4 * Copyright (C) 1991, 1992, 2002 Linus Torvalds
8 * Start bdflush() with kernel_thread not syscall - Paul Gortmaker, 12/95
10 * Removed a lot of unnecessary code and simplified things now that
11 * the buffer cache isn't our primary cache - Andrew Tridgell 12/96
13 * Speed up hash, lru, and free list operations. Use gfp() for allocating
14 * hash table, use SLAB cache for buffer heads. SMP threading. -DaveM
16 * Added 32k buffer block sizes - these are required older ARM systems. - RMK
18 * async buffer flushing, 1999 Andrea Arcangeli <andrea@suse.de>
21 #include <linux/kernel.h>
22 #include <linux/sched/signal.h>
23 #include <linux/syscalls.h>
25 #include <linux/iomap.h>
27 #include <linux/percpu.h>
28 #include <linux/slab.h>
29 #include <linux/capability.h>
30 #include <linux/blkdev.h>
31 #include <linux/file.h>
32 #include <linux/quotaops.h>
33 #include <linux/highmem.h>
34 #include <linux/export.h>
35 #include <linux/backing-dev.h>
36 #include <linux/writeback.h>
37 #include <linux/hash.h>
38 #include <linux/suspend.h>
39 #include <linux/buffer_head.h>
40 #include <linux/task_io_accounting_ops.h>
41 #include <linux/bio.h>
42 #include <linux/notifier.h>
43 #include <linux/cpu.h>
44 #include <linux/bitops.h>
45 #include <linux/mpage.h>
46 #include <linux/bit_spinlock.h>
47 #include <linux/pagevec.h>
48 #include <trace/events/block.h>
50 static int fsync_buffers_list(spinlock_t
*lock
, struct list_head
*list
);
51 static int submit_bh_wbc(int op
, int op_flags
, struct buffer_head
*bh
,
52 enum rw_hint hint
, struct writeback_control
*wbc
);
54 #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
56 void init_buffer(struct buffer_head
*bh
, bh_end_io_t
*handler
, void *private)
58 bh
->b_end_io
= handler
;
59 bh
->b_private
= private;
61 EXPORT_SYMBOL(init_buffer
);
63 inline void touch_buffer(struct buffer_head
*bh
)
65 trace_block_touch_buffer(bh
);
66 mark_page_accessed(bh
->b_page
);
68 EXPORT_SYMBOL(touch_buffer
);
70 void __lock_buffer(struct buffer_head
*bh
)
72 wait_on_bit_lock_io(&bh
->b_state
, BH_Lock
, TASK_UNINTERRUPTIBLE
);
74 EXPORT_SYMBOL(__lock_buffer
);
76 void unlock_buffer(struct buffer_head
*bh
)
78 clear_bit_unlock(BH_Lock
, &bh
->b_state
);
79 smp_mb__after_atomic();
80 wake_up_bit(&bh
->b_state
, BH_Lock
);
82 EXPORT_SYMBOL(unlock_buffer
);
85 * Returns if the page has dirty or writeback buffers. If all the buffers
86 * are unlocked and clean then the PageDirty information is stale. If
87 * any of the pages are locked, it is assumed they are locked for IO.
89 void buffer_check_dirty_writeback(struct page
*page
,
90 bool *dirty
, bool *writeback
)
92 struct buffer_head
*head
, *bh
;
96 BUG_ON(!PageLocked(page
));
98 if (!page_has_buffers(page
))
101 if (PageWriteback(page
))
104 head
= page_buffers(page
);
107 if (buffer_locked(bh
))
110 if (buffer_dirty(bh
))
113 bh
= bh
->b_this_page
;
114 } while (bh
!= head
);
116 EXPORT_SYMBOL(buffer_check_dirty_writeback
);
119 * Block until a buffer comes unlocked. This doesn't stop it
120 * from becoming locked again - you have to lock it yourself
121 * if you want to preserve its state.
123 void __wait_on_buffer(struct buffer_head
* bh
)
125 wait_on_bit_io(&bh
->b_state
, BH_Lock
, TASK_UNINTERRUPTIBLE
);
127 EXPORT_SYMBOL(__wait_on_buffer
);
130 __clear_page_buffers(struct page
*page
)
132 ClearPagePrivate(page
);
133 set_page_private(page
, 0);
137 static void buffer_io_error(struct buffer_head
*bh
, char *msg
)
139 if (!test_bit(BH_Quiet
, &bh
->b_state
))
140 printk_ratelimited(KERN_ERR
141 "Buffer I/O error on dev %pg, logical block %llu%s\n",
142 bh
->b_bdev
, (unsigned long long)bh
->b_blocknr
, msg
);
146 * End-of-IO handler helper function which does not touch the bh after
148 * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
149 * a race there is benign: unlock_buffer() only use the bh's address for
150 * hashing after unlocking the buffer, so it doesn't actually touch the bh
153 static void __end_buffer_read_notouch(struct buffer_head
*bh
, int uptodate
)
156 set_buffer_uptodate(bh
);
158 /* This happens, due to failed read-ahead attempts. */
159 clear_buffer_uptodate(bh
);
165 * Default synchronous end-of-IO handler.. Just mark it up-to-date and
166 * unlock the buffer. This is what ll_rw_block uses too.
168 void end_buffer_read_sync(struct buffer_head
*bh
, int uptodate
)
170 __end_buffer_read_notouch(bh
, uptodate
);
173 EXPORT_SYMBOL(end_buffer_read_sync
);
175 void end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
178 set_buffer_uptodate(bh
);
180 buffer_io_error(bh
, ", lost sync page write");
181 mark_buffer_write_io_error(bh
);
182 clear_buffer_uptodate(bh
);
187 EXPORT_SYMBOL(end_buffer_write_sync
);
190 * Various filesystems appear to want __find_get_block to be non-blocking.
191 * But it's the page lock which protects the buffers. To get around this,
192 * we get exclusion from try_to_free_buffers with the blockdev mapping's
195 * Hack idea: for the blockdev mapping, i_bufferlist_lock contention
196 * may be quite high. This code could TryLock the page, and if that
197 * succeeds, there is no need to take private_lock. (But if
198 * private_lock is contended then so is mapping->tree_lock).
200 static struct buffer_head
*
201 __find_get_block_slow(struct block_device
*bdev
, sector_t block
)
203 struct inode
*bd_inode
= bdev
->bd_inode
;
204 struct address_space
*bd_mapping
= bd_inode
->i_mapping
;
205 struct buffer_head
*ret
= NULL
;
207 struct buffer_head
*bh
;
208 struct buffer_head
*head
;
211 static DEFINE_RATELIMIT_STATE(last_warned
, HZ
, 1);
213 index
= block
>> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
214 page
= find_get_page_flags(bd_mapping
, index
, FGP_ACCESSED
);
218 spin_lock(&bd_mapping
->private_lock
);
219 if (!page_has_buffers(page
))
221 head
= page_buffers(page
);
224 if (!buffer_mapped(bh
))
226 else if (bh
->b_blocknr
== block
) {
231 bh
= bh
->b_this_page
;
232 } while (bh
!= head
);
234 /* we might be here because some of the buffers on this page are
235 * not mapped. This is due to various races between
236 * file io on the block device and getblk. It gets dealt with
237 * elsewhere, don't buffer_error if we had some unmapped buffers
239 ratelimit_set_flags(&last_warned
, RATELIMIT_MSG_ON_RELEASE
);
240 if (all_mapped
&& __ratelimit(&last_warned
)) {
241 printk("__find_get_block_slow() failed. block=%llu, "
242 "b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, "
243 "device %pg blocksize: %d\n",
244 (unsigned long long)block
,
245 (unsigned long long)bh
->b_blocknr
,
246 bh
->b_state
, bh
->b_size
, bdev
,
247 1 << bd_inode
->i_blkbits
);
250 spin_unlock(&bd_mapping
->private_lock
);
257 * Kick the writeback threads then try to free up some ZONE_NORMAL memory.
259 static void free_more_memory(void)
264 wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM
);
267 for_each_online_node(nid
) {
269 z
= first_zones_zonelist(node_zonelist(nid
, GFP_NOFS
),
270 gfp_zone(GFP_NOFS
), NULL
);
272 try_to_free_pages(node_zonelist(nid
, GFP_NOFS
), 0,
278 * I/O completion handler for block_read_full_page() - pages
279 * which come unlocked at the end of I/O.
281 static void end_buffer_async_read(struct buffer_head
*bh
, int uptodate
)
284 struct buffer_head
*first
;
285 struct buffer_head
*tmp
;
287 int page_uptodate
= 1;
289 BUG_ON(!buffer_async_read(bh
));
293 set_buffer_uptodate(bh
);
295 clear_buffer_uptodate(bh
);
296 buffer_io_error(bh
, ", async page read");
301 * Be _very_ careful from here on. Bad things can happen if
302 * two buffer heads end IO at almost the same time and both
303 * decide that the page is now completely done.
305 first
= page_buffers(page
);
306 local_irq_save(flags
);
307 bit_spin_lock(BH_Uptodate_Lock
, &first
->b_state
);
308 clear_buffer_async_read(bh
);
312 if (!buffer_uptodate(tmp
))
314 if (buffer_async_read(tmp
)) {
315 BUG_ON(!buffer_locked(tmp
));
318 tmp
= tmp
->b_this_page
;
320 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
321 local_irq_restore(flags
);
324 * If none of the buffers had errors and they are all
325 * uptodate then we can set the page uptodate.
327 if (page_uptodate
&& !PageError(page
))
328 SetPageUptodate(page
);
333 bit_spin_unlock(BH_Uptodate_Lock
, &first
->b_state
);
334 local_irq_restore(flags
);
339 * Completion handler for block_write_full_page() - pages which are unlocked
340 * during I/O, and which have PageWriteback cleared upon I/O completion.
342 void end_buffer_async_write(struct buffer_head
*bh
, int uptodate
)
345 struct buffer_head
*first
;
346 struct buffer_head
*tmp
;
349 BUG_ON(!buffer_async_write(bh
));
353 set_buffer_uptodate(bh
);
355 buffer_io_error(bh
, ", lost async page write");
356 mark_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 bh
->b_assoc_map
= NULL
;
487 int inode_has_buffers(struct inode
*inode
)
489 return !list_empty(&inode
->i_data
.private_list
);
493 * osync is designed to support O_SYNC io. It waits synchronously for
494 * all already-submitted IO to complete, but does not queue any new
495 * writes to the disk.
497 * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
498 * you dirty the buffers, and then use osync_inode_buffers to wait for
499 * completion. Any other dirty buffers which are not yet queued for
500 * write will not be flushed to disk by the osync.
502 static int osync_buffers_list(spinlock_t
*lock
, struct list_head
*list
)
504 struct buffer_head
*bh
;
510 list_for_each_prev(p
, list
) {
512 if (buffer_locked(bh
)) {
516 if (!buffer_uptodate(bh
))
527 static void do_thaw_one(struct super_block
*sb
, void *unused
)
529 while (sb
->s_bdev
&& !thaw_bdev(sb
->s_bdev
, sb
))
530 printk(KERN_WARNING
"Emergency Thaw on %pg\n", sb
->s_bdev
);
533 static void do_thaw_all(struct work_struct
*work
)
535 iterate_supers(do_thaw_one
, NULL
);
537 printk(KERN_WARNING
"Emergency Thaw complete\n");
541 * emergency_thaw_all -- forcibly thaw every frozen filesystem
543 * Used for emergency unfreeze of all filesystems via SysRq
545 void emergency_thaw_all(void)
547 struct work_struct
*work
;
549 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
551 INIT_WORK(work
, do_thaw_all
);
557 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
558 * @mapping: the mapping which wants those buffers written
560 * Starts I/O against the buffers at mapping->private_list, and waits upon
563 * Basically, this is a convenience function for fsync().
564 * @mapping is a file or directory which needs those buffers to be written for
565 * a successful fsync().
567 int sync_mapping_buffers(struct address_space
*mapping
)
569 struct address_space
*buffer_mapping
= mapping
->private_data
;
571 if (buffer_mapping
== NULL
|| list_empty(&mapping
->private_list
))
574 return fsync_buffers_list(&buffer_mapping
->private_lock
,
575 &mapping
->private_list
);
577 EXPORT_SYMBOL(sync_mapping_buffers
);
580 * Called when we've recently written block `bblock', and it is known that
581 * `bblock' was for a buffer_boundary() buffer. This means that the block at
582 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
583 * dirty, schedule it for IO. So that indirects merge nicely with their data.
