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/backing-dev.h>
34 #include <linux/writeback.h>
35 #include <linux/hash.h>
36 #include <linux/suspend.h>
37 #include <linux/buffer_head.h>
38 #include <linux/task_io_accounting_ops.h>
39 #include <linux/bio.h>
40 #include <linux/notifier.h>
41 #include <linux/cpu.h>
42 #include <linux/bitops.h>
43 #include <linux/mpage.h>
44 #include <linux/bit_spinlock.h>
45 #include <trace/events/block.h>
47 static int fsync_buffers_list(spinlock_t
*lock
, struct list_head
*list
);
48 static int submit_bh_wbc(int op
, int op_flags
, struct buffer_head
*bh
,
49 unsigned long bio_flags
,
50 struct writeback_control
*wbc
);
52 #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
54 void init_buffer(struct buffer_head
*bh
, bh_end_io_t
*handler
, void *private)
56 bh
->b_end_io
= handler
;
57 bh
->b_private
= private;
59 EXPORT_SYMBOL(init_buffer
);
61 inline void touch_buffer(struct buffer_head
*bh
)
63 trace_block_touch_buffer(bh
);
64 mark_page_accessed(bh
->b_page
);
66 EXPORT_SYMBOL(touch_buffer
);
68 void __lock_buffer(struct buffer_head
*bh
)
70 wait_on_bit_lock_io(&bh
->b_state
, BH_Lock
, TASK_UNINTERRUPTIBLE
);
72 EXPORT_SYMBOL(__lock_buffer
);
74 void unlock_buffer(struct buffer_head
*bh
)
76 clear_bit_unlock(BH_Lock
, &bh
->b_state
);
77 smp_mb__after_atomic();
78 wake_up_bit(&bh
->b_state
, BH_Lock
);
80 EXPORT_SYMBOL(unlock_buffer
);
83 * Returns if the page has dirty or writeback buffers. If all the buffers
84 * are unlocked and clean then the PageDirty information is stale. If
85 * any of the pages are locked, it is assumed they are locked for IO.
87 void buffer_check_dirty_writeback(struct page
*page
,
88 bool *dirty
, bool *writeback
)
90 struct buffer_head
*head
, *bh
;
94 BUG_ON(!PageLocked(page
));
96 if (!page_has_buffers(page
))
99 if (PageWriteback(page
))
102 head
= page_buffers(page
);
105 if (buffer_locked(bh
))
108 if (buffer_dirty(bh
))
111 bh
= bh
->b_this_page
;
112 } while (bh
!= head
);
114 EXPORT_SYMBOL(buffer_check_dirty_writeback
);
117 * Block until a buffer comes unlocked. This doesn't stop it
118 * from becoming locked again - you have to lock it yourself
119 * if you want to preserve its state.
121 void __wait_on_buffer(struct buffer_head
* bh
)
123 wait_on_bit_io(&bh
->b_state
, BH_Lock
, TASK_UNINTERRUPTIBLE
);
125 EXPORT_SYMBOL(__wait_on_buffer
);
128 __clear_page_buffers(struct page
*page
)
130 ClearPagePrivate(page
);
131 set_page_private(page
, 0);
135 static void buffer_io_error(struct buffer_head
*bh
, char *msg
)
137 if (!test_bit(BH_Quiet
, &bh
->b_state
))
138 printk_ratelimited(KERN_ERR
139 "Buffer I/O error on dev %pg, logical block %llu%s\n",
140 bh
->b_bdev
, (unsigned long long)bh
->b_blocknr
, msg
);
144 * End-of-IO handler helper function which does not touch the bh after
146 * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
147 * a race there is benign: unlock_buffer() only use the bh's address for
148 * hashing after unlocking the buffer, so it doesn't actually touch the bh
151 static void __end_buffer_read_notouch(struct buffer_head
*bh
, int uptodate
)
154 set_buffer_uptodate(bh
);
156 /* This happens, due to failed READA attempts. */
157 clear_buffer_uptodate(bh
);
163 * Default synchronous end-of-IO handler.. Just mark it up-to-date and
164 * unlock the buffer. This is what ll_rw_block uses too.
166 void end_buffer_read_sync(struct buffer_head
*bh
, int uptodate
)
168 __end_buffer_read_notouch(bh
, uptodate
);
171 EXPORT_SYMBOL(end_buffer_read_sync
);
173 void end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
176 set_buffer_uptodate(bh
);
178 buffer_io_error(bh
, ", lost sync page write");
179 set_buffer_write_io_error(bh
);
180 clear_buffer_uptodate(bh
);
185 EXPORT_SYMBOL(end_buffer_write_sync
);
188 * Various filesystems appear to want __find_get_block to be non-blocking.
189 * But it's the page lock which protects the buffers. To get around this,
190 * we get exclusion from try_to_free_buffers with the blockdev mapping's
193 * Hack idea: for the blockdev mapping, i_bufferlist_lock contention
194 * may be quite high. This code could TryLock the page, and if that
195 * succeeds, there is no need to take private_lock. (But if
196 * private_lock is contended then so is mapping->tree_lock).
198 static struct buffer_head
*
199 __find_get_block_slow(struct block_device
*bdev
, sector_t block
)
201 struct inode
*bd_inode
= bdev
->bd_inode
;
202 struct address_space
*bd_mapping
= bd_inode
->i_mapping
;
203 struct buffer_head
*ret
= NULL
;
205 struct buffer_head
*bh
;
206 struct buffer_head
*head
;
210 index
= block
>> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
211 page
= find_get_page_flags(bd_mapping
, index
, FGP_ACCESSED
);
215 spin_lock(&bd_mapping
->private_lock
);
216 if (!page_has_buffers(page
))
218 head
= page_buffers(page
);
221 if (!buffer_mapped(bh
))
223 else if (bh
->b_blocknr
== block
) {
228 bh
= bh
->b_this_page
;
229 } while (bh
!= head
);
231 /* we might be here because some of the buffers on this page are
232 * not mapped. This is due to various races between
233 * file io on the block device and getblk. It gets dealt with
234 * elsewhere, don't buffer_error if we had some unmapped buffers
237 printk("__find_get_block_slow() failed. "
238 "block=%llu, b_blocknr=%llu\n",
239 (unsigned long long)block
,
240 (unsigned long long)bh
->b_blocknr
);
241 printk("b_state=0x%08lx, b_size=%zu\n",
242 bh
->b_state
, bh
->b_size
);
243 printk("device %pg blocksize: %d\n", bdev
,
244 1 << bd_inode
->i_blkbits
);
247 spin_unlock(&bd_mapping
->private_lock
);
254 * Kick the writeback threads then try to free up some ZONE_NORMAL memory.
256 static void free_more_memory(void)
261 wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM
);
264 for_each_online_node(nid
) {
266 z
= first_zones_zonelist(node_zonelist(nid
, GFP_NOFS
),
267 gfp_zone(GFP_NOFS
), NULL
);
269 try_to_free_pages(node_zonelist(nid
, GFP_NOFS
), 0,
275 * I/O completion handler for block_read_full_page() - pages
276 * which come unlocked at the end of I/O.
278 static void end_buffer_async_read(struct buffer_head
*bh
, int uptodate
)
281 struct buffer_head
*first
;
282 struct buffer_head
*tmp
;
284 int page_uptodate
= 1;
286 BUG_ON(!buffer_async_read(bh
));
290 set_buffer_uptodate(bh
);
292 clear_buffer_uptodate(bh
);
293 buffer_io_error(bh
, ", async page read");
298 * Be _very_ careful from here on. Bad things can happen if
299 * two buffer heads end IO at almost the same time and both
300 * decide that the page is now completely done.
302 first
= page_buffers(page
);
303 local_irq_save(flags
);
304 bit_spin_lock(BH_Uptodate_Lock
, &first
->b_state
);
305 clear_buffer_async_read(bh
);
309 if (!buffer_uptodate(tmp
))
311 if (buffer_async_read(tmp
)) {
312 BUG_ON(!buffer_locked(tmp
));
315 tmp
= tmp
->b_this_page
;
317 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
318 local_irq_restore(flags
);
321 * If none of the buffers had errors and they are all
322 * uptodate then we can set the page uptodate.
324 if (page_uptodate
&& !PageError(page
))
325 SetPageUptodate(page
);
330 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
331 local_irq_restore(flags
);
336 * Completion handler for block_write_full_page() - pages which are unlocked
337 * during I/O, and which have PageWriteback cleared upon I/O completion.
339 void end_buffer_async_write(struct buffer_head
*bh
, int uptodate
)
342 struct buffer_head
*first
;
343 struct buffer_head
*tmp
;
346 BUG_ON(!buffer_async_write(bh
));
350 set_buffer_uptodate(bh
);
352 buffer_io_error(bh
, ", lost async page write");
353 set_bit(AS_EIO
, &page
->mapping
->flags
);
354 set_buffer_write_io_error(bh
);
355 clear_buffer_uptodate(bh
);
359 first
= page_buffers(page
);
360 local_irq_save(flags
);
361 bit_spin_lock(BH_Uptodate_Lock
, &first
->b_state
);
363 clear_buffer_async_write(bh
);
365 tmp
= bh
->b_this_page
;
367 if (buffer_async_write(tmp
)) {
368 BUG_ON(!buffer_locked(tmp
));
371 tmp
= tmp
->b_this_page
;
373 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
374 local_irq_restore(flags
);
375 end_page_writeback(page
);
379 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
380 local_irq_restore(flags
);
383 EXPORT_SYMBOL(end_buffer_async_write
);
386 * If a page's buffers are under async readin (end_buffer_async_read
387 * completion) then there is a possibility that another thread of
388 * control could lock one of the buffers after it has completed
389 * but while some of the other buffers have not completed. This
390 * locked buffer would confuse end_buffer_async_read() into not unlocking
391 * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
392 * that this buffer is not under async I/O.
394 * The page comes unlocked when it has no locked buffer_async buffers
397 * PageLocked prevents anyone starting new async I/O reads any of
400 * PageWriteback is used to prevent simultaneous writeout of the same
403 * PageLocked prevents anyone from starting writeback of a page which is
404 * under read I/O (PageWriteback is only ever set against a locked page).
406 static void mark_buffer_async_read(struct buffer_head
*bh
)
408 bh
->b_end_io
= end_buffer_async_read
;
409 set_buffer_async_read(bh
);
412 static void mark_buffer_async_write_endio(struct buffer_head
*bh
,
413 bh_end_io_t
*handler
)
415 bh
->b_end_io
= handler
;
416 set_buffer_async_write(bh
);
419 void mark_buffer_async_write(struct buffer_head
*bh
)
421 mark_buffer_async_write_endio(bh
, end_buffer_async_write
);
423 EXPORT_SYMBOL(mark_buffer_async_write
);
427 * fs/buffer.c contains helper functions for buffer-backed address space's
428 * fsync functions. A common requirement for buffer-based filesystems is
429 * that certain data from the backing blockdev needs to be written out for
430 * a successful fsync(). For example, ext2 indirect blocks need to be
431 * written back and waited upon before fsync() returns.
433 * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
434 * inode_has_buffers() and invalidate_inode_buffers() are provided for the
435 * management of a list of dependent buffers at ->i_mapping->private_list.
437 * Locking is a little subtle: try_to_free_buffers() will remove buffers
438 * from their controlling inode's queue when they are being freed. But
439 * try_to_free_buffers() will be operating against the *blockdev* mapping
440 * at the time, not against the S_ISREG file which depends on those buffers.
441 * So the locking for private_list is via the private_lock in the address_space
442 * which backs the buffers. Which is different from the address_space
443 * against which the buffers are listed. So for a particular address_space,
444 * mapping->private_lock does *not* protect mapping->private_list! In fact,
445 * mapping->private_list will always be protected by the backing blockdev's
448 * Which introduces a requirement: all buffers on an address_space's
449 * ->private_list must be from the same address_space: the blockdev's.
