4 * Copyright (C) 1991, 1992, 2002 Linus Torvalds
8 * Start bdflush() with kernel_thread not syscall - Paul Gortmaker, 12/95
10 * Removed a lot of unnecessary code and simplified things now that
11 * the buffer cache isn't our primary cache - Andrew Tridgell 12/96
13 * Speed up hash, lru, and free list operations. Use gfp() for allocating
14 * hash table, use SLAB cache for buffer heads. SMP threading. -DaveM
16 * Added 32k buffer block sizes - these are required older ARM systems. - RMK
18 * async buffer flushing, 1999 Andrea Arcangeli <andrea@suse.de>
21 #include <linux/kernel.h>
22 #include <linux/sched/signal.h>
23 #include <linux/syscalls.h>
25 #include <linux/iomap.h>
27 #include <linux/percpu.h>
28 #include <linux/slab.h>
29 #include <linux/capability.h>
30 #include <linux/blkdev.h>
31 #include <linux/file.h>
32 #include <linux/quotaops.h>
33 #include <linux/highmem.h>
34 #include <linux/export.h>
35 #include <linux/backing-dev.h>
36 #include <linux/writeback.h>
37 #include <linux/hash.h>
38 #include <linux/suspend.h>
39 #include <linux/buffer_head.h>
40 #include <linux/task_io_accounting_ops.h>
41 #include <linux/bio.h>
42 #include <linux/notifier.h>
43 #include <linux/cpu.h>
44 #include <linux/bitops.h>
45 #include <linux/mpage.h>
46 #include <linux/bit_spinlock.h>
47 #include <linux/pagevec.h>
48 #include <trace/events/block.h>
50 static int fsync_buffers_list(spinlock_t
*lock
, struct list_head
*list
);
51 static int submit_bh_wbc(int op
, int op_flags
, struct buffer_head
*bh
,
52 enum rw_hint hint
, struct writeback_control
*wbc
);
54 #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
56 void init_buffer(struct buffer_head
*bh
, bh_end_io_t
*handler
, void *private)
58 bh
->b_end_io
= handler
;
59 bh
->b_private
= private;
61 EXPORT_SYMBOL(init_buffer
);
63 inline void touch_buffer(struct buffer_head
*bh
)
65 trace_block_touch_buffer(bh
);
66 mark_page_accessed(bh
->b_page
);
68 EXPORT_SYMBOL(touch_buffer
);
70 void __lock_buffer(struct buffer_head
*bh
)
72 wait_on_bit_lock_io(&bh
->b_state
, BH_Lock
, TASK_UNINTERRUPTIBLE
);
74 EXPORT_SYMBOL(__lock_buffer
);
76 void unlock_buffer(struct buffer_head
*bh
)
78 clear_bit_unlock(BH_Lock
, &bh
->b_state
);
79 smp_mb__after_atomic();
80 wake_up_bit(&bh
->b_state
, BH_Lock
);
82 EXPORT_SYMBOL(unlock_buffer
);
85 * Returns if the page has dirty or writeback buffers. If all the buffers
86 * are unlocked and clean then the PageDirty information is stale. If
87 * any of the pages are locked, it is assumed they are locked for IO.
89 void buffer_check_dirty_writeback(struct page
*page
,
90 bool *dirty
, bool *writeback
)
92 struct buffer_head
*head
, *bh
;
96 BUG_ON(!PageLocked(page
));
98 if (!page_has_buffers(page
))
101 if (PageWriteback(page
))
104 head
= page_buffers(page
);
107 if (buffer_locked(bh
))
110 if (buffer_dirty(bh
))
113 bh
= bh
->b_this_page
;
114 } while (bh
!= head
);
116 EXPORT_SYMBOL(buffer_check_dirty_writeback
);
119 * Block until a buffer comes unlocked. This doesn't stop it
120 * from becoming locked again - you have to lock it yourself
121 * if you want to preserve its state.
123 void __wait_on_buffer(struct buffer_head
* bh
)
125 wait_on_bit_io(&bh
->b_state
, BH_Lock
, TASK_UNINTERRUPTIBLE
);
127 EXPORT_SYMBOL(__wait_on_buffer
);
130 __clear_page_buffers(struct page
*page
)
132 ClearPagePrivate(page
);
133 set_page_private(page
, 0);
137 static void buffer_io_error(struct buffer_head
*bh
, char *msg
)
139 if (!test_bit(BH_Quiet
, &bh
->b_state
))
140 printk_ratelimited(KERN_ERR
141 "Buffer I/O error on dev %pg, logical block %llu%s\n",
142 bh
->b_bdev
, (unsigned long long)bh
->b_blocknr
, msg
);
146 * End-of-IO handler helper function which does not touch the bh after
148 * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
149 * a race there is benign: unlock_buffer() only use the bh's address for
150 * hashing after unlocking the buffer, so it doesn't actually touch the bh
153 static void __end_buffer_read_notouch(struct buffer_head
*bh
, int uptodate
)
156 set_buffer_uptodate(bh
);
158 /* This happens, due to failed read-ahead attempts. */
159 clear_buffer_uptodate(bh
);
165 * Default synchronous end-of-IO handler.. Just mark it up-to-date and
166 * unlock the buffer. This is what ll_rw_block uses too.
168 void end_buffer_read_sync(struct buffer_head
*bh
, int uptodate
)
170 __end_buffer_read_notouch(bh
, uptodate
);
173 EXPORT_SYMBOL(end_buffer_read_sync
);
175 void end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
178 set_buffer_uptodate(bh
);
180 buffer_io_error(bh
, ", lost sync page write");
181 set_buffer_write_io_error(bh
);
182 clear_buffer_uptodate(bh
);
187 EXPORT_SYMBOL(end_buffer_write_sync
);
190 * Various filesystems appear to want __find_get_block to be non-blocking.
191 * But it's the page lock which protects the buffers. To get around this,
192 * we get exclusion from try_to_free_buffers with the blockdev mapping's
195 * Hack idea: for the blockdev mapping, i_bufferlist_lock contention
196 * may be quite high. This code could TryLock the page, and if that
197 * succeeds, there is no need to take private_lock. (But if
198 * private_lock is contended then so is mapping->tree_lock).
200 static struct buffer_head
*
201 __find_get_block_slow(struct block_device
*bdev
, sector_t block
)
203 struct inode
*bd_inode
= bdev
->bd_inode
;
204 struct address_space
*bd_mapping
= bd_inode
->i_mapping
;
205 struct buffer_head
*ret
= NULL
;
207 struct buffer_head
*bh
;
208 struct buffer_head
*head
;
212 index
= block
>> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
213 page
= find_get_page_flags(bd_mapping
, index
, FGP_ACCESSED
);
217 spin_lock(&bd_mapping
->private_lock
);
218 if (!page_has_buffers(page
))
220 head
= page_buffers(page
);
223 if (!buffer_mapped(bh
))
225 else if (bh
->b_blocknr
== block
) {
230 bh
= bh
->b_this_page
;
231 } while (bh
!= head
);
233 /* we might be here because some of the buffers on this page are
234 * not mapped. This is due to various races between
235 * file io on the block device and getblk. It gets dealt with
236 * elsewhere, don't buffer_error if we had some unmapped buffers
239 printk("__find_get_block_slow() failed. "
240 "block=%llu, b_blocknr=%llu\n",
241 (unsigned long long)block
,
242 (unsigned long long)bh
->b_blocknr
);
243 printk("b_state=0x%08lx, b_size=%zu\n",
244 bh
->b_state
, bh
->b_size
);
245 printk("device %pg blocksize: %d\n", bdev
,
246 1 << bd_inode
->i_blkbits
);
249 spin_unlock(&bd_mapping
->private_lock
);
256 * Kick the writeback threads then try to free up some ZONE_NORMAL memory.
258 static void free_more_memory(void)
263 wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM
);
266 for_each_online_node(nid
) {
268 z
= first_zones_zonelist(node_zonelist(nid
, GFP_NOFS
),
269 gfp_zone(GFP_NOFS
), NULL
);
271 try_to_free_pages(node_zonelist(nid
, GFP_NOFS
), 0,
277 * I/O completion handler for block_read_full_page() - pages
278 * which come unlocked at the end of I/O.
280 static void end_buffer_async_read(struct buffer_head
*bh
, int uptodate
)
283 struct buffer_head
*first
;
284 struct buffer_head
*tmp
;
286 int page_uptodate
= 1;
288 BUG_ON(!buffer_async_read(bh
));
292 set_buffer_uptodate(bh
);
294 clear_buffer_uptodate(bh
);
295 buffer_io_error(bh
, ", async page read");
300 * Be _very_ careful from here on. Bad things can happen if
301 * two buffer heads end IO at almost the same time and both
302 * decide that the page is now completely done.
304 first
= page_buffers(page
);
305 local_irq_save(flags
);
306 bit_spin_lock(BH_Uptodate_Lock
, &first
->b_state
);
307 clear_buffer_async_read(bh
);
311 if (!buffer_uptodate(tmp
))
313 if (buffer_async_read(tmp
)) {
314 BUG_ON(!buffer_locked(tmp
));
317 tmp
= tmp
->b_this_page
;
319 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
320 local_irq_restore(flags
);
323 * If none of the buffers had errors and they are all
324 * uptodate then we can set the page uptodate.
326 if (page_uptodate
&& !PageError(page
))
327 SetPageUptodate(page
);
332 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
333 local_irq_restore(flags
);
338 * Completion handler for block_write_full_page() - pages which are unlocked
339 * during I/O, and which have PageWriteback cleared upon I/O completion.
341 void end_buffer_async_write(struct buffer_head
*bh
, int uptodate
)
344 struct buffer_head
*first
;
345 struct buffer_head
*tmp
;
348 BUG_ON(!buffer_async_write(bh
));
352 set_buffer_uptodate(bh
);
354 buffer_io_error(bh
, ", lost async page write");
355 mapping_set_error(page
->mapping
, -EIO
);
356 set_buffer_write_io_error(bh
);
357 clear_buffer_uptodate(bh
);
361 first
= page_buffers(page
);
362 local_irq_save(flags
);
363 bit_spin_lock(BH_Uptodate_Lock
, &first
->b_state
);
365 clear_buffer_async_write(bh
);
367 tmp
= bh
->b_this_page
;
369 if (buffer_async_write(tmp
)) {
370 BUG_ON(!buffer_locked(tmp
));
373 tmp
= tmp
->b_this_page
;
375 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
376 local_irq_restore(flags
);
377 end_page_writeback(page
);
381 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
382 local_irq_restore(flags
);
385 EXPORT_SYMBOL(end_buffer_async_write
);
388 * If a page's buffers are under async readin (end_buffer_async_read
389 * completion) then there is a possibility that another thread of
390 * control could lock one of the buffers after it has completed
391 * but while some of the other buffers have not completed. This
392 * locked buffer would confuse end_buffer_async_read() into not unlocking
393 * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
394 * that this buffer is not under async I/O.
396 * The page comes unlocked when it has no locked buffer_async buffers
399 * PageLocked prevents anyone starting new async I/O reads any of
402 * PageWriteback is used to prevent simultaneous writeout of the same
405 * PageLocked prevents anyone from starting writeback of a page which is
406 * under read I/O (PageWriteback is only ever set against a locked page).
408 static void mark_buffer_async_read(struct buffer_head
*bh
)
410 bh
->b_end_io
= end_buffer_async_read
;
411 set_buffer_async_read(bh
);
414 static void mark_buffer_async_write_endio(struct buffer_head
*bh
,
415 bh_end_io_t
*handler
)
417 bh
->b_end_io
= handler
;
418 set_buffer_async_write(bh
);
421 void mark_buffer_async_write(struct buffer_head
*bh
)
423 mark_buffer_async_write_endio(bh
, end_buffer_async_write
);
425 EXPORT_SYMBOL(mark_buffer_async_write
);
429 * fs/buffer.c contains helper functions for buffer-backed address space's
430 * fsync functions. A common requirement for buffer-based filesystems is
431 * that certain data from the backing blockdev needs to be written out for
432 * a successful fsync(). For example, ext2 indirect blocks need to be
433 * written back and waited upon before fsync() returns.
