linux - fs/buffer.c

3444 lines · 89.4 KiB

1	// SPDX-License-Identifier: GPL-2.0-only
2	/*
3	* linux/fs/buffer.c
4	*
5	* Copyright (C) 1991, 1992, 2002 Linus Torvalds
6	*/
7
8	/*
9	* Start bdflush() with kernel_thread not syscall - Paul Gortmaker, 12/95
10	*
11	* Removed a lot of unnecessary code and simplified things now that
12	* the buffer cache isn't our primary cache - Andrew Tridgell 12/96
13	*
14	* Speed up hash, lru, and free list operations. Use gfp() for allocating
15	* hash table, use SLAB cache for buffer heads. SMP threading. -DaveM
16	*
17	* Added 32k buffer block sizes - these are required older ARM systems. - RMK
18	*
19	* async buffer flushing, 1999 Andrea Arcangeli <[email protected]>
20	*/
21
22	#include <linux/kernel.h>
23	#include <linux/sched/signal.h>
24	#include <linux/syscalls.h>
25	#include <linux/fs.h>
26	#include <linux/iomap.h>
27	#include <linux/mm.h>
28	#include <linux/percpu.h>
29	#include <linux/slab.h>
30	#include <linux/capability.h>
31	#include <linux/blkdev.h>
32	#include <linux/file.h>
33	#include <linux/quotaops.h>
34	#include <linux/highmem.h>
35	#include <linux/export.h>
36	#include <linux/backing-dev.h>
37	#include <linux/writeback.h>
38	#include <linux/hash.h>
39	#include <linux/suspend.h>
40	#include <linux/buffer_head.h>
41	#include <linux/task_io_accounting_ops.h>
42	#include <linux/bio.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 <linux/sched/mm.h>
49	#include <trace/events/block.h>
50	#include <linux/fscrypt.h>
51
52	#include "internal.h"
53
54	static int fsync_buffers_list(spinlock_t lock, struct list_head list);
55	static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
56	enum rw_hint hint, struct writeback_control *wbc);
57
58	#define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
59
60	inline void touch_buffer(struct buffer_head *bh)
61	{
62	trace_block_touch_buffer(bh);
63	mark_page_accessed(bh->b_page);
64	}
65	EXPORT_SYMBOL(touch_buffer);
66
67	void __lock_buffer(struct buffer_head *bh)
68	{
69	wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
70	}
71	EXPORT_SYMBOL(__lock_buffer);
72
73	void unlock_buffer(struct buffer_head *bh)
74	{
75	clear_bit_unlock(BH_Lock, &bh->b_state);
76	smp_mb__after_atomic();
77	wake_up_bit(&bh->b_state, BH_Lock);
78	}
79	EXPORT_SYMBOL(unlock_buffer);
80
81	/*
82	* Returns if the page has dirty or writeback buffers. If all the buffers
83	* are unlocked and clean then the PageDirty information is stale. If
84	* any of the pages are locked, it is assumed they are locked for IO.
85	*/
86	void buffer_check_dirty_writeback(struct page *page,
87	bool dirty, bool writeback)
88	{
89	struct buffer_head head, bh;
90	*dirty = false;
91	*writeback = false;
92
93	BUG_ON(!PageLocked(page));
94
95	if (!page_has_buffers(page))
96	return;
97
98	if (PageWriteback(page))
99	*writeback = true;
100
101	head = page_buffers(page);
102	bh = head;
103	do {
104	if (buffer_locked(bh))
105	*writeback = true;
106
107	if (buffer_dirty(bh))
108	*dirty = true;
109
110	bh = bh->b_this_page;
111	} while (bh != head);
112	}
113	EXPORT_SYMBOL(buffer_check_dirty_writeback);
114
115	/*
116	* Block until a buffer comes unlocked. This doesn't stop it
117	* from becoming locked again - you have to lock it yourself
118	* if you want to preserve its state.
119	*/
120	void __wait_on_buffer(struct buffer_head * bh)
121	{
122	wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
123	}
124	EXPORT_SYMBOL(__wait_on_buffer);
125
126	static void buffer_io_error(struct buffer_head bh, char msg)
127	{
128	if (!test_bit(BH_Quiet, &bh->b_state))
129	printk_ratelimited(KERN_ERR
130	"Buffer I/O error on dev %pg, logical block %llu%s\n",
131	bh->b_bdev, (unsigned long long)bh->b_blocknr, msg);
132	}
133
134	/*
135	* End-of-IO handler helper function which does not touch the bh after
136	* unlocking it.
137	* Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
138	* a race there is benign: unlock_buffer() only use the bh's address for
139	* hashing after unlocking the buffer, so it doesn't actually touch the bh
140	* itself.
141	*/
142	static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate)
143	{
144	if (uptodate) {
145	set_buffer_uptodate(bh);
146	} else {
147	/* This happens, due to failed read-ahead attempts. */
148	clear_buffer_uptodate(bh);
149	}
150	unlock_buffer(bh);
151	}
152
153	/*
154	* Default synchronous end-of-IO handler.. Just mark it up-to-date and
155	* unlock the buffer. This is what ll_rw_block uses too.
156	*/
157	void end_buffer_read_sync(struct buffer_head *bh, int uptodate)
158	{
159	__end_buffer_read_notouch(bh, uptodate);
160	put_bh(bh);
161	}
162	EXPORT_SYMBOL(end_buffer_read_sync);
163
164	void end_buffer_write_sync(struct buffer_head *bh, int uptodate)
165	{
166	if (uptodate) {
167	set_buffer_uptodate(bh);
168	} else {
169	buffer_io_error(bh, ", lost sync page write");
170	mark_buffer_write_io_error(bh);
171	clear_buffer_uptodate(bh);
172	}
173	unlock_buffer(bh);
174	put_bh(bh);
175	}
176	EXPORT_SYMBOL(end_buffer_write_sync);
177
178	/*
179	* Various filesystems appear to want __find_get_block to be non-blocking.
180	* But it's the page lock which protects the buffers. To get around this,
181	* we get exclusion from try_to_free_buffers with the blockdev mapping's
182	* private_lock.
183	*
184	* Hack idea: for the blockdev mapping, private_lock contention
185	* may be quite high. This code could TryLock the page, and if that
186	* succeeds, there is no need to take private_lock.
187	*/
188	static struct buffer_head *
189	__find_get_block_slow(struct block_device *bdev, sector_t block)
190	{
191	struct inode *bd_inode = bdev->bd_inode;
192	struct address_space *bd_mapping = bd_inode->i_mapping;
193	struct buffer_head *ret = NULL;
194	pgoff_t index;
195	struct buffer_head *bh;
196	struct buffer_head *head;
197	struct page *page;
198	int all_mapped = 1;
199	static DEFINE_RATELIMIT_STATE(last_warned, HZ, 1);
200
201	index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
202	page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED);
203	if (!page)
204	goto out;
205
206	spin_lock(&bd_mapping->private_lock);
207	if (!page_has_buffers(page))
208	goto out_unlock;
209	head = page_buffers(page);
210	bh = head;
211	do {
212	if (!buffer_mapped(bh))
213	all_mapped = 0;
214	else if (bh->b_blocknr == block) {
215	ret = bh;
216	get_bh(bh);
217	goto out_unlock;
218	}
219	bh = bh->b_this_page;
220	} while (bh != head);
221
222	/* we might be here because some of the buffers on this page are
223	* not mapped. This is due to various races between
224	* file io on the block device and getblk. It gets dealt with
225	* elsewhere, don't buffer_error if we had some unmapped buffers
226	*/
227	ratelimit_set_flags(&last_warned, RATELIMIT_MSG_ON_RELEASE);
228	if (all_mapped && __ratelimit(&last_warned)) {
229	printk("__find_get_block_slow() failed. block=%llu, "
230	"b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, "
231	"device %pg blocksize: %d\n",
232	(unsigned long long)block,
233	(unsigned long long)bh->b_blocknr,
234	bh->b_state, bh->b_size, bdev,
235	1 << bd_inode->i_blkbits);
236	}
237	out_unlock:
238	spin_unlock(&bd_mapping->private_lock);
239	put_page(page);
240	out:
241	return ret;
242	}
243
244	static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
245	{
246	unsigned long flags;
247	struct buffer_head *first;
248	struct buffer_head *tmp;
249	struct page *page;
250	int page_uptodate = 1;
251
252	BUG_ON(!buffer_async_read(bh));
253
254	page = bh->b_page;
255	if (uptodate) {
256	set_buffer_uptodate(bh);
257	} else {
258	clear_buffer_uptodate(bh);
259	buffer_io_error(bh, ", async page read");
260	SetPageError(page);
261	}
262
263	/*
264	* Be _very_ careful from here on. Bad things can happen if
265	* two buffer heads end IO at almost the same time and both
266	* decide that the page is now completely done.
267	*/
268	first = page_buffers(page);
269	spin_lock_irqsave(&first->b_uptodate_lock, flags);
270	clear_buffer_async_read(bh);
271	unlock_buffer(bh);
272	tmp = bh;
273	do {
274	if (!buffer_uptodate(tmp))
275	page_uptodate = 0;
276	if (buffer_async_read(tmp)) {
277	BUG_ON(!buffer_locked(tmp));
278	goto still_busy;
279	}
280	tmp = tmp->b_this_page;
281	} while (tmp != bh);
282	spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
283
284	/*
285	* If none of the buffers had errors and they are all
286	* uptodate then we can set the page uptodate.
287	*/
288	if (page_uptodate && !PageError(page))
289	SetPageUptodate(page);
290	unlock_page(page);
291	return;
292
293	still_busy:
294	spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
295	return;
296	}
297
298	struct decrypt_bh_ctx {
299	struct work_struct work;
300	struct buffer_head *bh;
301	};
302
303	static void decrypt_bh(struct work_struct *work)
304	{
305	struct decrypt_bh_ctx *ctx =
306	container_of(work, struct decrypt_bh_ctx, work);
307	struct buffer_head *bh = ctx->bh;
308	int err;
309
310	err = fscrypt_decrypt_pagecache_blocks(bh->b_page, bh->b_size,
311	bh_offset(bh));
312	end_buffer_async_read(bh, err == 0);
313	kfree(ctx);
314	}
315
316	/*
317	* I/O completion handler for block_read_full_page() - pages
318	* which come unlocked at the end of I/O.
319	*/
320	static void end_buffer_async_read_io(struct buffer_head *bh, int uptodate)
321	{
322	/* Decrypt if needed */
323	if (uptodate &&
324	fscrypt_inode_uses_fs_layer_crypto(bh->b_page->mapping->host)) {
325	struct decrypt_bh_ctx ctx = kmalloc(sizeof(ctx), GFP_ATOMIC);
326
327	if (ctx) {
328	INIT_WORK(&ctx->work, decrypt_bh);
329	ctx->bh = bh;
330	fscrypt_enqueue_decrypt_work(&ctx->work);
331	return;
332	}
333	uptodate = 0;
334	}
335	end_buffer_async_read(bh, uptodate);
336	}
337
338	/*
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.
341	*/
342	void end_buffer_async_write(struct buffer_head *bh, int uptodate)
343	{
344	unsigned long flags;
345	struct buffer_head *first;
346	struct buffer_head *tmp;
347	struct page *page;
348
349	BUG_ON(!buffer_async_write(bh));
350
351	page = bh->b_page;
352	if (uptodate) {
353	set_buffer_uptodate(bh);
354	} else {
355	buffer_io_error(bh, ", lost async page write");
356	mark_buffer_write_io_error(bh);
357	clear_buffer_uptodate(bh);
358	SetPageError(page);
359	}
360
361	first = page_buffers(page);
362	spin_lock_irqsave(&first->b_uptodate_lock, flags);
363
364	clear_buffer_async_write(bh);
365	unlock_buffer(bh);
366	tmp = bh->b_this_page;
367	while (tmp != bh) {
368	if (buffer_async_write(tmp)) {
369	BUG_ON(!buffer_locked(tmp));
370	goto still_busy;
371	}
372	tmp = tmp->b_this_page;
373	}
374	spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
375	end_page_writeback(page);
376	return;
377
378	still_busy:
379	spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
380	return;
381	}
382	EXPORT_SYMBOL(end_buffer_async_write);
383
384	/*
385	* If a page's buffers are under async readin (end_buffer_async_read
386	* completion) then there is a possibility that another thread of
387	* control could lock one of the buffers after it has completed
388	* but while some of the other buffers have not completed. This
389	* locked buffer would confuse end_buffer_async_read() into not unlocking
390	* the page. So the absence of BH_Async_Read tells end_buffer_async_read()
391	* that this buffer is not under async I/O.
392	*
393	* The page comes unlocked when it has no locked buffer_async buffers
394	* left.
395	*
396	* PageLocked prevents anyone starting new async I/O reads any of
397	* the buffers.
398	*
399	* PageWriteback is used to prevent simultaneous writeout of the same
400	* page.
401	*
402	* PageLocked prevents anyone from starting writeback of a page which is
403	* under read I/O (PageWriteback is only ever set against a locked page).
404	*/
405	static void mark_buffer_async_read(struct buffer_head *bh)
406	{
407	bh->b_end_io = end_buffer_async_read_io;
408	set_buffer_async_read(bh);
409	}
410
411	static void mark_buffer_async_write_endio(struct buffer_head *bh,
412	bh_end_io_t *handler)
413	{
414	bh->b_end_io = handler;
415	set_buffer_async_write(bh);
416	}
417
418	void mark_buffer_async_write(struct buffer_head *bh)
419	{
420	mark_buffer_async_write_endio(bh, end_buffer_async_write);
421	}
422	EXPORT_SYMBOL(mark_buffer_async_write);
423
424
425	/*
426	* fs/buffer.c contains helper functions for buffer-backed address space's
427	* fsync functions. A common requirement for buffer-based filesystems is
428	* that certain data from the backing blockdev needs to be written out for
429	* a successful fsync(). For example, ext2 indirect blocks need to be
430	* written back and waited upon before fsync() returns.
431	*
432	* The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
433	* inode_has_buffers() and invalidate_inode_buffers() are provided for the
434	* management of a list of dependent buffers at ->i_mapping->private_list.
435	*
436	* Locking is a little subtle: try_to_free_buffers() will remove buffers
437	* from their controlling inode's queue when they are being freed. But
438	* try_to_free_buffers() will be operating against the blockdev mapping
439	* at the time, not against the S_ISREG file which depends on those buffers.