585 void write_boundary_block(struct block_device
*bdev
,
586 sector_t bblock
, unsigned blocksize
)
588 struct buffer_head
*bh
= __find_get_block(bdev
, bblock
+ 1, blocksize
);
590 if (buffer_dirty(bh
))
591 ll_rw_block(REQ_OP_WRITE
, 0, 1, &bh
);
596 void mark_buffer_dirty_inode(struct buffer_head
*bh
, struct inode
*inode
)
598 struct address_space
*mapping
= inode
->i_mapping
;
599 struct address_space
*buffer_mapping
= bh
->b_page
->mapping
;
601 mark_buffer_dirty(bh
);
602 if (!mapping
->private_data
) {
603 mapping
->private_data
= buffer_mapping
;
605 BUG_ON(mapping
->private_data
!= buffer_mapping
);
607 if (!bh
->b_assoc_map
) {
608 spin_lock(&buffer_mapping
->private_lock
);
609 list_move_tail(&bh
->b_assoc_buffers
,
610 &mapping
->private_list
);
611 bh
->b_assoc_map
= mapping
;
612 spin_unlock(&buffer_mapping
->private_lock
);
615 EXPORT_SYMBOL(mark_buffer_dirty_inode
);
618 * Mark the page dirty, and set it dirty in the radix tree, and mark the inode
621 * If warn is true, then emit a warning if the page is not uptodate and has
622 * not been truncated.
624 * The caller must hold lock_page_memcg().
626 static void __set_page_dirty(struct page
*page
, struct address_space
*mapping
,
631 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
632 if (page
->mapping
) { /* Race with truncate? */
633 WARN_ON_ONCE(warn
&& !PageUptodate(page
));
634 account_page_dirtied(page
, mapping
);
635 radix_tree_tag_set(&mapping
->page_tree
,
636 page_index(page
), PAGECACHE_TAG_DIRTY
);
638 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
642 * Add a page to the dirty page list.
644 * It is a sad fact of life that this function is called from several places
645 * deeply under spinlocking. It may not sleep.
647 * If the page has buffers, the uptodate buffers are set dirty, to preserve
648 * dirty-state coherency between the page and the buffers. It the page does
649 * not have buffers then when they are later attached they will all be set
652 * The buffers are dirtied before the page is dirtied. There's a small race
653 * window in which a writepage caller may see the page cleanness but not the
654 * buffer dirtiness. That's fine. If this code were to set the page dirty
655 * before the buffers, a concurrent writepage caller could clear the page dirty
656 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
657 * page on the dirty page list.
659 * We use private_lock to lock against try_to_free_buffers while using the
660 * page's buffer list. Also use this to protect against clean buffers being
661 * added to the page after it was set dirty.
663 * FIXME: may need to call ->reservepage here as well. That's rather up to the
664 * address_space though.
666 int __set_page_dirty_buffers(struct page
*page
)
669 struct address_space
*mapping
= page_mapping(page
);
671 if (unlikely(!mapping
))
672 return !TestSetPageDirty(page
);
674 spin_lock(&mapping
->private_lock
);
675 if (page_has_buffers(page
)) {
676 struct buffer_head
*head
= page_buffers(page
);
677 struct buffer_head
*bh
= head
;
680 set_buffer_dirty(bh
);
681 bh
= bh
->b_this_page
;
682 } while (bh
!= head
);
685 * Lock out page->mem_cgroup migration to keep PageDirty
686 * synchronized with per-memcg dirty page counters.
688 lock_page_memcg(page
);
689 newly_dirty
= !TestSetPageDirty(page
);
690 spin_unlock(&mapping
->private_lock
);
693 __set_page_dirty(page
, mapping
, 1);
695 unlock_page_memcg(page
);
698 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
702 EXPORT_SYMBOL(__set_page_dirty_buffers
);
705 * Write out and wait upon a list of buffers.
707 * We have conflicting pressures: we want to make sure that all
708 * initially dirty buffers get waited on, but that any subsequently
709 * dirtied buffers don't. After all, we don't want fsync to last
710 * forever if somebody is actively writing to the file.
712 * Do this in two main stages: first we copy dirty buffers to a
713 * temporary inode list, queueing the writes as we go. Then we clean
714 * up, waiting for those writes to complete.
716 * During this second stage, any subsequent updates to the file may end
717 * up refiling the buffer on the original inode's dirty list again, so
718 * there is a chance we will end up with a buffer queued for write but
719 * not yet completed on that list. So, as a final cleanup we go through
720 * the osync code to catch these locked, dirty buffers without requeuing
721 * any newly dirty buffers for write.
723 static int fsync_buffers_list(spinlock_t
*lock
, struct list_head
*list
)
725 struct buffer_head
*bh
;
726 struct list_head tmp
;
727 struct address_space
*mapping
;
729 struct blk_plug plug
;
731 INIT_LIST_HEAD(&tmp
);
732 blk_start_plug(&plug
);
735 while (!list_empty(list
)) {
736 bh
= BH_ENTRY(list
->next
);
737 mapping
= bh
->b_assoc_map
;
738 __remove_assoc_queue(bh
);
739 /* Avoid race with mark_buffer_dirty_inode() which does
740 * a lockless check and we rely on seeing the dirty bit */
742 if (buffer_dirty(bh
) || buffer_locked(bh
)) {
743 list_add(&bh
->b_assoc_buffers
, &tmp
);
744 bh
->b_assoc_map
= mapping
;
745 if (buffer_dirty(bh
)) {
749 * Ensure any pending I/O completes so that
750 * write_dirty_buffer() actually writes the
751 * current contents - it is a noop if I/O is
752 * still in flight on potentially older
755 write_dirty_buffer(bh
, REQ_SYNC
);
758 * Kick off IO for the previous mapping. Note
759 * that we will not run the very last mapping,
760 * wait_on_buffer() will do that for us
761 * through sync_buffer().
770 blk_finish_plug(&plug
);
773 while (!list_empty(&tmp
)) {
774 bh
= BH_ENTRY(tmp
.prev
);
776 mapping
= bh
->b_assoc_map
;
777 __remove_assoc_queue(bh
);
778 /* Avoid race with mark_buffer_dirty_inode() which does
779 * a lockless check and we rely on seeing the dirty bit */
781 if (buffer_dirty(bh
)) {
782 list_add(&bh
->b_assoc_buffers
,
783 &mapping
->private_list
);
784 bh
->b_assoc_map
= mapping
;
788 if (!buffer_uptodate(bh
))
795 err2
= osync_buffers_list(lock
, list
);
803 * Invalidate any and all dirty buffers on a given inode. We are
804 * probably unmounting the fs, but that doesn't mean we have already
805 * done a sync(). Just drop the buffers from the inode list.
807 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
808 * assumes that all the buffers are against the blockdev. Not true
811 void invalidate_inode_buffers(struct inode
*inode
)
813 if (inode_has_buffers(inode
)) {
814 struct address_space
*mapping
= &inode
->i_data
;
815 struct list_head
*list
= &mapping
->private_list
;
816 struct address_space
*buffer_mapping
= mapping
->private_data
;
818 spin_lock(&buffer_mapping
->private_lock
);
819 while (!list_empty(list
))
820 __remove_assoc_queue(BH_ENTRY(list
->next
));
821 spin_unlock(&buffer_mapping
->private_lock
);
824 EXPORT_SYMBOL(invalidate_inode_buffers
);
827 * Remove any clean buffers from the inode's buffer list. This is called
828 * when we're trying to free the inode itself. Those buffers can pin it.
830 * Returns true if all buffers were removed.
832 int remove_inode_buffers(struct inode
*inode
)
836 if (inode_has_buffers(inode
)) {
837 struct address_space
*mapping
= &inode
->i_data
;
838 struct list_head
*list
= &mapping
->private_list
;
839 struct address_space
*buffer_mapping
= mapping
->private_data
;
841 spin_lock(&buffer_mapping
->private_lock
);
842 while (!list_empty(list
)) {
843 struct buffer_head
*bh
= BH_ENTRY(list
->next
);
844 if (buffer_dirty(bh
)) {
848 __remove_assoc_queue(bh
);
850 spin_unlock(&buffer_mapping
->private_lock
);
856 * Create the appropriate buffers when given a page for data area and
857 * the size of each buffer.. Use the bh->b_this_page linked list to
858 * follow the buffers created. Return NULL if unable to create more
861 * The retry flag is used to differentiate async IO (paging, swapping)
862 * which may not fail from ordinary buffer allocations.
864 struct buffer_head
*alloc_page_buffers(struct page
*page
, unsigned long size
,
867 struct buffer_head
*bh
, *head
;
873 while ((offset
-= size
) >= 0) {
874 bh
= alloc_buffer_head(GFP_NOFS
);
878 bh
->b_this_page
= head
;
884 /* Link the buffer to its page */
885 set_bh_page(bh
, page
, offset
);
889 * In case anything failed, we just free everything we got.
895 head
= head
->b_this_page
;
896 free_buffer_head(bh
);
901 * Return failure for non-async IO requests. Async IO requests
902 * are not allowed to fail, so we have to wait until buffer heads
903 * become available. But we don't want tasks sleeping with
904 * partially complete buffers, so all were released above.
909 /* We're _really_ low on memory. Now we just
910 * wait for old buffer heads to become free due to
911 * finishing IO. Since this is an async request and
912 * the reserve list is empty, we're sure there are
913 * async buffer heads in use.
918 EXPORT_SYMBOL_GPL(alloc_page_buffers
);
921 link_dev_buffers(struct page
*page
, struct buffer_head
*head
)
923 struct buffer_head
*bh
, *tail
;
928 bh
= bh
->b_this_page
;
930 tail
->b_this_page
= head
;
931 attach_page_buffers(page
, head
);
934 static sector_t
blkdev_max_block(struct block_device
*bdev
, unsigned int size
)
936 sector_t retval
= ~((sector_t
)0);
937 loff_t sz
= i_size_read(bdev
->bd_inode
);
940 unsigned int sizebits
= blksize_bits(size
);
941 retval
= (sz
>> sizebits
);
947 * Initialise the state of a blockdev page's buffers.
950 init_page_buffers(struct page
*page
, struct block_device
*bdev
,
951 sector_t block
, int size
)
953 struct buffer_head
*head
= page_buffers(page
);
954 struct buffer_head
*bh
= head
;
955 int uptodate
= PageUptodate(page
);
956 sector_t end_block
= blkdev_max_block(I_BDEV(bdev
->bd_inode
), size
);
959 if (!buffer_mapped(bh
)) {
960 init_buffer(bh
, NULL
, NULL
);
962 bh
->b_blocknr
= block
;
964 set_buffer_uptodate(bh
);
965 if (block
< end_block
)
966 set_buffer_mapped(bh
);
969 bh
= bh
->b_this_page
;
970 } while (bh
!= head
);
973 * Caller needs to validate requested block against end of device.
979 * Create the page-cache page that contains the requested block.
981 * This is used purely for blockdev mappings.
984 grow_dev_page(struct block_device
*bdev
, sector_t block
,
985 pgoff_t index
, int size
, int sizebits
, gfp_t gfp
)
987 struct inode
*inode
= bdev
->bd_inode
;
989 struct buffer_head
*bh
;
991 int ret
= 0; /* Will call free_more_memory() */
994 gfp_mask
= mapping_gfp_constraint(inode
->i_mapping
, ~__GFP_FS
) | gfp
;
997 * XXX: __getblk_slow() can not really deal with failure and
998 * will endlessly loop on improvised global reclaim. Prefer
999 * looping in the allocator rather than here, at least that
1000 * code knows what it's doing.