451 * address_spaces which do not place buffers at ->private_list via these
452 * utility functions are free to use private_lock and private_list for
453 * whatever they want. The only requirement is that list_empty(private_list)
454 * be true at clear_inode() time.
456 * FIXME: clear_inode should not call invalidate_inode_buffers(). The
457 * filesystems should do that. invalidate_inode_buffers() should just go
458 * BUG_ON(!list_empty).
460 * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
461 * take an address_space, not an inode. And it should be called
462 * mark_buffer_dirty_fsync() to clearly define why those buffers are being
465 * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
466 * list if it is already on a list. Because if the buffer is on a list,
467 * it *must* already be on the right one. If not, the filesystem is being
468 * silly. This will save a ton of locking. But first we have to ensure
469 * that buffers are taken *off* the old inode's list when they are freed
470 * (presumably in truncate). That requires careful auditing of all
471 * filesystems (do it inside bforget()). It could also be done by bringing
476 * The buffer's backing address_space's private_lock must be held
478 static void __remove_assoc_queue(struct buffer_head
*bh
)
480 list_del_init(&bh
->b_assoc_buffers
);
481 WARN_ON(!bh
->b_assoc_map
);
482 if (buffer_write_io_error(bh
))
483 set_bit(AS_EIO
, &bh
->b_assoc_map
->flags
);
484 bh
->b_assoc_map
= NULL
;
487 int inode_has_buffers(struct inode
*inode
)
489 return !list_empty(&inode
->i_data
.private_list
);
493 * osync is designed to support O_SYNC io. It waits synchronously for
494 * all already-submitted IO to complete, but does not queue any new
495 * writes to the disk.
497 * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
498 * you dirty the buffers, and then use osync_inode_buffers to wait for
499 * completion. Any other dirty buffers which are not yet queued for
500 * write will not be flushed to disk by the osync.
502 static int osync_buffers_list(spinlock_t
*lock
, struct list_head
*list
)
504 struct buffer_head
*bh
;
510 list_for_each_prev(p
, list
) {
512 if (buffer_locked(bh
)) {
516 if (!buffer_uptodate(bh
))
527 static void do_thaw_one(struct super_block
*sb
, void *unused
)
529 while (sb
->s_bdev
&& !thaw_bdev(sb
->s_bdev
, sb
))
530 printk(KERN_WARNING
"Emergency Thaw on %pg\n", sb
->s_bdev
);
533 static void do_thaw_all(struct work_struct
*work
)
535 iterate_supers(do_thaw_one
, NULL
);
537 printk(KERN_WARNING
"Emergency Thaw complete\n");
541 * emergency_thaw_all -- forcibly thaw every frozen filesystem
543 * Used for emergency unfreeze of all filesystems via SysRq
545 void emergency_thaw_all(void)
547 struct work_struct
*work
;
549 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
551 INIT_WORK(work
, do_thaw_all
);
557 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
558 * @mapping: the mapping which wants those buffers written
560 * Starts I/O against the buffers at mapping->private_list, and waits upon
563 * Basically, this is a convenience function for fsync().
564 * @mapping is a file or directory which needs those buffers to be written for
565 * a successful fsync().
567 int sync_mapping_buffers(struct address_space
*mapping
)
569 struct address_space
*buffer_mapping
= mapping
->private_data
;
571 if (buffer_mapping
== NULL
|| list_empty(&mapping
->private_list
))
574 return fsync_buffers_list(&buffer_mapping
->private_lock
,
575 &mapping
->private_list
);
577 EXPORT_SYMBOL(sync_mapping_buffers
);
580 * Called when we've recently written block `bblock', and it is known that
581 * `bblock' was for a buffer_boundary() buffer. This means that the block at
582 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
583 * dirty, schedule it for IO. So that indirects merge nicely with their data.
585 void write_boundary_block(struct block_device
*bdev
,
586 sector_t bblock
, unsigned blocksize
)
588 struct buffer_head
*bh
= __find_get_block(bdev
, bblock
+ 1, blocksize
);
590 if (buffer_dirty(bh
))
591 ll_rw_block(REQ_OP_WRITE
, 0, 1, &bh
);
596 void mark_buffer_dirty_inode(struct buffer_head
*bh
, struct inode
*inode
)
598 struct address_space
*mapping
= inode
->i_mapping
;
599 struct address_space
*buffer_mapping
= bh
->b_page
->mapping
;
601 mark_buffer_dirty(bh
);
602 if (!mapping
->private_data
) {
603 mapping
->private_data
= buffer_mapping
;
605 BUG_ON(mapping
->private_data
!= buffer_mapping
);
607 if (!bh
->b_assoc_map
) {
608 spin_lock(&buffer_mapping
->private_lock
);
609 list_move_tail(&bh
->b_assoc_buffers
,
610 &mapping
->private_list
);
611 bh
->b_assoc_map
= mapping
;
612 spin_unlock(&buffer_mapping
->private_lock
);
615 EXPORT_SYMBOL(mark_buffer_dirty_inode
);
618 * Mark the page dirty, and set it dirty in the radix tree, and mark the inode
621 * If warn is true, then emit a warning if the page is not uptodate and has
622 * not been truncated.
624 * The caller must hold lock_page_memcg().
626 static void __set_page_dirty(struct page
*page
, struct address_space
*mapping
,
631 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
632 if (page
->mapping
) { /* Race with truncate? */
633 WARN_ON_ONCE(warn
&& !PageUptodate(page
));
634 account_page_dirtied(page
, mapping
);
635 radix_tree_tag_set(&mapping
->page_tree
,
636 page_index(page
), PAGECACHE_TAG_DIRTY
);
638 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
642 * Add a page to the dirty page list.
644 * It is a sad fact of life that this function is called from several places
645 * deeply under spinlocking. It may not sleep.
647 * If the page has buffers, the uptodate buffers are set dirty, to preserve
648 * dirty-state coherency between the page and the buffers. It the page does
649 * not have buffers then when they are later attached they will all be set
652 * The buffers are dirtied before the page is dirtied. There's a small race
653 * window in which a writepage caller may see the page cleanness but not the
654 * buffer dirtiness. That's fine. If this code were to set the page dirty
655 * before the buffers, a concurrent writepage caller could clear the page dirty
656 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
657 * page on the dirty page list.
659 * We use private_lock to lock against try_to_free_buffers while using the
660 * page's buffer list. Also use this to protect against clean buffers being
661 * added to the page after it was set dirty.
663 * FIXME: may need to call ->reservepage here as well. That's rather up to the
664 * address_space though.
666 int __set_page_dirty_buffers(struct page
*page
)
669 struct address_space
*mapping
= page_mapping(page
);
671 if (unlikely(!mapping
))
672 return !TestSetPageDirty(page
);
674 spin_lock(&mapping
->private_lock
);
675 if (page_has_buffers(page
)) {
676 struct buffer_head
*head
= page_buffers(page
);
677 struct buffer_head
*bh
= head
;
680 set_buffer_dirty(bh
);
681 bh
= bh
->b_this_page
;
682 } while (bh
!= head
);
685 * Lock out page->mem_cgroup migration to keep PageDirty
686 * synchronized with per-memcg dirty page counters.
688 lock_page_memcg(page
);
689 newly_dirty
= !TestSetPageDirty(page
);
690 spin_unlock(&mapping
->private_lock
);
693 __set_page_dirty(page
, mapping
, 1);
695 unlock_page_memcg(page
);
698 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
702 EXPORT_SYMBOL(__set_page_dirty_buffers
);
705 * Write out and wait upon a list of buffers.
707 * We have conflicting pressures: we want to make sure that all
708 * initially dirty buffers get waited on, but that any subsequently
709 * dirtied buffers don't. After all, we don't want fsync to last
710 * forever if somebody is actively writing to the file.
712 * Do this in two main stages: first we copy dirty buffers to a
713 * temporary inode list, queueing the writes as we go. Then we clean
714 * up, waiting for those writes to complete.
716 * During this second stage, any subsequent updates to the file may end
717 * up refiling the buffer on the original inode's dirty list again, so
718 * there is a chance we will end up with a buffer queued for write but
719 * not yet completed on that list. So, as a final cleanup we go through
720 * the osync code to catch these locked, dirty buffers without requeuing
721 * any newly dirty buffers for write.
723 static int fsync_buffers_list(spinlock_t
*lock
, struct list_head
*list
)
725 struct buffer_head
*bh
;
726 struct list_head tmp
;
727 struct address_space
*mapping
;
729 struct blk_plug plug
;
731 INIT_LIST_HEAD(&tmp
);
732 blk_start_plug(&plug
);
735 while (!list_empty(list
)) {
736 bh
= BH_ENTRY(list
->next
);
737 mapping
= bh
->b_assoc_map
;
738 __remove_assoc_queue(bh
);
739 /* Avoid race with mark_buffer_dirty_inode() which does
740 * a lockless check and we rely on seeing the dirty bit */
742 if (buffer_dirty(bh
) || buffer_locked(bh
)) {
743 list_add(&bh
->b_assoc_buffers
, &tmp
);
744 bh
->b_assoc_map
= mapping
;
745 if (buffer_dirty(bh
)) {
749 * Ensure any pending I/O completes so that
750 * write_dirty_buffer() actually writes the
751 * current contents - it is a noop if I/O is
752 * still in flight on potentially older
755 write_dirty_buffer(bh
, WRITE_SYNC
);
758 * Kick off IO for the previous mapping. Note
759 * that we will not run the very last mapping,
760 * wait_on_buffer() will do that for us
761 * through sync_buffer().
770 blk_finish_plug(&plug
);
773 while (!list_empty(&tmp
)) {
774 bh
= BH_ENTRY(tmp
.prev
);
776 mapping
= bh
->b_assoc_map
;
777 __remove_assoc_queue(bh
);
778 /* Avoid race with mark_buffer_dirty_inode() which does
779 * a lockless check and we rely on seeing the dirty bit */
781 if (buffer_dirty(bh
)) {
782 list_add(&bh
->b_assoc_buffers
,
783 &mapping
->private_list
);
784 bh
->b_assoc_map
= mapping
;
788 if (!buffer_uptodate(bh
))
795 err2
= osync_buffers_list(lock
, list
);
803 * Invalidate any and all dirty buffers on a given inode. We are
804 * probably unmounting the fs, but that doesn't mean we have already
805 * done a sync(). Just drop the buffers from the inode list.
807 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
808 * assumes that all the buffers are against the blockdev. Not true
811 void invalidate_inode_buffers(struct inode
*inode
)
813 if (inode_has_buffers(inode
)) {
814 struct address_space
*mapping
= &inode
->i_data
;
815 struct list_head
*list
= &mapping
->private_list
;
816 struct address_space
*buffer_mapping
= mapping
->private_data
;
818 spin_lock(&buffer_mapping
->private_lock
);
819 while (!list_empty(list
))
820 __remove_assoc_queue(BH_ENTRY(list
->next
));
821 spin_unlock(&buffer_mapping
->private_lock
);
824 EXPORT_SYMBOL(invalidate_inode_buffers
);
827 * Remove any clean buffers from the inode's buffer list. This is called
828 * when we're trying to free the inode itself. Those buffers can pin it.
830 * Returns true if all buffers were removed.
832 int remove_inode_buffers(struct inode
*inode
)
836 if (inode_has_buffers(inode
)) {
837 struct address_space
*mapping
= &inode
->i_data
;
838 struct list_head
*list
= &mapping
->private_list
;
839 struct address_space
*buffer_mapping
= mapping
->private_data
;
841 spin_lock(&buffer_mapping
->private_lock
);
842 while (!list_empty(list
)) {
843 struct buffer_head
*bh
= BH_ENTRY(list
->next
);
844 if (buffer_dirty(bh
)) {
848 __remove_assoc_queue(bh
);
850 spin_unlock(&buffer_mapping
->private_lock
);
856 * Create the appropriate buffers when given a page for data area and
857 * the size of each buffer.. Use the bh->b_this_page linked list to
858 * follow the buffers created. Return NULL if unable to create more
861 * The retry flag is used to differentiate async IO (paging, swapping)
862 * which may not fail from ordinary buffer allocations.
