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
;
212 index
= block
>> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
213 page
= find_get_page_flags(bd_mapping
, index
, FGP_ACCESSED
);
217 spin_lock(&bd_mapping
->private_lock
);
218 if (!page_has_buffers(page
))
220 head
= page_buffers(page
);
223 if (!buffer_mapped(bh
))
225 else if (bh
->b_blocknr
== block
) {
230 bh
= bh
->b_this_page
;
231 } while (bh
!= head
);
233 /* we might be here because some of the buffers on this page are
234 * not mapped. This is due to various races between
235 * file io on the block device and getblk. It gets dealt with
236 * elsewhere, don't buffer_error if we had some unmapped buffers
239 printk("__find_get_block_slow() failed. "
240 "block=%llu, b_blocknr=%llu\n",
241 (unsigned long long)block
,
242 (unsigned long long)bh
->b_blocknr
);
243 printk("b_state=0x%08lx, b_size=%zu\n",
244 bh
->b_state
, bh
->b_size
);
245 printk("device %pg blocksize: %d\n", bdev
,
246 1 << bd_inode
->i_blkbits
);
249 spin_unlock(&bd_mapping
->private_lock
);
256 * Kick the writeback threads then try to free up some ZONE_NORMAL memory.
258 static void free_more_memory(void)
263 wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM
);
266 for_each_online_node(nid
) {
268 z
= first_zones_zonelist(node_zonelist(nid
, GFP_NOFS
),
269 gfp_zone(GFP_NOFS
), NULL
);
271 try_to_free_pages(node_zonelist(nid
, GFP_NOFS
), 0,
277 * I/O completion handler for block_read_full_page() - pages
278 * which come unlocked at the end of I/O.
280 static void end_buffer_async_read(struct buffer_head
*bh
, int uptodate
)
283 struct buffer_head
*first
;
284 struct buffer_head
*tmp
;
286 int page_uptodate
= 1;
288 BUG_ON(!buffer_async_read(bh
));
292 set_buffer_uptodate(bh
);
294 clear_buffer_uptodate(bh
);
295 buffer_io_error(bh
, ", async page read");
300 * Be _very_ careful from here on. Bad things can happen if
301 * two buffer heads end IO at almost the same time and both
302 * decide that the page is now completely done.
304 first
= page_buffers(page
);
305 local_irq_save(flags
);
306 bit_spin_lock(BH_Uptodate_Lock
, &first
->b_state
);
307 clear_buffer_async_read(bh
);
311 if (!buffer_uptodate(tmp
))
313 if (buffer_async_read(tmp
)) {
314 BUG_ON(!buffer_locked(tmp
));
317 tmp
= tmp
->b_this_page
;
319 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
320 local_irq_restore(flags
);
323 * If none of the buffers had errors and they are all
324 * uptodate then we can set the page uptodate.
326 if (page_uptodate
&& !PageError(page
))
327 SetPageUptodate(page
);
332 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
333 local_irq_restore(flags
);
338 * Completion handler for block_write_full_page() - pages which are unlocked
339 * during I/O, and which have PageWriteback cleared upon I/O completion.
341 void end_buffer_async_write(struct buffer_head
*bh
, int uptodate
)
344 struct buffer_head
*first
;
345 struct buffer_head
*tmp
;
348 BUG_ON(!buffer_async_write(bh
));
352 set_buffer_uptodate(bh
);
354 buffer_io_error(bh
, ", lost async page write");
355 mark_buffer_write_io_error(bh
);
356 clear_buffer_uptodate(bh
);
360 first
= page_buffers(page
);
361 local_irq_save(flags
);
362 bit_spin_lock(BH_Uptodate_Lock
, &first
->b_state
);
364 clear_buffer_async_write(bh
);
366 tmp
= bh
->b_this_page
;
368 if (buffer_async_write(tmp
)) {
369 BUG_ON(!buffer_locked(tmp
));
372 tmp
= tmp
->b_this_page
;
374 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
375 local_irq_restore(flags
);
376 end_page_writeback(page
);
380 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
381 local_irq_restore(flags
);
384 EXPORT_SYMBOL(end_buffer_async_write
);
387 * If a page's buffers are under async readin (end_buffer_async_read
388 * completion) then there is a possibility that another thread of
389 * control could lock one of the buffers after it has completed
390 * but while some of the other buffers have not completed. This
391 * locked buffer would confuse end_buffer_async_read() into not unlocking
392 * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
393 * that this buffer is not under async I/O.
395 * The page comes unlocked when it has no locked buffer_async buffers
398 * PageLocked prevents anyone starting new async I/O reads any of
401 * PageWriteback is used to prevent simultaneous writeout of the same
404 * PageLocked prevents anyone from starting writeback of a page which is
405 * under read I/O (PageWriteback is only ever set against a locked page).
407 static void mark_buffer_async_read(struct buffer_head
*bh
)
409 bh
->b_end_io
= end_buffer_async_read
;
410 set_buffer_async_read(bh
);
413 static void mark_buffer_async_write_endio(struct buffer_head
*bh
,
414 bh_end_io_t
*handler
)
416 bh
->b_end_io
= handler
;
417 set_buffer_async_write(bh
);
420 void mark_buffer_async_write(struct buffer_head
*bh
)
422 mark_buffer_async_write_endio(bh
, end_buffer_async_write
);
424 EXPORT_SYMBOL(mark_buffer_async_write
);
428 * fs/buffer.c contains helper functions for buffer-backed address space's
429 * fsync functions. A common requirement for buffer-based filesystems is
430 * that certain data from the backing blockdev needs to be written out for
431 * a successful fsync(). For example, ext2 indirect blocks need to be
432 * written back and waited upon before fsync() returns.
434 * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
435 * inode_has_buffers() and invalidate_inode_buffers() are provided for the
436 * management of a list of dependent buffers at ->i_mapping->private_list.
438 * Locking is a little subtle: try_to_free_buffers() will remove buffers
439 * from their controlling inode's queue when they are being freed. But
440 * try_to_free_buffers() will be operating against the *blockdev* mapping
441 * at the time, not against the S_ISREG file which depends on those buffers.
442 * So the locking for private_list is via the private_lock in the address_space
443 * which backs the buffers. Which is different from the address_space
444 * against which the buffers are listed. So for a particular address_space,
445 * mapping->private_lock does *not* protect mapping->private_list! In fact,
446 * mapping->private_list will always be protected by the backing blockdev's
449 * Which introduces a requirement: all buffers on an address_space's
450 * ->private_list must be from the same address_space: the blockdev's.
452 * address_spaces which do not place buffers at ->private_list via these
453 * utility functions are free to use private_lock and private_list for
454 * whatever they want. The only requirement is that list_empty(private_list)
455 * be true at clear_inode() time.
457 * FIXME: clear_inode should not call invalidate_inode_buffers(). The
458 * filesystems should do that. invalidate_inode_buffers() should just go
459 * BUG_ON(!list_empty).
461 * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
462 * take an address_space, not an inode. And it should be called
463 * mark_buffer_dirty_fsync() to clearly define why those buffers are being
466 * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
467 * list if it is already on a list. Because if the buffer is on a list,
468 * it *must* already be on the right one. If not, the filesystem is being
469 * silly. This will save a ton of locking. But first we have to ensure
470 * that buffers are taken *off* the old inode's list when they are freed
471 * (presumably in truncate). That requires careful auditing of all
472 * filesystems (do it inside bforget()). It could also be done by bringing
477 * The buffer's backing address_space's private_lock must be held
479 static void __remove_assoc_queue(struct buffer_head
*bh
)
481 list_del_init(&bh
->b_assoc_buffers
);
482 WARN_ON(!bh
->b_assoc_map
);
483 bh
->b_assoc_map
= NULL
;
486 int inode_has_buffers(struct inode
*inode
)
488 return !list_empty(&inode
->i_data
.private_list
);
492 * osync is designed to support O_SYNC io. It waits synchronously for
493 * all already-submitted IO to complete, but does not queue any new
494 * writes to the disk.
496 * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
497 * you dirty the buffers, and then use osync_inode_buffers to wait for
498 * completion. Any other dirty buffers which are not yet queued for
499 * write will not be flushed to disk by the osync.
501 static int osync_buffers_list(spinlock_t
*lock
, struct list_head
*list
)
503 struct buffer_head
*bh
;
509 list_for_each_prev(p
, list
) {
511 if (buffer_locked(bh
)) {
515 if (!buffer_uptodate(bh
))
526 static void do_thaw_one(struct super_block
*sb
, void *unused
)
528 while (sb
->s_bdev
&& !thaw_bdev(sb
->s_bdev
, sb
))
529 printk(KERN_WARNING
"Emergency Thaw on %pg\n", sb
->s_bdev
);
532 static void do_thaw_all(struct work_struct
*work
)
534 iterate_supers(do_thaw_one
, NULL
);
536 printk(KERN_WARNING
"Emergency Thaw complete\n");
540 * emergency_thaw_all -- forcibly thaw every frozen filesystem
542 * Used for emergency unfreeze of all filesystems via SysRq
544 void emergency_thaw_all(void)
546 struct work_struct
*work
;
548 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
550 INIT_WORK(work
, do_thaw_all
);
556 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
557 * @mapping: the mapping which wants those buffers written
559 * Starts I/O against the buffers at mapping->private_list, and waits upon
562 * Basically, this is a convenience function for fsync().
563 * @mapping is a file or directory which needs those buffers to be written for
564 * a successful fsync().
566 int sync_mapping_buffers(struct address_space
*mapping
)
568 struct address_space
*buffer_mapping
= mapping
->private_data
;
570 if (buffer_mapping
== NULL
|| list_empty(&mapping
->private_list
))
573 return fsync_buffers_list(&buffer_mapping
->private_lock
,
574 &mapping
->private_list
);
576 EXPORT_SYMBOL(sync_mapping_buffers
);
579 * Called when we've recently written block `bblock', and it is known that
580 * `bblock' was for a buffer_boundary() buffer. This means that the block at
581 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
582 * dirty, schedule it for IO. So that indirects merge nicely with their data.
584 void write_boundary_block(struct block_device
*bdev
,
585 sector_t bblock
, unsigned blocksize
)
587 struct buffer_head
*bh
= __find_get_block(bdev
, bblock
+ 1, blocksize
);
589 if (buffer_dirty(bh
))
590 ll_rw_block(REQ_OP_WRITE
, 0, 1, &bh
);
595 void mark_buffer_dirty_inode(struct buffer_head
*bh
, struct inode
*inode
)
597 struct address_space
*mapping
= inode
->i_mapping
;
598 struct address_space
*buffer_mapping
= bh
->b_page
->mapping
;
600 mark_buffer_dirty(bh
);
601 if (!mapping
->private_data
) {
602 mapping
->private_data
= buffer_mapping
;
604 BUG_ON(mapping
->private_data
!= buffer_mapping
);
606 if (!bh
->b_assoc_map
) {
607 spin_lock(&buffer_mapping
->private_lock
);
608 list_move_tail(&bh
->b_assoc_buffers
,
609 &mapping
->private_list
);
610 bh
->b_assoc_map
= mapping
;
611 spin_unlock(&buffer_mapping
->private_lock
);
614 EXPORT_SYMBOL(mark_buffer_dirty_inode
);
617 * Mark the page dirty, and set it dirty in the radix tree, and mark the inode
620 * If warn is true, then emit a warning if the page is not uptodate and has
621 * not been truncated.
623 * The caller must hold lock_page_memcg().
625 static void __set_page_dirty(struct page
*page
, struct address_space
*mapping
,
630 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
631 if (page
->mapping
) { /* Race with truncate? */
632 WARN_ON_ONCE(warn
&& !PageUptodate(page
));
633 account_page_dirtied(page
, mapping
);
634 radix_tree_tag_set(&mapping
->page_tree
,
635 page_index(page
), PAGECACHE_TAG_DIRTY
);
637 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
641 * Add a page to the dirty page list.
