Linux Audio

Check our new training course

Loading...
v5.4
  1// SPDX-License-Identifier: GPL-2.0+
  2/*
  3 * page.c - buffer/page management specific to NILFS
  4 *
  5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
  6 *
 
 
 
 
 
 
 
 
 
 
  7 * Written by Ryusuke Konishi and Seiji Kihara.
  8 */
  9
 10#include <linux/pagemap.h>
 11#include <linux/writeback.h>
 12#include <linux/swap.h>
 13#include <linux/bitops.h>
 14#include <linux/page-flags.h>
 15#include <linux/list.h>
 16#include <linux/highmem.h>
 17#include <linux/pagevec.h>
 18#include <linux/gfp.h>
 19#include "nilfs.h"
 20#include "page.h"
 21#include "mdt.h"
 22
 23
 24#define NILFS_BUFFER_INHERENT_BITS					\
 25	(BIT(BH_Uptodate) | BIT(BH_Mapped) | BIT(BH_NILFS_Node) |	\
 26	 BIT(BH_NILFS_Volatile) | BIT(BH_NILFS_Checked))
 27
 28static struct buffer_head *
 29__nilfs_get_page_block(struct page *page, unsigned long block, pgoff_t index,
 30		       int blkbits, unsigned long b_state)
 31
 32{
 33	unsigned long first_block;
 34	struct buffer_head *bh;
 35
 36	if (!page_has_buffers(page))
 37		create_empty_buffers(page, 1 << blkbits, b_state);
 38
 39	first_block = (unsigned long)index << (PAGE_SHIFT - blkbits);
 40	bh = nilfs_page_get_nth_block(page, block - first_block);
 41
 42	touch_buffer(bh);
 43	wait_on_buffer(bh);
 44	return bh;
 45}
 46
 47struct buffer_head *nilfs_grab_buffer(struct inode *inode,
 48				      struct address_space *mapping,
 49				      unsigned long blkoff,
 50				      unsigned long b_state)
 51{
 52	int blkbits = inode->i_blkbits;
 53	pgoff_t index = blkoff >> (PAGE_SHIFT - blkbits);
 54	struct page *page;
 55	struct buffer_head *bh;
 56
 57	page = grab_cache_page(mapping, index);
 58	if (unlikely(!page))
 59		return NULL;
 60
 61	bh = __nilfs_get_page_block(page, blkoff, index, blkbits, b_state);
 62	if (unlikely(!bh)) {
 63		unlock_page(page);
 64		put_page(page);
 65		return NULL;
 66	}
 67	return bh;
 68}
 69
 70/**
 71 * nilfs_forget_buffer - discard dirty state
 72 * @inode: owner inode of the buffer
 73 * @bh: buffer head of the buffer to be discarded
 74 */
 75void nilfs_forget_buffer(struct buffer_head *bh)
 76{
 77	struct page *page = bh->b_page;
 78	const unsigned long clear_bits =
 79		(BIT(BH_Uptodate) | BIT(BH_Dirty) | BIT(BH_Mapped) |
 80		 BIT(BH_Async_Write) | BIT(BH_NILFS_Volatile) |
 81		 BIT(BH_NILFS_Checked) | BIT(BH_NILFS_Redirected));
 82
 83	lock_buffer(bh);
 84	set_mask_bits(&bh->b_state, clear_bits, 0);
 85	if (nilfs_page_buffers_clean(page))
 86		__nilfs_clear_page_dirty(page);
 87
 88	bh->b_blocknr = -1;
 89	ClearPageUptodate(page);
 90	ClearPageMappedToDisk(page);
 91	unlock_buffer(bh);
 92	brelse(bh);
 93}
 94
 95/**
 96 * nilfs_copy_buffer -- copy buffer data and flags
 97 * @dbh: destination buffer
 98 * @sbh: source buffer
 99 */
100void nilfs_copy_buffer(struct buffer_head *dbh, struct buffer_head *sbh)
101{
102	void *kaddr0, *kaddr1;
103	unsigned long bits;
104	struct page *spage = sbh->b_page, *dpage = dbh->b_page;
105	struct buffer_head *bh;
106
107	kaddr0 = kmap_atomic(spage);
108	kaddr1 = kmap_atomic(dpage);
109	memcpy(kaddr1 + bh_offset(dbh), kaddr0 + bh_offset(sbh), sbh->b_size);
110	kunmap_atomic(kaddr1);
111	kunmap_atomic(kaddr0);
112
113	dbh->b_state = sbh->b_state & NILFS_BUFFER_INHERENT_BITS;
114	dbh->b_blocknr = sbh->b_blocknr;
115	dbh->b_bdev = sbh->b_bdev;
116
117	bh = dbh;
118	bits = sbh->b_state & (BIT(BH_Uptodate) | BIT(BH_Mapped));
119	while ((bh = bh->b_this_page) != dbh) {
120		lock_buffer(bh);
121		bits &= bh->b_state;
122		unlock_buffer(bh);
123	}
124	if (bits & BIT(BH_Uptodate))
125		SetPageUptodate(dpage);
126	else
127		ClearPageUptodate(dpage);
128	if (bits & BIT(BH_Mapped))
129		SetPageMappedToDisk(dpage);
130	else
131		ClearPageMappedToDisk(dpage);
132}
133
134/**
135 * nilfs_page_buffers_clean - check if a page has dirty buffers or not.
