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 rw
, 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);
132 page_cache_release(page
);
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_CACHE_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
);
248 page_cache_release(page
);
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
) {
265 (void)first_zones_zonelist(node_zonelist(nid
, GFP_NOFS
),
266 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(WRITE
, 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 mem_cgroup_begin_page_stat() lock.
626 static void __set_page_dirty(struct page
*page
, struct address_space
*mapping
,
627 struct mem_cgroup
*memcg
, int warn
)
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
, memcg
);
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 mem_cgroup
*memcg
;
670 struct address_space
*mapping
= page_mapping(page
);
672 if (unlikely(!mapping
))
673 return !TestSetPageDirty(page
);
675 spin_lock(&mapping
->private_lock
);
676 if (page_has_buffers(page
)) {
677 struct buffer_head
*head
= page_buffers(page
);
678 struct buffer_head
*bh
= head
;
681 set_buffer_dirty(bh
);
682 bh
= bh
->b_this_page
;
683 } while (bh
!= head
);
686 * Use mem_group_begin_page_stat() to keep PageDirty synchronized with
687 * per-memcg dirty page counters.
689 memcg
= mem_cgroup_begin_page_stat(page
);
690 newly_dirty
= !TestSetPageDirty(page
);
691 spin_unlock(&mapping
->private_lock
);
694 __set_page_dirty(page
, mapping
, memcg
, 1);
696 mem_cgroup_end_page_stat(memcg
);
699 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
703 EXPORT_SYMBOL(__set_page_dirty_buffers
);
706 * Write out and wait upon a list of buffers.
708 * We have conflicting pressures: we want to make sure that all
709 * initially dirty buffers get waited on, but that any subsequently
710 * dirtied buffers don't. After all, we don't want fsync to last
711 * forever if somebody is actively writing to the file.
713 * Do this in two main stages: first we copy dirty buffers to a
714 * temporary inode list, queueing the writes as we go. Then we clean
715 * up, waiting for those writes to complete.
717 * During this second stage, any subsequent updates to the file may end
718 * up refiling the buffer on the original inode's dirty list again, so
719 * there is a chance we will end up with a buffer queued for write but
720 * not yet completed on that list. So, as a final cleanup we go through
721 * the osync code to catch these locked, dirty buffers without requeuing
722 * any newly dirty buffers for write.
724 static int fsync_buffers_list(spinlock_t
*lock
, struct list_head
*list
)
726 struct buffer_head
*bh
;
727 struct list_head tmp
;
728 struct address_space
*mapping
;
730 struct blk_plug plug
;
732 INIT_LIST_HEAD(&tmp
);
733 blk_start_plug(&plug
);
736 while (!list_empty(list
)) {
737 bh
= BH_ENTRY(list
->next
);
738 mapping
= bh
->b_assoc_map
;
739 __remove_assoc_queue(bh
);
740 /* Avoid race with mark_buffer_dirty_inode() which does
741 * a lockless check and we rely on seeing the dirty bit */
743 if (buffer_dirty(bh
) || buffer_locked(bh
)) {
744 list_add(&bh
->b_assoc_buffers
, &tmp
);
745 bh
->b_assoc_map
= mapping
;
746 if (buffer_dirty(bh
)) {
750 * Ensure any pending I/O completes so that
751 * write_dirty_buffer() actually writes the
752 * current contents - it is a noop if I/O is
753 * still in flight on potentially older
756 write_dirty_buffer(bh
, WRITE_SYNC
);
759 * Kick off IO for the previous mapping. Note
760 * that we will not run the very last mapping,
761 * wait_on_buffer() will do that for us
762 * through sync_buffer().
771 blk_finish_plug(&plug
);
774 while (!list_empty(&tmp
)) {
775 bh
= BH_ENTRY(tmp
.prev
);
777 mapping
= bh
->b_assoc_map
;
778 __remove_assoc_queue(bh
);
779 /* Avoid race with mark_buffer_dirty_inode() which does
780 * a lockless check and we rely on seeing the dirty bit */
782 if (buffer_dirty(bh
)) {
783 list_add(&bh
->b_assoc_buffers
,
784 &mapping
->private_list
);
785 bh
->b_assoc_map
= mapping
;
789 if (!buffer_uptodate(bh
))
796 err2
= osync_buffers_list(lock
, list
);
804 * Invalidate any and all dirty buffers on a given inode. We are
805 * probably unmounting the fs, but that doesn't mean we have already
806 * done a sync(). Just drop the buffers from the inode list.
808 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
809 * assumes that all the buffers are against the blockdev. Not true
812 void invalidate_inode_buffers(struct inode
*inode
)
814 if (inode_has_buffers(inode
)) {
815 struct address_space
*mapping
= &inode
->i_data
;
816 struct list_head
*list
= &mapping
->private_list
;
817 struct address_space
*buffer_mapping
= mapping
->private_data
;
819 spin_lock(&buffer_mapping
->private_lock
);
820 while (!list_empty(list
))
821 __remove_assoc_queue(BH_ENTRY(list
->next
));
822 spin_unlock(&buffer_mapping
->private_lock
);
825 EXPORT_SYMBOL(invalidate_inode_buffers
);
828 * Remove any clean buffers from the inode's buffer list. This is called
829 * when we're trying to free the inode itself. Those buffers can pin it.
831 * Returns true if all buffers were removed.
833 int remove_inode_buffers(struct inode
*inode
)
837 if (inode_has_buffers(inode
)) {
838 struct address_space
*mapping
= &inode
->i_data
;
839 struct list_head
*list
= &mapping
->private_list
;
840 struct address_space
*buffer_mapping
= mapping
->private_data
;
842 spin_lock(&buffer_mapping
->private_lock
);
843 while (!list_empty(list
)) {
844 struct buffer_head
*bh
= BH_ENTRY(list
->next
);
845 if (buffer_dirty(bh
)) {
849 __remove_assoc_queue(bh
);
851 spin_unlock(&buffer_mapping
->private_lock
);
857 * Create the appropriate buffers when given a page for data area and
858 * the size of each buffer.. Use the bh->b_this_page linked list to
859 * follow the buffers created. Return NULL if unable to create more
862 * The retry flag is used to differentiate async IO (paging, swapping)
863 * which may not fail from ordinary buffer allocations.
865 struct buffer_head
*alloc_page_buffers(struct page
*page
, unsigned long size
,
868 struct buffer_head
*bh
, *head
;
874 while ((offset
-= size
) >= 0) {
875 bh
= alloc_buffer_head(GFP_NOFS
);
879 bh
->b_this_page
= head
;
885 /* Link the buffer to its page */
886 set_bh_page(bh
, page
, offset
);
890 * In case anything failed, we just free everything we got.
896 head
= head
->b_this_page
;
897 free_buffer_head(bh
);
902 * Return failure for non-async IO requests. Async IO requests
903 * are not allowed to fail, so we have to wait until buffer heads
904 * become available. But we don't want tasks sleeping with
905 * partially complete buffers, so all were released above.
910 /* We're _really_ low on memory. Now we just
911 * wait for old buffer heads to become free due to
912 * finishing IO. Since this is an async request and
913 * the reserve list is empty, we're sure there are
914 * async buffer heads in use.
919 EXPORT_SYMBOL_GPL(alloc_page_buffers
);
922 link_dev_buffers(struct page
*page
, struct buffer_head
*head
)
924 struct buffer_head
*bh
, *tail
;
929 bh
= bh
->b_this_page
;
931 tail
->b_this_page
= head
;
932 attach_page_buffers(page
, head
);
935 static sector_t
blkdev_max_block(struct block_device
*bdev
, unsigned int size
)
937 sector_t retval
= ~((sector_t
)0);
938 loff_t sz
= i_size_read(bdev
->bd_inode
);
941 unsigned int sizebits
= blksize_bits(size
);
942 retval
= (sz
>> sizebits
);
948 * Initialise the state of a blockdev page's buffers.
951 init_page_buffers(struct page
*page
, struct block_device
*bdev
,
952 sector_t block
, int size
)
954 struct buffer_head
*head
= page_buffers(page
);
955 struct buffer_head
*bh
= head
;
956 int uptodate
= PageUptodate(page
);
957 sector_t end_block
= blkdev_max_block(I_BDEV(bdev
->bd_inode
), size
);
960 if (!buffer_mapped(bh
)) {
961 init_buffer(bh
, NULL
, NULL
);
963 bh
->b_blocknr
= block
;
965 set_buffer_uptodate(bh
);
966 if (block
< end_block
)
967 set_buffer_mapped(bh
);
970 bh
= bh
->b_this_page
;
971 } while (bh
!= head
);
974 * Caller needs to validate requested block against end of device.
980 * Create the page-cache page that contains the requested block.
982 * This is used purely for blockdev mappings.
985 grow_dev_page(struct block_device
*bdev
, sector_t block
,
986 pgoff_t index
, int size
, int sizebits
, gfp_t gfp
)
988 struct inode
*inode
= bdev
->bd_inode
;
990 struct buffer_head
*bh
;
992 int ret
= 0; /* Will call free_more_memory() */
995 gfp_mask
= mapping_gfp_constraint(inode
->i_mapping
, ~__GFP_FS
) | gfp
;
998 * XXX: __getblk_slow() can not really deal with failure and
999 * will endlessly loop on improvised global reclaim. Prefer
1000 * looping in the allocator rather than here, at least that
1001 * code knows what it's doing.
