Linux Audio

Check our new training course

Loading...
v6.8
  1// SPDX-License-Identifier: GPL-2.0+
  2/*
  3 * 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 *__nilfs_get_folio_block(struct folio *folio,
 29		unsigned long block, pgoff_t index, int blkbits,
 30		unsigned long b_state)
 31
 32{
 33	unsigned long first_block;
 34	struct buffer_head *bh = folio_buffers(folio);
 35
 36	if (!bh)
 37		bh = create_empty_buffers(folio, 1 << blkbits, b_state);
 38
 39	first_block = (unsigned long)index << (PAGE_SHIFT - blkbits);
 40	bh = get_nth_bh(bh, 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 folio *folio;
 55	struct buffer_head *bh;
 56
 57	folio = filemap_grab_folio(mapping, index);
 58	if (IS_ERR(folio))
 59		return NULL;
 60
 61	bh = __nilfs_get_folio_block(folio, blkoff, index, blkbits, b_state);
 62	if (unlikely(!bh)) {
 63		folio_unlock(folio);
 64		folio_put(folio);
 65		return NULL;
 66	}
 67	return bh;
 68}
 69
 70/**
 71 * nilfs_forget_buffer - discard dirty state
 
 72 * @bh: buffer head of the buffer to be discarded
 73 */
 74void nilfs_forget_buffer(struct buffer_head *bh)
 75{
 76	struct folio *folio = bh->b_folio;
 77	const unsigned long clear_bits =
 78		(BIT(BH_Uptodate) | BIT(BH_Dirty) | BIT(BH_Mapped) |
 79		 BIT(BH_Async_Write) | BIT(BH_NILFS_Volatile) |
 80		 BIT(BH_NILFS_Checked) | BIT(BH_NILFS_Redirected));
 81
 82	lock_buffer(bh);
 83	set_mask_bits(&bh->b_state, clear_bits, 0);
 84	if (nilfs_folio_buffers_clean(folio))
 85		__nilfs_clear_folio_dirty(folio);
 86
 87	bh->b_blocknr = -1;
 88	folio_clear_uptodate(folio);
 89	folio_clear_mappedtodisk(folio);
 90	unlock_buffer(bh);
 91	brelse(bh);
 92}
 93
 94/**
 95 * nilfs_copy_buffer -- copy buffer data and flags
 96 * @dbh: destination buffer
 97 * @sbh: source buffer
 98 */
 99void nilfs_copy_buffer(struct buffer_head *dbh, struct buffer_head *sbh)
100{
101	void *kaddr0, *kaddr1;
102	unsigned long bits;
103	struct page *spage = sbh->b_page, *dpage = dbh->b_page;
104	struct buffer_head *bh;
105
106	kaddr0 = kmap_atomic(spage);
107	kaddr1 = kmap_atomic(dpage);
108	memcpy(kaddr1 + bh_offset(dbh), kaddr0 + bh_offset(sbh), sbh->b_size);
109	kunmap_atomic(kaddr1);
110	kunmap_atomic(kaddr0);
111
112	dbh->b_state = sbh->b_state & NILFS_BUFFER_INHERENT_BITS;
113	dbh->b_blocknr = sbh->b_blocknr;
114	dbh->b_bdev = sbh->b_bdev;
115
116	bh = dbh;
117	bits = sbh->b_state & (BIT(BH_Uptodate) | BIT(BH_Mapped));
118	while ((bh = bh->b_this_page) != dbh) {
119		lock_buffer(bh);
120		bits &= bh->b_state;
121		unlock_buffer(bh);
122	}
123	if (bits & BIT(BH_Uptodate))
124		SetPageUptodate(dpage);
125	else
126		ClearPageUptodate(dpage);
127	if (bits & BIT(BH_Mapped))
128		SetPageMappedToDisk(dpage);
129	else
130		ClearPageMappedToDisk(dpage);
131}
132
133/**
134 * nilfs_folio_buffers_clean - Check if a folio has dirty buffers or not.
135 * @folio: Folio to be checked.
136 *
137 * nilfs_folio_buffers_clean() returns false if the folio has dirty buffers.
138 * Otherwise, it returns true.
139 */
140bool nilfs_folio_buffers_clean(struct folio *folio)
141{
142	struct buffer_head *bh, *head;
143
144	bh = head = folio_buffers(folio);
145	do {
146		if (buffer_dirty(bh))
147			return false;
148		bh = bh->b_this_page;
149	} while (bh != head);
150	return true;
151}
152
153void nilfs_folio_bug(struct folio *folio)
154{
155	struct buffer_head *bh, *head;
156	struct address_space *m;
157	unsigned long ino;
158
159	if (unlikely(!folio)) {
160		printk(KERN_CRIT "NILFS_FOLIO_BUG(NULL)\n");
161		return;
162	}
163
164	m = folio->mapping;
165	ino = m ? m->host->i_ino : 0;
166
167	printk(KERN_CRIT "NILFS_FOLIO_BUG(%p): cnt=%d index#=%llu flags=0x%lx "
