock. So it needs a special end_io
2519 * handler which does not touch the bh after unlocking it.
2521 static void end_buffer_read_nobh(struct buffer_head
*bh
, int uptodate
)
2523 __end_buffer_read_notouch(bh
, uptodate
);
2527 * Attach the singly-linked list of buffers created by nobh_write_begin, to
2528 * the page (converting it to circular linked list and taking care of page
2531 static void attach_nobh_buffers(struct page
*page
, struct buffer_head
*head
)
2533 struct buffer_head
*bh
;
2535 BUG_ON(!PageLocked(page
));
2537 spin_lock(&page
->mapping
->private_lock
);
2540 if (PageDirty(page
))
2541 set_buffer_dirty(bh
);
2542 if (!bh
->b_this_page
)
2543 bh
->b_this_page
= head
;
2544 bh
= bh
->b_this_page
;
2545 } while (bh
!= head
);
2546 attach_page_buffers(page
, head
);
2547 spin_unlock(&page
->mapping
->private_lock
);
2551 * On entry, the page is fully not uptodate.
2552 * On exit the page is fully uptodate in the areas outside (from,to)
2553 * The filesystem needs to handle block truncation upon failure.
2555 int nobh_write_begin(struct address_space
*mapping
,
2556 loff_t pos
, unsigned len
, unsigned flags
,
2557 struct page
**pagep
, void **fsdata
,
2558 get_block_t
*get_block
)
2560 struct inode
*inode
= mapping
->host
;
2561 const unsigned blkbits
= inode
->i_blkbits
;
2562 const unsigned blocksize
= 1 << blkbits
;
2563 struct buffer_head
*head
, *bh
;
2567 unsigned block_in_page
;
2568 unsigned block_start
, block_end
;
2569 sector_t block_in_file
;
2572 int is_mapped_to_disk
= 1;
2574 index
= pos
>> PAGE_SHIFT
;
2575 from
= pos
& (PAGE_SIZE
- 1);
2578 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2584 if (page_has_buffers(page
)) {
2585 ret
= __block_write_begin(page
, pos
, len
, get_block
);
2591 if (PageMappedToDisk(page
))
2595 * Allocate buffers so that we can keep track of state, and potentially
2596 * attach them to the page if an error occurs. In the common case of
2597 * no error, they will just be freed again without ever being attached
2598 * to the page (which is all OK, because we're under the page lock).
2600 * Be careful: the buffer linked list is a NULL terminated one, rather
2601 * than the circular one we're used to.
2603 head
= alloc_page_buffers(page
, blocksize
, false);
2609 block_in_file
= (sector_t
)page
->index
<< (PAGE_SHIFT
- blkbits
);
2612 * We loop across all blocks in the page, whether or not they are
2613 * part of the affected region. This is so we can discover if the
2614 * page is fully mapped-to-disk.
2616 for (block_start
= 0, block_in_page
= 0, bh
= head
;
2617 block_start
< PAGE_SIZE
;
2618 block_in_page
++, block_start
+= blocksize
, bh
= bh
->b_this_page
) {
2621 block_end
= block_start
+ blocksize
;
2624 if (block_start
>= to
)
2626 ret
= get_block(inode
, block_in_file
+ block_in_page
,
2630 if (!buffer_mapped(bh
))
2631 is_mapped_to_disk
= 0;
2633 clean_bdev_bh_alias(bh
);
2634 if (PageUptodate(page
)) {
2635 set_buffer_uptodate(bh
);
2638 if (buffer_new(bh
) || !buffer_mapped(bh
)) {
2639 zero_user_segments(page
, block_start
, from
,
2643 if (buffer_uptodate(bh
))
2644 continue; /* reiserfs does this */
2645 if (block_start
< from
|| block_end
> to
) {
2647 bh
->b_end_io
= end_buffer_read_nobh
;
2648 submit_bh(REQ_OP_READ
, 0, bh
);
2655 * The page is locked, so these buffers are protected from
2656 * any VM or truncate activity. Hence we don't need to care
2657 * for the buffer_head refcounts.
2659 for (bh
= head
; bh
; bh
= bh
->b_this_page
) {
2661 if (!buffer_uptodate(bh
))
2668 if (is_mapped_to_disk
)
2669 SetPageMappedToDisk(page
);
2671 *fsdata
= head
; /* to be released by nobh_write_end */
2678 * Error recovery is a bit difficult. We need to zero out blocks that
2679 * were newly allocated, and dirty them to ensure they get written out.