1003 gfp_mask
|= __GFP_NOFAIL
;
1005 page
= find_or_create_page(inode
->i_mapping
, index
, gfp_mask
);
1009 BUG_ON(!PageLocked(page
));
1011 if (page_has_buffers(page
)) {
1012 bh
= page_buffers(page
);
1013 if (bh
->b_size
== size
) {
1014 end_block
= init_page_buffers(page
, bdev
,
1015 (sector_t
)index
<< sizebits
,
1019 if (!try_to_free_buffers(page
))
1024 * Allocate some buffers for this page
1026 bh
= alloc_page_buffers(page
, size
, 0);
1031 * Link the page to the buffers and initialise them. Take the
1032 * lock to be atomic wrt __find_get_block(), which does not
1033 * run under the page lock.
1035 spin_lock(&inode
->i_mapping
->private_lock
);
1036 link_dev_buffers(page
, bh
);
1037 end_block
= init_page_buffers(page
, bdev
, (sector_t
)index
<< sizebits
,
1039 spin_unlock(&inode
->i_mapping
->private_lock
);
1041 ret
= (block
< end_block
) ? 1 : -ENXIO
;
1044 page_cache_release(page
);
1049 * Create buffers for the specified block device block's page. If
1050 * that page was dirty, the buffers are set dirty also.
1053 grow_buffers(struct block_device
*bdev
, sector_t block
, int size
, gfp_t gfp
)
1061 } while ((size
<< sizebits
) < PAGE_SIZE
);
1063 index
= block
>> sizebits
;
1066 * Check for a block which wants to lie outside our maximum possible
1067 * pagecache index. (this comparison is done using sector_t types).
1069 if (unlikely(index
!= block
>> sizebits
)) {
1070 printk(KERN_ERR
"%s: requested out-of-range block %llu for "
1072 __func__
, (unsigned long long)block
,
1077 /* Create a page with the proper size buffers.. */
1078 return grow_dev_page(bdev
, block
, index
, size
, sizebits
, gfp
);
1081 struct buffer_head
*
1082 __getblk_slow(struct block_device
*bdev
, sector_t block
,
1083 unsigned size
, gfp_t gfp
)
1085 /* Size must be multiple of hard sectorsize */
1086 if (unlikely(size
& (bdev_logical_block_size(bdev
)-1) ||
1087 (size
< 512 || size
> PAGE_SIZE
))) {
1088 printk(KERN_ERR
"getblk(): invalid block size %d requested\n",
1090 printk(KERN_ERR
"logical block size: %d\n",
1091 bdev_logical_block_size(bdev
));
1098 struct buffer_head
*bh
;
1101 bh
= __find_get_block(bdev
, block
, size
);
1105 ret
= grow_buffers(bdev
, block
, size
, gfp
);
1112 EXPORT_SYMBOL(__getblk_slow
);
1115 * The relationship between dirty buffers and dirty pages:
1117 * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1118 * the page is tagged dirty in its radix tree.
1120 * At all times, the dirtiness of the buffers represents the dirtiness of
1121 * subsections of the page. If the page has buffers, the page dirty bit is
1122 * merely a hint about the true dirty state.
1124 * When a page is set dirty in its entirety, all its buffers are marked dirty
1125 * (if the page has buffers).
1127 * When a buffer is marked dirty, its page is dirtied, but the page's other
1130 * Also. When blockdev buffers are explicitly read with bread(), they
1131 * individually become uptodate. But their backing page remains not
1132 * uptodate - even if all of its buffers are uptodate. A subsequent
1133 * block_read_full_page() against that page will discover all the uptodate
1134 * buffers, will set the page uptodate and will perform no I/O.
1138 * mark_buffer_dirty - mark a buffer_head as needing writeout
1139 * @bh: the buffer_head to mark dirty
1141 * mark_buffer_dirty() will set the dirty bit against the buffer, then set its
1142 * backing page dirty, then tag the page as dirty in its address_space's radix
1143 * tree and then attach the address_space's inode to its superblock's dirty
1146 * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
1147 * mapping->tree_lock and mapping->host->i_lock.
1149 void mark_buffer_dirty(struct buffer_head
*bh
)
1151 WARN_ON_ONCE(!buffer_uptodate(bh
));
1153 trace_block_dirty_buffer(bh
);
1156 * Very *carefully* optimize the it-is-already-dirty case.
1158 * Don't let the final "is it dirty" escape to before we
1159 * perhaps modified the buffer.
1161 if (buffer_dirty(bh
)) {
1163 if (buffer_dirty(bh
))
1167 if (!test_set_buffer_dirty(bh
)) {
1168 struct page
*page
= bh
->b_page
;
1169 struct address_space
*mapping
= NULL
;
1170 struct mem_cgroup
*memcg
;
1172 memcg
= mem_cgroup_begin_page_stat(page
);
1173 if (!TestSetPageDirty(page
)) {
1174 mapping
= page_mapping(page
);
1176 __set_page_dirty(page
, mapping
, memcg
, 0);
1178 mem_cgroup_end_page_stat(memcg
);
1180 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
1183 EXPORT_SYMBOL(mark_buffer_dirty
);
1186 * Decrement a buffer_head's reference count. If all buffers against a page
1187 * have zero reference count, are clean and unlocked, and if the page is clean
1188 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1189 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1190 * a page but it ends up not being freed, and buffers may later be reattached).
1192 void __brelse(struct buffer_head
* buf
)
1194 if (atomic_read(&buf
->b_count
)) {
1198 WARN(1, KERN_ERR
"VFS: brelse: Trying to free free buffer\n");
1200 EXPORT_SYMBOL(__brelse
);
1203 * bforget() is like brelse(), except it discards any
1204 * potentially dirty data.
1206 void __bforget(struct buffer_head
*bh
)
1208 clear_buffer_dirty(bh
);
1209 if (bh
->b_assoc_map
) {
1210 struct address_space
*buffer_mapping
= bh
->b_page
->mapping
;
1212 spin_lock(&buffer_mapping
->private_lock
);
1213 list_del_init(&bh
->b_assoc_buffers
);
1214 bh
->b_assoc_map
= NULL
;
1215 spin_unlock(&buffer_mapping
->private_lock
);
1219 EXPORT_SYMBOL(__bforget
);
1221 static struct buffer_head
*__bread_slow(struct buffer_head
*bh
)
1224 if (buffer_uptodate(bh
)) {
1229 bh
->b_end_io
= end_buffer_read_sync
;
1230 submit_bh(READ
, bh
);
1232 if (buffer_uptodate(bh
))
1240 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1241 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1242 * refcount elevated by one when they're in an LRU. A buffer can only appear
1243 * once in a particular CPU's LRU. A single buffer can be present in multiple
1244 * CPU's LRUs at the same time.
1246 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1247 * sb_find_get_block().
1249 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1250 * a local interrupt disable for that.
1253 #define BH_LRU_SIZE 16
1256 struct buffer_head
*bhs
[BH_LRU_SIZE
];
1259 static DEFINE_PER_CPU(struct bh_lru
, bh_lrus
) = {{ NULL
}};
1262 #define bh_lru_lock() local_irq_disable()
1263 #define bh_lru_unlock() local_irq_enable()
1265 #define bh_lru_lock() preempt_disable()
1266 #define bh_lru_unlock() preempt_enable()
1269 static inline void check_irqs_on(void)
1271 #ifdef irqs_disabled
1272 BUG_ON(irqs_disabled());
1277 * The LRU management algorithm is dopey-but-simple. Sorry.
1279 static void bh_lru_install(struct buffer_head
*bh
)
1281 struct buffer_head
*evictee
= NULL
;
1285 if (__this_cpu_read(bh_lrus
.bhs
[0]) != bh
) {
1286 struct buffer_head
*bhs
[BH_LRU_SIZE
];
1292 for (in
= 0; in
< BH_LRU_SIZE
; in
++) {
1293 struct buffer_head
*bh2
=
1294 __this_cpu_read(bh_lrus
.bhs
[in
]);
1299 if (out
>= BH_LRU_SIZE
) {
1300 BUG_ON(evictee
!= NULL
);
1307 while (out
< BH_LRU_SIZE
)
1309 memcpy(this_cpu_ptr(&bh_lrus
.bhs
), bhs
, sizeof(bhs
));
1318 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1320 static struct buffer_head
*
1321 lookup_bh_lru(struct block_device
*bdev
, sector_t block
, unsigned size
)
1323 struct buffer_head
*ret
= NULL
;
1328 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1329 struct buffer_head
*bh
= __this_cpu_read(bh_lrus
.bhs
[i
]);
1331 if (bh
&& bh
->b_blocknr
== block
&& bh
->b_bdev
== bdev
&&
1332 bh
->b_size
== size
) {
1335 __this_cpu_write(bh_lrus
.bhs
[i
],
1336 __this_cpu_read(bh_lrus
.bhs
[i
- 1]));
1339 __this_cpu_write(bh_lrus
.bhs
[0], bh
);
1351 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1352 * it in the LRU and mark it as accessed. If it is not present then return
1355 struct buffer_head
*
1356 __find_get_block(struct block_device
*bdev
, sector_t block
, unsigned size
)
1358 struct buffer_head
*bh
= lookup_bh_lru(bdev
, block
, size
);
1361 /* __find_get_block_slow will mark the page accessed */
1362 bh
= __find_get_block_slow(bdev
, block
);
1370 EXPORT_SYMBOL(__find_get_block
);
1373 * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
1374 * which corresponds to the passed block_device, block and size. The
1375 * returned buffer has its reference count incremented.
1377 * __getblk_gfp() will lock up the machine if grow_dev_page's
1378 * try_to_free_buffers() attempt is failing. FIXME, perhaps?
