2 * page.c - buffer/page management specific to NILFS
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
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.
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.
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
20 * Written by Ryusuke Konishi <ryusuke@osrg.net>,
21 * Seiji Kihara <kihara@osrg.net>.
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>
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))
42 static struct buffer_head
*
43 __nilfs_get_page_block(struct page
*page
, unsigned long block
, pgoff_t index
,
44 int blkbits
, unsigned long b_state
)
47 unsigned long first_block
;
48 struct buffer_head
*bh
;
50 if (!page_has_buffers(page
))
51 create_empty_buffers(page
, 1 << blkbits
, b_state
);
53 first_block
= (unsigned long)index
<< (PAGE_CACHE_SHIFT
- blkbits
);
54 bh
= nilfs_page_get_nth_block(page
, block
- first_block
);
61 struct buffer_head
*nilfs_grab_buffer(struct inode
*inode
,
62 struct address_space
*mapping
,
64 unsigned long b_state
)
66 int blkbits
= inode
->i_blkbits
;
67 pgoff_t index
= blkoff
>> (PAGE_CACHE_SHIFT
- blkbits
);
69 struct buffer_head
*bh
;
71 page
= grab_cache_page(mapping
, index
);
75 bh
= __nilfs_get_page_block(page
, blkoff
, index
, blkbits
, b_state
);
78 page_cache_release(page
);
85 * nilfs_forget_buffer - discard dirty state
86 * @inode: owner inode of the buffer
87 * @bh: buffer head of the buffer to be discarded
89 void nilfs_forget_buffer(struct buffer_head
*bh
)
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
);
98 set_mask_bits(&bh
->b_state
, clear_bits
, 0);
99 if (nilfs_page_buffers_clean(page
))
100 __nilfs_clear_page_dirty(page
);
103 ClearPageUptodate(page
);
104 ClearPageMappedToDisk(page
);
110 * nilfs_copy_buffer -- copy buffer data and flags
111 * @dbh: destination buffer
112 * @sbh: source buffer
114 void nilfs_copy_buffer(struct buffer_head
*dbh
, struct buffer_head
*sbh
)
116 void *kaddr0
, *kaddr1
;
118 struct page
*spage
= sbh
->b_page
, *dpage
= dbh
->b_page
;
119 struct buffer_head
*bh
;
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
);
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
;
132 bits
= sbh
->b_state
& ((1UL << BH_Uptodate
) | (1UL << BH_Mapped
));
133 while ((bh
= bh
->b_this_page
) != dbh
) {
138 if (bits
& (1UL << BH_Uptodate
))
139 SetPageUptodate(dpage
);
141 ClearPageUptodate(dpage
);
142 if (bits
& (1UL << BH_Mapped
))
143 SetPageMappedToDisk(dpage
);
145 ClearPageMappedToDisk(dpage
);
149 * nilfs_page_buffers_clean - check if a page has dirty buffers or not.
150 * @page: page to be checked
152 * nilfs_page_buffers_clean() returns zero if the page has dirty buffers.
153 * Otherwise, it returns non-zero value.
155 int nilfs_page_buffers_clean(struct page
*page
)
157 struct buffer_head
*bh
, *head
;
159 bh
= head
= page_buffers(page
);
161 if (buffer_dirty(bh
))
163 bh
= bh
->b_this_page
;
164 } while (bh
!= head
);
168 void nilfs_page_bug(struct page
*page
)
170 struct address_space
*m
;
173 if (unlikely(!page
)) {
174 printk(KERN_CRIT
"NILFS_PAGE_BUG(NULL)\n");
179 ino
= m
? m
->host
->i_ino
: 0;
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
);
186 if (page_has_buffers(page
)) {
187 struct buffer_head
*bh
, *head
;
190 bh
= head
= page_buffers(page
);
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
);
202 * nilfs_copy_page -- copy the page with buffers
203 * @dst: destination page
205 * @copy_dirty: flag whether to copy dirty states on the page's buffer heads.
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
211 static void nilfs_copy_page(struct page
*dst
, struct page
*src
, int copy_dirty
)
213 struct buffer_head
*dbh
, *dbufs
, *sbh
, *sbufs
;
214 unsigned long mask
= NILFS_BUFFER_INHERENT_BITS
;
216 BUG_ON(PageWriteback(dst
));
218 sbh
= sbufs
= page_buffers(src
);
219 if (!page_has_buffers(dst
))
220 create_empty_buffers(dst
, sbh
->b_size
, 0);
223 mask
|= (1UL << BH_Dirty
);
225 dbh
= dbufs
= page_buffers(dst
);
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
);
236 copy_highpage(dst
, src
);
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
);
250 sbh
= sbh
->b_this_page
;
251 dbh
= dbh
->b_this_page
;
252 } while (dbh
!= dbufs
);
255 int nilfs_copy_dirty_pages(struct address_space
*dmap
,
256 struct address_space
*smap
)
263 pagevec_init(&pvec
, 0);
265 if (!pagevec_lookup_tag(&pvec
, smap
, &index
, PAGECACHE_TAG_DIRTY
,
269 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
270 struct page
*page
= pvec
.pages
[i
], *dpage
;
273 if (unlikely(!PageDirty(page
)))
274 NILFS_PAGE_BUG(page
, "inconsistent dirty state");
276 dpage
= grab_cache_page(dmap
, page
->index
);
277 if (unlikely(!dpage
)) {
278 /* No empty page is added to the page cache */
283 if (unlikely(!page_has_buffers(page
)))
285 "found empty page in dat page cache");
287 nilfs_copy_page(dpage
, page
, 1);
288 __set_page_dirty_nobuffers(dpage
);
291 page_cache_release(dpage
);
294 pagevec_release(&pvec
);
303 * nilfs_copy_back_pages -- copy back pages to original cache from shadow cache
304 * @dmap: destination page cache
305 * @smap: source page cache
307 * No pages must no be added to the cache during this process.