440	* So the locking for private_list is via the private_lock in the address_space
441	* which backs the buffers. Which is different from the address_space
442	* against which the buffers are listed. So for a particular address_space,
443	* mapping->private_lock does not protect mapping->private_list! In fact,
444	* mapping->private_list will always be protected by the backing blockdev's
445	* ->private_lock.
446	*
447	* Which introduces a requirement: all buffers on an address_space's
448	* ->private_list must be from the same address_space: the blockdev's.
449	*
450	* address_spaces which do not place buffers at ->private_list via these
451	* utility functions are free to use private_lock and private_list for
452	* whatever they want. The only requirement is that list_empty(private_list)
453	* be true at clear_inode() time.
454	*
455	* FIXME: clear_inode should not call invalidate_inode_buffers(). The
456	* filesystems should do that. invalidate_inode_buffers() should just go
457	* BUG_ON(!list_empty).
458	*
459	* FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
460	* take an address_space, not an inode. And it should be called
461	* mark_buffer_dirty_fsync() to clearly define why those buffers are being
462	* queued up.
463	*
464	* FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
465	* list if it is already on a list. Because if the buffer is on a list,
466	* it must already be on the right one. If not, the filesystem is being
467	* silly. This will save a ton of locking. But first we have to ensure
468	* that buffers are taken off the old inode's list when they are freed
469	* (presumably in truncate). That requires careful auditing of all
470	* filesystems (do it inside bforget()). It could also be done by bringing
471	* b_inode back.
472	*/
473
474	/*
475	* The buffer's backing address_space's private_lock must be held
476	*/
477	static void __remove_assoc_queue(struct buffer_head *bh)
478	{
479	list_del_init(&bh->b_assoc_buffers);
480	WARN_ON(!bh->b_assoc_map);
481	bh->b_assoc_map = NULL;
482	}
483
484	int inode_has_buffers(struct inode *inode)
485	{
486	return !list_empty(&inode->i_data.private_list);
487	}
488
489	/*
490	* osync is designed to support O_SYNC io. It waits synchronously for
491	* all already-submitted IO to complete, but does not queue any new
492	* writes to the disk.
493	*
494	* To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
495	* you dirty the buffers, and then use osync_inode_buffers to wait for
496	* completion. Any other dirty buffers which are not yet queued for
497	* write will not be flushed to disk by the osync.
498	*/
499	static int osync_buffers_list(spinlock_t lock, struct list_head list)
500	{
501	struct buffer_head *bh;
502	struct list_head *p;
503	int err = 0;
504
505	spin_lock(lock);
506	repeat:
507	list_for_each_prev(p, list) {
508	bh = BH_ENTRY(p);
509	if (buffer_locked(bh)) {
510	get_bh(bh);
511	spin_unlock(lock);
512	wait_on_buffer(bh);
513	if (!buffer_uptodate(bh))
514	err = -EIO;
515	brelse(bh);
516	spin_lock(lock);
517	goto repeat;
518	}
519	}
520	spin_unlock(lock);
521	return err;
522	}
523
524	void emergency_thaw_bdev(struct super_block *sb)
525	{
526	while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb))
527	printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev);
528	}
529
530	/**
531	* sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
532	* @mapping: the mapping which wants those buffers written
533	*
534	* Starts I/O against the buffers at mapping->private_list, and waits upon
535	* that I/O.
536	*
537	* Basically, this is a convenience function for fsync().
538	* @mapping is a file or directory which needs those buffers to be written for
539	* a successful fsync().
540	*/
541	int sync_mapping_buffers(struct address_space *mapping)
542	{
543	struct address_space *buffer_mapping = mapping->private_data;
544
545	if (buffer_mapping == NULL \|\| list_empty(&mapping->private_list))
546	return 0;
547
548	return fsync_buffers_list(&buffer_mapping->private_lock,
549	&mapping->private_list);
550	}
551	EXPORT_SYMBOL(sync_mapping_buffers);
552
553	/*
554	* Called when we've recently written block `bblock', and it is known that
555	* `bblock' was for a buffer_boundary() buffer. This means that the block at
556	* `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
557	* dirty, schedule it for IO. So that indirects merge nicely with their data.
558	*/
559	void write_boundary_block(struct block_device *bdev,
560	sector_t bblock, unsigned blocksize)
561	{
562	struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize);
563	if (bh) {
564	if (buffer_dirty(bh))
565	ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
566	put_bh(bh);
567	}
568	}
569
570	void mark_buffer_dirty_inode(struct buffer_head bh, struct inode inode)
571	{
572	struct address_space *mapping = inode->i_mapping;
573	struct address_space *buffer_mapping = bh->b_page->mapping;
574
575	mark_buffer_dirty(bh);
576	if (!mapping->private_data) {
577	mapping->private_data = buffer_mapping;
578	} else {
579	BUG_ON(mapping->private_data != buffer_mapping);
580	}
581	if (!bh->b_assoc_map) {
582	spin_lock(&buffer_mapping->private_lock);
583	list_move_tail(&bh->b_assoc_buffers,
584	&mapping->private_list);
585	bh->b_assoc_map = mapping;
586	spin_unlock(&buffer_mapping->private_lock);
587	}
588	}
589	EXPORT_SYMBOL(mark_buffer_dirty_inode);
590
591	/*
592	* Mark the page dirty, and set it dirty in the page cache, and mark the inode
593	* dirty.
594	*
595	* If warn is true, then emit a warning if the page is not uptodate and has
596	* not been truncated.
597	*
598	* The caller must hold lock_page_memcg().
599	*/
600	void __set_page_dirty(struct page page, struct address_space mapping,
601	int warn)
602	{
603	unsigned long flags;
604
605	xa_lock_irqsave(&mapping->i_pages, flags);
606	if (page->mapping) { /* Race with truncate? */
607	WARN_ON_ONCE(warn && !PageUptodate(page));
608	account_page_dirtied(page, mapping);
609	__xa_set_mark(&mapping->i_pages, page_index(page),
610	PAGECACHE_TAG_DIRTY);
611	}
612	xa_unlock_irqrestore(&mapping->i_pages, flags);
613	}
614	EXPORT_SYMBOL_GPL(__set_page_dirty);
615
616	/*
617	* Add a page to the dirty page list.
618	*
619	* It is a sad fact of life that this function is called from several places
620	* deeply under spinlocking. It may not sleep.
621	*
622	* If the page has buffers, the uptodate buffers are set dirty, to preserve
623	* dirty-state coherency between the page and the buffers. It the page does
624	* not have buffers then when they are later attached they will all be set
625	* dirty.
626	*
627	* The buffers are dirtied before the page is dirtied. There's a small race
628	* window in which a writepage caller may see the page cleanness but not the
629	* buffer dirtiness. That's fine. If this code were to set the page dirty
630	* before the buffers, a concurrent writepage caller could clear the page dirty
631	* bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
632	* page on the dirty page list.
633	*
634	* We use private_lock to lock against try_to_free_buffers while using the
635	* page's buffer list. Also use this to protect against clean buffers being
636	* added to the page after it was set dirty.
637	*
638	* FIXME: may need to call ->reservepage here as well. That's rather up to the
639	* address_space though.
640	*/
641	int __set_page_dirty_buffers(struct page *page)
642	{
643	int newly_dirty;
644	struct address_space *mapping = page_mapping(page);
645
646	if (unlikely(!mapping))
647	return !TestSetPageDirty(page);
648
649	spin_lock(&mapping->private_lock);
650	if (page_has_buffers(page)) {
651	struct buffer_head *head = page_buffers(page);
652	struct buffer_head *bh = head;
653
654	do {
655	set_buffer_dirty(bh);
656	bh = bh->b_this_page;
657	} while (bh != head);
658	}
659	/*
660	* Lock out page->mem_cgroup migration to keep PageDirty
661	* synchronized with per-memcg dirty page counters.
662	*/
663	lock_page_memcg(page);
664	newly_dirty = !TestSetPageDirty(page);
665	spin_unlock(&mapping->private_lock);
666
667	if (newly_dirty)
668	__set_page_dirty(page, mapping, 1);
669
670	unlock_page_memcg(page);
671
672	if (newly_dirty)
673	__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
674
675	return newly_dirty;
676	}
677	EXPORT_SYMBOL(__set_page_dirty_buffers);
678
679	/*
680	* Write out and wait upon a list of buffers.
681	*
682	* We have conflicting pressures: we want to make sure that all
683	* initially dirty buffers get waited on, but that any subsequently
684	* dirtied buffers don't. After all, we don't want fsync to last
685	* forever if somebody is actively writing to the file.
686	*
687	* Do this in two main stages: first we copy dirty buffers to a
688	* temporary inode list, queueing the writes as we go. Then we clean
689	* up, waiting for those writes to complete.
690	*
691	* During this second stage, any subsequent updates to the file may end
692	* up refiling the buffer on the original inode's dirty list again, so
693	* there is a chance we will end up with a buffer queued for write but
694	* not yet completed on that list. So, as a final cleanup we go through
695	* the osync code to catch these locked, dirty buffers without requeuing
696	* any newly dirty buffers for write.
697	*/
698	static int fsync_buffers_list(spinlock_t lock, struct list_head list)
699	{
700	struct buffer_head *bh;
701	struct list_head tmp;
702	struct address_space *mapping;
703	int err = 0, err2;
704	struct blk_plug plug;
705
706	INIT_LIST_HEAD(&tmp);
707	blk_start_plug(&plug);
708
709	spin_lock(lock);
710	while (!list_empty(list)) {
711	bh = BH_ENTRY(list->next);
712	mapping = bh->b_assoc_map;
713	__remove_assoc_queue(bh);
714	/* Avoid race with mark_buffer_dirty_inode() which does
715	* a lockless check and we rely on seeing the dirty bit */
716	smp_mb();
717	if (buffer_dirty(bh) \|\| buffer_locked(bh)) {
718	list_add(&bh->b_assoc_buffers, &tmp);
719	bh->b_assoc_map = mapping;
720	if (buffer_dirty(bh)) {
721	get_bh(bh);
722	spin_unlock(lock);
723	/*
724	* Ensure any pending I/O completes so that
725	* write_dirty_buffer() actually writes the
726	* current contents - it is a noop if I/O is
727	* still in flight on potentially older
728	* contents.
729	*/
730	write_dirty_buffer(bh, REQ_SYNC);
731
732	/*
733	* Kick off IO for the previous mapping. Note
734	* that we will not run the very last mapping,
735	* wait_on_buffer() will do that for us
736	* through sync_buffer().
737	*/
738	brelse(bh);
739	spin_lock(lock);
740	}
741	}
742	}
743
744	spin_unlock(lock);
745	blk_finish_plug(&plug);
746	spin_lock(lock);
747
748	while (!list_empty(&tmp)) {
749	bh = BH_ENTRY(tmp.prev);
750	get_bh(bh);
751	mapping = bh->b_assoc_map;
752	__remove_assoc_queue(bh);
753	/* Avoid race with mark_buffer_dirty_inode() which does
754	* a lockless check and we rely on seeing the dirty bit */
755	smp_mb();
756	if (buffer_dirty(bh)) {
757	list_add(&bh->b_assoc_buffers,
758	&mapping->private_list);
759	bh->b_assoc_map = mapping;
760	}
761	spin_unlock(lock);
762	wait_on_buffer(bh);
763	if (!buffer_uptodate(bh))
764	err = -EIO;
765	brelse(bh);
766	spin_lock(lock);
767	}
768
769	spin_unlock(lock);
770	err2 = osync_buffers_list(lock, list);
771	if (err)
772	return err;
773	else
774	return err2;
775	}
776
777	/*
778	* Invalidate any and all dirty buffers on a given inode. We are
779	* probably unmounting the fs, but that doesn't mean we have already
780	* done a sync(). Just drop the buffers from the inode list.
781	*
782	* NOTE: we take the inode's blockdev's mapping's private_lock. Which
783	* assumes that all the buffers are against the blockdev. Not true
784	* for reiserfs.
785	*/
786	void invalidate_inode_buffers(struct inode *inode)
787	{
788	if (inode_has_buffers(inode)) {
789	struct address_space *mapping = &inode->i_data;
790	struct list_head *list = &mapping->private_list;
791	struct address_space *buffer_mapping = mapping->private_data;
792
793	spin_lock(&buffer_mapping->private_lock);
794	while (!list_empty(list))
795	__remove_assoc_queue(BH_ENTRY(list->next));
796	spin_unlock(&buffer_mapping->private_lock);
797	}
798	}
799	EXPORT_SYMBOL(invalidate_inode_buffers);
800
801	/*
802	* Remove any clean buffers from the inode's buffer list. This is called
803	* when we're trying to free the inode itself. Those buffers can pin it.
804	*
805	* Returns true if all buffers were removed.
806	*/
807	int remove_inode_buffers(struct inode *inode)
808	{
809	int ret = 1;
810
811	if (inode_has_buffers(inode)) {
812	struct address_space *mapping = &inode->i_data;
813	struct list_head *list = &mapping->private_list;
814	struct address_space *buffer_mapping = mapping->private_data;
815
816	spin_lock(&buffer_mapping->private_lock);
817	while (!list_empty(list)) {
818	struct buffer_head *bh = BH_ENTRY(list->next);
819	if (buffer_dirty(bh)) {
820	ret = 0;
821	break;
822	}
823	__remove_assoc_queue(bh);
824	}
825	spin_unlock(&buffer_mapping->private_lock);
826	}
827	return ret;
828	}
829
830	/*
831	* Create the appropriate buffers when given a page for data area and
832	* the size of each buffer.. Use the bh->b_this_page linked list to
833	* follow the buffers created. Return NULL if unable to create more
834	* buffers.
835	*
836	* The retry flag is used to differentiate async IO (paging, swapping)
837	* which may not fail from ordinary buffer allocations.
838	*/
839	struct buffer_head alloc_page_buffers(struct page page, unsigned long size,
840	bool retry)
841	{
842	struct buffer_head bh, head;
843	gfp_t gfp = GFP_NOFS \| __GFP_ACCOUNT;
844	long offset;
845	struct mem_cgroup *memcg;
846
847	if (retry)
848	gfp \|= __GFP_NOFAIL;
849
850	memcg = get_mem_cgroup_from_page(page);
851	memalloc_use_memcg(memcg);
852
853	head = NULL;
854	offset = PAGE_SIZE;
855	while ((offset -= size) >= 0) {
856	bh = alloc_buffer_head(gfp);
857	if (!bh)
858	goto no_grow;
859
860	bh->b_this_page = head;
861	bh->b_blocknr = -1;
862	head = bh;
863
864	bh->b_size = size;
865
866	/* Link the buffer to its page */
867	set_bh_page(bh, page, offset);
868	}
869	out:
870	memalloc_unuse_memcg();
871	mem_cgroup_put(memcg);
872	return head;
873	/*
874	* In case anything failed, we just free everything we got.
