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/syscalls.h>
25 #include <linux/percpu.h>
26 #include <linux/slab.h>
27 #include <linux/capability.h>
28 #include <linux/blkdev.h>
29 #include <linux/file.h>
30 #include <linux/quotaops.h>
31 #include <linux/highmem.h>
32 #include <linux/export.h>
33 #include <linux/writeback.h>
34 #include <linux/hash.h>
35 #include <linux/suspend.h>
36 #include <linux/buffer_head.h>
37 #include <linux/task_io_accounting_ops.h>
38 #include <linux/bio.h>
39 #include <linux/notifier.h>
40 #include <linux/cpu.h>
41 #include <linux/bitops.h>
42 #include <linux/mpage.h>
43 #include <linux/bit_spinlock.h>
44 #include <trace/events/block.h>
46 static int fsync_buffers_list(spinlock_t
*lock
, struct list_head
*list
);
48 #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
50 void init_buffer(struct buffer_head
*bh
, bh_end_io_t
*handler
, void *private)
52 bh
->b_end_io
= handler
;
53 bh
->b_private
= private;
55 EXPORT_SYMBOL(init_buffer
);
57 inline void touch_buffer(struct buffer_head
*bh
)
59 trace_block_touch_buffer(bh
);
60 mark_page_accessed(bh
->b_page
);
62 EXPORT_SYMBOL(touch_buffer
);
64 void __lock_buffer(struct buffer_head
*bh
)
66 wait_on_bit_lock_io(&bh
->b_state
, BH_Lock
, TASK_UNINTERRUPTIBLE
);
68 EXPORT_SYMBOL(__lock_buffer
);
70 void unlock_buffer(struct buffer_head
*bh
)
72 clear_bit_unlock(BH_Lock
, &bh
->b_state
);
73 smp_mb__after_atomic();
74 wake_up_bit(&bh
->b_state
, BH_Lock
);
76 EXPORT_SYMBOL(unlock_buffer
);
79 * Returns if the page has dirty or writeback buffers. If all the buffers
80 * are unlocked and clean then the PageDirty information is stale. If
81 * any of the pages are locked, it is assumed they are locked for IO.
83 void buffer_check_dirty_writeback(struct page
*page
,
84 bool *dirty
, bool *writeback
)
86 struct buffer_head
*head
, *bh
;
90 BUG_ON(!PageLocked(page
));
92 if (!page_has_buffers(page
))
95 if (PageWriteback(page
))
98 head
= page_buffers(page
);
101 if (buffer_locked(bh
))
104 if (buffer_dirty(bh
))
107 bh
= bh
->b_this_page
;
108 } while (bh
!= head
);
110 EXPORT_SYMBOL(buffer_check_dirty_writeback
);
113 * Block until a buffer comes unlocked. This doesn't stop it
114 * from becoming locked again - you have to lock it yourself
115 * if you want to preserve its state.
117 void __wait_on_buffer(struct buffer_head
* bh
)
119 wait_on_bit_io(&bh
->b_state
, BH_Lock
, TASK_UNINTERRUPTIBLE
);
121 EXPORT_SYMBOL(__wait_on_buffer
);
124 __clear_page_buffers(struct page
*page
)
126 ClearPagePrivate(page
);
127 set_page_private(page
, 0);
128 page_cache_release(page
);
131 static void buffer_io_error(struct buffer_head
*bh
, char *msg
)
133 char b
[BDEVNAME_SIZE
];
135 if (!test_bit(BH_Quiet
, &bh
->b_state
))
136 printk_ratelimited(KERN_ERR
137 "Buffer I/O error on dev %s, logical block %llu%s\n",
138 bdevname(bh
->b_bdev
, b
),
139 (unsigned long long)bh
->b_blocknr
, msg
);
143 * End-of-IO handler helper function which does not touch the bh after
145 * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
146 * a race there is benign: unlock_buffer() only use the bh's address for
147 * hashing after unlocking the buffer, so it doesn't actually touch the bh
150 static void __end_buffer_read_notouch(struct buffer_head
*bh
, int uptodate
)
153 set_buffer_uptodate(bh
);
155 /* This happens, due to failed READA attempts. */
156 clear_buffer_uptodate(bh
);
162 * Default synchronous end-of-IO handler.. Just mark it up-to-date and
163 * unlock the buffer. This is what ll_rw_block uses too.
165 void end_buffer_read_sync(struct buffer_head
*bh
, int uptodate
)
167 __end_buffer_read_notouch(bh
, uptodate
);
170 EXPORT_SYMBOL(end_buffer_read_sync
);
172 void end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
175 set_buffer_uptodate(bh
);
177 buffer_io_error(bh
, ", lost sync page write");
178 set_buffer_write_io_error(bh
);
179 clear_buffer_uptodate(bh
);
184 EXPORT_SYMBOL(end_buffer_write_sync
);
187 * Various filesystems appear to want __find_get_block to be non-blocking.
188 * But it's the page lock which protects the buffers. To get around this,
189 * we get exclusion from try_to_free_buffers with the blockdev mapping's
192 * Hack idea: for the blockdev mapping, i_bufferlist_lock contention
193 * may be quite high. This code could TryLock the page, and if that
194 * succeeds, there is no need to take private_lock. (But if
195 * private_lock is contended then so is mapping->tree_lock).
197 static struct buffer_head
*
198 __find_get_block_slow(struct block_device
*bdev
, sector_t block
)
200 struct inode
*bd_inode
= bdev
->bd_inode
;
201 struct address_space
*bd_mapping
= bd_inode
->i_mapping
;
202 struct buffer_head
*ret
= NULL
;
204 struct buffer_head
*bh
;
205 struct buffer_head
*head
;
209 index
= block
>> (PAGE_CACHE_SHIFT
- bd_inode
->i_blkbits
);
210 page
= find_get_page_flags(bd_mapping
, index
, FGP_ACCESSED
);
214 spin_lock(&bd_mapping
->private_lock
);
215 if (!page_has_buffers(page
))
217 head
= page_buffers(page
);
220 if (!buffer_mapped(bh
))
222 else if (bh
->b_blocknr
== block
) {
227 bh
= bh
->b_this_page
;
228 } while (bh
!= head
);
230 /* we might be here because some of the buffers on this page are
231 * not mapped. This is due to various races between
232 * file io on the block device and getblk. It gets dealt with
233 * elsewhere, don't buffer_error if we had some unmapped buffers
236 char b
[BDEVNAME_SIZE
];
238 printk("__find_get_block_slow() failed. "
239 "block=%llu, b_blocknr=%llu\n",
240 (unsigned long long)block
,
241 (unsigned long long)bh
->b_blocknr
);
242 printk("b_state=0x%08lx, b_size=%zu\n",
243 bh
->b_state
, bh
->b_size
);
244 printk("device %s blocksize: %d\n", bdevname(bdev
, b
),
245 1 << bd_inode
->i_blkbits
);
248 spin_unlock(&bd_mapping
->private_lock
);
249 page_cache_release(page
);
255 * Kick the writeback threads then try to free up some ZONE_NORMAL memory.
257 static void free_more_memory(void)
262 wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM
);
265 for_each_online_node(nid
) {
266 (void)first_zones_zonelist(node_zonelist(nid
, GFP_NOFS
),
267 gfp_zone(GFP_NOFS
), NULL
,
270 try_to_free_pages(node_zonelist(nid
, GFP_NOFS
), 0,
276 * I/O completion handler for block_read_full_page() - pages
277 * which come unlocked at the end of I/O.
279 static void end_buffer_async_read(struct buffer_head
*bh
, int uptodate
)
282 struct buffer_head
*first
;
283 struct buffer_head
*tmp
;
285 int page_uptodate
= 1;
287 BUG_ON(!buffer_async_read(bh
));
291 set_buffer_uptodate(bh
);
293 clear_buffer_uptodate(bh
);
294 buffer_io_error(bh
, ", async page read");
299 * Be _very_ careful from here on. Bad things can happen if
300 * two buffer heads end IO at almost the same time and both
301 * decide that the page is now completely done.
303 first
= page_buffers(page
);
304 local_irq_save(flags
);
305 bit_spin_lock(BH_Uptodate_Lock
, &first
->b_state
);
306 clear_buffer_async_read(bh
);
310 if (!buffer_uptodate(tmp
))
312 if (buffer_async_read(tmp
)) {
313 BUG_ON(!buffer_locked(tmp
));
316 tmp
= tmp
->b_this_page
;
318 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
319 local_irq_restore(flags
);
322 * If none of the buffers had errors and they are all
323 * uptodate then we can set the page uptodate.
325 if (page_uptodate
&& !PageError(page
))
326 SetPageUptodate(page
);
331 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
332 local_irq_restore(flags
);
337 * Completion handler for block_write_full_page() - pages which are unlocked
338 * during I/O, and which have PageWriteback cleared upon I/O completion.
340 void end_buffer_async_write(struct buffer_head
*bh
, int uptodate
)
343 struct buffer_head
*first
;
344 struct buffer_head
*tmp
;
347 BUG_ON(!buffer_async_write(bh
));
351 set_buffer_uptodate(bh
);
353 buffer_io_error(bh
, ", lost async page write");
354 set_bit(AS_EIO
, &page
->mapping
->flags
);
355 set_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 if (buffer_write_io_error(bh
))
484 set_bit(AS_EIO
, &bh
->b_assoc_map
->flags
);
485 bh
->b_assoc_map
= NULL
;
488 int inode_has_buffers(struct inode
*inode
)
490 return !list_empty(&inode
->i_data
.private_list
);
494 * osync is designed to support O_SYNC io. It waits synchronously for
495 * all already-submitted IO to complete, but does not queue any new
496 * writes to the disk.
498 * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
499 * you dirty the buffers, and then use osync_inode_buffers to wait for
500 * completion. Any other dirty buffers which are not yet queued for
501 * write will not be flushed to disk by the osync.
503 static int osync_buffers_list(spinlock_t
*lock
, struct list_head
*list
)
505 struct buffer_head
*bh
;
511 list_for_each_prev(p
, list
) {
513 if (buffer_locked(bh
)) {
517 if (!buffer_uptodate(bh
))
528 static void do_thaw_one(struct super_block
*sb
, void *unused
)
530 char b
[BDEVNAME_SIZE
];
531 while (sb
->s_bdev
&& !thaw_bdev(sb
->s_bdev
, sb
))
532 printk(KERN_WARNING
"Emergency Thaw on %s\n",
533 bdevname(sb
->s_bdev
, b
));
536 static void do_thaw_all(struct work_struct
*work
)
538 iterate_supers(do_thaw_one
, NULL
);
540 printk(KERN_WARNING
"Emergency Thaw complete\n");
544 * emergency_thaw_all -- forcibly thaw every frozen filesystem
546 * Used for emergency unfreeze of all filesystems via SysRq
548 void emergency_thaw_all(void)
550 struct work_struct
*work
;
552 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
554 INIT_WORK(work
, do_thaw_all
);
560 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
561 * @mapping: the mapping which wants those buffers written
563 * Starts I/O against the buffers at mapping->private_list, and waits upon
566 * Basically, this is a convenience function for fsync().
567 * @mapping is a file or directory which needs those buffers to be written for
568 * a successful fsync().