1002 gfp_mask
|= __GFP_NOFAIL
;
1004 page
= find_or_create_page(inode
->i_mapping
, index
, gfp_mask
);
1008 BUG_ON(!PageLocked(page
));
1010 if (page_has_buffers(page
)) {
1011 bh
= page_buffers(page
);
1012 if (bh
->b_size
== size
) {
1013 end_block
= init_page_buffers(page
, bdev
,
1014 (sector_t
)index
<< sizebits
,
1018 if (!try_to_free_buffers(page
))
1023 * Allocate some buffers for this page
1025 bh
= alloc_page_buffers(page
, size
, 0);
1030 * Link the page to the buffers and initialise them. Take the
1031 * lock to be atomic wrt __find_get_block(), which does not
1032 * run under the page lock.
1034 spin_lock(&inode
->i_mapping
->private_lock
);
1035 link_dev_buffers(page
, bh
);
1036 end_block
= init_page_buffers(page
, bdev
, (sector_t
)index
<< sizebits
,
1038 spin_unlock(&inode
->i_mapping
->private_lock
);
1040 ret
= (block
< end_block
) ? 1 : -ENXIO
;
1048 * Create buffers for the specified block device block's page. If
1049 * that page was dirty, the buffers are set dirty also.
1052 grow_buffers(struct block_device
*bdev
, sector_t block
, int size
, gfp_t gfp
)
1060 } while ((size
<< sizebits
) < PAGE_SIZE
);
1062 index
= block
>> sizebits
;
1065 * Check for a block which wants to lie outside our maximum possible
1066 * pagecache index. (this comparison is done using sector_t types).
1068 if (unlikely(index
!= block
>> sizebits
)) {
1069 printk(KERN_ERR
"%s: requested out-of-range block %llu for "
1071 __func__
, (unsigned long long)block
,
1076 /* Create a page with the proper size buffers.. */
1077 return grow_dev_page(bdev
, block
, index
, size
, sizebits
, gfp
);
1080 static struct buffer_head
*
1081 __getblk_slow(struct block_device
*bdev
, sector_t block
,
1082 unsigned size
, gfp_t gfp
)
1084 /* Size must be multiple of hard sectorsize */
1085 if (unlikely(size
& (bdev_logical_block_size(bdev
)-1) ||
1086 (size
< 512 || size
> PAGE_SIZE
))) {
1087 printk(KERN_ERR
"getblk(): invalid block size %d requested\n",
1089 printk(KERN_ERR
"logical block size: %d\n",
1090 bdev_logical_block_size(bdev
));
1097 struct buffer_head
*bh
;
1100 bh
= __find_get_block(bdev
, block
, size
);
1104 ret
= grow_buffers(bdev
, block
, size
, gfp
);
1113 * The relationship between dirty buffers and dirty pages:
1115 * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1116 * the page is tagged dirty in its radix tree.
1118 * At all times, the dirtiness of the buffers represents the dirtiness of
1119 * subsections of the page. If the page has buffers, the page dirty bit is
1120 * merely a hint about the true dirty state.
1122 * When a page is set dirty in its entirety, all its buffers are marked dirty
1123 * (if the page has buffers).
1125 * When a buffer is marked dirty, its page is dirtied, but the page's other
1128 * Also. When blockdev buffers are explicitly read with bread(), they
1129 * individually become uptodate. But their backing page remains not
1130 * uptodate - even if all of its buffers are uptodate. A subsequent
1131 * block_read_full_page() against that page will discover all the uptodate
1132 * buffers, will set the page uptodate and will perform no I/O.
1136 * mark_buffer_dirty - mark a buffer_head as needing writeout
1137 * @bh: the buffer_head to mark dirty
1139 * mark_buffer_dirty() will set the dirty bit against the buffer, then set its
1140 * backing page dirty, then tag the page as dirty in its address_space's radix
1141 * tree and then attach the address_space's inode to its superblock's dirty
1144 * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
1145 * mapping->tree_lock and mapping->host->i_lock.
1147 void mark_buffer_dirty(struct buffer_head
*bh
)
1149 WARN_ON_ONCE(!buffer_uptodate(bh
));
1151 trace_block_dirty_buffer(bh
);
1154 * Very *carefully* optimize the it-is-already-dirty case.
1156 * Don't let the final "is it dirty" escape to before we
1157 * perhaps modified the buffer.
1159 if (buffer_dirty(bh
)) {
1161 if (buffer_dirty(bh
))
1165 if (!test_set_buffer_dirty(bh
)) {
1166 struct page
*page
= bh
->b_page
;
1167 struct address_space
*mapping
= NULL
;
1169 lock_page_memcg(page
);
1170 if (!TestSetPageDirty(page
)) {
1171 mapping
= page_mapping(page
);
1173 __set_page_dirty(page
, mapping
, 0);
1175 unlock_page_memcg(page
);
1177 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
1180 EXPORT_SYMBOL(mark_buffer_dirty
);
1182 void mark_buffer_write_io_error(struct buffer_head
*bh
)
1184 set_buffer_write_io_error(bh
);
1185 /* FIXME: do we need to set this in both places? */
1186 if (bh
->b_page
&& bh
->b_page
->mapping
)
1187 mapping_set_error(bh
->b_page
->mapping
, -EIO
);
1188 if (bh
->b_assoc_map
)
1189 mapping_set_error(bh
->b_assoc_map
, -EIO
);
1191 EXPORT_SYMBOL(mark_buffer_write_io_error
);
1194 * Decrement a buffer_head's reference count. If all buffers against a page
1195 * have zero reference count, are clean and unlocked, and if the page is clean
1196 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1197 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1198 * a page but it ends up not being freed, and buffers may later be reattached).
1200 void __brelse(struct buffer_head
* buf
)
1202 if (atomic_read(&buf
->b_count
)) {
1206 WARN(1, KERN_ERR
"VFS: brelse: Trying to free free buffer\n");
1208 EXPORT_SYMBOL(__brelse
);
1211 * bforget() is like brelse(), except it discards any
1212 * potentially dirty data.
1214 void __bforget(struct buffer_head
*bh
)
1216 clear_buffer_dirty(bh
);
1217 if (bh
->b_assoc_map
) {
1218 struct address_space
*buffer_mapping
= bh
->b_page
->mapping
;
1220 spin_lock(&buffer_mapping
->private_lock
);
1221 list_del_init(&bh
->b_assoc_buffers
);
1222 bh
->b_assoc_map
= NULL
;
1223 spin_unlock(&buffer_mapping
->private_lock
);
1227 EXPORT_SYMBOL(__bforget
);
1229 static struct buffer_head
*__bread_slow(struct buffer_head
*bh
)
1232 if (buffer_uptodate(bh
)) {
1237 bh
->b_end_io
= end_buffer_read_sync
;
1238 submit_bh(REQ_OP_READ
, 0, bh
);
1240 if (buffer_uptodate(bh
))
1248 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1249 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1250 * refcount elevated by one when they're in an LRU. A buffer can only appear
1251 * once in a particular CPU's LRU. A single buffer can be present in multiple
1252 * CPU's LRUs at the same time.
1254 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1255 * sb_find_get_block().
1257 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1258 * a local interrupt disable for that.
1261 #define BH_LRU_SIZE 16
1264 struct buffer_head
*bhs
[BH_LRU_SIZE
];
1267 static DEFINE_PER_CPU(struct bh_lru
, bh_lrus
) = {{ NULL
}};
1270 #define bh_lru_lock() local_irq_disable()
1271 #define bh_lru_unlock() local_irq_enable()
1273 #define bh_lru_lock() preempt_disable()
1274 #define bh_lru_unlock() preempt_enable()
1277 static inline void check_irqs_on(void)
1279 #ifdef irqs_disabled
1280 BUG_ON(irqs_disabled());
1285 * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is
1286 * inserted at the front, and the buffer_head at the back if any is evicted.
1287 * Or, if already in the LRU it is moved to the front.
1289 static void bh_lru_install(struct buffer_head
*bh
)
1291 struct buffer_head
*evictee
= bh
;
1298 b
= this_cpu_ptr(&bh_lrus
);
1299 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1300 swap(evictee
, b
->bhs
[i
]);
1301 if (evictee
== bh
) {
1313 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1315 static struct buffer_head
*
1316 lookup_bh_lru(struct block_device
*bdev
, sector_t block
, unsigned size
)
1318 struct buffer_head
*ret
= NULL
;
1323 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1324 struct buffer_head
*bh
= __this_cpu_read(bh_lrus
.bhs
[i
]);
1326 if (bh
&& bh
->b_blocknr
== block
&& bh
->b_bdev
== bdev
&&
1327 bh
->b_size
== size
) {
1330 __this_cpu_write(bh_lrus
.bhs
[i
],
1331 __this_cpu_read(bh_lrus
.bhs
[i
- 1]));
1334 __this_cpu_write(bh_lrus
.bhs
[0], bh
);
1346 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1347 * it in the LRU and mark it as accessed. If it is not present then return
1350 struct buffer_head
*
1351 __find_get_block(struct block_device
*bdev
, sector_t block
, unsigned size
)
1353 struct buffer_head
*bh
= lookup_bh_lru(bdev
, block
, size
);
1356 /* __find_get_block_slow will mark the page accessed */
1357 bh
= __find_get_block_slow(bdev
, block
);
1365 EXPORT_SYMBOL(__find_get_block
);
1368 * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
1369 * which corresponds to the passed block_device, block and size. The
1370 * returned buffer has its reference count incremented.
1372 * __getblk_gfp() will lock up the machine if grow_dev_page's
1373 * try_to_free_buffers() attempt is failing. FIXME, perhaps?
1375 struct buffer_head
*
1376 __getblk_gfp(struct block_device
*bdev
, sector_t block
,
1377 unsigned size
, gfp_t gfp
)
1379 struct buffer_head
*bh
= __find_get_block(bdev
, block
, size
);
1383 bh
= __getblk_slow(bdev
, block
, size
, gfp
);
1386 EXPORT_SYMBOL(__getblk_gfp
);
1389 * Do async read-ahead on a buffer..
1391 void __breadahead(struct block_device
*bdev
, sector_t block
, unsigned size
)
1393 struct buffer_head
*bh
= __getblk(bdev
, block
, size
);
1395 ll_rw_block(REQ_OP_READ
, REQ_RAHEAD
, 1, &bh
);
1399 EXPORT_SYMBOL(__breadahead
);
1402 * __bread_gfp() - reads a specified block and returns the bh
1403 * @bdev: the block_device to read from
1404 * @block: number of block
1405 * @size: size (in bytes) to read
1406 * @gfp: page allocation flag
1408 * Reads a specified block, and returns buffer head that contains it.
1409 * The page cache can be allocated from non-movable area
1410 * not to prevent page migration if you set gfp to zero.
1411 * It returns NULL if the block was unreadable.
1413 struct buffer_head
*
1414 __bread_gfp(struct block_device
*bdev
, sector_t block
,
1415 unsigned size
, gfp_t gfp
)
1417 struct buffer_head
*bh
= __getblk_gfp(bdev
, block
, size
, gfp
);
1419 if (likely(bh
) && !buffer_uptodate(bh
))
1420 bh
= __bread_slow(bh
);
1423 EXPORT_SYMBOL(__bread_gfp
);
1426 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1427 * This doesn't race because it runs in each cpu either in irq
1428 * or with preempt disabled.
1430 static void invalidate_bh_lru(void *arg
)
1432 struct bh_lru
*b
= &get_cpu_var(bh_lrus
);
1435 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1439 put_cpu_var(bh_lrus
);
1442 static bool has_bh_in_lru(int cpu
, void *dummy
)
1444 struct bh_lru
*b
= per_cpu_ptr(&bh_lrus
, cpu
);
1447 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
1455 void invalidate_bh_lrus(void)
1457 on_each_cpu_cond(has_bh_in_lru
, invalidate_bh_lru
, NULL
, 1, GFP_KERNEL
);
1459 EXPORT_SYMBOL_GPL(invalidate_bh_lrus
);
1461 void set_bh_page(struct buffer_head
*bh
,
1462 struct page
*page
, unsigned long offset
)
1465 BUG_ON(offset
>= PAGE_SIZE
);
1466 if (PageHighMem(page
))
1468 * This catches illegal uses and preserves the offset:
1470 bh
->b_data
= (char *)(0 + offset
);
1472 bh
->b_data
= page_address(page
) + offset
;
1474 EXPORT_SYMBOL(set_bh_page
);
1477 * Called when truncating a buffer on a page completely.