168	       "mapping=%p ino=%lu\n",
169	       folio, folio_ref_count(folio),
170	       (unsigned long long)folio->index, folio->flags, m, ino);
171
172	head = folio_buffers(folio);
173	if (head) {
174		int i = 0;
175
176		bh = head;
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_folio -- copy the folio with buffers
189 * @dst: destination folio
190 * @src: source folio
191 * @copy_dirty: flag whether to copy dirty states on the folio's buffer heads.
192 *
193 * This function is for both data folios and btnode folios.  The dirty flag
194 * should be treated by caller.  The folio must not be under i/o.
195 * Both src and dst folio must be locked
196 */
197static void nilfs_copy_folio(struct folio *dst, struct folio *src,
198		bool copy_dirty)
199{
200	struct buffer_head *dbh, *dbufs, *sbh;
201	unsigned long mask = NILFS_BUFFER_INHERENT_BITS;
202
203	BUG_ON(folio_test_writeback(dst));
204
205	sbh = folio_buffers(src);
206	dbh = folio_buffers(dst);
207	if (!dbh)
208		dbh = create_empty_buffers(dst, sbh->b_size, 0);
209
210	if (copy_dirty)
211		mask |= BIT(BH_Dirty);
212
213	dbufs = dbh;
214	do {
215		lock_buffer(sbh);
216		lock_buffer(dbh);
217		dbh->b_state = sbh->b_state & mask;
218		dbh->b_blocknr = sbh->b_blocknr;
219		dbh->b_bdev = sbh->b_bdev;
220		sbh = sbh->b_this_page;
221		dbh = dbh->b_this_page;
222	} while (dbh != dbufs);
223
224	folio_copy(dst, src);
225
226	if (folio_test_uptodate(src) && !folio_test_uptodate(dst))
227		folio_mark_uptodate(dst);
228	else if (!folio_test_uptodate(src) && folio_test_uptodate(dst))
229		folio_clear_uptodate(dst);
230	if (folio_test_mappedtodisk(src) && !folio_test_mappedtodisk(dst))
231		folio_set_mappedtodisk(dst);
232	else if (!folio_test_mappedtodisk(src) && folio_test_mappedtodisk(dst))
233		folio_clear_mappedtodisk(dst);
234
235	do {
236		unlock_buffer(sbh);
237		unlock_buffer(dbh);
238		sbh = sbh->b_this_page;
239		dbh = dbh->b_this_page;
240	} while (dbh != dbufs);
241}
242
243int nilfs_copy_dirty_pages(struct address_space *dmap,
244			   struct address_space *smap)
245{
246	struct folio_batch fbatch;
247	unsigned int i;
248	pgoff_t index = 0;
249	int err = 0;
250
251	folio_batch_init(&fbatch);
252repeat:
253	if (!filemap_get_folios_tag(smap, &index, (pgoff_t)-1,
254				PAGECACHE_TAG_DIRTY, &fbatch))
255		return 0;
256
257	for (i = 0; i < folio_batch_count(&fbatch); i++) {
258		struct folio *folio = fbatch.folios[i], *dfolio;
259
260		folio_lock(folio);
261		if (unlikely(!folio_test_dirty(folio)))
262			NILFS_FOLIO_BUG(folio, "inconsistent dirty state");
263
264		dfolio = filemap_grab_folio(dmap, folio->index);
265		if (unlikely(IS_ERR(dfolio))) {
266			/* No empty page is added to the page cache */
267			folio_unlock(folio);
268			err = PTR_ERR(dfolio);
269			break;
270		}
271		if (unlikely(!folio_buffers(folio)))
272			NILFS_FOLIO_BUG(folio,
273				       "found empty page in dat page cache");
274
275		nilfs_copy_folio(dfolio, folio, true);
276		filemap_dirty_folio(folio_mapping(dfolio), dfolio);
277
278		folio_unlock(dfolio);
279		folio_put(dfolio);
280		folio_unlock(folio);
281	}
282	folio_batch_release(&fbatch);
283	cond_resched();
284
285	if (likely(!err))
286		goto repeat;
287	return err;
288}
289
290/**
291 * nilfs_copy_back_pages -- copy back pages to original cache from shadow cache
292 * @dmap: destination page cache
293 * @smap: source page cache
294 *
295 * No pages must be added to the cache during this process.
296 * This must be ensured by the caller.
297 */
298void nilfs_copy_back_pages(struct address_space *dmap,
299			   struct address_space *smap)
300{
301	struct folio_batch fbatch;
302	unsigned int i, n;
303	pgoff_t start = 0;
 