875	*/
876	no_grow:
877	if (head) {
878	do {
879	bh = head;
880	head = head->b_this_page;
881	free_buffer_head(bh);
882	} while (head);
883	}
884
885	goto out;
886	}
887	EXPORT_SYMBOL_GPL(alloc_page_buffers);
888
889	static inline void
890	link_dev_buffers(struct page page, struct buffer_head head)
891	{
892	struct buffer_head bh, tail;
893
894	bh = head;
895	do {
896	tail = bh;
897	bh = bh->b_this_page;
898	} while (bh);
899	tail->b_this_page = head;
900	attach_page_private(page, head);
901	}
902
903	static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size)
904	{
905	sector_t retval = ~((sector_t)0);
906	loff_t sz = i_size_read(bdev->bd_inode);
907
908	if (sz) {
909	unsigned int sizebits = blksize_bits(size);
910	retval = (sz >> sizebits);
911	}
912	return retval;
913	}
914
915	/*
916	* Initialise the state of a blockdev page's buffers.
917	*/
918	static sector_t
919	init_page_buffers(struct page page, struct block_device bdev,
920	sector_t block, int size)
921	{
922	struct buffer_head *head = page_buffers(page);
923	struct buffer_head *bh = head;
924	int uptodate = PageUptodate(page);
925	sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size);
926
927	do {
928	if (!buffer_mapped(bh)) {
929	bh->b_end_io = NULL;
930	bh->b_private = NULL;
931	bh->b_bdev = bdev;
932	bh->b_blocknr = block;
933	if (uptodate)
934	set_buffer_uptodate(bh);
935	if (block < end_block)
936	set_buffer_mapped(bh);
937	}
938	block++;
939	bh = bh->b_this_page;
940	} while (bh != head);
941
942	/*
943	* Caller needs to validate requested block against end of device.
944	*/
945	return end_block;
946	}
947
948	/*
949	* Create the page-cache page that contains the requested block.
950	*
951	* This is used purely for blockdev mappings.
952	*/
953	static int
954	grow_dev_page(struct block_device *bdev, sector_t block,
955	pgoff_t index, int size, int sizebits, gfp_t gfp)
956	{
957	struct inode *inode = bdev->bd_inode;
958	struct page *page;
959	struct buffer_head *bh;
960	sector_t end_block;
961	int ret = 0;
962	gfp_t gfp_mask;
963
964	gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) \| gfp;
965
966	/*
967	* XXX: __getblk_slow() can not really deal with failure and
968	* will endlessly loop on improvised global reclaim. Prefer
969	* looping in the allocator rather than here, at least that
970	* code knows what it's doing.
971	*/
972	gfp_mask \|= __GFP_NOFAIL;
973
974	page = find_or_create_page(inode->i_mapping, index, gfp_mask);
975
976	BUG_ON(!PageLocked(page));
977
978	if (page_has_buffers(page)) {
979	bh = page_buffers(page);
980	if (bh->b_size == size) {
981	end_block = init_page_buffers(page, bdev,
982	(sector_t)index << sizebits,
983	size);
984	goto done;
985	}
986	if (!try_to_free_buffers(page))
987	goto failed;
988	}
989
990	/*
991	* Allocate some buffers for this page
992	*/
993	bh = alloc_page_buffers(page, size, true);
994
995	/*
996	* Link the page to the buffers and initialise them. Take the
997	* lock to be atomic wrt __find_get_block(), which does not
998	* run under the page lock.
999	*/
1000	spin_lock(&inode->i_mapping->private_lock);
1001	link_dev_buffers(page, bh);
1002	end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits,
1003	size);
1004	spin_unlock(&inode->i_mapping->private_lock);
1005	done:
1006	ret = (block < end_block) ? 1 : -ENXIO;
1007	failed:
1008	unlock_page(page);
1009	put_page(page);
1010	return ret;
1011	}
1012
1013	/*
1014	* Create buffers for the specified block device block's page. If
1015	* that page was dirty, the buffers are set dirty also.
1016	*/
1017	static int
1018	grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp)
1019	{
1020	pgoff_t index;
1021	int sizebits;
1022
1023	sizebits = -1;
1024	do {
1025	sizebits++;
1026	} while ((size << sizebits) < PAGE_SIZE);
1027
1028	index = block >> sizebits;
1029
1030	/*
1031	* Check for a block which wants to lie outside our maximum possible
1032	* pagecache index. (this comparison is done using sector_t types).
1033	*/
1034	if (unlikely(index != block >> sizebits)) {
1035	printk(KERN_ERR "%s: requested out-of-range block %llu for "
1036	"device %pg\n",
1037	__func__, (unsigned long long)block,
1038	bdev);
1039	return -EIO;
1040	}
1041
1042	/* Create a page with the proper size buffers.. */
1043	return grow_dev_page(bdev, block, index, size, sizebits, gfp);
1044	}
1045
1046	static struct buffer_head *
1047	__getblk_slow(struct block_device *bdev, sector_t block,
1048	unsigned size, gfp_t gfp)
1049	{
1050	/* Size must be multiple of hard sectorsize */
1051	if (unlikely(size & (bdev_logical_block_size(bdev)-1) \|\|
1052	(size < 512 \|\| size > PAGE_SIZE))) {
1053	printk(KERN_ERR "getblk(): invalid block size %d requested\n",
1054	size);
1055	printk(KERN_ERR "logical block size: %d\n",
1056	bdev_logical_block_size(bdev));
1057
1058	dump_stack();
1059	return NULL;
1060	}
1061
1062	for (;;) {
1063	struct buffer_head *bh;
1064	int ret;
1065
1066	bh = __find_get_block(bdev, block, size);
1067	if (bh)
1068	return bh;
1069
1070	ret = grow_buffers(bdev, block, size, gfp);
1071	if (ret < 0)
1072	return NULL;
1073	}
1074	}
1075
1076	/*
1077	* The relationship between dirty buffers and dirty pages:
1078	*
1079	* Whenever a page has any dirty buffers, the page's dirty bit is set, and
1080	* the page is tagged dirty in the page cache.
1081	*
1082	* At all times, the dirtiness of the buffers represents the dirtiness of
1083	* subsections of the page. If the page has buffers, the page dirty bit is
1084	* merely a hint about the true dirty state.
1085	*
1086	* When a page is set dirty in its entirety, all its buffers are marked dirty
1087	* (if the page has buffers).
1088	*
1089	* When a buffer is marked dirty, its page is dirtied, but the page's other
1090	* buffers are not.
1091	*
1092	* Also. When blockdev buffers are explicitly read with bread(), they
1093	* individually become uptodate. But their backing page remains not
1094	* uptodate - even if all of its buffers are uptodate. A subsequent
1095	* block_read_full_page() against that page will discover all the uptodate
1096	* buffers, will set the page uptodate and will perform no I/O.
1097	*/
1098
1099	/**
1100	* mark_buffer_dirty - mark a buffer_head as needing writeout
1101	* @bh: the buffer_head to mark dirty
1102	*
1103	* mark_buffer_dirty() will set the dirty bit against the buffer, then set
1104	* its backing page dirty, then tag the page as dirty in the page cache
1105	* and then attach the address_space's inode to its superblock's dirty
1106	* inode list.
1107	*
1108	* mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
1109	* i_pages lock and mapping->host->i_lock.
1110	*/
1111	void mark_buffer_dirty(struct buffer_head *bh)
1112	{
1113	WARN_ON_ONCE(!buffer_uptodate(bh));
1114
1115	trace_block_dirty_buffer(bh);
1116
1117	/*
1118	* Very carefully optimize the it-is-already-dirty case.
1119	*
1120	* Don't let the final "is it dirty" escape to before we
1121	* perhaps modified the buffer.
1122	*/
1123	if (buffer_dirty(bh)) {
1124	smp_mb();
1125	if (buffer_dirty(bh))
1126	return;
1127	}
1128
1129	if (!test_set_buffer_dirty(bh)) {
1130	struct page *page = bh->b_page;
1131	struct address_space *mapping = NULL;
1132
1133	lock_page_memcg(page);
1134	if (!TestSetPageDirty(page)) {
1135	mapping = page_mapping(page);
1136	if (mapping)
1137	__set_page_dirty(page, mapping, 0);
1138	}
1139	unlock_page_memcg(page);
1140	if (mapping)
1141	__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1142	}
1143	}
1144	EXPORT_SYMBOL(mark_buffer_dirty);
1145
1146	void mark_buffer_write_io_error(struct buffer_head *bh)
1147	{
1148	struct super_block *sb;
1149
1150	set_buffer_write_io_error(bh);
1151	/* FIXME: do we need to set this in both places? */
1152	if (bh->b_page && bh->b_page->mapping)
1153	mapping_set_error(bh->b_page->mapping, -EIO);
1154	if (bh->b_assoc_map)
1155	mapping_set_error(bh->b_assoc_map, -EIO);
1156	rcu_read_lock();
1157	sb = READ_ONCE(bh->b_bdev->bd_super);
1158	if (sb)
1159	errseq_set(&sb->s_wb_err, -EIO);
1160	rcu_read_unlock();
1161	}
1162	EXPORT_SYMBOL(mark_buffer_write_io_error);
1163
1164	/*
1165	* Decrement a buffer_head's reference count. If all buffers against a page
1166	* have zero reference count, are clean and unlocked, and if the page is clean
1167	* and unlocked then try_to_free_buffers() may strip the buffers from the page
1168	* in preparation for freeing it (sometimes, rarely, buffers are removed from
1169	* a page but it ends up not being freed, and buffers may later be reattached).
1170	*/
1171	void __brelse(struct buffer_head * buf)
1172	{
1173	if (atomic_read(&buf->b_count)) {
1174	put_bh(buf);
1175	return;
1176	}
1177	WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n");
1178	}
1179	EXPORT_SYMBOL(__brelse);
1180
1181	/*
1182	* bforget() is like brelse(), except it discards any
1183	* potentially dirty data.
1184	*/
1185	void __bforget(struct buffer_head *bh)
1186	{
1187	clear_buffer_dirty(bh);
1188	if (bh->b_assoc_map) {
1189	struct address_space *buffer_mapping = bh->b_page->mapping;
1190
1191	spin_lock(&buffer_mapping->private_lock);
1192	list_del_init(&bh->b_assoc_buffers);
1193	bh->b_assoc_map = NULL;
1194	spin_unlock(&buffer_mapping->private_lock);
1195	}
1196	__brelse(bh);
1197	}
1198	EXPORT_SYMBOL(__bforget);
1199
1200	static struct buffer_head __bread_slow(struct buffer_head bh)
1201	{
1202	lock_buffer(bh);
1203	if (buffer_uptodate(bh)) {
1204	unlock_buffer(bh);
1205	return bh;
1206	} else {
1207	get_bh(bh);
1208	bh->b_end_io = end_buffer_read_sync;
1209	submit_bh(REQ_OP_READ, 0, bh);
1210	wait_on_buffer(bh);
1211	if (buffer_uptodate(bh))
1212	return bh;
1213	}
1214	brelse(bh);
1215	return NULL;
1216	}
1217
1218	/*
1219	* Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1220	* The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1221	* refcount elevated by one when they're in an LRU. A buffer can only appear
1222	* once in a particular CPU's LRU. A single buffer can be present in multiple
1223	* CPU's LRUs at the same time.
1224	*
1225	* This is a transparent caching front-end to sb_bread(), sb_getblk() and
1226	* sb_find_get_block().
1227	*
1228	* The LRUs themselves only need locking against invalidate_bh_lrus. We use
1229	* a local interrupt disable for that.
1230	*/
1231
1232	#define BH_LRU_SIZE 16
1233
1234	struct bh_lru {
1235	struct buffer_head *bhs[BH_LRU_SIZE];
1236	};
1237
1238	static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }};
1239
1240	#ifdef CONFIG_SMP
1241	#define bh_lru_lock() local_irq_disable()
1242	#define bh_lru_unlock() local_irq_enable()
1243	#else
1244	#define bh_lru_lock() preempt_disable()
1245	#define bh_lru_unlock() preempt_enable()
1246	#endif
1247
1248	static inline void check_irqs_on(void)
1249	{
1250	#ifdef irqs_disabled
1251	BUG_ON(irqs_disabled());
1252	#endif
1253	}
1254
1255	/*
1256	* Install a buffer_head into this cpu's LRU. If not already in the LRU, it is
1257	* inserted at the front, and the buffer_head at the back if any is evicted.
1258	* Or, if already in the LRU it is moved to the front.
1259	*/
1260	static void bh_lru_install(struct buffer_head *bh)
1261	{
1262	struct buffer_head *evictee = bh;
1263	struct bh_lru *b;
1264	int i;
1265
1266	check_irqs_on();
1267	bh_lru_lock();
1268
1269	b = this_cpu_ptr(&bh_lrus);
1270	for (i = 0; i < BH_LRU_SIZE; i++) {
1271	swap(evictee, b->bhs[i]);
1272	if (evictee == bh) {
1273	bh_lru_unlock();
1274	return;
1275	}
1276	}
1277
1278	get_bh(bh);
1279	bh_lru_unlock();
1280	brelse(evictee);
1281	}
1282
1283	/*
1284	* Look up the bh in this cpu's LRU. If it's there, move it to the head.
1285	*/
1286	static struct buffer_head *
1287	lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
1288	{
1289	struct buffer_head *ret = NULL;
1290	unsigned int i;
1291
1292	check_irqs_on();
1293	bh_lru_lock();
1294	for (i = 0; i < BH_LRU_SIZE; i++) {
1295	struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]);
1296
1297	if (bh && bh->b_blocknr == block && bh->b_bdev == bdev &&
1298	bh->b_size == size) {
1299	if (i) {
1300	while (i) {
1301	__this_cpu_write(bh_lrus.bhs[i],
1302	__this_cpu_read(bh_lrus.bhs[i - 1]));
1303	i--;
1304	}
1305	__this_cpu_write(bh_lrus.bhs[0], bh);
1306	}
1307	get_bh(bh);
1308	ret = bh;
1309	break;
1310	}
1311	}
1312	bh_lru_unlock();
1313	return ret;
1314	}
1315
1316	/*
1317	* Perform a pagecache lookup for the matching buffer. If it's there, refresh
1318	* it in the LRU and mark it as accessed. If it is not present then return
1319	* NULL
1320	*/
1321	struct buffer_head *
1322	__find_get_block(struct block_device *bdev, sector_t block, unsigned size)
1323	{
1324	struct buffer_head *bh = lookup_bh_lru(bdev, block, size);
1325
1326	if (bh == NULL) {
1327	/* __find_get_block_slow will mark the page accessed */
1328	bh = __find_get_block_slow(bdev, block);
1329	if (bh)
1330	bh_lru_install(bh);
1331	} else
1332	touch_buffer(bh);
1333
1334	return bh;
1335	}
1336	EXPORT_SYMBOL(__find_get_block);
1337
1338	/*
1339	* __getblk_gfp() will locate (and, if necessary, create) the buffer_head
1340	* which corresponds to the passed block_device, block and size. The
1341	* returned buffer has its reference count incremented.