643 * It is a sad fact of life that this function is called from several places
644 * deeply under spinlocking. It may not sleep.
646 * If the page has buffers, the uptodate buffers are set dirty, to preserve
647 * dirty-state coherency between the page and the buffers. It the page does
648 * not have buffers then when they are later attached they will all be set
651 * The buffers are dirtied before the page is dirtied. There's a small race
652 * window in which a writepage caller may see the page cleanness but not the
653 * buffer dirtiness. That's fine. If this code were to set the page dirty
654 * before the buffers, a concurrent writepage caller could clear the page dirty
655 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
656 * page on the dirty page list.
658 * We use private_lock to lock against try_to_free_buffers while using the
659 * page's buffer list. Also use this to protect against clean buffers being
660 * added to the page after it was set dirty.
662 * FIXME: may need to call ->reservepage here as well. That's rather up to the
663 * address_space though.
665 int __set_page_dirty_buffers(struct page
*page
)
668 struct address_space
*mapping
= page_mapping(page
);
670 if (unlikely(!mapping
))
671 return !TestSetPageDirty(page
);
673 spin_lock(&mapping
->private_lock
);
674 if (page_has_buffers(page
)) {
675 struct buffer_head
*head
= page_buffers(page
);
676 struct buffer_head
*bh
= head
;
679 set_buffer_dirty(bh
);
680 bh
= bh
->b_this_page
;
681 } while (bh
!= head
);
684 * Lock out page->mem_cgroup migration to keep PageDirty
685 * synchronized with per-memcg dirty page counters.
687 lock_page_memcg(page
);
688 newly_dirty
= !TestSetPageDirty(page
);
689 spin_unlock(&mapping
->private_lock
);
692 __set_page_dirty(page
, mapping
, 1);
694 unlock_page_memcg(page
);
697 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
701 EXPORT_SYMBOL(__set_page_dirty_buffers
);
704 * Write out and wait upon a list of buffers.
706 * We have conflicting pressures: we want to make sure that all
707 * initially dirty buffers get waited on, but that any subsequently
708 * dirtied buffers don't. After all, we don't want fsync to last
709 * forever if somebody is actively writing to the file.
711 * Do this in two main stages: first we copy dirty buffers to a
712 * temporary inode list, queueing the writes as we go. Then we clean
713 * up, waiting for those writes to complete.
715 * During this second stage, any subsequent updates to the file may end
716 * up refiling the buffer on the original inode's dirty list again, so
717 * there is a chance we will end up with a buffer queued for write but
718 * not yet completed on that list. So, as a final cleanup we go through
719 * the osync code to catch these locked, dirty buffers without requeuing
720 * any newly dirty buffers for write.
722 static int fsync_buffers_list(spinlock_t
*lock
, struct list_head
*list
)
724 struct buffer_head
*bh
;
725 struct list_head tmp
;
726 struct address_space
*mapping
;
728 struct blk_plug plug
;
730 INIT_LIST_HEAD(&tmp
);
731 blk_start_plug(&plug
);
734 while (!list_empty(list
)) {
735 bh
= BH_ENTRY(list
->next
);
736 mapping
= bh
->b_assoc_map
;
737 __remove_assoc_queue(bh
);
738 /* Avoid race with mark_buffer_dirty_inode() which does
739 * a lockless check and we rely on seeing the dirty bit */
741 if (buffer_dirty(bh
) || buffer_locked(bh
)) {
742 list_add(&bh
->b_assoc_buffers
, &tmp
);
743 bh
->b_assoc_map
= mapping
;
744 if (buffer_dirty(bh
)) {
748 * Ensure any pending I/O completes so that
749 * write_dirty_buffer() actually writes the
750 * current contents - it is a noop if I/O is
751 * still in flight on potentially older
754 write_dirty_buffer(bh
, REQ_SYNC
);
757 * Kick off IO for the previous mapping. Note
758 * that we will not run the very last mapping,
759 * wait_on_buffer() will do that for us
760 * through sync_buffer().
769 blk_finish_plug(&plug
);
772 while (!list_empty(&tmp
)) {
773 bh
= BH_ENTRY(tmp
.prev
);
775 mapping
= bh
->b_assoc_map
;
776 __remove_assoc_queue(bh
);
777 /* Avoid race with mark_buffer_dirty_inode() which does
778 * a lockless check and we rely on seeing the dirty bit */
780 if (buffer_dirty(bh
)) {
781 list_add(&bh
->b_assoc_buffers
,
782 &mapping
->private_list
);
783 bh
->b_assoc_map
= mapping
;
787 if (!buffer_uptodate(bh
))
794 err2
= osync_buffers_list(lock
, list
);
802 * Invalidate any and all dirty buffers on a given inode. We are
803 * probably unmounting the fs, but that doesn't mean we have already
804 * done a sync(). Just drop the buffers from the inode list.
806 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
807 * assumes that all the buffers are against the blockdev. Not true
810 void invalidate_inode_buffers(struct inode
*inode
)
812 if (inode_has_buffers(inode
)) {
813 struct address_space
*mapping
= &inode
->i_data
;
814 struct list_head
*list
= &mapping
->private_list
;
815 struct address_space
*buffer_mapping
= mapping
->private_data
;
817 spin_lock(&buffer_mapping
->private_lock
);
818 while (!list_empty(list
))
819 __remove_assoc_queue(BH_ENTRY(list
->next
));
820 spin_unlock(&buffer_mapping
->private_lock
);
823 EXPORT_SYMBOL(invalidate_inode_buffers
);
826 * Remove any clean buffers from the inode's buffer list. This is called
827 * when we're trying to free the inode itself. Those buffers can pin it.
829 * Returns true if all buffers were removed.
831 int remove_inode_buffers(struct inode
*inode
)
835 if (inode_has_buffers(inode
)) {
836 struct address_space
*mapping
= &inode
->i_data
;
837 struct list_head
*list
= &mapping
->private_list
;
838 struct address_space
*buffer_mapping
= mapping
->private_data
;
840 spin_lock(&buffer_mapping
->private_lock
);
841 while (!list_empty(list
)) {
842 struct buffer_head
*bh
= BH_ENTRY(list
->next
);
843 if (buffer_dirty(bh
)) {
847 __remove_assoc_queue(bh
);
849 spin_unlock(&buffer_mapping
->private_lock
);
855 * Create the appropriate buffers when given a page for data area and
856 * the size of each buffer.. Use the bh->b_this_page linked list to
857 * follow the buffers created. Return NULL if unable to create more
860 * The retry flag is used to differentiate async IO (paging, swapping)
861 * which may not fail from ordinary buffer allocations.
863 struct buffer_head
*alloc_page_buffers(struct page
*page
, unsigned long size
,
866 struct buffer_head
*bh
, *head
;
867 gfp_t gfp
= GFP_NOFS
;
875 while ((offset
-= size
) >= 0) {
876 bh
= alloc_buffer_head(gfp
);
880 bh
->b_this_page
= head
;
886 /* Link the buffer to its page */
887 set_bh_page(bh
, page
, offset
);
891 * In case anything failed, we just free everything we got.
897 head
= head
->b_this_page
;
898 free_buffer_head(bh
);
904 EXPORT_SYMBOL_GPL(alloc_page_buffers
);
907 link_dev_buffers(struct page
*page
, struct buffer_head
*head
)
909 struct buffer_head
*bh
, *tail
;
914 bh
= bh
->b_this_page
;
916 tail
->b_this_page
= head
;
917 attach_page_buffers(page
, head
);
920 static sector_t
blkdev_max_block(struct block_device
*bdev
, unsigned int size
)
922 sector_t retval
= ~((sector_t
)0);
923 loff_t sz
= i_size_read(bdev
->bd_inode
);
926 unsigned int sizebits
= blksize_bits(size
);
927 retval
= (sz
>> sizebits
);
933 * Initialise the state of a blockdev page's buffers.
936 init_page_buffers(struct page
*page
, struct block_device
*bdev
,
937 sector_t block
, int size
)
939 struct buffer_head
*head
= page_buffers(page
);
940 struct buffer_head
*bh
= head
;
941 int uptodate
= PageUptodate(page
);
942 sector_t end_block
= blkdev_max_block(I_BDEV(bdev
->bd_inode
), size
);
945 if (!buffer_mapped(bh
)) {
946 init_buffer(bh
, NULL
, NULL
);
948 bh
->b_blocknr
= block
;
950 set_buffer_uptodate(bh
);
951 if (block
< end_block
)
952 set_buffer_mapped(bh
);
955 bh
= bh
->b_this_page
;
956 } while (bh
!= head
);
959 * Caller needs to validate requested block against end of device.
965 * Create the page-cache page that contains the requested block.
967 * This is used purely for blockdev mappings.
970 grow_dev_page(struct block_device
*bdev
, sector_t block
,
971 pgoff_t index
, int size
, int sizebits
, gfp_t gfp
)
973 struct inode
*inode
= bdev
->bd_inode
;
975 struct buffer_head
*bh
;
977 int ret
= 0; /* Will call free_more_memory() */
980 gfp_mask
= mapping_gfp_constraint(inode
->i_mapping
, ~__GFP_FS
) | gfp
;
983 * XXX: __getblk_slow() can not really deal with failure and
984 * will endlessly loop on improvised global reclaim. Prefer
985 * looping in the allocator rather than here, at least that
986 * code knows what it's doing.
988 gfp_mask
|= __GFP_NOFAIL
;
990 page
= find_or_create_page(inode
->i_mapping
, index
, gfp_mask
);
994 BUG_ON(!PageLocked(page
));
996 if (page_has_buffers(page
)) {
997 bh
= page_buffers(page
);
998 if (bh
->b_size
== size
) {
999 end_block
= init_page_buffers(page
, bdev
,
1000 (sector_t
)index
<< sizebits
,
1004 if (!try_to_free_buffers(page
))
1009 * Allocate some buffers for this page
1011 bh
= alloc_page_buffers(page
, size
, false);
1016 * Link the page to the buffers and initialise them. Take the
1017 * lock to be atomic wrt __find_get_block(), which does not
1018 * run under the page lock.
1020 spin_lock(&inode
->i_mapping
->private_lock
);
1021 link_dev_buffers(page
, bh
);
1022 end_block
= init_page_buffers(page
, bdev
, (sector_t
)index
<< sizebits
,
1024 spin_unlock(&inode
->i_mapping
->private_lock
);
1026 ret
= (block
< end_block
) ? 1 : -ENXIO
;
1034 * Create buffers for the specified block device block's page. If
1035 * that page was dirty, the buffers are set dirty also.
1038 grow_buffers(struct block_device
*bdev
, sector_t block
, int size
, gfp_t gfp
)
1046 } while ((size
<< sizebits
) < PAGE_SIZE
);
1048 index
= block
>> sizebits
;
1051 * Check for a block which wants to lie outside our maximum possible
1052 * pagecache index. (this comparison is done using sector_t types).
1054 if (unlikely(index
!= block
>> sizebits
)) {
1055 printk(KERN_ERR
"%s: requested out-of-range block %llu for "
1057 __func__
, (unsigned long long)block
,
1062 /* Create a page with the proper size buffers.. */
1063 return grow_dev_page(bdev
, block
, index
, size
, sizebits
, gfp
);
1066 static struct buffer_head
*
1067 __getblk_slow(struct block_device
*bdev
, sector_t block
,
1068 unsigned size
, gfp_t gfp
)
1070 /* Size must be multiple of hard sectorsize */
1071 if (unlikely(size
& (bdev_logical_block_size(bdev
)-1) ||
1072 (size
< 512 || size
> PAGE_SIZE
))) {
1073 printk(KERN_ERR
"getblk(): invalid block size %d requested\n",
1075 printk(KERN_ERR
"logical block size: %d\n",
1076 bdev_logical_block_size(bdev
));
1083 struct buffer_head
*bh
;
1086 bh
= __find_get_block(bdev
, block
, size
);
1090 ret
= grow_buffers(bdev
, block
, size
, gfp
);
1099 * The relationship between dirty buffers and dirty pages:
1101 * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1102 * the page is tagged dirty in its radix tree.