1480 /* Bits that are cleared during an invalidate */
1481 #define BUFFER_FLAGS_DISCARD \
1482 (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1483 1 << BH_Delay | 1 << BH_Unwritten)
1485 static void discard_buffer(struct buffer_head
* bh
)
1487 unsigned long b_state
, b_state_old
;
1490 clear_buffer_dirty(bh
);
1492 b_state
= bh
->b_state
;
1494 b_state_old
= cmpxchg(&bh
->b_state
, b_state
,
1495 (b_state
& ~BUFFER_FLAGS_DISCARD
));
1496 if (b_state_old
== b_state
)
1498 b_state
= b_state_old
;
1504 * block_invalidatepage - invalidate part or all of a buffer-backed page
1506 * @page: the page which is affected
1507 * @offset: start of the range to invalidate
1508 * @length: length of the range to invalidate
1510 * block_invalidatepage() is called when all or part of the page has become
1511 * invalidated by a truncate operation.
1513 * block_invalidatepage() does not have to release all buffers, but it must
1514 * ensure that no dirty buffer is left outside @offset and that no I/O
1515 * is underway against any of the blocks which are outside the truncation
1516 * point. Because the caller is about to free (and possibly reuse) those
1519 void block_invalidatepage(struct page
*page
, unsigned int offset
,
1520 unsigned int length
)
1522 struct buffer_head
*head
, *bh
, *next
;
1523 unsigned int curr_off
= 0;
1524 unsigned int stop
= length
+ offset
;
1526 BUG_ON(!PageLocked(page
));
1527 if (!page_has_buffers(page
))
1531 * Check for overflow
1533 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1535 head
= page_buffers(page
);
1538 unsigned int next_off
= curr_off
+ bh
->b_size
;
1539 next
= bh
->b_this_page
;
1542 * Are we still fully in range ?
1544 if (next_off
> stop
)
1548 * is this block fully invalidated?
1550 if (offset
<= curr_off
)
1552 curr_off
= next_off
;
1554 } while (bh
!= head
);
1557 * We release buffers only if the entire page is being invalidated.
1558 * The get_block cached value has been unconditionally invalidated,
1559 * so real IO is not possible anymore.
1562 try_to_release_page(page
, 0);
1566 EXPORT_SYMBOL(block_invalidatepage
);
1570 * We attach and possibly dirty the buffers atomically wrt
1571 * __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1572 * is already excluded via the page lock.
1574 void create_empty_buffers(struct page
*page
,
1575 unsigned long blocksize
, unsigned long b_state
)
1577 struct buffer_head
*bh
, *head
, *tail
;
1579 head
= alloc_page_buffers(page
, blocksize
, 1);
1582 bh
->b_state
|= b_state
;
1584 bh
= bh
->b_this_page
;
1586 tail
->b_this_page
= head
;
1588 spin_lock(&page
->mapping
->private_lock
);
1589 if (PageUptodate(page
) || PageDirty(page
)) {
1592 if (PageDirty(page
))
1593 set_buffer_dirty(bh
);
1594 if (PageUptodate(page
))
1595 set_buffer_uptodate(bh
);
1596 bh
= bh
->b_this_page
;
1597 } while (bh
!= head
);
1599 attach_page_buffers(page
, head
);
1600 spin_unlock(&page
->mapping
->private_lock
);
1602 EXPORT_SYMBOL(create_empty_buffers
);
1605 * clean_bdev_aliases: clean a range of buffers in block device
1606 * @bdev: Block device to clean buffers in
1607 * @block: Start of a range of blocks to clean
1608 * @len: Number of blocks to clean
1610 * We are taking a range of blocks for data and we don't want writeback of any
1611 * buffer-cache aliases starting from return from this function and until the
1612 * moment when something will explicitly mark the buffer dirty (hopefully that
1613 * will not happen until we will free that block ;-) We don't even need to mark
1614 * it not-uptodate - nobody can expect anything from a newly allocated buffer
1615 * anyway. We used to use unmap_buffer() for such invalidation, but that was
1616 * wrong. We definitely don't want to mark the alias unmapped, for example - it
1617 * would confuse anyone who might pick it with bread() afterwards...
1619 * Also.. Note that bforget() doesn't lock the buffer. So there can be
1620 * writeout I/O going on against recently-freed buffers. We don't wait on that
1621 * I/O in bforget() - it's more efficient to wait on the I/O only if we really
1622 * need to. That happens here.
1624 void clean_bdev_aliases(struct block_device
*bdev
, sector_t block
, sector_t len
)
1626 struct inode
*bd_inode
= bdev
->bd_inode
;
1627 struct address_space
*bd_mapping
= bd_inode
->i_mapping
;
1628 struct pagevec pvec
;
1629 pgoff_t index
= block
>> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
1632 struct buffer_head
*bh
;
1633 struct buffer_head
*head
;
1635 end
= (block
+ len
- 1) >> (PAGE_SHIFT
- bd_inode
->i_blkbits
);
1636 pagevec_init(&pvec
, 0);
1637 while (pagevec_lookup_range(&pvec
, bd_mapping
, &index
, end
)) {
1638 count
= pagevec_count(&pvec
);
1639 for (i
= 0; i
< count
; i
++) {
1640 struct page
*page
= pvec
.pages
[i
];
1642 if (!page_has_buffers(page
))
1645 * We use page lock instead of bd_mapping->private_lock
1646 * to pin buffers here since we can afford to sleep and
1647 * it scales better than a global spinlock lock.
1650 /* Recheck when the page is locked which pins bhs */
1651 if (!page_has_buffers(page
))
1653 head
= page_buffers(page
);
1656 if (!buffer_mapped(bh
) || (bh
->b_blocknr
< block
))
1658 if (bh
->b_blocknr
>= block
+ len
)
1660 clear_buffer_dirty(bh
);
1662 clear_buffer_req(bh
);
1664 bh
= bh
->b_this_page
;
1665 } while (bh
!= head
);
1669 pagevec_release(&pvec
);
1671 /* End of range already reached? */
1672 if (index
> end
|| !index
)
1676 EXPORT_SYMBOL(clean_bdev_aliases
);
1679 * Size is a power-of-two in the range 512..PAGE_SIZE,
1680 * and the case we care about most is PAGE_SIZE.
1682 * So this *could* possibly be written with those
1683 * constraints in mind (relevant mostly if some
1684 * architecture has a slow bit-scan instruction)
1686 static inline int block_size_bits(unsigned int blocksize
)
1688 return ilog2(blocksize
);
1691 static struct buffer_head
*create_page_buffers(struct page
*page
, struct inode
*inode
, unsigned int b_state
)
1693 BUG_ON(!PageLocked(page
));
1695 if (!page_has_buffers(page
))
1696 create_empty_buffers(page
, 1 << ACCESS_ONCE(inode
->i_blkbits
), b_state
);
1697 return page_buffers(page
);
1701 * NOTE! All mapped/uptodate combinations are valid:
1703 * Mapped Uptodate Meaning
1705 * No No "unknown" - must do get_block()
1706 * No Yes "hole" - zero-filled
1707 * Yes No "allocated" - allocated on disk, not read in
1708 * Yes Yes "valid" - allocated and up-to-date in memory.
1710 * "Dirty" is valid only with the last case (mapped+uptodate).
1714 * While block_write_full_page is writing back the dirty buffers under
1715 * the page lock, whoever dirtied the buffers may decide to clean them
1716 * again at any time. We handle that by only looking at the buffer
1717 * state inside lock_buffer().
1719 * If block_write_full_page() is called for regular writeback
1720 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1721 * locked buffer. This only can happen if someone has written the buffer
1722 * directly, with submit_bh(). At the address_space level PageWriteback
1723 * prevents this contention from occurring.
1725 * If block_write_full_page() is called with wbc->sync_mode ==
1726 * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this
1727 * causes the writes to be flagged as synchronous writes.
1729 int __block_write_full_page(struct inode
*inode
, struct page
*page
,
1730 get_block_t
*get_block
, struct writeback_control
*wbc
,
1731 bh_end_io_t
*handler
)
1735 sector_t last_block
;
1736 struct buffer_head
*bh
, *head
;
1737 unsigned int blocksize
, bbits
;
1738 int nr_underway
= 0;
1739 int write_flags
= wbc_to_write_flags(wbc
);
1741 head
= create_page_buffers(page
, inode
,
1742 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
1745 * Be very careful. We have no exclusion from __set_page_dirty_buffers
1746 * here, and the (potentially unmapped) buffers may become dirty at
1747 * any time. If a buffer becomes dirty here after we've inspected it
1748 * then we just miss that fact, and the page stays dirty.
1750 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1751 * handle that here by just cleaning them.
1755 blocksize
= bh
->b_size
;
1756 bbits
= block_size_bits(blocksize
);
1758 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1759 last_block
= (i_size_read(inode
) - 1) >> bbits
;
1762 * Get all the dirty buffers mapped to disk addresses and
1763 * handle any aliases from the underlying blockdev's mapping.
1766 if (block
> last_block
) {
1768 * mapped buffers outside i_size will occur, because
1769 * this page can be outside i_size when there is a
1770 * truncate in progress.
1773 * The buffer was zeroed by block_write_full_page()
1775 clear_buffer_dirty(bh
);
1776 set_buffer_uptodate(bh
);
1777 } else if ((!buffer_mapped(bh
) || buffer_delay(bh
)) &&
1779 WARN_ON(bh
->b_size
!= blocksize
);
1780 err
= get_block(inode
, block
, bh
, 1);
1783 clear_buffer_delay(bh
);
1784 if (buffer_new(bh
)) {
1785 /* blockdev mappings never come here */
1786 clear_buffer_new(bh
);
1787 clean_bdev_bh_alias(bh
);
1790 bh
= bh
->b_this_page
;
1792 } while (bh
!= head
);
1795 if (!buffer_mapped(bh
))
1798 * If it's a fully non-blocking write attempt and we cannot
1799 * lock the buffer then redirty the page. Note that this can
1800 * potentially cause a busy-wait loop from writeback threads
1801 * and kswapd activity, but those code paths have their own
1802 * higher-level throttling.
1804 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
1806 } else if (!trylock_buffer(bh
)) {
1807 redirty_page_for_writepage(wbc
, page
);
1810 if (test_clear_buffer_dirty(bh
)) {
1811 mark_buffer_async_write_endio(bh
, handler
);
1815 } while ((bh
= bh
->b_this_page
) != head
);
1818 * The page and its buffers are protected by PageWriteback(), so we can
1819 * drop the bh refcounts early.
1821 BUG_ON(PageWriteback(page
));
1822 set_page_writeback(page
);
1825 struct buffer_head
*next
= bh
->b_this_page
;
1826 if (buffer_async_write(bh
)) {
1827 submit_bh_wbc(REQ_OP_WRITE
, write_flags
, bh
,
1828 inode
->i_write_hint
, wbc
);
1832 } while (bh
!= head
);
1837 if (nr_underway
== 0) {
1839 * The page was marked dirty, but the buffers were
1840 * clean. Someone wrote them back by hand with
1841 * ll_rw_block/submit_bh. A rare case.
1843 end_page_writeback(page
);
1846 * The page and buffer_heads can be released at any time from
1854 * ENOSPC, or some other error. We may already have added some
1855 * blocks to the file, so we need to write these out to avoid
1856 * exposing stale data.
1857 * The page is currently locked and not marked for writeback
1860 /* Recovery: lock and submit the mapped buffers */
1862 if (buffer_mapped(bh
) && buffer_dirty(bh
) &&
1863 !buffer_delay(bh
)) {
1865 mark_buffer_async_write_endio(bh
, handler
);
1868 * The buffer may have been set dirty during
1869 * attachment to a dirty page.
1871 clear_buffer_dirty(bh
);
1873 } while ((bh
= bh
->b_this_page
) != head
);
1875 BUG_ON(PageWriteback(page
));
1876 mapping_set_error(page
->mapping
, err
);
1877 set_page_writeback(page
);
1879 struct buffer_head
*next
= bh
->b_this_page
;
1880 if (buffer_async_write(bh
)) {
1881 clear_buffer_dirty(bh
);
1882 submit_bh_wbc(REQ_OP_WRITE
, write_flags
, bh
,
1883 inode
->i_write_hint
, wbc
);
1887 } while (bh
!= head
);
1891 EXPORT_SYMBOL(__block_write_full_page
);
1894 * If a page has any new buffers, zero them out here, and mark them uptodate
1895 * and dirty so they'll be written out (in order to prevent uninitialised
1896 * block data from leaking). And clear the new bit.