1342	*
1343	* __getblk_gfp() will lock up the machine if grow_dev_page's
1344	* try_to_free_buffers() attempt is failing. FIXME, perhaps?
1345	*/
1346	struct buffer_head *
1347	__getblk_gfp(struct block_device *bdev, sector_t block,
1348	unsigned size, gfp_t gfp)
1349	{
1350	struct buffer_head *bh = __find_get_block(bdev, block, size);
1351
1352	might_sleep();
1353	if (bh == NULL)
1354	bh = __getblk_slow(bdev, block, size, gfp);
1355	return bh;
1356	}
1357	EXPORT_SYMBOL(__getblk_gfp);
1358
1359	/*
1360	* Do async read-ahead on a buffer..
1361	*/
1362	void __breadahead(struct block_device *bdev, sector_t block, unsigned size)
1363	{
1364	struct buffer_head *bh = __getblk(bdev, block, size);
1365	if (likely(bh)) {
1366	ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh);
1367	brelse(bh);
1368	}
1369	}
1370	EXPORT_SYMBOL(__breadahead);
1371
1372	void __breadahead_gfp(struct block_device *bdev, sector_t block, unsigned size,
1373	gfp_t gfp)
1374	{
1375	struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
1376	if (likely(bh)) {
1377	ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh);
1378	brelse(bh);
1379	}
1380	}
1381	EXPORT_SYMBOL(__breadahead_gfp);
1382
1383	/**
1384	* __bread_gfp() - reads a specified block and returns the bh
1385	* @bdev: the block_device to read from
1386	* @block: number of block
1387	* @size: size (in bytes) to read
1388	* @gfp: page allocation flag
1389	*
1390	* Reads a specified block, and returns buffer head that contains it.
1391	* The page cache can be allocated from non-movable area
1392	* not to prevent page migration if you set gfp to zero.
1393	* It returns NULL if the block was unreadable.
1394	*/
1395	struct buffer_head *
1396	__bread_gfp(struct block_device *bdev, sector_t block,
1397	unsigned size, gfp_t gfp)
1398	{
1399	struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
1400
1401	if (likely(bh) && !buffer_uptodate(bh))
1402	bh = __bread_slow(bh);
1403	return bh;
1404	}
1405	EXPORT_SYMBOL(__bread_gfp);
1406
1407	/*
1408	* invalidate_bh_lrus() is called rarely - but not only at unmount.
1409	* This doesn't race because it runs in each cpu either in irq
1410	* or with preempt disabled.
1411	*/
1412	static void invalidate_bh_lru(void *arg)
1413	{
1414	struct bh_lru *b = &get_cpu_var(bh_lrus);
1415	int i;
1416
1417	for (i = 0; i < BH_LRU_SIZE; i++) {
1418	brelse(b->bhs[i]);
1419	b->bhs[i] = NULL;
1420	}
1421	put_cpu_var(bh_lrus);
1422	}
1423
1424	static bool has_bh_in_lru(int cpu, void *dummy)
1425	{
1426	struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu);
1427	int i;
1428
1429	for (i = 0; i < BH_LRU_SIZE; i++) {
1430	if (b->bhs[i])
1431	return true;
1432	}
1433
1434	return false;
1435	}
1436
1437	void invalidate_bh_lrus(void)
1438	{
1439	on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1);
1440	}
1441	EXPORT_SYMBOL_GPL(invalidate_bh_lrus);
1442
1443	void set_bh_page(struct buffer_head *bh,
1444	struct page *page, unsigned long offset)
1445	{
1446	bh->b_page = page;
1447	BUG_ON(offset >= PAGE_SIZE);
1448	if (PageHighMem(page))
1449	/*
1450	* This catches illegal uses and preserves the offset:
1451	*/
1452	bh->b_data = (char *)(0 + offset);
1453	else
1454	bh->b_data = page_address(page) + offset;
1455	}
1456	EXPORT_SYMBOL(set_bh_page);
1457
1458	/*
1459	* Called when truncating a buffer on a page completely.
1460	*/
1461
1462	/* Bits that are cleared during an invalidate */
1463	#define BUFFER_FLAGS_DISCARD \
1464	(1 << BH_Mapped \| 1 << BH_New \| 1 << BH_Req \| \
1465	1 << BH_Delay \| 1 << BH_Unwritten)
1466
1467	static void discard_buffer(struct buffer_head * bh)
1468	{
1469	unsigned long b_state, b_state_old;
1470
1471	lock_buffer(bh);
1472	clear_buffer_dirty(bh);
1473	bh->b_bdev = NULL;
1474	b_state = bh->b_state;
1475	for (;;) {
1476	b_state_old = cmpxchg(&bh->b_state, b_state,
1477	(b_state & ~BUFFER_FLAGS_DISCARD));
1478	if (b_state_old == b_state)
1479	break;
1480	b_state = b_state_old;
1481	}
1482	unlock_buffer(bh);
1483	}
1484
1485	/**
1486	* block_invalidatepage - invalidate part or all of a buffer-backed page
1487	*
1488	* @page: the page which is affected
1489	* @offset: start of the range to invalidate
1490	* @length: length of the range to invalidate
1491	*
1492	* block_invalidatepage() is called when all or part of the page has become
1493	* invalidated by a truncate operation.
1494	*
1495	* block_invalidatepage() does not have to release all buffers, but it must
1496	* ensure that no dirty buffer is left outside @offset and that no I/O
1497	* is underway against any of the blocks which are outside the truncation
1498	* point. Because the caller is about to free (and possibly reuse) those
1499	* blocks on-disk.
1500	*/
1501	void block_invalidatepage(struct page *page, unsigned int offset,
1502	unsigned int length)
1503	{
1504	struct buffer_head head, bh, *next;
1505	unsigned int curr_off = 0;
1506	unsigned int stop = length + offset;
1507
1508	BUG_ON(!PageLocked(page));
1509	if (!page_has_buffers(page))
1510	goto out;
1511
1512	/*
1513	* Check for overflow
1514	*/
1515	BUG_ON(stop > PAGE_SIZE \|\| stop < length);
1516
1517	head = page_buffers(page);
1518	bh = head;
1519	do {
1520	unsigned int next_off = curr_off + bh->b_size;
1521	next = bh->b_this_page;
1522
1523	/*
1524	* Are we still fully in range ?
1525	*/
1526	if (next_off > stop)
1527	goto out;
1528
1529	/*
1530	* is this block fully invalidated?
1531	*/
1532	if (offset <= curr_off)
1533	discard_buffer(bh);
1534	curr_off = next_off;
1535	bh = next;
1536	} while (bh != head);
1537
1538	/*
1539	* We release buffers only if the entire page is being invalidated.
1540	* The get_block cached value has been unconditionally invalidated,
1541	* so real IO is not possible anymore.
1542	*/
1543	if (length == PAGE_SIZE)
1544	try_to_release_page(page, 0);
1545	out:
1546	return;
1547	}
1548	EXPORT_SYMBOL(block_invalidatepage);
1549
1550
1551	/*
1552	* We attach and possibly dirty the buffers atomically wrt
1553	* __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1554	* is already excluded via the page lock.
1555	*/
1556	void create_empty_buffers(struct page *page,
1557	unsigned long blocksize, unsigned long b_state)
1558	{
1559	struct buffer_head bh, head, *tail;
1560
1561	head = alloc_page_buffers(page, blocksize, true);
1562	bh = head;
1563	do {
1564	bh->b_state \|= b_state;
1565	tail = bh;
1566	bh = bh->b_this_page;
1567	} while (bh);
1568	tail->b_this_page = head;
1569
1570	spin_lock(&page->mapping->private_lock);
1571	if (PageUptodate(page) \|\| PageDirty(page)) {
1572	bh = head;
1573	do {
1574	if (PageDirty(page))
1575	set_buffer_dirty(bh);
1576	if (PageUptodate(page))
1577	set_buffer_uptodate(bh);
1578	bh = bh->b_this_page;
1579	} while (bh != head);
1580	}
1581	attach_page_private(page, head);
1582	spin_unlock(&page->mapping->private_lock);
1583	}
1584	EXPORT_SYMBOL(create_empty_buffers);
1585
1586	/**
1587	* clean_bdev_aliases: clean a range of buffers in block device
1588	* @bdev: Block device to clean buffers in
1589	* @block: Start of a range of blocks to clean
1590	* @len: Number of blocks to clean
1591	*
1592	* We are taking a range of blocks for data and we don't want writeback of any
1593	* buffer-cache aliases starting from return from this function and until the
1594	* moment when something will explicitly mark the buffer dirty (hopefully that
1595	* will not happen until we will free that block ;-) We don't even need to mark
1596	* it not-uptodate - nobody can expect anything from a newly allocated buffer
1597	* anyway. We used to use unmap_buffer() for such invalidation, but that was
1598	* wrong. We definitely don't want to mark the alias unmapped, for example - it
1599	* would confuse anyone who might pick it with bread() afterwards...
1600	*
1601	* Also.. Note that bforget() doesn't lock the buffer. So there can be
1602	* writeout I/O going on against recently-freed buffers. We don't wait on that
1603	* I/O in bforget() - it's more efficient to wait on the I/O only if we really
1604	* need to. That happens here.
1605	*/
1606	void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len)
1607	{
1608	struct inode *bd_inode = bdev->bd_inode;
1609	struct address_space *bd_mapping = bd_inode->i_mapping;
1610	struct pagevec pvec;
1611	pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
1612	pgoff_t end;
1613	int i, count;
1614	struct buffer_head *bh;
1615	struct buffer_head *head;
1616
1617	end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits);
1618	pagevec_init(&pvec);
1619	while (pagevec_lookup_range(&pvec, bd_mapping, &index, end)) {
1620	count = pagevec_count(&pvec);
1621	for (i = 0; i < count; i++) {
1622	struct page *page = pvec.pages[i];
1623
1624	if (!page_has_buffers(page))
1625	continue;
1626	/*
1627	* We use page lock instead of bd_mapping->private_lock
1628	* to pin buffers here since we can afford to sleep and
1629	* it scales better than a global spinlock lock.
1630	*/
1631	lock_page(page);
1632	/* Recheck when the page is locked which pins bhs */
1633	if (!page_has_buffers(page))
1634	goto unlock_page;
1635	head = page_buffers(page);
1636	bh = head;
1637	do {
1638	if (!buffer_mapped(bh) \|\| (bh->b_blocknr < block))
1639	goto next;
1640	if (bh->b_blocknr >= block + len)
1641	break;
1642	clear_buffer_dirty(bh);
1643	wait_on_buffer(bh);
1644	clear_buffer_req(bh);
1645	next:
1646	bh = bh->b_this_page;
1647	} while (bh != head);
1648	unlock_page:
1649	unlock_page(page);
1650	}
1651	pagevec_release(&pvec);
1652	cond_resched();
1653	/* End of range already reached? */
1654	if (index > end \|\| !index)
1655	break;
1656	}
1657	}
1658	EXPORT_SYMBOL(clean_bdev_aliases);
1659
1660	/*
1661	* Size is a power-of-two in the range 512..PAGE_SIZE,
1662	* and the case we care about most is PAGE_SIZE.
1663	*
1664	* So this could possibly be written with those
1665	* constraints in mind (relevant mostly if some
1666	* architecture has a slow bit-scan instruction)
1667	*/
1668	static inline int block_size_bits(unsigned int blocksize)
1669	{
1670	return ilog2(blocksize);
1671	}
1672
1673	static struct buffer_head create_page_buffers(struct page page, struct inode *inode, unsigned int b_state)
1674	{
1675	BUG_ON(!PageLocked(page));
1676
1677	if (!page_has_buffers(page))
1678	create_empty_buffers(page, 1 << READ_ONCE(inode->i_blkbits),
1679	b_state);
1680	return page_buffers(page);
1681	}
1682
1683	/*
1684	* NOTE! All mapped/uptodate combinations are valid:
1685	*
1686	* Mapped Uptodate Meaning
1687	*
1688	* No No "unknown" - must do get_block()
1689	* No Yes "hole" - zero-filled
1690	* Yes No "allocated" - allocated on disk, not read in
1691	* Yes Yes "valid" - allocated and up-to-date in memory.
1692	*
1693	* "Dirty" is valid only with the last case (mapped+uptodate).
1694	*/
1695
1696	/*
1697	* While block_write_full_page is writing back the dirty buffers under
1698	* the page lock, whoever dirtied the buffers may decide to clean them
1699	* again at any time. We handle that by only looking at the buffer
1700	* state inside lock_buffer().
1701	*
1702	* If block_write_full_page() is called for regular writeback
1703	* (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1704	* locked buffer. This only can happen if someone has written the buffer
1705	* directly, with submit_bh(). At the address_space level PageWriteback
1706	* prevents this contention from occurring.
1707	*
1708	* If block_write_full_page() is called with wbc->sync_mode ==
1709	* WB_SYNC_ALL, the writes are posted using REQ_SYNC; this
1710	* causes the writes to be flagged as synchronous writes.
1711	*/
1712	int __block_write_full_page(struct inode inode, struct page page,
1713	get_block_t get_block, struct writeback_control wbc,
1714	bh_end_io_t *handler)
1715	{
1716	int err;
1717	sector_t block;
1718	sector_t last_block;
1719	struct buffer_head bh, head;
1720	unsigned int blocksize, bbits;
1721	int nr_underway = 0;
1722	int write_flags = wbc_to_write_flags(wbc);
1723
1724	head = create_page_buffers(page, inode,
1725	(1 << BH_Dirty)\|(1 << BH_Uptodate));
1726
1727	/*
1728	* Be very careful. We have no exclusion from __set_page_dirty_buffers
1729	* here, and the (potentially unmapped) buffers may become dirty at
1730	* any time. If a buffer becomes dirty here after we've inspected it
1731	* then we just miss that fact, and the page stays dirty.
1732	*
1733	* Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1734	* handle that here by just cleaning them.
