blob: a585b35fd6bc201c9d3063005792101da769ab01 [file] [log] [blame]
Ryusuke Konishi0bd49f92009-04-06 19:01:27 -07001/*
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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090032#include <linux/gfp.h>
Ryusuke Konishi0bd49f92009-04-06 19:01:27 -070033#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) | \
Ryusuke Konishi4e13e662010-07-18 10:42:25 +090040 (1UL << BH_NILFS_Volatile) | (1UL << BH_NILFS_Allocated) | \
41 (1UL << BH_NILFS_Checked))
Ryusuke Konishi0bd49f92009-04-06 19:01:27 -070042
43static struct buffer_head *
44__nilfs_get_page_block(struct page *page, unsigned long block, pgoff_t index,
45 int blkbits, unsigned long b_state)
46
47{
48 unsigned long first_block;
49 struct buffer_head *bh;
50
51 if (!page_has_buffers(page))
52 create_empty_buffers(page, 1 << blkbits, b_state);
53
54 first_block = (unsigned long)index << (PAGE_CACHE_SHIFT - blkbits);
55 bh = nilfs_page_get_nth_block(page, block - first_block);
56
57 touch_buffer(bh);
58 wait_on_buffer(bh);
59 return bh;
60}
61
62/*
63 * Since the page cache of B-tree node pages or data page cache of pseudo
64 * inodes does not have a valid mapping->host pointer, calling
65 * mark_buffer_dirty() for their buffers causes a NULL pointer dereference;
66 * it calls __mark_inode_dirty(NULL) through __set_page_dirty().
67 * To avoid this problem, the old style mark_buffer_dirty() is used instead.
68 */
69void nilfs_mark_buffer_dirty(struct buffer_head *bh)
70{
71 if (!buffer_dirty(bh) && !test_set_buffer_dirty(bh))
72 __set_page_dirty_nobuffers(bh->b_page);
73}
74
75struct buffer_head *nilfs_grab_buffer(struct inode *inode,
76 struct address_space *mapping,
77 unsigned long blkoff,
78 unsigned long b_state)
79{
80 int blkbits = inode->i_blkbits;
81 pgoff_t index = blkoff >> (PAGE_CACHE_SHIFT - blkbits);
Ryusuke Konishic1c1d70922010-08-29 12:44:56 +090082 struct page *page;
83 struct buffer_head *bh;
Ryusuke Konishi0bd49f92009-04-06 19:01:27 -070084
85 page = grab_cache_page(mapping, index);
86 if (unlikely(!page))
87 return NULL;
88
89 bh = __nilfs_get_page_block(page, blkoff, index, blkbits, b_state);
90 if (unlikely(!bh)) {
91 unlock_page(page);
92 page_cache_release(page);
93 return NULL;
94 }
Ryusuke Konishi0bd49f92009-04-06 19:01:27 -070095 return bh;
96}
97
98/**
99 * nilfs_forget_buffer - discard dirty state
100 * @inode: owner inode of the buffer
101 * @bh: buffer head of the buffer to be discarded
102 */
103void nilfs_forget_buffer(struct buffer_head *bh)
104{
105 struct page *page = bh->b_page;
106
107 lock_buffer(bh);
108 clear_buffer_nilfs_volatile(bh);
Ryusuke Konishi4e13e662010-07-18 10:42:25 +0900109 clear_buffer_nilfs_checked(bh);
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900110 clear_buffer_nilfs_redirected(bh);
Ryusuke Konishi84338232009-05-05 21:52:06 +0900111 clear_buffer_dirty(bh);
112 if (nilfs_page_buffers_clean(page))
Ryusuke Konishi0bd49f92009-04-06 19:01:27 -0700113 __nilfs_clear_page_dirty(page);
114
115 clear_buffer_uptodate(bh);
116 clear_buffer_mapped(bh);
117 bh->b_blocknr = -1;
118 ClearPageUptodate(page);
119 ClearPageMappedToDisk(page);
120 unlock_buffer(bh);
121 brelse(bh);
122}
123
124/**
125 * nilfs_copy_buffer -- copy buffer data and flags
126 * @dbh: destination buffer
127 * @sbh: source buffer
128 */
129void nilfs_copy_buffer(struct buffer_head *dbh, struct buffer_head *sbh)
130{
131 void *kaddr0, *kaddr1;
132 unsigned long bits;
133 struct page *spage = sbh->b_page, *dpage = dbh->b_page;
134 struct buffer_head *bh;
135
136 kaddr0 = kmap_atomic(spage, KM_USER0);