1380 struct buffer_head
*
1381 __getblk_gfp(struct block_device
*bdev
, sector_t block
,
1382 unsigned size
, gfp_t gfp
)
1384 struct buffer_head
*bh
= __find_get_block(bdev
, block
, size
);
1388 bh
= __getblk_slow(bdev
, block
, size
, gfp
);
1391 EXPORT_SYMBOL(__getblk_gfp
);
1394 * Do async read-ahead on a buffer..
1396 void __breadahead(struct block_device
*bdev
, sector_t block
, unsigned size
)
1398 struct buffer_head
*bh
= __getblk(bdev
, block
, size
);
1400 ll_rw_block(READA
, 1, &bh
);
1404 EXPORT_SYMBOL(__breadahead
);
1407 * __bread_gfp() - reads a specified block and returns the bh
1408 * @bdev: the block_device to read from
1409 * @block: number of block
1410 * @size: size (in bytes) to read
1411 * @gfp: page allocation flag
1413 * Reads a specified block, and returns buffer head that contains it.
1414 * The page cache can be allocated from non-movable area
1415 * not to prevent page migration if you set gfp to zero.
1416 * It returns NULL if the block was unreadable.
1418 struct buffer_head
*
1419 __bread_gfp(struct block_device
*bdev
, sector_t block
,
1420 unsigned size
, gfp_t gfp
)
1422 struct buffer_head
*bh
= __getblk_gfp(bdev
, block
, size
, gfp
);
1424 if (likely(bh
) && !buffer_uptodate(bh
))
1425 bh
= __bread_slow(bh
);
1428 EXPORT_SYMBOL(__bread_gfp
);
1431 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1432 * This doesn't race because it runs in each cpu either in irq
1433 * or with preempt disabled.
1435 static void invalidate_bh_lru(void *arg
)
1437 struct bh_lru
*b
= &get_cpu_var(bh_lrus
);
1440 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1444 put_cpu_var(bh_lrus
);
1447 static bool has_bh_in_lru(int cpu
, void *dummy
)
1449 struct bh_lru
*b
= per_cpu_ptr(&bh_lrus
, cpu
);
1452 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1460 void invalidate_bh_lrus(void)
1462 on_each_cpu_cond(has_bh_in_lru
, invalidate_bh_lru
, NULL
, 1, GFP_KERNEL
);
1464 EXPORT_SYMBOL_GPL(invalidate_bh_lrus
);
1466 void set_bh_page(struct buffer_head
*bh
,
1467 struct page
*page
, unsigned long offset
)
1470 BUG_ON(offset
>= PAGE_SIZE
);
1471 if (PageHighMem(page
))
1473 * This catches illegal uses and preserves the offset:
1475 bh
->b_data
= (char *)(0 + offset
);
1477 bh
->b_data
= page_address(page
) + offset
;
1479 EXPORT_SYMBOL(set_bh_page
);
1482 * Called when truncating a buffer on a page completely.
1485 /* Bits that are cleared during an invalidate */
1486 #define BUFFER_FLAGS_DISCARD \
1487 (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1488 1 << BH_Delay | 1 << BH_Unwritten)
1490 static void discard_buffer(struct buffer_head
* bh
)
1492 unsigned long b_state
, b_state_old
;
1495 clear_buffer_dirty(bh
);
1497 b_state
= bh
->b_state
;
1499 b_state_old
= cmpxchg(&bh
->b_state
, b_state
,
1500 (b_state
& ~BUFFER_FLAGS_DISCARD
));
1501 if (b_state_old
== b_state
)
1503 b_state
= b_state_old
;
1509 * block_invalidatepage - invalidate part or all of a buffer-backed page
1511 * @page: the page which is affected
1512 * @offset: start of the range to invalidate
1513 * @length: length of the range to invalidate
1515 * block_invalidatepage() is called when all or part of the page has become
1516 * invalidated by a truncate operation.
1518 * block_invalidatepage() does not have to release all buffers, but it must
1519 * ensure that no dirty buffer is left outside @offset and that no I/O
1520 * is underway against any of the blocks which are outside the truncation
1521 * point. Because the caller is about to free (and possibly reuse) those
1524 void block_invalidatepage(struct page
*page
, unsigned int offset
,
1525 unsigned int length
)
1527 struct buffer_head
*head
, *bh
, *next
;
1528 unsigned int curr_off
= 0;
1529 unsigned int stop
= length
+ offset
;
1531 BUG_ON(!PageLocked(page
));
1532 if (!page_has_buffers(page
))
1536 * Check for overflow
1538 BUG_ON(stop
> PAGE_CACHE_SIZE
|| stop
< length
);
1540 head
= page_buffers(page
);
1543 unsigned int next_off
= curr_off
+ bh
->b_size
;
1544 next
= bh
->b_this_page
;
1547 * Are we still fully in range ?
1549 if (next_off
> stop
)
1553 * is this block fully invalidated?
1555 if (offset
<= curr_off
)
1557 curr_off
= next_off
;
1559 } while (bh
!= head
);
1562 * We release buffers only if the entire page is being invalidated.
1563 * The get_block cached value has been unconditionally invalidated,
1564 * so real IO is not possible anymore.
1567 try_to_release_page(page
, 0);
1571 EXPORT_SYMBOL(block_invalidatepage
);
1575 * We attach and possibly dirty the buffers atomically wrt
1576 * __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1577 * is already excluded via the page lock.
1579 void create_empty_buffers(struct page
*page
,
1580 unsigned long blocksize
, unsigned long b_state
)
1582 struct buffer_head
*bh
, *head
, *tail
;
1584 head
= alloc_page_buffers(page
, blocksize
, 1);
1587 bh
->b_state
|= b_state
;
1589 bh
= bh
->b_this_page
;
1591 tail
->b_this_page
= head
;
1593 spin_lock(&page
->mapping
->private_lock
);
1594 if (PageUptodate(page
) || PageDirty(page
)) {
1597 if (PageDirty(page
))
1598 set_buffer_dirty(bh
);
1599 if (PageUptodate(page
))
1600 set_buffer_uptodate(bh
);
1601 bh
= bh
->b_this_page
;
1602 } while (bh
!= head
);
1604 attach_page_buffers(page
, head
);
1605 spin_unlock(&page
->mapping
->private_lock
);
1607 EXPORT_SYMBOL(create_empty_buffers
);
1610 * We are taking a block for data and we don't want any output from any
1611 * buffer-cache aliases starting from return from that function and
1612 * until the moment when something will explicitly mark the buffer
1613 * dirty (hopefully that will not happen until we will free that block ;-)
1614 * We don't even need to mark it not-uptodate - nobody can expect
1615 * anything from a newly allocated buffer anyway. We used to used
1616 * unmap_buffer() for such invalidation, but that was wrong. We definitely
1617 * don't want to mark the alias unmapped, for example - it would confuse
1618 * anyone who might pick it with bread() afterwards...
1620 * Also.. Note that bforget() doesn't lock the buffer. So there can
1621 * be writeout I/O going on against recently-freed buffers. We don't
1622 * wait on that I/O in bforget() - it's more efficient to wait on the I/O
1623 * only if we really need to. That happens here.
1625 void unmap_underlying_metadata(struct block_device
*bdev
, sector_t block
)
1627 struct buffer_head
*old_bh
;
1631 old_bh
= __find_get_block_slow(bdev
, block
);
1633 clear_buffer_dirty(old_bh
);
1634 wait_on_buffer(old_bh
);
1635 clear_buffer_req(old_bh
);
1639 EXPORT_SYMBOL(unmap_underlying_metadata
);
1642 * Size is a power-of-two in the range 512..PAGE_SIZE,
1643 * and the case we care about most is PAGE_SIZE.
1645 * So this *could* possibly be written with those
1646 * constraints in mind (relevant mostly if some
1647 * architecture has a slow bit-scan instruction)
1649 static inline int block_size_bits(unsigned int blocksize
)
1651 return ilog2(blocksize
);
1654 static struct buffer_head
*create_page_buffers(struct page
*page
, struct inode
*inode
, unsigned int b_state
)
1656 BUG_ON(!PageLocked(page
));
1658 if (!page_has_buffers(page
))
1659 create_empty_buffers(page
, 1 << ACCESS_ONCE(inode
->i_blkbits
), b_state
);
1660 return page_buffers(page
);
1664 * NOTE! All mapped/uptodate combinations are valid:
1666 * Mapped Uptodate Meaning
1668 * No No "unknown" - must do get_block()
1669 * No Yes "hole" - zero-filled
1670 * Yes No "allocated" - allocated on disk, not read in
1671 * Yes Yes "valid" - allocated and up-to-date in memory.
1673 * "Dirty" is valid only with the last case (mapped+uptodate).
1677 * While block_write_full_page is writing back the dirty buffers under
1678 * the page lock, whoever dirtied the buffers may decide to clean them
1679 * again at any time. We handle that by only looking at the buffer
1680 * state inside lock_buffer().
1682 * If block_write_full_page() is called for regular writeback
1683 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1684 * locked buffer. This only can happen if someone has written the buffer
1685 * directly, with submit_bh(). At the address_space level PageWriteback
1686 * prevents this contention from occurring.
1688 * If block_write_full_page() is called with wbc->sync_mode ==
1689 * WB_SYNC_ALL, the writes are posted using WRITE_SYNC; this
1690 * causes the writes to be flagged as synchronous writes.