1104 * At all times, the dirtiness of the buffers represents the dirtiness of
1105 * subsections of the page. If the page has buffers, the page dirty bit is
1106 * merely a hint about the true dirty state.
1108 * When a page is set dirty in its entirety, all its buffers are marked dirty
1109 * (if the page has buffers).
1111 * When a buffer is marked dirty, its page is dirtied, but the page's other
1114 * Also. When blockdev buffers are explicitly read with bread(), they
1115 * individually become uptodate. But their backing page remains not
1116 * uptodate - even if all of its buffers are uptodate. A subsequent
1117 * block_read_full_page() against that page will discover all the uptodate
1118 * buffers, will set the page uptodate and will perform no I/O.
1122 * mark_buffer_dirty - mark a buffer_head as needing writeout
1123 * @bh: the buffer_head to mark dirty
1125 * mark_buffer_dirty() will set the dirty bit against the buffer, then set its
1126 * backing page dirty, then tag the page as dirty in its address_space's radix
1127 * tree and then attach the address_space's inode to its superblock's dirty
1130 * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
1131 * mapping->tree_lock and mapping->host->i_lock.
1133 void mark_buffer_dirty(struct buffer_head
*bh
)
1135 WARN_ON_ONCE(!buffer_uptodate(bh
));
1137 trace_block_dirty_buffer(bh
);
1140 * Very *carefully* optimize the it-is-already-dirty case.
1142 * Don't let the final "is it dirty" escape to before we
1143 * perhaps modified the buffer.
1145 if (buffer_dirty(bh
)) {
1147 if (buffer_dirty(bh
))
1151 if (!test_set_buffer_dirty(bh
)) {
1152 struct page
*page
= bh
->b_page
;
1153 struct address_space
*mapping
= NULL
;
1155 lock_page_memcg(page
);
1156 if (!TestSetPageDirty(page
)) {
1157 mapping
= page_mapping(page
);
1159 __set_page_dirty(page
, mapping
, 0);
1161 unlock_page_memcg(page
);
1163 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
1166 EXPORT_SYMBOL(mark_buffer_dirty
);
1168 void mark_buffer_write_io_error(struct buffer_head
*bh
)
1170 set_buffer_write_io_error(bh
);
1171 /* FIXME: do we need to set this in both places? */
1172 if (bh
->b_page
&& bh
->b_page
->mapping
)
1173 mapping_set_error(bh
->b_page
->mapping
, -EIO
);
1174 if (bh
->b_assoc_map
)
1175 mapping_set_error(bh
->b_assoc_map
, -EIO
);
1177 EXPORT_SYMBOL(mark_buffer_write_io_error
);
1180 * Decrement a buffer_head's reference count. If all buffers against a page
1181 * have zero reference count, are clean and unlocked, and if the page is clean
1182 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1183 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1184 * a page but it ends up not being freed, and buffers may later be reattached).
1186 void __brelse(struct buffer_head
* buf
)
1188 if (atomic_read(&buf
->b_count
)) {
1192 WARN(1, KERN_ERR
"VFS: brelse: Trying to free free buffer\n");
1194 EXPORT_SYMBOL(__brelse
);
1197 * bforget() is like brelse(), except it discards any
1198 * potentially dirty data.
1200 void __bforget(struct buffer_head
*bh
)
1202 clear_buffer_dirty(bh
);
1203 if (bh
->b_assoc_map
) {
1204 struct address_space
*buffer_mapping
= bh
->b_page
->mapping
;
1206 spin_lock(&buffer_mapping
->private_lock
);
1207 list_del_init(&bh
->b_assoc_buffers
);
1208 bh
->b_assoc_map
= NULL
;
1209 spin_unlock(&buffer_mapping
->private_lock
);
1213 EXPORT_SYMBOL(__bforget
);
1215 static struct buffer_head
*__bread_slow(struct buffer_head
*bh
)
1218 if (buffer_uptodate(bh
)) {
1223 bh
->b_end_io
= end_buffer_read_sync
;
1224 submit_bh(REQ_OP_READ
, 0, bh
);
1226 if (buffer_uptodate(bh
))
1234 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1235 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1236 * refcount elevated by one when they're in an LRU. A buffer can only appear
1237 * once in a particular CPU's LRU. A single buffer can be present in multiple
1238 * CPU's LRUs at the same time.
1240 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1241 * sb_find_get_block().
1243 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1244 * a local interrupt disable for that.
1247 #define BH_LRU_SIZE 16
1250 struct buffer_head
*bhs
[BH_LRU_SIZE
];
1253 static DEFINE_PER_CPU(struct bh_lru
, bh_lrus
) = {{ NULL
}};
1256 #define bh_lru_lock() local_irq_disable()
1257 #define bh_lru_unlock() local_irq_enable()
1259 #define bh_lru_lock() preempt_disable()
1260 #define bh_lru_unlock() preempt_enable()
1263 static inline void check_irqs_on(void)
1265 #ifdef irqs_disabled
1266 BUG_ON(irqs_disabled());
1271 * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is
1272 * inserted at the front, and the buffer_head at the back if any is evicted.
1273 * Or, if already in the LRU it is moved to the front.
1275 static void bh_lru_install(struct buffer_head
*bh
)
1277 struct buffer_head
*evictee
= bh
;
1284 b
= this_cpu_ptr(&bh_lrus
);
1285 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1286 swap(evictee
, b
->bhs
[i
]);
1287 if (evictee
== bh
) {
1299 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1301 static struct buffer_head
*
1302 lookup_bh_lru(struct block_device
*bdev
, sector_t block
, unsigned size
)
1304 struct buffer_head
*ret
= NULL
;
1309 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1310 struct buffer_head
*bh
= __this_cpu_read(bh_lrus
.bhs
[i
]);
1312 if (bh
&& bh
->b_blocknr
== block
&& bh
->b_bdev
== bdev
&&
1313 bh
->b_size
== size
) {
1316 __this_cpu_write(bh_lrus
.bhs
[i
],
1317 __this_cpu_read(bh_lrus
.bhs
[i
- 1]));
1320 __this_cpu_write(bh_lrus
.bhs
[0], bh
);
1332 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1333 * it in the LRU and mark it as accessed. If it is not present then return
1336 struct buffer_head
*
1337 __find_get_block(struct block_device
*bdev
, sector_t block
, unsigned size
)
1339 struct buffer_head
*bh
= lookup_bh_lru(bdev
, block
, size
);
1342 /* __find_get_block_slow will mark the page accessed */
1343 bh
= __find_get_block_slow(bdev
, block
);
1351 EXPORT_SYMBOL(__find_get_block
);
1354 * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
1355 * which corresponds to the passed block_device, block and size. The
1356 * returned buffer has its reference count incremented.
1358 * __getblk_gfp() will lock up the machine if grow_dev_page's
1359 * try_to_free_buffers() attempt is failing. FIXME, perhaps?
1361 struct buffer_head
*
1362 __getblk_gfp(struct block_device
*bdev
, sector_t block
,
1363 unsigned size
, gfp_t gfp
)
1365 struct buffer_head
*bh
= __find_get_block(bdev
, block
, size
);
1369 bh
= __getblk_slow(bdev
, block
, size
, gfp
);
1372 EXPORT_SYMBOL(__getblk_gfp
);
1375 * Do async read-ahead on a buffer..
1377 void __breadahead(struct block_device
*bdev
, sector_t block
, unsigned size
)
1379 struct buffer_head
*bh
= __getblk(bdev
, block
, size
);
1381 ll_rw_block(REQ_OP_READ
, REQ_RAHEAD
, 1, &bh
);
1385 EXPORT_SYMBOL(__breadahead
);
1388 * __bread_gfp() - reads a specified block and returns the bh
1389 * @bdev: the block_device to read from
1390 * @block: number of block
1391 * @size: size (in bytes) to read
1392 * @gfp: page allocation flag
1394 * Reads a specified block, and returns buffer head that contains it.
1395 * The page cache can be allocated from non-movable area
1396 * not to prevent page migration if you set gfp to zero.
1397 * It returns NULL if the block was unreadable.
1399 struct buffer_head
*
1400 __bread_gfp(struct block_device
*bdev
, sector_t block
,
1401 unsigned size
, gfp_t gfp
)
1403 struct buffer_head
*bh
= __getblk_gfp(bdev
, block
, size
, gfp
);
1405 if (likely(bh
) && !buffer_uptodate(bh
))
1406 bh
= __bread_slow(bh
);
1409 EXPORT_SYMBOL(__bread_gfp
);
1412 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1413 * This doesn't race because it runs in each cpu either in irq
1414 * or with preempt disabled.
1416 static void invalidate_bh_lru(void *arg
)
1418 struct bh_lru
*b
= &get_cpu_var(bh_lrus
);
1421 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1425 put_cpu_var(bh_lrus
);
1428 static bool has_bh_in_lru(int cpu
, void *dummy
)
1430 struct bh_lru
*b
= per_cpu_ptr(&bh_lrus
, cpu
);
1433 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1441 void invalidate_bh_lrus(void)
1443 on_each_cpu_cond(has_bh_in_lru
, invalidate_bh_lru
, NULL
, 1, GFP_KERNEL
);
1445 EXPORT_SYMBOL_GPL(invalidate_bh_lrus
);
1447 void set_bh_page(struct buffer_head
*bh
,
1448 struct page
*page
, unsigned long offset
)
1451 BUG_ON(offset
>= PAGE_SIZE
);
1452 if (PageHighMem(page
))
1454 * This catches illegal uses and preserves the offset:
1456 bh
->b_data
= (char *)(0 + offset
);
1458 bh
->b_data
= page_address(page
) + offset
;
1460 EXPORT_SYMBOL(set_bh_page
);
1463 * Called when truncating a buffer on a page completely.
1466 /* Bits that are cleared during an invalidate */
1467 #define BUFFER_FLAGS_DISCARD \
1468 (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1469 1 << BH_Delay | 1 << BH_Unwritten)
1471 static void discard_buffer(struct buffer_head
* bh
)
1473 unsigned long b_state
, b_state_old
;
1476 clear_buffer_dirty(bh
);
1478 b_state
= bh
->b_state
;
1480 b_state_old
= cmpxchg(&bh
->b_state
, b_state
,
1481 (b_state
& ~BUFFER_FLAGS_DISCARD
));
1482 if (b_state_old
== b_state
)
1484 b_state
= b_state_old
;
1490 * block_invalidatepage - invalidate part or all of a buffer-backed page
1492 * @page: the page which is affected
1493 * @offset: start of the range to invalidate
1494 * @length: length of the range to invalidate
1496 * block_invalidatepage() is called when all or part of the page has become
1497 * invalidated by a truncate operation.
1499 * block_invalidatepage() does not have to release all buffers, but it must
1500 * ensure that no dirty buffer is left outside @offset and that no I/O
1501 * is underway against any of the blocks which are outside the truncation
1502 * point. Because the caller is about to free (and possibly reuse) those
1505 void block_invalidatepage(struct page
*page
, unsigned int offset
,
1506 unsigned int length
)
1508 struct buffer_head
*head
, *bh
, *next
;
1509 unsigned int curr_off
= 0;
1510 unsigned int stop
= length
+ offset
;
1512 BUG_ON(!PageLocked(page
));
1513 if (!page_has_buffers(page
))
1517 * Check for overflow
1519 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1521 head
= page_buffers(page
);
1524 unsigned int next_off
= curr_off
+ bh
->b_size
;
1525 next
= bh
->b_this_page
;
1528 * Are we still fully in range ?