137 kaddr1 = kmap_atomic(dpage, KM_USER1);
138 memcpy(kaddr1 + bh_offset(dbh), kaddr0 + bh_offset(sbh), sbh->b_size);
139 kunmap_atomic(kaddr1, KM_USER1);
140 kunmap_atomic(kaddr0, KM_USER0);
141
142 dbh->b_state = sbh->b_state & NILFS_BUFFER_INHERENT_BITS;
143 dbh->b_blocknr = sbh->b_blocknr;
144 dbh->b_bdev = sbh->b_bdev;
145
146 bh = dbh;
147 bits = sbh->b_state & ((1UL << BH_Uptodate) | (1UL << BH_Mapped));
148 while ((bh = bh->b_this_page) != dbh) {
149 lock_buffer(bh);
150 bits &= bh->b_state;
151 unlock_buffer(bh);
152 }
153 if (bits & (1UL << BH_Uptodate))
154 SetPageUptodate(dpage);
155 else
156 ClearPageUptodate(dpage);
157 if (bits & (1UL << BH_Mapped))
158 SetPageMappedToDisk(dpage);
159 else
160 ClearPageMappedToDisk(dpage);
161}
162
163/**
164 * nilfs_page_buffers_clean - check if a page has dirty buffers or not.
165 * @page: page to be checked
166 *
167 * nilfs_page_buffers_clean() returns zero if the page has dirty buffers.
168 * Otherwise, it returns non-zero value.
169 */
170int nilfs_page_buffers_clean(struct page *page)
171{
172 struct buffer_head *bh, *head;
173
174 bh = head = page_buffers(page);
175 do {
176 if (buffer_dirty(bh))
177 return 0;
178 bh = bh->b_this_page;
179 } while (bh != head);
180 return 1;
181}
182
183void nilfs_page_bug(struct page *page)
184{
185 struct address_space *m;
186 unsigned long ino = 0;
187
188 if (unlikely(!page)) {
189 printk(KERN_CRIT "NILFS_PAGE_BUG(NULL)\n");
190 return;
191 }
192
193 m = page->mapping;
194 if (m) {
195 struct inode *inode = NILFS_AS_I(m);
196 if (inode != NULL)
197 ino = inode->i_ino;
198 }
199 printk(KERN_CRIT "NILFS_PAGE_BUG(%p): cnt=%d index#=%llu flags=0x%lx "
200 "mapping=%p ino=%lu\n",
201 page, atomic_read(&page->_count),
202 (unsigned long long)page->index, page->flags, m, ino);
203
204 if (page_has_buffers(page)) {
205 struct buffer_head *bh, *head;
206 int i = 0;
207
208 bh = head = page_buffers(page);
209 do {
210 printk(KERN_CRIT
211 " BH[%d] %p: cnt=%d block#=%llu state=0x%lx\n",
212 i++, bh, atomic_read(&bh->b_count),
213 (unsigned long long)bh->b_blocknr, bh->b_state);
214 bh = bh->b_this_page;
215 } while (bh != head);
216 }
217}
218
219/**
220 * nilfs_alloc_private_page - allocate a private page with buffer heads
221 *
222 * Return Value: On success, a pointer to the allocated page is returned.
223 * On error, NULL is returned.
224 */
225struct page *nilfs_alloc_private_page(struct block_device *bdev, int size,
226 unsigned long state)
227{
228 struct buffer_head *bh, *head, *tail;
229 struct page *page;
230
231 page = alloc_page(GFP_NOFS); /* page_count of the returned page is 1 */
232 if (unlikely(!page))
233 return NULL;
234
235 lock_page(page);
236 head = alloc_page_buffers(page, size, 0);
237 if (unlikely(!head)) {
238 unlock_page(page);
239 __free_page(page);
240 return NULL;
241 }
242
243 bh = head;
244 do {
245 bh->b_state = (1UL << BH_NILFS_Allocated) | state;
246 tail = bh;
247 bh->b_bdev = bdev;
248 bh = bh->b_this_page;
249 } while (bh);
250
251 tail->b_this_page = head;
252 attach_page_buffers(page, head);
253
254 return page;
255}
256
257void nilfs_free_private_page(struct page *page)
258{
259 BUG_ON(!PageLocked(page));
260 BUG_ON(page->mapping);
261
262 if (page_has_buffers(page) && !try_to_free_buffers(page))
263 NILFS_PAGE_BUG(page, "failed to free page");
264
265 unlock_page(page);
266 __free_page(page);
267}
268
269/**
270 * nilfs_copy_page -- copy the page with buffers
271 * @dst: destination page
272 * @src: source page
273 * @copy_dirty: flag whether to copy dirty states on the page's buffer heads.