1692 static int __block_write_full_page(struct inode
*inode
, struct page
*page
,
1693 get_block_t
*get_block
, struct writeback_control
*wbc
,
1694 bh_end_io_t
*handler
)
1698 sector_t last_block
;
1699 struct buffer_head
*bh
, *head
;
1700 unsigned int blocksize
, bbits
;
1701 int nr_underway
= 0;
1702 int write_op
= (wbc
->sync_mode
== WB_SYNC_ALL
? WRITE_SYNC
: WRITE
);
1704 head
= create_page_buffers(page
, inode
,
1705 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
1708 * Be very careful. We have no exclusion from __set_page_dirty_buffers
1709 * here, and the (potentially unmapped) buffers may become dirty at
1710 * any time. If a buffer becomes dirty here after we've inspected it
1711 * then we just miss that fact, and the page stays dirty.
1713 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1714 * handle that here by just cleaning them.
1718 blocksize
= bh
->b_size
;
1719 bbits
= block_size_bits(blocksize
);
1721 block
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- bbits
);
1722 last_block
= (i_size_read(inode
) - 1) >> bbits
;
1725 * Get all the dirty buffers mapped to disk addresses and
1726 * handle any aliases from the underlying blockdev's mapping.
1729 if (block
> last_block
) {
1731 * mapped buffers outside i_size will occur, because
1732 * this page can be outside i_size when there is a
1733 * truncate in progress.
1736 * The buffer was zeroed by block_write_full_page()
1738 clear_buffer_dirty(bh
);
1739 set_buffer_uptodate(bh
);
1740 } else if ((!buffer_mapped(bh
) || buffer_delay(bh
)) &&
1742 WARN_ON(bh
->b_size
!= blocksize
);
1743 err
= get_block(inode
, block
, bh
, 1);
1746 clear_buffer_delay(bh
);
1747 if (buffer_new(bh
)) {
1748 /* blockdev mappings never come here */
1749 clear_buffer_new(bh
);
1750 unmap_underlying_metadata(bh
->b_bdev
,
1754 bh
= bh
->b_this_page
;
1756 } while (bh
!= head
);
1759 if (!buffer_mapped(bh
))
1762 * If it's a fully non-blocking write attempt and we cannot
1763 * lock the buffer then redirty the page. Note that this can
1764 * potentially cause a busy-wait loop from writeback threads
1765 * and kswapd activity, but those code paths have their own
1766 * higher-level throttling.
1768 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
1770 } else if (!trylock_buffer(bh
)) {
1771 redirty_page_for_writepage(wbc
, page
);
1774 if (test_clear_buffer_dirty(bh
)) {
1775 mark_buffer_async_write_endio(bh
, handler
);
1779 } while ((bh
= bh
->b_this_page
) != head
);
1782 * The page and its buffers are protected by PageWriteback(), so we can
1783 * drop the bh refcounts early.
1785 BUG_ON(PageWriteback(page
));
1786 set_page_writeback(page
);
1789 struct buffer_head
*next
= bh
->b_this_page
;
1790 if (buffer_async_write(bh
)) {
1791 submit_bh_wbc(write_op
, bh
, 0, wbc
);
1795 } while (bh
!= head
);
1800 if (nr_underway
== 0) {
1802 * The page was marked dirty, but the buffers were
1803 * clean. Someone wrote them back by hand with
1804 * ll_rw_block/submit_bh. A rare case.
1806 end_page_writeback(page
);
1809 * The page and buffer_heads can be released at any time from
1817 * ENOSPC, or some other error. We may already have added some
1818 * blocks to the file, so we need to write these out to avoid
1819 * exposing stale data.
1820 * The page is currently locked and not marked for writeback
1823 /* Recovery: lock and submit the mapped buffers */
1825 if (buffer_mapped(bh
) && buffer_dirty(bh
) &&
1826 !buffer_delay(bh
)) {
1828 mark_buffer_async_write_endio(bh
, handler
);
1831 * The buffer may have been set dirty during
1832 * attachment to a dirty page.
1834 clear_buffer_dirty(bh
);
1836 } while ((bh
= bh
->b_this_page
) != head
);
1838 BUG_ON(PageWriteback(page
));
1839 mapping_set_error(page
->mapping
, err
);
1840 set_page_writeback(page
);
1842 struct buffer_head
*next
= bh
->b_this_page
;
1843 if (buffer_async_write(bh
)) {
1844 clear_buffer_dirty(bh
);
1845 submit_bh_wbc(write_op
, bh
, 0, wbc
);
1849 } while (bh
!= head
);
1855 * If a page has any new buffers, zero them out here, and mark them uptodate
1856 * and dirty so they'll be written out (in order to prevent uninitialised
1857 * block data from leaking). And clear the new bit.
1859 void page_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1861 unsigned int block_start
, block_end
;
1862 struct buffer_head
*head
, *bh
;
1864 BUG_ON(!PageLocked(page
));
1865 if (!page_has_buffers(page
))
1868 bh
= head
= page_buffers(page
);
1871 block_end
= block_start
+ bh
->b_size
;
1873 if (buffer_new(bh
)) {
1874 if (block_end
> from
&& block_start
< to
) {
1875 if (!PageUptodate(page
)) {
1876 unsigned start
, size
;
1878 start
= max(from
, block_start
);
1879 size
= min(to
, block_end
) - start
;
1881 zero_user(page
, start
, size
);
1882 set_buffer_uptodate(bh
);
1885 clear_buffer_new(bh
);
1886 mark_buffer_dirty(bh
);
1890 block_start
= block_end
;
1891 bh
= bh
->b_this_page
;
1892 } while (bh
!= head
);
1894 EXPORT_SYMBOL(page_zero_new_buffers
);
1896 int __block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1897 get_block_t
*get_block
)
1899 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1900 unsigned to
= from
+ len
;
1901 struct inode
*inode
= page
->mapping
->host
;
1902 unsigned block_start
, block_end
;
1905 unsigned blocksize
, bbits
;
1906 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
=wait
;
1908 BUG_ON(!PageLocked(page
));
1909 BUG_ON(from
> PAGE_CACHE_SIZE
);
1910 BUG_ON(to
> PAGE_CACHE_SIZE
);
1913 head
= create_page_buffers(page
, inode
, 0);
1914 blocksize
= head
->b_size
;
1915 bbits
= block_size_bits(blocksize
);
1917 block
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- bbits
);
1919 for(bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1920 block
++, block_start
=block_end
, bh
= bh
->b_this_page
) {
1921 block_end
= block_start
+ blocksize
;
1922 if (block_end
<= from
|| block_start
>= to
) {
1923 if (PageUptodate(page
)) {
1924 if (!buffer_uptodate(bh
))
1925 set_buffer_uptodate(bh
);
1930 clear_buffer_new(bh
);
1931 if (!buffer_mapped(bh
)) {
1932 WARN_ON(bh
->b_size
!= blocksize
);
1933 err
= get_block(inode
, block
, bh
, 1);
1936 if (buffer_new(bh
)) {
1937 unmap_underlying_metadata(bh
->b_bdev
,
1939 if (PageUptodate(page
)) {
1940 clear_buffer_new(bh
);
1941 set_buffer_uptodate(bh
);
1942 mark_buffer_dirty(bh
);
1945 if (block_end
> to
|| block_start
< from
)
1946 zero_user_segments(page
,
1952 if (PageUptodate(page
)) {
1953 if (!buffer_uptodate(bh
))
1954 set_buffer_uptodate(bh
);
1957 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1958 !buffer_unwritten(bh
) &&
1959 (block_start
< from
|| block_end
> to
)) {
1960 ll_rw_block(READ
, 1, &bh
);
1965 * If we issued read requests - let them complete.
1967 while(wait_bh
> wait
) {
1968 wait_on_buffer(*--wait_bh
);
1969 if (!buffer_uptodate(*wait_bh
))
1973 page_zero_new_buffers(page
, from
, to
);
1976 EXPORT_SYMBOL(__block_write_begin
);
1978 static int __block_commit_write(struct inode
*inode
, struct page
*page
,
1979 unsigned from
, unsigned to
)
1981 unsigned block_start
, block_end
;
1984 struct buffer_head
*bh
, *head
;
1986 bh
= head
= page_buffers(page
);
1987 blocksize
= bh
->b_size
;
1991 block_end
= block_start
+ blocksize
;
1992 if (block_end
<= from
|| block_start
>= to
) {
1993 if (!buffer_uptodate(bh
))
1996 set_buffer_uptodate(bh
);
1997 mark_buffer_dirty(bh
);
1999 clear_buffer_new(bh
);
2001 block_start
= block_end
;
2002 bh
= bh
->b_this_page
;
2003 } while (bh
!= head
);
2006 * If this is a partial write which happened to make all buffers
2007 * uptodate then we can optimize away a bogus readpage() for
2008 * the next read(). Here we 'discover' whether the page went
2009 * uptodate as a result of this (potentially partial) write.
2012 SetPageUptodate(page
);
2017 * block_write_begin takes care of the basic task of block allocation and
2018 * bringing partial write blocks uptodate first.
2020 * The filesystem needs to handle block truncation upon failure.
2022 int block_write_begin(struct address_space
*mapping
, loff_t pos
, unsigned len
,
2023 unsigned flags
, struct page
**pagep
, get_block_t
*get_block
)
2025 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
2029 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2033 status
= __block_write_begin(page
, pos
, len
, get_block
);
2034 if (unlikely(status
)) {
2036 page_cache_release(page
);
2043 EXPORT_SYMBOL(block_write_begin
);
2045 int block_write_end(struct file
*file
, struct address_space
*mapping
,
2046 loff_t pos
, unsigned len
, unsigned copied
,
2047 struct page
*page
, void *fsdata
)
2049 struct inode
*inode
= mapping
->host
;
2052 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2054 if (unlikely(copied
< len
)) {
2056 * The buffers that were written will now be uptodate, so we
2057 * don't have to worry about a readpage reading them and
2058 * overwriting a partial write. However if we have encountered
2059 * a short write and only partially written into a buffer, it
2060 * will not be marked uptodate, so a readpage might come in and
2061 * destroy our partial write.