136 * @page: page to be checked
137 *
138 * nilfs_page_buffers_clean() returns zero if the page has dirty buffers.
139 * Otherwise, it returns non-zero value.
140 */
141int nilfs_page_buffers_clean(struct page *page)
142{
143	struct buffer_head *bh, *head;
144
145	bh = head = page_buffers(page);
146	do {
147		if (buffer_dirty(bh))
148			return 0;
149		bh = bh->b_this_page;
150	} while (bh != head);
151	return 1;
152}
153
154void nilfs_page_bug(struct page *page)
155{
156	struct address_space *m;
157	unsigned long ino;
158
159	if (unlikely(!page)) {
160		printk(KERN_CRIT "NILFS_PAGE_BUG(NULL)\n");
161		return;
162	}
163
164	m = page->mapping;
165	ino = m ? m->host->i_ino : 0;
166
167	printk(KERN_CRIT "NILFS_PAGE_BUG(%p): cnt=%d index#=%llu flags=0x%lx "
168	       "mapping=%p ino=%lu\n",
169	       page, page_ref_count(page),
170	       (unsigned long long)page->index, page->flags, m, ino);
171
172	if (page_has_buffers(page)) {
173		struct buffer_head *bh, *head;
174		int i = 0;
175
176		bh = head = page_buffers(page);
177		do {
178			printk(KERN_CRIT
179			       " BH[%d] %p: cnt=%d block#=%llu state=0x%lx\n",
180			       i++, bh, atomic_read(&bh->b_count),
181			       (unsigned long long)bh->b_blocknr, bh->b_state);
182			bh = bh->b_this_page;
183		} while (bh != head);
184	}
185}
186
187/**
188 * nilfs_copy_page -- copy the page with buffers
189 * @dst: destination page
190 * @src: source page
191 * @copy_dirty: flag whether to copy dirty states on the page's buffer heads.
192 *
193 * This function is for both data pages and btnode pages.  The dirty flag
194 * should be treated by caller.  The page must not be under i/o.
195 * Both src and dst page must be locked
196 */
197static void nilfs_copy_page(struct page *dst, struct page *src, int copy_dirty)
198{
199	struct buffer_head *dbh, *dbufs, *sbh, *sbufs;
200	unsigned long mask = NILFS_BUFFER_INHERENT_BITS;
201
202	BUG_ON(PageWriteback(dst));
203
204	sbh = sbufs = page_buffers(src);
205	if (!page_has_buffers(dst))
206		create_empty_buffers(dst, sbh->b_size, 0);
207
208	if (copy_dirty)
209		mask |= BIT(BH_Dirty);
210
211	dbh = dbufs = page_buffers(dst);
212	do {
213		lock_buffer(sbh);
214		lock_buffer(dbh);
215		dbh->b_state = sbh->b_state & mask;
216		dbh->b_blocknr = sbh->b_blocknr;
217		dbh->b_bdev = sbh->b_bdev;
218		sbh = sbh->b_this_page;
219		dbh = dbh->b_this_page;
220	} while (dbh != dbufs);
221
222	copy_highpage(dst, src);
223
224	if (PageUptodate(src) && !PageUptodate(dst))
225		SetPageUptodate(dst);
226	else if (!PageUptodate(src) && PageUptodate(dst))
227		ClearPageUptodate(dst);
228	if (PageMappedToDisk(src) && !PageMappedToDisk(dst))
229		SetPageMappedToDisk(dst);
230	else if (!PageMappedToDisk(src) && PageMappedToDisk(dst))
231		ClearPageMappedToDisk(dst);
232
233	do {
234		unlock_buffer(sbh);
235		unlock_buffer(dbh);
236		sbh = sbh->b_this_page;
237		dbh = dbh->b_this_page;
238	} while (dbh != dbufs);
239}
240
241int nilfs_copy_dirty_pages(struct address_space *dmap,
242			   struct address_space *smap)
243{
244	struct pagevec pvec;
245	unsigned int i;
246	pgoff_t index = 0;
247	int err = 0;
248
249	pagevec_init(&pvec);
250repeat:
251	if (!pagevec_lookup_tag(&pvec, smap, &index, PAGECACHE_TAG_DIRTY))
252		return 0;
253
254	for (i = 0; i < pagevec_count(&pvec); i++) {
255		struct page *page = pvec.pages[i], *dpage;
256
257		lock_page(page);
258		if (unlikely(!PageDirty(page)))
259			NILFS_PAGE_BUG(page, "inconsistent dirty state");
260
261		dpage = grab_cache_page(dmap, page->index);
262		if (unlikely(!dpage)) {
263			/* No empty page is added to the page cache */
264			err = -ENOMEM;
265			unlock_page(page);
266			break;
267		}
268		if (unlikely(!page_has_buffers(page)))
269			NILFS_PAGE_BUG(page,
270				       "found empty page in dat page cache");
271
272		nilfs_copy_page(dpage, page, 1);
273		__set_page_dirty_nobuffers(dpage);
274
275		unlock_page(dpage);
276		put_page(dpage);
277		unlock_page(page);
278	}
279	pagevec_release(&pvec);
280	cond_resched();
281
282	if (likely(!err))
283		goto repeat;
284	return err;
285}
286
287/**
288 * nilfs_copy_back_pages -- copy back pages to original cache from shadow cache
289 * @dmap: destination page cache
290 * @smap: source page cache
291 *
292 * No pages must be added to the cache during this process.