864 struct buffer_head
*alloc_page_buffers(struct page
*page
, unsigned long size
,
867 struct buffer_head
*bh
, *head
;
873 while ((offset
-= size
) >= 0) {
874 bh
= alloc_buffer_head(GFP_NOFS
);
878 bh
->b_this_page
= head
;
884 /* Link the buffer to its page */
885 set_bh_page(bh
, page
, offset
);
889 * In case anything failed, we just free everything we got.
895 head
= head
->b_this_page
;
896 free_buffer_head(bh
);
901 * Return failure for non-async IO requests. Async IO requests
902 * are not allowed to fail, so we have to wait until buffer heads
903 * become available. But we don't want tasks sleeping with
904 * partially complete buffers, so all were released above.
909 /* We're _really_ low on memory. Now we just
910 * wait for old buffer heads to become free due to
911 * finishing IO. Since this is an async request and
912 * the reserve list is empty, we're sure there are
913 * async buffer heads in use.
918 EXPORT_SYMBOL_GPL(alloc_page_buffers
);
921 link_dev_buffers(struct page
*page
, struct buffer_head
*head
)
923 struct buffer_head
*bh
, *tail
;
928 bh
= bh
->b_this_page
;
930 tail
->b_this_page
= head
;
931 attach_page_buffers(page
, head
);
934 static sector_t
blkdev_max_block(struct block_device
*bdev
, unsigned int size
)
936 sector_t retval
= ~((sector_t
)0);
937 loff_t sz
= i_size_read(bdev
->bd_inode
);
940 unsigned int sizebits
= blksize_bits(size
);
941 retval
= (sz
>> sizebits
);
947 * Initialise the state of a blockdev page's buffers.
950 init_page_buffers(struct page
*page
, struct block_device
*bdev
,
951 sector_t block
, int size
)
953 struct buffer_head
*head
= page_buffers(page
);
954 struct buffer_head
*bh
= head
;
955 int uptodate
= PageUptodate(page
);
956 sector_t end_block
= blkdev_max_block(I_BDEV(bdev
->bd_inode
), size
);
959 if (!buffer_mapped(bh
)) {
960 init_buffer(bh
, NULL
, NULL
);
962 bh
->b_blocknr
= block
;
964 set_buffer_uptodate(bh
);
965 if (block
< end_block
)
966 set_buffer_mapped(bh
);
969 bh
= bh
->b_this_page
;
970 } while (bh
!= head
);
973 * Caller needs to validate requested block against end of device.
979 * Create the page-cache page that contains the requested block.
981 * This is used purely for blockdev mappings.
984 grow_dev_page(struct block_device
*bdev
, sector_t block
,
985 pgoff_t index
, int size
, int sizebits
, gfp_t gfp
)
987 struct inode
*inode
= bdev
->bd_inode
;
989 struct buffer_head
*bh
;
991 int ret
= 0; /* Will call free_more_memory() */
994 gfp_mask
= mapping_gfp_constraint(inode
->i_mapping
, ~__GFP_FS
) | gfp
;
997 * XXX: __getblk_slow() can not really deal with failure and
998 * will endlessly loop on improvised global reclaim. Prefer
999 * looping in the allocator rather than here, at least that
1000 * code knows what it's doing.
1002 gfp_mask
|= __GFP_NOFAIL
;
1004 page
= find_or_create_page(inode
->i_mapping
, index
, gfp_mask
);
1008 BUG_ON(!PageLocked(page
));
1010 if (page_has_buffers(page
)) {
1011 bh
= page_buffers(page
);
1012 if (bh
->b_size
== size
) {
1013 end_block
= init_page_buffers(page
, bdev
,
1014 (sector_t
)index
<< sizebits
,
1018 if (!try_to_free_buffers(page
))
1023 * Allocate some buffers for this page
1025 bh
= alloc_page_buffers(page
, size
, 0);
1030 * Link the page to the buffers and initialise them. Take the
1031 * lock to be atomic wrt __find_get_block(), which does not
1032 * run under the page lock.
1034 spin_lock(&inode
->i_mapping
->private_lock
);
1035 link_dev_buffers(page
, bh
);
1036 end_block
= init_page_buffers(page
, bdev
, (sector_t
)index
<< sizebits
,
1038 spin_unlock(&inode
->i_mapping
->private_lock
);
1040 ret
= (block
< end_block
) ? 1 : -ENXIO
;
1048 * Create buffers for the specified block device block's page. If
1049 * that page was dirty, the buffers are set dirty also.
1052 grow_buffers(struct block_device
*bdev
, sector_t block
, int size
, gfp_t gfp
)
1060 } while ((size
<< sizebits
) < PAGE_SIZE
);
1062 index
= block
>> sizebits
;
1065 * Check for a block which wants to lie outside our maximum possible
1066 * pagecache index. (this comparison is done using sector_t types).
1068 if (unlikely(index
!= block
>> sizebits
)) {
1069 printk(KERN_ERR
"%s: requested out-of-range block %llu for "
1071 __func__
, (unsigned long long)block
,
1076 /* Create a page with the proper size buffers.. */
1077 return grow_dev_page(bdev
, block
, index
, size
, sizebits
, gfp
);
1080 struct buffer_head
*
1081 __getblk_slow(struct block_device
*bdev
, sector_t block
,
1082 unsigned size
, gfp_t gfp
)
1084 /* Size must be multiple of hard sectorsize */
1085 if (unlikely(size
& (bdev_logical_block_size(bdev
)-1) ||
1086 (size
< 512 || size
> PAGE_SIZE
))) {
1087 printk(KERN_ERR
"getblk(): invalid block size %d requested\n",
1089 printk(KERN_ERR
"logical block size: %d\n",
1090 bdev_logical_block_size(bdev
));
1097 struct buffer_head
*bh
;
1100 bh
= __find_get_block(bdev
, block
, size
);
1104 ret
= grow_buffers(bdev
, block
, size
, gfp
);
1111 EXPORT_SYMBOL(__getblk_slow
);
1114 * The relationship between dirty buffers and dirty pages:
1116 * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1117 * the page is tagged dirty in its radix tree.
1119 * At all times, the dirtiness of the buffers represents the dirtiness of
1120 * subsections of the page. If the page has buffers, the page dirty bit is
1121 * merely a hint about the true dirty state.
1123 * When a page is set dirty in its entirety, all its buffers are marked dirty
1124 * (if the page has buffers).
1126 * When a buffer is marked dirty, its page is dirtied, but the page's other
1129 * Also. When blockdev buffers are explicitly read with bread(), they
1130 * individually become uptodate. But their backing page remains not
1131 * uptodate - even if all of its buffers are uptodate. A subsequent
1132 * block_read_full_page() against that page will discover all the uptodate
1133 * buffers, will set the page uptodate and will perform no I/O.
1137 * mark_buffer_dirty - mark a buffer_head as needing writeout
1138 * @bh: the buffer_head to mark dirty
1140 * mark_buffer_dirty() will set the dirty bit against the buffer, then set its
1141 * backing page dirty, then tag the page as dirty in its address_space's radix
1142 * tree and then attach the address_space's inode to its superblock's dirty
1145 * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
1146 * mapping->tree_lock and mapping->host->i_lock.
1148 void mark_buffer_dirty(struct buffer_head
*bh
)
1150 WARN_ON_ONCE(!buffer_uptodate(bh
));
1152 trace_block_dirty_buffer(bh
);
1155 * Very *carefully* optimize the it-is-already-dirty case.
1157 * Don't let the final "is it dirty" escape to before we
1158 * perhaps modified the buffer.
1160 if (buffer_dirty(bh
)) {
1162 if (buffer_dirty(bh
))
1166 if (!test_set_buffer_dirty(bh
)) {
1167 struct page
*page
= bh
->b_page
;
1168 struct address_space
*mapping
= NULL
;
1170 lock_page_memcg(page
);
1171 if (!TestSetPageDirty(page
)) {
1172 mapping
= page_mapping(page
);
1174 __set_page_dirty(page
, mapping
, 0);
1176 unlock_page_memcg(page
);
1178 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
1181 EXPORT_SYMBOL(mark_buffer_dirty
);
1184 * Decrement a buffer_head's reference count. If all buffers against a page
1185 * have zero reference count, are clean and unlocked, and if the page is clean
1186 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1187 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1188 * a page but it ends up not being freed, and buffers may later be reattached).
1190 void __brelse(struct buffer_head
* buf
)
1192 if (atomic_read(&buf
->b_count
)) {
1196 WARN(1, KERN_ERR
"VFS: brelse: Trying to free free buffer\n");
1198 EXPORT_SYMBOL(__brelse
);
1201 * bforget() is like brelse(), except it discards any
1202 * potentially dirty data.
1204 void __bforget(struct buffer_head
*bh
)
1206 clear_buffer_dirty(bh
);
1207 if (bh
->b_assoc_map
) {
1208 struct address_space
*buffer_mapping
= bh
->b_page
->mapping
;
1210 spin_lock(&buffer_mapping
->private_lock
);
1211 list_del_init(&bh
->b_assoc_buffers
);
1212 bh
->b_assoc_map
= NULL
;
1213 spin_unlock(&buffer_mapping
->private_lock
);
1217 EXPORT_SYMBOL(__bforget
);
1219 static struct buffer_head
*__bread_slow(struct buffer_head
*bh
)
1222 if (buffer_uptodate(bh
)) {
1227 bh
->b_end_io
= end_buffer_read_sync
;
1228 submit_bh(REQ_OP_READ
, 0, bh
);
1230 if (buffer_uptodate(bh
))
1238 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1239 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1240 * refcount elevated by one when they're in an LRU. A buffer can only appear
1241 * once in a particular CPU's LRU. A single buffer can be present in multiple
1242 * CPU's LRUs at the same time.
1244 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1245 * sb_find_get_block().
1247 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1248 * a local interrupt disable for that.
1251 #define BH_LRU_SIZE 16
1254 struct buffer_head
*bhs
[BH_LRU_SIZE
];
1257 static DEFINE_PER_CPU(struct bh_lru
, bh_lrus
) = {{ NULL
}};
1260 #define bh_lru_lock() local_irq_disable()
1261 #define bh_lru_unlock() local_irq_enable()
1263 #define bh_lru_lock() preempt_disable()
1264 #define bh_lru_unlock() preempt_enable()
1267 static inline void check_irqs_on(void)
1269 #ifdef irqs_disabled
1270 BUG_ON(irqs_disabled());
1275 * The LRU management algorithm is dopey-but-simple. Sorry.