435 * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
436 * inode_has_buffers() and invalidate_inode_buffers() are provided for the
437 * management of a list of dependent buffers at ->i_mapping->private_list.
439 * Locking is a little subtle: try_to_free_buffers() will remove buffers
440 * from their controlling inode's queue when they are being freed. But
441 * try_to_free_buffers() will be operating against the *blockdev* mapping
442 * at the time, not against the S_ISREG file which depends on those buffers.
443 * So the locking for private_list is via the private_lock in the address_space
444 * which backs the buffers. Which is different from the address_space
445 * against which the buffers are listed. So for a particular address_space,
446 * mapping->private_lock does *not* protect mapping->private_list! In fact,
447 * mapping->private_list will always be protected by the backing blockdev's
450 * Which introduces a requirement: all buffers on an address_space's
451 * ->private_list must be from the same address_space: the blockdev's.
453 * address_spaces which do not place buffers at ->private_list via these
454 * utility functions are free to use private_lock and private_list for
455 * whatever they want. The only requirement is that list_empty(private_list)
456 * be true at clear_inode() time.
458 * FIXME: clear_inode should not call invalidate_inode_buffers(). The
459 * filesystems should do that. invalidate_inode_buffers() should just go
460 * BUG_ON(!list_empty).
462 * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
463 * take an address_space, not an inode. And it should be called
464 * mark_buffer_dirty_fsync() to clearly define why those buffers are being
467 * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
468 * list if it is already on a list. Because if the buffer is on a list,
469 * it *must* already be on the right one. If not, the filesystem is being
470 * silly. This will save a ton of locking. But first we have to ensure
471 * that buffers are taken *off* the old inode's list when they are freed
472 * (presumably in truncate). That requires careful auditing of all
473 * filesystems (do it inside bforget()). It could also be done by bringing
478 * The buffer's backing address_space's private_lock must be held
480 static void __remove_assoc_queue(struct buffer_head
*bh
)
482 list_del_init(&bh
->b_assoc_buffers
);
483 WARN_ON(!bh
->b_assoc_map
);
484 if (buffer_write_io_error(bh
))
485 set_bit(AS_EIO
, &bh
->b_assoc_map
->flags
);
486 bh
->b_assoc_map
= NULL
;
489 int inode_has_buffers(struct inode
*inode
)
491 return !list_empty(&inode
->i_data
.private_list
);
495 * osync is designed to support O_SYNC io. It waits synchronously for
496 * all already-submitted IO to complete, but does not queue any new
497 * writes to the disk.
499 * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
500 * you dirty the buffers, and then use osync_inode_buffers to wait for
501 * completion. Any other dirty buffers which are not yet queued for
502 * write will not be flushed to disk by the osync.
504 static int osync_buffers_list(spinlock_t
*lock
, struct list_head
*list
)
506 struct buffer_head
*bh
;
512 list_for_each_prev(p
, list
) {
514 if (buffer_locked(bh
)) {
518 if (!buffer_uptodate(bh
))
529 static void do_thaw_one(struct super_block
*sb
, void *unused
)
531 while (sb
->s_bdev
&& !thaw_bdev(sb
->s_bdev
, sb
))
532 printk(KERN_WARNING
"Emergency Thaw on %pg\n", sb
->s_bdev
);
535 static void do_thaw_all(struct work_struct
*work
)
537 iterate_supers(do_thaw_one
, NULL
);
539 printk(KERN_WARNING
"Emergency Thaw complete\n");
543 * emergency_thaw_all -- forcibly thaw every frozen filesystem
545 * Used for emergency unfreeze of all filesystems via SysRq
547 void emergency_thaw_all(void)
549 struct work_struct
*work
;
551 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
553 INIT_WORK(work
, do_thaw_all
);
559 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
560 * @mapping: the mapping which wants those buffers written
562 * Starts I/O against the buffers at mapping->private_list, and waits upon
565 * Basically, this is a convenience function for fsync().
566 * @mapping is a file or directory which needs those buffers to be written for
567 * a successful fsync().
569 int sync_mapping_buffers(struct address_space
*mapping
)
571 struct address_space
*buffer_mapping
= mapping
->private_data
;
573 if (buffer_mapping
== NULL
|| list_empty(&mapping
->private_list
))
576 return fsync_buffers_list(&buffer_mapping
->private_lock
,
577 &mapping
->private_list
);
579 EXPORT_SYMBOL(sync_mapping_buffers
);
582 * Called when we've recently written block `bblock', and it is known that
583 * `bblock' was for a buffer_boundary() buffer. This means that the block at
584 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
585 * dirty, schedule it for IO. So that indirects merge nicely with their data.
587 void write_boundary_block(struct block_device
*bdev
,
588 sector_t bblock
, unsigned blocksize
)
590 struct buffer_head
*bh
= __find_get_block(bdev
, bblock
+ 1, blocksize
);
592 if (buffer_dirty(bh
))
593 ll_rw_block(REQ_OP_WRITE
, 0, 1, &bh
);
598 void mark_buffer_dirty_inode(struct buffer_head
*bh
, struct inode
*inode
)
600 struct address_space
*mapping
= inode
->i_mapping
;
601 struct address_space
*buffer_mapping
= bh
->b_page
->mapping
;
603 mark_buffer_dirty(bh
);
604 if (!mapping
->private_data
) {
605 mapping
->private_data
= buffer_mapping
;
607 BUG_ON(mapping
->private_data
!= buffer_mapping
);
609 if (!bh
->b_assoc_map
) {
610 spin_lock(&buffer_mapping
->private_lock
);
611 list_move_tail(&bh
->b_assoc_buffers
,
612 &mapping
->private_list
);
613 bh
->b_assoc_map
= mapping
;
614 spin_unlock(&buffer_mapping
->private_lock
);
617 EXPORT_SYMBOL(mark_buffer_dirty_inode
);
620 * Mark the page dirty, and set it dirty in the radix tree, and mark the inode
623 * If warn is true, then emit a warning if the page is not uptodate and has
624 * not been truncated.
626 * The caller must hold lock_page_memcg().
628 static void __set_page_dirty(struct page
*page
, struct address_space
*mapping
,
633 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
634 if (page
->mapping
) { /* Race with truncate? */
635 WARN_ON_ONCE(warn
&& !PageUptodate(page
));
636 account_page_dirtied(page
, mapping
);
637 radix_tree_tag_set(&mapping
->page_tree
,
638 page_index(page
), PAGECACHE_TAG_DIRTY
);
640 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
644 * Add a page to the dirty page list.
646 * It is a sad fact of life that this function is called from several places
647 * deeply under spinlocking. It may not sleep.
649 * If the page has buffers, the uptodate buffers are set dirty, to preserve
650 * dirty-state coherency between the page and the buffers. It the page does
651 * not have buffers then when they are later attached they will all be set
654 * The buffers are dirtied before the page is dirtied. There's a small race
655 * window in which a writepage caller may see the page cleanness but not the
656 * buffer dirtiness. That's fine. If this code were to set the page dirty
657 * before the buffers, a concurrent writepage caller could clear the page dirty
658 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
659 * page on the dirty page list.
661 * We use private_lock to lock against try_to_free_buffers while using the
662 * page's buffer list. Also use this to protect against clean buffers being
663 * added to the page after it was set dirty.
665 * FIXME: may need to call ->reservepage here as well. That's rather up to the
666 * address_space though.
668 int __set_page_dirty_buffers(struct page
*page
)
671 struct address_space
*mapping
= page_mapping(page
);
673 if (unlikely(!mapping
))
674 return !TestSetPageDirty(page
);
676 spin_lock(&mapping
->private_lock
);
677 if (page_has_buffers(page
)) {
678 struct buffer_head
*head
= page_buffers(page
);
679 struct buffer_head
*bh
= head
;
682 set_buffer_dirty(bh
);
683 bh
= bh
->b_this_page
;
684 } while (bh
!= head
);
687 * Lock out page->mem_cgroup migration to keep PageDirty
688 * synchronized with per-memcg dirty page counters.
690 lock_page_memcg(page
);
691 newly_dirty
= !TestSetPageDirty(page
);
692 spin_unlock(&mapping
->private_lock
);
695 __set_page_dirty(page
, mapping
, 1);
697 unlock_page_memcg(page
);
700 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
704 EXPORT_SYMBOL(__set_page_dirty_buffers
);
707 * Write out and wait upon a list of buffers.
709 * We have conflicting pressures: we want to make sure that all
710 * initially dirty buffers get waited on, but that any subsequently
711 * dirtied buffers don't. After all, we don't want fsync to last
712 * forever if somebody is actively writing to the file.
714 * Do this in two main stages: first we copy dirty buffers to a
715 * temporary inode list, queueing the writes as we go. Then we clean
716 * up, waiting for those writes to complete.
718 * During this second stage, any subsequent updates to the file may end
719 * up refiling the buffer on the original inode's dirty list again, so
720 * there is a chance we will end up with a buffer queued for write but
721 * not yet completed on that list. So, as a final cleanup we go through
722 * the osync code to catch these locked, dirty buffers without requeuing
723 * any newly dirty buffers for write.
725 static int fsync_buffers_list(spinlock_t
*lock
, struct list_head
*list
)
727 struct buffer_head
*bh
;
728 struct list_head tmp
;
729 struct address_space
*mapping
;
731 struct blk_plug plug
;
733 INIT_LIST_HEAD(&tmp
);
734 blk_start_plug(&plug
);
737 while (!list_empty(list
)) {
738 bh
= BH_ENTRY(list
->next
);
739 mapping
= bh
->b_assoc_map
;
740 __remove_assoc_queue(bh
);
741 /* Avoid race with mark_buffer_dirty_inode() which does
742 * a lockless check and we rely on seeing the dirty bit */
744 if (buffer_dirty(bh
) || buffer_locked(bh
)) {
745 list_add(&bh
->b_assoc_buffers
, &tmp
);
746 bh
->b_assoc_map
= mapping
;
747 if (buffer_dirty(bh
)) {
751 * Ensure any pending I/O completes so that
752 * write_dirty_buffer() actually writes the
753 * current contents - it is a noop if I/O is
754 * still in flight on potentially older
757 write_dirty_buffer(bh
, REQ_SYNC
);
760 * Kick off IO for the previous mapping. Note
761 * that we will not run the very last mapping,
762 * wait_on_buffer() will do that for us
763 * through sync_buffer().
772 blk_finish_plug(&plug
);
775 while (!list_empty(&tmp
)) {
776 bh
= BH_ENTRY(tmp
.prev
);
778 mapping
= bh
->b_assoc_map
;
779 __remove_assoc_queue(bh
);
780 /* Avoid race with mark_buffer_dirty_inode() which does
781 * a lockless check and we rely on seeing the dirty bit */
783 if (buffer_dirty(bh
)) {
784 list_add(&bh
->b_assoc_buffers
,
785 &mapping
->private_list
);
786 bh
->b_assoc_map
= mapping
;
790 if (!buffer_uptodate(bh
))
797 err2
= osync_buffers_list(lock
, list
);
805 * Invalidate any and all dirty buffers on a given inode. We are
806 * probably unmounting the fs, but that doesn't mean we have already
807 * done a sync(). Just drop the buffers from the inode list.
809 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
810 * assumes that all the buffers are against the blockdev. Not true
813 void invalidate_inode_buffers(struct inode
*inode
)
815 if (inode_has_buffers(inode
)) {
816 struct address_space
*mapping
= &inode
->i_data
;
817 struct list_head
*list
= &mapping
->private_list
;
818 struct address_space
*buffer_mapping
= mapping
->private_data
;
820 spin_lock(&buffer_mapping
->private_lock
);
821 while (!list_empty(list
))
822 __remove_assoc_queue(BH_ENTRY(list
->next
));
823 spin_unlock(&buffer_mapping
->private_lock
);
826 EXPORT_SYMBOL(invalidate_inode_buffers
);
829 * Remove any clean buffers from the inode's buffer list. This is called
830 * when we're trying to free the inode itself. Those buffers can pin it.