304
305	folio_batch_init(&fbatch);
306repeat:
307	n = filemap_get_folios(smap, &start, ~0UL, &fbatch);
308	if (!n)
309		return;
310
311	for (i = 0; i < folio_batch_count(&fbatch); i++) {
312		struct folio *folio = fbatch.folios[i], *dfolio;
313		pgoff_t index = folio->index;
314
315		folio_lock(folio);
316		dfolio = filemap_lock_folio(dmap, index);
317		if (!IS_ERR(dfolio)) {
318			/* overwrite existing folio in the destination cache */
319			WARN_ON(folio_test_dirty(dfolio));
320			nilfs_copy_folio(dfolio, folio, false);
321			folio_unlock(dfolio);
322			folio_put(dfolio);
323			/* Do we not need to remove folio from smap here? */
324		} else {
325			struct folio *f;
326
327			/* move the folio to the destination cache */
328			xa_lock_irq(&smap->i_pages);
329			f = __xa_erase(&smap->i_pages, index);
330			WARN_ON(folio != f);
 
331			smap->nrpages--;
332			xa_unlock_irq(&smap->i_pages);
333
334			xa_lock_irq(&dmap->i_pages);
335			f = __xa_store(&dmap->i_pages, index, folio, GFP_NOFS);
336			if (unlikely(f)) {
337				/* Probably -ENOMEM */
338				folio->mapping = NULL;
339				folio_put(folio);
340			} else {
341				folio->mapping = dmap;
342				dmap->nrpages++;
343				if (folio_test_dirty(folio))
344					__xa_set_mark(&dmap->i_pages, index,
345							PAGECACHE_TAG_DIRTY);
 
346			}
347			xa_unlock_irq(&dmap->i_pages);
348		}
349		folio_unlock(folio);
350	}
351	folio_batch_release(&fbatch);
352	cond_resched();
353
354	goto repeat;
355}
356
357/**
358 * nilfs_clear_dirty_pages - discard dirty pages in address space
359 * @mapping: address space with dirty pages for discarding
360 * @silent: suppress [true] or print [false] warning messages
361 */
362void nilfs_clear_dirty_pages(struct address_space *mapping, bool silent)
363{
364	struct folio_batch fbatch;
365	unsigned int i;
366	pgoff_t index = 0;
367
368	folio_batch_init(&fbatch);
369
370	while (filemap_get_folios_tag(mapping, &index, (pgoff_t)-1,
371				PAGECACHE_TAG_DIRTY, &fbatch)) {
372		for (i = 0; i < folio_batch_count(&fbatch); i++) {
373			struct folio *folio = fbatch.folios[i];
374
375			folio_lock(folio);
376
377			/*
378			 * This folio may have been removed from the address
379			 * space by truncation or invalidation when the lock
380			 * was acquired.  Skip processing in that case.
381			 */
382			if (likely(folio->mapping == mapping))
383				nilfs_clear_folio_dirty(folio, silent);
384
385			folio_unlock(folio);
386		}
387		folio_batch_release(&fbatch);
388		cond_resched();
389	}
390}
391
392/**
393 * nilfs_clear_folio_dirty - discard dirty folio
394 * @folio: dirty folio that will be discarded
395 * @silent: suppress [true] or print [false] warning messages
396 */
397void nilfs_clear_folio_dirty(struct folio *folio, bool silent)
398{
399	struct inode *inode = folio->mapping->host;
400	struct super_block *sb = inode->i_sb;
401	struct buffer_head *bh, *head;
402
403	BUG_ON(!folio_test_locked(folio));
404
405	if (!silent)
406		nilfs_warn(sb, "discard dirty page: offset=%lld, ino=%lu",
407			   folio_pos(folio), inode->i_ino);
 