293 * This must be ensured by the caller.
294 */
295void nilfs_copy_back_pages(struct address_space *dmap,
296			   struct address_space *smap)
297{
298	struct pagevec pvec;
299	unsigned int i, n;
300	pgoff_t index = 0;
 
301
302	pagevec_init(&pvec);
303repeat:
304	n = pagevec_lookup(&pvec, smap, &index);
305	if (!n)
306		return;
307
308	for (i = 0; i < pagevec_count(&pvec); i++) {
309		struct page *page = pvec.pages[i], *dpage;
310		pgoff_t offset = page->index;
311
312		lock_page(page);
313		dpage = find_lock_page(dmap, offset);
314		if (dpage) {
315			/* overwrite existing page in the destination cache */
316			WARN_ON(PageDirty(dpage));
317			nilfs_copy_page(dpage, page, 0);
318			unlock_page(dpage);
319			put_page(dpage);
320			/* Do we not need to remove page from smap here? */
321		} else {
322			struct page *p;
323
324			/* move the page to the destination cache */
325			xa_lock_irq(&smap->i_pages);
326			p = __xa_erase(&smap->i_pages, offset);
327			WARN_ON(page != p);
 
328			smap->nrpages--;
329			xa_unlock_irq(&smap->i_pages);
330
331			xa_lock_irq(&dmap->i_pages);
332			p = __xa_store(&dmap->i_pages, offset, page, GFP_NOFS);
333			if (unlikely(p)) {
334				/* Probably -ENOMEM */
335				page->mapping = NULL;
336				put_page(page);
337			} else {
338				page->mapping = dmap;
339				dmap->nrpages++;
340				if (PageDirty(page))
341					__xa_set_mark(&dmap->i_pages, offset,
342							PAGECACHE_TAG_DIRTY);
 
343			}
344			xa_unlock_irq(&dmap->i_pages);
345		}
346		unlock_page(page);
347	}
348	pagevec_release(&pvec);
349	cond_resched();
350
351	goto repeat;
352}
353
354/**
355 * nilfs_clear_dirty_pages - discard dirty pages in address space
356 * @mapping: address space with dirty pages for discarding
357 * @silent: suppress [true] or print [false] warning messages
358 */
359void nilfs_clear_dirty_pages(struct address_space *mapping, bool silent)
360{
361	struct pagevec pvec;
362	unsigned int i;
363	pgoff_t index = 0;
364
365	pagevec_init(&pvec);
366
367	while (pagevec_lookup_tag(&pvec, mapping, &index,
368					PAGECACHE_TAG_DIRTY)) {
369		for (i = 0; i < pagevec_count(&pvec); i++) {
370			struct page *page = pvec.pages[i];
371
372			lock_page(page);
373			nilfs_clear_dirty_page(page, silent);
374			unlock_page(page);
375		}
376		pagevec_release(&pvec);
377		cond_resched();
378	}
379}
380
381/**
382 * nilfs_clear_dirty_page - discard dirty page
383 * @page: dirty page that will be discarded
384 * @silent: suppress [true] or print [false] warning messages
385 */
386void nilfs_clear_dirty_page(struct page *page, bool silent)
387{
388	struct inode *inode = page->mapping->host;
389	struct super_block *sb = inode->i_sb;
390
391	BUG_ON(!PageLocked(page));
392
393	if (!silent)
394		nilfs_msg(sb, KERN_WARNING,
395			  "discard dirty page: offset=%lld, ino=%lu",
396			  page_offset(page), inode->i_ino);
397
398	ClearPageUptodate(page);
399	ClearPageMappedToDisk(page);
400
401	if (page_has_buffers(page)) {
402		struct buffer_head *bh, *head;
403		const unsigned long clear_bits =
404			(BIT(BH_Uptodate) | BIT(BH_Dirty) | BIT(BH_Mapped) |
405			 BIT(BH_Async_Write) | BIT(BH_NILFS_Volatile) |
406			 BIT(BH_NILFS_Checked) | BIT(BH_NILFS_Redirected));
407
408		bh = head = page_buffers(page);
409		do {
410			lock_buffer(bh);
411			if (!silent)
412				nilfs_msg(sb, KERN_WARNING,
413					  "discard dirty block: blocknr=%llu, size=%zu",
414					  (u64)bh->b_blocknr, bh->b_size);
415
416			set_mask_bits(&bh->b_state, clear_bits, 0);
417			unlock_buffer(bh);
418		} while (bh = bh->b_this_page, bh != head);
419	}
420
421	__nilfs_clear_page_dirty(page);
422}
423
424unsigned int nilfs_page_count_clean_buffers(struct page *page,
425					    unsigned int from, unsigned int to)
426{
427	unsigned int block_start, block_end;
428	struct buffer_head *bh, *head;
429	unsigned int nc = 0;
430
431	for (bh = head = page_buffers(page), block_start = 0;
432	     bh != head || !block_start;
433	     block_start = block_end, bh = bh->b_this_page) {