1898 void page_zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1900 unsigned int block_start
, block_end
;
1901 struct buffer_head
*head
, *bh
;
1903 BUG_ON(!PageLocked(page
));
1904 if (!page_has_buffers(page
))
1907 bh
= head
= page_buffers(page
);
1910 block_end
= block_start
+ bh
->b_size
;
1912 if (buffer_new(bh
)) {
1913 if (block_end
> from
&& block_start
< to
) {
1914 if (!PageUptodate(page
)) {
1915 unsigned start
, size
;
1917 start
= max(from
, block_start
);
1918 size
= min(to
, block_end
) - start
;
1920 zero_user(page
, start
, size
);
1921 set_buffer_uptodate(bh
);
1924 clear_buffer_new(bh
);
1925 mark_buffer_dirty(bh
);
1929 block_start
= block_end
;
1930 bh
= bh
->b_this_page
;
1931 } while (bh
!= head
);
1933 EXPORT_SYMBOL(page_zero_new_buffers
);
1936 iomap_to_bh(struct inode
*inode
, sector_t block
, struct buffer_head
*bh
,
1937 struct iomap
*iomap
)
1939 loff_t offset
= block
<< inode
->i_blkbits
;
1941 bh
->b_bdev
= iomap
->bdev
;
1944 * Block points to offset in file we need to map, iomap contains
1945 * the offset at which the map starts. If the map ends before the
1946 * current block, then do not map the buffer and let the caller
1949 BUG_ON(offset
>= iomap
->offset
+ iomap
->length
);
1951 switch (iomap
->type
) {
1954 * If the buffer is not up to date or beyond the current EOF,
1955 * we need to mark it as new to ensure sub-block zeroing is
1956 * executed if necessary.
1958 if (!buffer_uptodate(bh
) ||
1959 (offset
>= i_size_read(inode
)))
1962 case IOMAP_DELALLOC
:
1963 if (!buffer_uptodate(bh
) ||
1964 (offset
>= i_size_read(inode
)))
1966 set_buffer_uptodate(bh
);
1967 set_buffer_mapped(bh
);
1968 set_buffer_delay(bh
);
1970 case IOMAP_UNWRITTEN
:
1972 * For unwritten regions, we always need to ensure that
1973 * sub-block writes cause the regions in the block we are not
1974 * writing to are zeroed. Set the buffer as new to ensure this.
1977 set_buffer_unwritten(bh
);
1980 if (offset
>= i_size_read(inode
))
1982 bh
->b_blocknr
= (iomap
->blkno
>> (inode
->i_blkbits
- 9)) +
1983 ((offset
- iomap
->offset
) >> inode
->i_blkbits
);
1984 set_buffer_mapped(bh
);
1989 int __block_write_begin_int(struct page
*page
, loff_t pos
, unsigned len
,
1990 get_block_t
*get_block
, struct iomap
*iomap
)
1992 unsigned from
= pos
& (PAGE_SIZE
- 1);
1993 unsigned to
= from
+ len
;
1994 struct inode
*inode
= page
->mapping
->host
;
1995 unsigned block_start
, block_end
;
1998 unsigned blocksize
, bbits
;
1999 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
=wait
;
2001 BUG_ON(!PageLocked(page
));
2002 BUG_ON(from
> PAGE_SIZE
);
2003 BUG_ON(to
> PAGE_SIZE
);
2006 head
= create_page_buffers(page
, inode
, 0);
2007 blocksize
= head
->b_size
;
2008 bbits
= block_size_bits(blocksize
);
2010 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
2012 for(bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
2013 block
++, block_start
=block_end
, bh
= bh
->b_this_page
) {
2014 block_end
= block_start
+ blocksize
;
2015 if (block_end
<= from
|| block_start
>= to
) {
2016 if (PageUptodate(page
)) {
2017 if (!buffer_uptodate(bh
))
2018 set_buffer_uptodate(bh
);
2023 clear_buffer_new(bh
);
2024 if (!buffer_mapped(bh
)) {
2025 WARN_ON(bh
->b_size
!= blocksize
);
2027 err
= get_block(inode
, block
, bh
, 1);
2031 iomap_to_bh(inode
, block
, bh
, iomap
);
2034 if (buffer_new(bh
)) {
2035 clean_bdev_bh_alias(bh
);
2036 if (PageUptodate(page
)) {
2037 clear_buffer_new(bh
);
2038 set_buffer_uptodate(bh
);
2039 mark_buffer_dirty(bh
);
2042 if (block_end
> to
|| block_start
< from
)
2043 zero_user_segments(page
,
2049 if (PageUptodate(page
)) {
2050 if (!buffer_uptodate(bh
))
2051 set_buffer_uptodate(bh
);
2054 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
2055 !buffer_unwritten(bh
) &&
2056 (block_start
< from
|| block_end
> to
)) {
2057 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
2062 * If we issued read requests - let them complete.
2064 while(wait_bh
> wait
) {
2065 wait_on_buffer(*--wait_bh
);
2066 if (!buffer_uptodate(*wait_bh
))
2070 page_zero_new_buffers(page
, from
, to
);
2074 int __block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
2075 get_block_t
*get_block
)
2077 return __block_write_begin_int(page
, pos
, len
, get_block
, NULL
);
2079 EXPORT_SYMBOL(__block_write_begin
);
2081 static int __block_commit_write(struct inode
*inode
, struct page
*page
,
2082 unsigned from
, unsigned to
)
2084 unsigned block_start
, block_end
;
2087 struct buffer_head
*bh
, *head
;
2089 bh
= head
= page_buffers(page
);
2090 blocksize
= bh
->b_size
;
2094 block_end
= block_start
+ blocksize
;
2095 if (block_end
<= from
|| block_start
>= to
) {
2096 if (!buffer_uptodate(bh
))
2099 set_buffer_uptodate(bh
);
2100 mark_buffer_dirty(bh
);
2102 clear_buffer_new(bh
);
2104 block_start
= block_end
;
2105 bh
= bh
->b_this_page
;
2106 } while (bh
!= head
);
2109 * If this is a partial write which happened to make all buffers
2110 * uptodate then we can optimize away a bogus readpage() for
2111 * the next read(). Here we 'discover' whether the page went
2112 * uptodate as a result of this (potentially partial) write.
2115 SetPageUptodate(page
);
2120 * block_write_begin takes care of the basic task of block allocation and
2121 * bringing partial write blocks uptodate first.
2123 * The filesystem needs to handle block truncation upon failure.
2125 int block_write_begin(struct address_space
*mapping
, loff_t pos
, unsigned len
,
2126 unsigned flags
, struct page
**pagep
, get_block_t
*get_block
)
2128 pgoff_t index
= pos
>> PAGE_SHIFT
;
2132 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2136 status
= __block_write_begin(page
, pos
, len
, get_block
);
2137 if (unlikely(status
)) {
2146 EXPORT_SYMBOL(block_write_begin
);
2148 int block_write_end(struct file
*file
, struct address_space
*mapping
,
2149 loff_t pos
, unsigned len
, unsigned copied
,
2150 struct page
*page
, void *fsdata
)
2152 struct inode
*inode
= mapping
->host
;
2155 start
= pos
& (PAGE_SIZE
- 1);
2157 if (unlikely(copied
< len
)) {
2159 * The buffers that were written will now be uptodate, so we
2160 * don't have to worry about a readpage reading them and
2161 * overwriting a partial write. However if we have encountered
2162 * a short write and only partially written into a buffer, it
2163 * will not be marked uptodate, so a readpage might come in and
2164 * destroy our partial write.
2166 * Do the simplest thing, and just treat any short write to a
2167 * non uptodate page as a zero-length write, and force the
2168 * caller to redo the whole thing.
2170 if (!PageUptodate(page
))
2173 page_zero_new_buffers(page
, start
+copied
, start
+len
);
2175 flush_dcache_page(page
);
2177 /* This could be a short (even 0-length) commit */
2178 __block_commit_write(inode
, page
, start
, start
+copied
);
2182 EXPORT_SYMBOL(block_write_end
);
2184 int generic_write_end(struct file
*file
, struct address_space
*mapping
,
2185 loff_t pos
, unsigned len
, unsigned copied
,
2186 struct page
*page
, void *fsdata
)
2188 struct inode
*inode
= mapping
->host
;
2189 loff_t old_size
= inode
->i_size
;
2190 int i_size_changed
= 0;
2192 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
2195 * No need to use i_size_read() here, the i_size
2196 * cannot change under us because we hold i_mutex.
2198 * But it's important to update i_size while still holding page lock:
2199 * page writeout could otherwise come in and zero beyond i_size.
2201 if (pos
+copied
> inode
->i_size
) {
2202 i_size_write(inode
, pos
+copied
);
2210 pagecache_isize_extended(inode
, old_size
, pos
);
2212 * Don't mark the inode dirty under page lock. First, it unnecessarily
2213 * makes the holding time of page lock longer. Second, it forces lock
2214 * ordering of page lock and transaction start for journaling
2218 mark_inode_dirty(inode
);
2222 EXPORT_SYMBOL(generic_write_end
);
2225 * block_is_partially_uptodate checks whether buffers within a page are
2228 * Returns true if all buffers which correspond to a file portion
2229 * we want to read are uptodate.
2231 int block_is_partially_uptodate(struct page
*page
, unsigned long from
,
2232 unsigned long count
)
2234 unsigned block_start
, block_end
, blocksize
;
2236 struct buffer_head
*bh
, *head
;
2239 if (!page_has_buffers(page
))
2242 head
= page_buffers(page
);
2243 blocksize
= head
->b_size
;
2244 to
= min_t(unsigned, PAGE_SIZE
- from
, count
);
2246 if (from
< blocksize
&& to
> PAGE_SIZE
- blocksize
)
2252 block_end
= block_start
+ blocksize
;
2253 if (block_end
> from
&& block_start
< to
) {
2254 if (!buffer_uptodate(bh
)) {
2258 if (block_end
>= to
)
2261 block_start
= block_end
;
2262 bh
= bh
->b_this_page
;
2263 } while (bh
!= head
);
2267 EXPORT_SYMBOL(block_is_partially_uptodate
);
2270 * Generic "read page" function for block devices that have the normal
2271 * get_block functionality. This is most of the block device filesystems.
2272 * Reads the page asynchronously --- the unlock_buffer() and
2273 * set/clear_buffer_uptodate() functions propagate buffer state into the
2274 * page struct once IO has completed.
2276 int block_read_full_page(struct page
*page
, get_block_t
*get_block
)
2278 struct inode
*inode
= page
->mapping
->host
;
2279 sector_t iblock
, lblock
;
2280 struct buffer_head
*bh
, *head
, *arr
[MAX_BUF_PER_PAGE
];
2281 unsigned int blocksize
, bbits
;
2283 int fully_mapped
= 1;
2285 head
= create_page_buffers(page
, inode
, 0);
2286 blocksize
= head
->b_size
;
2287 bbits
= block_size_bits(blocksize
);
2289 iblock
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
2290 lblock
= (i_size_read(inode
)+blocksize
-1) >> bbits
;
2296 if (buffer_uptodate(bh
))
2299 if (!buffer_mapped(bh
)) {
2303 if (iblock
< lblock
) {
2304 WARN_ON(bh
->b_size
!= blocksize
);
2305 err
= get_block(inode
, iblock
, bh
, 0);
2309 if (!buffer_mapped(bh
)) {
2310 zero_user(page
, i
* blocksize
, blocksize
);
2312 set_buffer_uptodate(bh
);
2316 * get_block() might have updated the buffer
2319 if (buffer_uptodate(bh
))
2323 } while (i
++, iblock
++, (bh
= bh
->b_this_page
) != head
);
2326 SetPageMappedToDisk(page
);
2330 * All buffers are uptodate - we can set the page uptodate
2331 * as well. But not if get_block() returned an error.