1530 if (next_off
> stop
)
1534 * is this block fully invalidated?
1536 if (offset
<= curr_off
)
1538 curr_off
= next_off
;
1540 } while (bh
!= head
);
1543 * We release buffers only if the entire page is being invalidated.
1544 * The get_block cached value has been unconditionally invalidated,
1545 * so real IO is not possible anymore.
1548 try_to_release_page(page
, 0);
1552 EXPORT_SYMBOL(block_invalidatepage
);
1556 * We attach and possibly dirty the buffers atomically wrt
1557 * __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1558 * is already excluded via the page lock.
1560 void create_empty_buffers(struct page
*page
,
1561 unsigned long blocksize
, unsigned long b_state
)
1563 struct buffer_head
*bh
, *head
, *tail
;
1565 head
= alloc_page_buffers(page
, blocksize
, true);
1568 bh
->b_state
|= b_state
;
1570 bh
= bh
->b_this_page
;
1572 tail
->b_this_page
= head
;
1574 spin_lock(&page
->mapping
->private_lock
);
1575 if (PageUptodate(page
) || PageDirty(page
)) {
1578 if (PageDirty(page
))
1579 set_buffer_dirty(bh
);
1580 if (PageUptodate(page
))
1581 set_buffer_uptodate(bh
);
1582 bh
= bh
->b_this_page
;
1583 } while (bh
!= head
);
1585 attach_page_buffers(page
, head
);
1586 spin_unlock(&page
->mapping
->private_lock
);
1588 EXPORT_SYMBOL(create_empty_buffers
);
1591 * clean_bdev_aliases: clean a range of buffers in block device
1592 * @bdev: Block device to clean buffers in
1593 * @block: Start of a range of blocks to clean
1594 * @len: Number of blocks to clean
1596 * We are taking a range of blocks for data and we don't want writeback of any
1597 * buffer-cache aliases starting from return from this function and until the
1598 * moment when something will explicitly mark the buffer dirty (hopefully that
1599 * will not happen until we will free that block ;-) We don't even need to mark
1600 * it not-uptodate - nobody can expect anything from a newly allocated buffer
1601 * anyway. We used to use unmap_buffer() for such invalidation, but that was
1602 * wrong. We definitely don't want to mark the alias unmapped, for example - it
1603 * would confuse anyone who might pick it with bread() afterwards...
1605 * Also.. Note that bforget() doesn't lock the buffer. So there can be
1606 * writeout I/O going on against recently-freed buffers. We don't wait on that
1607 * I/O in bforget() - it's more efficient to wait on the I/O only if we really
1608 * need to. That happens here.
1610 void clean_bdev_aliases(struct block_device
*bdev
, sector_t block
, sector_t len
)
1612 struct inode
*bd_inode
= bdev
->bd_inode
;
1613 struct address_space
*bd_mapping
= bd_inode
->i_mapping
;
1614 struct pagevec pvec
;
1615 pgoff_t index
= block
>> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
1618 struct buffer_head
*bh
;
1619 struct buffer_head
*head
;
1621 end
= (block
+ len
- 1) >> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
1622 pagevec_init(&pvec
, 0);
1623 while (pagevec_lookup_range(&pvec
, bd_mapping
, &index
, end
)) {
1624 count
= pagevec_count(&pvec
);
1625 for (i
= 0; i
< count
; i
++) {
1626 struct page
*page
= pvec
.pages
[i
];
1628 if (!page_has_buffers(page
))
1631 * We use page lock instead of bd_mapping->private_lock
1632 * to pin buffers here since we can afford to sleep and
1633 * it scales better than a global spinlock lock.
1636 /* Recheck when the page is locked which pins bhs */
1637 if (!page_has_buffers(page
))
1639 head
= page_buffers(page
);
1642 if (!buffer_mapped(bh
) || (bh
->b_blocknr
< block
))
1644 if (bh
->b_blocknr
>= block
+ len
)
1646 clear_buffer_dirty(bh
);
1648 clear_buffer_req(bh
);
1650 bh
= bh
->b_this_page
;
1651 } while (bh
!= head
);
1655 pagevec_release(&pvec
);
1657 /* End of range already reached? */
1658 if (index
> end
|| !index
)
1662 EXPORT_SYMBOL(clean_bdev_aliases
);
1665 * Size is a power-of-two in the range 512..PAGE_SIZE,
1666 * and the case we care about most is PAGE_SIZE.
1668 * So this *could* possibly be written with those
1669 * constraints in mind (relevant mostly if some
1670 * architecture has a slow bit-scan instruction)
1672 static inline int block_size_bits(unsigned int blocksize
)
1674 return ilog2(blocksize
);
1677 static struct buffer_head
*create_page_buffers(struct page
*page
, struct inode
*inode
, unsigned int b_state
)
1679 BUG_ON(!PageLocked(page
));
1681 if (!page_has_buffers(page
))
1682 create_empty_buffers(page
, 1 << ACCESS_ONCE(inode
->i_blkbits
), b_state
);
1683 return page_buffers(page
);
1687 * NOTE! All mapped/uptodate combinations are valid:
1689 * Mapped Uptodate Meaning
1691 * No No "unknown" - must do get_block()
1692 * No Yes "hole" - zero-filled
1693 * Yes No "allocated" - allocated on disk, not read in
1694 * Yes Yes "valid" - allocated and up-to-date in memory.
1696 * "Dirty" is valid only with the last case (mapped+uptodate).
1700 * While block_write_full_page is writing back the dirty buffers under
1701 * the page lock, whoever dirtied the buffers may decide to clean them
1702 * again at any time. We handle that by only looking at the buffer
1703 * state inside lock_buffer().
1705 * If block_write_full_page() is called for regular writeback
1706 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1707 * locked buffer. This only can happen if someone has written the buffer
1708 * directly, with submit_bh(). At the address_space level PageWriteback
1709 * prevents this contention from occurring.
1711 * If block_write_full_page() is called with wbc->sync_mode ==
1712 * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this
1713 * causes the writes to be flagged as synchronous writes.
1715 int __block_write_full_page(struct inode
*inode
, struct page
*page
,
1716 get_block_t
*get_block
, struct writeback_control
*wbc
,
1717 bh_end_io_t
*handler
)
1721 sector_t last_block
;
1722 struct buffer_head
*bh
, *head
;
1723 unsigned int blocksize
, bbits
;
1724 int nr_underway
= 0;
1725 int write_flags
= wbc_to_write_flags(wbc
);
1727 head
= create_page_buffers(page
, inode
,
1728 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
1731 * Be very careful. We have no exclusion from __set_page_dirty_buffers
1732 * here, and the (potentially unmapped) buffers may become dirty at
1733 * any time. If a buffer becomes dirty here after we've inspected it
1734 * then we just miss that fact, and the page stays dirty.
1736 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1737 * handle that here by just cleaning them.
1741 blocksize
= bh
->b_size
;
1742 bbits
= block_size_bits(blocksize
);
1744 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1745 last_block
= (i_size_read(inode
) - 1) >> bbits
;
1748 * Get all the dirty buffers mapped to disk addresses and
1749 * handle any aliases from the underlying blockdev's mapping.
1752 if (block
> last_block
) {
1754 * mapped buffers outside i_size will occur, because
1755 * this page can be outside i_size when there is a
1756 * truncate in progress.
1759 * The buffer was zeroed by block_write_full_page()
1761 clear_buffer_dirty(bh
);
1762 set_buffer_uptodate(bh
);
1763 } else if ((!buffer_mapped(bh
) || buffer_delay(bh
)) &&
1765 WARN_ON(bh
->b_size
!= blocksize
);
1766 err
= get_block(inode
, block
, bh
, 1);
1769 clear_buffer_delay(bh
);
1770 if (buffer_new(bh
)) {
1771 /* blockdev mappings never come here */
1772 clear_buffer_new(bh
);
1773 clean_bdev_bh_alias(bh
);
1776 bh
= bh
->b_this_page
;
1778 } while (bh
!= head
);
1781 if (!buffer_mapped(bh
))
1784 * If it's a fully non-blocking write attempt and we cannot
1785 * lock the buffer then redirty the page. Note that this can
1786 * potentially cause a busy-wait loop from writeback threads
1787 * and kswapd activity, but those code paths have their own
1788 * higher-level throttling.
1790 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
1792 } else if (!trylock_buffer(bh
)) {
1793 redirty_page_for_writepage(wbc
, page
);
1796 if (test_clear_buffer_dirty(bh
)) {
1797 mark_buffer_async_write_endio(bh
, handler
);
1801 } while ((bh
= bh
->b_this_page
) != head
);
1804 * The page and its buffers are protected by PageWriteback(), so we can
1805 * drop the bh refcounts early.
1807 BUG_ON(PageWriteback(page
));
1808 set_page_writeback(page
);
1811 struct buffer_head
*next
= bh
->b_this_page
;
1812 if (buffer_async_write(bh
)) {
1813 submit_bh_wbc(REQ_OP_WRITE
, write_flags
, bh
,
1814 inode
->i_write_hint
, wbc
);
1818 } while (bh
!= head
);
1823 if (nr_underway
== 0) {
1825 * The page was marked dirty, but the buffers were
1826 * clean. Someone wrote them back by hand with
1827 * ll_rw_block/submit_bh. A rare case.
1829 end_page_writeback(page
);
1832 * The page and buffer_heads can be released at any time from
1840 * ENOSPC, or some other error. We may already have added some
1841 * blocks to the file, so we need to write these out to avoid
1842 * exposing stale data.
1843 * The page is currently locked and not marked for writeback
1846 /* Recovery: lock and submit the mapped buffers */
1848 if (buffer_mapped(bh
) && buffer_dirty(bh
) &&
1849 !buffer_delay(bh
)) {
1851 mark_buffer_async_write_endio(bh
, handler
);
1854 * The buffer may have been set dirty during
1855 * attachment to a dirty page.
1857 clear_buffer_dirty(bh
);
1859 } while ((bh
= bh
->b_this_page
) != head
);
1861 BUG_ON(PageWriteback(page
));
1862 mapping_set_error(page
->mapping
, err
);
1863 set_page_writeback(page
);
1865 struct buffer_head
*next
= bh
->b_this_page
;
1866 if (buffer_async_write(bh
)) {
1867 clear_buffer_dirty(bh
);
1868 submit_bh_wbc(REQ_OP_WRITE
, write_flags
, bh
,
1869 inode
->i_write_hint
, wbc
);
1873 } while (bh
!= head
);
1877 EXPORT_SYMBOL(__block_write_full_page
);
1880 * If a page has any new buffers, zero them out here, and mark them uptodate
1881 * and dirty so they'll be written out (in order to prevent uninitialised
1882 * block data from leaking). And clear the new bit.