434		block_end = block_start + bh->b_size;
435		if (block_end > from && block_start < to && !buffer_dirty(bh))
436			nc++;
437	}
438	return nc;
439}
440
441void nilfs_mapping_init(struct address_space *mapping, struct inode *inode)
442{
443	mapping->host = inode;
444	mapping->flags = 0;
445	mapping_set_gfp_mask(mapping, GFP_NOFS);
446	mapping->private_data = NULL;
447	mapping->a_ops = &empty_aops;
448}
449
450/*
451 * NILFS2 needs clear_page_dirty() in the following two cases:
452 *
453 * 1) For B-tree node pages and data pages of the dat/gcdat, NILFS2 clears
454 *    page dirty flags when it copies back pages from the shadow cache
455 *    (gcdat->{i_mapping,i_btnode_cache}) to its original cache
456 *    (dat->{i_mapping,i_btnode_cache}).
457 *
458 * 2) Some B-tree operations like insertion or deletion may dispose buffers
459 *    in dirty state, and this needs to cancel the dirty state of their pages.
460 */
461int __nilfs_clear_page_dirty(struct page *page)
462{
463	struct address_space *mapping = page->mapping;
464
465	if (mapping) {
466		xa_lock_irq(&mapping->i_pages);
467		if (test_bit(PG_dirty, &page->flags)) {
468			__xa_clear_mark(&mapping->i_pages, page_index(page),
 
469					     PAGECACHE_TAG_DIRTY);
470			xa_unlock_irq(&mapping->i_pages);
471			return clear_page_dirty_for_io(page);
472		}
473		xa_unlock_irq(&mapping->i_pages);
474		return 0;
475	}
476	return TestClearPageDirty(page);
477}
478
479/**
480 * nilfs_find_uncommitted_extent - find extent of uncommitted data
481 * @inode: inode
482 * @start_blk: start block offset (in)
483 * @blkoff: start offset of the found extent (out)
484 *
485 * This function searches an extent of buffers marked "delayed" which
486 * starts from a block offset equal to or larger than @start_blk.  If
487 * such an extent was found, this will store the start offset in
488 * @blkoff and return its length in blocks.  Otherwise, zero is
489 * returned.
490 */
491unsigned long nilfs_find_uncommitted_extent(struct inode *inode,
492					    sector_t start_blk,
493					    sector_t *blkoff)
494{
495	unsigned int i;
496	pgoff_t index;
497	unsigned int nblocks_in_page;
498	unsigned long length = 0;
499	sector_t b;
500	struct pagevec pvec;
501	struct page *page;
502
503	if (inode->i_mapping->nrpages == 0)
504		return 0;
505
506	index = start_blk >> (PAGE_SHIFT - inode->i_blkbits);
507	nblocks_in_page = 1U << (PAGE_SHIFT - inode->i_blkbits);
508
509	pagevec_init(&pvec);
510
511repeat:
512	pvec.nr = find_get_pages_contig(inode->i_mapping, index, PAGEVEC_SIZE,
513					pvec.pages);
514	if (pvec.nr == 0)
515		return length;
516
517	if (length > 0 && pvec.pages[0]->index > index)
518		goto out;
519
520	b = pvec.pages[0]->index << (PAGE_SHIFT - inode->i_blkbits);
521	i = 0;
522	do {
523		page = pvec.pages[i];
524
525		lock_page(page);
526		if (page_has_buffers(page)) {
527			struct buffer_head *bh, *head;
528
529			bh = head = page_buffers(page);
530			do {
531				if (b < start_blk)
532					continue;
533				if (buffer_delay(bh)) {
534					if (length == 0)
535						*blkoff = b;
536					length++;
537				} else if (length > 0) {
538					goto out_locked;
539				}
540			} while (++b, bh = bh->b_this_page, bh != head);
541		} else {
542			if (length > 0)
543				goto out_locked;
544
545			b += nblocks_in_page;
546		}
547		unlock_page(page);
548
549	} while (++i < pagevec_count(&pvec));
550
551	index = page->index + 1;
552	pagevec_release(&pvec);
553	cond_resched();
554	goto repeat;
555
556out_locked:
557	unlock_page(page);
558out:
559	pagevec_release(&pvec);
560	return length;
561}
v4.17
 
  1/*
  2 * page.c - buffer/page management specific to NILFS
  3 *
  4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
  5 *
  6 * This program is free software; you can redistribute it and/or modify
  7 * it under the terms of the GNU General Public License as published by
  8 * the Free Software Foundation; either version 2 of the License, or
  9 * (at your option) any later version.