1277 static void bh_lru_install(struct buffer_head
*bh
)
1279 struct buffer_head
*evictee
= NULL
;
1283 if (__this_cpu_read(bh_lrus
.bhs
[0]) != bh
) {
1284 struct buffer_head
*bhs
[BH_LRU_SIZE
];
1290 for (in
= 0; in
< BH_LRU_SIZE
; in
++) {
1291 struct buffer_head
*bh2
=
1292 __this_cpu_read(bh_lrus
.bhs
[in
]);
1297 if (out
>= BH_LRU_SIZE
) {
1298 BUG_ON(evictee
!= NULL
);
1305 while (out
< BH_LRU_SIZE
)
1307 memcpy(this_cpu_ptr(&bh_lrus
.bhs
), bhs
, sizeof(bhs
));
1316 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1318 static struct buffer_head
*
1319 lookup_bh_lru(struct block_device
*bdev
, sector_t block
, unsigned size
)
1321 struct buffer_head
*ret
= NULL
;
1326 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1327 struct buffer_head
*bh
= __this_cpu_read(bh_lrus
.bhs
[i
]);
1329 if (bh
&& bh
->b_blocknr
== block
&& bh
->b_bdev
== bdev
&&
1330 bh
->b_size
== size
) {
1333 __this_cpu_write(bh_lrus
.bhs
[i
],
1334 __this_cpu_read(bh_lrus
.bhs
[i
- 1]));
1337 __this_cpu_write(bh_lrus
.bhs
[0], bh
);
1349 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1350 * it in the LRU and mark it as accessed. If it is not present then return
1353 struct buffer_head
*
1354 __find_get_block(struct block_device
*bdev
, sector_t block
, unsigned size
)
1356 struct buffer_head
*bh
= lookup_bh_lru(bdev
, block
, size
);
1359 /* __find_get_block_slow will mark the page accessed */
1360 bh
= __find_get_block_slow(bdev
, block
);
1368 EXPORT_SYMBOL(__find_get_block
);
1371 * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
1372 * which corresponds to the passed block_device, block and size. The
1373 * returned buffer has its reference count incremented.
1375 * __getblk_gfp() will lock up the machine if grow_dev_page's
1376 * try_to_free_buffers() attempt is failing. FIXME, perhaps?
1378 struct buffer_head
*
1379 __getblk_gfp(struct block_device
*bdev
, sector_t block
,
1380 unsigned size
, gfp_t gfp
)
1382 struct buffer_head
*bh
= __find_get_block(bdev
, block
, size
);
1386 bh
= __getblk_slow(bdev
, block
, size
, gfp
);
1389 EXPORT_SYMBOL(__getblk_gfp
);
1392 * Do async read-ahead on a buffer..
1394 void __breadahead(struct block_device
*bdev
, sector_t block
, unsigned size
)
1396 struct buffer_head
*bh
= __getblk(bdev
, block
, size
);
1398 ll_rw_block(REQ_OP_READ
, READA
, 1, &bh
);
1402 EXPORT_SYMBOL(__breadahead
);
1405 * __bread_gfp() - reads a specified block and returns the bh
1406 * @bdev: the block_device to read from
1407 * @block: number of block
1408 * @size: size (in bytes) to read
1409 * @gfp: page allocation flag
1411 * Reads a specified block, and returns buffer head that contains it.
1412 * The page cache can be allocated from non-movable area
1413 * not to prevent page migration if you set gfp to zero.
1414 * It returns NULL if the block was unreadable.
1416 struct buffer_head
*
1417 __bread_gfp(struct block_device
*bdev
, sector_t block
,
1418 unsigned size
, gfp_t gfp
)
1420 struct buffer_head
*bh
= __getblk_gfp(bdev
, block
, size
, gfp
);
1422 if (likely(bh
) && !buffer_uptodate(bh
))
1423 bh
= __bread_slow(bh
);
1426 EXPORT_SYMBOL(__bread_gfp
);
1429 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1430 * This doesn't race because it runs in each cpu either in irq
1431 * or with preempt disabled.
1433 static void invalidate_bh_lru(void *arg
)
1435 struct bh_lru
*b
= &get_cpu_var(bh_lrus
);
1438 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1442 put_cpu_var(bh_lrus
);
1445 static bool has_bh_in_lru(int cpu
, void *dummy
)
1447 struct bh_lru
*b
= per_cpu_ptr(&bh_lrus
, cpu
);
1450 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1458 void invalidate_bh_lrus(void)
1460 on_each_cpu_cond(has_bh_in_lru
, invalidate_bh_lru
, NULL
, 1, GFP_KERNEL
);
1462 EXPORT_SYMBOL_GPL(invalidate_bh_lrus
);
1464 void set_bh_page(struct buffer_head
*bh
,
1465 struct page
*page
, unsigned long offset
)
1468 BUG_ON(offset
>= PAGE_SIZE
);
1469 if (PageHighMem(page
))
1471 * This catches illegal uses and preserves the offset:
1473 bh
->b_data
= (char *)(0 + offset
);
1475 bh
->b_data
= page_address(page
) + offset
;
1477 EXPORT_SYMBOL(set_bh_page
);
1480 * Called when truncating a buffer on a page completely.
1483 /* Bits that are cleared during an invalidate */
1484 #define BUFFER_FLAGS_DISCARD \
1485 (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1486 1 << BH_Delay | 1 << BH_Unwritten)
1488 static void discard_buffer(struct buffer_head
* bh
)
1490 unsigned long b_state
, b_state_old
;
1493 clear_buffer_dirty(bh
);
1495 b_state
= bh
->b_state
;
1497 b_state_old
= cmpxchg(&bh
->b_state
, b_state
,
1498 (b_state
& ~BUFFER_FLAGS_DISCARD
));
1499 if (b_state_old
== b_state
)
1501 b_state
= b_state_old
;
1507 * block_invalidatepage - invalidate part or all of a buffer-backed page
1509 * @page: the page which is affected
1510 * @offset: start of the range to invalidate
1511 * @length: length of the range to invalidate
1513 * block_invalidatepage() is called when all or part of the page has become
1514 * invalidated by a truncate operation.
1516 * block_invalidatepage() does not have to release all buffers, but it must
1517 * ensure that no dirty buffer is left outside @offset and that no I/O
1518 * is underway against any of the blocks which are outside the truncation
1519 * point. Because the caller is about to free (and possibly reuse) those
1522 void block_invalidatepage(struct page
*page
, unsigned int offset
,
1523 unsigned int length
)
1525 struct buffer_head
*head
, *bh
, *next
;
1526 unsigned int curr_off
= 0;
1527 unsigned int stop
= length
+ offset
;
1529 BUG_ON(!PageLocked(page
));
1530 if (!page_has_buffers(page
))
1534 * Check for overflow
1536 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1538 head
= page_buffers(page
);
1541 unsigned int next_off
= curr_off
+ bh
->b_size
;
1542 next
= bh
->b_this_page
;
1545 * Are we still fully in range ?
1547 if (next_off
> stop
)
1551 * is this block fully invalidated?
1553 if (offset
<= curr_off
)
1555 curr_off
= next_off
;
1557 } while (bh
!= head
);
1560 * We release buffers only if the entire page is being invalidated.
1561 * The get_block cached value has been unconditionally invalidated,
1562 * so real IO is not possible anymore.
1565 try_to_release_page(page
, 0);
1569 EXPORT_SYMBOL(block_invalidatepage
);
1573 * We attach and possibly dirty the buffers atomically wrt
1574 * __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1575 * is already excluded via the page lock.
1577 void create_empty_buffers(struct page
*page
,
1578 unsigned long blocksize
, unsigned long b_state
)
1580 struct buffer_head
*bh
, *head
, *tail
;
1582 head
= alloc_page_buffers(page
, blocksize
, 1);
1585 bh
->b_state
|= b_state
;
1587 bh
= bh
->b_this_page
;
1589 tail
->b_this_page
= head
;
1591 spin_lock(&page
->mapping
->private_lock
);
1592 if (PageUptodate(page
) || PageDirty(page
)) {
1595 if (PageDirty(page
))
1596 set_buffer_dirty(bh
);
1597 if (PageUptodate(page
))
1598 set_buffer_uptodate(bh
);
1599 bh
= bh
->b_this_page
;
1600 } while (bh
!= head
);
1602 attach_page_buffers(page
, head
);
1603 spin_unlock(&page
->mapping
->private_lock
);
1605 EXPORT_SYMBOL(create_empty_buffers
);
1608 * We are taking a block for data and we don't want any output from any
1609 * buffer-cache aliases starting from return from that function and
1610 * until the moment when something will explicitly mark the buffer
1611 * dirty (hopefully that will not happen until we will free that block ;-)
1612 * We don't even need to mark it not-uptodate - nobody can expect
1613 * anything from a newly allocated buffer anyway. We used to used
1614 * unmap_buffer() for such invalidation, but that was wrong. We definitely
1615 * don't want to mark the alias unmapped, for example - it would confuse
1616 * anyone who might pick it with bread() afterwards...
1618 * Also.. Note that bforget() doesn't lock the buffer. So there can
1619 * be writeout I/O going on against recently-freed buffers. We don't
1620 * wait on that I/O in bforget() - it's more efficient to wait on the I/O
1621 * only if we really need to. That happens here.
1623 void unmap_underlying_metadata(struct block_device
*bdev
, sector_t block
)
1625 struct buffer_head
*old_bh
;
1629 old_bh
= __find_get_block_slow(bdev
, block
);
1631 clear_buffer_dirty(old_bh
);
1632 wait_on_buffer(old_bh
);
1633 clear_buffer_req(old_bh
);
1637 EXPORT_SYMBOL(unmap_underlying_metadata
);
1640 * Size is a power-of-two in the range 512..PAGE_SIZE,
1641 * and the case we care about most is PAGE_SIZE.
1643 * So this *could* possibly be written with those
1644 * constraints in mind (relevant mostly if some
1645 * architecture has a slow bit-scan instruction)
1647 static inline int block_size_bits(unsigned int blocksize
)
1649 return ilog2(blocksize
);
1652 static struct buffer_head
*create_page_buffers(struct page
*page
, struct inode
*inode
, unsigned int b_state
)
1654 BUG_ON(!PageLocked(page
));
1656 if (!page_has_buffers(page
))
1657 create_empty_buffers(page
, 1 << ACCESS_ONCE(inode
->i_blkbits
), b_state
);
1658 return page_buffers(page
);
1662 * NOTE! All mapped/uptodate combinations are valid:
1664 * Mapped Uptodate Meaning
1666 * No No "unknown" - must do get_block()
1667 * No Yes "hole" - zero-filled
1668 * Yes No "allocated" - allocated on disk, not read in
1669 * Yes Yes "valid" - allocated and up-to-date in memory.
1671 * "Dirty" is valid only with the last case (mapped+uptodate).
1675 * While block_write_full_page is writing back the dirty buffers under
1676 * the page lock, whoever dirtied the buffers may decide to clean them
1677 * again at any time. We handle that by only looking at the buffer
1678 * state inside lock_buffer().
1680 * If block_write_full_page() is called for regular writeback
1681 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1682 * locked buffer. This only can happen if someone has written the buffer
1683 * directly, with submit_bh(). At the address_space level PageWriteback
1684 * prevents this contention from occurring.
1686 * If block_write_full_page() is called with wbc->sync_mode ==
1687 * WB_SYNC_ALL, the writes are posted using WRITE_SYNC; this
1688 * causes the writes to be flagged as synchronous writes.
1690 int __block_write_full_page(struct inode
*inode
, struct page
*page
,
1691 get_block_t
*get_block
, struct writeback_control
*wbc
,
1692 bh_end_io_t
*handler
)
1696 sector_t last_block
;
1697 struct buffer_head
*bh
, *head
;
1698 unsigned int blocksize
, bbits
;
1699 int nr_underway
= 0;
1700 int write_flags
= (wbc
->sync_mode
== WB_SYNC_ALL
? WRITE_SYNC
: 0);
1702 head
= create_page_buffers(page
, inode
,
1703 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
1706 * Be very careful. We have no exclusion from __set_page_dirty_buffers
1707 * here, and the (potentially unmapped) buffers may become dirty at
1708 * any time. If a buffer becomes dirty here after we've inspected it
1709 * then we just miss that fact, and the page stays dirty.
1711 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1712 * handle that here by just cleaning them.
1716 blocksize
= bh
->b_size
;
1717 bbits
= block_size_bits(blocksize
);
1719 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1720 last_block
= (i_size_read(inode
) - 1) >> bbits
;
1723 * Get all the dirty buffers mapped to disk addresses and
1724 * handle any aliases from the underlying blockdev's mapping.
1727 if (block
> last_block
) {
1729 * mapped buffers outside i_size will occur, because
1730 * this page can be outside i_size when there is a
1731 * truncate in progress.