2063 * Do the simplest thing, and just treat any short write to a
2064 * non uptodate page as a zero-length write, and force the
2065 * caller to redo the whole thing.
2067 if (!PageUptodate(page
))
2070 page_zero_new_buffers(page
, start
+copied
, start
+len
);
2072 flush_dcache_page(page
);
2074 /* This could be a short (even 0-length) commit */
2075 __block_commit_write(inode
, page
, start
, start
+copied
);
2079 EXPORT_SYMBOL(block_write_end
);
2081 int generic_write_end(struct file
*file
, struct address_space
*mapping
,
2082 loff_t pos
, unsigned len
, unsigned copied
,
2083 struct page
*page
, void *fsdata
)
2085 struct inode
*inode
= mapping
->host
;
2086 loff_t old_size
= inode
->i_size
;
2087 int i_size_changed
= 0;
2089 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2092 * No need to use i_size_read() here, the i_size
2093 * cannot change under us because we hold i_mutex.
2095 * But it's important to update i_size while still holding page lock:
2096 * page writeout could otherwise come in and zero beyond i_size.
2098 if (pos
+copied
> inode
->i_size
) {
2099 i_size_write(inode
, pos
+copied
);
2104 page_cache_release(page
);
2107 pagecache_isize_extended(inode
, old_size
, pos
);
2109 * Don't mark the inode dirty under page lock. First, it unnecessarily
2110 * makes the holding time of page lock longer. Second, it forces lock
2111 * ordering of page lock and transaction start for journaling
2115 mark_inode_dirty(inode
);
2119 EXPORT_SYMBOL(generic_write_end
);
2122 * block_is_partially_uptodate checks whether buffers within a page are
2125 * Returns true if all buffers which correspond to a file portion
2126 * we want to read are uptodate.
2128 int block_is_partially_uptodate(struct page
*page
, unsigned long from
,
2129 unsigned long count
)
2131 unsigned block_start
, block_end
, blocksize
;
2133 struct buffer_head
*bh
, *head
;
2136 if (!page_has_buffers(page
))
2139 head
= page_buffers(page
);
2140 blocksize
= head
->b_size
;
2141 to
= min_t(unsigned, PAGE_CACHE_SIZE
- from
, count
);
2143 if (from
< blocksize
&& to
> PAGE_CACHE_SIZE
- blocksize
)
2149 block_end
= block_start
+ blocksize
;
2150 if (block_end
> from
&& block_start
< to
) {
2151 if (!buffer_uptodate(bh
)) {
2155 if (block_end
>= to
)
2158 block_start
= block_end
;
2159 bh
= bh
->b_this_page
;
2160 } while (bh
!= head
);
2164 EXPORT_SYMBOL(block_is_partially_uptodate
);
2167 * Generic "read page" function for block devices that have the normal
2168 * get_block functionality. This is most of the block device filesystems.
2169 * Reads the page asynchronously --- the unlock_buffer() and
2170 * set/clear_buffer_uptodate() functions propagate buffer state into the
2171 * page struct once IO has completed.
2173 int block_read_full_page(struct page
*page
, get_block_t
*get_block
)
2175 struct inode
*inode
= page
->mapping
->host
;
2176 sector_t iblock
, lblock
;
2177 struct buffer_head
*bh
, *head
, *arr
[MAX_BUF_PER_PAGE
];
2178 unsigned int blocksize
, bbits
;
2180 int fully_mapped
= 1;
2182 head
= create_page_buffers(page
, inode
, 0);
2183 blocksize
= head
->b_size
;
2184 bbits
= block_size_bits(blocksize
);
2186 iblock
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- bbits
);
2187 lblock
= (i_size_read(inode
)+blocksize
-1) >> bbits
;
2193 if (buffer_uptodate(bh
))
2196 if (!buffer_mapped(bh
)) {
2200 if (iblock
< lblock
) {
2201 WARN_ON(bh
->b_size
!= blocksize
);
2202 err
= get_block(inode
, iblock
, bh
, 0);
2206 if (!buffer_mapped(bh
)) {
2207 zero_user(page
, i
* blocksize
, blocksize
);
2209 set_buffer_uptodate(bh
);
2213 * get_block() might have updated the buffer
2216 if (buffer_uptodate(bh
))
2220 } while (i
++, iblock
++, (bh
= bh
->b_this_page
) != head
);
2223 SetPageMappedToDisk(page
);
2227 * All buffers are uptodate - we can set the page uptodate
2228 * as well. But not if get_block() returned an error.
2230 if (!PageError(page
))
2231 SetPageUptodate(page
);
2236 /* Stage two: lock the buffers */
2237 for (i
= 0; i
< nr
; i
++) {
2240 mark_buffer_async_read(bh
);
2244 * Stage 3: start the IO. Check for uptodateness
2245 * inside the buffer lock in case another process reading
2246 * the underlying blockdev brought it uptodate (the sct fix).
2248 for (i
= 0; i
< nr
; i
++) {
2250 if (buffer_uptodate(bh
))
2251 end_buffer_async_read(bh
, 1);
2253 submit_bh(READ
, bh
);
2257 EXPORT_SYMBOL(block_read_full_page
);
2259 /* utility function for filesystems that need to do work on expanding
2260 * truncates. Uses filesystem pagecache writes to allow the filesystem to
2261 * deal with the hole.
2263 int generic_cont_expand_simple(struct inode
*inode
, loff_t size
)
2265 struct address_space
*mapping
= inode
->i_mapping
;
2270 err
= inode_newsize_ok(inode
, size
);
2274 err
= pagecache_write_begin(NULL
, mapping
, size
, 0,
2275 AOP_FLAG_UNINTERRUPTIBLE
|AOP_FLAG_CONT_EXPAND
,
2280 err
= pagecache_write_end(NULL
, mapping
, size
, 0, 0, page
, fsdata
);
2286 EXPORT_SYMBOL(generic_cont_expand_simple
);
2288 static int cont_expand_zero(struct file
*file
, struct address_space
*mapping
,
2289 loff_t pos
, loff_t
*bytes
)
2291 struct inode
*inode
= mapping
->host
;
2292 unsigned blocksize
= 1 << inode
->i_blkbits
;
2295 pgoff_t index
, curidx
;
2297 unsigned zerofrom
, offset
, len
;
2300 index
= pos
>> PAGE_CACHE_SHIFT
;
2301 offset
= pos
& ~PAGE_CACHE_MASK
;
2303 while (index
> (curidx
= (curpos
= *bytes
)>>PAGE_CACHE_SHIFT
)) {
2304 zerofrom
= curpos
& ~PAGE_CACHE_MASK
;
2305 if (zerofrom
& (blocksize
-1)) {
2306 *bytes
|= (blocksize
-1);
2309 len
= PAGE_CACHE_SIZE
- zerofrom
;
2311 err
= pagecache_write_begin(file
, mapping
, curpos
, len
,
2312 AOP_FLAG_UNINTERRUPTIBLE
,
2316 zero_user(page
, zerofrom
, len
);
2317 err
= pagecache_write_end(file
, mapping
, curpos
, len
, len
,
2324 balance_dirty_pages_ratelimited(mapping
);
2326 if (unlikely(fatal_signal_pending(current
))) {
2332 /* page covers the boundary, find the boundary offset */
2333 if (index
== curidx
) {
2334 zerofrom
= curpos
& ~PAGE_CACHE_MASK
;
2335 /* if we will expand the thing last block will be filled */
2336 if (offset
<= zerofrom
) {
2339 if (zerofrom
& (blocksize
-1)) {
2340 *bytes
|= (blocksize
-1);
2343 len
= offset
- zerofrom
;
2345 err
= pagecache_write_begin(file
, mapping
, curpos
, len
,
2346 AOP_FLAG_UNINTERRUPTIBLE
,
2350 zero_user(page
, zerofrom
, len
);
2351 err
= pagecache_write_end(file
, mapping
, curpos
, len
, len
,
2363 * For moronic filesystems that do not allow holes in file.
2364 * We may have to extend the file.
2366 int cont_write_begin(struct file
*file
, struct address_space
*mapping
,
2367 loff_t pos
, unsigned len
, unsigned flags
,
2368 struct page
**pagep
, void **fsdata
,
2369 get_block_t
*get_block
, loff_t
*bytes
)
2371 struct inode
*inode
= mapping
->host
;
2372 unsigned blocksize
= 1 << inode
->i_blkbits
;
2376 err
= cont_expand_zero(file
, mapping
, pos
, bytes
);
2380 zerofrom
= *bytes
& ~PAGE_CACHE_MASK
;
2381 if (pos
+len
> *bytes
&& zerofrom
& (blocksize
-1)) {
2382 *bytes
|= (blocksize
-1);
2386 return block_write_begin(mapping
, pos
, len
, flags
, pagep
, get_block
);
2388 EXPORT_SYMBOL(cont_write_begin
);
2390 int block_commit_write(struct page
*page
, unsigned from
, unsigned to
)
2392 struct inode
*inode
= page
->mapping
->host
;
2393 __block_commit_write(inode
,page
,from
,to
);
2396 EXPORT_SYMBOL(block_commit_write
);
2399 * block_page_mkwrite() is not allowed to change the file size as it gets
2400 * called from a page fault handler when a page is first dirtied. Hence we must
2401 * be careful to check for EOF conditions here. We set the page up correctly
2402 * for a written page which means we get ENOSPC checking when writing into
2403 * holes and correct delalloc and unwritten extent mapping on filesystems that
2404 * support these features.