832 * Returns true if all buffers were removed.
834 int remove_inode_buffers(struct inode
*inode
)
838 if (inode_has_buffers(inode
)) {
839 struct address_space
*mapping
= &inode
->i_data
;
840 struct list_head
*list
= &mapping
->private_list
;
841 struct address_space
*buffer_mapping
= mapping
->private_data
;
843 spin_lock(&buffer_mapping
->private_lock
);
844 while (!list_empty(list
)) {
845 struct buffer_head
*bh
= BH_ENTRY(list
->next
);
846 if (buffer_dirty(bh
)) {
850 __remove_assoc_queue(bh
);
852 spin_unlock(&buffer_mapping
->private_lock
);
858 * Create the appropriate buffers when given a page for data area and
859 * the size of each buffer.. Use the bh->b_this_page linked list to
860 * follow the buffers created. Return NULL if unable to create more
863 * The retry flag is used to differentiate async IO (paging, swapping)
864 * which may not fail from ordinary buffer allocations.
866 struct buffer_head
*alloc_page_buffers(struct page
*page
, unsigned long size
,
869 struct buffer_head
*bh
, *head
;
875 while ((offset
-= size
) >= 0) {
876 bh
= alloc_buffer_head(GFP_NOFS
);
880 bh
->b_this_page
= head
;
886 /* Link the buffer to its page */
887 set_bh_page(bh
, page
, offset
);
891 * In case anything failed, we just free everything we got.
897 head
= head
->b_this_page
;
898 free_buffer_head(bh
);
903 * Return failure for non-async IO requests. Async IO requests
904 * are not allowed to fail, so we have to wait until buffer heads
905 * become available. But we don't want tasks sleeping with
906 * partially complete buffers, so all were released above.
911 /* We're _really_ low on memory. Now we just
912 * wait for old buffer heads to become free due to
913 * finishing IO. Since this is an async request and
914 * the reserve list is empty, we're sure there are
915 * async buffer heads in use.
920 EXPORT_SYMBOL_GPL(alloc_page_buffers
);
923 link_dev_buffers(struct page
*page
, struct buffer_head
*head
)
925 struct buffer_head
*bh
, *tail
;
930 bh
= bh
->b_this_page
;
932 tail
->b_this_page
= head
;
933 attach_page_buffers(page
, head
);
936 static sector_t
blkdev_max_block(struct block_device
*bdev
, unsigned int size
)
938 sector_t retval
= ~((sector_t
)0);
939 loff_t sz
= i_size_read(bdev
->bd_inode
);
942 unsigned int sizebits
= blksize_bits(size
);
943 retval
= (sz
>> sizebits
);
949 * Initialise the state of a blockdev page's buffers.
952 init_page_buffers(struct page
*page
, struct block_device
*bdev
,
953 sector_t block
, int size
)
955 struct buffer_head
*head
= page_buffers(page
);
956 struct buffer_head
*bh
= head
;
957 int uptodate
= PageUptodate(page
);
958 sector_t end_block
= blkdev_max_block(I_BDEV(bdev
->bd_inode
), size
);
961 if (!buffer_mapped(bh
)) {
962 init_buffer(bh
, NULL
, NULL
);
964 bh
->b_blocknr
= block
;
966 set_buffer_uptodate(bh
);
967 if (block
< end_block
)
968 set_buffer_mapped(bh
);
971 bh
= bh
->b_this_page
;
972 } while (bh
!= head
);
975 * Caller needs to validate requested block against end of device.
981 * Create the page-cache page that contains the requested block.
983 * This is used purely for blockdev mappings.
986 grow_dev_page(struct block_device
*bdev
, sector_t block
,
987 pgoff_t index
, int size
, int sizebits
, gfp_t gfp
)
989 struct inode
*inode
= bdev
->bd_inode
;
991 struct buffer_head
*bh
;
993 int ret
= 0; /* Will call free_more_memory() */
996 gfp_mask
= mapping_gfp_constraint(inode
->i_mapping
, ~__GFP_FS
) | gfp
;
999 * XXX: __getblk_slow() can not really deal with failure and
1000 * will endlessly loop on improvised global reclaim. Prefer
1001 * looping in the allocator rather than here, at least that
1002 * code knows what it's doing.
1004 gfp_mask
|= __GFP_NOFAIL
;
1006 page
= find_or_create_page(inode
->i_mapping
, index
, gfp_mask
);
1010 BUG_ON(!PageLocked(page
));
1012 if (page_has_buffers(page
)) {
1013 bh
= page_buffers(page
);
1014 if (bh
->b_size
== size
) {
1015 end_block
= init_page_buffers(page
, bdev
,
1016 (sector_t
)index
<< sizebits
,
1020 if (!try_to_free_buffers(page
))
1025 * Allocate some buffers for this page
1027 bh
= alloc_page_buffers(page
, size
, 0);
1032 * Link the page to the buffers and initialise them. Take the
1033 * lock to be atomic wrt __find_get_block(), which does not
1034 * run under the page lock.
1036 spin_lock(&inode
->i_mapping
->private_lock
);
1037 link_dev_buffers(page
, bh
);
1038 end_block
= init_page_buffers(page
, bdev
, (sector_t
)index
<< sizebits
,
1040 spin_unlock(&inode
->i_mapping
->private_lock
);
1042 ret
= (block
< end_block
) ? 1 : -ENXIO
;
1050 * Create buffers for the specified block device block's page. If
1051 * that page was dirty, the buffers are set dirty also.
1054 grow_buffers(struct block_device
*bdev
, sector_t block
, int size
, gfp_t gfp
)
1062 } while ((size
<< sizebits
) < PAGE_SIZE
);
1064 index
= block
>> sizebits
;
1067 * Check for a block which wants to lie outside our maximum possible
1068 * pagecache index. (this comparison is done using sector_t types).
1070 if (unlikely(index
!= block
>> sizebits
)) {
1071 printk(KERN_ERR
"%s: requested out-of-range block %llu for "
1073 __func__
, (unsigned long long)block
,
1078 /* Create a page with the proper size buffers.. */
1079 return grow_dev_page(bdev
, block
, index
, size
, sizebits
, gfp
);
1082 static struct buffer_head
*
1083 __getblk_slow(struct block_device
*bdev
, sector_t block
,
1084 unsigned size
, gfp_t gfp
)
1086 /* Size must be multiple of hard sectorsize */
1087 if (unlikely(size
& (bdev_logical_block_size(bdev
)-1) ||
1088 (size
< 512 || size
> PAGE_SIZE
))) {
1089 printk(KERN_ERR
"getblk(): invalid block size %d requested\n",
1091 printk(KERN_ERR
"logical block size: %d\n",
1092 bdev_logical_block_size(bdev
));
1099 struct buffer_head
*bh
;
1102 bh
= __find_get_block(bdev
, block
, size
);
1106 ret
= grow_buffers(bdev
, block
, size
, gfp
);
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
;
1171 lock_page_memcg(page
);
1172 if (!TestSetPageDirty(page
)) {
1173 mapping
= page_mapping(page
);
1175 __set_page_dirty(page
, mapping
, 0);
1177 unlock_page_memcg(page
);
1179 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
1182 EXPORT_SYMBOL(mark_buffer_dirty
);
1185 * Decrement a buffer_head's reference count. If all buffers against a page
1186 * have zero reference count, are clean and unlocked, and if the page is clean
1187 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1188 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1189 * a page but it ends up not being freed, and buffers may later be reattached).
1191 void __brelse(struct buffer_head
* buf
)
1193 if (atomic_read(&buf
->b_count
)) {
1197 WARN(1, KERN_ERR
"VFS: brelse: Trying to free free buffer\n");
1199 EXPORT_SYMBOL(__brelse
);
1202 * bforget() is like brelse(), except it discards any
1203 * potentially dirty data.
1205 void __bforget(struct buffer_head
*bh
)
1207 clear_buffer_dirty(bh
);
1208 if (bh
->b_assoc_map
) {
1209 struct address_space
*buffer_mapping
= bh
->b_page
->mapping
;
1211 spin_lock(&buffer_mapping
->private_lock
);
1212 list_del_init(&bh
->b_assoc_buffers
);
1213 bh
->b_assoc_map
= NULL
;
1214 spin_unlock(&buffer_mapping
->private_lock
);
1218 EXPORT_SYMBOL(__bforget
);
1220 static struct buffer_head
*__bread_slow(struct buffer_head
*bh
)
1223 if (buffer_uptodate(bh
)) {
1228 bh
->b_end_io
= end_buffer_read_sync
;
1229 submit_bh(REQ_OP_READ
, 0, bh
);
1231 if (buffer_uptodate(bh
))
1239 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1240 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1241 * refcount elevated by one when they're in an LRU. A buffer can only appear
1242 * once in a particular CPU's LRU. A single buffer can be present in multiple
1243 * CPU's LRUs at the same time.
1245 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1246 * sb_find_get_block().
1248 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1249 * a local interrupt disable for that.
1252 #define BH_LRU_SIZE 16
1255 struct buffer_head
*bhs
[BH_LRU_SIZE
];
1258 static DEFINE_PER_CPU(struct bh_lru
, bh_lrus
) = {{ NULL
}};
1261 #define bh_lru_lock() local_irq_disable()
1262 #define bh_lru_unlock() local_irq_enable()
1264 #define bh_lru_lock() preempt_disable()
1265 #define bh_lru_unlock() preempt_enable()
1268 static inline void check_irqs_on(void)
1270 #ifdef irqs_disabled
1271 BUG_ON(irqs_disabled());
1276 * The LRU management algorithm is dopey-but-simple. Sorry.
1278 static void bh_lru_install(struct buffer_head
*bh
)
1280 struct buffer_head
*evictee
= NULL
;
1284 if (__this_cpu_read(bh_lrus
.bhs
[0]) != bh
) {
1285 struct buffer_head
*bhs
[BH_LRU_SIZE
];
1291 for (in
= 0; in
< BH_LRU_SIZE
; in
++) {
1292 struct buffer_head
*bh2
=
1293 __this_cpu_read(bh_lrus
.bhs
[in
]);
1298 if (out
>= BH_LRU_SIZE
) {
1299 BUG_ON(evictee
!= NULL
);
1306 while (out
< BH_LRU_SIZE
)
1308 memcpy(this_cpu_ptr(&bh_lrus
.bhs
), bhs
, sizeof(bhs
));
1317 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1319 static struct buffer_head
*
1320 lookup_bh_lru(struct block_device
*bdev
, sector_t block
, unsigned size
)
1322 struct buffer_head
*ret
= NULL
;
1327 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1328 struct buffer_head
*bh
= __this_cpu_read(bh_lrus
.bhs
[i
]);
1330 if (bh
&& bh
->b_blocknr
== block
&& bh
->b_bdev
== bdev
&&
1331 bh
->b_size
== size
) {
1334 __this_cpu_write(bh_lrus
.bhs
[i
],
1335 __this_cpu_read(bh_lrus
.bhs
[i
- 1]));
1338 __this_cpu_write(bh_lrus
.bhs
[0], bh
);
1350 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1351 * it in the LRU and mark it as accessed. If it is not present then return
1354 struct buffer_head
*
1355 __find_get_block(struct block_device
*bdev
, sector_t block
, unsigned size
)
1357 struct buffer_head
*bh
= lookup_bh_lru(bdev
, block
, size
);
1360 /* __find_get_block_slow will mark the page accessed */
1361 bh
= __find_get_block_slow(bdev
, block
);
1369 EXPORT_SYMBOL(__find_get_block
);
1372 * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
1373 * which corresponds to the passed block_device, block and size. The
1374 * returned buffer has its reference count incremented.
1376 * __getblk_gfp() will lock up the machine if grow_dev_page's
1377 * try_to_free_buffers() attempt is failing. FIXME, perhaps?
1379 struct buffer_head
*
1380 __getblk_gfp(struct block_device
*bdev
, sector_t block
,
1381 unsigned size
, gfp_t gfp
)
1383 struct buffer_head
*bh
= __find_get_block(bdev
, block
, size
);
1387 bh
= __getblk_slow(bdev
, block
, size
, gfp
);
1390 EXPORT_SYMBOL(__getblk_gfp
);
1393 * Do async read-ahead on a buffer..