1734 * The buffer was zeroed by block_write_full_page()
1736 clear_buffer_dirty(bh
);
1737 set_buffer_uptodate(bh
);
1738 } else if ((!buffer_mapped(bh
) || buffer_delay(bh
)) &&
1740 WARN_ON(bh
->b_size
!= blocksize
);
1741 err
= get_block(inode
, block
, bh
, 1);
1744 clear_buffer_delay(bh
);
1745 if (buffer_new(bh
)) {
1746 /* blockdev mappings never come here */
1747 clear_buffer_new(bh
);
1748 unmap_underlying_metadata(bh
->b_bdev
,
1752 bh
= bh
->b_this_page
;
1754 } while (bh
!= head
);
1757 if (!buffer_mapped(bh
))
1760 * If it's a fully non-blocking write attempt and we cannot
1761 * lock the buffer then redirty the page. Note that this can
1762 * potentially cause a busy-wait loop from writeback threads
1763 * and kswapd activity, but those code paths have their own
1764 * higher-level throttling.
1766 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
1768 } else if (!trylock_buffer(bh
)) {
1769 redirty_page_for_writepage(wbc
, page
);
1772 if (test_clear_buffer_dirty(bh
)) {
1773 mark_buffer_async_write_endio(bh
, handler
);
1777 } while ((bh
= bh
->b_this_page
) != head
);
1780 * The page and its buffers are protected by PageWriteback(), so we can
1781 * drop the bh refcounts early.
1783 BUG_ON(PageWriteback(page
));
1784 set_page_writeback(page
);
1787 struct buffer_head
*next
= bh
->b_this_page
;
1788 if (buffer_async_write(bh
)) {
1789 submit_bh_wbc(REQ_OP_WRITE
, write_flags
, bh
, 0, wbc
);
1793 } while (bh
!= head
);
1798 if (nr_underway
== 0) {
1800 * The page was marked dirty, but the buffers were
1801 * clean. Someone wrote them back by hand with
1802 * ll_rw_block/submit_bh. A rare case.
1804 end_page_writeback(page
);
1807 * The page and buffer_heads can be released at any time from
1815 * ENOSPC, or some other error. We may already have added some
1816 * blocks to the file, so we need to write these out to avoid
1817 * exposing stale data.
1818 * The page is currently locked and not marked for writeback
1821 /* Recovery: lock and submit the mapped buffers */
1823 if (buffer_mapped(bh
) && buffer_dirty(bh
) &&
1824 !buffer_delay(bh
)) {
1826 mark_buffer_async_write_endio(bh
, handler
);
1829 * The buffer may have been set dirty during
1830 * attachment to a dirty page.
1832 clear_buffer_dirty(bh
);
1834 } while ((bh
= bh
->b_this_page
) != head
);
1836 BUG_ON(PageWriteback(page
));
1837 mapping_set_error(page
->mapping
, err
);
1838 set_page_writeback(page
);
1840 struct buffer_head
*next
= bh
->b_this_page
;
1841 if (buffer_async_write(bh
)) {
1842 clear_buffer_dirty(bh
);
1843 submit_bh_wbc(REQ_OP_WRITE
, write_flags
, bh
, 0, wbc
);
1847 } while (bh
!= head
);
1851 EXPORT_SYMBOL(__block_write_full_page
);
1854 * If a page has any new buffers, zero them out here, and mark them uptodate
1855 * and dirty so they'll be written out (in order to prevent uninitialised
1856 * block data from leaking). And clear the new bit.
1858 void page_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1860 unsigned int block_start
, block_end
;
1861 struct buffer_head
*head
, *bh
;
1863 BUG_ON(!PageLocked(page
));
1864 if (!page_has_buffers(page
))
1867 bh
= head
= page_buffers(page
);
1870 block_end
= block_start
+ bh
->b_size
;
1872 if (buffer_new(bh
)) {
1873 if (block_end
> from
&& block_start
< to
) {
1874 if (!PageUptodate(page
)) {
1875 unsigned start
, size
;
1877 start
= max(from
, block_start
);
1878 size
= min(to
, block_end
) - start
;
1880 zero_user(page
, start
, size
);
1881 set_buffer_uptodate(bh
);
1884 clear_buffer_new(bh
);
1885 mark_buffer_dirty(bh
);
1889 block_start
= block_end
;
1890 bh
= bh
->b_this_page
;
1891 } while (bh
!= head
);
1893 EXPORT_SYMBOL(page_zero_new_buffers
);
1895 int __block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1896 get_block_t
*get_block
)
1898 unsigned from
= pos
& (PAGE_SIZE
- 1);
1899 unsigned to
= from
+ len
;
1900 struct inode
*inode
= page
->mapping
->host
;
1901 unsigned block_start
, block_end
;
1904 unsigned blocksize
, bbits
;
1905 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
=wait
;
1907 BUG_ON(!PageLocked(page
));
1908 BUG_ON(from
> PAGE_SIZE
);
1909 BUG_ON(to
> PAGE_SIZE
);
1912 head
= create_page_buffers(page
, inode
, 0);
1913 blocksize
= head
->b_size
;
1914 bbits
= block_size_bits(blocksize
);
1916 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1918 for(bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1919 block
++, block_start
=block_end
, bh
= bh
->b_this_page
) {
1920 block_end
= block_start
+ blocksize
;
1921 if (block_end
<= from
|| block_start
>= to
) {
1922 if (PageUptodate(page
)) {
1923 if (!buffer_uptodate(bh
))
1924 set_buffer_uptodate(bh
);
1929 clear_buffer_new(bh
);
1930 if (!buffer_mapped(bh
)) {
1931 WARN_ON(bh
->b_size
!= blocksize
);
1932 err
= get_block(inode
, block
, bh
, 1);
1935 if (buffer_new(bh
)) {
1936 unmap_underlying_metadata(bh
->b_bdev
,
1938 if (PageUptodate(page
)) {
1939 clear_buffer_new(bh
);
1940 set_buffer_uptodate(bh
);
1941 mark_buffer_dirty(bh
);
1944 if (block_end
> to
|| block_start
< from
)
1945 zero_user_segments(page
,
1951 if (PageUptodate(page
)) {
1952 if (!buffer_uptodate(bh
))
1953 set_buffer_uptodate(bh
);
1956 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1957 !buffer_unwritten(bh
) &&
1958 (block_start
< from
|| block_end
> to
)) {
1959 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
1964 * If we issued read requests - let them complete.
1966 while(wait_bh
> wait
) {
1967 wait_on_buffer(*--wait_bh
);
1968 if (!buffer_uptodate(*wait_bh
))
1972 page_zero_new_buffers(page
, from
, to
);
1975 EXPORT_SYMBOL(__block_write_begin
);
1977 static int __block_commit_write(struct inode
*inode
, struct page
*page
,
1978 unsigned from
, unsigned to
)
1980 unsigned block_start
, block_end
;
1983 struct buffer_head
*bh
, *head
;
1985 bh
= head
= page_buffers(page
);
1986 blocksize
= bh
->b_size
;
1990 block_end
= block_start
+ blocksize
;
1991 if (block_end
<= from
|| block_start
>= to
) {
1992 if (!buffer_uptodate(bh
))
1995 set_buffer_uptodate(bh
);
1996 mark_buffer_dirty(bh
);
1998 clear_buffer_new(bh
);
2000 block_start
= block_end
;
2001 bh
= bh
->b_this_page
;
2002 } while (bh
!= head
);
2005 * If this is a partial write which happened to make all buffers
2006 * uptodate then we can optimize away a bogus readpage() for
2007 * the next read(). Here we 'discover' whether the page went
2008 * uptodate as a result of this (potentially partial) write.
2011 SetPageUptodate(page
);
2016 * block_write_begin takes care of the basic task of block allocation and
2017 * bringing partial write blocks uptodate first.
2019 * The filesystem needs to handle block truncation upon failure.
2021 int block_write_begin(struct address_space
*mapping
, loff_t pos
, unsigned len
,
2022 unsigned flags
, struct page
**pagep
, get_block_t
*get_block
)
2024 pgoff_t index
= pos
>> PAGE_SHIFT
;
2028 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2032 status
= __block_write_begin(page
, pos
, len
, get_block
);
2033 if (unlikely(status
)) {
2042 EXPORT_SYMBOL(block_write_begin
);
2044 int block_write_end(struct file
*file
, struct address_space
*mapping
,
2045 loff_t pos
, unsigned len
, unsigned copied
,
2046 struct page
*page
, void *fsdata
)
2048 struct inode
*inode
= mapping
->host
;
2051 start
= pos
& (PAGE_SIZE
- 1);
2053 if (unlikely(copied
< len
)) {
2055 * The buffers that were written will now be uptodate, so we
2056 * don't have to worry about a readpage reading them and
2057 * overwriting a partial write. However if we have encountered
2058 * a short write and only partially written into a buffer, it
2059 * will not be marked uptodate, so a readpage might come in and
2060 * destroy our partial write.
2062 * Do the simplest thing, and just treat any short write to a
2063 * non uptodate page as a zero-length write, and force the
2064 * caller to redo the whole thing.
2066 if (!PageUptodate(page
))
2069 page_zero_new_buffers(page
, start
+copied
, start
+len
);
2071 flush_dcache_page(page
);
2073 /* This could be a short (even 0-length) commit */
2074 __block_commit_write(inode
, page
, start
, start
+copied
);
2078 EXPORT_SYMBOL(block_write_end
);
2080 int generic_write_end(struct file
*file
, struct address_space
*mapping
,
2081 loff_t pos
, unsigned len
, unsigned copied
,
2082 struct page
*page
, void *fsdata
)
2084 struct inode
*inode
= mapping
->host
;
2085 loff_t old_size
= inode
->i_size
;
2086 int i_size_changed
= 0;
2088 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2091 * No need to use i_size_read() here, the i_size
2092 * cannot change under us because we hold i_mutex.
2094 * But it's important to update i_size while still holding page lock:
2095 * page writeout could otherwise come in and zero beyond i_size.
2097 if (pos
+copied
> inode
->i_size
) {
2098 i_size_write(inode
, pos
+copied
);
2106 pagecache_isize_extended(inode
, old_size
, pos
);
2108 * Don't mark the inode dirty under page lock. First, it unnecessarily
2109 * makes the holding time of page lock longer. Second, it forces lock
2110 * ordering of page lock and transaction start for journaling
2114 mark_inode_dirty(inode
);
2118 EXPORT_SYMBOL(generic_write_end
);
2121 * block_is_partially_uptodate checks whether buffers within a page are
2124 * Returns true if all buffers which correspond to a file portion
2125 * we want to read are uptodate.
2127 int block_is_partially_uptodate(struct page
*page
, unsigned long from
,
2128 unsigned long count
)
2130 unsigned block_start
, block_end
, blocksize
;
2132 struct buffer_head
*bh
, *head
;
2135 if (!page_has_buffers(page
))
2138 head
= page_buffers(page
);
2139 blocksize
= head
->b_size
;
2140 to
= min_t(unsigned, PAGE_SIZE
- from
, count
);
2142 if (from
< blocksize
&& to
> PAGE_SIZE
- blocksize
)
2148 block_end
= block_start
+ blocksize
;
2149 if (block_end
> from
&& block_start
< to
) {
2150 if (!buffer_uptodate(bh
)) {
2154 if (block_end
>= to
)
2157 block_start
= block_end
;
2158 bh
= bh
->b_this_page
;
2159 } while (bh
!= head
);
2163 EXPORT_SYMBOL(block_is_partially_uptodate
);
2166 * Generic "read page" function for block devices that have the normal
2167 * get_block functionality. This is most of the block device filesystems.
2168 * Reads the page asynchronously --- the unlock_buffer() and
2169 * set/clear_buffer_uptodate() functions propagate buffer state into the
2170 * page struct once IO has completed.