 10 *
 11 * This program is distributed in the hope that it will be useful,
 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 14 * GNU General Public License for more details.
 15 *
 16 * Written by Ryusuke Konishi and Seiji Kihara.
 17 */
 18
 19#include <linux/pagemap.h>
 20#include <linux/writeback.h>
 21#include <linux/swap.h>
 22#include <linux/bitops.h>
 23#include <linux/page-flags.h>
 24#include <linux/list.h>
 25#include <linux/highmem.h>
 26#include <linux/pagevec.h>
 27#include <linux/gfp.h>
 28#include "nilfs.h"
 29#include "page.h"
 30#include "mdt.h"
 31
 32
 33#define NILFS_BUFFER_INHERENT_BITS					\
 34	(BIT(BH_Uptodate) | BIT(BH_Mapped) | BIT(BH_NILFS_Node) |	\
 35	 BIT(BH_NILFS_Volatile) | BIT(BH_NILFS_Checked))
 36
 37static struct buffer_head *
 38__nilfs_get_page_block(struct page *page, unsigned long block, pgoff_t index,
 39		       int blkbits, unsigned long b_state)
 40
 41{
 42	unsigned long first_block;
 43	struct buffer_head *bh;
 44
 45	if (!page_has_buffers(page))
 46		create_empty_buffers(page, 1 << blkbits, b_state);
 47
 48	first_block = (unsigned long)index << (PAGE_SHIFT - blkbits);
 49	bh = nilfs_page_get_nth_block(page, block - first_block);
 50
 51	touch_buffer(bh);
 52	wait_on_buffer(bh);
 53	return bh;
 54}
 55
 56struct buffer_head *nilfs_grab_buffer(struct inode *inode,
 57				      struct address_space *mapping,
 58				      unsigned long blkoff,
 59				      unsigned long b_state)
 60{
 61	int blkbits = inode->i_blkbits;
 62	pgoff_t index = blkoff >> (PAGE_SHIFT - blkbits);
 63	struct page *page;
 64	struct buffer_head *bh;
 65
 66	page = grab_cache_page(mapping, index);
 67	if (unlikely(!page))
 68		return NULL;
 69
 70	bh = __nilfs_get_page_block(page, blkoff, index, blkbits, b_state);
 71	if (unlikely(!bh)) {
 72		unlock_page(page);
 73		put_page(page);
 74		return NULL;
 75	}
 76	return bh;
 77}
 78
 79/**
 80 * nilfs_forget_buffer - discard dirty state
 81 * @inode: owner inode of the buffer
 82 * @bh: buffer head of the buffer to be discarded
 83 */
 84void nilfs_forget_buffer(struct buffer_head *bh)
 85{
 86	struct page *page = bh->b_page;
 87	const unsigned long clear_bits =
 88		(BIT(BH_Uptodate) | BIT(BH_Dirty) | BIT(BH_Mapped) |
 89		 BIT(BH_Async_Write) | BIT(BH_NILFS_Volatile) |
 90		 BIT(BH_NILFS_Checked) | BIT(BH_NILFS_Redirected));
 91
 92	lock_buffer(bh);
 93	set_mask_bits(&bh->b_state, clear_bits, 0);
 94	if (nilfs_page_buffers_clean(page))
 95		__nilfs_clear_page_dirty(page);
 96
 97	bh->b_blocknr = -1;
 98	ClearPageUptodate(page);
 99	ClearPageMappedToDisk(page);
100	unlock_buffer(bh);
101	brelse(bh);
102}
103
104/**
105 * nilfs_copy_buffer -- copy buffer data and flags
106 * @dbh: destination buffer
107 * @sbh: source buffer
108 */
109void nilfs_copy_buffer(struct buffer_head *dbh, struct buffer_head *sbh)
110{
111	void *kaddr0, *kaddr1;
112	unsigned long bits;
113	struct page *spage = sbh->b_page, *dpage = dbh->b_page;
114	struct buffer_head *bh;
115
116	kaddr0 = kmap_atomic(spage);
117	kaddr1 = kmap_atomic(dpage);
118	memcpy(kaddr1 + bh_offset(dbh), kaddr0 + bh_offset(sbh), sbh->b_size);
119	kunmap_atomic(kaddr1);
120	kunmap_atomic(kaddr0);
121
122	dbh->b_state = sbh->b_state & NILFS_BUFFER_INHERENT_BITS;
123	dbh->b_blocknr = sbh->b_blocknr;
124	dbh->b_bdev = sbh->b_bdev;
125
126	bh = dbh;
127	bits = sbh->b_state & (BIT(BH_Uptodate) | BIT(BH_Mapped));
128	while ((bh = bh->b_this_page) != dbh) {
129		lock_buffer(bh);
130		bits &= bh->b_state;
131		unlock_buffer(bh);
132	}
133	if (bits & BIT(BH_Uptodate))
134		SetPageUptodate(dpage);
135	else
136		ClearPageUptodate(dpage);
137	if (bits & BIT(BH_Mapped))
138		SetPageMappedToDisk(dpage);
139	else
140		ClearPageMappedToDisk(dpage);
141}
142
143/**
144 * nilfs_page_buffers_clean - check if a page has dirty buffers or not.