570 int sync_mapping_buffers(struct address_space
*mapping
)
572 struct address_space
*buffer_mapping
= mapping
->private_data
;
574 if (buffer_mapping
== NULL
|| list_empty(&mapping
->private_list
))
577 return fsync_buffers_list(&buffer_mapping
->private_lock
,
578 &mapping
->private_list
);
580 EXPORT_SYMBOL(sync_mapping_buffers
);
583 * Called when we've recently written block `bblock', and it is known that
584 * `bblock' was for a buffer_boundary() buffer. This means that the block at
585 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
586 * dirty, schedule it for IO. So that indirects merge nicely with their data.
588 void write_boundary_block(struct block_device
*bdev
,
589 sector_t bblock
, unsigned blocksize
)
591 struct buffer_head
*bh
= __find_get_block(bdev
, bblock
+ 1, blocksize
);
593 if (buffer_dirty(bh
))
594 ll_rw_block(WRITE
, 1, &bh
);
599 void mark_buffer_dirty_inode(struct buffer_head
*bh
, struct inode
*inode
)
601 struct address_space
*mapping
= inode
->i_mapping
;
602 struct address_space
*buffer_mapping
= bh
->b_page
->mapping
;
604 mark_buffer_dirty(bh
);
605 if (!mapping
->private_data
) {
606 mapping
->private_data
= buffer_mapping
;
608 BUG_ON(mapping
->private_data
!= buffer_mapping
);
610 if (!bh
->b_assoc_map
) {
611 spin_lock(&buffer_mapping
->private_lock
);
612 list_move_tail(&bh
->b_assoc_buffers
,
613 &mapping
->private_list
);
614 bh
->b_assoc_map
= mapping
;
615 spin_unlock(&buffer_mapping
->private_lock
);
618 EXPORT_SYMBOL(mark_buffer_dirty_inode
);
621 * Mark the page dirty, and set it dirty in the radix tree, and mark the inode
624 * If warn is true, then emit a warning if the page is not uptodate and has
625 * not been truncated.
627 * The caller must hold mem_cgroup_begin_page_stat() lock.
629 static void __set_page_dirty(struct page
*page
, struct address_space
*mapping
,
630 struct mem_cgroup
*memcg
, int warn
)
634 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
635 if (page
->mapping
) { /* Race with truncate? */
636 WARN_ON_ONCE(warn
&& !PageUptodate(page
));
637 account_page_dirtied(page
, mapping
, memcg
);
638 radix_tree_tag_set(&mapping
->page_tree
,
639 page_index(page
), PAGECACHE_TAG_DIRTY
);
641 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
645 * Add a page to the dirty page list.
647 * It is a sad fact of life that this function is called from several places
648 * deeply under spinlocking. It may not sleep.
650 * If the page has buffers, the uptodate buffers are set dirty, to preserve
651 * dirty-state coherency between the page and the buffers. It the page does
652 * not have buffers then when they are later attached they will all be set
655 * The buffers are dirtied before the page is dirtied. There's a small race
656 * window in which a writepage caller may see the page cleanness but not the
657 * buffer dirtiness. That's fine. If this code were to set the page dirty
658 * before the buffers, a concurrent writepage caller could clear the page dirty
659 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
660 * page on the dirty page list.
662 * We use private_lock to lock against try_to_free_buffers while using the
663 * page's buffer list. Also use this to protect against clean buffers being
664 * added to the page after it was set dirty.
666 * FIXME: may need to call ->reservepage here as well. That's rather up to the
667 * address_space though.
669 int __set_page_dirty_buffers(struct page
*page
)
672 struct mem_cgroup
*memcg
;
673 struct address_space
*mapping
= page_mapping(page
);
675 if (unlikely(!mapping
))
676 return !TestSetPageDirty(page
);
678 spin_lock(&mapping
->private_lock
);
679 if (page_has_buffers(page
)) {
680 struct buffer_head
*head
= page_buffers(page
);
681 struct buffer_head
*bh
= head
;
684 set_buffer_dirty(bh
);
685 bh
= bh
->b_this_page
;
686 } while (bh
!= head
);
689 * Use mem_group_begin_page_stat() to keep PageDirty synchronized with
690 * per-memcg dirty page counters.
692 memcg
= mem_cgroup_begin_page_stat(page
);
693 newly_dirty
= !TestSetPageDirty(page
);
694 spin_unlock(&mapping
->private_lock
);
697 __set_page_dirty(page
, mapping
, memcg
, 1);
699 mem_cgroup_end_page_stat(memcg
);
702 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
706 EXPORT_SYMBOL(__set_page_dirty_buffers
);
709 * Write out and wait upon a list of buffers.
711 * We have conflicting pressures: we want to make sure that all
712 * initially dirty buffers get waited on, but that any subsequently
713 * dirtied buffers don't. After all, we don't want fsync to last
714 * forever if somebody is actively writing to the file.
716 * Do this in two main stages: first we copy dirty buffers to a
717 * temporary inode list, queueing the writes as we go. Then we clean
718 * up, waiting for those writes to complete.
720 * During this second stage, any subsequent updates to the file may end
721 * up refiling the buffer on the original inode's dirty list again, so
722 * there is a chance we will end up with a buffer queued for write but
723 * not yet completed on that list. So, as a final cleanup we go through
724 * the osync code to catch these locked, dirty buffers without requeuing
725 * any newly dirty buffers for write.
727 static int fsync_buffers_list(spinlock_t
*lock
, struct list_head
*list
)
729 struct buffer_head
*bh
;
730 struct list_head tmp
;
731 struct address_space
*mapping
;
733 struct blk_plug plug
;
735 INIT_LIST_HEAD(&tmp
);
736 blk_start_plug(&plug
);
739 while (!list_empty(list
)) {
740 bh
= BH_ENTRY(list
->next
);
741 mapping
= bh
->b_assoc_map
;
742 __remove_assoc_queue(bh
);
743 /* Avoid race with mark_buffer_dirty_inode() which does
744 * a lockless check and we rely on seeing the dirty bit */
746 if (buffer_dirty(bh
) || buffer_locked(bh
)) {
747 list_add(&bh
->b_assoc_buffers
, &tmp
);
748 bh
->b_assoc_map
= mapping
;
749 if (buffer_dirty(bh
)) {
753 * Ensure any pending I/O completes so that
754 * write_dirty_buffer() actually writes the
755 * current contents - it is a noop if I/O is
756 * still in flight on potentially older
759 write_dirty_buffer(bh
, WRITE_SYNC
);
762 * Kick off IO for the previous mapping. Note
763 * that we will not run the very last mapping,
764 * wait_on_buffer() will do that for us
765 * through sync_buffer().
774 blk_finish_plug(&plug
);
777 while (!list_empty(&tmp
)) {
778 bh
= BH_ENTRY(tmp
.prev
);
780 mapping
= bh
->b_assoc_map
;
781 __remove_assoc_queue(bh
);
782 /* Avoid race with mark_buffer_dirty_inode() which does
783 * a lockless check and we rely on seeing the dirty bit */
785 if (buffer_dirty(bh
)) {
786 list_add(&bh
->b_assoc_buffers
,
787 &mapping
->private_list
);
788 bh
->b_assoc_map
= mapping
;
792 if (!buffer_uptodate(bh
))
799 err2
= osync_buffers_list(lock
, list
);
807 * Invalidate any and all dirty buffers on a given inode. We are
808 * probably unmounting the fs, but that doesn't mean we have already
809 * done a sync(). Just drop the buffers from the inode list.
811 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
812 * assumes that all the buffers are against the blockdev. Not true
815 void invalidate_inode_buffers(struct inode
*inode
)
817 if (inode_has_buffers(inode
)) {
818 struct address_space
*mapping
= &inode
->i_data
;
819 struct list_head
*list
= &mapping
->private_list
;
820 struct address_space
*buffer_mapping
= mapping
->private_data
;
822 spin_lock(&buffer_mapping
->private_lock
);
823 while (!list_empty(list
))
824 __remove_assoc_queue(BH_ENTRY(list
->next
));
825 spin_unlock(&buffer_mapping
->private_lock
);
828 EXPORT_SYMBOL(invalidate_inode_buffers
);
831 * Remove any clean buffers from the inode's buffer list. This is called
832 * when we're trying to free the inode itself. Those buffers can pin it.
834 * Returns true if all buffers were removed.
836 int remove_inode_buffers(struct inode
*inode
)
840 if (inode_has_buffers(inode
)) {
841 struct address_space
*mapping
= &inode
->i_data
;
842 struct list_head
*list
= &mapping
->private_list
;
843 struct address_space
*buffer_mapping
= mapping
->private_data
;
845 spin_lock(&buffer_mapping
->private_lock
);
846 while (!list_empty(list
)) {
847 struct buffer_head
*bh
= BH_ENTRY(list
->next
);
848 if (buffer_dirty(bh
)) {
852 __remove_assoc_queue(bh
);
854 spin_unlock(&buffer_mapping
->private_lock
);
860 * Create the appropriate buffers when given a page for data area and
861 * the size of each buffer.. Use the bh->b_this_page linked list to
862 * follow the buffers created. Return NULL if unable to create more
865 * The retry flag is used to differentiate async IO (paging, swapping)
866 * which may not fail from ordinary buffer allocations.
868 struct buffer_head
*alloc_page_buffers(struct page
*page
, unsigned long size
,
871 struct buffer_head
*bh
, *head
;
877 while ((offset
-= size
) >= 0) {
878 bh
= alloc_buffer_head(GFP_NOFS
);
882 bh
->b_this_page
= head
;
888 /* Link the buffer to its page */
889 set_bh_page(bh
, page
, offset
);
893 * In case anything failed, we just free everything we got.
899 head
= head
->b_this_page
;
900 free_buffer_head(bh
);
905 * Return failure for non-async IO requests. Async IO requests
906 * are not allowed to fail, so we have to wait until buffer heads
907 * become available. But we don't want tasks sleeping with
908 * partially complete buffers, so all were released above.
913 /* We're _really_ low on memory. Now we just
914 * wait for old buffer heads to become free due to
915 * finishing IO. Since this is an async request and
916 * the reserve list is empty, we're sure there are
917 * async buffer heads in use.
922 EXPORT_SYMBOL_GPL(alloc_page_buffers
);
925 link_dev_buffers(struct page
*page
, struct buffer_head
*head
)
927 struct buffer_head
*bh
, *tail
;
932 bh
= bh
->b_this_page
;
934 tail
->b_this_page
= head
;
935 attach_page_buffers(page
, head
);
938 static sector_t
blkdev_max_block(struct block_device
*bdev
, unsigned int size
)
940 sector_t retval
= ~((sector_t
)0);
941 loff_t sz
= i_size_read(bdev
->bd_inode
);
944 unsigned int sizebits
= blksize_bits(size
);
945 retval
= (sz
>> sizebits
);
951 * Initialise the state of a blockdev page's buffers.
954 init_page_buffers(struct page
*page
, struct block_device
*bdev
,
955 sector_t block
, int size
)
957 struct buffer_head
*head
= page_buffers(page
);
958 struct buffer_head
*bh
= head
;
959 int uptodate
= PageUptodate(page
);
960 sector_t end_block
= blkdev_max_block(I_BDEV(bdev
->bd_inode
), size
);
963 if (!buffer_mapped(bh
)) {
964 init_buffer(bh
, NULL
, NULL
);
966 bh
->b_blocknr
= block
;
968 set_buffer_uptodate(bh
);
969 if (block
< end_block
)
970 set_buffer_mapped(bh
);
973 bh
= bh
->b_this_page
;
974 } while (bh
!= head
);
977 * Caller needs to validate requested block against end of device.