2172 int block_read_full_page(struct page
*page
, get_block_t
*get_block
)
2174 struct inode
*inode
= page
->mapping
->host
;
2175 sector_t iblock
, lblock
;
2176 struct buffer_head
*bh
, *head
, *arr
[MAX_BUF_PER_PAGE
];
2177 unsigned int blocksize
, bbits
;
2179 int fully_mapped
= 1;
2181 head
= create_page_buffers(page
, inode
, 0);
2182 blocksize
= head
->b_size
;
2183 bbits
= block_size_bits(blocksize
);
2185 iblock
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
2186 lblock
= (i_size_read(inode
)+blocksize
-1) >> bbits
;
2192 if (buffer_uptodate(bh
))
2195 if (!buffer_mapped(bh
)) {
2199 if (iblock
< lblock
) {
2200 WARN_ON(bh
->b_size
!= blocksize
);
2201 err
= get_block(inode
, iblock
, bh
, 0);
2205 if (!buffer_mapped(bh
)) {
2206 zero_user(page
, i
* blocksize
, blocksize
);
2208 set_buffer_uptodate(bh
);
2212 * get_block() might have updated the buffer
2215 if (buffer_uptodate(bh
))
2219 } while (i
++, iblock
++, (bh
= bh
->b_this_page
) != head
);
2222 SetPageMappedToDisk(page
);
2226 * All buffers are uptodate - we can set the page uptodate
2227 * as well. But not if get_block() returned an error.
2229 if (!PageError(page
))
2230 SetPageUptodate(page
);
2235 /* Stage two: lock the buffers */
2236 for (i
= 0; i
< nr
; i
++) {
2239 mark_buffer_async_read(bh
);
2243 * Stage 3: start the IO. Check for uptodateness
2244 * inside the buffer lock in case another process reading
2245 * the underlying blockdev brought it uptodate (the sct fix).
2247 for (i
= 0; i
< nr
; i
++) {
2249 if (buffer_uptodate(bh
))
2250 end_buffer_async_read(bh
, 1);
2252 submit_bh(REQ_OP_READ
, 0, bh
);
2256 EXPORT_SYMBOL(block_read_full_page
);
2258 /* utility function for filesystems that need to do work on expanding
2259 * truncates. Uses filesystem pagecache writes to allow the filesystem to
2260 * deal with the hole.
2262 int generic_cont_expand_simple(struct inode
*inode
, loff_t size
)
2264 struct address_space
*mapping
= inode
->i_mapping
;
2269 err
= inode_newsize_ok(inode
, size
);
2273 err
= pagecache_write_begin(NULL
, mapping
, size
, 0,
2274 AOP_FLAG_UNINTERRUPTIBLE
|AOP_FLAG_CONT_EXPAND
,
2279 err
= pagecache_write_end(NULL
, mapping
, size
, 0, 0, page
, fsdata
);
2285 EXPORT_SYMBOL(generic_cont_expand_simple
);
2287 static int cont_expand_zero(struct file
*file
, struct address_space
*mapping
,
2288 loff_t pos
, loff_t
*bytes
)
2290 struct inode
*inode
= mapping
->host
;
2291 unsigned blocksize
= 1 << inode
->i_blkbits
;
2294 pgoff_t index
, curidx
;
2296 unsigned zerofrom
, offset
, len
;
2299 index
= pos
>> PAGE_SHIFT
;
2300 offset
= pos
& ~PAGE_MASK
;
2302 while (index
> (curidx
= (curpos
= *bytes
)>>PAGE_SHIFT
)) {
2303 zerofrom
= curpos
& ~PAGE_MASK
;
2304 if (zerofrom
& (blocksize
-1)) {
2305 *bytes
|= (blocksize
-1);
2308 len
= PAGE_SIZE
- zerofrom
;
2310 err
= pagecache_write_begin(file
, mapping
, curpos
, len
,
2311 AOP_FLAG_UNINTERRUPTIBLE
,
2315 zero_user(page
, zerofrom
, len
);
2316 err
= pagecache_write_end(file
, mapping
, curpos
, len
, len
,
2323 balance_dirty_pages_ratelimited(mapping
);
2325 if (unlikely(fatal_signal_pending(current
))) {
2331 /* page covers the boundary, find the boundary offset */
2332 if (index
== curidx
) {
2333 zerofrom
= curpos
& ~PAGE_MASK
;
2334 /* if we will expand the thing last block will be filled */
2335 if (offset
<= zerofrom
) {
2338 if (zerofrom
& (blocksize
-1)) {
2339 *bytes
|= (blocksize
-1);
2342 len
= offset
- zerofrom
;
2344 err
= pagecache_write_begin(file
, mapping
, curpos
, len
,
2345 AOP_FLAG_UNINTERRUPTIBLE
,
2349 zero_user(page
, zerofrom
, len
);
2350 err
= pagecache_write_end(file
, mapping
, curpos
, len
, len
,
2362 * For moronic filesystems that do not allow holes in file.
2363 * We may have to extend the file.
2365 int cont_write_begin(struct file
*file
, struct address_space
*mapping
,
2366 loff_t pos
, unsigned len
, unsigned flags
,
2367 struct page
**pagep
, void **fsdata
,
2368 get_block_t
*get_block
, loff_t
*bytes
)
2370 struct inode
*inode
= mapping
->host
;
2371 unsigned blocksize
= 1 << inode
->i_blkbits
;
2375 err
= cont_expand_zero(file
, mapping
, pos
, bytes
);
2379 zerofrom
= *bytes
& ~PAGE_MASK
;
2380 if (pos
+len
> *bytes
&& zerofrom
& (blocksize
-1)) {
2381 *bytes
|= (blocksize
-1);
2385 return block_write_begin(mapping
, pos
, len
, flags
, pagep
, get_block
);
2387 EXPORT_SYMBOL(cont_write_begin
);
2389 int block_commit_write(struct page
*page
, unsigned from
, unsigned to
)
2391 struct inode
*inode
= page
->mapping
->host
;
2392 __block_commit_write(inode
,page
,from
,to
);
2395 EXPORT_SYMBOL(block_commit_write
);
2398 * block_page_mkwrite() is not allowed to change the file size as it gets
2399 * called from a page fault handler when a page is first dirtied. Hence we must
2400 * be careful to check for EOF conditions here. We set the page up correctly
2401 * for a written page which means we get ENOSPC checking when writing into
2402 * holes and correct delalloc and unwritten extent mapping on filesystems that
2403 * support these features.
2405 * We are not allowed to take the i_mutex here so we have to play games to
2406 * protect against truncate races as the page could now be beyond EOF. Because
2407 * truncate writes the inode size before removing pages, once we have the
2408 * page lock we can determine safely if the page is beyond EOF. If it is not
2409 * beyond EOF, then the page is guaranteed safe against truncation until we
2412 * Direct callers of this function should protect against filesystem freezing
2413 * using sb_start_pagefault() - sb_end_pagefault() functions.
2415 int block_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
2416 get_block_t get_block
)
2418 struct page
*page
= vmf
->page
;
2419 struct inode
*inode
= file_inode(vma
->vm_file
);
2425 size
= i_size_read(inode
);
2426 if ((page
->mapping
!= inode
->i_mapping
) ||
2427 (page_offset(page
) > size
)) {
2428 /* We overload EFAULT to mean page got truncated */
2433 /* page is wholly or partially inside EOF */
2434 if (((page
->index
+ 1) << PAGE_SHIFT
) > size
)
2435 end
= size
& ~PAGE_MASK
;
2439 ret
= __block_write_begin(page
, 0, end
, get_block
);
2441 ret
= block_commit_write(page
, 0, end
);
2443 if (unlikely(ret
< 0))
2445 set_page_dirty(page
);
2446 wait_for_stable_page(page
);
2452 EXPORT_SYMBOL(block_page_mkwrite
);
2455 * nobh_write_begin()'s prereads are special: the buffer_heads are freed
2456 * immediately, while under the page lock. So it needs a special end_io
2457 * handler which does not touch the bh after unlocking it.
2459 static void end_buffer_read_nobh(struct buffer_head
*bh
, int uptodate
)
2461 __end_buffer_read_notouch(bh
, uptodate
);
2465 * Attach the singly-linked list of buffers created by nobh_write_begin, to
2466 * the page (converting it to circular linked list and taking care of page
2469 static void attach_nobh_buffers(struct page
*page
, struct buffer_head
*head
)
2471 struct buffer_head
*bh
;
2473 BUG_ON(!PageLocked(page
));
2475 spin_lock(&page
->mapping
->private_lock
);
2478 if (PageDirty(page
))
2479 set_buffer_dirty(bh
);
2480 if (!bh
->b_this_page
)
2481 bh
->b_this_page
= head
;
2482 bh
= bh
->b_this_page
;
2483 } while (bh
!= head
);
2484 attach_page_buffers(page
, head
);
2485 spin_unlock(&page
->mapping
->private_lock
);
2489 * On entry, the page is fully not uptodate.
2490 * On exit the page is fully uptodate in the areas outside (from,to)
2491 * The filesystem needs to handle block truncation upon failure.
2493 int nobh_write_begin(struct address_space
*mapping
,
2494 loff_t pos
, unsigned len
, unsigned flags
,
2495 struct page
**pagep
, void **fsdata
,
2496 get_block_t
*get_block
)
2498 struct inode
*inode
= mapping
->host
;
2499 const unsigned blkbits
= inode
->i_blkbits
;
2500 const unsigned blocksize
= 1 << blkbits
;
2501 struct buffer_head
*head
, *bh
;
2505 unsigned block_in_page
;
2506 unsigned block_start
, block_end
;
2507 sector_t block_in_file
;
2510 int is_mapped_to_disk
= 1;
2512 index
= pos
>> PAGE_SHIFT
;
2513 from
= pos
& (PAGE_SIZE
- 1);
2516 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2522 if (page_has_buffers(page
)) {
2523 ret
= __block_write_begin(page
, pos
, len
, get_block
);
2529 if (PageMappedToDisk(page
))
2533 * Allocate buffers so that we can keep track of state, and potentially
2534 * attach them to the page if an error occurs. In the common case of
2535 * no error, they will just be freed again without ever being attached
2536 * to the page (which is all OK, because we're under the page lock).
2538 * Be careful: the buffer linked list is a NULL terminated one, rather
2539 * than the circular one we're used to.
2541 head
= alloc_page_buffers(page
, blocksize
, 0);
2547 block_in_file
= (sector_t
)page
->index
<< (PAGE_SHIFT
- blkbits
);
2550 * We loop across all blocks in the page, whether or not they are
2551 * part of the affected region. This is so we can discover if the
2552 * page is fully mapped-to-disk.
2554 for (block_start
= 0, block_in_page
= 0, bh
= head
;
2555 block_start
< PAGE_SIZE
;
2556 block_in_page
++, block_start
+= blocksize
, bh
= bh
->b_this_page
) {
2559 block_end
= block_start
+ blocksize
;
2562 if (block_start
>= to
)
2564 ret
= get_block(inode
, block_in_file
+ block_in_page
,
2568 if (!buffer_mapped(bh
))
2569 is_mapped_to_disk
= 0;
2571 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
2572 if (PageUptodate(page
)) {
2573 set_buffer_uptodate(bh
);
2576 if (buffer_new(bh
) || !buffer_mapped(bh
)) {
2577 zero_user_segments(page
, block_start
, from
,
2581 if (buffer_uptodate(bh
))
2582 continue; /* reiserfs does this */
2583 if (block_start
< from
|| block_end
> to
) {
2585 bh
->b_end_io
= end_buffer_read_nobh
;
2586 submit_bh(REQ_OP_READ
, 0, bh
);
2593 * The page is locked, so these buffers are protected from
2594 * any VM or truncate activity. Hence we don't need to care
2595 * for the buffer_head refcounts.