145 * @page: page to be checked
146 *
147 * nilfs_page_buffers_clean() returns zero if the page has dirty buffers.
148 * Otherwise, it returns non-zero value.
149 */
150int nilfs_page_buffers_clean(struct page *page)
151{
152	struct buffer_head *bh, *head;
153
154	bh = head = page_buffers(page);
155	do {
156		if (buffer_dirty(bh))
157			return 0;
158		bh = bh->b_this_page;
159	} while (bh != head);
160	return 1;
161}
162
163void nilfs_page_bug(struct page *page)
164{
165	struct address_space *m;
166	unsigned long ino;
167
168	if (unlikely(!page)) {
169		printk(KERN_CRIT "NILFS_PAGE_BUG(NULL)\n");
170		return;
171	}
172
173	m = page->mapping;
174	ino = m ? m->host->i_ino : 0;
175
176	printk(KERN_CRIT "NILFS_PAGE_BUG(%p): cnt=%d index#=%llu flags=0x%lx "
177	       "mapping=%p ino=%lu\n",
178	       page, page_ref_count(page),
179	       (unsigned long long)page->index, page->flags, m, ino);
180
181	if (page_has_buffers(page)) {
182		struct buffer_head *bh, *head;
183		int i = 0;
184
185		bh = head = page_buffers(page);
186		do {
187			printk(KERN_CRIT
188			       " BH[%d] %p: cnt=%d block#=%llu state=0x%lx\n",
189			       i++, bh, atomic_read(&bh->b_count),
190			       (unsigned long long)bh->b_blocknr, bh->b_state);
191			bh = bh->b_this_page;
192		} while (bh != head);
193	}
194}
195
196/**
197 * nilfs_copy_page -- copy the page with buffers
198 * @dst: destination page
199 * @src: source page
200 * @copy_dirty: flag whether to copy dirty states on the page's buffer heads.
201 *
202 * This function is for both data pages and btnode pages.  The dirty flag
203 * should be treated by caller.  The page must not be under i/o.
204 * Both src and dst page must be locked
205 */
206static void nilfs_copy_page(struct page *dst, struct page *src, int copy_dirty)
207{
208	struct buffer_head *dbh, *dbufs, *sbh, *sbufs;
209	unsigned long mask = NILFS_BUFFER_INHERENT_BITS;
210
211	BUG_ON(PageWriteback(dst));
212
213	sbh = sbufs = page_buffers(src);
214	if (!page_has_buffers(dst))
215		create_empty_buffers(dst, sbh->b_size, 0);
216
217	if (copy_dirty)
218		mask |= BIT(BH_Dirty);
219
220	dbh = dbufs = page_buffers(dst);
221	do {
222		lock_buffer(sbh);
223		lock_buffer(dbh);
224		dbh->b_state = sbh->b_state & mask;
225		dbh->b_blocknr = sbh->b_blocknr;
226		dbh->b_bdev = sbh->b_bdev;
227		sbh = sbh->b_this_page;
228		dbh = dbh->b_this_page;
229	} while (dbh != dbufs);
230
231	copy_highpage(dst, src);
232
233	if (PageUptodate(src) && !PageUptodate(dst))
234		SetPageUptodate(dst);
235	else if (!PageUptodate(src) && PageUptodate(dst))
236		ClearPageUptodate(dst);
237	if (PageMappedToDisk(src) && !PageMappedToDisk(dst))
238		SetPageMappedToDisk(dst);
239	else if (!PageMappedToDisk(src) && PageMappedToDisk(dst))
240		ClearPageMappedToDisk(dst);
241
242	do {
243		unlock_buffer(sbh);
244		unlock_buffer(dbh);
245		sbh = sbh->b_this_page;
246		dbh = dbh->b_this_page;
247	} while (dbh != dbufs);
248}
249
250int nilfs_copy_dirty_pages(struct address_space *dmap,
251			   struct address_space *smap)
252{
253	struct pagevec pvec;
254	unsigned int i;
255	pgoff_t index = 0;
256	int err = 0;
257
258	pagevec_init(&pvec);
259repeat:
260	if (!pagevec_lookup_tag(&pvec, smap, &index, PAGECACHE_TAG_DIRTY))
261		return 0;
262
263	for (i = 0; i < pagevec_count(&pvec); i++) {
264		struct page *page = pvec.pages[i], *dpage;
265
266		lock_page(page);
267		if (unlikely(!PageDirty(page)))
268			NILFS_PAGE_BUG(page, "inconsistent dirty state");
269
270		dpage = grab_cache_page(dmap, page->index);
271		if (unlikely(!dpage)) {
272			/* No empty page is added to the page cache */
273			err = -ENOMEM;
274			unlock_page(page);
275			break;
276		}
277		if (unlikely(!page_has_buffers(page)))
278			NILFS_PAGE_BUG(page,
279				       "found empty page in dat page cache");
280
281		nilfs_copy_page(dpage, page, 1);
282		__set_page_dirty_nobuffers(dpage);
283
284		unlock_page(dpage);
285		put_page(dpage);
286		unlock_page(page);
287	}
288	pagevec_release(&pvec);
289	cond_resched();
290
291	if (likely(!err))
292		goto repeat;
293	return err;
294}
295
296/**
297 * nilfs_copy_back_pages -- copy back pages to original cache from shadow cache
298 * @dmap: destination page cache
299 * @smap: source page cache
300 *
301 * No pages must no be added to the cache during this process.