408
409	folio_clear_uptodate(folio);
410	folio_clear_mappedtodisk(folio);
411
412	head = folio_buffers(folio);
413	if (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;
420		do {
421			lock_buffer(bh);
422			if (!silent)
423				nilfs_warn(sb,
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_folio_dirty(folio);
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
 
 
 
 
 
 
 
 
 
452/*
453 * NILFS2 needs clear_page_dirty() in the following two cases:
454 *
455 * 1) For B-tree node pages and data pages of DAT file, NILFS2 clears dirty
456 *    flag of pages when it copies back pages from shadow cache to the
457 *    original cache.
 
458 *
459 * 2) Some B-tree operations like insertion or deletion may dispose buffers
460 *    in dirty state, and this needs to cancel the dirty state of their pages.
461 */
462void __nilfs_clear_folio_dirty(struct folio *folio)
463{
464	struct address_space *mapping = folio->mapping;
465
466	if (mapping) {
467		xa_lock_irq(&mapping->i_pages);
468		if (folio_test_dirty(folio)) {
469			__xa_clear_mark(&mapping->i_pages, folio->index,
 
470					     PAGECACHE_TAG_DIRTY);
471			xa_unlock_irq(&mapping->i_pages);
472			folio_clear_dirty_for_io(folio);
473			return;
474		}
475		xa_unlock_irq(&mapping->i_pages);
476		return;
477	}
478	folio_clear_dirty(folio);
479}
480
481/**
482 * nilfs_find_uncommitted_extent - find extent of uncommitted data
483 * @inode: inode
484 * @start_blk: start block offset (in)
485 * @blkoff: start offset of the found extent (out)
486 *
487 * This function searches an extent of buffers marked "delayed" which
488 * starts from a block offset equal to or larger than @start_blk.  If
489 * such an extent was found, this will store the start offset in
490 * @blkoff and return its length in blocks.  Otherwise, zero is
491 * returned.
492 */
493unsigned long nilfs_find_uncommitted_extent(struct inode *inode,
494					    sector_t start_blk,
495					    sector_t *blkoff)
496{
497	unsigned int i, nr_folios;
498	pgoff_t index;
 
499	unsigned long length = 0;
500	struct folio_batch fbatch;
501	struct folio *folio;
 
502
503	if (inode->i_mapping->nrpages == 0)
504		return 0;
505
506	index = start_blk >> (PAGE_SHIFT - inode->i_blkbits);
 
507
508	folio_batch_init(&fbatch);
509
510repeat:
511	nr_folios = filemap_get_folios_contig(inode->i_mapping, &index, ULONG_MAX,
512			&fbatch);
513	if (nr_folios == 0)
514		return length;
515
 
 
 
 
516	i = 0;
517	do {
518		folio = fbatch.folios[i];
519
520		folio_lock(folio);
521		if (folio_buffers(folio)) {
522			struct buffer_head *bh, *head;
523			sector_t b;
524
525			b = folio->index << (PAGE_SHIFT - inode->i_blkbits);
526			bh = head = folio_buffers(folio);
527			do {
528				if (b < start_blk)
529					continue;
530				if (buffer_delay(bh)) {
531					if (length == 0)
532						*blkoff = b;
533					length++;
534				} else if (length > 0) {
535					goto out_locked;
536				}
537			} while (++b, bh = bh->b_this_page, bh != head);
538		} else {
539			if (length > 0)
540				goto out_locked;
 
 
541		}
542		folio_unlock(folio);
543
544	} while (++i < nr_folios);
545
546	folio_batch_release(&fbatch);
 
547	cond_resched();
548	goto repeat;
549
550out_locked:
551	folio_unlock(folio);
552	folio_batch_release(&fbatch);
 
553	return length;
554}
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}