2680 * Buffers need to be attached to the page at this point, otherwise
2681 * the handling of potential IO errors during writeout would be hard
2682 * (could try doing synchronous writeout, but what if that fails too?)
2684 attach_nobh_buffers(page
, head
);
2685 page_zero_new_buffers(page
, from
, to
);
2694 EXPORT_SYMBOL(nobh_write_begin
);
2696 int nobh_write_end(struct file
*file
, struct address_space
*mapping
,
2697 loff_t pos
, unsigned len
, unsigned copied
,
2698 struct page
*page
, void *fsdata
)
2700 struct inode
*inode
= page
->mapping
->host
;
2701 struct buffer_head
*head
= fsdata
;
2702 struct buffer_head
*bh
;
2703 BUG_ON(fsdata
!= NULL
&& page_has_buffers(page
));
2705 if (unlikely(copied
< len
) && head
)
2706 attach_nobh_buffers(page
, head
);
2707 if (page_has_buffers(page
))
2708 return generic_write_end(file
, mapping
, pos
, len
,
2709 copied
, page
, fsdata
);
2711 SetPageUptodate(page
);
2712 set_page_dirty(page
);
2713 if (pos
+copied
> inode
->i_size
) {
2714 i_size_write(inode
, pos
+copied
);
2715 mark_inode_dirty(inode
);
2723 head
= head
->b_this_page
;
2724 free_buffer_head(bh
);
2729 EXPORT_SYMBOL(nobh_write_end
);
2732 * nobh_writepage() - based on block_full_write_page() except
2733 * that it tries to operate without attaching bufferheads to
2736 int nobh_writepage(struct page
*page
, get_block_t
*get_block
,
2737 struct writeback_control
*wbc
)
2739 struct inode
* const inode
= page
->mapping
->host
;
2740 loff_t i_size
= i_size_read(inode
);
2741 const pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2745 /* Is the page fully inside i_size? */
2746 if (page
->index
< end_index
)
2749 /* Is the page fully outside i_size? (truncate in progress) */
2750 offset
= i_size
& (PAGE_SIZE
-1);
2751 if (page
->index
>= end_index
+1 || !offset
) {
2753 * The page may have dirty, unmapped buffers. For example,
2754 * they may have been added in ext3_writepage(). Make them
2755 * freeable here, so the page does not leak.
2758 /* Not really sure about this - do we need this ? */
2759 if (page
->mapping
->a_ops
->invalidatepage
)
2760 page
->mapping
->a_ops
->invalidatepage(page
, offset
);
2763 return 0; /* don't care */
2767 * The page straddles i_size. It must be zeroed out on each and every
2768 * writepage invocation because it may be mmapped. "A file is mapped
2769 * in multiples of the page size. For a file that is not a multiple of
2770 * the page size, the remaining memory is zeroed when mapped, and
2771 * writes to that region are not written out to the file."
2773 zero_user_segment(page
, offset
, PAGE_SIZE
);
2775 ret
= mpage_writepage(page
, get_block
, wbc
);
2777 ret
= __block_write_full_page(inode
, page
, get_block
, wbc
,
2778 end_buffer_async_write
);
2781 EXPORT_SYMBOL(nobh_writepage
);
2783 int nobh_truncate_page(struct address_space
*mapping
,
2784 loff_t from
, get_block_t
*get_block
)
2786 pgoff_t index
= from
>> PAGE_SHIFT
;
2787 unsigned offset
= from
& (PAGE_SIZE
-1);
2790 unsigned length
, pos
;
2791 struct inode
*inode
= mapping
->host
;
2793 struct buffer_head map_bh
;
2796 blocksize
= i_blocksize(inode
);
2797 length
= offset
& (blocksize
- 1);
2799 /* Block boundary? Nothing to do */
2803 length
= blocksize
- length
;
2804 iblock
= (sector_t
)index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
2806 page
= grab_cache_page(mapping
, index
);
2811 if (page_has_buffers(page
)) {
2815 return block_truncate_page(mapping
, from
, get_block
);
2818 /* Find the buffer that contains "offset" */
2820 while (offset
>= pos
) {
2825 map_bh
.b_size
= blocksize
;
2827 err
= get_block(inode
, iblock
, &map_bh
, 0);
2830 /* unmapped? It's a hole - nothing to do */
2831 if (!buffer_mapped(&map_bh
))
2834 /* Ok, it's mapped. Make sure it's up-to-date */
2835 if (!PageUptodate(page
)) {
2836 err
= mapping
->a_ops
->readpage(NULL
, page
);