1735	*/
1736
1737	bh = head;
1738	blocksize = bh->b_size;
1739	bbits = block_size_bits(blocksize);
1740
1741	block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1742	last_block = (i_size_read(inode) - 1) >> bbits;
1743
1744	/*
1745	* Get all the dirty buffers mapped to disk addresses and
1746	* handle any aliases from the underlying blockdev's mapping.
1747	*/
1748	do {
1749	if (block > last_block) {
1750	/*
1751	* mapped buffers outside i_size will occur, because
1752	* this page can be outside i_size when there is a
1753	* truncate in progress.
1754	*/
1755	/*
1756	* The buffer was zeroed by block_write_full_page()
1757	*/
1758	clear_buffer_dirty(bh);
1759	set_buffer_uptodate(bh);
1760	} else if ((!buffer_mapped(bh) \|\| buffer_delay(bh)) &&
1761	buffer_dirty(bh)) {
1762	WARN_ON(bh->b_size != blocksize);
1763	err = get_block(inode, block, bh, 1);
1764	if (err)
1765	goto recover;
1766	clear_buffer_delay(bh);
1767	if (buffer_new(bh)) {
1768	/* blockdev mappings never come here */
1769	clear_buffer_new(bh);
1770	clean_bdev_bh_alias(bh);
1771	}
1772	}
1773	bh = bh->b_this_page;
1774	block++;
1775	} while (bh != head);
1776
1777	do {
1778	if (!buffer_mapped(bh))
1779	continue;
1780	/*
1781	* If it's a fully non-blocking write attempt and we cannot
1782	* lock the buffer then redirty the page. Note that this can
1783	* potentially cause a busy-wait loop from writeback threads
1784	* and kswapd activity, but those code paths have their own
1785	* higher-level throttling.
1786	*/
1787	if (wbc->sync_mode != WB_SYNC_NONE) {
1788	lock_buffer(bh);
1789	} else if (!trylock_buffer(bh)) {
1790	redirty_page_for_writepage(wbc, page);
1791	continue;
1792	}
1793	if (test_clear_buffer_dirty(bh)) {
1794	mark_buffer_async_write_endio(bh, handler);
1795	} else {
1796	unlock_buffer(bh);
1797	}
1798	} while ((bh = bh->b_this_page) != head);
1799
1800	/*
1801	* The page and its buffers are protected by PageWriteback(), so we can
1802	* drop the bh refcounts early.
1803	*/
1804	BUG_ON(PageWriteback(page));
1805	set_page_writeback(page);
1806
1807	do {
1808	struct buffer_head *next = bh->b_this_page;
1809	if (buffer_async_write(bh)) {
1810	submit_bh_wbc(REQ_OP_WRITE, write_flags, bh,
1811	inode->i_write_hint, wbc);
1812	nr_underway++;
1813	}
1814	bh = next;
1815	} while (bh != head);
1816	unlock_page(page);
1817
1818	err = 0;
1819	done:
1820	if (nr_underway == 0) {
1821	/*
1822	* The page was marked dirty, but the buffers were
1823	* clean. Someone wrote them back by hand with
1824	* ll_rw_block/submit_bh. A rare case.
1825	*/
1826	end_page_writeback(page);
1827
1828	/*
1829	* The page and buffer_heads can be released at any time from
1830	* here on.
1831	*/
1832	}
1833	return err;
1834
1835	recover:
1836	/*
1837	* ENOSPC, or some other error. We may already have added some
1838	* blocks to the file, so we need to write these out to avoid
1839	* exposing stale data.
1840	* The page is currently locked and not marked for writeback
1841	*/
1842	bh = head;
1843	/* Recovery: lock and submit the mapped buffers */
1844	do {
1845	if (buffer_mapped(bh) && buffer_dirty(bh) &&
1846	!buffer_delay(bh)) {
1847	lock_buffer(bh);
1848	mark_buffer_async_write_endio(bh, handler);
1849	} else {
1850	/*
1851	* The buffer may have been set dirty during
1852	* attachment to a dirty page.
1853	*/
1854	clear_buffer_dirty(bh);
1855	}
1856	} while ((bh = bh->b_this_page) != head);
1857	SetPageError(page);
1858	BUG_ON(PageWriteback(page));
1859	mapping_set_error(page->mapping, err);
1860	set_page_writeback(page);
1861	do {
1862	struct buffer_head *next = bh->b_this_page;
1863	if (buffer_async_write(bh)) {
1864	clear_buffer_dirty(bh);
1865	submit_bh_wbc(REQ_OP_WRITE, write_flags, bh,
1866	inode->i_write_hint, wbc);
1867	nr_underway++;
1868	}
1869	bh = next;
1870	} while (bh != head);
1871	unlock_page(page);
1872	goto done;
1873	}
1874	EXPORT_SYMBOL(__block_write_full_page);
1875
1876	/*
1877	* If a page has any new buffers, zero them out here, and mark them uptodate
1878	* and dirty so they'll be written out (in order to prevent uninitialised
1879	* block data from leaking). And clear the new bit.
1880	*/
1881	void page_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1882	{
1883	unsigned int block_start, block_end;
1884	struct buffer_head head, bh;
1885
1886	BUG_ON(!PageLocked(page));
1887	if (!page_has_buffers(page))
1888	return;
1889
1890	bh = head = page_buffers(page);
1891	block_start = 0;
1892	do {
1893	block_end = block_start + bh->b_size;
1894
1895	if (buffer_new(bh)) {
1896	if (block_end > from && block_start < to) {
1897	if (!PageUptodate(page)) {
1898	unsigned start, size;
1899
1900	start = max(from, block_start);
1901	size = min(to, block_end) - start;
1902
1903	zero_user(page, start, size);
1904	set_buffer_uptodate(bh);
1905	}
1906
1907	clear_buffer_new(bh);
1908	mark_buffer_dirty(bh);
1909	}
1910	}
1911
1912	block_start = block_end;
1913	bh = bh->b_this_page;
1914	} while (bh != head);
1915	}
1916	EXPORT_SYMBOL(page_zero_new_buffers);
1917
1918	static void
1919	iomap_to_bh(struct inode inode, sector_t block, struct buffer_head bh,
1920	struct iomap *iomap)
1921	{
1922	loff_t offset = block << inode->i_blkbits;
1923
1924	bh->b_bdev = iomap->bdev;
1925
1926	/*
1927	* Block points to offset in file we need to map, iomap contains
1928	* the offset at which the map starts. If the map ends before the
1929	* current block, then do not map the buffer and let the caller
1930	* handle it.
1931	*/
1932	BUG_ON(offset >= iomap->offset + iomap->length);
1933
1934	switch (iomap->type) {
1935	case IOMAP_HOLE:
1936	/*
1937	* If the buffer is not up to date or beyond the current EOF,
1938	* we need to mark it as new to ensure sub-block zeroing is
1939	* executed if necessary.
1940	*/
1941	if (!buffer_uptodate(bh) \|\|
1942	(offset >= i_size_read(inode)))
1943	set_buffer_new(bh);
1944	break;
1945	case IOMAP_DELALLOC:
1946	if (!buffer_uptodate(bh) \|\|
1947	(offset >= i_size_read(inode)))
1948	set_buffer_new(bh);
1949	set_buffer_uptodate(bh);
1950	set_buffer_mapped(bh);
1951	set_buffer_delay(bh);
1952	break;
1953	case IOMAP_UNWRITTEN:
1954	/*
1955	* For unwritten regions, we always need to ensure that regions
1956	* in the block we are not writing to are zeroed. Mark the
1957	* buffer as new to ensure this.
1958	*/
1959	set_buffer_new(bh);
1960	set_buffer_unwritten(bh);
1961	/* FALLTHRU */
1962	case IOMAP_MAPPED:
1963	if ((iomap->flags & IOMAP_F_NEW) \|\|
1964	offset >= i_size_read(inode))
1965	set_buffer_new(bh);
1966	bh->b_blocknr = (iomap->addr + offset - iomap->offset) >>
1967	inode->i_blkbits;
1968	set_buffer_mapped(bh);
1969	break;
1970	}
1971	}
1972
1973	int __block_write_begin_int(struct page *page, loff_t pos, unsigned len,
1974	get_block_t get_block, struct iomap iomap)
1975	{
1976	unsigned from = pos & (PAGE_SIZE - 1);
1977	unsigned to = from + len;
1978	struct inode *inode = page->mapping->host;
1979	unsigned block_start, block_end;
1980	sector_t block;
1981	int err = 0;
1982	unsigned blocksize, bbits;
1983	struct buffer_head bh, head, wait[2], *wait_bh=wait;
1984
1985	BUG_ON(!PageLocked(page));
1986	BUG_ON(from > PAGE_SIZE);
1987	BUG_ON(to > PAGE_SIZE);
1988	BUG_ON(from > to);
1989
1990	head = create_page_buffers(page, inode, 0);
1991	blocksize = head->b_size;
1992	bbits = block_size_bits(blocksize);
1993
1994	block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1995
1996	for(bh = head, block_start = 0; bh != head \|\| !block_start;
1997	block++, block_start=block_end, bh = bh->b_this_page) {
1998	block_end = block_start + blocksize;
1999	if (block_end <= from \|\| block_start >= to) {
2000	if (PageUptodate(page)) {
2001	if (!buffer_uptodate(bh))
2002	set_buffer_uptodate(bh);
2003	}
2004	continue;
2005	}
2006	if (buffer_new(bh))
2007	clear_buffer_new(bh);
2008	if (!buffer_mapped(bh)) {
2009	WARN_ON(bh->b_size != blocksize);
2010	if (get_block) {
2011	err = get_block(inode, block, bh, 1);
2012	if (err)
2013	break;
2014	} else {
2015	iomap_to_bh(inode, block, bh, iomap);
2016	}
2017
2018	if (buffer_new(bh)) {
2019	clean_bdev_bh_alias(bh);
2020	if (PageUptodate(page)) {
2021	clear_buffer_new(bh);
2022	set_buffer_uptodate(bh);
2023	mark_buffer_dirty(bh);
2024	continue;
2025	}
2026	if (block_end > to \|\| block_start < from)
2027	zero_user_segments(page,
2028	to, block_end,
2029	block_start, from);
2030	continue;
2031	}
2032	}
2033	if (PageUptodate(page)) {
2034	if (!buffer_uptodate(bh))
2035	set_buffer_uptodate(bh);
2036	continue;
2037	}
2038	if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
2039	!buffer_unwritten(bh) &&
2040	(block_start < from \|\| block_end > to)) {
2041	ll_rw_block(REQ_OP_READ, 0, 1, &bh);
2042	*wait_bh++=bh;
2043	}
2044	}
2045	/*
2046	* If we issued read requests - let them complete.
2047	*/
2048	while(wait_bh > wait) {
2049	wait_on_buffer(*--wait_bh);
2050	if (!buffer_uptodate(*wait_bh))
2051	err = -EIO;
2052	}
2053	if (unlikely(err))
2054	page_zero_new_buffers(page, from, to);
2055	return err;
2056	}
2057
2058	int __block_write_begin(struct page *page, loff_t pos, unsigned len,
2059	get_block_t *get_block)
2060	{
2061	return __block_write_begin_int(page, pos, len, get_block, NULL);
2062	}
2063	EXPORT_SYMBOL(__block_write_begin);
2064
2065	static int __block_commit_write(struct inode inode, struct page page,
2066	unsigned from, unsigned to)
2067	{
2068	unsigned block_start, block_end;
2069	int partial = 0;
2070	unsigned blocksize;
2071	struct buffer_head bh, head;
2072
2073	bh = head = page_buffers(page);
2074	blocksize = bh->b_size;
2075
2076	block_start = 0;
2077	do {
2078	block_end = block_start + blocksize;
2079	if (block_end <= from \|\| block_start >= to) {
2080	if (!buffer_uptodate(bh))
2081	partial = 1;
2082	} else {
2083	set_buffer_uptodate(bh);
2084	mark_buffer_dirty(bh);
2085	}
2086	clear_buffer_new(bh);
2087
2088	block_start = block_end;
2089	bh = bh->b_this_page;
2090	} while (bh != head);
2091
2092	/*
2093	* If this is a partial write which happened to make all buffers
2094	* uptodate then we can optimize away a bogus readpage() for
2095	* the next read(). Here we 'discover' whether the page went
2096	* uptodate as a result of this (potentially partial) write.
2097	*/
2098	if (!partial)
2099	SetPageUptodate(page);
2100	return 0;
2101	}
2102
2103	/*
2104	* block_write_begin takes care of the basic task of block allocation and
2105	* bringing partial write blocks uptodate first.
2106	*
2107	* The filesystem needs to handle block truncation upon failure.
2108	*/
2109	int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
2110	unsigned flags, struct page *pagep, get_block_t get_block)
2111	{
2112	pgoff_t index = pos >> PAGE_SHIFT;
2113	struct page *page;
2114	int status;
2115
2116	page = grab_cache_page_write_begin(mapping, index, flags);
2117	if (!page)
2118	return -ENOMEM;
2119
2120	status = __block_write_begin(page, pos, len, get_block);
2121	if (unlikely(status)) {
2122	unlock_page(page);
2123	put_page(page);
2124	page = NULL;
2125	}
2126
2127	*pagep = page;
2128	return status;
2129	}
2130	EXPORT_SYMBOL(block_write_begin);
2131
2132	int block_write_end(struct file file, struct address_space mapping,
2133	loff_t pos, unsigned len, unsigned copied,
2134	struct page page, void fsdata)
2135	{
2136	struct inode *inode = mapping->host;
2137	unsigned start;
2138
2139	start = pos & (PAGE_SIZE - 1);
2140
2141	if (unlikely(copied < len)) {
2142	/*
2143	* The buffers that were written will now be uptodate, so we
2144	* don't have to worry about a readpage reading them and
2145	* overwriting a partial write. However if we have encountered
2146	* a short write and only partially written into a buffer, it
2147	* will not be marked uptodate, so a readpage might come in and
2148	* destroy our partial write.
2149	*
2150	* Do the simplest thing, and just treat any short write to a
2151	* non uptodate page as a zero-length write, and force the
2152	* caller to redo the whole thing.