302 * This must be ensured by the caller.
303 */
304void nilfs_copy_back_pages(struct address_space *dmap,
305			   struct address_space *smap)
306{
307	struct pagevec pvec;
308	unsigned int i, n;
309	pgoff_t index = 0;
310	int err;
311
312	pagevec_init(&pvec);
313repeat:
314	n = pagevec_lookup(&pvec, smap, &index);
315	if (!n)
316		return;
317
318	for (i = 0; i < pagevec_count(&pvec); i++) {
319		struct page *page = pvec.pages[i], *dpage;
320		pgoff_t offset = page->index;
321
322		lock_page(page);
323		dpage = find_lock_page(dmap, offset);
324		if (dpage) {
325			/* override existing page on the destination cache */
326			WARN_ON(PageDirty(dpage));
327			nilfs_copy_page(dpage, page, 0);
328			unlock_page(dpage);
329			put_page(dpage);
 
330		} else {
331			struct page *page2;
332
333			/* move the page to the destination cache */
334			xa_lock_irq(&smap->i_pages);
335			page2 = radix_tree_delete(&smap->i_pages, offset);
336			WARN_ON(page2 != page);
337
338			smap->nrpages--;
339			xa_unlock_irq(&smap->i_pages);
340
341			xa_lock_irq(&dmap->i_pages);
342			err = radix_tree_insert(&dmap->i_pages, offset, page);
343			if (unlikely(err < 0)) {
344				WARN_ON(err == -EEXIST);
345				page->mapping = NULL;
346				put_page(page); /* for cache */
347			} else {
348				page->mapping = dmap;
349				dmap->nrpages++;
350				if (PageDirty(page))
351					radix_tree_tag_set(&dmap->i_pages,
352							   offset,
353							   PAGECACHE_TAG_DIRTY);
354			}
355			xa_unlock_irq(&dmap->i_pages);
356		}
357		unlock_page(page);
358	}
359	pagevec_release(&pvec);
360	cond_resched();
361
362	goto repeat;
363}
364
365/**
366 * nilfs_clear_dirty_pages - discard dirty pages in address space
367 * @mapping: address space with dirty pages for discarding
368 * @silent: suppress [true] or print [false] warning messages
369 */
370void nilfs_clear_dirty_pages(struct address_space *mapping, bool silent)
371{
372	struct pagevec pvec;
373	unsigned int i;
374	pgoff_t index = 0;
375
376	pagevec_init(&pvec);
377
378	while (pagevec_lookup_tag(&pvec, mapping, &index,
379					PAGECACHE_TAG_DIRTY)) {
380		for (i = 0; i < pagevec_count(&pvec); i++) {
381			struct page *page = pvec.pages[i];
382
383			lock_page(page);
384			nilfs_clear_dirty_page(page, silent);
385			unlock_page(page);
386		}
387		pagevec_release(&pvec);
388		cond_resched();
389	}
390}
391
392/**
393 * nilfs_clear_dirty_page - discard dirty page
394 * @page: dirty page that will be discarded
395 * @silent: suppress [true] or print [false] warning messages
396 */
397void nilfs_clear_dirty_page(struct page *page, bool silent)
398{
399	struct inode *inode = page->mapping->host;
400	struct super_block *sb = inode->i_sb;
401
402	BUG_ON(!PageLocked(page));
403
404	if (!silent)
405		nilfs_msg(sb, KERN_WARNING,
406			  "discard dirty page: offset=%lld, ino=%lu",
407			  page_offset(page), inode->i_ino);
408
409	ClearPageUptodate(page);
410	ClearPageMappedToDisk(page);
411
412	if (page_has_buffers(page)) {
413		struct buffer_head *bh, *head;
414		const unsigned long clear_bits =
415			(BIT(BH_Uptodate) | BIT(BH_Dirty) | BIT(BH_Mapped) |
416			 BIT(BH_Async_Write) | BIT(BH_NILFS_Volatile) |
417			 BIT(BH_NILFS_Checked) | BIT(BH_NILFS_Redirected));
418
419		bh = head = page_buffers(page);
420		do {
421			lock_buffer(bh);
422			if (!silent)
423				nilfs_msg(sb, KERN_WARNING,
424					  "discard dirty block: blocknr=%llu, size=%zu",
425					  (u64)bh->b_blocknr, bh->b_size);
426
427			set_mask_bits(&bh->b_state, clear_bits, 0);
428			unlock_buffer(bh);
429		} while (bh = bh->b_this_page, bh != head);
430	}
431
432	__nilfs_clear_page_dirty(page);
433}
434
435unsigned int nilfs_page_count_clean_buffers(struct page *page,
436					    unsigned int from, unsigned int to)
437{
438	unsigned int block_start, block_end;
