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