1884 void page_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1886 unsigned int block_start
, block_end
;
1887 struct buffer_head
*head
, *bh
;
1889 BUG_ON(!PageLocked(page
));
1890 if (!page_has_buffers(page
))
1893 bh
= head
= page_buffers(page
);
1896 block_end
= block_start
+ bh
->b_size
;
1898 if (buffer_new(bh
)) {
1899 if (block_end
> from
&& block_start
< to
) {
1900 if (!PageUptodate(page
)) {
1901 unsigned start
, size
;
1903 start
= max(from
, block_start
);
1904 size
= min(to
, block_end
) - start
;
1906 zero_user(page
, start
, size
);
1907 set_buffer_uptodate(bh
);
1910 clear_buffer_new(bh
);
1911 mark_buffer_dirty(bh
);
1915 block_start
= block_end
;
1916 bh
= bh
->b_this_page
;
1917 } while (bh
!= head
);
1919 EXPORT_SYMBOL(page_zero_new_buffers
);
1922 iomap_to_bh(struct inode
*inode
, sector_t block
, struct buffer_head
*bh
,
1923 struct iomap
*iomap
)
1925 loff_t offset
= block
<< inode
->i_blkbits
;
1927 bh
->b_bdev
= iomap
->bdev
;
1930 * Block points to offset in file we need to map, iomap contains
1931 * the offset at which the map starts. If the map ends before the
1932 * current block, then do not map the buffer and let the caller
1935 BUG_ON(offset
>= iomap
->offset
+ iomap
->length
);
1937 switch (iomap
->type
) {
1940 * If the buffer is not up to date or beyond the current EOF,
1941 * we need to mark it as new to ensure sub-block zeroing is
1942 * executed if necessary.
1944 if (!buffer_uptodate(bh
) ||
1945 (offset
>= i_size_read(inode
)))
1948 case IOMAP_DELALLOC
:
1949 if (!buffer_uptodate(bh
) ||
1950 (offset
>= i_size_read(inode
)))
1952 set_buffer_uptodate(bh
);
1953 set_buffer_mapped(bh
);
1954 set_buffer_delay(bh
);
1956 case IOMAP_UNWRITTEN
:
1958 * For unwritten regions, we always need to ensure that
1959 * sub-block writes cause the regions in the block we are not
1960 * writing to are zeroed. Set the buffer as new to ensure this.
1963 set_buffer_unwritten(bh
);
1966 if (offset
>= i_size_read(inode
))
1968 bh
->b_blocknr
= (iomap
->blkno
>> (inode
->i_blkbits
- 9)) +
1969 ((offset
- iomap
->offset
) >> inode
->i_blkbits
);
1970 set_buffer_mapped(bh
);
1975 int __block_write_begin_int(struct page
*page
, loff_t pos
, unsigned len
,
1976 get_block_t
*get_block
, struct iomap
*iomap
)
1978 unsigned from
= pos
& (PAGE_SIZE
- 1);
1979 unsigned to
= from
+ len
;
1980 struct inode
*inode
= page
->mapping
->host
;
1981 unsigned block_start
, block_end
;
1984 unsigned blocksize
, bbits
;
1985 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
=wait
;
1987 BUG_ON(!PageLocked(page
));
1988 BUG_ON(from
> PAGE_SIZE
);
1989 BUG_ON(to
> PAGE_SIZE
);
1992 head
= create_page_buffers(page
, inode
, 0);
1993 blocksize
= head
->b_size
;
1994 bbits
= block_size_bits(blocksize
);
1996 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1998 for(bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1999 block
++, block_start
=block_end
, bh
= bh
->b_this_page
) {
2000 block_end
= block_start
+ blocksize
;
2001 if (block_end
<= from
|| block_start
>= to
) {
2002 if (PageUptodate(page
)) {
2003 if (!buffer_uptodate(bh
))
2004 set_buffer_uptodate(bh
);
2009 clear_buffer_new(bh
);
2010 if (!buffer_mapped(bh
)) {
2011 WARN_ON(bh
->b_size
!= blocksize
);
2013 err
= get_block(inode
, block
, bh
, 1);
2017 iomap_to_bh(inode
, block
, bh
, iomap
);
2020 if (buffer_new(bh
)) {
2021 clean_bdev_bh_alias(bh
);
2022 if (PageUptodate(page
)) {
2023 clear_buffer_new(bh
);
2024 set_buffer_uptodate(bh
);
2025 mark_buffer_dirty(bh
);
2028 if (block_end
> to
|| block_start
< from
)
2029 zero_user_segments(page
,
2035 if (PageUptodate(page
)) {
2036 if (!buffer_uptodate(bh
))
2037 set_buffer_uptodate(bh
);
2040 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
2041 !buffer_unwritten(bh
) &&
2042 (block_start
< from
|| block_end
> to
)) {
2043 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
2048 * If we issued read requests - let them complete.
2050 while(wait_bh
> wait
) {
2051 wait_on_buffer(*--wait_bh
);
2052 if (!buffer_uptodate(*wait_bh
))
2056 page_zero_new_buffers(page
, from
, to
);
2060 int __block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
2061 get_block_t
*get_block
)
2063 return __block_write_begin_int(page
, pos
, len
, get_block
, NULL
);
2065 EXPORT_SYMBOL(__block_write_begin
);
2067 static int __block_commit_write(struct inode
*inode
, struct page
*page
,
2068 unsigned from
, unsigned to
)
2070 unsigned block_start
, block_end
;
2073 struct buffer_head
*bh
, *head
;
2075 bh
= head
= page_buffers(page
);
2076 blocksize
= bh
->b_size
;
2080 block_end
= block_start
+ blocksize
;
2081 if (block_end
<= from
|| block_start
>= to
) {
2082 if (!buffer_uptodate(bh
))
2085 set_buffer_uptodate(bh
);
2086 mark_buffer_dirty(bh
);
2088 clear_buffer_new(bh
);
2090 block_start
= block_end
;
2091 bh
= bh
->b_this_page
;
2092 } while (bh
!= head
);
2095 * If this is a partial write which happened to make all buffers
2096 * uptodate then we can optimize away a bogus readpage() for
2097 * the next read(). Here we 'discover' whether the page went
2098 * uptodate as a result of this (potentially partial) write.
2101 SetPageUptodate(page
);
2106 * block_write_begin takes care of the basic task of block allocation and
2107 * bringing partial write blocks uptodate first.
2109 * The filesystem needs to handle block truncation upon failure.
2111 int block_write_begin(struct address_space
*mapping
, loff_t pos
, unsigned len
,
2112 unsigned flags
, struct page
**pagep
, get_block_t
*get_block
)
2114 pgoff_t index
= pos
>> PAGE_SHIFT
;
2118 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2122 status
= __block_write_begin(page
, pos
, len
, get_block
);
2123 if (unlikely(status
)) {
2132 EXPORT_SYMBOL(block_write_begin
);
2134 int block_write_end(struct file
*file
, struct address_space
*mapping
,
2135 loff_t pos
, unsigned len
, unsigned copied
,
2136 struct page
*page
, void *fsdata
)
2138 struct inode
*inode
= mapping
->host
;
2141 start
= pos
& (PAGE_SIZE
- 1);
2143 if (unlikely(copied
< len
)) {
2145 * The buffers that were written will now be uptodate, so we
2146 * don't have to worry about a readpage reading them and
2147 * overwriting a partial write. However if we have encountered
2148 * a short write and only partially written into a buffer, it
2149 * will not be marked uptodate, so a readpage might come in and
2150 * destroy our partial write.
2152 * Do the simplest thing, and just treat any short write to a
2153 * non uptodate page as a zero-length write, and force the
2154 * caller to redo the whole thing.
2156 if (!PageUptodate(page
))
2159 page_zero_new_buffers(page
, start
+copied
, start
+len
);
2161 flush_dcache_page(page
);
2163 /* This could be a short (even 0-length) commit */
2164 __block_commit_write(inode
, page
, start
, start
+copied
);
2168 EXPORT_SYMBOL(block_write_end
);
2170 int generic_write_end(struct file
*file
, struct address_space
*mapping
,
2171 loff_t pos
, unsigned len
, unsigned copied
,
2172 struct page
*page
, void *fsdata
)
2174 struct inode
*inode
= mapping
->host
;
2175 loff_t old_size
= inode
->i_size
;
2176 int i_size_changed
= 0;
2178 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2181 * No need to use i_size_read() here, the i_size
2182 * cannot change under us because we hold i_mutex.
2184 * But it's important to update i_size while still holding page lock:
2185 * page writeout could otherwise come in and zero beyond i_size.
2187 if (pos
+copied
> inode
->i_size
) {
2188 i_size_write(inode
, pos
+copied
);
2196 pagecache_isize_extended(inode
, old_size
, pos
);
2198 * Don't mark the inode dirty under page lock. First, it unnecessarily
2199 * makes the holding time of page lock longer. Second, it forces lock
2200 * ordering of page lock and transaction start for journaling
2204 mark_inode_dirty(inode
);
2208 EXPORT_SYMBOL(generic_write_end
);
2211 * block_is_partially_uptodate checks whether buffers within a page are
2214 * Returns true if all buffers which correspond to a file portion
2215 * we want to read are uptodate.
2217 int block_is_partially_uptodate(struct page
*page
, unsigned long from
,
2218 unsigned long count
)
2220 unsigned block_start
, block_end
, blocksize
;
2222 struct buffer_head
*bh
, *head
;
2225 if (!page_has_buffers(page
))
2228 head
= page_buffers(page
);
2229 blocksize
= head
->b_size
;
2230 to
= min_t(unsigned, PAGE_SIZE
- from
, count
);
2232 if (from
< blocksize
&& to
> PAGE_SIZE
- blocksize
)
2238 block_end
= block_start
+ blocksize
;
2239 if (block_end
> from
&& block_start
< to
) {
2240 if (!buffer_uptodate(bh
)) {
2244 if (block_end
>= to
)
2247 block_start
= block_end
;
2248 bh
= bh
->b_this_page
;
2249 } while (bh
!= head
);
2253 EXPORT_SYMBOL(block_is_partially_uptodate
);
2256 * Generic "read page" function for block devices that have the normal
2257 * get_block functionality. This is most of the block device filesystems.
2258 * Reads the page asynchronously --- the unlock_buffer() and
2259 * set/clear_buffer_uptodate() functions propagate buffer state into the
2260 * page struct once IO has completed.
2262 int block_read_full_page(struct page
*page
, get_block_t
*get_block
)
2264 struct inode
*inode
= page
->mapping
->host
;
2265 sector_t iblock
, lblock
;
2266 struct buffer_head
*bh
, *head
, *arr
[MAX_BUF_PER_PAGE
];
2267 unsigned int blocksize
, bbits
;
2269 int fully_mapped
= 1;
2271 head
= create_page_buffers(page
, inode
, 0);
2272 blocksize
= head
->b_size
;
2273 bbits
= block_size_bits(blocksize
);
2275 iblock
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
2276 lblock
= (i_size_read(inode
)+blocksize
-1) >> bbits
;
2282 if (buffer_uptodate(bh
))
2285 if (!buffer_mapped(bh
)) {
2289 if (iblock
< lblock
) {
2290 WARN_ON(bh
->b_size
!= blocksize
);
2291 err
= get_block(inode
, iblock
, bh
, 0);
2295 if (!buffer_mapped(bh
)) {
2296 zero_user(page
, i
* blocksize
, blocksize
);
2298 set_buffer_uptodate(bh
);
2302 * get_block() might have updated the buffer
2305 if (buffer_uptodate(bh
))
2309 } while (i
++, iblock
++, (bh
= bh
->b_this_page
) != head
);
2312 SetPageMappedToDisk(page
);
2316 * All buffers are uptodate - we can set the page uptodate
2317 * as well. But not if get_block() returned an error.
2319 if (!PageError(page
))
2320 SetPageUptodate(page
);
2325 /* Stage two: lock the buffers */
2326 for (i
= 0; i
< nr
; i
++) {
2329 mark_buffer_async_read(bh
);
2333 * Stage 3: start the IO. Check for uptodateness
2334 * inside the buffer lock in case another process reading
2335 * the underlying blockdev brought it uptodate (the sct fix).
