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}
v3.15
 
  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 * You should have received a copy of the GNU General Public License
 17 * along with this program; if not, write to the Free Software
 18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 19 *
 20 * Written by Ryusuke Konishi <ryusuke@osrg.net>,
 21 *            Seiji Kihara <kihara@osrg.net>.
 22 */
 23
 24#include <linux/pagemap.h>
 25#include <linux/writeback.h>
 26#include <linux/swap.h>
 27#include <linux/bitops.h>
 28#include <linux/page-flags.h>
 29#include <linux/list.h>
 30#include <linux/highmem.h>
 31#include <linux/pagevec.h>
 32#include <linux/gfp.h>
 33#include "nilfs.h"
 34#include "page.h"
 35#include "mdt.h"
 36
 37
 38#define NILFS_BUFFER_INHERENT_BITS  \
 39	((1UL << BH_Uptodate) | (1UL << BH_Mapped) | (1UL << BH_NILFS_Node) | \
 40	 (1UL << BH_NILFS_Volatile) | (1UL << BH_NILFS_Checked))
 41
 42static struct buffer_head *
 43__nilfs_get_page_block(struct page *page, unsigned long block, pgoff_t index,
 44		       int blkbits, unsigned long b_state)
 45
 46{
 47	unsigned long first_block;
 48	struct buffer_head *bh;
 49
 50	if (!page_has_buffers(page))
 51		create_empty_buffers(page, 1 << blkbits, b_state);
 52
 53	first_block = (unsigned long)index << (PAGE_CACHE_SHIFT - blkbits);
 54	bh = nilfs_page_get_nth_block(page, block - first_block);
 55
 56	touch_buffer(bh);
 57	wait_on_buffer(bh);
 58	return bh;
 59}
 60
 61struct buffer_head *nilfs_grab_buffer(struct inode *inode,
 62				      struct address_space *mapping,
 63				      unsigned long blkoff,
 64				      unsigned long b_state)
 65{
 66	int blkbits = inode->i_blkbits;
 67	pgoff_t index = blkoff >> (PAGE_CACHE_SHIFT - blkbits);
 68	struct page *page;
 69	struct buffer_head *bh;
 70
 71	page = grab_cache_page(mapping, index);
 72	if (unlikely(!page))
 73		return NULL;
 74
 75	bh = __nilfs_get_page_block(page, blkoff, index, blkbits, b_state);
 76	if (unlikely(!bh)) {
 77		unlock_page(page);
 78		page_cache_release(page);
 79		return NULL;
 80	}
 81	return bh;
 82}
 83
 84/**
 85 * nilfs_forget_buffer - discard dirty state
 86 * @inode: owner inode of the buffer
 87 * @bh: buffer head of the buffer to be discarded
 88 */
 89void nilfs_forget_buffer(struct buffer_head *bh)
 90{
 91	struct page *page = bh->b_page;
 
 
 
 
 92
 93	lock_buffer(bh);
 94	clear_buffer_nilfs_volatile(bh);
 95	clear_buffer_nilfs_checked(bh);
 96	clear_buffer_nilfs_redirected(bh);
 97	clear_buffer_async_write(bh);
 98	clear_buffer_dirty(bh);
 99	if (nilfs_page_buffers_clean(page))
100		__nilfs_clear_page_dirty(page);
101
102	clear_buffer_uptodate(bh);
103	clear_buffer_mapped(bh);
104	bh->b_blocknr = -1;
105	ClearPageUptodate(page);
106	ClearPageMappedToDisk(page);
107	unlock_buffer(bh);
108	brelse(bh);
109}
110
111/**