2333 if (!PageError(page
))
2334 SetPageUptodate(page
);
2339 /* Stage two: lock the buffers */
2340 for (i
= 0; i
< nr
; i
++) {
2343 mark_buffer_async_read(bh
);
2347 * Stage 3: start the IO. Check for uptodateness
2348 * inside the buffer lock in case another process reading
2349 * the underlying blockdev brought it uptodate (the sct fix).
2351 for (i
= 0; i
< nr
; i
++) {
2353 if (buffer_uptodate(bh
))
2354 end_buffer_async_read(bh
, 1);
2356 submit_bh(REQ_OP_READ
, 0, bh
);
2360 EXPORT_SYMBOL(block_read_full_page
);
2362 /* utility function for filesystems that need to do work on expanding
2363 * truncates. Uses filesystem pagecache writes to allow the filesystem to
2364 * deal with the hole.
2366 int generic_cont_expand_simple(struct inode
*inode
, loff_t size
)
2368 struct address_space
*mapping
= inode
->i_mapping
;
2373 err
= inode_newsize_ok(inode
, size
);
2377 err
= pagecache_write_begin(NULL
, mapping
, size
, 0,
2378 AOP_FLAG_CONT_EXPAND
, &page
, &fsdata
);
2382 err
= pagecache_write_end(NULL
, mapping
, size
, 0, 0, page
, fsdata
);
2388 EXPORT_SYMBOL(generic_cont_expand_simple
);
2390 static int cont_expand_zero(struct file
*file
, struct address_space
*mapping
,
2391 loff_t pos
, loff_t
*bytes
)
2393 struct inode
*inode
= mapping
->host
;
2394 unsigned int blocksize
= i_blocksize(inode
);
2397 pgoff_t index
, curidx
;
2399 unsigned zerofrom
, offset
, len
;
2402 index
= pos
>> PAGE_SHIFT
;
2403 offset
= pos
& ~PAGE_MASK
;
2405 while (index
> (curidx
= (curpos
= *bytes
)>>PAGE_SHIFT
)) {
2406 zerofrom
= curpos
& ~PAGE_MASK
;
2407 if (zerofrom
& (blocksize
-1)) {
2408 *bytes
|= (blocksize
-1);
2411 len
= PAGE_SIZE
- zerofrom
;
2413 err
= pagecache_write_begin(file
, mapping
, curpos
, len
, 0,
2417 zero_user(page
, zerofrom
, len
);
2418 err
= pagecache_write_end(file
, mapping
, curpos
, len
, len
,
2425 balance_dirty_pages_ratelimited(mapping
);
2427 if (unlikely(fatal_signal_pending(current
))) {
2433 /* page covers the boundary, find the boundary offset */
2434 if (index
== curidx
) {
2435 zerofrom
= curpos
& ~PAGE_MASK
;
2436 /* if we will expand the thing last block will be filled */
2437 if (offset
<= zerofrom
) {
2440 if (zerofrom
& (blocksize
-1)) {
2441 *bytes
|= (blocksize
-1);
2444 len
= offset
- zerofrom
;
2446 err
= pagecache_write_begin(file
, mapping
, curpos
, len
, 0,
2450 zero_user(page
, zerofrom
, len
);
2451 err
= pagecache_write_end(file
, mapping
, curpos
, len
, len
,
2463 * For moronic filesystems that do not allow holes in file.
2464 * We may have to extend the file.
2466 int cont_write_begin(struct file
*file
, struct address_space
*mapping
,
2467 loff_t pos
, unsigned len
, unsigned flags
,
2468 struct page
**pagep
, void **fsdata
,
2469 get_block_t
*get_block
, loff_t
*bytes
)
2471 struct inode
*inode
= mapping
->host
;
2472 unsigned int blocksize
= i_blocksize(inode
);
2473 unsigned int zerofrom
;
2476 err
= cont_expand_zero(file
, mapping
, pos
, bytes
);
2480 zerofrom
= *bytes
& ~PAGE_MASK
;
2481 if (pos
+len
> *bytes
&& zerofrom
& (blocksize
-1)) {
2482 *bytes
|= (blocksize
-1);
2486 return block_write_begin(mapping
, pos
, len
, flags
, pagep
, get_block
);
2488 EXPORT_SYMBOL(cont_write_begin
);
2490 int block_commit_write(struct page
*page
, unsigned from
, unsigned to
)
2492 struct inode
*inode
= page
->mapping
->host
;
2493 __block_commit_write(inode
,page
,from
,to
);
2496 EXPORT_SYMBOL(block_commit_write
);
2499 * block_page_mkwrite() is not allowed to change the file size as it gets
2500 * called from a page fault handler when a page is first dirtied. Hence we must
2501 * be careful to check for EOF conditions here. We set the page up correctly
2502 * for a written page which means we get ENOSPC checking when writing into
2503 * holes and correct delalloc and unwritten extent mapping on filesystems that
2504 * support these features.
2506 * We are not allowed to take the i_mutex here so we have to play games to
2507 * protect against truncate races as the page could now be beyond EOF. Because
2508 * truncate writes the inode size before removing pages, once we have the
2509 * page lock we can determine safely if the page is beyond EOF. If it is not
2510 * beyond EOF, then the page is guaranteed safe against truncation until we
2513 * Direct callers of this function should protect against filesystem freezing
2514 * using sb_start_pagefault() - sb_end_pagefault() functions.
2516 int block_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
2517 get_block_t get_block
)
2519 struct page
*page
= vmf
->page
;
2520 struct inode
*inode
= file_inode(vma
->vm_file
);
2526 size
= i_size_read(inode
);
2527 if ((page
->mapping
!= inode
->i_mapping
) ||
2528 (page_offset(page
) > size
)) {
2529 /* We overload EFAULT to mean page got truncated */
2534 /* page is wholly or partially inside EOF */
2535 if (((page
->index
+ 1) << PAGE_SHIFT
) > size
)
2536 end
= size
& ~PAGE_MASK
;
2540 ret
= __block_write_begin(page
, 0, end
, get_block
);
2542 ret
= block_commit_write(page
, 0, end
);
2544 if (unlikely(ret
< 0))
2546 set_page_dirty(page
);
2547 wait_for_stable_page(page
);
2553 EXPORT_SYMBOL(block_page_mkwrite
);
2556 * nobh_write_begin()'s prereads are special: the buffer_heads are freed
2557 * immediately, while under the page lock. So it needs a special end_io
2558 * handler which does not touch the bh after unlocking it.
2560 static void end_buffer_read_nobh(struct buffer_head
*bh
, int uptodate
)
2562 __end_buffer_read_notouch(bh
, uptodate
);
2566 * Attach the singly-linked list of buffers created by nobh_write_begin, to
2567 * the page (converting it to circular linked list and taking care of page
2570 static void attach_nobh_buffers(struct page
*page
, struct buffer_head
*head
)
2572 struct buffer_head
*bh
;
2574 BUG_ON(!PageLocked(page
));
2576 spin_lock(&page
->mapping
->private_lock
);
2579 if (PageDirty(page
))
2580 set_buffer_dirty(bh
);
2581 if (!bh
->b_this_page
)
2582 bh
->b_this_page
= head
;
2583 bh
= bh
->b_this_page
;
2584 } while (bh
!= head
);
2585 attach_page_buffers(page
, head
);
2586 spin_unlock(&page
->mapping
->private_lock
);
2590 * On entry, the page is fully not uptodate.
2591 * On exit the page is fully uptodate in the areas outside (from,to)
2592 * The filesystem needs to handle block truncation upon failure.
2594 int nobh_write_begin(struct address_space
*mapping
,
2595 loff_t pos
, unsigned len
, unsigned flags
,
2596 struct page
**pagep
, void **fsdata
,
2597 get_block_t
*get_block
)
2599 struct inode
*inode
= mapping
->host
;
2600 const unsigned blkbits
= inode
->i_blkbits
;
2601 const unsigned blocksize
= 1 << blkbits
;
2602 struct buffer_head
*head
, *bh
;
2606 unsigned block_in_page
;
2607 unsigned block_start
, block_end
;
2608 sector_t block_in_file
;
2611 int is_mapped_to_disk
= 1;
2613 index
= pos
>> PAGE_SHIFT
;
2614 from
= pos
& (PAGE_SIZE
- 1);
2617 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2623 if (page_has_buffers(page
)) {
2624 ret
= __block_write_begin(page
, pos
, len
, get_block
);
2630 if (PageMappedToDisk(page
))
2634 * Allocate buffers so that we can keep track of state, and potentially
2635 * attach them to the page if an error occurs. In the common case of
2636 * no error, they will just be freed again without ever being attached
2637 * to the page (which is all OK, because we're under the page lock).
2639 * Be careful: the buffer linked list is a NULL terminated one, rather
2640 * than the circular one we're used to.
2642 head
= alloc_page_buffers(page
, blocksize
, 0);
2648 block_in_file
= (sector_t
)page
->index
<< (PAGE_SHIFT
- blkbits
);
2651 * We loop across all blocks in the page, whether or not they are
2652 * part of the affected region. This is so we can discover if the
2653 * page is fully mapped-to-disk.
2655 for (block_start
= 0, block_in_page
= 0, bh
= head
;
2656 block_start
< PAGE_SIZE
;
2657 block_in_page
++, block_start
+= blocksize
, bh
= bh
->b_this_page
) {
2660 block_end
= block_start
+ blocksize
;
2663 if (block_start
>= to
)
2665 ret
= get_block(inode
, block_in_file
+ block_in_page
,
2669 if (!buffer_mapped(bh
))
2670 is_mapped_to_disk
= 0;
2672 clean_bdev_bh_alias(bh
);
2673 if (PageUptodate(page
)) {
2674 set_buffer_uptodate(bh
);
2677 if (buffer_new(bh
) || !buffer_mapped(bh
)) {
2678 zero_user_segments(page
, block_start
, from
,
2682 if (buffer_uptodate(bh
))
2683 continue; /* reiserfs does this */
2684 if (block_start
< from
|| block_end
> to
) {
2686 bh
->b_end_io
= end_buffer_read_nobh
;
2687 submit_bh(REQ_OP_READ
, 0, bh
);
2694 * The page is locked, so these buffers are protected from
2695 * any VM or truncate activity. Hence we don't need to care
2696 * for the buffer_head refcounts.
2698 for (bh
= head
; bh
; bh
= bh
->b_this_page
) {
2700 if (!buffer_uptodate(bh
))
2707 if (is_mapped_to_disk
)
2708 SetPageMappedToDisk(page
);
2710 *fsdata
= head
; /* to be released by nobh_write_end */
2717 * Error recovery is a bit difficult. We need to zero out blocks that
2718 * were newly allocated, and dirty them to ensure they get written out.
2719 * Buffers need to be attached to the page at this point, otherwise
2720 * the handling of potential IO errors during writeout would be hard
2721 * (could try doing synchronous writeout, but what if that fails too?)
2723 attach_nobh_buffers(page
, head
);
2724 page_zero_new_buffers(page
, from
, to
);
2733 EXPORT_SYMBOL(nobh_write_begin
);
2735 int nobh_write_end(struct file
*file
, struct address_space
*mapping
,
2736 loff_t pos
, unsigned len
, unsigned copied
,
2737 struct page
*page
, void *fsdata
)
2739 struct inode
*inode
= page
->mapping
->host
;
2740 struct buffer_head
*head
= fsdata
;
2741 struct buffer_head
*bh
;
2742 BUG_ON(fsdata
!= NULL
&& page_has_buffers(page
));
2744 if (unlikely(copied
< len
) && head
)
2745 attach_nobh_buffers(page
, head
);
2746 if (page_has_buffers(page
))
2747 return generic_write_end(file
, mapping
, pos
, len
,
2748 copied
, page
, fsdata
);
2750 SetPageUptodate(page
);
2751 set_page_dirty(page
);
2752 if (pos
+copied
> inode
->i_size
) {
2753 i_size_write(inode
, pos
+copied
);
2754 mark_inode_dirty(inode
);
2762 head
= head
->b_this_page
;
2763 free_buffer_head(bh
);
2768 EXPORT_SYMBOL(nobh_write_end
);
2771 * nobh_writepage() - based on block_full_write_page() except
2772 * that it tries to operate without attaching bufferheads to
2775 int nobh_writepage(struct page
*page
, get_block_t
*get_block
,
2776 struct writeback_control
*wbc
)
2778 struct inode
* const inode
= page
->mapping
->host
;
2779 loff_t i_size
= i_size_read(inode
);
2780 const pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2784 /* Is the page fully inside i_size? */
2785 if (page
->index
< end_index
)
2788 /* Is the page fully outside i_size? (truncate in progress) */
2789 offset
= i_size
& (PAGE_SIZE
-1);
2790 if (page
->index
>= end_index
+1 || !offset
) {
2792 * The page may have dirty, unmapped buffers. For example,
2793 * they may have been added in ext3_writepage(). Make them
2794 * freeable here, so the page does not leak.