1395 void __breadahead(struct block_device
*bdev
, sector_t block
, unsigned size
)
1397 struct buffer_head
*bh
= __getblk(bdev
, block
, size
);
1399 ll_rw_block(REQ_OP_READ
, REQ_RAHEAD
, 1, &bh
);
1403 EXPORT_SYMBOL(__breadahead
);
1406 * __bread_gfp() - reads a specified block and returns the bh
1407 * @bdev: the block_device to read from
1408 * @block: number of block
1409 * @size: size (in bytes) to read
1410 * @gfp: page allocation flag
1412 * Reads a specified block, and returns buffer head that contains it.
1413 * The page cache can be allocated from non-movable area
1414 * not to prevent page migration if you set gfp to zero.
1415 * It returns NULL if the block was unreadable.
1417 struct buffer_head
*
1418 __bread_gfp(struct block_device
*bdev
, sector_t block
,
1419 unsigned size
, gfp_t gfp
)
1421 struct buffer_head
*bh
= __getblk_gfp(bdev
, block
, size
, gfp
);
1423 if (likely(bh
) && !buffer_uptodate(bh
))
1424 bh
= __bread_slow(bh
);
1427 EXPORT_SYMBOL(__bread_gfp
);
1430 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1431 * This doesn't race because it runs in each cpu either in irq
1432 * or with preempt disabled.
1434 static void invalidate_bh_lru(void *arg
)
1436 struct bh_lru
*b
= &get_cpu_var(bh_lrus
);
1439 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1443 put_cpu_var(bh_lrus
);
1446 static bool has_bh_in_lru(int cpu
, void *dummy
)
1448 struct bh_lru
*b
= per_cpu_ptr(&bh_lrus
, cpu
);
1451 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1459 void invalidate_bh_lrus(void)
1461 on_each_cpu_cond(has_bh_in_lru
, invalidate_bh_lru
, NULL
, 1, GFP_KERNEL
);
1463 EXPORT_SYMBOL_GPL(invalidate_bh_lrus
);
1465 void set_bh_page(struct buffer_head
*bh
,
1466 struct page
*page
, unsigned long offset
)
1469 BUG_ON(offset
>= PAGE_SIZE
);
1470 if (PageHighMem(page
))
1472 * This catches illegal uses and preserves the offset:
1474 bh
->b_data
= (char *)(0 + offset
);
1476 bh
->b_data
= page_address(page
) + offset
;
1478 EXPORT_SYMBOL(set_bh_page
);
1481 * Called when truncating a buffer on a page completely.
1484 /* Bits that are cleared during an invalidate */
1485 #define BUFFER_FLAGS_DISCARD \
1486 (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1487 1 << BH_Delay | 1 << BH_Unwritten)
1489 static void discard_buffer(struct buffer_head
* bh
)
1491 unsigned long b_state
, b_state_old
;
1494 clear_buffer_dirty(bh
);
1496 b_state
= bh
->b_state
;
1498 b_state_old
= cmpxchg(&bh
->b_state
, b_state
,
1499 (b_state
& ~BUFFER_FLAGS_DISCARD
));
1500 if (b_state_old
== b_state
)
1502 b_state
= b_state_old
;
1508 * block_invalidatepage - invalidate part or all of a buffer-backed page
1510 * @page: the page which is affected
1511 * @offset: start of the range to invalidate
1512 * @length: length of the range to invalidate
1514 * block_invalidatepage() is called when all or part of the page has become
1515 * invalidated by a truncate operation.
1517 * block_invalidatepage() does not have to release all buffers, but it must
1518 * ensure that no dirty buffer is left outside @offset and that no I/O
1519 * is underway against any of the blocks which are outside the truncation
1520 * point. Because the caller is about to free (and possibly reuse) those
1523 void block_invalidatepage(struct page
*page
, unsigned int offset
,
1524 unsigned int length
)
1526 struct buffer_head
*head
, *bh
, *next
;
1527 unsigned int curr_off
= 0;
1528 unsigned int stop
= length
+ offset
;
1530 BUG_ON(!PageLocked(page
));
1531 if (!page_has_buffers(page
))
1535 * Check for overflow
1537 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1539 head
= page_buffers(page
);
1542 unsigned int next_off
= curr_off
+ bh
->b_size
;
1543 next
= bh
->b_this_page
;
1546 * Are we still fully in range ?
1548 if (next_off
> stop
)
1552 * is this block fully invalidated?
1554 if (offset
<= curr_off
)
1556 curr_off
= next_off
;
1558 } while (bh
!= head
);
1561 * We release buffers only if the entire page is being invalidated.
1562 * The get_block cached value has been unconditionally invalidated,
1563 * so real IO is not possible anymore.
1566 try_to_release_page(page
, 0);
1570 EXPORT_SYMBOL(block_invalidatepage
);
1574 * We attach and possibly dirty the buffers atomically wrt
1575 * __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1576 * is already excluded via the page lock.
1578 void create_empty_buffers(struct page
*page
,
1579 unsigned long blocksize
, unsigned long b_state
)
1581 struct buffer_head
*bh
, *head
, *tail
;
1583 head
= alloc_page_buffers(page
, blocksize
, 1);
1586 bh
->b_state
|= b_state
;
1588 bh
= bh
->b_this_page
;
1590 tail
->b_this_page
= head
;
1592 spin_lock(&page
->mapping
->private_lock
);
1593 if (PageUptodate(page
) || PageDirty(page
)) {
1596 if (PageDirty(page
))
1597 set_buffer_dirty(bh
);
1598 if (PageUptodate(page
))
1599 set_buffer_uptodate(bh
);
1600 bh
= bh
->b_this_page
;
1601 } while (bh
!= head
);
1603 attach_page_buffers(page
, head
);
1604 spin_unlock(&page
->mapping
->private_lock
);
1606 EXPORT_SYMBOL(create_empty_buffers
);
1609 * clean_bdev_aliases: clean a range of buffers in block device
1610 * @bdev: Block device to clean buffers in
1611 * @block: Start of a range of blocks to clean
1612 * @len: Number of blocks to clean
1614 * We are taking a range of blocks for data and we don't want writeback of any
1615 * buffer-cache aliases starting from return from this function and until the
1616 * moment when something will explicitly mark the buffer dirty (hopefully that
1617 * will not happen until we will free that block ;-) We don't even need to mark
1618 * it not-uptodate - nobody can expect anything from a newly allocated buffer
1619 * anyway. We used to use unmap_buffer() for such invalidation, but that was
1620 * wrong. We definitely don't want to mark the alias unmapped, for example - it
1621 * would confuse anyone who might pick it with bread() afterwards...
1623 * Also.. Note that bforget() doesn't lock the buffer. So there can be
1624 * writeout I/O going on against recently-freed buffers. We don't wait on that
1625 * I/O in bforget() - it's more efficient to wait on the I/O only if we really
1626 * need to. That happens here.
1628 void clean_bdev_aliases(struct block_device
*bdev
, sector_t block
, sector_t len
)
1630 struct inode
*bd_inode
= bdev
->bd_inode
;
1631 struct address_space
*bd_mapping
= bd_inode
->i_mapping
;
1632 struct pagevec pvec
;
1633 pgoff_t index
= block
>> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
1636 struct buffer_head
*bh
;
1637 struct buffer_head
*head
;
1639 end
= (block
+ len
- 1) >> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
1640 pagevec_init(&pvec
, 0);
1641 while (index
<= end
&& pagevec_lookup(&pvec
, bd_mapping
, index
,
1642 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1)) {
1643 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
1644 struct page
*page
= pvec
.pages
[i
];
1646 index
= page
->index
;
1649 if (!page_has_buffers(page
))
1652 * We use page lock instead of bd_mapping->private_lock
1653 * to pin buffers here since we can afford to sleep and
1654 * it scales better than a global spinlock lock.
1657 /* Recheck when the page is locked which pins bhs */
1658 if (!page_has_buffers(page
))
1660 head
= page_buffers(page
);
1663 if (!buffer_mapped(bh
) || (bh
->b_blocknr
< block
))
1665 if (bh
->b_blocknr
>= block
+ len
)
1667 clear_buffer_dirty(bh
);
1669 clear_buffer_req(bh
);
1671 bh
= bh
->b_this_page
;
1672 } while (bh
!= head
);
1676 pagevec_release(&pvec
);
1681 EXPORT_SYMBOL(clean_bdev_aliases
);
1684 * Size is a power-of-two in the range 512..PAGE_SIZE,
1685 * and the case we care about most is PAGE_SIZE.
1687 * So this *could* possibly be written with those
1688 * constraints in mind (relevant mostly if some
1689 * architecture has a slow bit-scan instruction)
1691 static inline int block_size_bits(unsigned int blocksize
)
1693 return ilog2(blocksize
);
1696 static struct buffer_head
*create_page_buffers(struct page
*page
, struct inode
*inode
, unsigned int b_state
)
1698 BUG_ON(!PageLocked(page
));
1700 if (!page_has_buffers(page
))
1701 create_empty_buffers(page
, 1 << ACCESS_ONCE(inode
->i_blkbits
), b_state
);
1702 return page_buffers(page
);
1706 * NOTE! All mapped/uptodate combinations are valid:
1708 * Mapped Uptodate Meaning
1710 * No No "unknown" - must do get_block()
1711 * No Yes "hole" - zero-filled
1712 * Yes No "allocated" - allocated on disk, not read in
1713 * Yes Yes "valid" - allocated and up-to-date in memory.
1715 * "Dirty" is valid only with the last case (mapped+uptodate).
1719 * While block_write_full_page is writing back the dirty buffers under
1720 * the page lock, whoever dirtied the buffers may decide to clean them
1721 * again at any time. We handle that by only looking at the buffer
1722 * state inside lock_buffer().
1724 * If block_write_full_page() is called for regular writeback
1725 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1726 * locked buffer. This only can happen if someone has written the buffer
1727 * directly, with submit_bh(). At the address_space level PageWriteback
1728 * prevents this contention from occurring.
1730 * If block_write_full_page() is called with wbc->sync_mode ==
1731 * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this
1732 * causes the writes to be flagged as synchronous writes.
1734 int __block_write_full_page(struct inode
*inode
, struct page
*page
,
1735 get_block_t
*get_block
, struct writeback_control
*wbc
,
1736 bh_end_io_t
*handler
)
1740 sector_t last_block
;
1741 struct buffer_head
*bh
, *head
;
1742 unsigned int blocksize
, bbits
;
1743 int nr_underway
= 0;
1744 int write_flags
= wbc_to_write_flags(wbc
);
1746 head
= create_page_buffers(page
, inode
,
1747 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
1750 * Be very careful. We have no exclusion from __set_page_dirty_buffers
1751 * here, and the (potentially unmapped) buffers may become dirty at
1752 * any time. If a buffer becomes dirty here after we've inspected it
1753 * then we just miss that fact, and the page stays dirty.
1755 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1756 * handle that here by just cleaning them.
1760 blocksize
= bh
->b_size
;
1761 bbits
= block_size_bits(blocksize
);
1763 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1764 last_block
= (i_size_read(inode
) - 1) >> bbits
;
1767 * Get all the dirty buffers mapped to disk addresses and
1768 * handle any aliases from the underlying blockdev's mapping.
1771 if (block
> last_block
) {
1773 * mapped buffers outside i_size will occur, because
1774 * this page can be outside i_size when there is a
1775 * truncate in progress.