439	struct buffer_head *bh, *head;
440	unsigned int nc = 0;
441
442	for (bh = head = page_buffers(page), block_start = 0;
443	     bh != head || !block_start;
444	     block_start = block_end, bh = bh->b_this_page) {
445		block_end = block_start + bh->b_size;
446		if (block_end > from && block_start < to && !buffer_dirty(bh))
447			nc++;
448	}
449	return nc;
450}
451
452void nilfs_mapping_init(struct address_space *mapping, struct inode *inode)
453{
454	mapping->host = inode;
455	mapping->flags = 0;
456	mapping_set_gfp_mask(mapping, GFP_NOFS);
457	mapping->private_data = NULL;
458	mapping->a_ops = &empty_aops;
459}
460
461/*
462 * NILFS2 needs clear_page_dirty() in the following two cases:
463 *
464 * 1) For B-tree node pages and data pages of the dat/gcdat, NILFS2 clears
465 *    page dirty flags when it copies back pages from the shadow cache
466 *    (gcdat->{i_mapping,i_btnode_cache}) to its original cache
467 *    (dat->{i_mapping,i_btnode_cache}).
468 *
469 * 2) Some B-tree operations like insertion or deletion may dispose buffers
470 *    in dirty state, and this needs to cancel the dirty state of their pages.
471 */
472int __nilfs_clear_page_dirty(struct page *page)
473{
474	struct address_space *mapping = page->mapping;
475
476	if (mapping) {
477		xa_lock_irq(&mapping->i_pages);
478		if (test_bit(PG_dirty, &page->flags)) {
479			radix_tree_tag_clear(&mapping->i_pages,
480					     page_index(page),
481					     PAGECACHE_TAG_DIRTY);
482			xa_unlock_irq(&mapping->i_pages);
483			return clear_page_dirty_for_io(page);
484		}
485		xa_unlock_irq(&mapping->i_pages);
486		return 0;
487	}
488	return TestClearPageDirty(page);
489}
490
491/**
492 * nilfs_find_uncommitted_extent - find extent of uncommitted data
493 * @inode: inode
494 * @start_blk: start block offset (in)
495 * @blkoff: start offset of the found extent (out)
496 *
497 * This function searches an extent of buffers marked "delayed" which
498 * starts from a block offset equal to or larger than @start_blk.  If
499 * such an extent was found, this will store the start offset in
500 * @blkoff and return its length in blocks.  Otherwise, zero is
501 * returned.
502 */
503unsigned long nilfs_find_uncommitted_extent(struct inode *inode,
504					    sector_t start_blk,
505					    sector_t *blkoff)
506{
507	unsigned int i;
508	pgoff_t index;
509	unsigned int nblocks_in_page;
510	unsigned long length = 0;
511	sector_t b;
512	struct pagevec pvec;
513	struct page *page;
514
515	if (inode->i_mapping->nrpages == 0)
516		return 0;
517
518	index = start_blk >> (PAGE_SHIFT - inode->i_blkbits);
519	nblocks_in_page = 1U << (PAGE_SHIFT - inode->i_blkbits);
520
521	pagevec_init(&pvec);
522
523repeat:
524	pvec.nr = find_get_pages_contig(inode->i_mapping, index, PAGEVEC_SIZE,
525					pvec.pages);
526	if (pvec.nr == 0)
527		return length;
528
529	if (length > 0 && pvec.pages[0]->index > index)
530		goto out;
531
532	b = pvec.pages[0]->index << (PAGE_SHIFT - inode->i_blkbits);
533	i = 0;
534	do {
535		page = pvec.pages[i];
536
537		lock_page(page);
538		if (page_has_buffers(page)) {
539			struct buffer_head *bh, *head;
540
541			bh = head = page_buffers(page);
542			do {
543				if (b < start_blk)
544					continue;
545				if (buffer_delay(bh)) {
546					if (length == 0)
547						*blkoff = b;
548					length++;
549				} else if (length > 0) {
550					goto out_locked;
551				}
552			} while (++b, bh = bh->b_this_page, bh != head);
553		} else {
554			if (length > 0)
555				goto out_locked;
556
557			b += nblocks_in_page;
558		}
559		unlock_page(page);
560
561	} while (++i < pagevec_count(&pvec));
562
563	index = page->index + 1;
564	pagevec_release(&pvec);
565	cond_resched();
566	goto repeat;
567
568out_locked:
569	unlock_page(page);
570out:
571	pagevec_release(&pvec);
572	return length;
573}