2337 for (i
= 0; i
< nr
; i
++) {
2339 if (buffer_uptodate(bh
))
2340 end_buffer_async_read(bh
, 1);
2342 submit_bh(REQ_OP_READ
, 0, bh
);
2346 EXPORT_SYMBOL(block_read_full_page
);
2348 /* utility function for filesystems that need to do work on expanding
2349 * truncates. Uses filesystem pagecache writes to allow the filesystem to
2350 * deal with the hole.
2352 int generic_cont_expand_simple(struct inode
*inode
, loff_t size
)
2354 struct address_space
*mapping
= inode
->i_mapping
;
2359 err
= inode_newsize_ok(inode
, size
);
2363 err
= pagecache_write_begin(NULL
, mapping
, size
, 0,
2364 AOP_FLAG_CONT_EXPAND
, &page
, &fsdata
);
2368 err
= pagecache_write_end(NULL
, mapping
, size
, 0, 0, page
, fsdata
);
2374 EXPORT_SYMBOL(generic_cont_expand_simple
);
2376 static int cont_expand_zero(struct file
*file
, struct address_space
*mapping
,
2377 loff_t pos
, loff_t
*bytes
)
2379 struct inode
*inode
= mapping
->host
;
2380 unsigned int blocksize
= i_blocksize(inode
);
2383 pgoff_t index
, curidx
;
2385 unsigned zerofrom
, offset
, len
;
2388 index
= pos
>> PAGE_SHIFT
;
2389 offset
= pos
& ~PAGE_MASK
;
2391 while (index
> (curidx
= (curpos
= *bytes
)>>PAGE_SHIFT
)) {
2392 zerofrom
= curpos
& ~PAGE_MASK
;
2393 if (zerofrom
& (blocksize
-1)) {
2394 *bytes
|= (blocksize
-1);
2397 len
= PAGE_SIZE
- zerofrom
;
2399 err
= pagecache_write_begin(file
, mapping
, curpos
, len
, 0,
2403 zero_user(page
, zerofrom
, len
);
2404 err
= pagecache_write_end(file
, mapping
, curpos
, len
, len
,
2411 balance_dirty_pages_ratelimited(mapping
);
2413 if (unlikely(fatal_signal_pending(current
))) {
2419 /* page covers the boundary, find the boundary offset */
2420 if (index
== curidx
) {
2421 zerofrom
= curpos
& ~PAGE_MASK
;
2422 /* if we will expand the thing last block will be filled */
2423 if (offset
<= zerofrom
) {
2426 if (zerofrom
& (blocksize
-1)) {
2427 *bytes
|= (blocksize
-1);
2430 len
= offset
- zerofrom
;
2432 err
= pagecache_write_begin(file
, mapping
, curpos
, len
, 0,
2436 zero_user(page
, zerofrom
, len
);
2437 err
= pagecache_write_end(file
, mapping
, curpos
, len
, len
,
2449 * For moronic filesystems that do not allow holes in file.
2450 * We may have to extend the file.
2452 int cont_write_begin(struct file
*file
, struct address_space
*mapping
,
2453 loff_t pos
, unsigned len
, unsigned flags
,
2454 struct page
**pagep
, void **fsdata
,
2455 get_block_t
*get_block
, loff_t
*bytes
)
2457 struct inode
*inode
= mapping
->host
;
2458 unsigned int blocksize
= i_blocksize(inode
);
2459 unsigned int zerofrom
;
2462 err
= cont_expand_zero(file
, mapping
, pos
, bytes
);
2466 zerofrom
= *bytes
& ~PAGE_MASK
;
2467 if (pos
+len
> *bytes
&& zerofrom
& (blocksize
-1)) {
2468 *bytes
|= (blocksize
-1);
2472 return block_write_begin(mapping
, pos
, len
, flags
, pagep
, get_block
);
2474 EXPORT_SYMBOL(cont_write_begin
);
2476 int block_commit_write(struct page
*page
, unsigned from
, unsigned to
)
2478 struct inode
*inode
= page
->mapping
->host
;
2479 __block_commit_write(inode
,page
,from
,to
);
2482 EXPORT_SYMBOL(block_commit_write
);
2485 * block_page_mkwrite() is not allowed to change the file size as it gets
2486 * called from a page fault handler when a page is first dirtied. Hence we must
2487 * be careful to check for EOF conditions here. We set the page up correctly
2488 * for a written page which means we get ENOSPC checking when writing into
2489 * holes and correct delalloc and unwritten extent mapping on filesystems that
2490 * support these features.
2492 * We are not allowed to take the i_mutex here so we have to play games to
2493 * protect against truncate races as the page could now be beyond EOF. Because
2494 * truncate writes the inode size before removing pages, once we have the
2495 * page lock we can determine safely if the page is beyond EOF. If it is not
2496 * beyond EOF, then the page is guaranteed safe against truncation until we
2499 * Direct callers of this function should protect against filesystem freezing
2500 * using sb_start_pagefault() - sb_end_pagefault() functions.
2502 int block_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
2503 get_block_t get_block
)
2505 struct page
*page
= vmf
->page
;
2506 struct inode
*inode
= file_inode(vma
->vm_file
);
2512 size
= i_size_read(inode
);
2513 if ((page
->mapping
!= inode
->i_mapping
) ||
2514 (page_offset(page
) > size
)) {
2515 /* We overload EFAULT to mean page got truncated */
2520 /* page is wholly or partially inside EOF */
2521 if (((page
->index
+ 1) << PAGE_SHIFT
) > size
)
2522 end
= size
& ~PAGE_MASK
;
2526 ret
= __block_write_begin(page
, 0, end
, get_block
);
2528 ret
= block_commit_write(page
, 0, end
);
2530 if (unlikely(ret
< 0))
2532 set_page_dirty(page
);
2533 wait_for_stable_page(page
);
2539 EXPORT_SYMBOL(block_page_mkwrite
);
2542 * nobh_write_begin()'s prereads are special: the buffer_heads are freed
2543 * immediately, while under the page lock. So it needs a special end_io
2544 * handler which does not touch the bh after unlocking it.
2546 static void end_buffer_read_nobh(struct buffer_head
*bh
, int uptodate
)
2548 __end_buffer_read_notouch(bh
, uptodate
);
2552 * Attach the singly-linked list of buffers created by nobh_write_begin, to
2553 * the page (converting it to circular linked list and taking care of page
2556 static void attach_nobh_buffers(struct page
*page
, struct buffer_head
*head
)
2558 struct buffer_head
*bh
;
2560 BUG_ON(!PageLocked(page
));
2562 spin_lock(&page
->mapping
->private_lock
);
2565 if (PageDirty(page
))
2566 set_buffer_dirty(bh
);
2567 if (!bh
->b_this_page
)
2568 bh
->b_this_page
= head
;
2569 bh
= bh
->b_this_page
;
2570 } while (bh
!= head
);
2571 attach_page_buffers(page
, head
);
2572 spin_unlock(&page
->mapping
->private_lock
);
2576 * On entry, the page is fully not uptodate.
2577 * On exit the page is fully uptodate in the areas outside (from,to)
2578 * The filesystem needs to handle block truncation upon failure.
2580 int nobh_write_begin(struct address_space
*mapping
,
2581 loff_t pos
, unsigned len
, unsigned flags
,
2582 struct page
**pagep
, void **fsdata
,
2583 get_block_t
*get_block
)
2585 struct inode
*inode
= mapping
->host
;
2586 const unsigned blkbits
= inode
->i_blkbits
;
2587 const unsigned blocksize
= 1 << blkbits
;
2588 struct buffer_head
*head
, *bh
;
2592 unsigned block_in_page
;
2593 unsigned block_start
, block_end
;
2594 sector_t block_in_file
;
2597 int is_mapped_to_disk
= 1;
2599 index
= pos
>> PAGE_SHIFT
;
2600 from
= pos
& (PAGE_SIZE
- 1);
2603 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2609 if (page_has_buffers(page
)) {
2610 ret
= __block_write_begin(page
, pos
, len
, get_block
);
2616 if (PageMappedToDisk(page
))
2620 * Allocate buffers so that we can keep track of state, and potentially
2621 * attach them to the page if an error occurs. In the common case of
2622 * no error, they will just be freed again without ever being attached
2623 * to the page (which is all OK, because we're under the page lock).
2625 * Be careful: the buffer linked list is a NULL terminated one, rather
2626 * than the circular one we're used to.
2628 head
= alloc_page_buffers(page
, blocksize
, false);
2634 block_in_file
= (sector_t
)page
->index
<< (PAGE_SHIFT
- blkbits
);
2637 * We loop across all blocks in the page, whether or not they are
2638 * part of the affected region. This is so we can discover if the
2639 * page is fully mapped-to-disk.
2641 for (block_start
= 0, block_in_page
= 0, bh
= head
;
2642 block_start
< PAGE_SIZE
;
2643 block_in_page
++, block_start
+= blocksize
, bh
= bh
->b_this_page
) {
2646 block_end
= block_start
+ blocksize
;
2649 if (block_start
>= to
)
2651 ret
= get_block(inode
, block_in_file
+ block_in_page
,
2655 if (!buffer_mapped(bh
))
2656 is_mapped_to_disk
= 0;
2658 clean_bdev_bh_alias(bh
);
2659 if (PageUptodate(page
)) {
2660 set_buffer_uptodate(bh
);
2663 if (buffer_new(bh
) || !buffer_mapped(bh
)) {
2664 zero_user_segments(page
, block_start
, from
,
2668 if (buffer_uptodate(bh
))
2669 continue; /* reiserfs does this */
2670 if (block_start
< from
|| block_end
> to
) {
2672 bh
->b_end_io
= end_buffer_read_nobh
;
2673 submit_bh(REQ_OP_READ
, 0, bh
);
2680 * The page is locked, so these buffers are protected from
2681 * any VM or truncate activity. Hence we don't need to care
2682 * for the buffer_head refcounts.
2684 for (bh
= head
; bh
; bh
= bh
->b_this_page
) {
2686 if (!buffer_uptodate(bh
))
2693 if (is_mapped_to_disk
)
2694 SetPageMappedToDisk(page
);
2696 *fsdata
= head
; /* to be released by nobh_write_end */
2703 * Error recovery is a bit difficult. We need to zero out blocks that
2704 * were newly allocated, and dirty them to ensure they get written out.
2705 * Buffers need to be attached to the page at this point, otherwise
2706 * the handling of potential IO errors during writeout would be hard
2707 * (could try doing synchronous writeout, but what if that fails too?)
2709 attach_nobh_buffers(page
, head
);
2710 page_zero_new_buffers(page
, from
, to
);
2719 EXPORT_SYMBOL(nobh_write_begin
);
2721 int nobh_write_end(struct file
*file
, struct address_space
*mapping
,
2722 loff_t pos
, unsigned len
, unsigned copied
,
2723 struct page
*page
, void *fsdata
)
2725 struct inode
*inode
= page
->mapping
->host
;
2726 struct buffer_head
*head
= fsdata
;
2727 struct buffer_head
*bh
;
2728 BUG_ON(fsdata
!= NULL
&& page_has_buffers(page
));
2730 if (unlikely(copied
< len
) && head
)
2731 attach_nobh_buffers(page
, head
);
2732 if (page_has_buffers(page
))
2733 return generic_write_end(file
, mapping
, pos
, len
,
2734 copied
, page
, fsdata
);
2736 SetPageUptodate(page
);
2737 set_page_dirty(page
);
2738 if (pos
+copied
> inode
->i_size
) {
2739 i_size_write(inode
, pos
+copied
);
2740 mark_inode_dirty(inode
);
2748 head
= head
->b_this_page
;
2749 free_buffer_head(bh
);
2754 EXPORT_SYMBOL(nobh_write_end
);
2757 * nobh_writepage() - based on block_full_write_page() except
2758 * that it tries to operate without attaching bufferheads to
2761 int nobh_writepage(struct page
*page
, get_block_t
*get_block
,
2762 struct writeback_control
*wbc
)
2764 struct inode
* const inode
= page
->mapping
->host
;
2765 loff_t i_size
= i_size_read(inode
);
2766 const pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2770 /* Is the page fully inside i_size? */
2771 if (page
->index
< end_index
)
2774 /* Is the page fully outside i_size? (truncate in progress) */
2775 offset
= i_size
& (PAGE_SIZE
-1);
2776 if (page
->index
>= end_index
+1 || !offset
) {
2778 * The page may have dirty, unmapped buffers. For example,
2779 * they may have been added in ext3_writepage(). Make them
2780 * freeable here, so the page does not leak.
