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
42 static struct buffer_head *
__nilfs_get_page_block(struct page * page,unsigned long block,pgoff_t index,int blkbits,unsigned long b_state)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
nilfs_grab_buffer(struct inode * inode,struct address_space * mapping,unsigned long blkoff,unsigned long b_state)61 struct 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 */
nilfs_forget_buffer(struct buffer_head * bh)89 void 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 */
nilfs_copy_buffer(struct buffer_head * dbh,struct buffer_head * sbh)116 void 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 */
nilfs_page_buffers_clean(struct page * page)157 int 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
nilfs_page_bug(struct page * page)170 void 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 */
nilfs_copy_page(struct page * dst,struct page * src,int copy_dirty)213 static 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
nilfs_copy_dirty_pages(struct address_space * dmap,struct address_space * smap)257 int 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);
266 repeat:
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 */
nilfs_copy_back_pages(struct address_space * dmap,struct address_space * smap)312 void 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);
321 repeat:
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 */
nilfs_clear_dirty_pages(struct address_space * mapping,bool silent)379 void 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 */
nilfs_clear_dirty_page(struct page * page,bool silent)406 void 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
nilfs_page_count_clean_buffers(struct page * page,unsigned from,unsigned to)447 unsigned 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
nilfs_mapping_init(struct address_space * mapping,struct inode * inode,struct backing_dev_info * bdi)464 void 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 */
__nilfs_clear_page_dirty(struct page * page)486 int __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 */
nilfs_find_uncommitted_extent(struct inode * inode,sector_t start_blk,sector_t * blkoff)517 unsigned 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
537 repeat:
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
582 out_locked:
583 unlock_page(page);
584 out:
585 pagevec_release(&pvec);
586 return length;
587 }
588