983 * Create the page-cache page that contains the requested block.
985 * This is used purely for blockdev mappings.
988 grow_dev_page(struct block_device
*bdev
, sector_t block
,
989 pgoff_t index
, int size
, int sizebits
, gfp_t gfp
)
991 struct inode
*inode
= bdev
->bd_inode
;
993 struct buffer_head
*bh
;
995 int ret
= 0; /* Will call free_more_memory() */
998 gfp_mask
= (mapping_gfp_mask(inode
->i_mapping
) & ~__GFP_FS
) | gfp
;
1001 * XXX: __getblk_slow() can not really deal with failure and
1002 * will endlessly loop on improvised global reclaim. Prefer
1003 * looping in the allocator rather than here, at least that
1004 * code knows what it's doing.
1006 gfp_mask
|= __GFP_NOFAIL
;
1008 page
= find_or_create_page(inode
->i_mapping
, index
, gfp_mask
);
1012 BUG_ON(!PageLocked(page
));
1014 if (page_has_buffers(page
)) {
1015 bh
= page_buffers(page
);
1016 if (bh
->b_size
== size
) {
1017 end_block
= init_page_buffers(page
, bdev
,
1018 (sector_t
)index
<< sizebits
,
1022 if (!try_to_free_buffers(page
))
1027 * Allocate some buffers for this page
1029 bh
= alloc_page_buffers(page
, size
, 0);
1034 * Link the page to the buffers and initialise them. Take the
1035 * lock to be atomic wrt __find_get_block(), which does not
1036 * run under the page lock.
1038 spin_lock(&inode
->i_mapping
->private_lock
);
1039 link_dev_buffers(page
, bh
);
1040 end_block
= init_page_buffers(page
, bdev
, (sector_t
)index
<< sizebits
,
1042 spin_unlock(&inode
->i_mapping
->private_lock
);
1044 ret
= (block
< end_block
) ? 1 : -ENXIO
;
1047 page_cache_release(page
);
1052 * Create buffers for the specified block device block's page. If
1053 * that page was dirty, the buffers are set dirty also.
1056 grow_buffers(struct block_device
*bdev
, sector_t block
, int size
, gfp_t gfp
)
1064 } while ((size
<< sizebits
) < PAGE_SIZE
);
1066 index
= block
>> sizebits
;
1069 * Check for a block which wants to lie outside our maximum possible
1070 * pagecache index. (this comparison is done using sector_t types).
1072 if (unlikely(index
!= block
>> sizebits
)) {
1073 char b
[BDEVNAME_SIZE
];
1075 printk(KERN_ERR
"%s: requested out-of-range block %llu for "
1077 __func__
, (unsigned long long)block
,
1082 /* Create a page with the proper size buffers.. */
1083 return grow_dev_page(bdev
, block
, index
, size
, sizebits
, gfp
);
1086 struct buffer_head
*
1087 __getblk_slow(struct block_device
*bdev
, sector_t block
,
1088 unsigned size
, gfp_t gfp
)
1090 /* Size must be multiple of hard sectorsize */
1091 if (unlikely(size
& (bdev_logical_block_size(bdev
)-1) ||
1092 (size
< 512 || size
> PAGE_SIZE
))) {
1093 printk(KERN_ERR
"getblk(): invalid block size %d requested\n",
1095 printk(KERN_ERR
"logical block size: %d\n",
1096 bdev_logical_block_size(bdev
));
1103 struct buffer_head
*bh
;
1106 bh
= __find_get_block(bdev
, block
, size
);
1110 ret
= grow_buffers(bdev
, block
, size
, gfp
);
1117 EXPORT_SYMBOL(__getblk_slow
);
1120 * The relationship between dirty buffers and dirty pages:
1122 * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1123 * the page is tagged dirty in its radix tree.
1125 * At all times, the dirtiness of the buffers represents the dirtiness of
1126 * subsections of the page. If the page has buffers, the page dirty bit is
1127 * merely a hint about the true dirty state.
1129 * When a page is set dirty in its entirety, all its buffers are marked dirty
1130 * (if the page has buffers).
1132 * When a buffer is marked dirty, its page is dirtied, but the page's other
1135 * Also. When blockdev buffers are explicitly read with bread(), they
1136 * individually become uptodate. But their backing page remains not
1137 * uptodate - even if all of its buffers are uptodate. A subsequent
1138 * block_read_full_page() against that page will discover all the uptodate
1139 * buffers, will set the page uptodate and will perform no I/O.
1143 * mark_buffer_dirty - mark a buffer_head as needing writeout
1144 * @bh: the buffer_head to mark dirty
1146 * mark_buffer_dirty() will set the dirty bit against the buffer, then set its
1147 * backing page dirty, then tag the page as dirty in its address_space's radix
1148 * tree and then attach the address_space's inode to its superblock's dirty
1151 * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
1152 * mapping->tree_lock and mapping->host->i_lock.
1154 void mark_buffer_dirty(struct buffer_head
*bh
)
1156 WARN_ON_ONCE(!buffer_uptodate(bh
));
1158 trace_block_dirty_buffer(bh
);
1161 * Very *carefully* optimize the it-is-already-dirty case.
1163 * Don't let the final "is it dirty" escape to before we
1164 * perhaps modified the buffer.
1166 if (buffer_dirty(bh
)) {
1168 if (buffer_dirty(bh
))
1172 if (!test_set_buffer_dirty(bh
)) {
1173 struct page
*page
= bh
->b_page
;
1174 struct address_space
*mapping
= NULL
;
1175 struct mem_cgroup
*memcg
;
1177 memcg
= mem_cgroup_begin_page_stat(page
);
1178 if (!TestSetPageDirty(page
)) {
1179 mapping
= page_mapping(page
);
1181 __set_page_dirty(page
, mapping
, memcg
, 0);
1183 mem_cgroup_end_page_stat(memcg
);
1185 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
1188 EXPORT_SYMBOL(mark_buffer_dirty
);
1191 * Decrement a buffer_head's reference count. If all buffers against a page
1192 * have zero reference count, are clean and unlocked, and if the page is clean
1193 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1194 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1195 * a page but it ends up not being freed, and buffers may later be reattached).
1197 void __brelse(struct buffer_head
* buf
)
1199 if (atomic_read(&buf
->b_count
)) {
1203 WARN(1, KERN_ERR
"VFS: brelse: Trying to free free buffer\n");
1205 EXPORT_SYMBOL(__brelse
);
1208 * bforget() is like brelse(), except it discards any
1209 * potentially dirty data.
1211 void __bforget(struct buffer_head
*bh
)
1213 clear_buffer_dirty(bh
);
1214 if (bh
->b_assoc_map
) {
1215 struct address_space
*buffer_mapping
= bh
->b_page
->mapping
;
1217 spin_lock(&buffer_mapping
->private_lock
);
1218 list_del_init(&bh
->b_assoc_buffers
);
1219 bh
->b_assoc_map
= NULL
;
1220 spin_unlock(&buffer_mapping
->private_lock
);
1224 EXPORT_SYMBOL(__bforget
);
1226 static struct buffer_head
*__bread_slow(struct buffer_head
*bh
)
1229 if (buffer_uptodate(bh
)) {
1234 bh
->b_end_io
= end_buffer_read_sync
;
1235 submit_bh(READ
, bh
);
1237 if (buffer_uptodate(bh
))
1245 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1246 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1247 * refcount elevated by one when they're in an LRU. A buffer can only appear
1248 * once in a particular CPU's LRU. A single buffer can be present in multiple
1249 * CPU's LRUs at the same time.
1251 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1252 * sb_find_get_block().
1254 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1255 * a local interrupt disable for that.
1258 #define BH_LRU_SIZE 16
1261 struct buffer_head
*bhs
[BH_LRU_SIZE
];
1264 static DEFINE_PER_CPU(struct bh_lru
, bh_lrus
) = {{ NULL
}};
1267 #define bh_lru_lock() local_irq_disable()
1268 #define bh_lru_unlock() local_irq_enable()
1270 #define bh_lru_lock() preempt_disable()
1271 #define bh_lru_unlock() preempt_enable()
1274 static inline void check_irqs_on(void)
1276 #ifdef irqs_disabled
1277 BUG_ON(irqs_disabled());
1282 * The LRU management algorithm is dopey-but-simple. Sorry.
1284 static void bh_lru_install(struct buffer_head
*bh
)
1286 struct buffer_head
*evictee
= NULL
;
1290 if (__this_cpu_read(bh_lrus
.bhs
[0]) != bh
) {
1291 struct buffer_head
*bhs
[BH_LRU_SIZE
];
1297 for (in
= 0; in
< BH_LRU_SIZE
; in
++) {
1298 struct buffer_head
*bh2
=
1299 __this_cpu_read(bh_lrus
.bhs
[in
]);
1304 if (out
>= BH_LRU_SIZE
) {
1305 BUG_ON(evictee
!= NULL
);
1312 while (out
< BH_LRU_SIZE
)
1314 memcpy(this_cpu_ptr(&bh_lrus
.bhs
), bhs
, sizeof(bhs
));
1323 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1325 static struct buffer_head
*
1326 lookup_bh_lru(struct block_device
*bdev
, sector_t block
, unsigned size
)
1328 struct buffer_head
*ret
= NULL
;
1333 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1334 struct buffer_head
*bh
= __this_cpu_read(bh_lrus
.bhs
[i
]);
1336 if (bh
&& bh
->b_blocknr
== block
&& bh
->b_bdev
== bdev
&&
1337 bh
->b_size
== size
) {
1340 __this_cpu_write(bh_lrus
.bhs
[i
],
1341 __this_cpu_read(bh_lrus
.bhs
[i
- 1]));
1344 __this_cpu_write(bh_lrus
.bhs
[0], bh
);
1356 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1357 * it in the LRU and mark it as accessed. If it is not present then return
1360 struct buffer_head
*
1361 __find_get_block(struct block_device
*bdev
, sector_t block
, unsigned size
)
1363 struct buffer_head
*bh
= lookup_bh_lru(bdev
, block
, size
);
1366 /* __find_get_block_slow will mark the page accessed */
1367 bh
= __find_get_block_slow(bdev
, block
);
1375 EXPORT_SYMBOL(__find_get_block
);
1378 * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
1379 * which corresponds to the passed block_device, block and size. The
1380 * returned buffer has its reference count incremented.
1382 * __getblk_gfp() will lock up the machine if grow_dev_page's
1383 * try_to_free_buffers() attempt is failing. FIXME, perhaps?
1385 struct buffer_head
*
1386 __getblk_gfp(struct block_device
*bdev
, sector_t block
,
1387 unsigned size
, gfp_t gfp
)
1389 struct buffer_head
*bh
= __find_get_block(bdev
, block
, size
);
1393 bh
= __getblk_slow(bdev
, block
, size
, gfp
);
1396 EXPORT_SYMBOL(__getblk_gfp
);
1399 * Do async read-ahead on a buffer..
1401 void __breadahead(struct block_device
*bdev
, sector_t block
, unsigned size
)
1403 struct buffer_head
*bh
= __getblk(bdev
, block
, size
);
1405 ll_rw_block(READA
, 1, &bh
);
1409 EXPORT_SYMBOL(__breadahead
);
1412 * __bread_gfp() - reads a specified block and returns the bh
1413 * @bdev: the block_device to read from
1414 * @block: number of block
1415 * @size: size (in bytes) to read
1416 * @gfp: page allocation flag
1418 * Reads a specified block, and returns buffer head that contains it.