2153	*/
2154	if (!PageUptodate(page))
2155	copied = 0;
2156
2157	page_zero_new_buffers(page, start+copied, start+len);
2158	}
2159	flush_dcache_page(page);
2160
2161	/* This could be a short (even 0-length) commit */
2162	__block_commit_write(inode, page, start, start+copied);
2163
2164	return copied;
2165	}
2166	EXPORT_SYMBOL(block_write_end);
2167
2168	int generic_write_end(struct file file, struct address_space mapping,
2169	loff_t pos, unsigned len, unsigned copied,
2170	struct page page, void fsdata)
2171	{
2172	struct inode *inode = mapping->host;
2173	loff_t old_size = inode->i_size;
2174	bool i_size_changed = false;
2175
2176	copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
2177
2178	/*
2179	* No need to use i_size_read() here, the i_size cannot change under us
2180	* because we hold i_rwsem.
2181	*
2182	* But it's important to update i_size while still holding page lock:
2183	* page writeout could otherwise come in and zero beyond i_size.
2184	*/
2185	if (pos + copied > inode->i_size) {
2186	i_size_write(inode, pos + copied);
2187	i_size_changed = true;
2188	}
2189
2190	unlock_page(page);
2191	put_page(page);
2192
2193	if (old_size < pos)
2194	pagecache_isize_extended(inode, old_size, pos);
2195	/*
2196	* Don't mark the inode dirty under page lock. First, it unnecessarily
2197	* makes the holding time of page lock longer. Second, it forces lock
2198	* ordering of page lock and transaction start for journaling
2199	* filesystems.
2200	*/
2201	if (i_size_changed)
2202	mark_inode_dirty(inode);
2203	return copied;
2204	}
2205	EXPORT_SYMBOL(generic_write_end);
2206
2207	/*
2208	* block_is_partially_uptodate checks whether buffers within a page are
2209	* uptodate or not.
2210	*
2211	* Returns true if all buffers which correspond to a file portion
2212	* we want to read are uptodate.
2213	*/
2214	int block_is_partially_uptodate(struct page *page, unsigned long from,
2215	unsigned long count)
2216	{
2217	unsigned block_start, block_end, blocksize;
2218	unsigned to;
2219	struct buffer_head bh, head;
2220	int ret = 1;
2221
2222	if (!page_has_buffers(page))
2223	return 0;
2224
2225	head = page_buffers(page);
2226	blocksize = head->b_size;
2227	to = min_t(unsigned, PAGE_SIZE - from, count);
2228	to = from + to;
2229	if (from < blocksize && to > PAGE_SIZE - blocksize)
2230	return 0;
2231
2232	bh = head;
2233	block_start = 0;
2234	do {
2235	block_end = block_start + blocksize;
2236	if (block_end > from && block_start < to) {
2237	if (!buffer_uptodate(bh)) {
2238	ret = 0;
2239	break;
2240	}
2241	if (block_end >= to)
2242	break;
2243	}
2244	block_start = block_end;
2245	bh = bh->b_this_page;
2246	} while (bh != head);
2247
2248	return ret;
2249	}
2250	EXPORT_SYMBOL(block_is_partially_uptodate);
2251
2252	/*
2253	* Generic "read page" function for block devices that have the normal
2254	* get_block functionality. This is most of the block device filesystems.
2255	* Reads the page asynchronously --- the unlock_buffer() and
2256	* set/clear_buffer_uptodate() functions propagate buffer state into the
2257	* page struct once IO has completed.
2258	*/
2259	int block_read_full_page(struct page page, get_block_t get_block)
2260	{
2261	struct inode *inode = page->mapping->host;
2262	sector_t iblock, lblock;
2263	struct buffer_head bh, head, *arr[MAX_BUF_PER_PAGE];
2264	unsigned int blocksize, bbits;
2265	int nr, i;
2266	int fully_mapped = 1;
2267
2268	head = create_page_buffers(page, inode, 0);
2269	blocksize = head->b_size;
2270	bbits = block_size_bits(blocksize);
2271
2272	iblock = (sector_t)page->index << (PAGE_SHIFT - bbits);
2273	lblock = (i_size_read(inode)+blocksize-1) >> bbits;
2274	bh = head;
2275	nr = 0;
2276	i = 0;
2277
2278	do {
2279	if (buffer_uptodate(bh))
2280	continue;
2281
2282	if (!buffer_mapped(bh)) {
2283	int err = 0;
2284
2285	fully_mapped = 0;
2286	if (iblock < lblock) {
2287	WARN_ON(bh->b_size != blocksize);
2288	err = get_block(inode, iblock, bh, 0);
2289	if (err)
2290	SetPageError(page);
2291	}
2292	if (!buffer_mapped(bh)) {
2293	zero_user(page, i * blocksize, blocksize);
2294	if (!err)
2295	set_buffer_uptodate(bh);
2296	continue;
2297	}
2298	/*
2299	* get_block() might have updated the buffer
2300	* synchronously
2301	*/
2302	if (buffer_uptodate(bh))
2303	continue;
2304	}
2305	arr[nr++] = bh;
2306	} while (i++, iblock++, (bh = bh->b_this_page) != head);
2307
2308	if (fully_mapped)
2309	SetPageMappedToDisk(page);
2310
2311	if (!nr) {
2312	/*
2313	* All buffers are uptodate - we can set the page uptodate
2314	* as well. But not if get_block() returned an error.
2315	*/
2316	if (!PageError(page))
2317	SetPageUptodate(page);
2318	unlock_page(page);
2319	return 0;
2320	}
2321
2322	/* Stage two: lock the buffers */
2323	for (i = 0; i < nr; i++) {
2324	bh = arr[i];
2325	lock_buffer(bh);
2326	mark_buffer_async_read(bh);
2327	}
2328
2329	/*
2330	* Stage 3: start the IO. Check for uptodateness
2331	* inside the buffer lock in case another process reading
2332	* the underlying blockdev brought it uptodate (the sct fix).
2333	*/
2334	for (i = 0; i < nr; i++) {
2335	bh = arr[i];
2336	if (buffer_uptodate(bh))
2337	end_buffer_async_read(bh, 1);
2338	else
2339	submit_bh(REQ_OP_READ, 0, bh);
2340	}
2341	return 0;
2342	}
2343	EXPORT_SYMBOL(block_read_full_page);
2344
2345	/* utility function for filesystems that need to do work on expanding
2346	* truncates. Uses filesystem pagecache writes to allow the filesystem to
2347	* deal with the hole.
2348	*/
2349	int generic_cont_expand_simple(struct inode *inode, loff_t size)
2350	{
2351	struct address_space *mapping = inode->i_mapping;
2352	struct page *page;
2353	void *fsdata;
2354	int err;
2355
2356	err = inode_newsize_ok(inode, size);
2357	if (err)
2358	goto out;
2359
2360	err = pagecache_write_begin(NULL, mapping, size, 0,
2361	AOP_FLAG_CONT_EXPAND, &page, &fsdata);
2362	if (err)
2363	goto out;
2364
2365	err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata);
2366	BUG_ON(err > 0);
2367
2368	out:
2369	return err;
2370	}
2371	EXPORT_SYMBOL(generic_cont_expand_simple);
2372
2373	static int cont_expand_zero(struct file file, struct address_space mapping,
2374	loff_t pos, loff_t *bytes)
2375	{
2376	struct inode *inode = mapping->host;
2377	unsigned int blocksize = i_blocksize(inode);
2378	struct page *page;
2379	void *fsdata;
2380	pgoff_t index, curidx;
2381	loff_t curpos;
2382	unsigned zerofrom, offset, len;
2383	int err = 0;
2384
2385	index = pos >> PAGE_SHIFT;
2386	offset = pos & ~PAGE_MASK;
2387
2388	while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) {
2389	zerofrom = curpos & ~PAGE_MASK;
2390	if (zerofrom & (blocksize-1)) {
2391	*bytes \|= (blocksize-1);
2392	(*bytes)++;
2393	}
2394	len = PAGE_SIZE - zerofrom;
2395
2396	err = pagecache_write_begin(file, mapping, curpos, len, 0,
2397	&page, &fsdata);
2398	if (err)
2399	goto out;
2400	zero_user(page, zerofrom, len);
2401	err = pagecache_write_end(file, mapping, curpos, len, len,
2402	page, fsdata);
2403	if (err < 0)
2404	goto out;
2405	BUG_ON(err != len);
2406	err = 0;
2407
2408	balance_dirty_pages_ratelimited(mapping);
2409
2410	if (fatal_signal_pending(current)) {
2411	err = -EINTR;
2412	goto out;
2413	}
2414	}
2415
2416	/* page covers the boundary, find the boundary offset */
2417	if (index == curidx) {
2418	zerofrom = curpos & ~PAGE_MASK;
2419	/* if we will expand the thing last block will be filled */
2420	if (offset <= zerofrom) {
2421	goto out;
2422	}
2423	if (zerofrom & (blocksize-1)) {
2424	*bytes \|= (blocksize-1);
2425	(*bytes)++;
2426	}
2427	len = offset - zerofrom;
2428
2429	err = pagecache_write_begin(file, mapping, curpos, len, 0,
2430	&page, &fsdata);
2431	if (err)
2432	goto out;
2433	zero_user(page, zerofrom, len);
2434	err = pagecache_write_end(file, mapping, curpos, len, len,
2435	page, fsdata);
2436	if (err < 0)
2437	goto out;
2438	BUG_ON(err != len);
2439	err = 0;
2440	}
2441	out:
2442	return err;
2443	}
2444
2445	/*
2446	* For moronic filesystems that do not allow holes in file.
2447	* We may have to extend the file.
2448	*/
2449	int cont_write_begin(struct file file, struct address_space mapping,
2450	loff_t pos, unsigned len, unsigned flags,
2451	struct page pagep, void fsdata,
2452	get_block_t get_block, loff_t bytes)
2453	{
2454	struct inode *inode = mapping->host;
2455	unsigned int blocksize = i_blocksize(inode);
2456	unsigned int zerofrom;
2457	int err;
2458
2459	err = cont_expand_zero(file, mapping, pos, bytes);
2460	if (err)
2461	return err;
2462
2463	zerofrom = *bytes & ~PAGE_MASK;
2464	if (pos+len > *bytes && zerofrom & (blocksize-1)) {
2465	*bytes \|= (blocksize-1);
2466	(*bytes)++;
2467	}
2468
2469	return block_write_begin(mapping, pos, len, flags, pagep, get_block);
2470	}
2471	EXPORT_SYMBOL(cont_write_begin);
2472
2473	int block_commit_write(struct page *page, unsigned from, unsigned to)
2474	{
2475	struct inode *inode = page->mapping->host;
2476	__block_commit_write(inode,page,from,to);
2477	return 0;
2478	}
2479	EXPORT_SYMBOL(block_commit_write);
2480
2481	/*
2482	* block_page_mkwrite() is not allowed to change the file size as it gets
2483	* called from a page fault handler when a page is first dirtied. Hence we must
2484	* be careful to check for EOF conditions here. We set the page up correctly
2485	* for a written page which means we get ENOSPC checking when writing into
2486	* holes and correct delalloc and unwritten extent mapping on filesystems that
2487	* support these features.
2488	*
2489	* We are not allowed to take the i_mutex here so we have to play games to
2490	* protect against truncate races as the page could now be beyond EOF. Because
2491	* truncate writes the inode size before removing pages, once we have the
2492	* page lock we can determine safely if the page is beyond EOF. If it is not
2493	* beyond EOF, then the page is guaranteed safe against truncation until we
2494	* unlock the page.
2495	*
2496	* Direct callers of this function should protect against filesystem freezing
2497	* using sb_start_pagefault() - sb_end_pagefault() functions.
2498	*/
2499	int block_page_mkwrite(struct vm_area_struct vma, struct vm_fault vmf,
2500	get_block_t get_block)
2501	{
2502	struct page *page = vmf->page;
2503	struct inode *inode = file_inode(vma->vm_file);
2504	unsigned long end;
2505	loff_t size;
2506	int ret;
2507
2508	lock_page(page);
2509	size = i_size_read(inode);
2510	if ((page->mapping != inode->i_mapping) \|\|
2511	(page_offset(page) > size)) {
2512	/* We overload EFAULT to mean page got truncated */
2513	ret = -EFAULT;
2514	goto out_unlock;
2515	}
2516
2517	/* page is wholly or partially inside EOF */
2518	if (((page->index + 1) << PAGE_SHIFT) > size)
2519	end = size & ~PAGE_MASK;
2520	else
2521	end = PAGE_SIZE;
2522
2523	ret = __block_write_begin(page, 0, end, get_block);
2524	if (!ret)
2525	ret = block_commit_write(page, 0, end);
2526
2527	if (unlikely(ret < 0))
2528	goto out_unlock;
2529	set_page_dirty(page);
2530	wait_for_stable_page(page);
2531	return 0;
2532	out_unlock:
2533	unlock_page(page);
2534	return ret;
2535	}
2536	EXPORT_SYMBOL(block_page_mkwrite);
2537
2538	/*
2539	* nobh_write_begin()'s prereads are special: the buffer_heads are freed
2540	* immediately, while under the page lock. So it needs a special end_io
2541	* handler which does not touch the bh after unlocking it.
2542	*/
2543	static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate)
2544	{
2545	__end_buffer_read_notouch(bh, uptodate);
2546	}
2547
2548	/*
2549	* Attach the singly-linked list of buffers created by nobh_write_begin, to
2550	* the page (converting it to circular linked list and taking care of page
2551	* dirty races).
2552	*/
2553	static void attach_nobh_buffers(struct page page, struct buffer_head head)
2554	{
2555	struct buffer_head *bh;
2556
2557	BUG_ON(!PageLocked(page));
2558
2559	spin_lock(&page->mapping->private_lock);
2560	bh = head;
2561	do {
2562	if (PageDirty(page))
2563	set_buffer_dirty(bh);
2564	if (!bh->b_this_page)
2565	bh->b_this_page = head;
2566	bh = bh->b_this_page;
2567	} while (bh != head);
2568	attach_page_private(page, head);
2569	spin_unlock(&page->mapping->private_lock);
2570	}
2571
2572	/*
2573	* On entry, the page is fully not uptodate.
2574	* On exit the page is fully uptodate in the areas outside (from,to)
2575	* The filesystem needs to handle block truncation upon failure.