112 * nilfs_copy_buffer -- copy buffer data and flags
113 * @dbh: destination buffer
114 * @sbh: source buffer
115 */
116void nilfs_copy_buffer(struct buffer_head *dbh, struct buffer_head *sbh)
117{
118	void *kaddr0, *kaddr1;
119	unsigned long bits;
120	struct page *spage = sbh->b_page, *dpage = dbh->b_page;
121	struct buffer_head *bh;
122
123	kaddr0 = kmap_atomic(spage);
124	kaddr1 = kmap_atomic(dpage);
125	memcpy(kaddr1 + bh_offset(dbh), kaddr0 + bh_offset(sbh), sbh->b_size);
126	kunmap_atomic(kaddr1);
127	kunmap_atomic(kaddr0);
128
129	dbh->b_state = sbh->b_state & NILFS_BUFFER_INHERENT_BITS;
130	dbh->b_blocknr = sbh->b_blocknr;
131	dbh->b_bdev = sbh->b_bdev;
132
133	bh = dbh;
134	bits = sbh->b_state & ((1UL << BH_Uptodate) | (1UL << BH_Mapped));
135	while ((bh = bh->b_this_page) != dbh) {
136		lock_buffer(bh);
137		bits &= bh->b_state;
138		unlock_buffer(bh);
139	}
140	if (bits & (1UL << BH_Uptodate))
141		SetPageUptodate(dpage);
142	else
143		ClearPageUptodate(dpage);
144	if (bits & (1UL << BH_Mapped))
145		SetPageMappedToDisk(dpage);
146	else
147		ClearPageMappedToDisk(dpage);
148}
149
150/**
151 * nilfs_page_buffers_clean - check if a page has dirty buffers or not.
152 * @page: page to be checked
153 *
154 * nilfs_page_buffers_clean() returns zero if the page has dirty buffers.
155 * Otherwise, it returns non-zero value.
156 */
157int nilfs_page_buffers_clean(struct page *page)
158{
159	struct buffer_head *bh, *head;
160
161	bh = head = page_buffers(page);
162	do {
163		if (buffer_dirty(bh))
164			return 0;
165		bh = bh->b_this_page;
166	} while (bh != head);
167	return 1;
168}
169
170void nilfs_page_bug(struct page *page)
171{
172	struct address_space *m;
173	unsigned long ino;
174
175	if (unlikely(!page)) {
176		printk(KERN_CRIT "NILFS_PAGE_BUG(NULL)\n");
177		return;
178	}
179
180	m = page->mapping;
181	ino = m ? m->host->i_ino : 0;
182
183	printk(KERN_CRIT "NILFS_PAGE_BUG(%p): cnt=%d index#=%llu flags=0x%lx "
184	       "mapping=%p ino=%lu\n",
185	       page, atomic_read(&page->_count),
186	       (unsigned long long)page->index, page->flags, m, ino);
187
188	if (page_has_buffers(page)) {
189		struct buffer_head *bh, *head;
190		int i = 0;
191
192		bh = head = page_buffers(page);
193		do {
194			printk(KERN_CRIT
195			       " BH[%d] %p: cnt=%d block#=%llu state=0x%lx\n",
196			       i++, bh, atomic_read(&bh->b_count),
197			       (unsigned long long)bh->b_blocknr, bh->b_state);
198			bh = bh->b_this_page;
199		} while (bh != head);
200	}
201}
202
203/**
204 * nilfs_copy_page -- copy the page with buffers
205 * @dst: destination page
206 * @src: source page
207 * @copy_dirty: flag whether to copy dirty states on the page's buffer heads.
208 *
209 * This function is for both data pages and btnode pages.  The dirty flag
210 * should be treated by caller.  The page must not be under i/o.