274 *
Ryusuke Konishi7a650042010-03-14 03:32:40 +0900275 * This function is for both data pages and btnode pages. The dirty flag
Ryusuke Konishi0bd49f92009-04-06 19:01:27 -0700276 * should be treated by caller. The page must not be under i/o.
277 * Both src and dst page must be locked
278 */
279static void nilfs_copy_page(struct page *dst, struct page *src, int copy_dirty)
280{
281 struct buffer_head *dbh, *dbufs, *sbh, *sbufs;
282 unsigned long mask = NILFS_BUFFER_INHERENT_BITS;
283
284 BUG_ON(PageWriteback(dst));
285
286 sbh = sbufs = page_buffers(src);
287 if (!page_has_buffers(dst))
288 create_empty_buffers(dst, sbh->b_size, 0);
289
290 if (copy_dirty)
291 mask |= (1UL << BH_Dirty);
292
293 dbh = dbufs = page_buffers(dst);
294 do {
295 lock_buffer(sbh);
296 lock_buffer(dbh);
297 dbh->b_state = sbh->b_state & mask;
298 dbh->b_blocknr = sbh->b_blocknr;
299 dbh->b_bdev = sbh->b_bdev;
300 sbh = sbh->b_this_page;
301 dbh = dbh->b_this_page;
302 } while (dbh != dbufs);
303
304 copy_highpage(dst, src);
305
306 if (PageUptodate(src) && !PageUptodate(dst))
307 SetPageUptodate(dst);
308 else if (!PageUptodate(src) && PageUptodate(dst))
309 ClearPageUptodate(dst);
310 if (PageMappedToDisk(src) && !PageMappedToDisk(dst))
311 SetPageMappedToDisk(dst);
312 else if (!PageMappedToDisk(src) && PageMappedToDisk(dst))
313 ClearPageMappedToDisk(dst);
314
315 do {
316 unlock_buffer(sbh);
317 unlock_buffer(dbh);
318 sbh = sbh->b_this_page;
319 dbh = dbh->b_this_page;
320 } while (dbh != dbufs);
321}
322
323int nilfs_copy_dirty_pages(struct address_space *dmap,
324 struct address_space *smap)
325{
326 struct pagevec pvec;
327 unsigned int i;
328 pgoff_t index = 0;
329 int err = 0;
330
331 pagevec_init(&pvec, 0);
332repeat:
333 if (!pagevec_lookup_tag(&pvec, smap, &index, PAGECACHE_TAG_DIRTY,
334 PAGEVEC_SIZE))
335 return 0;
336
337 for (i = 0; i < pagevec_count(&pvec); i++) {
338 struct page *page = pvec.pages[i], *dpage;
339
340 lock_page(page);
341 if (unlikely(!PageDirty(page)))
342 NILFS_PAGE_BUG(page, "inconsistent dirty state");
343
344 dpage = grab_cache_page(dmap, page->index);
345 if (unlikely(!dpage)) {
346 /* No empty page is added to the page cache */
347 err = -ENOMEM;
348 unlock_page(page);
349 break;
350 }
351 if (unlikely(!page_has_buffers(page)))
352 NILFS_PAGE_BUG(page,
353 "found empty page in dat page cache");
354
355 nilfs_copy_page(dpage, page, 1);
356 __set_page_dirty_nobuffers(dpage);
357
358 unlock_page(dpage);
359 page_cache_release(dpage);
360 unlock_page(page);
361 }
362 pagevec_release(&pvec);
363 cond_resched();
364
365 if (likely(!err))
366 goto repeat;
367 return err;
368}
369
370/**
Ryusuke Konishi7a650042010-03-14 03:32:40 +0900371 * nilfs_copy_back_pages -- copy back pages to original cache from shadow cache
Ryusuke Konishi0bd49f92009-04-06 19:01:27 -0700372 * @dmap: destination page cache
373 * @smap: source page cache
374 *
375 * No pages must no be added to the cache during this process.