2842 if (!PageUptodate(page
)) {
2846 if (page_has_buffers(page
))
2849 zero_user(page
, offset
, length
);
2850 set_page_dirty(page
);
2859 EXPORT_SYMBOL(nobh_truncate_page
);
2861 int block_truncate_page(struct address_space
*mapping
,
2862 loff_t from
, get_block_t
*get_block
)
2864 pgoff_t index
= from
>> PAGE_SHIFT
;
2865 unsigned offset
= from
& (PAGE_SIZE
-1);
2868 unsigned length
, pos
;
2869 struct inode
*inode
= mapping
->host
;
2871 struct buffer_head
*bh
;
2874 blocksize
= i_blocksize(inode
);
2875 length
= offset
& (blocksize
- 1);
2877 /* Block boundary? Nothing to do */
2881 length
= blocksize
- length
;
2882 iblock
= (sector_t
)index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
2884 page
= grab_cache_page(mapping
, index
);
2889 if (!page_has_buffers(page
))
2890 create_empty_buffers(page
, blocksize
, 0);
2892 /* Find the buffer that contains "offset" */
2893 bh
= page_buffers(page
);
2895 while (offset
>= pos
) {
2896 bh
= bh
->b_this_page
;
2902 if (!buffer_mapped(bh
)) {
2903 WARN_ON(bh
->b_size
!= blocksize
);
2904 err
= get_block(inode
, iblock
, bh
, 0);
2907 /* unmapped? It's a hole - nothing to do */
2908 if (!buffer_mapped(bh
))
2912 /* Ok, it's mapped. Make sure it's up-to-date */
2913 if (PageUptodate(page
))
2914 set_buffer_uptodate(bh
);
2916 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) && !buffer_unwritten(bh
)) {
2918 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
2920 /* Uhhuh. Read error. Complain and punt. */
2921 if (!buffer_uptodate(bh
))
2925 zero_user(page
, offset
, length
);
2926 mark_buffer_dirty(bh
);
2935 EXPORT_SYMBOL(block_truncate_page
);
2938 * The generic ->writepage function for buffer-backed address_spaces
2940 int block_write_full_page(struct page
*page
, get_block_t
*get_block
,
2941 struct writeback_control
*wbc
)
2943 struct inode
* const inode
= page
->mapping
->host
;
2944 loff_t i_size
= i_size_read(inode
);
2945 const pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2948 /* Is the page fully inside i_size? */
2949 if (page
->index
< end_index
)
2950 return __block_write_full_page(inode
, page
, get_block
, wbc
,
2951 end_buffer_async_write
);
2953 /* Is the page fully outside i_size? (truncate in progress) */
2954 offset
= i_size
& (PAGE_SIZE
-1);
2955 if (page
->index
>= end_index
+1 || !offset
) {
2957 * The page may have dirty, unmapped buffers. For example,
2958 * they may have been added in ext3_writepage(). Make them
2959 * freeable here, so the page does not leak.
2961 do_invalidatepage(page
, 0, PAGE_SIZE
);
2963 return 0; /* don't care */
2967 * The page straddles i_size. It must be zeroed out on each and every
2968 * writepage invocation because it may be mmapped. "A file is mapped
2969 * in multiples of the page size. For a file that is not a multiple of
2970 * the page size, the remaining memory is zeroed when mapped, and
2971 * writes to that region are not written out to the file."
2973 zero_user_segment(page
, offset
, PAGE_SIZE
);
2974 return __block_write_full_page(inode
, page
, get_block
, wbc
,
2975 end_buffer_async_write
);
2977 EXPORT_SYMBOL(block_write_full_page
);
2979 sector_t
generic_block_bmap(struct address_space
*mapping
, sector_t block
,
2980 get_block_t
*get_block
)
2982 struct inode
*inode
= mapping
->host
;
2983 struct buffer_head tmp
= {
2984 .b_size
= i_blocksize(inode
),
2987 get_block(inode
, block
, &tmp
, 0);
2988 return tmp
.b_blocknr
;
2990 EXPORT_SYMBOL(generic_block_bmap
);
2992 static void end_bio_bh_io_sync(struct bio
*bio
)
2994 struct buffer_head
*bh
= bio
->bi_private
;
2996 if (unlikely(bio_flagged(bio
, BIO_QUIET
)))
2997 set_bit(BH_Quiet
, &bh
->b_state
);
2999 bh
->b_end_io(bh
, !bio
->bi_status
);
3004 * This allows us to do IO even on the odd last sectors
3005 * of a device, even if the block size is some multiple
3006 * of the physical sector size.