1778 * The buffer was zeroed by block_write_full_page()
1780 clear_buffer_dirty(bh
);
1781 set_buffer_uptodate(bh
);
1782 } else if ((!buffer_mapped(bh
) || buffer_delay(bh
)) &&
1784 WARN_ON(bh
->b_size
!= blocksize
);
1785 err
= get_block(inode
, block
, bh
, 1);
1788 clear_buffer_delay(bh
);
1789 if (buffer_new(bh
)) {
1790 /* blockdev mappings never come here */
1791 clear_buffer_new(bh
);
1792 clean_bdev_bh_alias(bh
);
1795 bh
= bh
->b_this_page
;
1797 } while (bh
!= head
);
1800 if (!buffer_mapped(bh
))
1803 * If it's a fully non-blocking write attempt and we cannot
1804 * lock the buffer then redirty the page. Note that this can
1805 * potentially cause a busy-wait loop from writeback threads
1806 * and kswapd activity, but those code paths have their own
1807 * higher-level throttling.
1809 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
1811 } else if (!trylock_buffer(bh
)) {
1812 redirty_page_for_writepage(wbc
, page
);
1815 if (test_clear_buffer_dirty(bh
)) {
1816 mark_buffer_async_write_endio(bh
, handler
);
1820 } while ((bh
= bh
->b_this_page
) != head
);
1823 * The page and its buffers are protected by PageWriteback(), so we can
1824 * drop the bh refcounts early.
1826 BUG_ON(PageWriteback(page
));
1827 set_page_writeback(page
);
1830 struct buffer_head
*next
= bh
->b_this_page
;
1831 if (buffer_async_write(bh
)) {
1832 submit_bh_wbc(REQ_OP_WRITE
, write_flags
, bh
,
1833 inode
->i_write_hint
, wbc
);
1837 } while (bh
!= head
);
1842 if (nr_underway
== 0) {
1844 * The page was marked dirty, but the buffers were
1845 * clean. Someone wrote them back by hand with
1846 * ll_rw_block/submit_bh. A rare case.
1848 end_page_writeback(page
);
1851 * The page and buffer_heads can be released at any time from
1859 * ENOSPC, or some other error. We may already have added some
1860 * blocks to the file, so we need to write these out to avoid
1861 * exposing stale data.
1862 * The page is currently locked and not marked for writeback
1865 /* Recovery: lock and submit the mapped buffers */
1867 if (buffer_mapped(bh
) && buffer_dirty(bh
) &&
1868 !buffer_delay(bh
)) {
1870 mark_buffer_async_write_endio(bh
, handler
);
1873 * The buffer may have been set dirty during
1874 * attachment to a dirty page.
1876 clear_buffer_dirty(bh
);
1878 } while ((bh
= bh
->b_this_page
) != head
);
1880 BUG_ON(PageWriteback(page
));
1881 mapping_set_error(page
->mapping
, err
);
1882 set_page_writeback(page
);
1884 struct buffer_head
*next
= bh
->b_this_page
;
1885 if (buffer_async_write(bh
)) {
1886 clear_buffer_dirty(bh
);
1887 submit_bh_wbc(REQ_OP_WRITE
, write_flags
, bh
,
1888 inode
->i_write_hint
, wbc
);
1892 } while (bh
!= head
);
1896 EXPORT_SYMBOL(__block_write_full_page
);
1899 * If a page has any new buffers, zero them out here, and mark them uptodate
1900 * and dirty so they'll be written out (in order to prevent uninitialised
1901 * block data from leaking). And clear the new bit.
1903 void page_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1905 unsigned int block_start
, block_end
;
1906 struct buffer_head
*head
, *bh
;
1908 BUG_ON(!PageLocked(page
));
1909 if (!page_has_buffers(page
))
1912 bh
= head
= page_buffers(page
);
1915 block_end
= block_start
+ bh
->b_size
;
1917 if (buffer_new(bh
)) {
1918 if (block_end
> from
&& block_start
< to
) {
1919 if (!PageUptodate(page
)) {
1920 unsigned start
, size
;
1922 start
= max(from
, block_start
);
1923 size
= min(to
, block_end
) - start
;
1925 zero_user(page
, start
, size
);
1926 set_buffer_uptodate(bh
);
1929 clear_buffer_new(bh
);
1930 mark_buffer_dirty(bh
);
1934 block_start
= block_end
;
1935 bh
= bh
->b_this_page
;
1936 } while (bh
!= head
);
1938 EXPORT_SYMBOL(page_zero_new_buffers
);
1941 iomap_to_bh(struct inode
*inode
, sector_t block
, struct buffer_head
*bh
,
1942 struct iomap
*iomap
)
1944 loff_t offset
= block
<< inode
->i_blkbits
;
1946 bh
->b_bdev
= iomap
->bdev
;
1949 * Block points to offset in file we need to map, iomap contains
1950 * the offset at which the map starts. If the map ends before the
1951 * current block, then do not map the buffer and let the caller
1954 BUG_ON(offset
>= iomap
->offset
+ iomap
->length
);
1956 switch (iomap
->type
) {
1959 * If the buffer is not up to date or beyond the current EOF,
1960 * we need to mark it as new to ensure sub-block zeroing is
1961 * executed if necessary.
1963 if (!buffer_uptodate(bh
) ||
1964 (offset
>= i_size_read(inode
)))
1967 case IOMAP_DELALLOC
:
1968 if (!buffer_uptodate(bh
) ||
1969 (offset
>= i_size_read(inode
)))
1971 set_buffer_uptodate(bh
);
1972 set_buffer_mapped(bh
);
1973 set_buffer_delay(bh
);
1975 case IOMAP_UNWRITTEN
:
1977 * For unwritten regions, we always need to ensure that
1978 * sub-block writes cause the regions in the block we are not
1979 * writing to are zeroed. Set the buffer as new to ensure this.
1982 set_buffer_unwritten(bh
);
1985 if (offset
>= i_size_read(inode
))
1987 bh
->b_blocknr
= (iomap
->blkno
>> (inode
->i_blkbits
- 9)) +
1988 ((offset
- iomap
->offset
) >> inode
->i_blkbits
);
1989 set_buffer_mapped(bh
);
1994 int __block_write_begin_int(struct page
*page
, loff_t pos
, unsigned len
,
1995 get_block_t
*get_block
, struct iomap
*iomap
)
1997 unsigned from
= pos
& (PAGE_SIZE
- 1);
1998 unsigned to
= from
+ len
;
1999 struct inode
*inode
= page
->mapping
->host
;
2000 unsigned block_start
, block_end
;
2003 unsigned blocksize
, bbits
;
2004 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
=wait
;
2006 BUG_ON(!PageLocked(page
));
2007 BUG_ON(from
> PAGE_SIZE
);
2008 BUG_ON(to
> PAGE_SIZE
);
2011 head
= create_page_buffers(page
, inode
, 0);
2012 blocksize
= head
->b_size
;
2013 bbits
= block_size_bits(blocksize
);
2015 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
2017 for(bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
2018 block
++, block_start
=block_end
, bh
= bh
->b_this_page
) {
2019 block_end
= block_start
+ blocksize
;
2020 if (block_end
<= from
|| block_start
>= to
) {
2021 if (PageUptodate(page
)) {
2022 if (!buffer_uptodate(bh
))
2023 set_buffer_uptodate(bh
);
2028 clear_buffer_new(bh
);
2029 if (!buffer_mapped(bh
)) {
2030 WARN_ON(bh
->b_size
!= blocksize
);
2032 err
= get_block(inode
, block
, bh
, 1);
2036 iomap_to_bh(inode
, block
, bh
, iomap
);
2039 if (buffer_new(bh
)) {
2040 clean_bdev_bh_alias(bh
);
2041 if (PageUptodate(page
)) {
2042 clear_buffer_new(bh
);
2043 set_buffer_uptodate(bh
);
2044 mark_buffer_dirty(bh
);
2047 if (block_end
> to
|| block_start
< from
)
2048 zero_user_segments(page
,
2054 if (PageUptodate(page
)) {
2055 if (!buffer_uptodate(bh
))
2056 set_buffer_uptodate(bh
);
2059 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
2060 !buffer_unwritten(bh
) &&
2061 (block_start
< from
|| block_end
> to
)) {
2062 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
2067 * If we issued read requests - let them complete.
2069 while(wait_bh
> wait
) {
2070 wait_on_buffer(*--wait_bh
);
2071 if (!buffer_uptodate(*wait_bh
))
2075 page_zero_new_buffers(page
, from
, to
);
2079 int __block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
2080 get_block_t
*get_block
)
2082 return __block_write_begin_int(page
, pos
, len
, get_block
, NULL
);
2084 EXPORT_SYMBOL(__block_write_begin
);
2086 static int __block_commit_write(struct inode
*inode
, struct page
*page
,
2087 unsigned from
, unsigned to
)
2089 unsigned block_start
, block_end
;
2092 struct buffer_head
*bh
, *head
;
2094 bh
= head
= page_buffers(page
);
2095 blocksize
= bh
->b_size
;
2099 block_end
= block_start
+ blocksize
;
2100 if (block_end
<= from
|| block_start
>= to
) {
2101 if (!buffer_uptodate(bh
))
2104 set_buffer_uptodate(bh
);
2105 mark_buffer_dirty(bh
);
2107 clear_buffer_new(bh
);
2109 block_start
= block_end
;
2110 bh
= bh
->b_this_page
;
2111 } while (bh
!= head
);
2114 * If this is a partial write which happened to make all buffers
2115 * uptodate then we can optimize away a bogus readpage() for
2116 * the next read(). Here we 'discover' whether the page went
2117 * uptodate as a result of this (potentially partial) write.
2120 SetPageUptodate(page
);
2125 * block_write_begin takes care of the basic task of block allocation and
2126 * bringing partial write blocks uptodate first.
2128 * The filesystem needs to handle block truncation upon failure.
2130 int block_write_begin(struct address_space
*mapping
, loff_t pos
, unsigned len
,
2131 unsigned flags
, struct page
**pagep
, get_block_t
*get_block
)
2133 pgoff_t index
= pos
>> PAGE_SHIFT
;
2137 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2141 status
= __block_write_begin(page
, pos
, len
, get_block
);
2142 if (unlikely(status
)) {
2151 EXPORT_SYMBOL(block_write_begin
);
2153 int block_write_end(struct file
*file
, struct address_space
*mapping
,
2154 loff_t pos
, unsigned len
, unsigned copied
,
2155 struct page
*page
, void *fsdata
)
2157 struct inode
*inode
= mapping
->host
;
2160 start
= pos
& (PAGE_SIZE
- 1);
2162 if (unlikely(copied
< len
)) {
2164 * The buffers that were written will now be uptodate, so we
2165 * don't have to worry about a readpage reading them and
2166 * overwriting a partial write. However if we have encountered
2167 * a short write and only partially written into a buffer, it
2168 * will not be marked uptodate, so a readpage might come in and
2169 * destroy our partial write.
2171 * Do the simplest thing, and just treat any short write to a
2172 * non uptodate page as a zero-length write, and force the
2173 * caller to redo the whole thing.
2175 if (!PageUptodate(page
))
2178 page_zero_new_buffers(page
, start
+copied
, start
+len
);
2180 flush_dcache_page(page
);
2182 /* This could be a short (even 0-length) commit */
2183 __block_commit_write(inode
, page
, start
, start
+copied
);
2187 EXPORT_SYMBOL(block_write_end
);
2189 int generic_write_end(struct file
*file
, struct address_space
*mapping
,
2190 loff_t pos
, unsigned len
, unsigned copied
,
2191 struct page
*page
, void *fsdata
)
2193 struct inode
*inode
= mapping
->host
;
2194 loff_t old_size
= inode
->i_size
;
2195 int i_size_changed
= 0;
2197 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2200 * No need to use i_size_read() here, the i_size
2201 * cannot change under us because we hold i_mutex.
2203 * But it's important to update i_size while still holding page lock:
2204 * page writeout could otherwise come in and zero beyond i_size.
2206 if (pos
+copied
> inode
->i_size
) {
2207 i_size_write(inode
, pos
+copied
);
2215 pagecache_isize_extended(inode
, old_size
, pos
);
2217 * Don't mark the inode dirty under page lock. First, it unnecessarily
2218 * makes the holding time of page lock longer. Second, it forces lock
2219 * ordering of page lock and transaction start for journaling
2223 mark_inode_dirty(inode
);
2227 EXPORT_SYMBOL(generic_write_end
);
2230 * block_is_partially_uptodate checks whether buffers within a page are
2233 * Returns true if all buffers which correspond to a file portion
2234 * we want to read are uptodate.