308 * This must be ensured by the caller.
310 void nilfs_copy_back_pages(struct address_space
*dmap
,
311 struct address_space
*smap
)
318 pagevec_init(&pvec
, 0);
320 n
= pagevec_lookup(&pvec
, smap
, index
, PAGEVEC_SIZE
);
323 index
= pvec
.pages
[n
- 1]->index
+ 1;
325 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
326 struct page
*page
= pvec
.pages
[i
], *dpage
;
327 pgoff_t offset
= page
->index
;
330 dpage
= find_lock_page(dmap
, offset
);
332 /* override existing page on the destination cache */
333 WARN_ON(PageDirty(dpage
));
334 nilfs_copy_page(dpage
, page
, 0);
336 page_cache_release(dpage
);
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
);
346 spin_unlock_irq(&smap
->tree_lock
);
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 */
355 page
->mapping
= dmap
;
358 radix_tree_tag_set(&dmap
->page_tree
,
360 PAGECACHE_TAG_DIRTY
);
362 spin_unlock_irq(&dmap
->tree_lock
);
366 pagevec_release(&pvec
);
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
377 void nilfs_clear_dirty_pages(struct address_space
*mapping
, bool silent
)
383 pagevec_init(&pvec
, 0);
385 while (pagevec_lookup_tag(&pvec
, mapping
, &index
, PAGECACHE_TAG_DIRTY
,
387 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
388 struct page
*page
= pvec
.pages
[i
];
391 nilfs_clear_dirty_page(page
, silent
);
394 pagevec_release(&pvec
);
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
404 void nilfs_clear_dirty_page(struct page
*page
, bool silent
)
406 struct inode
*inode
= page
->mapping
->host
;
407 struct super_block
*sb
= inode
->i_sb
;
409 BUG_ON(!PageLocked(page
));
412 nilfs_warning(sb
, __func__
,
413 "discard page: offset %lld, ino %lu",
414 page_offset(page
), inode
->i_ino
);
417 ClearPageUptodate(page
);
418 ClearPageMappedToDisk(page
);
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
);
427 bh
= head
= page_buffers(page
);
431 nilfs_warning(sb
, __func__
,
432 "discard block %llu, size %zu",
433 (u64
)bh
->b_blocknr
, bh
->b_size
);
435 set_mask_bits(&bh
->b_state
, clear_bits
, 0);
437 } while (bh
= bh
->b_this_page
, bh
!= head
);
440 __nilfs_clear_page_dirty(page
);
443 unsigned nilfs_page_count_clean_buffers(struct page
*page
,
444 unsigned from
, unsigned to
)
446 unsigned block_start
, block_end
;
447 struct buffer_head
*bh
, *head
;
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
))
460 void nilfs_mapping_init(struct address_space
*mapping
, struct inode
*inode
)
462 mapping
->host
= inode
;
464 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
465 mapping
->private_data
= NULL
;
466 mapping
->a_ops
= &empty_aops
;
470 * NILFS2 needs clear_page_dirty() in the following two cases:
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}).
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.
480 int __nilfs_clear_page_dirty(struct page
*page
)
482 struct address_space
*mapping
= page
->mapping
;
485 spin_lock_irq(&mapping
->tree_lock
);
486 if (test_bit(PG_dirty
, &page
->flags
)) {
487 radix_tree_tag_clear(&mapping
->page_tree
,
489 PAGECACHE_TAG_DIRTY
);
490 spin_unlock_irq(&mapping
->tree_lock
);
491 return clear_page_dirty_for_io(page
);
493 spin_unlock_irq(&mapping
->tree_lock
);
496 return TestClearPageDirty(page
);
500 * nilfs_find_uncommitted_extent - find extent of uncommitted data
502 * @start_blk: start block offset (in)
503 * @blkoff: start offset of the found extent (out)
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
511 unsigned long nilfs_find_uncommitted_extent(struct inode
*inode
,
517 unsigned int nblocks_in_page
;
518 unsigned long length
= 0;
523 if (inode
->i_mapping
->nrpages
== 0)
526 index
= start_blk
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
527 nblocks_in_page
= 1U << (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
529 pagevec_init(&pvec
, 0);
532 pvec
.nr
= find_get_pages_contig(inode
->i_mapping
, index
, PAGEVEC_SIZE
,
537 if (length
> 0 && pvec
.pages
[0]->index
> index
)
540 b
= pvec
.pages
[0]->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
543 page
= pvec
.pages
[i
];
546 if (page_has_buffers(page
)) {
547 struct buffer_head
*bh
, *head
;
549 bh
= head
= page_buffers(page
);
553 if (buffer_delay(bh
)) {
557 } else if (length
> 0) {
560 } while (++b
, bh
= bh
->b_this_page
, bh
!= head
);
565 b
+= nblocks_in_page
;
569 } while (++i
< pagevec_count(&pvec
));
571 index
= page
->index
+ 1;
572 pagevec_release(&pvec
);
579 pagevec_release(&pvec
);