3008 * We'll just truncate the bio to the size of the device,
3009 * and clear the end of the buffer head manually.
3011 * Truly out-of-range accesses will turn into actual IO
3012 * errors, this only handles the "we need to be able to
3013 * do IO at the final sector" case.
3015 void guard_bio_eod(int op
, struct bio
*bio
)
3018 struct bio_vec
*bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
3019 unsigned truncated_bytes
;
3020 struct hd_struct
*part
;
3023 part
= __disk_get_part(bio
->bi_disk
, bio
->bi_partno
);
3025 maxsector
= part_nr_sects_read(part
);
3027 maxsector
= get_capacity(bio
->bi_disk
);
3034 * If the *whole* IO is past the end of the device,
3035 * let it through, and the IO layer will turn it into
3038 if (unlikely(bio
->bi_iter
.bi_sector
>= maxsector
))
3041 maxsector
-= bio
->bi_iter
.bi_sector
;
3042 if (likely((bio
->bi_iter
.bi_size
>> 9) <= maxsector
))
3045 /* Uhhuh. We've got a bio that straddles the device size! */
3046 truncated_bytes
= bio
->bi_iter
.bi_size
- (maxsector
<< 9);
3048 /* Truncate the bio.. */
3049 bio
->bi_iter
.bi_size
-= truncated_bytes
;
3050 bvec
->bv_len
-= truncated_bytes
;
3052 /* ..and clear the end of the buffer for reads */
3053 if (op
== REQ_OP_READ
) {
3054 zero_user(bvec
->bv_page
, bvec
->bv_offset
+ bvec
->bv_len
,
3059 static int submit_bh_wbc(int op
, int op_flags
, struct buffer_head
*bh
,
3060 enum rw_hint write_hint
, struct writeback_control
*wbc
)
3064 BUG_ON(!buffer_locked(bh
));
3065 BUG_ON(!buffer_mapped(bh
));
3066 BUG_ON(!bh
->b_end_io
);
3067 BUG_ON(buffer_delay(bh
));
3068 BUG_ON(buffer_unwritten(bh
));
3071 * Only clear out a write error when rewriting
3073 if (test_set_buffer_req(bh
) && (op
== REQ_OP_WRITE
))
3074 clear_buffer_write_io_error(bh
);
3077 * from here on down, it's all bio -- do the initial mapping,
3078 * submit_bio -> generic_make_request may further map this bio around
3080 bio
= bio_alloc(GFP_NOIO
, 1);
3083 wbc_init_bio(wbc
, bio
);
3084 wbc_account_io(wbc
, bh
->b_page
, bh
->b_size
);
3087 bio
->bi_iter
.bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
3088 bio_set_dev(bio
, bh
->b_bdev
);
3089 bio
->bi_write_hint
= write_hint
;
3091 bio_add_page(bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
3092 BUG_ON(bio
->bi_iter
.bi_size
!= bh
->b_size
);
3094 bio
->bi_end_io
= end_bio_bh_io_sync
;
3095 bio
->bi_private
= bh
;
3097 /* Take care of bh's that straddle the end of the device */
3098 guard_bio_eod(op
, bio
);
3100 if (buffer_meta(bh
))
3101 op_flags
|= REQ_META
;
3102 if (buffer_prio(bh
))
3103 op_flags
|= REQ_PRIO
;
3104 bio_set_op_attrs(bio
, op
, op_flags
);
3110 int submit_bh(int op
, int op_flags
, struct buffer_head
*bh
)
3112 return submit_bh_wbc(op
, op_flags
, bh
, 0, NULL
);
3114 EXPORT_SYMBOL(submit_bh
);
3117 * ll_rw_block: low-level access to block devices (DEPRECATED)
3118 * @op: whether to %READ or %WRITE
3119 * @op_flags: req_flag_bits
3120 * @nr: number of &struct buffer_heads in the array
3121 * @bhs: array of pointers to &struct buffer_head
3123 * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
3124 * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE.