2597 for (bh
= head
; bh
; bh
= bh
->b_this_page
) {
2599 if (!buffer_uptodate(bh
))
2606 if (is_mapped_to_disk
)
2607 SetPageMappedToDisk(page
);
2609 *fsdata
= head
; /* to be released by nobh_write_end */
2616 * Error recovery is a bit difficult. We need to zero out blocks that
2617 * were newly allocated, and dirty them to ensure they get written out.
2618 * Buffers need to be attached to the page at this point, otherwise
2619 * the handling of potential IO errors during writeout would be hard
2620 * (could try doing synchronous writeout, but what if that fails too?)
2622 attach_nobh_buffers(page
, head
);
2623 page_zero_new_buffers(page
, from
, to
);
2632 EXPORT_SYMBOL(nobh_write_begin
);
2634 int nobh_write_end(struct file
*file
, struct address_space
*mapping
,
2635 loff_t pos
, unsigned len
, unsigned copied
,
2636 struct page
*page
, void *fsdata
)
2638 struct inode
*inode
= page
->mapping
->host
;
2639 struct buffer_head
*head
= fsdata
;
2640 struct buffer_head
*bh
;
2641 BUG_ON(fsdata
!= NULL
&& page_has_buffers(page
));
2643 if (unlikely(copied
< len
) && head
)
2644 attach_nobh_buffers(page
, head
);
2645 if (page_has_buffers(page
))
2646 return generic_write_end(file
, mapping
, pos
, len
,
2647 copied
, page
, fsdata
);
2649 SetPageUptodate(page
);
2650 set_page_dirty(page
);
2651 if (pos
+copied
> inode
->i_size
) {
2652 i_size_write(inode
, pos
+copied
);
2653 mark_inode_dirty(inode
);
2661 head
= head
->b_this_page
;
2662 free_buffer_head(bh
);
2667 EXPORT_SYMBOL(nobh_write_end
);
2670 * nobh_writepage() - based on block_full_write_page() except
2671 * that it tries to operate without attaching bufferheads to
2674 int nobh_writepage(struct page
*page
, get_block_t
*get_block
,
2675 struct writeback_control
*wbc
)
2677 struct inode
* const inode
= page
->mapping
->host
;
2678 loff_t i_size
= i_size_read(inode
);
2679 const pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2683 /* Is the page fully inside i_size? */
2684 if (page
->index
< end_index
)
2687 /* Is the page fully outside i_size? (truncate in progress) */
2688 offset
= i_size
& (PAGE_SIZE
-1);
2689 if (page
->index
>= end_index
+1 || !offset
) {
2691 * The page may have dirty, unmapped buffers. For example,
2692 * they may have been added in ext3_writepage(). Make them
2693 * freeable here, so the page does not leak.
2696 /* Not really sure about this - do we need this ? */
2697 if (page
->mapping
->a_ops
->invalidatepage
)
2698 page
->mapping
->a_ops
->invalidatepage(page
, offset
);
2701 return 0; /* don't care */
2705 * The page straddles i_size. It must be zeroed out on each and every
2706 * writepage invocation because it may be mmapped. "A file is mapped
2707 * in multiples of the page size. For a file that is not a multiple of
2708 * the page size, the remaining memory is zeroed when mapped, and
2709 * writes to that region are not written out to the file."
2711 zero_user_segment(page
, offset
, PAGE_SIZE
);
2713 ret
= mpage_writepage(page
, get_block
, wbc
);
2715 ret
= __block_write_full_page(inode
, page
, get_block
, wbc
,
2716 end_buffer_async_write
);
2719 EXPORT_SYMBOL(nobh_writepage
);
2721 int nobh_truncate_page(struct address_space
*mapping
,
2722 loff_t from
, get_block_t
*get_block
)
2724 pgoff_t index
= from
>> PAGE_SHIFT
;
2725 unsigned offset
= from
& (PAGE_SIZE
-1);
2728 unsigned length
, pos
;
2729 struct inode
*inode
= mapping
->host
;
2731 struct buffer_head map_bh
;
2734 blocksize
= 1 << inode
->i_blkbits
;
2735 length
= offset
& (blocksize
- 1);
2737 /* Block boundary? Nothing to do */
2741 length
= blocksize
- length
;
2742 iblock
= (sector_t
)index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
2744 page
= grab_cache_page(mapping
, index
);
2749 if (page_has_buffers(page
)) {
2753 return block_truncate_page(mapping
, from
, get_block
);
2756 /* Find the buffer that contains "offset" */
2758 while (offset
>= pos
) {
2763 map_bh
.b_size
= blocksize
;
2765 err
= get_block(inode
, iblock
, &map_bh
, 0);
2768 /* unmapped? It's a hole - nothing to do */
2769 if (!buffer_mapped(&map_bh
))
2772 /* Ok, it's mapped. Make sure it's up-to-date */
2773 if (!PageUptodate(page
)) {
2774 err
= mapping
->a_ops
->readpage(NULL
, page
);
2780 if (!PageUptodate(page
)) {
2784 if (page_has_buffers(page
))
2787 zero_user(page
, offset
, length
);
2788 set_page_dirty(page
);
2797 EXPORT_SYMBOL(nobh_truncate_page
);
2799 int block_truncate_page(struct address_space
*mapping
,
2800 loff_t from
, get_block_t
*get_block
)
2802 pgoff_t index
= from
>> PAGE_SHIFT
;
2803 unsigned offset
= from
& (PAGE_SIZE
-1);
2806 unsigned length
, pos
;
2807 struct inode
*inode
= mapping
->host
;
2809 struct buffer_head
*bh
;
2812 blocksize
= 1 << inode
->i_blkbits
;
2813 length
= offset
& (blocksize
- 1);
2815 /* Block boundary? Nothing to do */
2819 length
= blocksize
- length
;
2820 iblock
= (sector_t
)index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
2822 page
= grab_cache_page(mapping
, index
);
2827 if (!page_has_buffers(page
))
2828 create_empty_buffers(page
, blocksize
, 0);
2830 /* Find the buffer that contains "offset" */
2831 bh
= page_buffers(page
);
2833 while (offset
>= pos
) {
2834 bh
= bh
->b_this_page
;
2840 if (!buffer_mapped(bh
)) {
2841 WARN_ON(bh
->b_size
!= blocksize
);
2842 err
= get_block(inode
, iblock
, bh
, 0);
2845 /* unmapped? It's a hole - nothing to do */
2846 if (!buffer_mapped(bh
))
2850 /* Ok, it's mapped. Make sure it's up-to-date */
2851 if (PageUptodate(page
))
2852 set_buffer_uptodate(bh
);
2854 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) && !buffer_unwritten(bh
)) {
2856 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
2858 /* Uhhuh. Read error. Complain and punt. */
2859 if (!buffer_uptodate(bh
))
2863 zero_user(page
, offset
, length
);
2864 mark_buffer_dirty(bh
);
2873 EXPORT_SYMBOL(block_truncate_page
);
2876 * The generic ->writepage function for buffer-backed address_spaces
2878 int block_write_full_page(struct page
*page
, get_block_t
*get_block
,
2879 struct writeback_control
*wbc
)
2881 struct inode
* const inode
= page
->mapping
->host
;
2882 loff_t i_size
= i_size_read(inode
);
2883 const pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2886 /* Is the page fully inside i_size? */
2887 if (page
->index
< end_index
)
2888 return __block_write_full_page(inode
, page
, get_block
, wbc
,
2889 end_buffer_async_write
);
2891 /* Is the page fully outside i_size? (truncate in progress) */
2892 offset
= i_size
& (PAGE_SIZE
-1);
2893 if (page
->index
>= end_index
+1 || !offset
) {
2895 * The page may have dirty, unmapped buffers. For example,
2896 * they may have been added in ext3_writepage(). Make them
2897 * freeable here, so the page does not leak.
2899 do_invalidatepage(page
, 0, PAGE_SIZE
);
2901 return 0; /* don't care */
2905 * The page straddles i_size. It must be zeroed out on each and every
2906 * writepage invocation because it may be mmapped. "A file is mapped
2907 * in multiples of the page size. For a file that is not a multiple of
2908 * the page size, the remaining memory is zeroed when mapped, and
2909 * writes to that region are not written out to the file."
2911 zero_user_segment(page
, offset
, PAGE_SIZE
);
2912 return __block_write_full_page(inode
, page
, get_block
, wbc
,
2913 end_buffer_async_write
);
2915 EXPORT_SYMBOL(block_write_full_page
);
2917 sector_t
generic_block_bmap(struct address_space
*mapping
, sector_t block
,
2918 get_block_t
*get_block
)
2920 struct buffer_head tmp
;
2921 struct inode
*inode
= mapping
->host
;
2924 tmp
.b_size
= 1 << inode
->i_blkbits
;
2925 get_block(inode
, block
, &tmp
, 0);
2926 return tmp
.b_blocknr
;
2928 EXPORT_SYMBOL(generic_block_bmap
);
2930 static void end_bio_bh_io_sync(struct bio
*bio
)
2932 struct buffer_head
*bh
= bio
->bi_private
;
2934 if (unlikely(bio_flagged(bio
, BIO_QUIET
)))
2935 set_bit(BH_Quiet
, &bh
->b_state
);
2937 bh
->b_end_io(bh
, !bio
->bi_error
);
2942 * This allows us to do IO even on the odd last sectors
2943 * of a device, even if the block size is some multiple
2944 * of the physical sector size.
2946 * We'll just truncate the bio to the size of the device,
2947 * and clear the end of the buffer head manually.
2949 * Truly out-of-range accesses will turn into actual IO
2950 * errors, this only handles the "we need to be able to
2951 * do IO at the final sector" case.
2953 void guard_bio_eod(int op
, struct bio
*bio
)
2956 struct bio_vec
*bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
2957 unsigned truncated_bytes
;
2959 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
2964 * If the *whole* IO is past the end of the device,
2965 * let it through, and the IO layer will turn it into
2968 if (unlikely(bio
->bi_iter
.bi_sector
>= maxsector
))
2971 maxsector
-= bio
->bi_iter
.bi_sector
;
2972 if (likely((bio
->bi_iter
.bi_size
>> 9) <= maxsector
))
2975 /* Uhhuh. We've got a bio that straddles the device size! */
2976 truncated_bytes
= bio
->bi_iter
.bi_size
- (maxsector
<< 9);
2978 /* Truncate the bio.. */
2979 bio
->bi_iter
.bi_size
-= truncated_bytes
;
2980 bvec
->bv_len
-= truncated_bytes
;
2982 /* ..and clear the end of the buffer for reads */
2983 if (op
== REQ_OP_READ
) {
2984 zero_user(bvec
->bv_page
, bvec
->bv_offset
+ bvec
->bv_len
,
2989 static int submit_bh_wbc(int op
, int op_flags
, struct buffer_head
*bh
,
2990 unsigned long bio_flags
, struct writeback_control
*wbc
)
2994 BUG_ON(!buffer_locked(bh
));
2995 BUG_ON(!buffer_mapped(bh
));
2996 BUG_ON(!bh
->b_end_io
);
2997 BUG_ON(buffer_delay(bh
));
2998 BUG_ON(buffer_unwritten(bh
));
3001 * Only clear out a write error when rewriting
3003 if (test_set_buffer_req(bh
) && (op
== REQ_OP_WRITE
))
3004 clear_buffer_write_io_error(bh
);
3007 * from here on down, it's all bio -- do the initial mapping,
3008 * submit_bio -> generic_make_request may further map this bio around
3010 bio
= bio_alloc(GFP_NOIO
, 1);
3013 wbc_init_bio(wbc
, bio
);
3014 wbc_account_io(wbc
, bh
->b_page
, bh
->b_size
);
3017 bio
->bi_iter
.bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
3018 bio
->bi_bdev
= bh
->b_bdev
;
3020 bio_add_page(bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
3021 BUG_ON(bio
->bi_iter
.bi_size
!= bh
->b_size
);
3023 bio
->bi_end_io
= end_bio_bh_io_sync
;
3024 bio
->bi_private
= bh
;
3025 bio
->bi_flags
|= bio_flags
;
3027 /* Take care of bh's that straddle the end of the device */
3028 guard_bio_eod(op
, bio
);
3030 if (buffer_meta(bh
))
3031 op_flags
|= REQ_META
;
3032 if (buffer_prio(bh
))
3033 op_flags
|= REQ_PRIO
;
3034 bio_set_op_attrs(bio
, op
, op_flags
);
3040 int _submit_bh(int op
, int op_flags
, struct buffer_head
*bh
,
3041 unsigned long bio_flags
)
3043 return submit_bh_wbc(op
, op_flags
, bh
, bio_flags
, NULL
);
3045 EXPORT_SYMBOL_GPL(_submit_bh
);
3047 int submit_bh(int op
, int op_flags
, struct buffer_head
*bh
)
3049 return submit_bh_wbc(op
, op_flags
, bh
, 0, NULL
);
3051 EXPORT_SYMBOL(submit_bh
);
3054 * ll_rw_block: low-level access to block devices (DEPRECATED)
3055 * @op: whether to %READ or %WRITE
3056 * @op_flags: rq_flag_bits or %READA (readahead)
3057 * @nr: number of &struct buffer_heads in the array
3058 * @bhs: array of pointers to &struct buffer_head
3060 * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
3061 * requests an I/O operation on them, either a %READ or a %WRITE. The third
3062 * %READA option is described in the documentation for generic_make_request()
3063 * which ll_rw_block() calls.