2406 * We are not allowed to take the i_mutex here so we have to play games to
2407 * protect against truncate races as the page could now be beyond EOF. Because
2408 * truncate writes the inode size before removing pages, once we have the
2409 * page lock we can determine safely if the page is beyond EOF. If it is not
2410 * beyond EOF, then the page is guaranteed safe against truncation until we
2413 * Direct callers of this function should protect against filesystem freezing
2414 * using sb_start_pagefault() - sb_end_pagefault() functions.
2416 int block_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
2417 get_block_t get_block
)
2419 struct page
*page
= vmf
->page
;
2420 struct inode
*inode
= file_inode(vma
->vm_file
);
2426 size
= i_size_read(inode
);
2427 if ((page
->mapping
!= inode
->i_mapping
) ||
2428 (page_offset(page
) > size
)) {
2429 /* We overload EFAULT to mean page got truncated */
2434 /* page is wholly or partially inside EOF */
2435 if (((page
->index
+ 1) << PAGE_CACHE_SHIFT
) > size
)
2436 end
= size
& ~PAGE_CACHE_MASK
;
2438 end
= PAGE_CACHE_SIZE
;
2440 ret
= __block_write_begin(page
, 0, end
, get_block
);
2442 ret
= block_commit_write(page
, 0, end
);
2444 if (unlikely(ret
< 0))
2446 set_page_dirty(page
);
2447 wait_for_stable_page(page
);
2453 EXPORT_SYMBOL(block_page_mkwrite
);
2456 * nobh_write_begin()'s prereads are special: the buffer_heads are freed
2457 * immediately, while under the page lock. So it needs a special end_io
2458 * handler which does not touch the bh after unlocking it.
2460 static void end_buffer_read_nobh(struct buffer_head
*bh
, int uptodate
)
2462 __end_buffer_read_notouch(bh
, uptodate
);
2466 * Attach the singly-linked list of buffers created by nobh_write_begin, to
2467 * the page (converting it to circular linked list and taking care of page
2470 static void attach_nobh_buffers(struct page
*page
, struct buffer_head
*head
)
2472 struct buffer_head
*bh
;
2474 BUG_ON(!PageLocked(page
));
2476 spin_lock(&page
->mapping
->private_lock
);
2479 if (PageDirty(page
))
2480 set_buffer_dirty(bh
);
2481 if (!bh
->b_this_page
)
2482 bh
->b_this_page
= head
;
2483 bh
= bh
->b_this_page
;
2484 } while (bh
!= head
);
2485 attach_page_buffers(page
, head
);
2486 spin_unlock(&page
->mapping
->private_lock
);
2490 * On entry, the page is fully not uptodate.
2491 * On exit the page is fully uptodate in the areas outside (from,to)
2492 * The filesystem needs to handle block truncation upon failure.
2494 int nobh_write_begin(struct address_space
*mapping
,
2495 loff_t pos
, unsigned len
, unsigned flags
,
2496 struct page
**pagep
, void **fsdata
,
2497 get_block_t
*get_block
)
2499 struct inode
*inode
= mapping
->host
;
2500 const unsigned blkbits
= inode
->i_blkbits
;
2501 const unsigned blocksize
= 1 << blkbits
;
2502 struct buffer_head
*head
, *bh
;
2506 unsigned block_in_page
;
2507 unsigned block_start
, block_end
;
2508 sector_t block_in_file
;
2511 int is_mapped_to_disk
= 1;
2513 index
= pos
>> PAGE_CACHE_SHIFT
;
2514 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2517 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2523 if (page_has_buffers(page
)) {
2524 ret
= __block_write_begin(page
, pos
, len
, get_block
);
2530 if (PageMappedToDisk(page
))
2534 * Allocate buffers so that we can keep track of state, and potentially
2535 * attach them to the page if an error occurs. In the common case of
2536 * no error, they will just be freed again without ever being attached
2537 * to the page (which is all OK, because we're under the page lock).
2539 * Be careful: the buffer linked list is a NULL terminated one, rather
2540 * than the circular one we're used to.
2542 head
= alloc_page_buffers(page
, blocksize
, 0);
2548 block_in_file
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- blkbits
);
2551 * We loop across all blocks in the page, whether or not they are
2552 * part of the affected region. This is so we can discover if the
2553 * page is fully mapped-to-disk.
2555 for (block_start
= 0, block_in_page
= 0, bh
= head
;
2556 block_start
< PAGE_CACHE_SIZE
;
2557 block_in_page
++, block_start
+= blocksize
, bh
= bh
->b_this_page
) {
2560 block_end
= block_start
+ blocksize
;
2563 if (block_start
>= to
)
2565 ret
= get_block(inode
, block_in_file
+ block_in_page
,
2569 if (!buffer_mapped(bh
))
2570 is_mapped_to_disk
= 0;
2572 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
2573 if (PageUptodate(page
)) {
2574 set_buffer_uptodate(bh
);
2577 if (buffer_new(bh
) || !buffer_mapped(bh
)) {
2578 zero_user_segments(page
, block_start
, from
,
2582 if (buffer_uptodate(bh
))
2583 continue; /* reiserfs does this */
2584 if (block_start
< from
|| block_end
> to
) {
2586 bh
->b_end_io
= end_buffer_read_nobh
;
2587 submit_bh(READ
, bh
);
2594 * The page is locked, so these buffers are protected from
2595 * any VM or truncate activity. Hence we don't need to care
2596 * for the buffer_head refcounts.
2598 for (bh
= head
; bh
; bh
= bh
->b_this_page
) {
2600 if (!buffer_uptodate(bh
))
2607 if (is_mapped_to_disk
)
2608 SetPageMappedToDisk(page
);
2610 *fsdata
= head
; /* to be released by nobh_write_end */
2617 * Error recovery is a bit difficult. We need to zero out blocks that
2618 * were newly allocated, and dirty them to ensure they get written out.
2619 * Buffers need to be attached to the page at this point, otherwise
2620 * the handling of potential IO errors during writeout would be hard
2621 * (could try doing synchronous writeout, but what if that fails too?)
2623 attach_nobh_buffers(page
, head
);
2624 page_zero_new_buffers(page
, from
, to
);
2628 page_cache_release(page
);
2633 EXPORT_SYMBOL(nobh_write_begin
);
2635 int nobh_write_end(struct file
*file
, struct address_space
*mapping
,
2636 loff_t pos
, unsigned len
, unsigned copied
,
2637 struct page
*page
, void *fsdata
)
2639 struct inode
*inode
= page
->mapping
->host
;
2640 struct buffer_head
*head
= fsdata
;
2641 struct buffer_head
*bh
;
2642 BUG_ON(fsdata
!= NULL
&& page_has_buffers(page
));
2644 if (unlikely(copied
< len
) && head
)
2645 attach_nobh_buffers(page
, head
);
2646 if (page_has_buffers(page
))
2647 return generic_write_end(file
, mapping
, pos
, len
,
2648 copied
, page
, fsdata
);
2650 SetPageUptodate(page
);
2651 set_page_dirty(page
);
2652 if (pos
+copied
> inode
->i_size
) {
2653 i_size_write(inode
, pos
+copied
);
2654 mark_inode_dirty(inode
);
2658 page_cache_release(page
);
2662 head
= head
->b_this_page
;
2663 free_buffer_head(bh
);
2668 EXPORT_SYMBOL(nobh_write_end
);
2671 * nobh_writepage() - based on block_full_write_page() except
2672 * that it tries to operate without attaching bufferheads to
2675 int nobh_writepage(struct page
*page
, get_block_t
*get_block
,
2676 struct writeback_control
*wbc
)
2678 struct inode
* const inode
= page
->mapping
->host
;
2679 loff_t i_size
= i_size_read(inode
);
2680 const pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2684 /* Is the page fully inside i_size? */
2685 if (page
->index
< end_index
)
2688 /* Is the page fully outside i_size? (truncate in progress) */
2689 offset
= i_size
& (PAGE_CACHE_SIZE
-1);
2690 if (page
->index
>= end_index
+1 || !offset
) {
2692 * The page may have dirty, unmapped buffers. For example,
2693 * they may have been added in ext3_writepage(). Make them
2694 * freeable here, so the page does not leak.
2697 /* Not really sure about this - do we need this ? */
2698 if (page
->mapping
->a_ops
->invalidatepage
)
2699 page
->mapping
->a_ops
->invalidatepage(page
, offset
);
2702 return 0; /* don't care */
2706 * The page straddles i_size. It must be zeroed out on each and every
2707 * writepage invocation because it may be mmapped. "A file is mapped
2708 * in multiples of the page size. For a file that is not a multiple of
2709 * the page size, the remaining memory is zeroed when mapped, and
2710 * writes to that region are not written out to the file."