2783 /* Not really sure about this - do we need this ? */
2784 if (page
->mapping
->a_ops
->invalidatepage
)
2785 page
->mapping
->a_ops
->invalidatepage(page
, offset
);
2788 return 0; /* don't care */
2792 * The page straddles i_size. It must be zeroed out on each and every
2793 * writepage invocation because it may be mmapped. "A file is mapped
2794 * in multiples of the page size. For a file that is not a multiple of
2795 * the page size, the remaining memory is zeroed when mapped, and
2796 * writes to that region are not written out to the file."
2798 zero_user_segment(page
, offset
, PAGE_SIZE
);
2800 ret
= mpage_writepage(page
, get_block
, wbc
);
2802 ret
= __block_write_full_page(inode
, page
, get_block
, wbc
,
2803 end_buffer_async_write
);
2806 EXPORT_SYMBOL(nobh_writepage
);
2808 int nobh_truncate_page(struct address_space
*mapping
,
2809 loff_t from
, get_block_t
*get_block
)
2811 pgoff_t index
= from
>> PAGE_SHIFT
;
2812 unsigned offset
= from
& (PAGE_SIZE
-1);
2815 unsigned length
, pos
;
2816 struct inode
*inode
= mapping
->host
;
2818 struct buffer_head map_bh
;
2821 blocksize
= i_blocksize(inode
);
2822 length
= offset
& (blocksize
- 1);
2824 /* Block boundary? Nothing to do */
2828 length
= blocksize
- length
;
2829 iblock
= (sector_t
)index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
2831 page
= grab_cache_page(mapping
, index
);
2836 if (page_has_buffers(page
)) {
2840 return block_truncate_page(mapping
, from
, get_block
);
2843 /* Find the buffer that contains "offset" */
2845 while (offset
>= pos
) {
2850 map_bh
.b_size
= blocksize
;
2852 err
= get_block(inode
, iblock
, &map_bh
, 0);
2855 /* unmapped? It's a hole - nothing to do */
2856 if (!buffer_mapped(&map_bh
))
2859 /* Ok, it's mapped. Make sure it's up-to-date */
2860 if (!PageUptodate(page
)) {
2861 err
= mapping
->a_ops
->readpage(NULL
, page
);
2867 if (!PageUptodate(page
)) {
2871 if (page_has_buffers(page
))
2874 zero_user(page
, offset
, length
);
2875 set_page_dirty(page
);
2884 EXPORT_SYMBOL(nobh_truncate_page
);
2886 int block_truncate_page(struct address_space
*mapping
,
2887 loff_t from
, get_block_t
*get_block
)
2889 pgoff_t index
= from
>> PAGE_SHIFT
;
2890 unsigned offset
= from
& (PAGE_SIZE
-1);
2893 unsigned length
, pos
;
2894 struct inode
*inode
= mapping
->host
;
2896 struct buffer_head
*bh
;
2899 blocksize
= i_blocksize(inode
);
2900 length
= offset
& (blocksize
- 1);
2902 /* Block boundary? Nothing to do */
2906 length
= blocksize
- length
;
2907 iblock
= (sector_t
)index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
2909 page
= grab_cache_page(mapping
, index
);
2914 if (!page_has_buffers(page
))
2915 create_empty_buffers(page
, blocksize
, 0);
2917 /* Find the buffer that contains "offset" */
2918 bh
= page_buffers(page
);
2920 while (offset
>= pos
) {
2921 bh
= bh
->b_this_page
;
2927 if (!buffer_mapped(bh
)) {
2928 WARN_ON(bh
->b_size
!= blocksize
);
2929 err
= get_block(inode
, iblock
, bh
, 0);
2932 /* unmapped? It's a hole - nothing to do */
2933 if (!buffer_mapped(bh
))
2937 /* Ok, it's mapped. Make sure it's up-to-date */
2938 if (PageUptodate(page
))
2939 set_buffer_uptodate(bh
);
2941 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) && !buffer_unwritten(bh
)) {
2943 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
2945 /* Uhhuh. Read error. Complain and punt. */
2946 if (!buffer_uptodate(bh
))
2950 zero_user(page
, offset
, length
);
2951 mark_buffer_dirty(bh
);
2960 EXPORT_SYMBOL(block_truncate_page
);
2963 * The generic ->writepage function for buffer-backed address_spaces
2965 int block_write_full_page(struct page
*page
, get_block_t
*get_block
,
2966 struct writeback_control
*wbc
)
2968 struct inode
* const inode
= page
->mapping
->host
;
2969 loff_t i_size
= i_size_read(inode
);
2970 const pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2973 /* Is the page fully inside i_size? */
2974 if (page
->index
< end_index
)
2975 return __block_write_full_page(inode
, page
, get_block
, wbc
,
2976 end_buffer_async_write
);
2978 /* Is the page fully outside i_size? (truncate in progress) */
2979 offset
= i_size
& (PAGE_SIZE
-1);
2980 if (page
->index
>= end_index
+1 || !offset
) {
2982 * The page may have dirty, unmapped buffers. For example,
2983 * they may have been added in ext3_writepage(). Make them
2984 * freeable here, so the page does not leak.
2986 do_invalidatepage(page
, 0, PAGE_SIZE
);
2988 return 0; /* don't care */
2992 * The page straddles i_size. It must be zeroed out on each and every
2993 * writepage invocation because it may be mmapped. "A file is mapped
2994 * in multiples of the page size. For a file that is not a multiple of
2995 * the page size, the remaining memory is zeroed when mapped, and
2996 * writes to that region are not written out to the file."
2998 zero_user_segment(page
, offset
, PAGE_SIZE
);
2999 return __block_write_full_page(inode
, page
, get_block
, wbc
,
3000 end_buffer_async_write
);
3002 EXPORT_SYMBOL(block_write_full_page
);
3004 sector_t
generic_block_bmap(struct address_space
*mapping
, sector_t block
,
3005 get_block_t
*get_block
)
3007 struct inode
*inode
= mapping
->host
;
3008 struct buffer_head tmp
= {
3009 .b_size
= i_blocksize(inode
),
3012 get_block(inode
, block
, &tmp
, 0);
3013 return tmp
.b_blocknr
;
3015 EXPORT_SYMBOL(generic_block_bmap
);
3017 static void end_bio_bh_io_sync(struct bio
*bio
)
3019 struct buffer_head
*bh
= bio
->bi_private
;
3021 if (unlikely(bio_flagged(bio
, BIO_QUIET
)))
3022 set_bit(BH_Quiet
, &bh
->b_state
);
3024 bh
->b_end_io(bh
, !bio
->bi_status
);
3029 * This allows us to do IO even on the odd last sectors
3030 * of a device, even if the block size is some multiple
3031 * of the physical sector size.
3033 * We'll just truncate the bio to the size of the device,
3034 * and clear the end of the buffer head manually.
3036 * Truly out-of-range accesses will turn into actual IO
3037 * errors, this only handles the "we need to be able to
3038 * do IO at the final sector" case.
3040 void guard_bio_eod(int op
, struct bio
*bio
)
3043 struct bio_vec
*bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
3044 unsigned truncated_bytes
;
3046 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);
3065 /* Truncate the bio.. */
3066 bio
->bi_iter
.bi_size
-= truncated_bytes
;
3067 bvec
->bv_len
-= truncated_bytes
;
3069 /* ..and clear the end of the buffer for reads */
3070 if (op
== REQ_OP_READ
) {
3071 zero_user(bvec
->bv_page
, bvec
->bv_offset
+ bvec
->bv_len
,
3076 static int submit_bh_wbc(int op
, int op_flags
, struct buffer_head
*bh
,
3077 enum rw_hint write_hint
, struct writeback_control
*wbc
)
3081 BUG_ON(!buffer_locked(bh
));
3082 BUG_ON(!buffer_mapped(bh
));
3083 BUG_ON(!bh
->b_end_io
);
3084 BUG_ON(buffer_delay(bh
));
3085 BUG_ON(buffer_unwritten(bh
));
3088 * Only clear out a write error when rewriting
3090 if (test_set_buffer_req(bh
) && (op
== REQ_OP_WRITE
))
3091 clear_buffer_write_io_error(bh
);
3094 * from here on down, it's all bio -- do the initial mapping,
3095 * submit_bio -> generic_make_request may further map this bio around
3097 bio
= bio_alloc(GFP_NOIO
, 1);
3100 wbc_init_bio(wbc
, bio
);
3101 wbc_account_io(wbc
, bh
->b_page
, bh
->b_size
);
3104 bio
->bi_iter
.bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
3105 bio_set_dev(bio
, bh
->b_bdev
);
3106 bio
->bi_write_hint
= write_hint
;
3108 bio_add_page(bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
3109 BUG_ON(bio
->bi_iter
.bi_size
!= bh
->b_size
);
3111 bio
->bi_end_io
= end_bio_bh_io_sync
;
3112 bio
->bi_private
= bh
;
3114 /* Take care of bh's that straddle the end of the device */
3115 guard_bio_eod(op
, bio
);
3117 if (buffer_meta(bh
))
3118 op_flags
|= REQ_META
;
3119 if (buffer_prio(bh
))
3120 op_flags
|= REQ_PRIO
;
3121 bio_set_op_attrs(bio
, op
, op_flags
);
3127 int submit_bh(int op
, int op_flags
, struct buffer_head
*bh
)
3129 return submit_bh_wbc(op
, op_flags
, bh
, 0, NULL
);
3131 EXPORT_SYMBOL(submit_bh
);
3134 * ll_rw_block: low-level access to block devices (DEPRECATED)
3135 * @op: whether to %READ or %WRITE
3136 * @op_flags: req_flag_bits
3137 * @nr: number of &struct buffer_heads in the array
3138 * @bhs: array of pointers to &struct buffer_head
3140 * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
3141 * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE.
3142 * @op_flags contains flags modifying the detailed I/O behavior, most notably
3145 * This function drops any buffer that it cannot get a lock on (with the
3146 * BH_Lock state bit), any buffer that appears to be clean when doing a write
3147 * request, and any buffer that appears to be up-to-date when doing read
3148 * request. Further it marks as clean buffers that are processed for
3149 * writing (the buffer cache won't assume that they are actually clean
3150 * until the buffer gets unlocked).
3152 * ll_rw_block sets b_end_io to simple completion handler that marks
3153 * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
3156 * All of the buffers must be for the same device, and must also be a
3157 * multiple of the current approved size for the device.