1419 * The page cache can be allocated from non-movable area
1420 * not to prevent page migration if you set gfp to zero.
1421 * It returns NULL if the block was unreadable.
1423 struct buffer_head
*
1424 __bread_gfp(struct block_device
*bdev
, sector_t block
,
1425 unsigned size
, gfp_t gfp
)
1427 struct buffer_head
*bh
= __getblk_gfp(bdev
, block
, size
, gfp
);
1429 if (likely(bh
) && !buffer_uptodate(bh
))
1430 bh
= __bread_slow(bh
);
1433 EXPORT_SYMBOL(__bread_gfp
);
1436 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1437 * This doesn't race because it runs in each cpu either in irq
1438 * or with preempt disabled.
1440 static void invalidate_bh_lru(void *arg
)
1442 struct bh_lru
*b
= &get_cpu_var(bh_lrus
);
1445 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1449 put_cpu_var(bh_lrus
);
1452 static bool has_bh_in_lru(int cpu
, void *dummy
)
1454 struct bh_lru
*b
= per_cpu_ptr(&bh_lrus
, cpu
);
1457 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1465 void invalidate_bh_lrus(void)
1467 on_each_cpu_cond(has_bh_in_lru
, invalidate_bh_lru
, NULL
, 1, GFP_KERNEL
);
1469 EXPORT_SYMBOL_GPL(invalidate_bh_lrus
);
1471 void set_bh_page(struct buffer_head
*bh
,
1472 struct page
*page
, unsigned long offset
)
1475 BUG_ON(offset
>= PAGE_SIZE
);
1476 if (PageHighMem(page
))
1478 * This catches illegal uses and preserves the offset:
1480 bh
->b_data
= (char *)(0 + offset
);
1482 bh
->b_data
= page_address(page
) + offset
;
1484 EXPORT_SYMBOL(set_bh_page
);
1487 * Called when truncating a buffer on a page completely.
1490 /* Bits that are cleared during an invalidate */
1491 #define BUFFER_FLAGS_DISCARD \
1492 (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1493 1 << BH_Delay | 1 << BH_Unwritten)
1495 static void discard_buffer(struct buffer_head
* bh
)
1497 unsigned long b_state
, b_state_old
;
1500 clear_buffer_dirty(bh
);
1502 b_state
= bh
->b_state
;
1504 b_state_old
= cmpxchg(&bh
->b_state
, b_state
,
1505 (b_state
& ~BUFFER_FLAGS_DISCARD
));
1506 if (b_state_old
== b_state
)
1508 b_state
= b_state_old
;
1514 * block_invalidatepage - invalidate part or all of a buffer-backed page
1516 * @page: the page which is affected
1517 * @offset: start of the range to invalidate
1518 * @length: length of the range to invalidate
1520 * block_invalidatepage() is called when all or part of the page has become
1521 * invalidated by a truncate operation.
1523 * block_invalidatepage() does not have to release all buffers, but it must
1524 * ensure that no dirty buffer is left outside @offset and that no I/O
1525 * is underway against any of the blocks which are outside the truncation
1526 * point. Because the caller is about to free (and possibly reuse) those
1529 void block_invalidatepage(struct page
*page
, unsigned int offset
,
1530 unsigned int length
)
1532 struct buffer_head
*head
, *bh
, *next
;
1533 unsigned int curr_off
= 0;
1534 unsigned int stop
= length
+ offset
;
1536 BUG_ON(!PageLocked(page
));
1537 if (!page_has_buffers(page
))
1541 * Check for overflow
1543 BUG_ON(stop
> PAGE_CACHE_SIZE
|| stop
< length
);
1545 head
= page_buffers(page
);
1548 unsigned int next_off
= curr_off
+ bh
->b_size
;
1549 next
= bh
->b_this_page
;
1552 * Are we still fully in range ?
1554 if (next_off
> stop
)
1558 * is this block fully invalidated?
1560 if (offset
<= curr_off
)
1562 curr_off
= next_off
;
1564 } while (bh
!= head
);
1567 * We release buffers only if the entire page is being invalidated.
1568 * The get_block cached value has been unconditionally invalidated,
1569 * so real IO is not possible anymore.
1572 try_to_release_page(page
, 0);
1576 EXPORT_SYMBOL(block_invalidatepage
);
1580 * We attach and possibly dirty the buffers atomically wrt
1581 * __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1582 * is already excluded via the page lock.
1584 void create_empty_buffers(struct page
*page
,
1585 unsigned long blocksize
, unsigned long b_state
)
1587 struct buffer_head
*bh
, *head
, *tail
;
1589 head
= alloc_page_buffers(page
, blocksize
, 1);
1592 bh
->b_state
|= b_state
;
1594 bh
= bh
->b_this_page
;
1596 tail
->b_this_page
= head
;
1598 spin_lock(&page
->mapping
->private_lock
);
1599 if (PageUptodate(page
) || PageDirty(page
)) {
1602 if (PageDirty(page
))
1603 set_buffer_dirty(bh
);
1604 if (PageUptodate(page
))
1605 set_buffer_uptodate(bh
);
1606 bh
= bh
->b_this_page
;
1607 } while (bh
!= head
);
1609 attach_page_buffers(page
, head
);
1610 spin_unlock(&page
->mapping
->private_lock
);
1612 EXPORT_SYMBOL(create_empty_buffers
);
1615 * We are taking a block for data and we don't want any output from any
1616 * buffer-cache aliases starting from return from that function and
1617 * until the moment when something will explicitly mark the buffer
1618 * dirty (hopefully that will not happen until we will free that block ;-)
1619 * We don't even need to mark it not-uptodate - nobody can expect
1620 * anything from a newly allocated buffer anyway. We used to used
1621 * unmap_buffer() for such invalidation, but that was wrong. We definitely
1622 * don't want to mark the alias unmapped, for example - it would confuse
1623 * anyone who might pick it with bread() afterwards...
1625 * Also.. Note that bforget() doesn't lock the buffer. So there can
1626 * be writeout I/O going on against recently-freed buffers. We don't
1627 * wait on that I/O in bforget() - it's more efficient to wait on the I/O
1628 * only if we really need to. That happens here.
1630 void unmap_underlying_metadata(struct block_device
*bdev
, sector_t block
)
1632 struct buffer_head
*old_bh
;
1636 old_bh
= __find_get_block_slow(bdev
, block
);
1638 clear_buffer_dirty(old_bh
);
1639 wait_on_buffer(old_bh
);
1640 clear_buffer_req(old_bh
);
1644 EXPORT_SYMBOL(unmap_underlying_metadata
);
1647 * Size is a power-of-two in the range 512..PAGE_SIZE,
1648 * and the case we care about most is PAGE_SIZE.
1650 * So this *could* possibly be written with those
1651 * constraints in mind (relevant mostly if some
1652 * architecture has a slow bit-scan instruction)
1654 static inline int block_size_bits(unsigned int blocksize
)
1656 return ilog2(blocksize
);
1659 static struct buffer_head
*create_page_buffers(struct page
*page
, struct inode
*inode
, unsigned int b_state
)
1661 BUG_ON(!PageLocked(page
));
1663 if (!page_has_buffers(page
))
1664 create_empty_buffers(page
, 1 << ACCESS_ONCE(inode
->i_blkbits
), b_state
);
1665 return page_buffers(page
);
1669 * NOTE! All mapped/uptodate combinations are valid:
1671 * Mapped Uptodate Meaning
1673 * No No "unknown" - must do get_block()
1674 * No Yes "hole" - zero-filled
1675 * Yes No "allocated" - allocated on disk, not read in
1676 * Yes Yes "valid" - allocated and up-to-date in memory.
1678 * "Dirty" is valid only with the last case (mapped+uptodate).
1682 * While block_write_full_page is writing back the dirty buffers under
1683 * the page lock, whoever dirtied the buffers may decide to clean them
1684 * again at any time. We handle that by only looking at the buffer
1685 * state inside lock_buffer().
1687 * If block_write_full_page() is called for regular writeback
1688 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1689 * locked buffer. This only can happen if someone has written the buffer
1690 * directly, with submit_bh(). At the address_space level PageWriteback
1691 * prevents this contention from occurring.
1693 * If block_write_full_page() is called with wbc->sync_mode ==
1694 * WB_SYNC_ALL, the writes are posted using WRITE_SYNC; this
1695 * causes the writes to be flagged as synchronous writes.
1697 static int __block_write_full_page(struct inode
*inode
, struct page
*page
,
1698 get_block_t
*get_block
, struct writeback_control
*wbc
,
1699 bh_end_io_t
*handler
)
1703 sector_t last_block
;
1704 struct buffer_head
*bh
, *head
;
1705 unsigned int blocksize
, bbits
;
1706 int nr_underway
= 0;
1707 int write_op
= (wbc
->sync_mode
== WB_SYNC_ALL
?
1708 WRITE_SYNC
: WRITE
);
1710 head
= create_page_buffers(page
, inode
,
1711 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
1714 * Be very careful. We have no exclusion from __set_page_dirty_buffers
1715 * here, and the (potentially unmapped) buffers may become dirty at
1716 * any time. If a buffer becomes dirty here after we've inspected it
1717 * then we just miss that fact, and the page stays dirty.
1719 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1720 * handle that here by just cleaning them.
1724 blocksize
= bh
->b_size
;
1725 bbits
= block_size_bits(blocksize
);
1727 block
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- bbits
);
1728 last_block
= (i_size_read(inode
) - 1) >> bbits
;
1731 * Get all the dirty buffers mapped to disk addresses and
1732 * handle any aliases from the underlying blockdev's mapping.
1735 if (block
> last_block
) {
1737 * mapped buffers outside i_size will occur, because
1738 * this page can be outside i_size when there is a
1739 * truncate in progress.
1742 * The buffer was zeroed by block_write_full_page()
1744 clear_buffer_dirty(bh
);
1745 set_buffer_uptodate(bh
);
1746 } else if ((!buffer_mapped(bh
) || buffer_delay(bh
)) &&
1748 WARN_ON(bh
->b_size
!= blocksize
);
1749 err
= get_block(inode
, block
, bh
, 1);
1752 clear_buffer_delay(bh
);
1753 if (buffer_new(bh
)) {
1754 /* blockdev mappings never come here */
1755 clear_buffer_new(bh
);
1756 unmap_underlying_metadata(bh
->b_bdev
,
1760 bh
= bh
->b_this_page
;
1762 } while (bh
!= head
);
1765 if (!buffer_mapped(bh
))
1768 * If it's a fully non-blocking write attempt and we cannot
1769 * lock the buffer then redirty the page. Note that this can
1770 * potentially cause a busy-wait loop from writeback threads
1771 * and kswapd activity, but those code paths have their own
1772 * higher-level throttling.
1774 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
1776 } else if (!trylock_buffer(bh
)) {
1777 redirty_page_for_writepage(wbc
, page
);
1780 if (test_clear_buffer_dirty(bh
)) {
1781 mark_buffer_async_write_endio(bh
, handler
);
1785 } while ((bh
= bh
->b_this_page
) != head
);
1788 * The page and its buffers are protected by PageWriteback(), so we can
1789 * drop the bh refcounts early.