2236 int block_is_partially_uptodate(struct page
*page
, unsigned long from
,
2237 unsigned long count
)
2239 unsigned block_start
, block_end
, blocksize
;
2241 struct buffer_head
*bh
, *head
;
2244 if (!page_has_buffers(page
))
2247 head
= page_buffers(page
);
2248 blocksize
= head
->b_size
;
2249 to
= min_t(unsigned, PAGE_SIZE
- from
, count
);
2251 if (from
< blocksize
&& to
> PAGE_SIZE
- blocksize
)
2257 block_end
= block_start
+ blocksize
;
2258 if (block_end
> from
&& block_start
< to
) {
2259 if (!buffer_uptodate(bh
)) {
2263 if (block_end
>= to
)
2266 block_start
= block_end
;
2267 bh
= bh
->b_this_page
;
2268 } while (bh
!= head
);
2272 EXPORT_SYMBOL(block_is_partially_uptodate
);
2275 * Generic "read page" function for block devices that have the normal
2276 * get_block functionality. This is most of the block device filesystems.
2277 * Reads the page asynchronously --- the unlock_buffer() and
2278 * set/clear_buffer_uptodate() functions propagate buffer state into the
2279 * page struct once IO has completed.
2281 int block_read_full_page(struct page
*page
, get_block_t
*get_block
)
2283 struct inode
*inode
= page
->mapping
->host
;
2284 sector_t iblock
, lblock
;
2285 struct buffer_head
*bh
, *head
, *arr
[MAX_BUF_PER_PAGE
];
2286 unsigned int blocksize
, bbits
;
2288 int fully_mapped
= 1;
2290 head
= create_page_buffers(page
, inode
, 0);
2291 blocksize
= head
->b_size
;
2292 bbits
= block_size_bits(blocksize
);
2294 iblock
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
2295 lblock
= (i_size_read(inode
)+blocksize
-1) >> bbits
;
2301 if (buffer_uptodate(bh
))
2304 if (!buffer_mapped(bh
)) {
2308 if (iblock
< lblock
) {
2309 WARN_ON(bh
->b_size
!= blocksize
);
2310 err
= get_block(inode
, iblock
, bh
, 0);
2314 if (!buffer_mapped(bh
)) {
2315 zero_user(page
, i
* blocksize
, blocksize
);
2317 set_buffer_uptodate(bh
);
2321 * get_block() might have updated the buffer
2324 if (buffer_uptodate(bh
))
2328 } while (i
++, iblock
++, (bh
= bh
->b_this_page
) != head
);
2331 SetPageMappedToDisk(page
);
2335 * All buffers are uptodate - we can set the page uptodate
2336 * as well. But not if get_block() returned an error.
2338 if (!PageError(page
))
2339 SetPageUptodate(page
);
2344 /* Stage two: lock the buffers */
2345 for (i
= 0; i
< nr
; i
++) {
2348 mark_buffer_async_read(bh
);
2352 * Stage 3: start the IO. Check for uptodateness
2353 * inside the buffer lock in case another process reading
2354 * the underlying blockdev brought it uptodate (the sct fix).
2356 for (i
= 0; i
< nr
; i
++) {
2358 if (buffer_uptodate(bh
))
2359 end_buffer_async_read(bh
, 1);
2361 submit_bh(REQ_OP_READ
, 0, bh
);
2365 EXPORT_SYMBOL(block_read_full_page
);
2367 /* utility function for filesystems that need to do work on expanding
2368 * truncates. Uses filesystem pagecache writes to allow the filesystem to
2369 * deal with the hole.
2371 int generic_cont_expand_simple(struct inode
*inode
, loff_t size
)
2373 struct address_space
*mapping
= inode
->i_mapping
;
2378 err
= inode_newsize_ok(inode
, size
);
2382 err
= pagecache_write_begin(NULL
, mapping
, size
, 0,
2383 AOP_FLAG_CONT_EXPAND
, &page
, &fsdata
);
2387 err
= pagecache_write_end(NULL
, mapping
, size
, 0, 0, page
, fsdata
);
2393 EXPORT_SYMBOL(generic_cont_expand_simple
);
2395 static int cont_expand_zero(struct file
*file
, struct address_space
*mapping
,
2396 loff_t pos
, loff_t
*bytes
)
2398 struct inode
*inode
= mapping
->host
;
2399 unsigned int blocksize
= i_blocksize(inode
);
2402 pgoff_t index
, curidx
;
2404 unsigned zerofrom
, offset
, len
;
2407 index
= pos
>> PAGE_SHIFT
;
2408 offset
= pos
& ~PAGE_MASK
;
2410 while (index
> (curidx
= (curpos
= *bytes
)>>PAGE_SHIFT
)) {
2411 zerofrom
= curpos
& ~PAGE_MASK
;
2412 if (zerofrom
& (blocksize
-1)) {
2413 *bytes
|= (blocksize
-1);
2416 len
= PAGE_SIZE
- zerofrom
;
2418 err
= pagecache_write_begin(file
, mapping
, curpos
, len
, 0,
2422 zero_user(page
, zerofrom
, len
);
2423 err
= pagecache_write_end(file
, mapping
, curpos
, len
, len
,
2430 balance_dirty_pages_ratelimited(mapping
);
2432 if (unlikely(fatal_signal_pending(current
))) {
2438 /* page covers the boundary, find the boundary offset */
2439 if (index
== curidx
) {
2440 zerofrom
= curpos
& ~PAGE_MASK
;
2441 /* if we will expand the thing last block will be filled */
2442 if (offset
<= zerofrom
) {
2445 if (zerofrom
& (blocksize
-1)) {
2446 *bytes
|= (blocksize
-1);
2449 len
= offset
- zerofrom
;
2451 err
= pagecache_write_begin(file
, mapping
, curpos
, len
, 0,
2455 zero_user(page
, zerofrom
, len
);
2456 err
= pagecache_write_end(file
, mapping
, curpos
, len
, len
,
2468 * For moronic filesystems that do not allow holes in file.
2469 * We may have to extend the file.
2471 int cont_write_begin(struct file
*file
, struct address_space
*mapping
,
2472 loff_t pos
, unsigned len
, unsigned flags
,
2473 struct page
**pagep
, void **fsdata
,
2474 get_block_t
*get_block
, loff_t
*bytes
)
2476 struct inode
*inode
= mapping
->host
;
2477 unsigned int blocksize
= i_blocksize(inode
);
2478 unsigned int zerofrom
;
2481 err
= cont_expand_zero(file
, mapping
, pos
, bytes
);
2485 zerofrom
= *bytes
& ~PAGE_MASK
;
2486 if (pos
+len
> *bytes
&& zerofrom
& (blocksize
-1)) {
2487 *bytes
|= (blocksize
-1);
2491 return block_write_begin(mapping
, pos
, len
, flags
, pagep
, get_block
);
2493 EXPORT_SYMBOL(cont_write_begin
);
2495 int block_commit_write(struct page
*page
, unsigned from
, unsigned to
)
2497 struct inode
*inode
= page
->mapping
->host
;
2498 __block_commit_write(inode
,page
,from
,to
);
2501 EXPORT_SYMBOL(block_commit_write
);
2504 * block_page_mkwrite() is not allowed to change the file size as it gets
2505 * called from a page fault handler when a page is first dirtied. Hence we must
2506 * be careful to check for EOF conditions here. We set the page up correctly
2507 * for a written page which means we get ENOSPC checking when writing into
2508 * holes and correct delalloc and unwritten extent mapping on filesystems that
2509 * support these features.
2511 * We are not allowed to take the i_mutex here so we have to play games to
2512 * protect against truncate races as the page could now be beyond EOF. Because
2513 * truncate writes the inode size before removing pages, once we have the
2514 * page lock we can determine safely if the page is beyond EOF. If it is not
2515 * beyond EOF, then the page is guaranteed safe against truncation until we
2518 * Direct callers of this function should protect against filesystem freezing
2519 * using sb_start_pagefault() - sb_end_pagefault() functions.
2521 int block_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
2522 get_block_t get_block
)
2524 struct page
*page
= vmf
->page
;
2525 struct inode
*inode
= file_inode(vma
->vm_file
);
2531 size
= i_size_read(inode
);
2532 if ((page
->mapping
!= inode
->i_mapping
) ||
2533 (page_offset(page
) > size
)) {
2534 /* We overload EFAULT to mean page got truncated */
2539 /* page is wholly or partially inside EOF */
2540 if (((page
->index
+ 1) << PAGE_SHIFT
) > size
)
2541 end
= size
& ~PAGE_MASK
;
2545 ret
= __block_write_begin(page
, 0, end
, get_block
);
2547 ret
= block_commit_write(page
, 0, end
);
2549 if (unlikely(ret
< 0))
2551 set_page_dirty(page
);
2552 wait_for_stable_page(page
);
2558 EXPORT_SYMBOL(block_page_mkwrite
);
2561 * nobh_write_begin()'s prereads are special: the buffer_heads are freed
2562 * immediately, while under the page lock. So it needs a special end_io
2563 * handler which does not touch the bh after unlocking it.
2565 static void end_buffer_read_nobh(struct buffer_head
*bh
, int uptodate
)
2567 __end_buffer_read_notouch(bh
, uptodate
);
2571 * Attach the singly-linked list of buffers created by nobh_write_begin, to
2572 * the page (converting it to circular linked list and taking care of page
2575 static void attach_nobh_buffers(struct page
*page
, struct buffer_head
*head
)
2577 struct buffer_head
*bh
;
2579 BUG_ON(!PageLocked(page
));
2581 spin_lock(&page
->mapping
->private_lock
);
2584 if (PageDirty(page
))
2585 set_buffer_dirty(bh
);
2586 if (!bh
->b_this_page
)
2587 bh
->b_this_page
= head
;
2588 bh
= bh
->b_this_page
;
2589 } while (bh
!= head
);
2590 attach_page_buffers(page
, head
);
2591 spin_unlock(&page
->mapping
->private_lock
);
2595 * On entry, the page is fully not uptodate.
2596 * On exit the page is fully uptodate in the areas outside (from,to)
2597 * The filesystem needs to handle block truncation upon failure.
2599 int nobh_write_begin(struct address_space
*mapping
,
2600 loff_t pos
, unsigned len
, unsigned flags
,
2601 struct page
**pagep
, void **fsdata
,
2602 get_block_t
*get_block
)
2604 struct inode
*inode
= mapping
->host
;
2605 const unsigned blkbits
= inode
->i_blkbits
;
2606 const unsigned blocksize
= 1 << blkbits
;
2607 struct buffer_head
*head
, *bh
;
2611 unsigned block_in_page
;
2612 unsigned block_start
, block_end
;
2613 sector_t block_in_file
;
2616 int is_mapped_to_disk
= 1;
2618 index
= pos
>> PAGE_SHIFT
;
2619 from
= pos
& (PAGE_SIZE
- 1);
2622 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2628 if (page_has_buffers(page
)) {
2629 ret
= __block_write_begin(page
, pos
, len
, get_block
);
2635 if (PageMappedToDisk(page
))
2639 * Allocate buffers so that we can keep track of state, and potentially
2640 * attach them to the page if an error occurs. In the common case of
2641 * no error, they will just be freed again without ever being attached
2642 * to the page (which is all OK, because we're under the page lock).
2644 * Be careful: the buffer linked list is a NULL terminated one, rather
2645 * than the circular one we're used to.
2647 head
= alloc_page_buffers(page
, blocksize
, 0);
2653 block_in_file
= (sector_t
)page
->index
<< (PAGE_SHIFT
- blkbits
);
2656 * We loop across all blocks in the page, whether or not they are
2657 * part of the affected region. This is so we can discover if the
2658 * page is fully mapped-to-disk.