2576	*/
2577	int nobh_write_begin(struct address_space *mapping,
2578	loff_t pos, unsigned len, unsigned flags,
2579	struct page pagep, void fsdata,
2580	get_block_t *get_block)
2581	{
2582	struct inode *inode = mapping->host;
2583	const unsigned blkbits = inode->i_blkbits;
2584	const unsigned blocksize = 1 << blkbits;
2585	struct buffer_head head, bh;
2586	struct page *page;
2587	pgoff_t index;
2588	unsigned from, to;
2589	unsigned block_in_page;
2590	unsigned block_start, block_end;
2591	sector_t block_in_file;
2592	int nr_reads = 0;
2593	int ret = 0;
2594	int is_mapped_to_disk = 1;
2595
2596	index = pos >> PAGE_SHIFT;
2597	from = pos & (PAGE_SIZE - 1);
2598	to = from + len;
2599
2600	page = grab_cache_page_write_begin(mapping, index, flags);
2601	if (!page)
2602	return -ENOMEM;
2603	*pagep = page;
2604	*fsdata = NULL;
2605
2606	if (page_has_buffers(page)) {
2607	ret = __block_write_begin(page, pos, len, get_block);
2608	if (unlikely(ret))
2609	goto out_release;
2610	return ret;
2611	}
2612
2613	if (PageMappedToDisk(page))
2614	return 0;
2615
2616	/*
2617	* Allocate buffers so that we can keep track of state, and potentially
2618	* attach them to the page if an error occurs. In the common case of
2619	* no error, they will just be freed again without ever being attached
2620	* to the page (which is all OK, because we're under the page lock).
2621	*
2622	* Be careful: the buffer linked list is a NULL terminated one, rather
2623	* than the circular one we're used to.
2624	*/
2625	head = alloc_page_buffers(page, blocksize, false);
2626	if (!head) {
2627	ret = -ENOMEM;
2628	goto out_release;
2629	}
2630
2631	block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
2632
2633	/*
2634	* We loop across all blocks in the page, whether or not they are
2635	* part of the affected region. This is so we can discover if the
2636	* page is fully mapped-to-disk.
2637	*/
2638	for (block_start = 0, block_in_page = 0, bh = head;
2639	block_start < PAGE_SIZE;
2640	block_in_page++, block_start += blocksize, bh = bh->b_this_page) {
2641	int create;
2642
2643	block_end = block_start + blocksize;
2644	bh->b_state = 0;
2645	create = 1;
2646	if (block_start >= to)
2647	create = 0;
2648	ret = get_block(inode, block_in_file + block_in_page,
2649	bh, create);
2650	if (ret)
2651	goto failed;
2652	if (!buffer_mapped(bh))
2653	is_mapped_to_disk = 0;
2654	if (buffer_new(bh))
2655	clean_bdev_bh_alias(bh);
2656	if (PageUptodate(page)) {
2657	set_buffer_uptodate(bh);
2658	continue;
2659	}
2660	if (buffer_new(bh) \|\| !buffer_mapped(bh)) {
2661	zero_user_segments(page, block_start, from,
2662	to, block_end);
2663	continue;
2664	}
2665	if (buffer_uptodate(bh))
2666	continue; /* reiserfs does this */
2667	if (block_start < from \|\| block_end > to) {
2668	lock_buffer(bh);
2669	bh->b_end_io = end_buffer_read_nobh;
2670	submit_bh(REQ_OP_READ, 0, bh);
2671	nr_reads++;
2672	}
2673	}
2674
2675	if (nr_reads) {
2676	/*
2677	* The page is locked, so these buffers are protected from
2678	* any VM or truncate activity. Hence we don't need to care
2679	* for the buffer_head refcounts.
2680	*/
2681	for (bh = head; bh; bh = bh->b_this_page) {
2682	wait_on_buffer(bh);
2683	if (!buffer_uptodate(bh))
2684	ret = -EIO;
2685	}
2686	if (ret)
2687	goto failed;
2688	}
2689
2690	if (is_mapped_to_disk)
2691	SetPageMappedToDisk(page);
2692
2693	fsdata = head; / to be released by nobh_write_end */
2694
2695	return 0;
2696
2697	failed:
2698	BUG_ON(!ret);
2699	/*
2700	* Error recovery is a bit difficult. We need to zero out blocks that
2701	* were newly allocated, and dirty them to ensure they get written out.
2702	* Buffers need to be attached to the page at this point, otherwise
2703	* the handling of potential IO errors during writeout would be hard
2704	* (could try doing synchronous writeout, but what if that fails too?)
2705	*/
2706	attach_nobh_buffers(page, head);
2707	page_zero_new_buffers(page, from, to);
2708
2709	out_release:
2710	unlock_page(page);
2711	put_page(page);
2712	*pagep = NULL;
2713
2714	return ret;
2715	}
2716	EXPORT_SYMBOL(nobh_write_begin);
2717
2718	int nobh_write_end(struct file file, struct address_space mapping,
2719	loff_t pos, unsigned len, unsigned copied,
2720	struct page page, void fsdata)
2721	{
2722	struct inode *inode = page->mapping->host;
2723	struct buffer_head *head = fsdata;
2724	struct buffer_head *bh;
2725	BUG_ON(fsdata != NULL && page_has_buffers(page));
2726
2727	if (unlikely(copied < len) && head)
2728	attach_nobh_buffers(page, head);
2729	if (page_has_buffers(page))
2730	return generic_write_end(file, mapping, pos, len,
2731	copied, page, fsdata);
2732
2733	SetPageUptodate(page);
2734	set_page_dirty(page);
2735	if (pos+copied > inode->i_size) {
2736	i_size_write(inode, pos+copied);
2737	mark_inode_dirty(inode);
2738	}
2739
2740	unlock_page(page);
2741	put_page(page);
2742
2743	while (head) {
2744	bh = head;
2745	head = head->b_this_page;
2746	free_buffer_head(bh);
2747	}
2748
2749	return copied;
2750	}
2751	EXPORT_SYMBOL(nobh_write_end);
2752
2753	/*
2754	* nobh_writepage() - based on block_full_write_page() except
2755	* that it tries to operate without attaching bufferheads to
2756	* the page.
2757	*/
2758	int nobh_writepage(struct page page, get_block_t get_block,
2759	struct writeback_control *wbc)
2760	{
2761	struct inode * const inode = page->mapping->host;
2762	loff_t i_size = i_size_read(inode);
2763	const pgoff_t end_index = i_size >> PAGE_SHIFT;
2764	unsigned offset;
2765	int ret;
2766
2767	/* Is the page fully inside i_size? */
2768	if (page->index < end_index)
2769	goto out;
2770
2771	/* Is the page fully outside i_size? (truncate in progress) */
2772	offset = i_size & (PAGE_SIZE-1);
2773	if (page->index >= end_index+1 \|\| !offset) {
2774	/*
2775	* The page may have dirty, unmapped buffers. For example,
2776	* they may have been added in ext3_writepage(). Make them
2777	* freeable here, so the page does not leak.
2778	*/
2779	#if 0
2780	/* Not really sure about this - do we need this ? */
2781	if (page->mapping->a_ops->invalidatepage)
2782	page->mapping->a_ops->invalidatepage(page, offset);
2783	#endif
2784	unlock_page(page);
2785	return 0; /* don't care */
2786	}
2787
2788	/*
2789	* The page straddles i_size. It must be zeroed out on each and every
2790	* writepage invocation because it may be mmapped. "A file is mapped
2791	* in multiples of the page size. For a file that is not a multiple of
2792	* the page size, the remaining memory is zeroed when mapped, and
2793	* writes to that region are not written out to the file."
2794	*/
2795	zero_user_segment(page, offset, PAGE_SIZE);
2796	out:
2797	ret = mpage_writepage(page, get_block, wbc);
2798	if (ret == -EAGAIN)
2799	ret = __block_write_full_page(inode, page, get_block, wbc,
2800	end_buffer_async_write);
2801	return ret;
2802	}
2803	EXPORT_SYMBOL(nobh_writepage);
2804
2805	int nobh_truncate_page(struct address_space *mapping,
2806	loff_t from, get_block_t *get_block)
2807	{
2808	pgoff_t index = from >> PAGE_SHIFT;
2809	unsigned offset = from & (PAGE_SIZE-1);
2810	unsigned blocksize;
2811	sector_t iblock;
2812	unsigned length, pos;
2813	struct inode *inode = mapping->host;
2814	struct page *page;
2815	struct buffer_head map_bh;
2816	int err;
2817
2818	blocksize = i_blocksize(inode);
2819	length = offset & (blocksize - 1);
2820
2821	/* Block boundary? Nothing to do */
2822	if (!length)
2823	return 0;
2824
2825	length = blocksize - length;
2826	iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
2827
2828	page = grab_cache_page(mapping, index);
2829	err = -ENOMEM;
2830	if (!page)
2831	goto out;
2832
2833	if (page_has_buffers(page)) {
2834	has_buffers:
2835	unlock_page(page);
2836	put_page(page);
2837	return block_truncate_page(mapping, from, get_block);
2838	}
2839
2840	/* Find the buffer that contains "offset" */
2841	pos = blocksize;
2842	while (offset >= pos) {
2843	iblock++;
2844	pos += blocksize;
2845	}
2846
2847	map_bh.b_size = blocksize;
2848	map_bh.b_state = 0;
2849	err = get_block(inode, iblock, &map_bh, 0);
2850	if (err)
2851	goto unlock;
2852	/* unmapped? It's a hole - nothing to do */
2853	if (!buffer_mapped(&map_bh))
2854	goto unlock;
2855
2856	/* Ok, it's mapped. Make sure it's up-to-date */
2857	if (!PageUptodate(page)) {
2858	err = mapping->a_ops->readpage(NULL, page);
2859	if (err) {
2860	put_page(page);
2861	goto out;
2862	}
2863	lock_page(page);
2864	if (!PageUptodate(page)) {
2865	err = -EIO;
2866	goto unlock;
2867	}
2868	if (page_has_buffers(page))
2869	goto has_buffers;
2870	}
2871	zero_user(page, offset, length);
2872	set_page_dirty(page);
2873	err = 0;
2874
2875	unlock:
2876	unlock_page(page);
2877	put_page(page);
2878	out:
2879	return err;
2880	}
2881	EXPORT_SYMBOL(nobh_truncate_page);
2882
2883	int block_truncate_page(struct address_space *mapping,
2884	loff_t from, get_block_t *get_block)
2885	{
2886	pgoff_t index = from >> PAGE_SHIFT;
2887	unsigned offset = from & (PAGE_SIZE-1);
2888	unsigned blocksize;
2889	sector_t iblock;
2890	unsigned length, pos;
2891	struct inode *inode = mapping->host;
2892	struct page *page;
2893	struct buffer_head *bh;
2894	int err;
2895
2896	blocksize = i_blocksize(inode);
2897	length = offset & (blocksize - 1);
2898
2899	/* Block boundary? Nothing to do */
2900	if (!length)
2901	return 0;
2902
2903	length = blocksize - length;
2904	iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
2905
2906	page = grab_cache_page(mapping, index);
2907	err = -ENOMEM;
2908	if (!page)
2909	goto out;
2910
2911	if (!page_has_buffers(page))
2912	create_empty_buffers(page, blocksize, 0);
2913
2914	/* Find the buffer that contains "offset" */
2915	bh = page_buffers(page);
2916	pos = blocksize;
2917	while (offset >= pos) {
2918	bh = bh->b_this_page;
2919	iblock++;
2920	pos += blocksize;
2921	}
2922
2923	err = 0;
2924	if (!buffer_mapped(bh)) {
2925	WARN_ON(bh->b_size != blocksize);
2926	err = get_block(inode, iblock, bh, 0);
2927	if (err)
2928	goto unlock;
2929	/* unmapped? It's a hole - nothing to do */
2930	if (!buffer_mapped(bh))
2931	goto unlock;
2932	}
2933
2934	/* Ok, it's mapped. Make sure it's up-to-date */
2935	if (PageUptodate(page))
2936	set_buffer_uptodate(bh);
2937
2938	if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
2939	err = -EIO;
2940	ll_rw_block(REQ_OP_READ, 0, 1, &bh);
2941	wait_on_buffer(bh);
2942	/* Uhhuh. Read error. Complain and punt. */
2943	if (!buffer_uptodate(bh))
2944	goto unlock;
2945	}
2946
2947	zero_user(page, offset, length);
2948	mark_buffer_dirty(bh);
2949	err = 0;
2950
2951	unlock:
2952	unlock_page(page);
2953	put_page(page);
2954	out:
2955	return err;
2956	}
2957	EXPORT_SYMBOL(block_truncate_page);
2958
2959	/*
2960	* The generic ->writepage function for buffer-backed address_spaces
2961	*/
2962	int block_write_full_page(struct page page, get_block_t get_block,
2963	struct writeback_control *wbc)
2964	{
2965	struct inode * const inode = page->mapping->host;
2966	loff_t i_size = i_size_read(inode);
2967	const pgoff_t end_index = i_size >> PAGE_SHIFT;
2968	unsigned offset;
2969
2970	/* Is the page fully inside i_size? */
2971	if (page->index < end_index)
2972	return __block_write_full_page(inode, page, get_block, wbc,
2973	end_buffer_async_write);
2974
2975	/* Is the page fully outside i_size? (truncate in progress) */
2976	offset = i_size & (PAGE_SIZE-1);
2977	if (page->index >= end_index+1 \|\| !offset) {
2978	/*
2979	* The page may have dirty, unmapped buffers. For example,
2980	* they may have been added in ext3_writepage(). Make them
2981	* freeable here, so the page does not leak.
2982	*/
2983	do_invalidatepage(page, 0, PAGE_SIZE);
2984	unlock_page(page);
2985	return 0; /* don't care */
2986	}
2987
2988	/*
2989	* The page straddles i_size. It must be zeroed out on each and every
2990	* writepage invocation because it may be mmapped. "A file is mapped
2991	* in multiples of the page size. For a file that is not a multiple of
2992	* the page size, the remaining memory is zeroed when mapped, and
2993	* writes to that region are not written out to the file."