2712 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
2714 ret
= mpage_writepage(page
, get_block
, wbc
);
2716 ret
= __block_write_full_page(inode
, page
, get_block
, wbc
,
2717 end_buffer_async_write
);
2720 EXPORT_SYMBOL(nobh_writepage
);
2722 int nobh_truncate_page(struct address_space
*mapping
,
2723 loff_t from
, get_block_t
*get_block
)
2725 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
2726 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2729 unsigned length
, pos
;
2730 struct inode
*inode
= mapping
->host
;
2732 struct buffer_head map_bh
;
2735 blocksize
= 1 << inode
->i_blkbits
;
2736 length
= offset
& (blocksize
- 1);
2738 /* Block boundary? Nothing to do */
2742 length
= blocksize
- length
;
2743 iblock
= (sector_t
)index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2745 page
= grab_cache_page(mapping
, index
);
2750 if (page_has_buffers(page
)) {
2753 page_cache_release(page
);
2754 return block_truncate_page(mapping
, from
, get_block
);
2757 /* Find the buffer that contains "offset" */
2759 while (offset
>= pos
) {
2764 map_bh
.b_size
= blocksize
;
2766 err
= get_block(inode
, iblock
, &map_bh
, 0);
2769 /* unmapped? It's a hole - nothing to do */
2770 if (!buffer_mapped(&map_bh
))
2773 /* Ok, it's mapped. Make sure it's up-to-date */
2774 if (!PageUptodate(page
)) {
2775 err
= mapping
->a_ops
->readpage(NULL
, page
);
2777 page_cache_release(page
);
2781 if (!PageUptodate(page
)) {
2785 if (page_has_buffers(page
))
2788 zero_user(page
, offset
, length
);
2789 set_page_dirty(page
);
2794 page_cache_release(page
);
2798 EXPORT_SYMBOL(nobh_truncate_page
);
2800 int block_truncate_page(struct address_space
*mapping
,
2801 loff_t from
, get_block_t
*get_block
)
2803 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
2804 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2807 unsigned length
, pos
;
2808 struct inode
*inode
= mapping
->host
;
2810 struct buffer_head
*bh
;
2813 blocksize
= 1 << inode
->i_blkbits
;
2814 length
= offset
& (blocksize
- 1);
2816 /* Block boundary? Nothing to do */
2820 length
= blocksize
- length
;
2821 iblock
= (sector_t
)index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2823 page
= grab_cache_page(mapping
, index
);
2828 if (!page_has_buffers(page
))
2829 create_empty_buffers(page
, blocksize
, 0);
2831 /* Find the buffer that contains "offset" */
2832 bh
= page_buffers(page
);
2834 while (offset
>= pos
) {
2835 bh
= bh
->b_this_page
;
2841 if (!buffer_mapped(bh
)) {
2842 WARN_ON(bh
->b_size
!= blocksize
);
2843 err
= get_block(inode
, iblock
, bh
, 0);
2846 /* unmapped? It's a hole - nothing to do */
2847 if (!buffer_mapped(bh
))
2851 /* Ok, it's mapped. Make sure it's up-to-date */
2852 if (PageUptodate(page
))
2853 set_buffer_uptodate(bh
);
2855 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) && !buffer_unwritten(bh
)) {
2857 ll_rw_block(READ
, 1, &bh
);
2859 /* Uhhuh. Read error. Complain and punt. */
2860 if (!buffer_uptodate(bh
))
2864 zero_user(page
, offset
, length
);
2865 mark_buffer_dirty(bh
);
2870 page_cache_release(page
);
2874 EXPORT_SYMBOL(block_truncate_page
);
2877 * The generic ->writepage function for buffer-backed address_spaces
2879 int block_write_full_page(struct page
*page
, get_block_t
*get_block
,
2880 struct writeback_control
*wbc
)
2882 struct inode
* const inode
= page
->mapping
->host
;
2883 loff_t i_size
= i_size_read(inode
);
2884 const pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
2887 /* Is the page fully inside i_size? */
2888 if (page
->index
< end_index
)
2889 return __block_write_full_page(inode
, page
, get_block
, wbc
,
2890 end_buffer_async_write
);
2892 /* Is the page fully outside i_size? (truncate in progress) */
2893 offset
= i_size
& (PAGE_CACHE_SIZE
-1);
2894 if (page
->index
>= end_index
+1 || !offset
) {
2896 * The page may have dirty, unmapped buffers. For example,
2897 * they may have been added in ext3_writepage(). Make them
2898 * freeable here, so the page does not leak.
2900 do_invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
2902 return 0; /* don't care */
2906 * The page straddles i_size. It must be zeroed out on each and every
2907 * writepage invocation because it may be mmapped. "A file is mapped
2908 * in multiples of the page size. For a file that is not a multiple of
2909 * the page size, the remaining memory is zeroed when mapped, and
2910 * writes to that region are not written out to the file."
2912 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
2913 return __block_write_full_page(inode
, page
, get_block
, wbc
,
2914 end_buffer_async_write
);
2916 EXPORT_SYMBOL(block_write_full_page
);
2918 sector_t
generic_block_bmap(struct address_space
*mapping
, sector_t block
,
2919 get_block_t
*get_block
)
2921 struct buffer_head tmp
;
2922 struct inode
*inode
= mapping
->host
;
2925 tmp
.b_size
= 1 << inode
->i_blkbits
;
2926 get_block(inode
, block
, &tmp
, 0);
2927 return tmp
.b_blocknr
;
2929 EXPORT_SYMBOL(generic_block_bmap
);
2931 static void end_bio_bh_io_sync(struct bio
*bio
)
2933 struct buffer_head
*bh
= bio
->bi_private
;
2935 if (unlikely(bio_flagged(bio
, BIO_QUIET
)))
2936 set_bit(BH_Quiet
, &bh
->b_state
);
2938 bh
->b_end_io(bh
, !bio
->bi_error
);
2943 * This allows us to do IO even on the odd last sectors
2944 * of a device, even if the block size is some multiple
2945 * of the physical sector size.
2947 * We'll just truncate the bio to the size of the device,
2948 * and clear the end of the buffer head manually.
2950 * Truly out-of-range accesses will turn into actual IO
2951 * errors, this only handles the "we need to be able to
2952 * do IO at the final sector" case.
2954 void guard_bio_eod(int rw
, struct bio
*bio
)
2957 struct bio_vec
*bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
2958 unsigned truncated_bytes
;
2960 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
2965 * If the *whole* IO is past the end of the device,
2966 * let it through, and the IO layer will turn it into
2969 if (unlikely(bio
->bi_iter
.bi_sector
>= maxsector
))
2972 maxsector
-= bio
->bi_iter
.bi_sector
;
2973 if (likely((bio
->bi_iter
.bi_size
>> 9) <= maxsector
))
2976 /* Uhhuh. We've got a bio that straddles the device size! */
2977 truncated_bytes
= bio
->bi_iter
.bi_size
- (maxsector
<< 9);
2979 /* Truncate the bio.. */
2980 bio
->bi_iter
.bi_size
-= truncated_bytes
;
2981 bvec
->bv_len
-= truncated_bytes
;
2983 /* ..and clear the end of the buffer for reads */
2984 if ((rw
& RW_MASK
) == READ
) {
2985 zero_user(bvec
->bv_page
, bvec
->bv_offset
+ bvec
->bv_len
,
2990 static int submit_bh_wbc(int rw
, struct buffer_head
*bh
,
2991 unsigned long bio_flags
, struct writeback_control
*wbc
)
2995 BUG_ON(!buffer_locked(bh
));
2996 BUG_ON(!buffer_mapped(bh
));
2997 BUG_ON(!bh
->b_end_io
);
2998 BUG_ON(buffer_delay(bh
));
2999 BUG_ON(buffer_unwritten(bh
));
3002 * Only clear out a write error when rewriting
3004 if (test_set_buffer_req(bh
) && (rw
& WRITE
))
3005 clear_buffer_write_io_error(bh
);
3008 * from here on down, it's all bio -- do the initial mapping,
3009 * submit_bio -> generic_make_request may further map this bio around
3011 bio
= bio_alloc(GFP_NOIO
, 1);
3014 wbc_init_bio(wbc
, bio
);
3015 wbc_account_io(wbc
, bh
->b_page
, bh
->b_size
);
3018 bio
->bi_iter
.bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
3019 bio
->bi_bdev
= bh
->b_bdev
;
3021 bio_add_page(bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
3022 BUG_ON(bio
->bi_iter
.bi_size
!= bh
->b_size
);
3024 bio
->bi_end_io
= end_bio_bh_io_sync
;
3025 bio
->bi_private
= bh
;
3026 bio
->bi_flags
|= bio_flags
;
3028 /* Take care of bh's that straddle the end of the device */
3029 guard_bio_eod(rw
, bio
);
3031 if (buffer_meta(bh
))
3033 if (buffer_prio(bh
))
3036 submit_bio(rw
, bio
);
3040 int _submit_bh(int rw
, struct buffer_head
*bh
, unsigned long bio_flags
)
3042 return submit_bh_wbc(rw
, bh
, bio_flags
, NULL
);
3044 EXPORT_SYMBOL_GPL(_submit_bh
);
3046 int submit_bh(int rw
, struct buffer_head
*bh
)
3048 return submit_bh_wbc(rw
, bh
, 0, NULL
);
3050 EXPORT_SYMBOL(submit_bh
);
3053 * ll_rw_block: low-level access to block devices (DEPRECATED)
3054 * @rw: whether to %READ or %WRITE or maybe %READA (readahead)
3055 * @nr: number of &struct buffer_heads in the array
3056 * @bhs: array of pointers to &struct buffer_head
3058 * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
3059 * requests an I/O operation on them, either a %READ or a %WRITE. The third
3060 * %READA option is described in the documentation for generic_make_request()
3061 * which ll_rw_block() calls.
3063 * This function drops any buffer that it cannot get a lock on (with the
3064 * BH_Lock state bit), any buffer that appears to be clean when doing a write
3065 * request, and any buffer that appears to be up-to-date when doing read
3066 * request. Further it marks as clean buffers that are processed for
3067 * writing (the buffer cache won't assume that they are actually clean
3068 * until the buffer gets unlocked).