211 * Both src and dst page must be locked
212 */
213static void nilfs_copy_page(struct page *dst, struct page *src, int copy_dirty)
214{
215	struct buffer_head *dbh, *dbufs, *sbh, *sbufs;
216	unsigned long mask = NILFS_BUFFER_INHERENT_BITS;
217
218	BUG_ON(PageWriteback(dst));
219
220	sbh = sbufs = page_buffers(src);
221	if (!page_has_buffers(dst))
222		create_empty_buffers(dst, sbh->b_size, 0);
223
224	if (copy_dirty)
225		mask |= (1UL << BH_Dirty);
226
227	dbh = dbufs = page_buffers(dst);
228	do {
229		lock_buffer(sbh);
230		lock_buffer(dbh);
231		dbh->b_state = sbh->b_state & mask;
232		dbh->b_blocknr = sbh->b_blocknr;
233		dbh->b_bdev = sbh->b_bdev;
234		sbh = sbh->b_this_page;
235		dbh = dbh->b_this_page;
236	} while (dbh != dbufs);
237
238	copy_highpage(dst, src);
239
240	if (PageUptodate(src) && !PageUptodate(dst))
241		SetPageUptodate(dst);
242	else if (!PageUptodate(src) && PageUptodate(dst))
243		ClearPageUptodate(dst);
244	if (PageMappedToDisk(src) && !PageMappedToDisk(dst))
245		SetPageMappedToDisk(dst);
246	else if (!PageMappedToDisk(src) && PageMappedToDisk(dst))
247		ClearPageMappedToDisk(dst);
248
249	do {
250		unlock_buffer(sbh);
251		unlock_buffer(dbh);
252		sbh = sbh->b_this_page;
253		dbh = dbh->b_this_page;
254	} while (dbh != dbufs);
255}
256
257int nilfs_copy_dirty_pages(struct address_space *dmap,
258			   struct address_space *smap)
259{
260	struct pagevec pvec;
261	unsigned int i;
262	pgoff_t index = 0;
263	int err = 0;
264
265	pagevec_init(&pvec, 0);
266repeat:
267	if (!pagevec_lookup_tag(&pvec, smap, &index, PAGECACHE_TAG_DIRTY,
268				PAGEVEC_SIZE))
269		return 0;
270
271	for (i = 0; i < pagevec_count(&pvec); i++) {
272		struct page *page = pvec.pages[i], *dpage;
273
274		lock_page(page);
275		if (unlikely(!PageDirty(page)))
276			NILFS_PAGE_BUG(page, "inconsistent dirty state");
277
278		dpage = grab_cache_page(dmap, page->index);
279		if (unlikely(!dpage)) {
280			/* No empty page is added to the page cache */
281			err = -ENOMEM;
282			unlock_page(page);
283			break;
284		}
285		if (unlikely(!page_has_buffers(page)))
286			NILFS_PAGE_BUG(page,
287				       "found empty page in dat page cache");
288
289		nilfs_copy_page(dpage, page, 1);
290		__set_page_dirty_nobuffers(dpage);
291
292		unlock_page(dpage);
293		page_cache_release(dpage);
294		unlock_page(page);
295	}
296	pagevec_release(&pvec);
297	cond_resched();
298
299	if (likely(!err))
300		goto repeat;
301	return err;
302}
303
304/**
305 * nilfs_copy_back_pages -- copy back pages to original cache from shadow cache
306 * @dmap: destination page cache
307 * @smap: source page cache
308 *
309 * No pages must no be added to the cache during this process.
310 * This must be ensured by the caller.