3125 * @op_flags contains flags modifying the detailed I/O behavior, most notably
3128 * This function drops any buffer that it cannot get a lock on (with the
3129 * BH_Lock state bit), any buffer that appears to be clean when doing a write
3130 * request, and any buffer that appears to be up-to-date when doing read
3131 * request. Further it marks as clean buffers that are processed for
3132 * writing (the buffer cache won't assume that they are actually clean
3133 * until the buffer gets unlocked).
3135 * ll_rw_block sets b_end_io to simple completion handler that marks
3136 * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
3139 * All of the buffers must be for the same device, and must also be a
3140 * multiple of the current approved size for the device.
3142 void ll_rw_block(int op
, int op_flags
, int nr
, struct buffer_head
*bhs
[])
3146 for (i
= 0; i
< nr
; i
++) {
3147 struct buffer_head
*bh
= bhs
[i
];
3149 if (!trylock_buffer(bh
))
3152 if (test_clear_buffer_dirty(bh
)) {
3153 bh
->b_end_io
= end_buffer_write_sync
;
3155 submit_bh(op
, op_flags
, bh
);
3159 if (!buffer_uptodate(bh
)) {
3160 bh
->b_end_io
= end_buffer_read_sync
;
3162 submit_bh(op
, op_flags
, bh
);
3169 EXPORT_SYMBOL(ll_rw_block
);
3171 void write_dirty_buffer(struct buffer_head
*bh
, int op_flags
)
3174 if (!test_clear_buffer_dirty(bh
)) {
3178 bh
->b_end_io
= end_buffer_write_sync
;
3180 submit_bh(REQ_OP_WRITE
, op_flags
, bh
);
3182 EXPORT_SYMBOL(write_dirty_buffer
);
3185 * For a data-integrity writeout, we need to wait upon any in-progress I/O
3186 * and then start new I/O and then wait upon it. The caller must have a ref on
3189 int __sync_dirty_buffer(struct buffer_head
*bh
, int op_flags
)
3193 WARN_ON(atomic_read(&bh
->b_count
) < 1);
3195 if (test_clear_buffer_dirty(bh
)) {
3197 bh
->b_end_io
= end_buffer_write_sync
;
3198 ret
= submit_bh(REQ_OP_WRITE
, op_flags
, bh
);
3200 if (!ret
&& !buffer_uptodate(bh
))
3207 EXPORT_SYMBOL(__sync_dirty_buffer
);
3209 int sync_dirty_buffer(struct buffer_head
*bh
)
3211 return __sync_dirty_buffer(bh
, REQ_SYNC
);
3213 EXPORT_SYMBOL(sync_dirty_buffer
);
3216 * try_to_free_buffers() checks if all the buffers on this particular page
3217 * are unused, and releases them if so.
3219 * Exclusion against try_to_free_buffers may be obtained by either
3220 * locking the page or by holding its mapping's private_lock.
3222 * If the page is dirty but all the buffers are clean then we need to
3223 * be sure to mark the page clean as well. This is because the page
3224 * may be against a block device, and a later reattachment of buffers
3225 * to a dirty page will set *all* buffers dirty. Which would corrupt
3226 * filesystem data on the same device.
3228 * The same applies to regular filesystem pages: if all the buffers are
3229 * clean then we set the page clean and proceed. To do that, we require
3230 * total exclusion from __set_page_dirty_buffers(). That is obtained with
3233 * try_to_free_buffers() is non-blocking.
3235 static inline int buffer_busy(struct buffer_head
*bh
)
3237 return atomic_read(&bh
->b_count
) |
3238 (bh
->b_state
& ((1 << BH_Dirty
) | (1 << BH_Lock
)));
3242 drop_buffers(struct page
*page
, struct buffer_head
**buffers_to_free
)
3244 struct buffer_head
*head
= page_buffers(page
);
3245 struct buffer_head
*bh
;
3249 if (buffer_busy(bh
))
3251 bh
= bh
->b_this_page
;
3252 } while (bh
!= head
);
3255 struct buffer_head
*next
= bh
->b_this_page
;
3257 if (bh
->b_assoc_map
)
3258 __remove_assoc_queue(bh
);
3260 } while (bh
!= head
);
3261 *buffers_to_free
= head
;
3262 __clear_page_buffers(page
);
3268 int try_to_free_buffers(struct page
*page
)
3270 struct address_space
* const mapping
= page
->mapping
;
3271 struct buffer_head
*buffers_to_free
= NULL
;
3274 BUG_ON(!PageLocked(page
));
3275 if (PageWriteback(page
))
3278 if (mapping
== NULL
) { /* can this still happen? */
3279 ret
= drop_buffers(page
, &buffers_to_free
);
3283 spin_lock(&mapping
->private_lock
);
3284 ret
= drop_buffers(page
, &buffers_to_free
);
3287 * If the filesystem writes its buffers by hand (eg ext3)
3288 * then we can have clean buffers against a dirty page. We
3289 * clean the page here; otherwise the VM will never notice
3290 * that the filesystem did any IO at all.