2660 for (block_start
= 0, block_in_page
= 0, bh
= head
;
2661 block_start
< PAGE_SIZE
;
2662 block_in_page
++, block_start
+= blocksize
, bh
= bh
->b_this_page
) {
2665 block_end
= block_start
+ blocksize
;
2668 if (block_start
>= to
)
2670 ret
= get_block(inode
, block_in_file
+ block_in_page
,
2674 if (!buffer_mapped(bh
))
2675 is_mapped_to_disk
= 0;
2677 clean_bdev_bh_alias(bh
);
2678 if (PageUptodate(page
)) {
2679 set_buffer_uptodate(bh
);
2682 if (buffer_new(bh
) || !buffer_mapped(bh
)) {
2683 zero_user_segments(page
, block_start
, from
,
2687 if (buffer_uptodate(bh
))
2688 continue; /* reiserfs does this */
2689 if (block_start
< from
|| block_end
> to
) {
2691 bh
->b_end_io
= end_buffer_read_nobh
;
2692 submit_bh(REQ_OP_READ
, 0, bh
);
2699 * The page is locked, so these buffers are protected from
2700 * any VM or truncate activity. Hence we don't need to care
2701 * for the buffer_head refcounts.
2703 for (bh
= head
; bh
; bh
= bh
->b_this_page
) {
2705 if (!buffer_uptodate(bh
))
2712 if (is_mapped_to_disk
)
2713 SetPageMappedToDisk(page
);
2715 *fsdata
= head
; /* to be released by nobh_write_end */
2722 * Error recovery is a bit difficult. We need to zero out blocks that
2723 * were newly allocated, and dirty them to ensure they get written out.
2724 * Buffers need to be attached to the page at this point, otherwise
2725 * the handling of potential IO errors during writeout would be hard
2726 * (could try doing synchronous writeout, but what if that fails too?)
2728 attach_nobh_buffers(page
, head
);
2729 page_zero_new_buffers(page
, from
, to
);
2738 EXPORT_SYMBOL(nobh_write_begin
);
2740 int nobh_write_end(struct file
*file
, struct address_space
*mapping
,
2741 loff_t pos
, unsigned len
, unsigned copied
,
2742 struct page
*page
, void *fsdata
)
2744 struct inode
*inode
= page
->mapping
->host
;
2745 struct buffer_head
*head
= fsdata
;
2746 struct buffer_head
*bh
;
2747 BUG_ON(fsdata
!= NULL
&& page_has_buffers(page
));
2749 if (unlikely(copied
< len
) && head
)
2750 attach_nobh_buffers(page
, head
);
2751 if (page_has_buffers(page
))
2752 return generic_write_end(file
, mapping
, pos
, len
,
2753 copied
, page
, fsdata
);
2755 SetPageUptodate(page
);
2756 set_page_dirty(page
);
2757 if (pos
+copied
> inode
->i_size
) {
2758 i_size_write(inode
, pos
+copied
);
2759 mark_inode_dirty(inode
);
2767 head
= head
->b_this_page
;
2768 free_buffer_head(bh
);
2773 EXPORT_SYMBOL(nobh_write_end
);
2776 * nobh_writepage() - based on block_full_write_page() except
2777 * that it tries to operate without attaching bufferheads to
2780 int nobh_writepage(struct page
*page
, get_block_t
*get_block
,
2781 struct writeback_control
*wbc
)
2783 struct inode
* const inode
= page
->mapping
->host
;
2784 loff_t i_size
= i_size_read(inode
);
2785 const pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2789 /* Is the page fully inside i_size? */
2790 if (page
->index
< end_index
)
2793 /* Is the page fully outside i_size? (truncate in progress) */
2794 offset
= i_size
& (PAGE_SIZE
-1);
2795 if (page
->index
>= end_index
+1 || !offset
) {
2797 * The page may have dirty, unmapped buffers. For example,
2798 * they may have been added in ext3_writepage(). Make them
2799 * freeable here, so the page does not leak.
2802 /* Not really sure about this - do we need this ? */
2803 if (page
->mapping
->a_ops
->invalidatepage
)
2804 page
->mapping
->a_ops
->invalidatepage(page
, offset
);
2807 return 0; /* don't care */
2811 * The page straddles i_size. It must be zeroed out on each and every
2812 * writepage invocation because it may be mmapped. "A file is mapped
2813 * in multiples of the page size. For a file that is not a multiple of
2814 * the page size, the remaining memory is zeroed when mapped, and
2815 * writes to that region are not written out to the file."
2817 zero_user_segment(page
, offset
, PAGE_SIZE
);
2819 ret
= mpage_writepage(page
, get_block
, wbc
);
2821 ret
= __block_write_full_page(inode
, page
, get_block
, wbc
,
2822 end_buffer_async_write
);
2825 EXPORT_SYMBOL(nobh_writepage
);
2827 int nobh_truncate_page(struct address_space
*mapping
,
2828 loff_t from
, get_block_t
*get_block
)
2830 pgoff_t index
= from
>> PAGE_SHIFT
;
2831 unsigned offset
= from
& (PAGE_SIZE
-1);
2834 unsigned length
, pos
;
2835 struct inode
*inode
= mapping
->host
;
2837 struct buffer_head map_bh
;
2840 blocksize
= i_blocksize(inode
);
2841 length
= offset
& (blocksize
- 1);
2843 /* Block boundary? Nothing to do */
2847 length
= blocksize
- length
;
2848 iblock
= (sector_t
)index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
2850 page
= grab_cache_page(mapping
, index
);
2855 if (page_has_buffers(page
)) {
2859 return block_truncate_page(mapping
, from
, get_block
);
2862 /* Find the buffer that contains "offset" */
2864 while (offset
>= pos
) {
2869 map_bh
.b_size
= blocksize
;
2871 err
= get_block(inode
, iblock
, &map_bh
, 0);
2874 /* unmapped? It's a hole - nothing to do */
2875 if (!buffer_mapped(&map_bh
))
2878 /* Ok, it's mapped. Make sure it's up-to-date */
2879 if (!PageUptodate(page
)) {
2880 err
= mapping
->a_ops
->readpage(NULL
, page
);
2886 if (!PageUptodate(page
)) {
2890 if (page_has_buffers(page
))
2893 zero_user(page
, offset
, length
);
2894 set_page_dirty(page
);
2903 EXPORT_SYMBOL(nobh_truncate_page
);
2905 int block_truncate_page(struct address_space
*mapping
,
2906 loff_t from
, get_block_t
*get_block
)
2908 pgoff_t index
= from
>> PAGE_SHIFT
;
2909 unsigned offset
= from
& (PAGE_SIZE
-1);
2912 unsigned length
, pos
;
2913 struct inode
*inode
= mapping
->host
;
2915 struct buffer_head
*bh
;
2918 blocksize
= i_blocksize(inode
);
2919 length
= offset
& (blocksize
- 1);
2921 /* Block boundary? Nothing to do */
2925 length
= blocksize
- length
;
2926 iblock
= (sector_t
)index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
2928 page
= grab_cache_page(mapping
, index
);
2933 if (!page_has_buffers(page
))
2934 create_empty_buffers(page
, blocksize
, 0);
2936 /* Find the buffer that contains "offset" */
2937 bh
= page_buffers(page
);
2939 while (offset
>= pos
) {
2940 bh
= bh
->b_this_page
;
2946 if (!buffer_mapped(bh
)) {
2947 WARN_ON(bh
->b_size
!= blocksize
);
2948 err
= get_block(inode
, iblock
, bh
, 0);
2951 /* unmapped? It's a hole - nothing to do */
2952 if (!buffer_mapped(bh
))
2956 /* Ok, it's mapped. Make sure it's up-to-date */
2957 if (PageUptodate(page
))
2958 set_buffer_uptodate(bh
);
2960 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) && !buffer_unwritten(bh
)) {
2962 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
2964 /* Uhhuh. Read error. Complain and punt. */
2965 if (!buffer_uptodate(bh
))
2969 zero_user(page
, offset
, length
);
2970 mark_buffer_dirty(bh
);
2979 EXPORT_SYMBOL(block_truncate_page
);
2982 * The generic ->writepage function for buffer-backed address_spaces
2984 int block_write_full_page(struct page
*page
, get_block_t
*get_block
,
2985 struct writeback_control
*wbc
)
2987 struct inode
* const inode
= page
->mapping
->host
;
2988 loff_t i_size
= i_size_read(inode
);
2989 const pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2992 /* Is the page fully inside i_size? */
2993 if (page
->index
< end_index
)
2994 return __block_write_full_page(inode
, page
, get_block
, wbc
,
2995 end_buffer_async_write
);
2997 /* Is the page fully outside i_size? (truncate in progress) */
2998 offset
= i_size
& (PAGE_SIZE
-1);
2999 if (page
->index
>= end_index
+1 || !offset
) {
3001 * The page may have dirty, unmapped buffers. For example,
3002 * they may have been added in ext3_writepage(). Make them
3003 * freeable here, so the page does not leak.
3005 do_invalidatepage(page
, 0, PAGE_SIZE
);
3007 return 0; /* don't care */
3011 * The page straddles i_size. It must be zeroed out on each and every
3012 * writepage invocation because it may be mmapped. "A file is mapped
3013 * in multiples of the page size. For a file that is not a multiple of
3014 * the page size, the remaining memory is zeroed when mapped, and
3015 * writes to that region are not written out to the file."
3017 zero_user_segment(page
, offset
, PAGE_SIZE
);
3018 return __block_write_full_page(inode
, page
, get_block
, wbc
,
3019 end_buffer_async_write
);
3021 EXPORT_SYMBOL(block_write_full_page
);
3023 sector_t
generic_block_bmap(struct address_space
*mapping
, sector_t block
,
3024 get_block_t
*get_block
)
3026 struct buffer_head tmp
;
3027 struct inode
*inode
= mapping
->host
;
3030 tmp
.b_size
= i_blocksize(inode
);
3031 get_block(inode
, block
, &tmp
, 0);
3032 return tmp
.b_blocknr
;
3034 EXPORT_SYMBOL(generic_block_bmap
);
3036 static void end_bio_bh_io_sync(struct bio
*bio
)
3038 struct buffer_head
*bh
= bio
->bi_private
;
3040 if (unlikely(bio_flagged(bio
, BIO_QUIET
)))
3041 set_bit(BH_Quiet
, &bh
->b_state
);
3043 bh
->b_end_io(bh
, !bio
->bi_status
);
3048 * This allows us to do IO even on the odd last sectors
3049 * of a device, even if the block size is some multiple
3050 * of the physical sector size.
3052 * We'll just truncate the bio to the size of the device,
3053 * and clear the end of the buffer head manually.
3055 * Truly out-of-range accesses will turn into actual IO
3056 * errors, this only handles the "we need to be able to
3057 * do IO at the final sector" case.