2994	*/
2995	zero_user_segment(page, offset, PAGE_SIZE);
2996	return __block_write_full_page(inode, page, get_block, wbc,
2997	end_buffer_async_write);
2998	}
2999	EXPORT_SYMBOL(block_write_full_page);
3000
3001	sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
3002	get_block_t *get_block)
3003	{
3004	struct inode *inode = mapping->host;
3005	struct buffer_head tmp = {
3006	.b_size = i_blocksize(inode),
3007	};
3008
3009	get_block(inode, block, &tmp, 0);
3010	return tmp.b_blocknr;
3011	}
3012	EXPORT_SYMBOL(generic_block_bmap);
3013
3014	static void end_bio_bh_io_sync(struct bio *bio)
3015	{
3016	struct buffer_head *bh = bio->bi_private;
3017
3018	if (unlikely(bio_flagged(bio, BIO_QUIET)))
3019	set_bit(BH_Quiet, &bh->b_state);
3020
3021	bh->b_end_io(bh, !bio->bi_status);
3022	bio_put(bio);
3023	}
3024
3025	static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
3026	enum rw_hint write_hint, struct writeback_control *wbc)
3027	{
3028	struct bio *bio;
3029
3030	BUG_ON(!buffer_locked(bh));
3031	BUG_ON(!buffer_mapped(bh));
3032	BUG_ON(!bh->b_end_io);
3033	BUG_ON(buffer_delay(bh));
3034	BUG_ON(buffer_unwritten(bh));
3035
3036	/*
3037	* Only clear out a write error when rewriting
3038	*/
3039	if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE))
3040	clear_buffer_write_io_error(bh);
3041
3042	bio = bio_alloc(GFP_NOIO, 1);
3043
3044	fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO);
3045
3046	bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
3047	bio_set_dev(bio, bh->b_bdev);
3048	bio->bi_write_hint = write_hint;
3049
3050	bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
3051	BUG_ON(bio->bi_iter.bi_size != bh->b_size);
3052
3053	bio->bi_end_io = end_bio_bh_io_sync;
3054	bio->bi_private = bh;
3055
3056	if (buffer_meta(bh))
3057	op_flags \|= REQ_META;
3058	if (buffer_prio(bh))
3059	op_flags \|= REQ_PRIO;
3060	bio_set_op_attrs(bio, op, op_flags);
3061
3062	/* Take care of bh's that straddle the end of the device */
3063	guard_bio_eod(bio);
3064
3065	if (wbc) {
3066	wbc_init_bio(wbc, bio);
3067	wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size);
3068	}
3069
3070	submit_bio(bio);
3071	return 0;
3072	}
3073
3074	int submit_bh(int op, int op_flags, struct buffer_head *bh)
3075	{
3076	return submit_bh_wbc(op, op_flags, bh, 0, NULL);
3077	}
3078	EXPORT_SYMBOL(submit_bh);
3079
3080	/**
3081	* ll_rw_block: low-level access to block devices (DEPRECATED)
3082	* @op: whether to %READ or %WRITE
3083	* @op_flags: req_flag_bits
3084	* @nr: number of &struct buffer_heads in the array
3085	* @bhs: array of pointers to &struct buffer_head
3086	*
3087	* ll_rw_block() takes an array of pointers to &struct buffer_heads, and
3088	* requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE.
3089	* @op_flags contains flags modifying the detailed I/O behavior, most notably
3090	* %REQ_RAHEAD.
3091	*
3092	* This function drops any buffer that it cannot get a lock on (with the
3093	* BH_Lock state bit), any buffer that appears to be clean when doing a write
3094	* request, and any buffer that appears to be up-to-date when doing read
3095	* request. Further it marks as clean buffers that are processed for
3096	* writing (the buffer cache won't assume that they are actually clean
3097	* until the buffer gets unlocked).
3098	*
3099	* ll_rw_block sets b_end_io to simple completion handler that marks
3100	* the buffer up-to-date (if appropriate), unlocks the buffer and wakes
3101	* any waiters.
3102	*
3103	* All of the buffers must be for the same device, and must also be a
3104	* multiple of the current approved size for the device.
3105	*/
3106	void ll_rw_block(int op, int op_flags, int nr, struct buffer_head *bhs[])
3107	{
3108	int i;
3109
3110	for (i = 0; i < nr; i++) {
3111	struct buffer_head *bh = bhs[i];
3112
3113	if (!trylock_buffer(bh))
3114	continue;
3115	if (op == WRITE) {
3116	if (test_clear_buffer_dirty(bh)) {
3117	bh->b_end_io = end_buffer_write_sync;
3118	get_bh(bh);
3119	submit_bh(op, op_flags, bh);
3120	continue;
3121	}
3122	} else {
3123	if (!buffer_uptodate(bh)) {
3124	bh->b_end_io = end_buffer_read_sync;
3125	get_bh(bh);
3126	submit_bh(op, op_flags, bh);
3127	continue;
3128	}
3129	}
3130	unlock_buffer(bh);
3131	}
3132	}
3133	EXPORT_SYMBOL(ll_rw_block);
3134
3135	void write_dirty_buffer(struct buffer_head *bh, int op_flags)
3136	{
3137	lock_buffer(bh);
3138	if (!test_clear_buffer_dirty(bh)) {
3139	unlock_buffer(bh);
3140	return;
3141	}
3142	bh->b_end_io = end_buffer_write_sync;
3143	get_bh(bh);
3144	submit_bh(REQ_OP_WRITE, op_flags, bh);
3145	}
3146	EXPORT_SYMBOL(write_dirty_buffer);
3147
3148	/*
3149	* For a data-integrity writeout, we need to wait upon any in-progress I/O
3150	* and then start new I/O and then wait upon it. The caller must have a ref on
3151	* the buffer_head.
3152	*/
3153	int __sync_dirty_buffer(struct buffer_head *bh, int op_flags)
3154	{
3155	int ret = 0;
3156
3157	WARN_ON(atomic_read(&bh->b_count) < 1);
3158	lock_buffer(bh);
3159	if (test_clear_buffer_dirty(bh)) {
3160	get_bh(bh);
3161	bh->b_end_io = end_buffer_write_sync;
3162	ret = submit_bh(REQ_OP_WRITE, op_flags, bh);
3163	wait_on_buffer(bh);
3164	if (!ret && !buffer_uptodate(bh))
3165	ret = -EIO;
3166	} else {
3167	unlock_buffer(bh);
3168	}
3169	return ret;
3170	}
3171	EXPORT_SYMBOL(__sync_dirty_buffer);
3172
3173	int sync_dirty_buffer(struct buffer_head *bh)
3174	{
3175	return __sync_dirty_buffer(bh, REQ_SYNC);
3176	}
3177	EXPORT_SYMBOL(sync_dirty_buffer);
3178
3179	/*
3180	* try_to_free_buffers() checks if all the buffers on this particular page
3181	* are unused, and releases them if so.
3182	*
3183	* Exclusion against try_to_free_buffers may be obtained by either
3184	* locking the page or by holding its mapping's private_lock.
3185	*
3186	* If the page is dirty but all the buffers are clean then we need to
3187	* be sure to mark the page clean as well. This is because the page
3188	* may be against a block device, and a later reattachment of buffers
3189	* to a dirty page will set all buffers dirty. Which would corrupt
3190	* filesystem data on the same device.
3191	*
3192	* The same applies to regular filesystem pages: if all the buffers are
3193	* clean then we set the page clean and proceed. To do that, we require
3194	* total exclusion from __set_page_dirty_buffers(). That is obtained with
3195	* private_lock.
3196	*
3197	* try_to_free_buffers() is non-blocking.
3198	*/
3199	static inline int buffer_busy(struct buffer_head *bh)
3200	{
3201	return atomic_read(&bh->b_count) \|
3202	(bh->b_state & ((1 << BH_Dirty) \| (1 << BH_Lock)));
3203	}
3204
3205	static int
3206	drop_buffers(struct page page, struct buffer_head *buffers_to_free)
3207	{
3208	struct buffer_head *head = page_buffers(page);
3209	struct buffer_head *bh;
3210
3211	bh = head;
3212	do {
3213	if (buffer_busy(bh))
3214	goto failed;
3215	bh = bh->b_this_page;
3216	} while (bh != head);
3217
3218	do {
3219	struct buffer_head *next = bh->b_this_page;
3220
3221	if (bh->b_assoc_map)
3222	__remove_assoc_queue(bh);
3223	bh = next;
3224	} while (bh != head);
3225	*buffers_to_free = head;
3226	detach_page_private(page);
3227	return 1;
3228	failed:
3229	return 0;
3230	}
3231
3232	int try_to_free_buffers(struct page *page)
3233	{
3234	struct address_space * const mapping = page->mapping;
3235	struct buffer_head *buffers_to_free = NULL;
3236	int ret = 0;
3237
3238	BUG_ON(!PageLocked(page));
3239	if (PageWriteback(page))
3240	return 0;
3241
3242	if (mapping == NULL) { /* can this still happen? */
3243	ret = drop_buffers(page, &buffers_to_free);
3244	goto out;
3245	}
3246
3247	spin_lock(&mapping->private_lock);
3248	ret = drop_buffers(page, &buffers_to_free);
3249
3250	/*
3251	* If the filesystem writes its buffers by hand (eg ext3)
3252	* then we can have clean buffers against a dirty page. We
3253	* clean the page here; otherwise the VM will never notice
3254	* that the filesystem did any IO at all.
3255	*
3256	* Also, during truncate, discard_buffer will have marked all
3257	* the page's buffers clean. We discover that here and clean
3258	* the page also.
3259	*
3260	* private_lock must be held over this entire operation in order
3261	* to synchronise against __set_page_dirty_buffers and prevent the
3262	* dirty bit from being lost.
3263	*/
3264	if (ret)
3265	cancel_dirty_page(page);
3266	spin_unlock(&mapping->private_lock);
3267	out:
3268	if (buffers_to_free) {
3269	struct buffer_head *bh = buffers_to_free;
3270
3271	do {
3272	struct buffer_head *next = bh->b_this_page;
3273	free_buffer_head(bh);
3274	bh = next;
3275	} while (bh != buffers_to_free);
3276	}
3277	return ret;
3278	}
3279	EXPORT_SYMBOL(try_to_free_buffers);
3280
3281	/*
3282	* There are no bdflush tunables left. But distributions are
3283	* still running obsolete flush daemons, so we terminate them here.
3284	*
3285	* Use of bdflush() is deprecated and will be removed in a future kernel.
3286	* The `flush-X' kernel threads fully replace bdflush daemons and this call.
3287	*/
3288	SYSCALL_DEFINE2(bdflush, int, func, long, data)
3289	{
3290	static int msg_count;
3291
3292	if (!capable(CAP_SYS_ADMIN))
3293	return -EPERM;
3294
3295	if (msg_count < 5) {
3296	msg_count++;
3297	printk(KERN_INFO
3298	"warning: process `%s' used the obsolete bdflush"
3299	" system call\n", current->comm);
3300	printk(KERN_INFO "Fix your initscripts?\n");
3301	}
3302
3303	if (func == 1)
3304	do_exit(0);
3305	return 0;
3306	}
3307
3308	/*
3309	* Buffer-head allocation
3310	*/
3311	static struct kmem_cache *bh_cachep __read_mostly;
3312
3313	/*
3314	* Once the number of bh's in the machine exceeds this level, we start
3315	* stripping them in writeback.
3316	*/
3317	static unsigned long max_buffer_heads;
3318
3319	int buffer_heads_over_limit;
3320
3321	struct bh_accounting {
3322	int nr; /* Number of live bh's */
3323	int ratelimit; /* Limit cacheline bouncing */
3324	};
3325
3326	static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0};
3327
3328	static void recalc_bh_state(void)
3329	{
3330	int i;
3331	int tot = 0;
3332
3333	if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096)
3334	return;
3335	__this_cpu_write(bh_accounting.ratelimit, 0);
3336	for_each_online_cpu(i)
3337	tot += per_cpu(bh_accounting, i).nr;
3338	buffer_heads_over_limit = (tot > max_buffer_heads);
3339	}
3340
3341	struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
3342	{
3343	struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags);
3344	if (ret) {
3345	INIT_LIST_HEAD(&ret->b_assoc_buffers);
3346	spin_lock_init(&ret->b_uptodate_lock);
3347	preempt_disable();
3348	__this_cpu_inc(bh_accounting.nr);
3349	recalc_bh_state();
3350	preempt_enable();
3351	}
3352	return ret;
3353	}
3354	EXPORT_SYMBOL(alloc_buffer_head);
3355
3356	void free_buffer_head(struct buffer_head *bh)
3357	{
3358	BUG_ON(!list_empty(&bh->b_assoc_buffers));
3359	kmem_cache_free(bh_cachep, bh);
3360	preempt_disable();
3361	__this_cpu_dec(bh_accounting.nr);
3362	recalc_bh_state();
3363	preempt_enable();
3364	}
3365	EXPORT_SYMBOL(free_buffer_head);
3366
3367	static int buffer_exit_cpu_dead(unsigned int cpu)
3368	{
3369	int i;
3370	struct bh_lru *b = &per_cpu(bh_lrus, cpu);
3371
3372	for (i = 0; i < BH_LRU_SIZE; i++) {
3373	brelse(b->bhs[i]);
3374	b->bhs[i] = NULL;
3375	}
3376	this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr);
3377	per_cpu(bh_accounting, cpu).nr = 0;
3378	return 0;
3379	}
3380
3381	/**
3382	* bh_uptodate_or_lock - Test whether the buffer is uptodate
3383	* @bh: struct buffer_head
3384	*
3385	* Return true if the buffer is up-to-date and false,
3386	* with the buffer locked, if not.
3387	*/
3388	int bh_uptodate_or_lock(struct buffer_head *bh)
3389	{
3390	if (!buffer_uptodate(bh)) {
3391	lock_buffer(bh);
3392	if (!buffer_uptodate(bh))
3393	return 0;
3394	unlock_buffer(bh);
3395	}
3396	return 1;
3397	}
3398	EXPORT_SYMBOL(bh_uptodate_or_lock);
3399
3400	/**
3401	* bh_submit_read - Submit a locked buffer for reading
3402	* @bh: struct buffer_head
3403	*
3404	* Returns zero on success and -EIO on error.
3405	*/
3406	int bh_submit_read(struct buffer_head *bh)
3407	{
3408	BUG_ON(!buffer_locked(bh));
3409
3410	if (buffer_uptodate(bh)) {
3411	unlock_buffer(bh);
3412	return 0;
3413	}
3414
3415	get_bh(bh);
3416	bh->b_end_io = end_buffer_read_sync;
3417	submit_bh(REQ_OP_READ, 0, bh);
3418	wait_on_buffer(bh);
3419	if (buffer_uptodate(bh))
3420	return 0;
3421	return -EIO;
3422	}
3423	EXPORT_SYMBOL(bh_submit_read);
3424
3425	void __init buffer_init(void)
3426	{
3427	unsigned long nrpages;
3428	int ret;
3429
3430	bh_cachep = kmem_cache_create("buffer_head",
3431	sizeof(struct buffer_head), 0,
3432	(SLAB_RECLAIM_ACCOUNT\|SLAB_PANIC\|
3433	SLAB_MEM_SPREAD),
3434	NULL);
3435
3436	/*
3437	* Limit the bh occupancy to 10% of ZONE_NORMAL
3438	*/
3439	nrpages = (nr_free_buffer_pages() * 10) / 100;
3440	max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head));
3441	ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead",
3442	NULL, buffer_exit_cpu_dead);
3443	WARN_ON(ret < 0);
3444	}