3292 * Also, during truncate, discard_buffer will have marked all
3293 * the page's buffers clean. We discover that here and clean
3296 * private_lock must be held over this entire operation in order
3297 * to synchronise against __set_page_dirty_buffers and prevent the
3298 * dirty bit from being lost.
3301 cancel_dirty_page(page
);
3302 spin_unlock(&mapping
->private_lock
);
3304 if (buffers_to_free
) {
3305 struct buffer_head
*bh
= buffers_to_free
;
3308 struct buffer_head
*next
= bh
->b_this_page
;
3309 free_buffer_head(bh
);
3311 } while (bh
!= buffers_to_free
);
3315 EXPORT_SYMBOL(try_to_free_buffers
);
3318 * There are no bdflush tunables left. But distributions are
3319 * still running obsolete flush daemons, so we terminate them here.
3321 * Use of bdflush() is deprecated and will be removed in a future kernel.
3322 * The `flush-X' kernel threads fully replace bdflush daemons and this call.
3324 SYSCALL_DEFINE2(bdflush
, int, func
, long, data
)
3326 static int msg_count
;
3328 if (!capable(CAP_SYS_ADMIN
))
3331 if (msg_count
< 5) {
3334 "warning: process `%s' used the obsolete bdflush"
3335 " system call\n", current
->comm
);
3336 printk(KERN_INFO
"Fix your initscripts?\n");
3345 * Buffer-head allocation
3347 static struct kmem_cache
*bh_cachep __read_mostly
;
3350 * Once the number of bh's in the machine exceeds this level, we start
3351 * stripping them in writeback.
3353 static unsigned long max_buffer_heads
;
3355 int buffer_heads_over_limit
;
3357 struct bh_accounting
{
3358 int nr
; /* Number of live bh's */
3359 int ratelimit
; /* Limit cacheline bouncing */
3362 static DEFINE_PER_CPU(struct bh_accounting
, bh_accounting
) = {0, 0};
3364 static void recalc_bh_state(void)
3369 if (__this_cpu_inc_return(bh_accounting
.ratelimit
) - 1 < 4096)
3371 __this_cpu_write(bh_accounting
.ratelimit
, 0);
3372 for_each_online_cpu(i
)
3373 tot
+= per_cpu(bh_accounting
, i
).nr
;
3374 buffer_heads_over_limit
= (tot
> max_buffer_heads
);
3377 struct buffer_head
*alloc_buffer_head(gfp_t gfp_flags
)
3379 struct buffer_head
*ret
= kmem_cache_zalloc(bh_cachep
, gfp_flags
);
3381 INIT_LIST_HEAD(&ret
->b_assoc_buffers
);
3383 __this_cpu_inc(bh_accounting
.nr
);
3389 EXPORT_SYMBOL(alloc_buffer_head
);
3391 void free_buffer_head(struct buffer_head
*bh
)
3393 BUG_ON(!list_empty(&bh
->b_assoc_buffers
));
3394 kmem_cache_free(bh_cachep
, bh
);
3396 __this_cpu_dec(bh_accounting
.nr
);
3400 EXPORT_SYMBOL(free_buffer_head
);
3402 static int buffer_exit_cpu_dead(unsigned int cpu
)
3405 struct bh_lru
*b
= &per_cpu(bh_lrus
, cpu
);
3407 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
3411 this_cpu_add(bh_accounting
.nr
, per_cpu(bh_accounting
, cpu
).nr
);
3412 per_cpu(bh_accounting
, cpu
).nr
= 0;
3417 * bh_uptodate_or_lock - Test whether the buffer is uptodate
3418 * @bh: struct buffer_head
3420 * Return true if the buffer is up-to-date and false,
3421 * with the buffer locked, if not.