376 * This must be ensured by the caller.
377 */
378void nilfs_copy_back_pages(struct address_space *dmap,
379 struct address_space *smap)
380{
381 struct pagevec pvec;
382 unsigned int i, n;
383 pgoff_t index = 0;
384 int err;
385
386 pagevec_init(&pvec, 0);
387repeat:
388 n = pagevec_lookup(&pvec, smap, index, PAGEVEC_SIZE);
389 if (!n)
390 return;
391 index = pvec.pages[n - 1]->index + 1;
392
393 for (i = 0; i < pagevec_count(&pvec); i++) {
394 struct page *page = pvec.pages[i], *dpage;
395 pgoff_t offset = page->index;
396
397 lock_page(page);
398 dpage = find_lock_page(dmap, offset);
399 if (dpage) {
400 /* override existing page on the destination cache */
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -0700401 WARN_ON(PageDirty(dpage));
Ryusuke Konishi0bd49f92009-04-06 19:01:27 -0700402 nilfs_copy_page(dpage, page, 0);
403 unlock_page(dpage);
404 page_cache_release(dpage);
405 } else {
406 struct page *page2;
407
408 /* move the page to the destination cache */
409 spin_lock_irq(&smap->tree_lock);
410 page2 = radix_tree_delete(&smap->page_tree, offset);
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -0700411 WARN_ON(page2 != page);
412
Ryusuke Konishi0bd49f92009-04-06 19:01:27 -0700413 smap->nrpages--;
414 spin_unlock_irq(&smap->tree_lock);
415
416 spin_lock_irq(&dmap->tree_lock);
417 err = radix_tree_insert(&dmap->page_tree, offset, page);
418 if (unlikely(err < 0)) {
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -0700419 WARN_ON(err == -EEXIST);
Ryusuke Konishi0bd49f92009-04-06 19:01:27 -0700420 page->mapping = NULL;
421 page_cache_release(page); /* for cache */
422 } else {
423 page->mapping = dmap;
424 dmap->nrpages++;
425 if (PageDirty(page))
426 radix_tree_tag_set(&dmap->page_tree,
427 offset,
428 PAGECACHE_TAG_DIRTY);
429 }
430 spin_unlock_irq(&dmap->tree_lock);
431 }
432 unlock_page(page);
433 }
434 pagevec_release(&pvec);
435 cond_resched();
436
437 goto repeat;
438}
439
440void nilfs_clear_dirty_pages(struct address_space *mapping)
441{
442 struct pagevec pvec;
443 unsigned int i;
444 pgoff_t index = 0;
445
446 pagevec_init(&pvec, 0);
447
448 while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
449 PAGEVEC_SIZE)) {
450 for (i = 0; i < pagevec_count(&pvec); i++) {
451 struct page *page = pvec.pages[i];
452 struct buffer_head *bh, *head;
453
454 lock_page(page);
455 ClearPageUptodate(page);
456 ClearPageMappedToDisk(page);
457 bh = head = page_buffers(page);
458 do {
459 lock_buffer(bh);
460 clear_buffer_dirty(bh);
461 clear_buffer_nilfs_volatile(bh);
Ryusuke Konishi4e13e662010-07-18 10:42:25 +0900462 clear_buffer_nilfs_checked(bh);
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900463 clear_buffer_nilfs_redirected(bh);
Ryusuke Konishi0bd49f92009-04-06 19:01:27 -0700464 clear_buffer_uptodate(bh);
465 clear_buffer_mapped(bh);
466 unlock_buffer(bh);
467 bh = bh->b_this_page;
468 } while (bh != head);
469
470 __nilfs_clear_page_dirty(page);
471 unlock_page(page);
472 }
473 pagevec_release(&pvec);
474 cond_resched();
475 }
476}
477
478unsigned nilfs_page_count_clean_buffers(struct page *page,
479 unsigned from, unsigned to)
480{
481 unsigned block_start, block_end;
482 struct buffer_head *bh, *head;
483 unsigned nc = 0;
484
485 for (bh = head = page_buffers(page), block_start = 0;
486 bh != head || !block_start;
487 block_start = block_end, bh = bh->b_this_page) {
488 block_end = block_start + bh->b_size;
489 if (block_end > from && block_start < to && !buffer_dirty(bh))
490 nc++;
491 }
492 return nc;
493}
Ryusuke Konishiae53a0a2010-12-26 23:30:02 +0900494
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900495void nilfs_mapping_init(struct address_space *mapping,
496 struct backing_dev_info *bdi,
497 const struct address_space_operations *aops)
498{
499 mapping->host = NULL;
500 mapping->flags = 0;
501 mapping_set_gfp_mask(mapping, GFP_NOFS);
502 mapping->assoc_mapping = NULL;
503 mapping->backing_dev_info = bdi;
504 mapping->a_ops = aops;
505}
Ryusuke Konishi0bd49f92009-04-06 19:01:27 -0700506
507/*
508 * NILFS2 needs clear_page_dirty() in the following two cases:
509 *
510 * 1) For B-tree node pages and data pages of the dat/gcdat, NILFS2 clears
511 * page dirty flags when it copies back pages from the shadow cache
512 * (gcdat->{i_mapping,i_btnode_cache}) to its original cache
513 * (dat->{i_mapping,i_btnode_cache}).