1791 BUG_ON(PageWriteback(page
));
1792 set_page_writeback(page
);
1795 struct buffer_head
*next
= bh
->b_this_page
;
1796 if (buffer_async_write(bh
)) {
1797 submit_bh(write_op
, bh
);
1801 } while (bh
!= head
);
1806 if (nr_underway
== 0) {
1808 * The page was marked dirty, but the buffers were
1809 * clean. Someone wrote them back by hand with
1810 * ll_rw_block/submit_bh. A rare case.
1812 end_page_writeback(page
);
1815 * The page and buffer_heads can be released at any time from
1823 * ENOSPC, or some other error. We may already have added some
1824 * blocks to the file, so we need to write these out to avoid
1825 * exposing stale data.
1826 * The page is currently locked and not marked for writeback
1829 /* Recovery: lock and submit the mapped buffers */
1831 if (buffer_mapped(bh
) && buffer_dirty(bh
) &&
1832 !buffer_delay(bh
)) {
1834 mark_buffer_async_write_endio(bh
, handler
);
1837 * The buffer may have been set dirty during
1838 * attachment to a dirty page.
1840 clear_buffer_dirty(bh
);
1842 } while ((bh
= bh
->b_this_page
) != head
);
1844 BUG_ON(PageWriteback(page
));
1845 mapping_set_error(page
->mapping
, err
);
1846 set_page_writeback(page
);
1848 struct buffer_head
*next
= bh
->b_this_page
;
1849 if (buffer_async_write(bh
)) {
1850 clear_buffer_dirty(bh
);
1851 submit_bh(write_op
, bh
);
1855 } while (bh
!= head
);
1861 * If a page has any new buffers, zero them out here, and mark them uptodate
1862 * and dirty so they'll be written out (in order to prevent uninitialised
1863 * block data from leaking). And clear the new bit.
1865 void page_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1867 unsigned int block_start
, block_end
;
1868 struct buffer_head
*head
, *bh
;
1870 BUG_ON(!PageLocked(page
));
1871 if (!page_has_buffers(page
))
1874 bh
= head
= page_buffers(page
);
1877 block_end
= block_start
+ bh
->b_size
;
1879 if (buffer_new(bh
)) {
1880 if (block_end
> from
&& block_start
< to
) {
1881 if (!PageUptodate(page
)) {
1882 unsigned start
, size
;
1884 start
= max(from
, block_start
);
1885 size
= min(to
, block_end
) - start
;
1887 zero_user(page
, start
, size
);
1888 set_buffer_uptodate(bh
);
1891 clear_buffer_new(bh
);
1892 mark_buffer_dirty(bh
);
1896 block_start
= block_end
;
1897 bh
= bh
->b_this_page
;
1898 } while (bh
!= head
);
1900 EXPORT_SYMBOL(page_zero_new_buffers
);
1902 int __block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1903 get_block_t
*get_block
)
1905 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1906 unsigned to
= from
+ len
;
1907 struct inode
*inode
= page
->mapping
->host
;
1908 unsigned block_start
, block_end
;
1911 unsigned blocksize
, bbits
;
1912 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
=wait
;
1914 BUG_ON(!PageLocked(page
));
1915 BUG_ON(from
> PAGE_CACHE_SIZE
);
1916 BUG_ON(to
> PAGE_CACHE_SIZE
);
1919 head
= create_page_buffers(page
, inode
, 0);
1920 blocksize
= head
->b_size
;
1921 bbits
= block_size_bits(blocksize
);
1923 block
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- bbits
);
1925 for(bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1926 block
++, block_start
=block_end
, bh
= bh
->b_this_page
) {
1927 block_end
= block_start
+ blocksize
;
1928 if (block_end
<= from
|| block_start
>= to
) {
1929 if (PageUptodate(page
)) {
1930 if (!buffer_uptodate(bh
))
1931 set_buffer_uptodate(bh
);
1936 clear_buffer_new(bh
);
1937 if (!buffer_mapped(bh
)) {
1938 WARN_ON(bh
->b_size
!= blocksize
);
1939 err
= get_block(inode
, block
, bh
, 1);
1942 if (buffer_new(bh
)) {
1943 unmap_underlying_metadata(bh
->b_bdev
,
1945 if (PageUptodate(page
)) {
1946 clear_buffer_new(bh
);
1947 set_buffer_uptodate(bh
);
1948 mark_buffer_dirty(bh
);
1951 if (block_end
> to
|| block_start
< from
)
1952 zero_user_segments(page
,
1958 if (PageUptodate(page
)) {
1959 if (!buffer_uptodate(bh
))
1960 set_buffer_uptodate(bh
);
1963 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1964 !buffer_unwritten(bh
) &&
1965 (block_start
< from
|| block_end
> to
)) {
1966 ll_rw_block(READ
, 1, &bh
);
1971 * If we issued read requests - let them complete.
1973 while(wait_bh
> wait
) {
1974 wait_on_buffer(*--wait_bh
);
1975 if (!buffer_uptodate(*wait_bh
))
1979 page_zero_new_buffers(page
, from
, to
);
1982 EXPORT_SYMBOL(__block_write_begin
);
1984 static int __block_commit_write(struct inode
*inode
, struct page
*page
,
1985 unsigned from
, unsigned to
)
1987 unsigned block_start
, block_end
;
1990 struct buffer_head
*bh
, *head
;
1992 bh
= head
= page_buffers(page
);
1993 blocksize
= bh
->b_size
;
1997 block_end
= block_start
+ blocksize
;
1998 if (block_end
<= from
|| block_start
>= to
) {
1999 if (!buffer_uptodate(bh
))
2002 set_buffer_uptodate(bh
);
2003 mark_buffer_dirty(bh
);
2005 clear_buffer_new(bh
);
2007 block_start
= block_end
;
2008 bh
= bh
->b_this_page
;
2009 } while (bh
!= head
);
2012 * If this is a partial write which happened to make all buffers
2013 * uptodate then we can optimize away a bogus readpage() for
2014 * the next read(). Here we 'discover' whether the page went
2015 * uptodate as a result of this (potentially partial) write.
2018 SetPageUptodate(page
);
2023 * block_write_begin takes care of the basic task of block allocation and
2024 * bringing partial write blocks uptodate first.
2026 * The filesystem needs to handle block truncation upon failure.
2028 int block_write_begin(struct address_space
*mapping
, loff_t pos
, unsigned len
,
2029 unsigned flags
, struct page
**pagep
, get_block_t
*get_block
)
2031 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
2035 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2039 status
= __block_write_begin(page
, pos
, len
, get_block
);
2040 if (unlikely(status
)) {
2042 page_cache_release(page
);
2049 EXPORT_SYMBOL(block_write_begin
);
2051 int block_write_end(struct file
*file
, struct address_space
*mapping
,
2052 loff_t pos
, unsigned len
, unsigned copied
,
2053 struct page
*page
, void *fsdata
)
2055 struct inode
*inode
= mapping
->host
;
2058 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2060 if (unlikely(copied
< len
)) {
2062 * The buffers that were written will now be uptodate, so we
2063 * don't have to worry about a readpage reading them and
2064 * overwriting a partial write. However if we have encountered
2065 * a short write and only partially written into a buffer, it
2066 * will not be marked uptodate, so a readpage might come in and
2067 * destroy our partial write.
2069 * Do the simplest thing, and just treat any short write to a
2070 * non uptodate page as a zero-length write, and force the
2071 * caller to redo the whole thing.
2073 if (!PageUptodate(page
))
2076 page_zero_new_buffers(page
, start
+copied
, start
+len
);
2078 flush_dcache_page(page
);
2080 /* This could be a short (even 0-length) commit */
2081 __block_commit_write(inode
, page
, start
, start
+copied
);
2085 EXPORT_SYMBOL(block_write_end
);
2087 int generic_write_end(struct file
*file
, struct address_space
*mapping
,
2088 loff_t pos
, unsigned len
, unsigned copied
,
2089 struct page
*page
, void *fsdata
)
2091 struct inode
*inode
= mapping
->host
;
2092 loff_t old_size
= inode
->i_size
;
2093 int i_size_changed
= 0;
2095 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2098 * No need to use i_size_read() here, the i_size
2099 * cannot change under us because we hold i_mutex.
2101 * But it's important to update i_size while still holding page lock:
2102 * page writeout could otherwise come in and zero beyond i_size.
2104 if (pos
+copied
> inode
->i_size
) {
2105 i_size_write(inode
, pos
+copied
);
2110 page_cache_release(page
);
2113 pagecache_isize_extended(inode
, old_size
, pos
);
2115 * Don't mark the inode dirty under page lock. First, it unnecessarily
2116 * makes the holding time of page lock longer. Second, it forces lock
2117 * ordering of page lock and transaction start for journaling
2121 mark_inode_dirty(inode
);
2125 EXPORT_SYMBOL(generic_write_end
);
2128 * block_is_partially_uptodate checks whether buffers within a page are
2131 * Returns true if all buffers which correspond to a file portion
2132 * we want to read are uptodate.
2134 int block_is_partially_uptodate(struct page
*page
, unsigned long from
,
2135 unsigned long count
)
2137 unsigned block_start
, block_end
, blocksize
;
2139 struct buffer_head
*bh
, *head
;
2142 if (!page_has_buffers(page
))
2145 head
= page_buffers(page
);
2146 blocksize
= head
->b_size
;
2147 to
= min_t(unsigned, PAGE_CACHE_SIZE
- from
, count
);
2149 if (from
< blocksize
&& to
> PAGE_CACHE_SIZE
- blocksize
)
2155 block_end
= block_start
+ blocksize
;
2156 if (block_end
> from
&& block_start
< to
) {
2157 if (!buffer_uptodate(bh
)) {
2161 if (block_end
>= to
)
2164 block_start
= block_end
;
2165 bh
= bh
->b_this_page
;
2166 } while (bh
!= head
);
2170 EXPORT_SYMBOL(block_is_partially_uptodate
);
2173 * Generic "read page" function for block devices that have the normal
2174 * get_block functionality. This is most of the block device filesystems.
2175 * Reads the page asynchronously --- the unlock_buffer() and
2176 * set/clear_buffer_uptodate() functions propagate buffer state into the
2177 * page struct once IO has completed.
2179 int block_read_full_page(struct page
*page
, get_block_t
*get_block
)
2181 struct inode
*inode
= page
->mapping
->host
;
2182 sector_t iblock
, lblock
;
2183 struct buffer_head
*bh
, *head
, *arr
[MAX_BUF_PER_PAGE
];
2184 unsigned int blocksize
, bbits
;
2186 int fully_mapped
= 1;
2188 head
= create_page_buffers(page
, inode
, 0);
2189 blocksize
= head
->b_size
;
2190 bbits
= block_size_bits(blocksize
);
2192 iblock
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- bbits
);
2193 lblock
= (i_size_read(inode
)+blocksize
-1) >> bbits
;
2199 if (buffer_uptodate(bh
))
2202 if (!buffer_mapped(bh
)) {
2206 if (iblock
< lblock
) {
2207 WARN_ON(bh
->b_size
!= blocksize
);
2208 err
= get_block(inode
, iblock
, bh
, 0);
2212 if (!buffer_mapped(bh
)) {
2213 zero_user(page
, i
* blocksize
, blocksize
);
2215 set_buffer_uptodate(bh
);
2219 * get_block() might have updated the buffer
2222 if (buffer_uptodate(bh
))
2226 } while (i
++, iblock
++, (bh
= bh
->b_this_page
) != head
);
2229 SetPageMappedToDisk(page
);
2233 * All buffers are uptodate - we can set the page uptodate
2234 * as well. But not if get_block() returned an error.