3423 int bh_uptodate_or_lock(struct buffer_head
*bh
)
3425 if (!buffer_uptodate(bh
)) {
3427 if (!buffer_uptodate(bh
))
3433 EXPORT_SYMBOL(bh_uptodate_or_lock
);
3436 * bh_submit_read - Submit a locked buffer for reading
3437 * @bh: struct buffer_head
3439 * Returns zero on success and -EIO on error.
3441 int bh_submit_read(struct buffer_head
*bh
)
3443 BUG_ON(!buffer_locked(bh
));
3445 if (buffer_uptodate(bh
)) {
3451 bh
->b_end_io
= end_buffer_read_sync
;
3452 submit_bh(REQ_OP_READ
, 0, bh
);
3454 if (buffer_uptodate(bh
))
3458 EXPORT_SYMBOL(bh_submit_read
);
3461 * Seek for SEEK_DATA / SEEK_HOLE within @page, starting at @lastoff.
3463 * Returns the offset within the file on success, and -ENOENT otherwise.
3466 page_seek_hole_data(struct page
*page
, loff_t lastoff
, int whence
)
3468 loff_t offset
= page_offset(page
);
3469 struct buffer_head
*bh
, *head
;
3470 bool seek_data
= whence
== SEEK_DATA
;
3472 if (lastoff
< offset
)
3475 bh
= head
= page_buffers(page
);
3477 offset
+= bh
->b_size
;
3478 if (lastoff
>= offset
)
3482 * Unwritten extents that have data in the page cache covering
3483 * them can be identified by the BH_Unwritten state flag.
3484 * Pages with multiple buffers might have a mix of holes, data
3485 * and unwritten extents - any buffer with valid data in it
3486 * should have BH_Uptodate flag set on it.
3489 if ((buffer_unwritten(bh
) || buffer_uptodate(bh
)) == seek_data
)
3493 } while ((bh
= bh
->b_this_page
) != head
);
3498 * Seek for SEEK_DATA / SEEK_HOLE in the page cache.
3500 * Within unwritten extents, the page cache determines which parts are holes
3501 * and which are data: unwritten and uptodate buffer heads count as data;
3502 * everything else counts as a hole.
3504 * Returns the resulting offset on successs, and -ENOENT otherwise.
3507 page_cache_seek_hole_data(struct inode
*inode
, loff_t offset
, loff_t length
,
3510 pgoff_t index
= offset
>> PAGE_SHIFT
;
3511 pgoff_t end
= DIV_ROUND_UP(offset
+ length
, PAGE_SIZE
);
3512 loff_t lastoff
= offset
;
3513 struct pagevec pvec
;
3518 pagevec_init(&pvec
);
3521 unsigned nr_pages
, i
;
3523 nr_pages
= pagevec_lookup_range(&pvec
, inode
->i_mapping
, &index
,
3528 for (i
= 0; i
< nr_pages
; i
++) {
3529 struct page
*page
= pvec
.pages
[i
];
3532 * At this point, the page may be truncated or
3533 * invalidated (changing page->mapping to NULL), or
3534 * even swizzled back from swapper_space to tmpfs file
3535 * mapping. However, page->index will not change
3536 * because we have a reference on the page.
3538 * If current page offset is beyond where we've ended,
3539 * we've found a hole.
3541 if (whence
== SEEK_HOLE
&&
3542 lastoff
< page_offset(page
))
3546 if (likely(page
->mapping
== inode
->i_mapping
) &&
3547 page_has_buffers(page
)) {
3548 lastoff
= page_seek_hole_data(page
, lastoff
, whence
);
3555 lastoff
= page_offset(page
) + PAGE_SIZE
;
3557 pagevec_release(&pvec
);
3558 } while (index
< end
);
3560 /* When no page at lastoff and we are not done, we found a hole. */
3561 if (whence
!= SEEK_HOLE
)
3565 if (lastoff
< offset
+ length
)
3570 pagevec_release(&pvec
);
3574 void __init
buffer_init(void)
3576 unsigned long nrpages
;
3579 bh_cachep
= kmem_cache_create("buffer_head",
3580 sizeof(struct buffer_head
), 0,
3581 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
3586 * Limit the bh occupancy to 10% of ZONE_NORMAL
3588 nrpages
= (nr_free_buffer_pages() * 10) / 100;
3589 max_buffer_heads
= nrpages
* (PAGE_SIZE
/ sizeof(struct buffer_head
));
3590 ret
= cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD
, "fs/buffer:dead",
3591 NULL
, buffer_exit_cpu_dead
);