2797 /* Not really sure about this - do we need this ? */
2798 if (page
->mapping
->a_ops
->invalidatepage
)
2799 page
->mapping
->a_ops
->invalidatepage(page
, offset
);
2802 return 0; /* don't care */
2806 * The page straddles i_size. It must be zeroed out on each and every
2807 * writepage invocation because it may be mmapped. "A file is mapped
2808 * in multiples of the page size. For a file that is not a multiple of
2809 * the page size, the remaining memory is zeroed when mapped, and
2810 * writes to that region are not written out to the file."
2812 zero_user_segment(page
, offset
, PAGE_SIZE
);
2814 ret
= mpage_writepage(page
, get_block
, wbc
);
2816 ret
= __block_write_full_page(inode
, page
, get_block
, wbc
,
2817 end_buffer_async_write
);
2820 EXPORT_SYMBOL(nobh_writepage
);
2822 int nobh_truncate_page(struct address_space
*mapping
,
2823 loff_t from
, get_block_t
*get_block
)
2825 pgoff_t index
= from
>> PAGE_SHIFT
;
2826 unsigned offset
= from
& (PAGE_SIZE
-1);
2829 unsigned length
, pos
;
2830 struct inode
*inode
= mapping
->host
;
2832 struct buffer_head map_bh
;
2835 blocksize
= i_blocksize(inode
);
2836 length
= offset
& (blocksize
- 1);
2838 /* Block boundary? Nothing to do */
2842 length
= blocksize
- length
;
2843 iblock
= (sector_t
)index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
2845 page
= grab_cache_page(mapping
, index
);
2850 if (page_has_buffers(page
)) {
2854 return block_truncate_page(mapping
, from
, get_block
);
2857 /* Find the buffer that contains "offset" */
2859 while (offset
>= pos
) {
2864 map_bh
.b_size
= blocksize
;
2866 err
= get_block(inode
, iblock
, &map_bh
, 0);
2869 /* unmapped? It's a hole - nothing to do */
2870 if (!buffer_mapped(&map_bh
))
2873 /* Ok, it's mapped. Make sure it's up-to-date */
2874 if (!PageUptodate(page
)) {
2875 err
= mapping
->a_ops
->readpage(NULL
, page
);
2881 if (!PageUptodate(page
)) {
2885 if (page_has_buffers(page
))
2888 zero_user(page
, offset
, length
);
2889 set_page_dirty(page
);
2898 EXPORT_SYMBOL(nobh_truncate_page
);
2900 int block_truncate_page(struct address_space
*mapping
,
2901 loff_t from
, get_block_t
*get_block
)
2903 pgoff_t index
= from
>> PAGE_SHIFT
;
2904 unsigned offset
= from
& (PAGE_SIZE
-1);
2907 unsigned length
, pos
;
2908 struct inode
*inode
= mapping
->host
;
2910 struct buffer_head
*bh
;
2913 blocksize
= i_blocksize(inode
);
2914 length
= offset
& (blocksize
- 1);
2916 /* Block boundary? Nothing to do */
2920 length
= blocksize
- length
;
2921 iblock
= (sector_t
)index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
2923 page
= grab_cache_page(mapping
, index
);
2928 if (!page_has_buffers(page
))
2929 create_empty_buffers(page
, blocksize
, 0);
2931 /* Find the buffer that contains "offset" */
2932 bh
= page_buffers(page
);
2934 while (offset
>= pos
) {
2935 bh
= bh
->b_this_page
;
2941 if (!buffer_mapped(bh
)) {
2942 WARN_ON(bh
->b_size
!= blocksize
);
2943 err
= get_block(inode
, iblock
, bh
, 0);
2946 /* unmapped? It's a hole - nothing to do */
2947 if (!buffer_mapped(bh
))
2951 /* Ok, it's mapped. Make sure it's up-to-date */
2952 if (PageUptodate(page
))
2953 set_buffer_uptodate(bh
);
2955 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) && !buffer_unwritten(bh
)) {
2957 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
2959 /* Uhhuh. Read error. Complain and punt. */
2960 if (!buffer_uptodate(bh
))
2964 zero_user(page
, offset
, length
);
2965 mark_buffer_dirty(bh
);
2974 EXPORT_SYMBOL(block_truncate_page
);
2977 * The generic ->writepage function for buffer-backed address_spaces
2979 int block_write_full_page(struct page
*page
, get_block_t
*get_block
,
2980 struct writeback_control
*wbc
)
2982 struct inode
* const inode
= page
->mapping
->host
;
2983 loff_t i_size
= i_size_read(inode
);
2984 const pgoff_t end_index
= i_size
>> PAGE_SHIFT
;
2987 /* Is the page fully inside i_size? */
2988 if (page
->index
< end_index
)
2989 return __block_write_full_page(inode
, page
, get_block
, wbc
,
2990 end_buffer_async_write
);
2992 /* Is the page fully outside i_size? (truncate in progress) */
2993 offset
= i_size
& (PAGE_SIZE
-1);
2994 if (page
->index
>= end_index
+1 || !offset
) {
2996 * The page may have dirty, unmapped buffers. For example,
2997 * they may have been added in ext3_writepage(). Make them
2998 * freeable here, so the page does not leak.
3000 do_invalidatepage(page
, 0, PAGE_SIZE
);
3002 return 0; /* don't care */
3006 * The page straddles i_size. It must be zeroed out on each and every
3007 * writepage invocation because it may be mmapped. "A file is mapped
3008 * in multiples of the page size. For a file that is not a multiple of
3009 * the page size, the remaining memory is zeroed when mapped, and
3010 * writes to that region are not written out to the file."
3012 zero_user_segment(page
, offset
, PAGE_SIZE
);
3013 return __block_write_full_page(inode
, page
, get_block
, wbc
,
3014 end_buffer_async_write
);
3016 EXPORT_SYMBOL(block_write_full_page
);
3018 sector_t
generic_block_bmap(struct address_space
*mapping
, sector_t block
,
3019 get_block_t
*get_block
)
3021 struct inode
*inode
= mapping
->host
;
3022 struct buffer_head tmp
= {
3023 .b_size
= i_blocksize(inode
),
3026 get_block(inode
, block
, &tmp
, 0);
3027 return tmp
.b_blocknr
;
3029 EXPORT_SYMBOL(generic_block_bmap
);
3031 static void end_bio_bh_io_sync(struct bio
*bio
)
3033 struct buffer_head
*bh
= bio
->bi_private
;
3035 if (unlikely(bio_flagged(bio
, BIO_QUIET
)))
3036 set_bit(BH_Quiet
, &bh
->b_state
);
3038 bh
->b_end_io(bh
, !bio
->bi_status
);
3043 * This allows us to do IO even on the odd last sectors
3044 * of a device, even if the block size is some multiple
3045 * of the physical sector size.
3047 * We'll just truncate the bio to the size of the device,
3048 * and clear the end of the buffer head manually.
3050 * Truly out-of-range accesses will turn into actual IO
3051 * errors, this only handles the "we need to be able to
3052 * do IO at the final sector" case.
3054 void guard_bio_eod(int op
, struct bio
*bio
)
3057 struct bio_vec
*bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
3058 unsigned truncated_bytes
;
3059 struct hd_struct
*part
;
3062 part
= __disk_get_part(bio
->bi_disk
, bio
->bi_partno
);
3064 maxsector
= part_nr_sects_read(part
);
3066 maxsector
= get_capacity(bio
->bi_disk
);
3073 * If the *whole* IO is past the end of the device,
3074 * let it through, and the IO layer will turn it into
3077 if (unlikely(bio
->bi_iter
.bi_sector
>= maxsector
))
3080 maxsector
-= bio
->bi_iter
.bi_sector
;
3081 if (likely((bio
->bi_iter
.bi_size
>> 9) <= maxsector
))
3084 /* Uhhuh. We've got a bio that straddles the device size! */
3085 truncated_bytes
= bio
->bi_iter
.bi_size
- (maxsector
<< 9);
3088 * The bio contains more than one segment which spans EOD, just return
3089 * and let IO layer turn it into an EIO
3091 if (truncated_bytes
> bvec
->bv_len
)
3094 /* Truncate the bio.. */
3095 bio
->bi_iter
.bi_size
-= truncated_bytes
;
3096 bvec
->bv_len
-= truncated_bytes
;
3098 /* ..and clear the end of the buffer for reads */
3099 if (op
== REQ_OP_READ
) {
3100 zero_user(bvec
->bv_page
, bvec
->bv_offset
+ bvec
->bv_len
,
3105 static int submit_bh_wbc(int op
, int op_flags
, struct buffer_head
*bh
,
3106 enum rw_hint write_hint
, struct writeback_control
*wbc
)
3110 BUG_ON(!buffer_locked(bh
));
3111 BUG_ON(!buffer_mapped(bh
));
3112 BUG_ON(!bh
->b_end_io
);
3113 BUG_ON(buffer_delay(bh
));
3114 BUG_ON(buffer_unwritten(bh
));
3117 * Only clear out a write error when rewriting
3119 if (test_set_buffer_req(bh
) && (op
== REQ_OP_WRITE
))
3120 clear_buffer_write_io_error(bh
);
3123 * from here on down, it's all bio -- do the initial mapping,
3124 * submit_bio -> generic_make_request may further map this bio around
3126 bio
= bio_alloc(GFP_NOIO
, 1);
3129 wbc_init_bio(wbc
, bio
);
3130 wbc_account_io(wbc
, bh
->b_page
, bh
->b_size
);
3133 bio
->bi_iter
.bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
3134 bio_set_dev(bio
, bh
->b_bdev
);
3135 bio
->bi_write_hint
= write_hint
;
3137 bio_add_page(bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
3138 BUG_ON(bio
->bi_iter
.bi_size
!= bh
->b_size
);
3140 bio
->bi_end_io
= end_bio_bh_io_sync
;
3141 bio
->bi_private
= bh
;
3143 /* Take care of bh's that straddle the end of the device */
3144 guard_bio_eod(op
, bio
);
3146 if (buffer_meta(bh
))
3147 op_flags
|= REQ_META
;
3148 if (buffer_prio(bh
))
3149 op_flags
|= REQ_PRIO
;
3150 bio_set_op_attrs(bio
, op
, op_flags
);
3156 int submit_bh(int op
, int op_flags
, struct buffer_head
*bh
)
3158 return submit_bh_wbc(op
, op_flags
, bh
, 0, NULL
);
3160 EXPORT_SYMBOL(submit_bh
);
3163 * ll_rw_block: low-level access to block devices (DEPRECATED)
3164 * @op: whether to %READ or %WRITE
3165 * @op_flags: req_flag_bits
3166 * @nr: number of &struct buffer_heads in the array
3167 * @bhs: array of pointers to &struct buffer_head
3169 * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
3170 * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE.
3171 * @op_flags contains flags modifying the detailed I/O behavior, most notably
3174 * This function drops any buffer that it cannot get a lock on (with the
3175 * BH_Lock state bit), any buffer that appears to be clean when doing a write
3176 * request, and any buffer that appears to be up-to-date when doing read
3177 * request. Further it marks as clean buffers that are processed for
3178 * writing (the buffer cache won't assume that they are actually clean
3179 * until the buffer gets unlocked).
3181 * ll_rw_block sets b_end_io to simple completion handler that marks
3182 * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
3185 * All of the buffers must be for the same device, and must also be a
3186 * multiple of the current approved size for the device.