311 */
312void nilfs_copy_back_pages(struct address_space *dmap,
313			   struct address_space *smap)
314{
315	struct pagevec pvec;
316	unsigned int i, n;
317	pgoff_t index = 0;
318	int err;
319
320	pagevec_init(&pvec, 0);
321repeat:
322	n = pagevec_lookup(&pvec, smap, index, PAGEVEC_SIZE);
323	if (!n)
324		return;
325	index = pvec.pages[n - 1]->index + 1;
326
327	for (i = 0; i < pagevec_count(&pvec); i++) {
328		struct page *page = pvec.pages[i], *dpage;
329		pgoff_t offset = page->index;
330
331		lock_page(page);
332		dpage = find_lock_page(dmap, offset);
333		if (dpage) {
334			/* override existing page on the destination cache */
335			WARN_ON(PageDirty(dpage));
336			nilfs_copy_page(dpage, page, 0);
337			unlock_page(dpage);
338			page_cache_release(dpage);
 
339		} else {
340			struct page *page2;
341
342			/* move the page to the destination cache */
343			spin_lock_irq(&smap->tree_lock);
344			page2 = radix_tree_delete(&smap->page_tree, offset);
345			WARN_ON(page2 != page);
346
347			smap->nrpages--;
348			spin_unlock_irq(&smap->tree_lock);
349
350			spin_lock_irq(&dmap->tree_lock);
351			err = radix_tree_insert(&dmap->page_tree, offset, page);
352			if (unlikely(err < 0)) {
353				WARN_ON(err == -EEXIST);
354				page->mapping = NULL;
355				page_cache_release(page); /* for cache */
356			} else {
357				page->mapping = dmap;
358				dmap->nrpages++;
359				if (PageDirty(page))
360					radix_tree_tag_set(&dmap->page_tree,
361							   offset,
362							   PAGECACHE_TAG_DIRTY);
363			}
364			spin_unlock_irq(&dmap->tree_lock);
365		}
366		unlock_page(page);
367	}
368	pagevec_release(&pvec);
369	cond_resched();
370
371	goto repeat;
372}
373
374/**
375 * nilfs_clear_dirty_pages - discard dirty pages in address space
376 * @mapping: address space with dirty pages for discarding
377 * @silent: suppress [true] or print [false] warning messages
378 */
379void nilfs_clear_dirty_pages(struct address_space *mapping, bool silent)
380{
381	struct pagevec pvec;
382	unsigned int i;
383	pgoff_t index = 0;
384
385	pagevec_init(&pvec, 0);
386
387	while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
388				  PAGEVEC_SIZE)) {
389		for (i = 0; i < pagevec_count(&pvec); i++) {
390			struct page *page = pvec.pages[i];
391
392			lock_page(page);
393			nilfs_clear_dirty_page(page, silent);
394			unlock_page(page);
395		}
396		pagevec_release(&pvec);
397		cond_resched();
398	}
399}
400
401/**
402 * nilfs_clear_dirty_page - discard dirty page
403 * @page: dirty page that will be discarded
404 * @silent: suppress [true] or print [false] warning messages
405 */
406void nilfs_clear_dirty_page(struct page *page, bool silent)
407{
408	struct inode *inode = page->mapping->host;
409	struct super_block *sb = inode->i_sb;
410
411	BUG_ON(!PageLocked(page));
412
413	if (!silent) {
414		nilfs_warning(sb, __func__,
415				"discard page: offset %lld, ino %lu",
416				page_offset(page), inode->i_ino);
417	}
418
419	ClearPageUptodate(page);
420	ClearPageMappedToDisk(page);
421
422	if (page_has_buffers(page)) {
423		struct buffer_head *bh, *head;
 
 
 
 
424
425		bh = head = page_buffers(page);
426		do {
427			lock_buffer(bh);
428			if (!silent) {
429				nilfs_warning(sb, __func__,
430					"discard block %llu, size %zu",
431					(u64)bh->b_blocknr, bh->b_size);
432			}
433			clear_buffer_async_write(bh);
434			clear_buffer_dirty(bh);
435			clear_buffer_nilfs_volatile(bh);
436			clear_buffer_nilfs_checked(bh);
437			clear_buffer_nilfs_redirected(bh);
438			clear_buffer_uptodate(bh);
439			clear_buffer_mapped(bh);
440			unlock_buffer(bh);
441		} while (bh = bh->b_this_page, bh != head);
442	}
443
444	__nilfs_clear_page_dirty(page);
445}
446
447unsigned nilfs_page_count_clean_buffers(struct page *page,
448					unsigned from, unsigned to)
449{
450	unsigned block_start, block_end;
451	struct buffer_head *bh, *head;
452	unsigned nc = 0;
453
454	for (bh = head = page_buffers(page), block_start = 0;
455	     bh != head || !block_start;
456	     block_start = block_end, bh = bh->b_this_page) {
457		block_end = block_start + bh->b_size;
458		if (block_end > from && block_start < to && !buffer_dirty(bh))
459			nc++;
460	}
461	return nc;
462}
463
464void nilfs_mapping_init(struct address_space *mapping, struct inode *inode,
465			struct backing_dev_info *bdi)
466{
467	mapping->host = inode;
468	mapping->flags = 0;
469	mapping_set_gfp_mask(mapping, GFP_NOFS);
470	mapping->private_data = NULL;
471	mapping->backing_dev_info = bdi;
472	mapping->a_ops = &empty_aops;
473}
474
475/*
476 * NILFS2 needs clear_page_dirty() in the following two cases:
477 *
478 * 1) For B-tree node pages and data pages of the dat/gcdat, NILFS2 clears
479 *    page dirty flags when it copies back pages from the shadow cache
480 *    (gcdat->{i_mapping,i_btnode_cache}) to its original cache
481 *    (dat->{i_mapping,i_btnode_cache}).