514 *
515 * 2) Some B-tree operations like insertion or deletion may dispose buffers
516 * in dirty state, and this needs to cancel the dirty state of their pages.
517 */
518int __nilfs_clear_page_dirty(struct page *page)
519{
520 struct address_space *mapping = page->mapping;
521
522 if (mapping) {
523 spin_lock_irq(&mapping->tree_lock);
524 if (test_bit(PG_dirty, &page->flags)) {
525 radix_tree_tag_clear(&mapping->page_tree,
526 page_index(page),
527 PAGECACHE_TAG_DIRTY);
528 spin_unlock_irq(&mapping->tree_lock);
529 return clear_page_dirty_for_io(page);
530 }
531 spin_unlock_irq(&mapping->tree_lock);
532 return 0;
533 }
534 return TestClearPageDirty(page);
535}
Ryusuke Konishi622daaff2010-12-26 16:38:43 +0900536
537/**
538 * nilfs_find_uncommitted_extent - find extent of uncommitted data
539 * @inode: inode
540 * @start_blk: start block offset (in)
541 * @blkoff: start offset of the found extent (out)
542 *
543 * This function searches an extent of buffers marked "delayed" which
544 * starts from a block offset equal to or larger than @start_blk. If
545 * such an extent was found, this will store the start offset in
546 * @blkoff and return its length in blocks. Otherwise, zero is
547 * returned.
548 */
549unsigned long nilfs_find_uncommitted_extent(struct inode *inode,
550 sector_t start_blk,
551 sector_t *blkoff)
552{
553 unsigned int i;
554 pgoff_t index;
555 unsigned int nblocks_in_page;
556 unsigned long length = 0;
557 sector_t b;
558 struct pagevec pvec;
559 struct page *page;
560
561 if (inode->i_mapping->nrpages == 0)
562 return 0;
563
564 index = start_blk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
565 nblocks_in_page = 1U << (PAGE_CACHE_SHIFT - inode->i_blkbits);
566
567 pagevec_init(&pvec, 0);
568
569repeat:
570 pvec.nr = find_get_pages_contig(inode->i_mapping, index, PAGEVEC_SIZE,
571 pvec.pages);
572 if (pvec.nr == 0)
573 return length;
574
575 if (length > 0 && pvec.pages[0]->index > index)
576 goto out;
577
578 b = pvec.pages[0]->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
579 i = 0;
580 do {
581 page = pvec.pages[i];
582
583 lock_page(page);
584 if (page_has_buffers(page)) {
585 struct buffer_head *bh, *head;
586
587 bh = head = page_buffers(page);
588 do {
589 if (b < start_blk)
590 continue;
591 if (buffer_delay(bh)) {
592 if (length == 0)
593 *blkoff = b;
594 length++;
595 } else if (length > 0) {
596 goto out_locked;
597 }
598 } while (++b, bh = bh->b_this_page, bh != head);
599 } else {
600 if (length > 0)
601 goto out_locked;
602
603 b += nblocks_in_page;
604 }
605 unlock_page(page);
606
607 } while (++i < pagevec_count(&pvec));
608
609 index = page->index + 1;
610 pagevec_release(&pvec);
611 cond_resched();
612 goto repeat;
613
614out_locked:
615 unlock_page(page);
616out:
617 pagevec_release(&pvec);
618 return length;
619}