3159 void ll_rw_block(int op
, int op_flags
, int nr
, struct buffer_head
*bhs
[])
3163 for (i
= 0; i
< nr
; i
++) {
3164 struct buffer_head
*bh
= bhs
[i
];
3166 if (!trylock_buffer(bh
))
3169 if (test_clear_buffer_dirty(bh
)) {
3170 bh
->b_end_io
= end_buffer_write_sync
;
3172 submit_bh(op
, op_flags
, bh
);
3176 if (!buffer_uptodate(bh
)) {
3177 bh
->b_end_io
= end_buffer_read_sync
;
3179 submit_bh(op
, op_flags
, bh
);
3186 EXPORT_SYMBOL(ll_rw_block
);
3188 void write_dirty_buffer(struct buffer_head
*bh
, int op_flags
)
3191 if (!test_clear_buffer_dirty(bh
)) {
3195 bh
->b_end_io
= end_buffer_write_sync
;
3197 submit_bh(REQ_OP_WRITE
, op_flags
, bh
);
3199 EXPORT_SYMBOL(write_dirty_buffer
);
3202 * For a data-integrity writeout, we need to wait upon any in-progress I/O
3203 * and then start new I/O and then wait upon it. The caller must have a ref on
3206 int __sync_dirty_buffer(struct buffer_head
*bh
, int op_flags
)
3210 WARN_ON(atomic_read(&bh
->b_count
) < 1);
3212 if (test_clear_buffer_dirty(bh
)) {
3214 bh
->b_end_io
= end_buffer_write_sync
;
3215 ret
= submit_bh(REQ_OP_WRITE
, op_flags
, bh
);
3217 if (!ret
&& !buffer_uptodate(bh
))
3224 EXPORT_SYMBOL(__sync_dirty_buffer
);
3226 int sync_dirty_buffer(struct buffer_head
*bh
)
3228 return __sync_dirty_buffer(bh
, REQ_SYNC
);
3230 EXPORT_SYMBOL(sync_dirty_buffer
);
3233 * try_to_free_buffers() checks if all the buffers on this particular page
3234 * are unused, and releases them if so.
3236 * Exclusion against try_to_free_buffers may be obtained by either
3237 * locking the page or by holding its mapping's private_lock.
3239 * If the page is dirty but all the buffers are clean then we need to
3240 * be sure to mark the page clean as well. This is because the page
3241 * may be against a block device, and a later reattachment of buffers
3242 * to a dirty page will set *all* buffers dirty. Which would corrupt
3243 * filesystem data on the same device.
3245 * The same applies to regular filesystem pages: if all the buffers are
3246 * clean then we set the page clean and proceed. To do that, we require
3247 * total exclusion from __set_page_dirty_buffers(). That is obtained with
3250 * try_to_free_buffers() is non-blocking.
3252 static inline int buffer_busy(struct buffer_head
*bh
)
3254 return atomic_read(&bh
->b_count
) |
3255 (bh
->b_state
& ((1 << BH_Dirty
) | (1 << BH_Lock
)));
3259 drop_buffers(struct page
*page
, struct buffer_head
**buffers_to_free
)
3261 struct buffer_head
*head
= page_buffers(page
);
3262 struct buffer_head
*bh
;
3266 if (buffer_busy(bh
))
3268 bh
= bh
->b_this_page
;
3269 } while (bh
!= head
);
3272 struct buffer_head
*next
= bh
->b_this_page
;
3274 if (bh
->b_assoc_map
)
3275 __remove_assoc_queue(bh
);
3277 } while (bh
!= head
);
3278 *buffers_to_free
= head
;
3279 __clear_page_buffers(page
);
3285 int try_to_free_buffers(struct page
*page
)
3287 struct address_space
* const mapping
= page
->mapping
;
3288 struct buffer_head
*buffers_to_free
= NULL
;
3291 BUG_ON(!PageLocked(page
));
3292 if (PageWriteback(page
))
3295 if (mapping
== NULL
) { /* can this still happen? */
3296 ret
= drop_buffers(page
, &buffers_to_free
);
3300 spin_lock(&mapping
->private_lock
);
3301 ret
= drop_buffers(page
, &buffers_to_free
);
3304 * If the filesystem writes its buffers by hand (eg ext3)
3305 * then we can have clean buffers against a dirty page. We
3306 * clean the page here; otherwise the VM will never notice
3307 * that the filesystem did any IO at all.
3309 * Also, during truncate, discard_buffer will have marked all
3310 * the page's buffers clean. We discover that here and clean
3313 * private_lock must be held over this entire operation in order
3314 * to synchronise against __set_page_dirty_buffers and prevent the
3315 * dirty bit from being lost.
3318 cancel_dirty_page(page
);
3319 spin_unlock(&mapping
->private_lock
);
3321 if (buffers_to_free
) {
3322 struct buffer_head
*bh
= buffers_to_free
;
3325 struct buffer_head
*next
= bh
->b_this_page
;
3326 free_buffer_head(bh
);
3328 } while (bh
!= buffers_to_free
);
3332 EXPORT_SYMBOL(try_to_free_buffers
);
3335 * There are no bdflush tunables left. But distributions are
3336 * still running obsolete flush daemons, so we terminate them here.
3338 * Use of bdflush() is deprecated and will be removed in a future kernel.
3339 * The `flush-X' kernel threads fully replace bdflush daemons and this call.
3341 SYSCALL_DEFINE2(bdflush
, int, func
, long, data
)
3343 static int msg_count
;
3345 if (!capable(CAP_SYS_ADMIN
))
3348 if (msg_count
< 5) {
3351 "warning: process `%s' used the obsolete bdflush"
3352 " system call\n", current
->comm
);
3353 printk(KERN_INFO
"Fix your initscripts?\n");
3362 * Buffer-head allocation
3364 static struct kmem_cache
*bh_cachep __read_mostly
;
3367 * Once the number of bh's in the machine exceeds this level, we start
3368 * stripping them in writeback.
3370 static unsigned long max_buffer_heads
;
3372 int buffer_heads_over_limit
;
3374 struct bh_accounting
{
3375 int nr
; /* Number of live bh's */
3376 int ratelimit
; /* Limit cacheline bouncing */
3379 static DEFINE_PER_CPU(struct bh_accounting
, bh_accounting
) = {0, 0};
3381 static void recalc_bh_state(void)
3386 if (__this_cpu_inc_return(bh_accounting
.ratelimit
) - 1 < 4096)
3388 __this_cpu_write(bh_accounting
.ratelimit
, 0);
3389 for_each_online_cpu(i
)
3390 tot
+= per_cpu(bh_accounting
, i
).nr
;
3391 buffer_heads_over_limit
= (tot
> max_buffer_heads
);
3394 struct buffer_head
*alloc_buffer_head(gfp_t gfp_flags
)
3396 struct buffer_head
*ret
= kmem_cache_zalloc(bh_cachep
, gfp_flags
);
3398 INIT_LIST_HEAD(&ret
->b_assoc_buffers
);
3400 __this_cpu_inc(bh_accounting
.nr
);
3406 EXPORT_SYMBOL(alloc_buffer_head
);
3408 void free_buffer_head(struct buffer_head
*bh
)
3410 BUG_ON(!list_empty(&bh
->b_assoc_buffers
));
3411 kmem_cache_free(bh_cachep
, bh
);
3413 __this_cpu_dec(bh_accounting
.nr
);
3417 EXPORT_SYMBOL(free_buffer_head
);
3419 static int buffer_exit_cpu_dead(unsigned int cpu
)
3422 struct bh_lru
*b
= &per_cpu(bh_lrus
, cpu
);
3424 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
3428 this_cpu_add(bh_accounting
.nr
, per_cpu(bh_accounting
, cpu
).nr
);
3429 per_cpu(bh_accounting
, cpu
).nr
= 0;
3434 * bh_uptodate_or_lock - Test whether the buffer is uptodate
3435 * @bh: struct buffer_head
3437 * Return true if the buffer is up-to-date and false,
3438 * with the buffer locked, if not.
3440 int bh_uptodate_or_lock(struct buffer_head
*bh
)
3442 if (!buffer_uptodate(bh
)) {
3444 if (!buffer_uptodate(bh
))
3450 EXPORT_SYMBOL(bh_uptodate_or_lock
);
3453 * bh_submit_read - Submit a locked buffer for reading
3454 * @bh: struct buffer_head
3456 * Returns zero on success and -EIO on error.
3458 int bh_submit_read(struct buffer_head
*bh
)
3460 BUG_ON(!buffer_locked(bh
));
3462 if (buffer_uptodate(bh
)) {
3468 bh
->b_end_io
= end_buffer_read_sync
;
3469 submit_bh(REQ_OP_READ
, 0, bh
);
3471 if (buffer_uptodate(bh
))
3475 EXPORT_SYMBOL(bh_submit_read
);
3478 * Seek for SEEK_DATA / SEEK_HOLE within @page, starting at @lastoff.
3480 * Returns the offset within the file on success, and -ENOENT otherwise.
3483 page_seek_hole_data(struct page
*page
, loff_t lastoff
, int whence
)
3485 loff_t offset
= page_offset(page
);
3486 struct buffer_head
*bh
, *head
;
3487 bool seek_data
= whence
== SEEK_DATA
;
3489 if (lastoff
< offset
)
3492 bh
= head
= page_buffers(page
);
3494 offset
+= bh
->b_size
;
3495 if (lastoff
>= offset
)
3499 * Unwritten extents that have data in the page cache covering
3500 * them can be identified by the BH_Unwritten state flag.
3501 * Pages with multiple buffers might have a mix of holes, data
3502 * and unwritten extents - any buffer with valid data in it
3503 * should have BH_Uptodate flag set on it.
3506 if ((buffer_unwritten(bh
) || buffer_uptodate(bh
)) == seek_data
)
3510 } while ((bh
= bh
->b_this_page
) != head
);
3515 * Seek for SEEK_DATA / SEEK_HOLE in the page cache.
3517 * Within unwritten extents, the page cache determines which parts are holes
3518 * and which are data: unwritten and uptodate buffer heads count as data;
3519 * everything else counts as a hole.
3521 * Returns the resulting offset on successs, and -ENOENT otherwise.
3524 page_cache_seek_hole_data(struct inode
*inode
, loff_t offset
, loff_t length
,
3527 pgoff_t index
= offset
>> PAGE_SHIFT
;
3528 pgoff_t end
= DIV_ROUND_UP(offset
+ length
, PAGE_SIZE
);
3529 loff_t lastoff
= offset
;
3530 struct pagevec pvec
;
3535 pagevec_init(&pvec
, 0);
3538 unsigned nr_pages
, i
;
3540 nr_pages
= pagevec_lookup_range(&pvec
, inode
->i_mapping
, &index
,
3545 for (i
= 0; i
< nr_pages
; i
++) {
3546 struct page
*page
= pvec
.pages
[i
];
3549 * At this point, the page may be truncated or
3550 * invalidated (changing page->mapping to NULL), or
3551 * even swizzled back from swapper_space to tmpfs file
3552 * mapping. However, page->index will not change
3553 * because we have a reference on the page.
3555 * If current page offset is beyond where we've ended,
3556 * we've found a hole.
3558 if (whence
== SEEK_HOLE
&&
3559 lastoff
< page_offset(page
))
3563 if (likely(page
->mapping
== inode
->i_mapping
) &&
3564 page_has_buffers(page
)) {
3565 lastoff
= page_seek_hole_data(page
, lastoff
, whence
);
3572 lastoff
= page_offset(page
) + PAGE_SIZE
;
3574 pagevec_release(&pvec
);
3575 } while (index
< end
);
3577 /* When no page at lastoff and we are not done, we found a hole. */
3578 if (whence
!= SEEK_HOLE
)
3582 if (lastoff
< offset
+ length
)
3587 pagevec_release(&pvec
);
3591 void __init
buffer_init(void)
3593 unsigned long nrpages
;
3596 bh_cachep
= kmem_cache_create("buffer_head",
3597 sizeof(struct buffer_head
), 0,
3598 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
3603 * Limit the bh occupancy to 10% of ZONE_NORMAL
3605 nrpages
= (nr_free_buffer_pages() * 10) / 100;
3606 max_buffer_heads
= nrpages
* (PAGE_SIZE
/ sizeof(struct buffer_head
));
3607 ret
= cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD
, "fs/buffer:dead",
3608 NULL
, buffer_exit_cpu_dead
);