3070 * ll_rw_block sets b_end_io to simple completion handler that marks
3071 * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
3074 * All of the buffers must be for the same device, and must also be a
3075 * multiple of the current approved size for the device.
3077 void ll_rw_block(int rw
, int nr
, struct buffer_head
*bhs
[])
3081 for (i
= 0; i
< nr
; i
++) {
3082 struct buffer_head
*bh
= bhs
[i
];
3084 if (!trylock_buffer(bh
))
3087 if (test_clear_buffer_dirty(bh
)) {
3088 bh
->b_end_io
= end_buffer_write_sync
;
3090 submit_bh(WRITE
, bh
);
3094 if (!buffer_uptodate(bh
)) {
3095 bh
->b_end_io
= end_buffer_read_sync
;
3104 EXPORT_SYMBOL(ll_rw_block
);
3106 void write_dirty_buffer(struct buffer_head
*bh
, int rw
)
3109 if (!test_clear_buffer_dirty(bh
)) {
3113 bh
->b_end_io
= end_buffer_write_sync
;
3117 EXPORT_SYMBOL(write_dirty_buffer
);
3120 * For a data-integrity writeout, we need to wait upon any in-progress I/O
3121 * and then start new I/O and then wait upon it. The caller must have a ref on
3124 int __sync_dirty_buffer(struct buffer_head
*bh
, int rw
)
3128 WARN_ON(atomic_read(&bh
->b_count
) < 1);
3130 if (test_clear_buffer_dirty(bh
)) {
3132 bh
->b_end_io
= end_buffer_write_sync
;
3133 ret
= submit_bh(rw
, bh
);
3135 if (!ret
&& !buffer_uptodate(bh
))
3142 EXPORT_SYMBOL(__sync_dirty_buffer
);
3144 int sync_dirty_buffer(struct buffer_head
*bh
)
3146 return __sync_dirty_buffer(bh
, WRITE_SYNC
);
3148 EXPORT_SYMBOL(sync_dirty_buffer
);
3151 * try_to_free_buffers() checks if all the buffers on this particular page
3152 * are unused, and releases them if so.
3154 * Exclusion against try_to_free_buffers may be obtained by either
3155 * locking the page or by holding its mapping's private_lock.
3157 * If the page is dirty but all the buffers are clean then we need to
3158 * be sure to mark the page clean as well. This is because the page
3159 * may be against a block device, and a later reattachment of buffers
3160 * to a dirty page will set *all* buffers dirty. Which would corrupt
3161 * filesystem data on the same device.
3163 * The same applies to regular filesystem pages: if all the buffers are
3164 * clean then we set the page clean and proceed. To do that, we require
3165 * total exclusion from __set_page_dirty_buffers(). That is obtained with
3168 * try_to_free_buffers() is non-blocking.
3170 static inline int buffer_busy(struct buffer_head
*bh
)
3172 return atomic_read(&bh
->b_count
) |
3173 (bh
->b_state
& ((1 << BH_Dirty
) | (1 << BH_Lock
)));
3177 drop_buffers(struct page
*page
, struct buffer_head
**buffers_to_free
)
3179 struct buffer_head
*head
= page_buffers(page
);
3180 struct buffer_head
*bh
;
3184 if (buffer_write_io_error(bh
) && page
->mapping
)
3185 set_bit(AS_EIO
, &page
->mapping
->flags
);
3186 if (buffer_busy(bh
))
3188 bh
= bh
->b_this_page
;
3189 } while (bh
!= head
);
3192 struct buffer_head
*next
= bh
->b_this_page
;
3194 if (bh
->b_assoc_map
)
3195 __remove_assoc_queue(bh
);
3197 } while (bh
!= head
);
3198 *buffers_to_free
= head
;
3199 __clear_page_buffers(page
);
3205 int try_to_free_buffers(struct page
*page
)
3207 struct address_space
* const mapping
= page
->mapping
;
3208 struct buffer_head
*buffers_to_free
= NULL
;
3211 BUG_ON(!PageLocked(page
));
3212 if (PageWriteback(page
))
3215 if (mapping
== NULL
) { /* can this still happen? */
3216 ret
= drop_buffers(page
, &buffers_to_free
);
3220 spin_lock(&mapping
->private_lock
);
3221 ret
= drop_buffers(page
, &buffers_to_free
);
3224 * If the filesystem writes its buffers by hand (eg ext3)
3225 * then we can have clean buffers against a dirty page. We
3226 * clean the page here; otherwise the VM will never notice
3227 * that the filesystem did any IO at all.
3229 * Also, during truncate, discard_buffer will have marked all
3230 * the page's buffers clean. We discover that here and clean
3233 * private_lock must be held over this entire operation in order
3234 * to synchronise against __set_page_dirty_buffers and prevent the
3235 * dirty bit from being lost.
3238 cancel_dirty_page(page
);
3239 spin_unlock(&mapping
->private_lock
);
3241 if (buffers_to_free
) {
3242 struct buffer_head
*bh
= buffers_to_free
;
3245 struct buffer_head
*next
= bh
->b_this_page
;
3246 free_buffer_head(bh
);
3248 } while (bh
!= buffers_to_free
);
3252 EXPORT_SYMBOL(try_to_free_buffers
);
3255 * There are no bdflush tunables left. But distributions are
3256 * still running obsolete flush daemons, so we terminate them here.
3258 * Use of bdflush() is deprecated and will be removed in a future kernel.
3259 * The `flush-X' kernel threads fully replace bdflush daemons and this call.
3261 SYSCALL_DEFINE2(bdflush
, int, func
, long, data
)
3263 static int msg_count
;
3265 if (!capable(CAP_SYS_ADMIN
))
3268 if (msg_count
< 5) {
3271 "warning: process `%s' used the obsolete bdflush"
3272 " system call\n", current
->comm
);
3273 printk(KERN_INFO
"Fix your initscripts?\n");
3282 * Buffer-head allocation
3284 static struct kmem_cache
*bh_cachep __read_mostly
;
3287 * Once the number of bh's in the machine exceeds this level, we start
3288 * stripping them in writeback.
3290 static unsigned long max_buffer_heads
;
3292 int buffer_heads_over_limit
;
3294 struct bh_accounting
{
3295 int nr
; /* Number of live bh's */
3296 int ratelimit
; /* Limit cacheline bouncing */
3299 static DEFINE_PER_CPU(struct bh_accounting
, bh_accounting
) = {0, 0};
3301 static void recalc_bh_state(void)
3306 if (__this_cpu_inc_return(bh_accounting
.ratelimit
) - 1 < 4096)
3308 __this_cpu_write(bh_accounting
.ratelimit
, 0);
3309 for_each_online_cpu(i
)
3310 tot
+= per_cpu(bh_accounting
, i
).nr
;
3311 buffer_heads_over_limit
= (tot
> max_buffer_heads
);
3314 struct buffer_head
*alloc_buffer_head(gfp_t gfp_flags
)
3316 struct buffer_head
*ret
= kmem_cache_zalloc(bh_cachep
, gfp_flags
);
3318 INIT_LIST_HEAD(&ret
->b_assoc_buffers
);
3320 __this_cpu_inc(bh_accounting
.nr
);
3326 EXPORT_SYMBOL(alloc_buffer_head
);
3328 void free_buffer_head(struct buffer_head
*bh
)
3330 BUG_ON(!list_empty(&bh
->b_assoc_buffers
));
3331 kmem_cache_free(bh_cachep
, bh
);
3333 __this_cpu_dec(bh_accounting
.nr
);
3337 EXPORT_SYMBOL(free_buffer_head
);
3339 static void buffer_exit_cpu(int cpu
)
3342 struct bh_lru
*b
= &per_cpu(bh_lrus
, cpu
);
3344 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
3348 this_cpu_add(bh_accounting
.nr
, per_cpu(bh_accounting
, cpu
).nr
);
3349 per_cpu(bh_accounting
, cpu
).nr
= 0;
3352 static int buffer_cpu_notify(struct notifier_block
*self
,
3353 unsigned long action
, void *hcpu
)
3355 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
)
3356 buffer_exit_cpu((unsigned long)hcpu
);
3361 * bh_uptodate_or_lock - Test whether the buffer is uptodate
3362 * @bh: struct buffer_head
3364 * Return true if the buffer is up-to-date and false,
3365 * with the buffer locked, if not.
3367 int bh_uptodate_or_lock(struct buffer_head
*bh
)
3369 if (!buffer_uptodate(bh
)) {
3371 if (!buffer_uptodate(bh
))
3377 EXPORT_SYMBOL(bh_uptodate_or_lock
);
3380 * bh_submit_read - Submit a locked buffer for reading
3381 * @bh: struct buffer_head
3383 * Returns zero on success and -EIO on error.
3385 int bh_submit_read(struct buffer_head
*bh
)
3387 BUG_ON(!buffer_locked(bh
));
3389 if (buffer_uptodate(bh
)) {
3395 bh
->b_end_io
= end_buffer_read_sync
;
3396 submit_bh(READ
, bh
);
3398 if (buffer_uptodate(bh
))
3402 EXPORT_SYMBOL(bh_submit_read
);
3404 void __init
buffer_init(void)
3406 unsigned long nrpages
;
3408 bh_cachep
= kmem_cache_create("buffer_head",
3409 sizeof(struct buffer_head
), 0,
3410 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
3415 * Limit the bh occupancy to 10% of ZONE_NORMAL
3417 nrpages
= (nr_free_buffer_pages() * 10) / 100;
3418 max_buffer_heads
= nrpages
* (PAGE_SIZE
/ sizeof(struct buffer_head
));
3419 hotcpu_notifier(buffer_cpu_notify
, 0);