2236 if (!PageError(page
))
2237 SetPageUptodate(page
);
2242 /* Stage two: lock the buffers */
2243 for (i
= 0; i
< nr
; i
++) {
2246 mark_buffer_async_read(bh
);
2250 * Stage 3: start the IO. Check for uptodateness
2251 * inside the buffer lock in case another process reading
2252 * the underlying blockdev brought it uptodate (the sct fix).
2254 for (i
= 0; i
< nr
; i
++) {
2256 if (buffer_uptodate(bh
))
2257 end_buffer_async_read(bh
, 1);
2259 submit_bh(READ
, bh
);
2263 EXPORT_SYMBOL(block_read_full_page
);
2265 /* utility function for filesystems that need to do work on expanding
2266 * truncates. Uses filesystem pagecache writes to allow the filesystem to
2267 * deal with the hole.
2269 int generic_cont_expand_simple(struct inode
*inode
, loff_t size
)
2271 struct address_space
*mapping
= inode
->i_mapping
;
2276 err
= inode_newsize_ok(inode
, size
);
2280 err
= pagecache_write_begin(NULL
, mapping
, size
, 0,
2281 AOP_FLAG_UNINTERRUPTIBLE
|AOP_FLAG_CONT_EXPAND
,
2286 err
= pagecache_write_end(NULL
, mapping
, size
, 0, 0, page
, fsdata
);
2292 EXPORT_SYMBOL(generic_cont_expand_simple
);
2294 static int cont_expand_zero(struct file
*file
, struct address_space
*mapping
,
2295 loff_t pos
, loff_t
*bytes
)
2297 struct inode
*inode
= mapping
->host
;
2298 unsigned blocksize
= 1 << inode
->i_blkbits
;
2301 pgoff_t index
, curidx
;
2303 unsigned zerofrom
, offset
, len
;
2306 index
= pos
>> PAGE_CACHE_SHIFT
;
2307 offset
= pos
& ~PAGE_CACHE_MASK
;
2309 while (index
> (curidx
= (curpos
= *bytes
)>>PAGE_CACHE_SHIFT
)) {
2310 zerofrom
= curpos
& ~PAGE_CACHE_MASK
;
2311 if (zerofrom
& (blocksize
-1)) {
2312 *bytes
|= (blocksize
-1);
2315 len
= PAGE_CACHE_SIZE
- zerofrom
;
2317 err
= pagecache_write_begin(file
, mapping
, curpos
, len
,
2318 AOP_FLAG_UNINTERRUPTIBLE
,
2322 zero_user(page
, zerofrom
, len
);
2323 err
= pagecache_write_end(file
, mapping
, curpos
, len
, len
,
2330 balance_dirty_pages_ratelimited(mapping
);
2332 if (unlikely(fatal_signal_pending(current
))) {
2338 /* page covers the boundary, find the boundary offset */
2339 if (index
== curidx
) {
2340 zerofrom
= curpos
& ~PAGE_CACHE_MASK
;
2341 /* if we will expand the thing last block will be filled */
2342 if (offset
<= zerofrom
) {
2345 if (zerofrom
& (blocksize
-1)) {
2346 *bytes
|= (blocksize
-1);
2349 len
= offset
- zerofrom
;
2351 err
= pagecache_write_begin(file
, mapping
, curpos
, len
,
2352 AOP_FLAG_UNINTERRUPTIBLE
,
2356 zero_user(page
, zerofrom
, len
);
2357 err
= pagecache_write_end(file
, mapping
, curpos
, len
, len
,
2369 * For moronic filesystems that do not allow holes in file.
2370 * We may have to extend the file.
2372 int cont_write_begin(struct file
*file
, struct address_space
*mapping
,
2373 loff_t pos
, unsigned len
, unsigned flags
,
2374 struct page
**pagep
, void **fsdata
,
2375 get_block_t
*get_block
, loff_t
*bytes
)
2377 struct inode
*inode
= mapping
->host
;
2378 unsigned blocksize
= 1 << inode
->i_blkbits
;
2382 err
= cont_expand_zero(file
, mapping
, pos
, bytes
);
2386 zerofrom
= *bytes
& ~PAGE_CACHE_MASK
;
2387 if (pos
+len
> *bytes
&& zerofrom
& (blocksize
-1)) {
2388 *bytes
|= (blocksize
-1);
2392 return block_write_begin(mapping
, pos
, len
, flags
, pagep
, get_block
);
2394 EXPORT_SYMBOL(cont_write_begin
);
2396 int block_commit_write(struct page
*page
, unsigned from
, unsigned to
)
2398 struct inode
*inode
= page
->mapping
->host
;
2399 __block_commit_write(inode
,page
,from
,to
);
2402 EXPORT_SYMBOL(block_commit_write
);
2405 * block_page_mkwrite() is not allowed to change the file size as it gets
2406 * called from a page fault handler when a page is first dirtied. Hence we must
2407 * be careful to check for EOF conditions here. We set the page up correctly
2408 * for a written page which means we get ENOSPC checking when writing into
2409 * holes and correct delalloc and unwritten extent mapping on filesystems that
2410 * support these features.
2412 * We are not allowed to take the i_mutex here so we have to play games to
2413 * protect against truncate races as the page could now be beyond EOF. Because
2414 * truncate writes the inode size before removing pages, once we have the
2415 * page lock we can determine safely if the page is beyond EOF. If it is not
2416 * beyond EOF, then the page is guaranteed safe against truncation until we
2419 * Direct callers of this function should protect against filesystem freezing
2420 * using sb_start_write() - sb_end_write() functions.
2422 int __block_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
2423 get_block_t get_block
)
2425 struct page
*page
= vmf
->page
;
2426 struct inode
*inode
= file_inode(vma
->vm_file
);
2432 size
= i_size_read(inode
);
2433 if ((page
->mapping
!= inode
->i_mapping
) ||
2434 (page_offset(page
) > size
)) {
2435 /* We overload EFAULT to mean page got truncated */
2440 /* page is wholly or partially inside EOF */
2441 if (((page
->index
+ 1) << PAGE_CACHE_SHIFT
) > size
)
2442 end
= size
& ~PAGE_CACHE_MASK
;
2444 end
= PAGE_CACHE_SIZE
;
2446 ret
= __block_write_begin(page
, 0, end
, get_block
);
2448 ret
= block_commit_write(page
, 0, end
);
2450 if (unlikely(ret
< 0))
2452 set_page_dirty(page
);
2453 wait_for_stable_page(page
);
2459 EXPORT_SYMBOL(__block_page_mkwrite
);
2461 int block_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
2462 get_block_t get_block
)
2465 struct super_block
*sb
= file_inode(vma
->vm_file
)->i_sb
;
2467 sb_start_pagefault(sb
);
2470 * Update file times before taking page lock. We may end up failing the
2471 * fault so this update may be superfluous but who really cares...
2473 file_update_time(vma
->vm_file
);
2475 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
2476 sb_end_pagefault(sb
);
2477 return block_page_mkwrite_return(ret
);
2479 EXPORT_SYMBOL(block_page_mkwrite
);
2482 * nobh_write_begin()'s prereads are special: the buffer_heads are freed
2483 * immediately, while under the page lock. So it needs a special end_io
2484 * handler which does not touch the bh after unlocking it.
2486 static void end_buffer_read_nobh(struct buffer_head
*bh
, int uptodate
)
2488 __end_buffer_read_notouch(bh
, uptodate
);
2492 * Attach the singly-linked list of buffers created by nobh_write_begin, to
2493 * the page (converting it to circular linked list and taking care of page
2496 static void attach_nobh_buffers(struct page
*page
, struct buffer_head
*head
)
2498 struct buffer_head
*bh
;
2500 BUG_ON(!PageLocked(page
));
2502 spin_lock(&page
->mapping
->private_lock
);
2505 if (PageDirty(page
))
2506 set_buffer_dirty(bh
);
2507 if (!bh
->b_this_page
)
2508 bh
->b_this_page
= head
;
2509 bh
= bh
->b_this_page
;
2510 } while (bh
!= head
);
2511 attach_page_buffers(page
, head
);
2512 spin_unlock(&page
->mapping
->private_lock
);
2516 * On entry, the page is fully not uptodate.
2517 * On exit the page is fully uptodate in the areas outside (from,to)
2518 * The filesystem needs to handle block truncation upon failure.
2520 int nobh_write_begin(struct address_space
*mapping
,
2521 loff_t pos
, unsigned len
, unsigned flags
,
2522 struct page
**pagep
, void **fsdata
,
2523 get_block_t
*get_block
)
2525 struct inode
*inode
= mapping
->host
;
2526 const unsigned blkbits
= inode
->i_blkbits
;
2527 const unsigned blocksize
= 1 << blkbits
;
2528 struct buffer_head
*head
, *bh
;
2532 unsigned block_in_page
;
2533 unsigned block_start
, block_end
;
2534 sector_t block_in_file
;
2537 int is_mapped_to_disk
= 1;
2539 index
= pos
>> PAGE_CACHE_SHIFT
;
2540 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2543 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2549 if (page_has_buffers(page
)) {
2550 ret
= __block_write_begin(page
, pos
, len
, get_block
);
2556 if (PageMappedToDisk(page
))
2560 * Allocate buffers so that we can keep track of state, and potentially
2561 * attach them to the page if an error occurs. In the common case of
2562 * no error, they will just be freed again without ever being attached
2563 * to the page (which is all OK, because we're under the page lock).
2565 * Be careful: the buffer linked list is a NULL terminated one, rather
2566 * than the circular one we're used to.
2568 head
= alloc_page_buffers(page
, blocksize
, 0);
2574 block_in_file
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- blkbits
);
2577 * We loop across all blocks in the page, whether or not they are
2578 * part of the affected region. This is so we can discover if the
2579 * page is fully mapped-to-disk.
2581 for (block_start
= 0, block_in_page
= 0, bh
= head
;
2582 block_start
< PAGE_CACHE_SIZE
;
2583 block_in_page
++, block_start
+= blocksize
, bh
= bh
->b_this_page
) {
2586 block_end
= block_start
+ blocksize
;
2589 if (block_start
>= to
)
2591 ret
= get_block(inode
, block_in_file
+ block_in_page
,
2595 if (!buffer_mapped(bh
))
2596 is_mapped_to_disk
= 0;
2598 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
2599 if (PageUptodate(page
)) {
2600 set_buffer_uptodate(bh
);
2603 if (buffer_new(bh
) || !buffer_mapped(bh
)) {
2604 zero_user_segments(page
, block_start
, from
,
2608 if (buffer_uptodate(bh
))
2609 continue; /* reiserfs does this */
2610 if (block_start
< from
|| block_end
> to
) {
2612 bh
->b_end_io
= end_buffer_read_nobh
;
2613 submit_bh(READ
, bh
);
2620 * The page is locked, so these buffers are protected from
2621 * any VM or truncate activity. Hence we don't need to care
2622 * for the buffer_head refcounts.