3188 void ll_rw_block(int op
, int op_flags
, int nr
, struct buffer_head
*bhs
[])
3192 for (i
= 0; i
< nr
; i
++) {
3193 struct buffer_head
*bh
= bhs
[i
];
3195 if (!trylock_buffer(bh
))
3198 if (test_clear_buffer_dirty(bh
)) {
3199 bh
->b_end_io
= end_buffer_write_sync
;
3201 submit_bh(op
, op_flags
, bh
);
3205 if (!buffer_uptodate(bh
)) {
3206 bh
->b_end_io
= end_buffer_read_sync
;
3208 submit_bh(op
, op_flags
, bh
);
3215 EXPORT_SYMBOL(ll_rw_block
);
3217 void write_dirty_buffer(struct buffer_head
*bh
, int op_flags
)
3220 if (!test_clear_buffer_dirty(bh
)) {
3224 bh
->b_end_io
= end_buffer_write_sync
;
3226 submit_bh(REQ_OP_WRITE
, op_flags
, bh
);
3228 EXPORT_SYMBOL(write_dirty_buffer
);
3231 * For a data-integrity writeout, we need to wait upon any in-progress I/O
3232 * and then start new I/O and then wait upon it. The caller must have a ref on
3235 int __sync_dirty_buffer(struct buffer_head
*bh
, int op_flags
)
3239 WARN_ON(atomic_read(&bh
->b_count
) < 1);
3241 if (test_clear_buffer_dirty(bh
)) {
3243 bh
->b_end_io
= end_buffer_write_sync
;
3244 ret
= submit_bh(REQ_OP_WRITE
, op_flags
, bh
);
3246 if (!ret
&& !buffer_uptodate(bh
))
3253 EXPORT_SYMBOL(__sync_dirty_buffer
);
3255 int sync_dirty_buffer(struct buffer_head
*bh
)
3257 return __sync_dirty_buffer(bh
, REQ_SYNC
);
3259 EXPORT_SYMBOL(sync_dirty_buffer
);
3262 * try_to_free_buffers() checks if all the buffers on this particular page
3263 * are unused, and releases them if so.
3265 * Exclusion against try_to_free_buffers may be obtained by either
3266 * locking the page or by holding its mapping's private_lock.
3268 * If the page is dirty but all the buffers are clean then we need to
3269 * be sure to mark the page clean as well. This is because the page
3270 * may be against a block device, and a later reattachment of buffers
3271 * to a dirty page will set *all* buffers dirty. Which would corrupt
3272 * filesystem data on the same device.
3274 * The same applies to regular filesystem pages: if all the buffers are
3275 * clean then we set the page clean and proceed. To do that, we require
3276 * total exclusion from __set_page_dirty_buffers(). That is obtained with
3279 * try_to_free_buffers() is non-blocking.
3281 static inline int buffer_busy(struct buffer_head
*bh
)
3283 return atomic_read(&bh
->b_count
) |
3284 (bh
->b_state
& ((1 << BH_Dirty
) | (1 << BH_Lock
)));
3288 drop_buffers(struct page
*page
, struct buffer_head
**buffers_to_free
)
3290 struct buffer_head
*head
= page_buffers(page
);
3291 struct buffer_head
*bh
;
3295 if (buffer_busy(bh
))
3297 bh
= bh
->b_this_page
;
3298 } while (bh
!= head
);
3301 struct buffer_head
*next
= bh
->b_this_page
;
3303 if (bh
->b_assoc_map
)
3304 __remove_assoc_queue(bh
);
3306 } while (bh
!= head
);
3307 *buffers_to_free
= head
;
3308 __clear_page_buffers(page
);
3314 int try_to_free_buffers(struct page
*page
)
3316 struct address_space
* const mapping
= page
->mapping
;
3317 struct buffer_head
*buffers_to_free
= NULL
;
3320 BUG_ON(!PageLocked(page
));
3321 if (PageWriteback(page
))
3324 if (mapping
== NULL
) { /* can this still happen? */
3325 ret
= drop_buffers(page
, &buffers_to_free
);
3329 spin_lock(&mapping
->private_lock
);
3330 ret
= drop_buffers(page
, &buffers_to_free
);
3333 * If the filesystem writes its buffers by hand (eg ext3)
3334 * then we can have clean buffers against a dirty page. We
3335 * clean the page here; otherwise the VM will never notice
3336 * that the filesystem did any IO at all.
3338 * Also, during truncate, discard_buffer will have marked all
3339 * the page's buffers clean. We discover that here and clean
3342 * private_lock must be held over this entire operation in order
3343 * to synchronise against __set_page_dirty_buffers and prevent the
3344 * dirty bit from being lost.
3347 cancel_dirty_page(page
);
3348 spin_unlock(&mapping
->private_lock
);
3350 if (buffers_to_free
) {
3351 struct buffer_head
*bh
= buffers_to_free
;
3354 struct buffer_head
*next
= bh
->b_this_page
;
3355 free_buffer_head(bh
);
3357 } while (bh
!= buffers_to_free
);
3361 EXPORT_SYMBOL(try_to_free_buffers
);
3364 * There are no bdflush tunables left. But distributions are
3365 * still running obsolete flush daemons, so we terminate them here.
3367 * Use of bdflush() is deprecated and will be removed in a future kernel.
3368 * The `flush-X' kernel threads fully replace bdflush daemons and this call.
3370 SYSCALL_DEFINE2(bdflush
, int, func
, long, data
)
3372 static int msg_count
;
3374 if (!capable(CAP_SYS_ADMIN
))
3377 if (msg_count
< 5) {
3380 "warning: process `%s' used the obsolete bdflush"
3381 " system call\n", current
->comm
);
3382 printk(KERN_INFO
"Fix your initscripts?\n");
3391 * Buffer-head allocation
3393 static struct kmem_cache
*bh_cachep __read_mostly
;
3396 * Once the number of bh's in the machine exceeds this level, we start
3397 * stripping them in writeback.
3399 static unsigned long max_buffer_heads
;
3401 int buffer_heads_over_limit
;
3403 struct bh_accounting
{
3404 int nr
; /* Number of live bh's */
3405 int ratelimit
; /* Limit cacheline bouncing */
3408 static DEFINE_PER_CPU(struct bh_accounting
, bh_accounting
) = {0, 0};
3410 static void recalc_bh_state(void)
3415 if (__this_cpu_inc_return(bh_accounting
.ratelimit
) - 1 < 4096)
3417 __this_cpu_write(bh_accounting
.ratelimit
, 0);
3418 for_each_online_cpu(i
)
3419 tot
+= per_cpu(bh_accounting
, i
).nr
;
3420 buffer_heads_over_limit
= (tot
> max_buffer_heads
);
3423 struct buffer_head
*alloc_buffer_head(gfp_t gfp_flags
)
3425 struct buffer_head
*ret
= kmem_cache_zalloc(bh_cachep
, gfp_flags
);
3427 INIT_LIST_HEAD(&ret
->b_assoc_buffers
);
3429 __this_cpu_inc(bh_accounting
.nr
);
3435 EXPORT_SYMBOL(alloc_buffer_head
);
3437 void free_buffer_head(struct buffer_head
*bh
)
3439 BUG_ON(!list_empty(&bh
->b_assoc_buffers
));
3440 kmem_cache_free(bh_cachep
, bh
);
3442 __this_cpu_dec(bh_accounting
.nr
);
3446 EXPORT_SYMBOL(free_buffer_head
);
3448 static int buffer_exit_cpu_dead(unsigned int cpu
)
3451 struct bh_lru
*b
= &per_cpu(bh_lrus
, cpu
);
3453 for (i
= 0; i
< BH_LRU_SIZE
; i
++) {
3457 this_cpu_add(bh_accounting
.nr
, per_cpu(bh_accounting
, cpu
).nr
);
3458 per_cpu(bh_accounting
, cpu
).nr
= 0;
3463 * bh_uptodate_or_lock - Test whether the buffer is uptodate
3464 * @bh: struct buffer_head
3466 * Return true if the buffer is up-to-date and false,
3467 * with the buffer locked, if not.
3469 int bh_uptodate_or_lock(struct buffer_head
*bh
)
3471 if (!buffer_uptodate(bh
)) {
3473 if (!buffer_uptodate(bh
))
3479 EXPORT_SYMBOL(bh_uptodate_or_lock
);
3482 * bh_submit_read - Submit a locked buffer for reading
3483 * @bh: struct buffer_head
3485 * Returns zero on success and -EIO on error.
3487 int bh_submit_read(struct buffer_head
*bh
)
3489 BUG_ON(!buffer_locked(bh
));
3491 if (buffer_uptodate(bh
)) {
3497 bh
->b_end_io
= end_buffer_read_sync
;
3498 submit_bh(REQ_OP_READ
, 0, bh
);
3500 if (buffer_uptodate(bh
))
3504 EXPORT_SYMBOL(bh_submit_read
);
3507 * Seek for SEEK_DATA / SEEK_HOLE within @page, starting at @lastoff.
3509 * Returns the offset within the file on success, and -ENOENT otherwise.
3512 page_seek_hole_data(struct page
*page
, loff_t lastoff
, int whence
)
3514 loff_t offset
= page_offset(page
);
3515 struct buffer_head
*bh
, *head
;
3516 bool seek_data
= whence
== SEEK_DATA
;
3518 if (lastoff
< offset
)
3521 bh
= head
= page_buffers(page
);
3523 offset
+= bh
->b_size
;
3524 if (lastoff
>= offset
)
3528 * Unwritten extents that have data in the page cache covering
3529 * them can be identified by the BH_Unwritten state flag.
3530 * Pages with multiple buffers might have a mix of holes, data
3531 * and unwritten extents - any buffer with valid data in it
3532 * should have BH_Uptodate flag set on it.
3535 if ((buffer_unwritten(bh
) || buffer_uptodate(bh
)) == seek_data
)
3539 } while ((bh
= bh
->b_this_page
) != head
);
3544 * Seek for SEEK_DATA / SEEK_HOLE in the page cache.
3546 * Within unwritten extents, the page cache determines which parts are holes
3547 * and which are data: unwritten and uptodate buffer heads count as data;
3548 * everything else counts as a hole.
3550 * Returns the resulting offset on successs, and -ENOENT otherwise.
3553 page_cache_seek_hole_data(struct inode
*inode
, loff_t offset
, loff_t length
,
3556 pgoff_t index
= offset
>> PAGE_SHIFT
;
3557 pgoff_t end
= DIV_ROUND_UP(offset
+ length
, PAGE_SIZE
);
3558 loff_t lastoff
= offset
;
3559 struct pagevec pvec
;
3564 pagevec_init(&pvec
, 0);
3567 unsigned nr_pages
, i
;
3569 nr_pages
= pagevec_lookup_range(&pvec
, inode
->i_mapping
, &index
,
3574 for (i
= 0; i
< nr_pages
; i
++) {
3575 struct page
*page
= pvec
.pages
[i
];
3578 * At this point, the page may be truncated or
3579 * invalidated (changing page->mapping to NULL), or
3580 * even swizzled back from swapper_space to tmpfs file
3581 * mapping. However, page->index will not change
3582 * because we have a reference on the page.
3584 * If current page offset is beyond where we've ended,
3585 * we've found a hole.
3587 if (whence
== SEEK_HOLE
&&
3588 lastoff
< page_offset(page
))
3592 if (likely(page
->mapping
== inode
->i_mapping
) &&
3593 page_has_buffers(page
)) {
3594 lastoff
= page_seek_hole_data(page
, lastoff
, whence
);
3601 lastoff
= page_offset(page
) + PAGE_SIZE
;
3603 pagevec_release(&pvec
);
3604 } while (index
< end
);
3606 /* When no page at lastoff and we are not done, we found a hole. */
3607 if (whence
!= SEEK_HOLE
)
3611 if (lastoff
< offset
+ length
)
3616 pagevec_release(&pvec
);
3620 void __init
buffer_init(void)
3622 unsigned long nrpages
;
3625 bh_cachep
= kmem_cache_create("buffer_head",
3626 sizeof(struct buffer_head
), 0,
3627 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
3632 * Limit the bh occupancy to 10% of ZONE_NORMAL
3634 nrpages
= (nr_free_buffer_pages() * 10) / 100;
3635 max_buffer_heads
= nrpages
* (PAGE_SIZE
/ sizeof(struct buffer_head
));
3636 ret
= cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD
, "fs/buffer:dead",
3637 NULL
, buffer_exit_cpu_dead
);