3065 * This function drops any buffer that it cannot get a lock on (with the
3066 * BH_Lock state bit), any buffer that appears to be clean when doing a write
3067 * request, and any buffer that appears to be up-to-date when doing read
3068 * request. Further it marks as clean buffers that are processed for
3069 * writing (the buffer cache won't assume that they are actually clean
3070 * until the buffer gets unlocked).
3072 * ll_rw_block sets b_end_io to simple completion handler that marks
3073 * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
3076 * All of the buffers must be for the same device, and must also be a
3077 * multiple of the current approved size for the device.
3079 void ll_rw_block(int op
, int op_flags
, int nr
, struct buffer_head
*bhs
[])
3083 for (i
= 0; i
< nr
; i
++) {
3084 struct buffer_head
*bh
= bhs
[i
];
3086 if (!trylock_buffer(bh
))
3089 if (test_clear_buffer_dirty(bh
)) {
3090 bh
->b_end_io
= end_buffer_write_sync
;
3092 submit_bh(op
, op_flags
, bh
);
3096 if (!buffer_uptodate(bh
)) {
3097 bh
->b_end_io
= end_buffer_read_sync
;
3099 submit_bh(op
, op_flags
, bh
);
3106 EXPORT_SYMBOL(ll_rw_block
);
3108 void write_dirty_buffer(struct buffer_head
*bh
, int op_flags
)
3111 if (!test_clear_buffer_dirty(bh
)) {
3115 bh
->b_end_io
= end_buffer_write_sync
;
3117 submit_bh(REQ_OP_WRITE
, op_flags
, bh
);
3119 EXPORT_SYMBOL(write_dirty_buffer
);
3122 * For a data-integrity writeout, we need to wait upon any in-progress I/O
3123 * and then start new I/O and then wait upon it. The caller must have a ref on
3126 int __sync_dirty_buffer(struct buffer_head
*bh
, int op_flags
)
3130 WARN_ON(atomic_read(&bh
->b_count
) < 1);
3132 if (test_clear_buffer_dirty(bh
)) {
3134 bh
->b_end_io
= end_buffer_write_sync
;
3135 ret
= submit_bh(REQ_OP_WRITE
, op_flags
, bh
);
3137 if (!ret
&& !buffer_uptodate(bh
))
3144 EXPORT_SYMBOL(__sync_dirty_buffer
);
3146 int sync_dirty_buffer(struct buffer_head
*bh
)
3148 return __sync_dirty_buffer(bh
, WRITE_SYNC
);
3150 EXPORT_SYMBOL(sync_dirty_buffer
);
3153 * try_to_free_buffers() checks if all the buffers on this particular page
3154 * are unused, and releases them if so.
3156 * Exclusion against try_to_free_buffers may be obtained by either
3157 * locking the page or by holding its mapping's private_lock.
3159 * If the page is dirty but all the buffers are clean then we need to
3160 * be sure to mark the page clean as well. This is because the page
3161 * may be against a block device, and a later reattachment of buffers
3162 * to a dirty page will set *all* buffers dirty. Which would corrupt
3163 * filesystem data on the same device.
3165 * The same applies to regular filesystem pages: if all the buffers are
3166 * clean then we set the page clean and proceed. To do that, we require
3167 * total exclusion from __set_page_dirty_buffers(). That is obtained with
3170 * try_to_free_buffers() is non-blocking.
3172 static inline int buffer_busy(struct buffer_head
*bh
)
3174 return atomic_read(&bh
->b_count
) |
3175 (bh
->b_state
& ((1 << BH_Dirty
) | (1 << BH_Lock
)));
3179 drop_buffers(struct page
*page
, struct buffer_head
**buffers_to_free
)
3181 struct buffer_head
*head
= page_buffers(page
);
3182 struct buffer_head
*bh
;
3186 if (buffer_write_io_error(bh
) && page
->mapping
)
3187 set_bit(AS_EIO
, &page
->mapping
->flags
);
3188 if (buffer_busy(bh
))
3190 bh
= bh
->b_this_page
;
3191 } while (bh
!= head
);
3194 struct buffer_head
*next
= bh
->b_this_page
;
3196 if (bh
->b_assoc_map
)
3197 __remove_assoc_queue(bh
);
3199 } while (bh
!= head
);
3200 *buffers_to_free
= head
;
3201 __clear_page_buffers(page
);
3207 int try_to_free_buffers(struct page
*page
)
3209 struct address_space
* const mapping
= page
->mapping
;
3210 struct buffer_head
*buffers_to_free
= NULL
;
3213 BUG_ON(!PageLocked(page
));
3214 if (PageWriteback(page
))
3217 if (mapping
== NULL
) { /* can this still happen? */
3218 ret
= drop_buffers(page
, &buffers_to_free
);
3222 spin_lock(&mapping
->private_lock
);
3223 ret
= drop_buffers(page
, &buffers_to_free
);
3226 * If the filesystem writes its buffers by hand (eg ext3)
3227 * then we can have clean buffers against a dirty page. We
3228 * clean the page here; otherwise the VM will never notice
3229 * that the filesystem did any IO at all.
3231 * Also, during truncate, discard_buffer will have marked all
3232 * the page's buffers clean. We discover that here and clean
3235 * private_lock must be held over this entire operation in order
3236 * to synchronise against __set_page_dirty_buffers and prevent the
3237 * dirty bit from being lost.
3240 cancel_dirty_page(page
);
3241 spin_unlock(&mapping
->private_lock
);
3243 if (buffers_to_free
) {
3244 struct buffer_head
*bh
= buffers_to_free
;
3247 struct buffer_head
*next
= bh
->b_this_page
;
3248 free_buffer_head(bh
);
3250 } while (bh
!= buffers_to_free
);
3254 EXPORT_SYMBOL(try_to_free_buffers
);
3257 * There are no bdflush tunables left. But distributions are
3258 * still running obsolete flush daemons, so we terminate them here.
3260 * Use of bdflush() is deprecated and will be removed in a future kernel.
3261 * The `flush-X' kernel threads fully replace bdflush daemons and this call.
3263 SYSCALL_DEFINE2(bdflush
, int, func
, long, data
)
3265 static int msg_count
;
3267 if (!capable(CAP_SYS_ADMIN
))
3270 if (msg_count
< 5) {
3273 "warning: process `%s' used the obsolete bdflush"
3274 " system call\n", current
->comm
);
3275 printk(KERN_INFO
"Fix your initscripts?\n");
3284 * Buffer-head allocation
3286 static struct kmem_cache
*bh_cachep __read_mostly
;
3289 * Once the number of bh's in the machine exceeds this level, we start
3290 * stripping them in writeback.
3292 static unsigned long max_buffer_heads
;
3294 int buffer_heads_over_limit
;
3296 struct bh_accounting
{
3297 int nr
; /* Number of live bh's */
3298 int ratelimit
; /* Limit cacheline bouncing */
3301 static DEFINE_PER_CPU(struct bh_accounting
, bh_accounting
) = {0, 0};
3303 static void recalc_bh_state(void)
3308 if (__this_cpu_inc_return(bh_accounting
.ratelimit
) - 1 < 4096)
3310 __this_cpu_write(bh_accounting
.ratelimit
, 0);
3311 for_each_online_cpu(i
)
3312 tot
+= per_cpu(bh_accounting
, i
).nr
;
3313 buffer_heads_over_limit
= (tot
> max_buffer_heads
);
3316 struct buffer_head
*alloc_buffer_head(gfp_t gfp_flags
)
3318 struct buffer_head
*ret
= kmem_cache_zalloc(bh_cachep
, gfp_flags
);
3320 INIT_LIST_HEAD(&ret
->b_assoc_buffers
);
3322 __this_cpu_inc(bh_accounting
.nr
);
3328 EXPORT_SYMBOL(alloc_buffer_head
);
3330 void free_buffer_head(struct buffer_head
*bh
)
3332 BUG_ON(!list_empty(&bh
->b_assoc_buffers
));
3333 kmem_cache_free(bh_cachep
, bh
);
3335 __this_cpu_dec(bh_accounting
.nr
);
3339 EXPORT_SYMBOL(free_buffer_head
);
3341 static void buffer_exit_cpu(int cpu
)
3344 struct bh_lru
*b
= &per_cpu(bh_lrus
, cpu
);
3346 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
3350 this_cpu_add(bh_accounting
.nr
, per_cpu(bh_accounting
, cpu
).nr
);
3351 per_cpu(bh_accounting
, cpu
).nr
= 0;
3354 static int buffer_cpu_notify(struct notifier_block
*self
,
3355 unsigned long action
, void *hcpu
)
3357 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
)
3358 buffer_exit_cpu((unsigned long)hcpu
);
3363 * bh_uptodate_or_lock - Test whether the buffer is uptodate
3364 * @bh: struct buffer_head
3366 * Return true if the buffer is up-to-date and false,
3367 * with the buffer locked, if not.
3369 int bh_uptodate_or_lock(struct buffer_head
*bh
)
3371 if (!buffer_uptodate(bh
)) {
3373 if (!buffer_uptodate(bh
))
3379 EXPORT_SYMBOL(bh_uptodate_or_lock
);
3382 * bh_submit_read - Submit a locked buffer for reading
3383 * @bh: struct buffer_head
3385 * Returns zero on success and -EIO on error.
3387 int bh_submit_read(struct buffer_head
*bh
)
3389 BUG_ON(!buffer_locked(bh
));
3391 if (buffer_uptodate(bh
)) {
3397 bh
->b_end_io
= end_buffer_read_sync
;
3398 submit_bh(REQ_OP_READ
, 0, bh
);
3400 if (buffer_uptodate(bh
))
3404 EXPORT_SYMBOL(bh_submit_read
);
3406 void __init
buffer_init(void)
3408 unsigned long nrpages
;
3410 bh_cachep
= kmem_cache_create("buffer_head",
3411 sizeof(struct buffer_head
), 0,
3412 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
3417 * Limit the bh occupancy to 10% of ZONE_NORMAL
3419 nrpages
= (nr_free_buffer_pages() * 10) / 100;
3420 max_buffer_heads
= nrpages
* (PAGE_SIZE
/ sizeof(struct buffer_head
));
3421 hotcpu_notifier(buffer_cpu_notify
, 0);