2624 for (bh
= head
; bh
; bh
= bh
->b_this_page
) {
2626 if (!buffer_uptodate(bh
))
2633 if (is_mapped_to_disk
)
2634 SetPageMappedToDisk(page
);
2636 *fsdata
= head
; /* to be released by nobh_write_end */
2643 * Error recovery is a bit difficult. We need to zero out blocks that
2644 * were newly allocated, and dirty them to ensure they get written out.
2645 * Buffers need to be attached to the page at this point, otherwise
2646 * the handling of potential IO errors during writeout would be hard
2647 * (could try doing synchronous writeout, but what if that fails too?)
2649 attach_nobh_buffers(page
, head
);
2650 page_zero_new_buffers(page
, from
, to
);
2654 page_cache_release(page
);
2659 EXPORT_SYMBOL(nobh_write_begin
);
2661 int nobh_write_end(struct file
*file
, struct address_space
*mapping
,
2662 loff_t pos
, unsigned len
, unsigned copied
,
2663 struct page
*page
, void *fsdata
)
2665 struct inode
*inode
= page
->mapping
->host
;
2666 struct buffer_head
*head
= fsdata
;
2667 struct buffer_head
*bh
;
2668 BUG_ON(fsdata
!= NULL
&& page_has_buffers(page
));
2670 if (unlikely(copied
< len
) && head
)
2671 attach_nobh_buffers(page
, head
);
2672 if (page_has_buffers(page
))
2673 return generic_write_end(file
, mapping
, pos
, len
,
2674 copied
, page
, fsdata
);
2676 SetPageUptodate(page
);
2677 set_page_dirty(page
);
2678 if (pos
+copied
> inode
->i_size
) {
2679 i_size_write(inode
, pos
+copied
);
2680 mark_inode_dirty(inode
);
2684 page_cache_release(page
);
2688 head
= head
->b_this_page
;
2689 free_buffer_head(bh
);
2694 EXPORT_SYMBOL(nobh_write_end
);
2697 * nobh_writepage() - based on block_full_write_page() except
2698 * that it tries to operate without attaching bufferheads to
2701 int nobh_writepage(struct page
*page
, get_block_t
*get_block
,
2702 struct writeback_control
*wbc
)
2704 struct inode
* const inode
= page
->mapping
->host
;
2705 loff_t i_size
= i_size_read(inode
);
2706 const pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2710 /* Is the page fully inside i_size? */
2711 if (page
->index
< end_index
)
2714 /* Is the page fully outside i_size? (truncate in progress) */
2715 offset
= i_size
& (PAGE_CACHE_SIZE
-1);
2716 if (page
->index
>= end_index
+1 || !offset
) {
2718 * The page may have dirty, unmapped buffers. For example,
2719 * they may have been added in ext3_writepage(). Make them
2720 * freeable here, so the page does not leak.
2723 /* Not really sure about this - do we need this ? */
2724 if (page
->mapping
->a_ops
->invalidatepage
)
2725 page
->mapping
->a_ops
->invalidatepage(page
, offset
);
2728 return 0; /* don't care */
2732 * The page straddles i_size. It must be zeroed out on each and every
2733 * writepage invocation because it may be mmapped. "A file is mapped
2734 * in multiples of the page size. For a file that is not a multiple of
2735 * the page size, the remaining memory is zeroed when mapped, and
2736 * writes to that region are not written out to the file."
2738 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
2740 ret
= mpage_writepage(page
, get_block
, wbc
);
2742 ret
= __block_write_full_page(inode
, page
, get_block
, wbc
,
2743 end_buffer_async_write
);
2746 EXPORT_SYMBOL(nobh_writepage
);
2748 int nobh_truncate_page(struct address_space
*mapping
,
2749 loff_t from
, get_block_t
*get_block
)
2751 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
2752 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2755 unsigned length
, pos
;
2756 struct inode
*inode
= mapping
->host
;
2758 struct buffer_head map_bh
;
2761 blocksize
= 1 << inode
->i_blkbits
;
2762 length
= offset
& (blocksize
- 1);
2764 /* Block boundary? Nothing to do */
2768 length
= blocksize
- length
;
2769 iblock
= (sector_t
)index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2771 page
= grab_cache_page(mapping
, index
);
2776 if (page_has_buffers(page
)) {
2779 page_cache_release(page
);
2780 return block_truncate_page(mapping
, from
, get_block
);
2783 /* Find the buffer that contains "offset" */
2785 while (offset
>= pos
) {
2790 map_bh
.b_size
= blocksize
;
2792 err
= get_block(inode
, iblock
, &map_bh
, 0);
2795 /* unmapped? It's a hole - nothing to do */
2796 if (!buffer_mapped(&map_bh
))
2799 /* Ok, it's mapped. Make sure it's up-to-date */
2800 if (!PageUptodate(page
)) {
2801 err
= mapping
->a_ops
->readpage(NULL
, page
);
2803 page_cache_release(page
);
2807 if (!PageUptodate(page
)) {
2811 if (page_has_buffers(page
))
2814 zero_user(page
, offset
, length
);
2815 set_page_dirty(page
);
2820 page_cache_release(page
);
2824 EXPORT_SYMBOL(nobh_truncate_page
);
2826 int block_truncate_page(struct address_space
*mapping
,
2827 loff_t from
, get_block_t
*get_block
)
2829 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
2830 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2833 unsigned length
, pos
;
2834 struct inode
*inode
= mapping
->host
;
2836 struct buffer_head
*bh
;
2839 blocksize
= 1 << inode
->i_blkbits
;
2840 length
= offset
& (blocksize
- 1);
2842 /* Block boundary? Nothing to do */
2846 length
= blocksize
- length
;
2847 iblock
= (sector_t
)index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2849 page
= grab_cache_page(mapping
, index
);
2854 if (!page_has_buffers(page
))
2855 create_empty_buffers(page
, blocksize
, 0);
2857 /* Find the buffer that contains "offset" */
2858 bh
= page_buffers(page
);
2860 while (offset
>= pos
) {
2861 bh
= bh
->b_this_page
;
2867 if (!buffer_mapped(bh
)) {
2868 WARN_ON(bh
->b_size
!= blocksize
);
2869 err
= get_block(inode
, iblock
, bh
, 0);
2872 /* unmapped? It's a hole - nothing to do */
2873 if (!buffer_mapped(bh
))
2877 /* Ok, it's mapped. Make sure it's up-to-date */
2878 if (PageUptodate(page
))
2879 set_buffer_uptodate(bh
);
2881 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) && !buffer_unwritten(bh
)) {
2883 ll_rw_block(READ
, 1, &bh
);
2885 /* Uhhuh. Read error. Complain and punt. */
2886 if (!buffer_uptodate(bh
))
2890 zero_user(page
, offset
, length
);
2891 mark_buffer_dirty(bh
);
2896 page_cache_release(page
);
2900 EXPORT_SYMBOL(block_truncate_page
);
2903 * The generic ->writepage function for buffer-backed address_spaces
2905 int block_write_full_page(struct page
*page
, get_block_t
*get_block
,
2906 struct writeback_control
*wbc
)
2908 struct inode
* const inode
= page
->mapping
->host
;
2909 loff_t i_size
= i_size_read(inode
);
2910 const pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2913 /* Is the page fully inside i_size? */
2914 if (page
->index
< end_index
)
2915 return __block_write_full_page(inode
, page
, get_block
, wbc
,
2916 end_buffer_async_write
);
2918 /* Is the page fully outside i_size? (truncate in progress) */
2919 offset
= i_size
& (PAGE_CACHE_SIZE
-1);
2920 if (page
->index
>= end_index
+1 || !offset
) {
2922 * The page may have dirty, unmapped buffers. For example,
2923 * they may have been added in ext3_writepage(). Make them
2924 * freeable here, so the page does not leak.
2926 do_invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
2928 return 0; /* don't care */
2932 * The page straddles i_size. It must be zeroed out on each and every
2933 * writepage invocation because it may be mmapped. "A file is mapped
2934 * in multiples of the page size. For a file that is not a multiple of
2935 * the page size, the remaining memory is zeroed when mapped, and
2936 * writes to that region are not written out to the file."
2938 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
2939 return __block_write_full_page(inode
, page
, get_block
, wbc
,
2940 end_buffer_async_write
);
2942 EXPORT_SYMBOL(block_write_full_page
);
2944 sector_t
generic_block_bmap(struct address_space
*mapping
, sector_t block
,
2945 get_block_t
*get_block
)
2947 struct buffer_head tmp
;
2948 struct inode
*inode
= mapping
->host
;
2951 tmp
.b_size
= 1 << inode
->i_blkbits
;
2952 get_block(inode
, block
, &tmp
, 0);
2953 return tmp
.b_blocknr
;
2955 EXPORT_SYMBOL(generic_block_bmap
);
2957 static void end_bio_bh_io_sync(struct bio
*bio
, int err
)
2959 struct buffer_head
*bh
= bio
->bi_private
;
2961 if (unlikely (test_bit(BIO_QUIET
,&bio
->bi_flags
)))
2962 set_bit(BH_Quiet
, &bh
->b_state
);
2964 bh
->b_end_io(bh
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
2969 * This allows us to do IO even on the odd last sectors
2970 * of a device, even if the block size is some multiple
2971 * of the physical sector size.
2973 * We'll just truncate the bio to the size of the device,
2974 * and clear the end of the buffer head manually.
2976 * Truly out-of-range accesses will turn into actual IO
2977 * errors, this only handles the "we need to be able to
2978 * do IO at the final sector" case.
2980 void guard_bio_eod(int rw
, struct bio
*bio
)
2983 struct bio_vec
*bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
2984 unsigned truncated_bytes
;
2986 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
2991 * If the *whole* IO is past the end of the device,
2992 * let it through, and the IO layer will turn it into
2995 if (unlikely(bio
->bi_iter
.bi_sector
>= maxsector
))
2998 maxsector
-= bio
->bi_iter
.bi_sector
;
2999 if (likely((bio
->bi_iter
.bi_size
>> 9) <= maxsector
))
3002 /* Uhhuh. We've got a bio that straddles the device size! */
3003 truncated_bytes
= bio
->bi_iter
.bi_size
- (maxsector
<< 9);
3005 /* Truncate the bio.. */
3006 bio
->bi_iter
.bi_size
-= truncated_bytes
;
3007 bvec
->bv_len
-= truncated_bytes
;
3009 /* ..and clear the end of the buffer for reads */
3010 if ((rw
& RW_MASK
) == READ
) {
3011 zero_user(bvec
->bv_page
, bvec
->bv_offset
+ bvec
->bv_len
,
3016 int _submit_bh(int rw
, struct buffer_head
*bh
, unsigned long bio_flags
)
3020 BUG_ON(!buffer_locked(bh
));
3021 BUG_ON(!buffer_mapped(bh
));
3022 BUG_ON(!bh
->b_end_io
);
3023 BUG_ON(buffer_delay(bh
));
3024 BUG_ON(buffer_unwritten(bh
));
3027 * Only clear out a write error when rewriting
3029 if (test_set_buffer_req(bh
) && (rw
& WRITE
))
3030 clear_buffer_write_io_error(bh
);
3033 * from here on down, it's all bio -- do the initial mapping,
3034 * submit_bio -> generic_make_request may further map this bio around
3036 bio
= bio_alloc(GFP_NOIO
, 1);
3038 bio
->bi_iter
.bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
3039 bio
->bi_bdev
= bh
->b_bdev
;
3040 bio
->bi_io_vec
[0].bv_page
= bh
->b_page
;
3041 bio
->bi_io_vec
[0].bv_len
= bh
->b_size
;
3042 bio
->bi_io_vec
[0].bv_offset
= bh_offset(bh
);
3045 bio
->bi_iter
.bi_size
= bh
->b_size
;
3047 bio
->bi_end_io
= end_bio_bh_io_sync
;
3048 bio
->bi_private
= bh
;
3049 bio
->bi_flags
|= bio_flags
;
3051 /* Take care of bh's that straddle the end of the device */
3052 guard_bio_eod(rw
, bio
);
3054 if (buffer_meta(bh
))
3056 if (buffer_prio(bh
))
3059 submit_bio(rw
, bio
);
3062 EXPORT_SYMBOL_GPL(_submit_bh
);
3064 int submit_bh(int rw
, struct buffer_head
*bh
)
3066 return _submit_bh(rw
, bh
, 0);
3068 EXPORT_SYMBOL(submit_bh
);
3071 * ll_rw_block: low-level access to block devices (DEPRECATED)
3072 * @rw: whether to %READ or %WRITE or maybe %READA (readahead)
3073 * @nr: number of &struct buffer_heads in the array
3074 * @bhs: array of pointers to &struct buffer_head
3076 * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
3077 * requests an I/O operation on them, either a %READ or a %WRITE. The third
3078 * %READA option is described in the documentation for generic_make_request()
3079 * which ll_rw_block() calls.