482 *
483 * 2) Some B-tree operations like insertion or deletion may dispose buffers
484 *    in dirty state, and this needs to cancel the dirty state of their pages.
485 */
486int __nilfs_clear_page_dirty(struct page *page)
487{
488	struct address_space *mapping = page->mapping;
489
490	if (mapping) {
491		spin_lock_irq(&mapping->tree_lock);
492		if (test_bit(PG_dirty, &page->flags)) {
493			radix_tree_tag_clear(&mapping->page_tree,
494					     page_index(page),
495					     PAGECACHE_TAG_DIRTY);
496			spin_unlock_irq(&mapping->tree_lock);
497			return clear_page_dirty_for_io(page);
498		}
499		spin_unlock_irq(&mapping->tree_lock);
500		return 0;
501	}
502	return TestClearPageDirty(page);
503}
504
505/**
506 * nilfs_find_uncommitted_extent - find extent of uncommitted data
507 * @inode: inode
508 * @start_blk: start block offset (in)
509 * @blkoff: start offset of the found extent (out)
510 *
511 * This function searches an extent of buffers marked "delayed" which
512 * starts from a block offset equal to or larger than @start_blk.  If
513 * such an extent was found, this will store the start offset in
514 * @blkoff and return its length in blocks.  Otherwise, zero is
515 * returned.
516 */
517unsigned long nilfs_find_uncommitted_extent(struct inode *inode,
518					    sector_t start_blk,
519					    sector_t *blkoff)
520{
521	unsigned int i;
522	pgoff_t index;
523	unsigned int nblocks_in_page;
524	unsigned long length = 0;
525	sector_t b;
526	struct pagevec pvec;
527	struct page *page;
528
529	if (inode->i_mapping->nrpages == 0)
530		return 0;
531
532	index = start_blk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
533	nblocks_in_page = 1U << (PAGE_CACHE_SHIFT - inode->i_blkbits);
534
535	pagevec_init(&pvec, 0);
536
537repeat:
538	pvec.nr = find_get_pages_contig(inode->i_mapping, index, PAGEVEC_SIZE,
539					pvec.pages);
540	if (pvec.nr == 0)
541		return length;
542
543	if (length > 0 && pvec.pages[0]->index > index)
544		goto out;
545
546	b = pvec.pages[0]->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
547	i = 0;
548	do {
549		page = pvec.pages[i];
550
551		lock_page(page);
552		if (page_has_buffers(page)) {
553			struct buffer_head *bh, *head;
554
555			bh = head = page_buffers(page);
556			do {
557				if (b < start_blk)
558					continue;
559				if (buffer_delay(bh)) {
560					if (length == 0)
561						*blkoff = b;
562					length++;
563				} else if (length > 0) {
564					goto out_locked;
565				}
566			} while (++b, bh = bh->b_this_page, bh != head);
567		} else {
568			if (length > 0)
569				goto out_locked;
570
571			b += nblocks_in_page;
572		}
573		unlock_page(page);
574
575	} while (++i < pagevec_count(&pvec));
576
577	index = page->index + 1;
578	pagevec_release(&pvec);
579	cond_resched();
580	goto repeat;
581
582out_locked:
583	unlock_page(page);
584out:
585	pagevec_release(&pvec);
586	return length;
587}