3059 void guard_bio_eod(int op
, struct bio
*bio
)
3062 struct bio_vec
*bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
3063 unsigned truncated_bytes
;
3065 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
3070 * If the *whole* IO is past the end of the device,
3071 * let it through, and the IO layer will turn it into
3074 if (unlikely(bio
->bi_iter
.bi_sector
>= maxsector
))
3077 maxsector
-= bio
->bi_iter
.bi_sector
;
3078 if (likely((bio
->bi_iter
.bi_size
>> 9) <= maxsector
))
3081 /* Uhhuh. We've got a bio that straddles the device size! */
3082 truncated_bytes
= bio
->bi_iter
.bi_size
- (maxsector
<< 9);
3084 /* Truncate the bio.. */
3085 bio
->bi_iter
.bi_size
-= truncated_bytes
;
3086 bvec
->bv_len
-= truncated_bytes
;
3088 /* ..and clear the end of the buffer for reads */
3089 if (op
== REQ_OP_READ
) {
3090 zero_user(bvec
->bv_page
, bvec
->bv_offset
+ bvec
->bv_len
,
3095 static int submit_bh_wbc(int op
, int op_flags
, struct buffer_head
*bh
,
3096 enum rw_hint write_hint
, struct writeback_control
*wbc
)
3100 BUG_ON(!buffer_locked(bh
));
3101 BUG_ON(!buffer_mapped(bh
));
3102 BUG_ON(!bh
->b_end_io
);
3103 BUG_ON(buffer_delay(bh
));
3104 BUG_ON(buffer_unwritten(bh
));
3107 * Only clear out a write error when rewriting
3109 if (test_set_buffer_req(bh
) && (op
== REQ_OP_WRITE
))
3110 clear_buffer_write_io_error(bh
);
3113 * from here on down, it's all bio -- do the initial mapping,
3114 * submit_bio -> generic_make_request may further map this bio around
3116 bio
= bio_alloc(GFP_NOIO
, 1);
3119 wbc_init_bio(wbc
, bio
);
3120 wbc_account_io(wbc
, bh
->b_page
, bh
->b_size
);
3123 bio
->bi_iter
.bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
3124 bio
->bi_bdev
= bh
->b_bdev
;
3125 bio
->bi_write_hint
= write_hint
;
3127 bio_add_page(bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
3128 BUG_ON(bio
->bi_iter
.bi_size
!= bh
->b_size
);
3130 bio
->bi_end_io
= end_bio_bh_io_sync
;
3131 bio
->bi_private
= bh
;
3133 /* Take care of bh's that straddle the end of the device */
3134 guard_bio_eod(op
, bio
);
3136 if (buffer_meta(bh
))
3137 op_flags
|= REQ_META
;
3138 if (buffer_prio(bh
))
3139 op_flags
|= REQ_PRIO
;
3140 bio_set_op_attrs(bio
, op
, op_flags
);
3146 int submit_bh(int op
, int op_flags
, struct buffer_head
*bh
)
3148 return submit_bh_wbc(op
, op_flags
, bh
, 0, NULL
);
3150 EXPORT_SYMBOL(submit_bh
);
3153 * ll_rw_block: low-level access to block devices (DEPRECATED)
3154 * @op: whether to %READ or %WRITE
3155 * @op_flags: req_flag_bits
3156 * @nr: number of &struct buffer_heads in the array
3157 * @bhs: array of pointers to &struct buffer_head
3159 * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
3160 * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE.
3161 * @op_flags contains flags modifying the detailed I/O behavior, most notably
3164 * This function drops any buffer that it cannot get a lock on (with the
3165 * BH_Lock state bit), any buffer that appears to be clean when doing a write
3166 * request, and any buffer that appears to be up-to-date when doing read
3167 * request. Further it marks as clean buffers that are processed for
3168 * writing (the buffer cache won't assume that they are actually clean
3169 * until the buffer gets unlocked).
3171 * ll_rw_block sets b_end_io to simple completion handler that marks
3172 * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
3175 * All of the buffers must be for the same device, and must also be a
3176 * multiple of the current approved size for the device.
3178 void ll_rw_block(int op
, int op_flags
, int nr
, struct buffer_head
*bhs
[])
3182 for (i
= 0; i
< nr
; i
++) {
3183 struct buffer_head
*bh
= bhs
[i
];
3185 if (!trylock_buffer(bh
))
3188 if (test_clear_buffer_dirty(bh
)) {
3189 bh
->b_end_io
= end_buffer_write_sync
;
3191 submit_bh(op
, op_flags
, bh
);
3195 if (!buffer_uptodate(bh
)) {
3196 bh
->b_end_io
= end_buffer_read_sync
;
3198 submit_bh(op
, op_flags
, bh
);
3205 EXPORT_SYMBOL(ll_rw_block
);
3207 void write_dirty_buffer(struct buffer_head
*bh
, int op_flags
)
3210 if (!test_clear_buffer_dirty(bh
)) {
3214 bh
->b_end_io
= end_buffer_write_sync
;
3216 submit_bh(REQ_OP_WRITE
, op_flags
, bh
);
3218 EXPORT_SYMBOL(write_dirty_buffer
);
3221 * For a data-integrity writeout, we need to wait upon any in-progress I/O
3222 * and then start new I/O and then wait upon it. The caller must have a ref on
3225 int __sync_dirty_buffer(struct buffer_head
*bh
, int op_flags
)
3229 WARN_ON(atomic_read(&bh
->b_count
) < 1);
3231 if (test_clear_buffer_dirty(bh
)) {
3233 bh
->b_end_io
= end_buffer_write_sync
;
3234 ret
= submit_bh(REQ_OP_WRITE
, op_flags
, bh
);
3236 if (!ret
&& !buffer_uptodate(bh
))
3243 EXPORT_SYMBOL(__sync_dirty_buffer
);
3245 int sync_dirty_buffer(struct buffer_head
*bh
)
3247 return __sync_dirty_buffer(bh
, REQ_SYNC
);
3249 EXPORT_SYMBOL(sync_dirty_buffer
);
3252 * try_to_free_buffers() checks if all the buffers on this particular page
3253 * are unused, and releases them if so.
3255 * Exclusion against try_to_free_buffers may be obtained by either
3256 * locking the page or by holding its mapping's private_lock.
3258 * If the page is dirty but all the buffers are clean then we need to
3259 * be sure to mark the page clean as well. This is because the page
3260 * may be against a block device, and a later reattachment of buffers
3261 * to a dirty page will set *all* buffers dirty. Which would corrupt
3262 * filesystem data on the same device.
3264 * The same applies to regular filesystem pages: if all the buffers are
3265 * clean then we set the page clean and proceed. To do that, we require
3266 * total exclusion from __set_page_dirty_buffers(). That is obtained with
3269 * try_to_free_buffers() is non-blocking.
3271 static inline int buffer_busy(struct buffer_head
*bh
)
3273 return atomic_read(&bh
->b_count
) |
3274 (bh
->b_state
& ((1 << BH_Dirty
) | (1 << BH_Lock
)));
3278 drop_buffers(struct page
*page
, struct buffer_head
**buffers_to_free
)
3280 struct buffer_head
*head
= page_buffers(page
);
3281 struct buffer_head
*bh
;
3285 if (buffer_write_io_error(bh
) && page
->mapping
)
3286 mapping_set_error(page
->mapping
, -EIO
);
3287 if (buffer_busy(bh
))
3289 bh
= bh
->b_this_page
;
3290 } while (bh
!= head
);
3293 struct buffer_head
*next
= bh
->b_this_page
;
3295 if (bh
->b_assoc_map
)
3296 __remove_assoc_queue(bh
);
3298 } while (bh
!= head
);
3299 *buffers_to_free
= head
;
3300 __clear_page_buffers(page
);
3306 int try_to_free_buffers(struct page
*page
)
3308 struct address_space
* const mapping
= page
->mapping
;
3309 struct buffer_head
*buffers_to_free
= NULL
;
3312 BUG_ON(!PageLocked(page
));
3313 if (PageWriteback(page
))
3316 if (mapping
== NULL
) { /* can this still happen? */
3317 ret
= drop_buffers(page
, &buffers_to_free
);
3321 spin_lock(&mapping
->private_lock
);
3322 ret
= drop_buffers(page
, &buffers_to_free
);
3325 * If the filesystem writes its buffers by hand (eg ext3)
3326 * then we can have clean buffers against a dirty page. We
3327 * clean the page here; otherwise the VM will never notice
3328 * that the filesystem did any IO at all.
3330 * Also, during truncate, discard_buffer will have marked all
3331 * the page's buffers clean. We discover that here and clean
3334 * private_lock must be held over this entire operation in order
3335 * to synchronise against __set_page_dirty_buffers and prevent the
3336 * dirty bit from being lost.
3339 cancel_dirty_page(page
);
3340 spin_unlock(&mapping
->private_lock
);
3342 if (buffers_to_free
) {
3343 struct buffer_head
*bh
= buffers_to_free
;
3346 struct buffer_head
*next
= bh
->b_this_page
;
3347 free_buffer_head(bh
);
3349 } while (bh
!= buffers_to_free
);
3353 EXPORT_SYMBOL(try_to_free_buffers
);
3356 * There are no bdflush tunables left. But distributions are
3357 * still running obsolete flush daemons, so we terminate them here.
3359 * Use of bdflush() is deprecated and will be removed in a future kernel.
3360 * The `flush-X' kernel threads fully replace bdflush daemons and this call.
3362 SYSCALL_DEFINE2(bdflush
, int, func
, long, data
)
3364 static int msg_count
;
3366 if (!capable(CAP_SYS_ADMIN
))
3369 if (msg_count
< 5) {
3372 "warning: process `%s' used the obsolete bdflush"
3373 " system call\n", current
->comm
);
3374 printk(KERN_INFO
"Fix your initscripts?\n");
3383 * Buffer-head allocation
3385 static struct kmem_cache
*bh_cachep __read_mostly
;
3388 * Once the number of bh's in the machine exceeds this level, we start
3389 * stripping them in writeback.
3391 static unsigned long max_buffer_heads
;
3393 int buffer_heads_over_limit
;
3395 struct bh_accounting
{
3396 int nr
; /* Number of live bh's */
3397 int ratelimit
; /* Limit cacheline bouncing */
3400 static DEFINE_PER_CPU(struct bh_accounting
, bh_accounting
) = {0, 0};
3402 static void recalc_bh_state(void)
3407 if (__this_cpu_inc_return(bh_accounting
.ratelimit
) - 1 < 4096)
3409 __this_cpu_write(bh_accounting
.ratelimit
, 0);
3410 for_each_online_cpu(i
)
3411 tot
+= per_cpu(bh_accounting
, i
).nr
;
3412 buffer_heads_over_limit
= (tot
> max_buffer_heads
);
3415 struct buffer_head
*alloc_buffer_head(gfp_t gfp_flags
)
3417 struct buffer_head
*ret
= kmem_cache_zalloc(bh_cachep
, gfp_flags
);
3419 INIT_LIST_HEAD(&ret
->b_assoc_buffers
);
3421 __this_cpu_inc(bh_accounting
.nr
);
3427 EXPORT_SYMBOL(alloc_buffer_head
);
3429 void free_buffer_head(struct buffer_head
*bh
)
3431 BUG_ON(!list_empty(&bh
->b_assoc_buffers
));
3432 kmem_cache_free(bh_cachep
, bh
);
3434 __this_cpu_dec(bh_accounting
.nr
);
3438 EXPORT_SYMBOL(free_buffer_head
);
3440 static int buffer_exit_cpu_dead(unsigned int cpu
)
3443 struct bh_lru
*b
= &per_cpu(bh_lrus
, cpu
);
3445 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
3449 this_cpu_add(bh_accounting
.nr
, per_cpu(bh_accounting
, cpu
).nr
);
3450 per_cpu(bh_accounting
, cpu
).nr
= 0;
3455 * bh_uptodate_or_lock - Test whether the buffer is uptodate
3456 * @bh: struct buffer_head
3458 * Return true if the buffer is up-to-date and false,
3459 * with the buffer locked, if not.
3461 int bh_uptodate_or_lock(struct buffer_head
*bh
)
3463 if (!buffer_uptodate(bh
)) {
3465 if (!buffer_uptodate(bh
))
3471 EXPORT_SYMBOL(bh_uptodate_or_lock
);
3474 * bh_submit_read - Submit a locked buffer for reading
3475 * @bh: struct buffer_head
3477 * Returns zero on success and -EIO on error.
3479 int bh_submit_read(struct buffer_head
*bh
)
3481 BUG_ON(!buffer_locked(bh
));
3483 if (buffer_uptodate(bh
)) {
3489 bh
->b_end_io
= end_buffer_read_sync
;
3490 submit_bh(REQ_OP_READ
, 0, bh
);
3492 if (buffer_uptodate(bh
))
3496 EXPORT_SYMBOL(bh_submit_read
);
3498 void __init
buffer_init(void)
3500 unsigned long nrpages
;
3503 bh_cachep
= kmem_cache_create("buffer_head",
3504 sizeof(struct buffer_head
), 0,
3505 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
3510 * Limit the bh occupancy to 10% of ZONE_NORMAL
3512 nrpages
= (nr_free_buffer_pages() * 10) / 100;
3513 max_buffer_heads
= nrpages
* (PAGE_SIZE
/ sizeof(struct buffer_head
));
3514 ret
= cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD
, "fs/buffer:dead",
3515 NULL
, buffer_exit_cpu_dead
);