3081 * This function drops any buffer that it cannot get a lock on (with the
3082 * BH_Lock state bit), any buffer that appears to be clean when doing a write
3083 * request, and any buffer that appears to be up-to-date when doing read
3084 * request. Further it marks as clean buffers that are processed for
3085 * writing (the buffer cache won't assume that they are actually clean
3086 * until the buffer gets unlocked).
3088 * ll_rw_block sets b_end_io to simple completion handler that marks
3089 * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
3092 * All of the buffers must be for the same device, and must also be a
3093 * multiple of the current approved size for the device.
3095 void ll_rw_block(int rw
, int nr
, struct buffer_head
*bhs
[])
3099 for (i
= 0; i
< nr
; i
++) {
3100 struct buffer_head
*bh
= bhs
[i
];
3102 if (!trylock_buffer(bh
))
3105 if (test_clear_buffer_dirty(bh
)) {
3106 bh
->b_end_io
= end_buffer_write_sync
;
3108 submit_bh(WRITE
, bh
);
3112 if (!buffer_uptodate(bh
)) {
3113 bh
->b_end_io
= end_buffer_read_sync
;
3122 EXPORT_SYMBOL(ll_rw_block
);
3124 void write_dirty_buffer(struct buffer_head
*bh
, int rw
)
3127 if (!test_clear_buffer_dirty(bh
)) {
3131 bh
->b_end_io
= end_buffer_write_sync
;
3135 EXPORT_SYMBOL(write_dirty_buffer
);
3138 * For a data-integrity writeout, we need to wait upon any in-progress I/O
3139 * and then start new I/O and then wait upon it. The caller must have a ref on
3142 int __sync_dirty_buffer(struct buffer_head
*bh
, int rw
)
3146 WARN_ON(atomic_read(&bh
->b_count
) < 1);
3148 if (test_clear_buffer_dirty(bh
)) {
3150 bh
->b_end_io
= end_buffer_write_sync
;
3151 ret
= submit_bh(rw
, bh
);
3153 if (!ret
&& !buffer_uptodate(bh
))
3160 EXPORT_SYMBOL(__sync_dirty_buffer
);
3162 int sync_dirty_buffer(struct buffer_head
*bh
)
3164 return __sync_dirty_buffer(bh
, WRITE_SYNC
);
3166 EXPORT_SYMBOL(sync_dirty_buffer
);
3169 * try_to_free_buffers() checks if all the buffers on this particular page
3170 * are unused, and releases them if so.
3172 * Exclusion against try_to_free_buffers may be obtained by either
3173 * locking the page or by holding its mapping's private_lock.
3175 * If the page is dirty but all the buffers are clean then we need to
3176 * be sure to mark the page clean as well. This is because the page
3177 * may be against a block device, and a later reattachment of buffers
3178 * to a dirty page will set *all* buffers dirty. Which would corrupt
3179 * filesystem data on the same device.
3181 * The same applies to regular filesystem pages: if all the buffers are
3182 * clean then we set the page clean and proceed. To do that, we require
3183 * total exclusion from __set_page_dirty_buffers(). That is obtained with
3186 * try_to_free_buffers() is non-blocking.
3188 static inline int buffer_busy(struct buffer_head
*bh
)
3190 return atomic_read(&bh
->b_count
) |
3191 (bh
->b_state
& ((1 << BH_Dirty
) | (1 << BH_Lock
)));
3195 drop_buffers(struct page
*page
, struct buffer_head
**buffers_to_free
)
3197 struct buffer_head
*head
= page_buffers(page
);
3198 struct buffer_head
*bh
;
3202 if (buffer_write_io_error(bh
) && page
->mapping
)
3203 set_bit(AS_EIO
, &page
->mapping
->flags
);
3204 if (buffer_busy(bh
))
3206 bh
= bh
->b_this_page
;
3207 } while (bh
!= head
);
3210 struct buffer_head
*next
= bh
->b_this_page
;
3212 if (bh
->b_assoc_map
)
3213 __remove_assoc_queue(bh
);
3215 } while (bh
!= head
);
3216 *buffers_to_free
= head
;
3217 __clear_page_buffers(page
);
3223 int try_to_free_buffers(struct page
*page
)
3225 struct address_space
* const mapping
= page
->mapping
;
3226 struct buffer_head
*buffers_to_free
= NULL
;
3229 BUG_ON(!PageLocked(page
));
3230 if (PageWriteback(page
))
3233 if (mapping
== NULL
) { /* can this still happen? */
3234 ret
= drop_buffers(page
, &buffers_to_free
);
3238 spin_lock(&mapping
->private_lock
);
3239 ret
= drop_buffers(page
, &buffers_to_free
);
3242 * If the filesystem writes its buffers by hand (eg ext3)
3243 * then we can have clean buffers against a dirty page. We
3244 * clean the page here; otherwise the VM will never notice
3245 * that the filesystem did any IO at all.
3247 * Also, during truncate, discard_buffer will have marked all
3248 * the page's buffers clean. We discover that here and clean
3251 * private_lock must be held over this entire operation in order
3252 * to synchronise against __set_page_dirty_buffers and prevent the
3253 * dirty bit from being lost.
3256 cancel_dirty_page(page
);
3257 spin_unlock(&mapping
->private_lock
);
3259 if (buffers_to_free
) {
3260 struct buffer_head
*bh
= buffers_to_free
;
3263 struct buffer_head
*next
= bh
->b_this_page
;
3264 free_buffer_head(bh
);
3266 } while (bh
!= buffers_to_free
);
3270 EXPORT_SYMBOL(try_to_free_buffers
);
3273 * There are no bdflush tunables left. But distributions are
3274 * still running obsolete flush daemons, so we terminate them here.
3276 * Use of bdflush() is deprecated and will be removed in a future kernel.
3277 * The `flush-X' kernel threads fully replace bdflush daemons and this call.
3279 SYSCALL_DEFINE2(bdflush
, int, func
, long, data
)
3281 static int msg_count
;
3283 if (!capable(CAP_SYS_ADMIN
))
3286 if (msg_count
< 5) {
3289 "warning: process `%s' used the obsolete bdflush"
3290 " system call\n", current
->comm
);
3291 printk(KERN_INFO
"Fix your initscripts?\n");
3300 * Buffer-head allocation
3302 static struct kmem_cache
*bh_cachep __read_mostly
;
3305 * Once the number of bh's in the machine exceeds this level, we start
3306 * stripping them in writeback.
3308 static unsigned long max_buffer_heads
;
3310 int buffer_heads_over_limit
;
3312 struct bh_accounting
{
3313 int nr
; /* Number of live bh's */
3314 int ratelimit
; /* Limit cacheline bouncing */
3317 static DEFINE_PER_CPU(struct bh_accounting
, bh_accounting
) = {0, 0};
3319 static void recalc_bh_state(void)
3324 if (__this_cpu_inc_return(bh_accounting
.ratelimit
) - 1 < 4096)
3326 __this_cpu_write(bh_accounting
.ratelimit
, 0);
3327 for_each_online_cpu(i
)
3328 tot
+= per_cpu(bh_accounting
, i
).nr
;
3329 buffer_heads_over_limit
= (tot
> max_buffer_heads
);
3332 struct buffer_head
*alloc_buffer_head(gfp_t gfp_flags
)
3334 struct buffer_head
*ret
= kmem_cache_zalloc(bh_cachep
, gfp_flags
);
3336 INIT_LIST_HEAD(&ret
->b_assoc_buffers
);
3338 __this_cpu_inc(bh_accounting
.nr
);
3344 EXPORT_SYMBOL(alloc_buffer_head
);
3346 void free_buffer_head(struct buffer_head
*bh
)
3348 BUG_ON(!list_empty(&bh
->b_assoc_buffers
));
3349 kmem_cache_free(bh_cachep
, bh
);
3351 __this_cpu_dec(bh_accounting
.nr
);
3355 EXPORT_SYMBOL(free_buffer_head
);
3357 static void buffer_exit_cpu(int cpu
)
3360 struct bh_lru
*b
= &per_cpu(bh_lrus
, cpu
);
3362 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
3366 this_cpu_add(bh_accounting
.nr
, per_cpu(bh_accounting
, cpu
).nr
);
3367 per_cpu(bh_accounting
, cpu
).nr
= 0;
3370 static int buffer_cpu_notify(struct notifier_block
*self
,
3371 unsigned long action
, void *hcpu
)
3373 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
)
3374 buffer_exit_cpu((unsigned long)hcpu
);
3379 * bh_uptodate_or_lock - Test whether the buffer is uptodate
3380 * @bh: struct buffer_head
3382 * Return true if the buffer is up-to-date and false,
3383 * with the buffer locked, if not.
3385 int bh_uptodate_or_lock(struct buffer_head
*bh
)
3387 if (!buffer_uptodate(bh
)) {
3389 if (!buffer_uptodate(bh
))
3395 EXPORT_SYMBOL(bh_uptodate_or_lock
);
3398 * bh_submit_read - Submit a locked buffer for reading
3399 * @bh: struct buffer_head
3401 * Returns zero on success and -EIO on error.
3403 int bh_submit_read(struct buffer_head
*bh
)
3405 BUG_ON(!buffer_locked(bh
));
3407 if (buffer_uptodate(bh
)) {
3413 bh
->b_end_io
= end_buffer_read_sync
;
3414 submit_bh(READ
, bh
);
3416 if (buffer_uptodate(bh
))
3420 EXPORT_SYMBOL(bh_submit_read
);
3422 void __init
buffer_init(void)
3424 unsigned long nrpages
;
3426 bh_cachep
= kmem_cache_create("buffer_head",
3427 sizeof(struct buffer_head
), 0,
3428 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
3433 * Limit the bh occupancy to 10% of ZONE_NORMAL
3435 nrpages
= (nr_free_buffer_pages() * 10) / 100;
3436 max_buffer_heads
= nrpages
* (PAGE_SIZE
/ sizeof(struct buffer_head
));
3437 hotcpu_notifier(buffer_cpu_notify
, 0);