1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3 * page.c - 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
28 static struct buffer_head *
__nilfs_get_page_block(struct page * page,unsigned long block,pgoff_t index,int blkbits,unsigned long b_state)29 __nilfs_get_page_block(struct page *page, unsigned long block, pgoff_t index,
30 int blkbits, unsigned long b_state)
31
32 {
33 unsigned long first_block;
34 struct buffer_head *bh;
35
36 if (!page_has_buffers(page))
37 create_empty_buffers(page, 1 << blkbits, b_state);
38
39 first_block = (unsigned long)index << (PAGE_SHIFT - blkbits);
40 bh = nilfs_page_get_nth_block(page, block - first_block);
41
42 touch_buffer(bh);
43 wait_on_buffer(bh);
44 return bh;
45 }
46
nilfs_grab_buffer(struct inode * inode,struct address_space * mapping,unsigned long blkoff,unsigned long b_state)47 struct 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 page *page;
55 struct buffer_head *bh;
56
57 page = grab_cache_page(mapping, index);
58 if (unlikely(!page))
59 return NULL;
60
61 bh = __nilfs_get_page_block(page, blkoff, index, blkbits, b_state);
62 if (unlikely(!bh)) {
63 unlock_page(page);
64 put_page(page);
65 return NULL;
66 }
67 return bh;
68 }
69
70 /**
71 * nilfs_forget_buffer - discard dirty state
72 * @inode: owner inode of the buffer
73 * @bh: buffer head of the buffer to be discarded
74 */
nilfs_forget_buffer(struct buffer_head * bh)75 void nilfs_forget_buffer(struct buffer_head *bh)
76 {
77 struct page *page = bh->b_page;
78 const unsigned long clear_bits =
79 (BIT(BH_Uptodate) | BIT(BH_Dirty) | BIT(BH_Mapped) |
80 BIT(BH_Async_Write) | BIT(BH_NILFS_Volatile) |
81 BIT(BH_NILFS_Checked) | BIT(BH_NILFS_Redirected));
82
83 lock_buffer(bh);
84 set_mask_bits(&bh->b_state, clear_bits, 0);
85 if (nilfs_page_buffers_clean(page))
86 __nilfs_clear_page_dirty(page);
87
88 bh->b_blocknr = -1;
89 ClearPageUptodate(page);
90 ClearPageMappedToDisk(page);
91 unlock_buffer(bh);
92 brelse(bh);
93 }
94
95 /**
96 * nilfs_copy_buffer -- copy buffer data and flags
97 * @dbh: destination buffer
98 * @sbh: source buffer
99 */
nilfs_copy_buffer(struct buffer_head * dbh,struct buffer_head * sbh)100 void nilfs_copy_buffer(struct buffer_head *dbh, struct buffer_head *sbh)
101 {
102 void *kaddr0, *kaddr1;
103 unsigned long bits;
104 struct page *spage = sbh->b_page, *dpage = dbh->b_page;
105 struct buffer_head *bh;
106
107 kaddr0 = kmap_atomic(spage);
108 kaddr1 = kmap_atomic(dpage);
109 memcpy(kaddr1 + bh_offset(dbh), kaddr0 + bh_offset(sbh), sbh->b_size);
110 kunmap_atomic(kaddr1);
111 kunmap_atomic(kaddr0);
112
113 dbh->b_state = sbh->b_state & NILFS_BUFFER_INHERENT_BITS;
114 dbh->b_blocknr = sbh->b_blocknr;
115 dbh->b_bdev = sbh->b_bdev;
116
117 bh = dbh;
118 bits = sbh->b_state & (BIT(BH_Uptodate) | BIT(BH_Mapped));
119 while ((bh = bh->b_this_page) != dbh) {
120 lock_buffer(bh);
121 bits &= bh->b_state;
122 unlock_buffer(bh);
123 }
124 if (bits & BIT(BH_Uptodate))
125 SetPageUptodate(dpage);
126 else
127 ClearPageUptodate(dpage);
128 if (bits & BIT(BH_Mapped))
129 SetPageMappedToDisk(dpage);
130 else
131 ClearPageMappedToDisk(dpage);
132 }
133
134 /**
135 * nilfs_page_buffers_clean - check if a page has dirty buffers or not.
136 * @page: page to be checked
137 *
138 * nilfs_page_buffers_clean() returns zero if the page has dirty buffers.
139 * Otherwise, it returns non-zero value.
140 */
nilfs_page_buffers_clean(struct page * page)141 int nilfs_page_buffers_clean(struct page *page)
142 {
143 struct buffer_head *bh, *head;
144
145 bh = head = page_buffers(page);
146 do {
147 if (buffer_dirty(bh))
148 return 0;
149 bh = bh->b_this_page;
150 } while (bh != head);
151 return 1;
152 }
153
nilfs_page_bug(struct page * page)154 void nilfs_page_bug(struct page *page)
155 {
156 struct address_space *m;
157 unsigned long ino;
158
159 if (unlikely(!page)) {
160 printk(KERN_CRIT "NILFS_PAGE_BUG(NULL)\n");
161 return;
162 }
163
164 m = page->mapping;
165 ino = m ? m->host->i_ino : 0;
166
167 printk(KERN_CRIT "NILFS_PAGE_BUG(%p): cnt=%d index#=%llu flags=0x%lx "
168 "mapping=%p ino=%lu\n",
169 page, page_ref_count(page),
170 (unsigned long long)page->index, page->flags, m, ino);
171
172 if (page_has_buffers(page)) {
173 struct buffer_head *bh, *head;
174 int i = 0;
175
176 bh = head = page_buffers(page);
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_page -- copy the page with buffers
189 * @dst: destination page
190 * @src: source page
191 * @copy_dirty: flag whether to copy dirty states on the page's buffer heads.
192 *
193 * This function is for both data pages and btnode pages. The dirty flag
194 * should be treated by caller. The page must not be under i/o.
195 * Both src and dst page must be locked
196 */
nilfs_copy_page(struct page * dst,struct page * src,int copy_dirty)197 static void nilfs_copy_page(struct page *dst, struct page *src, int copy_dirty)
198 {
199 struct buffer_head *dbh, *dbufs, *sbh, *sbufs;
200 unsigned long mask = NILFS_BUFFER_INHERENT_BITS;
201
202 BUG_ON(PageWriteback(dst));
203
204 sbh = sbufs = page_buffers(src);
205 if (!page_has_buffers(dst))
206 create_empty_buffers(dst, sbh->b_size, 0);
207
208 if (copy_dirty)
209 mask |= BIT(BH_Dirty);
210
211 dbh = dbufs = page_buffers(dst);
212 do {
213 lock_buffer(sbh);
214 lock_buffer(dbh);
215 dbh->b_state = sbh->b_state & mask;
216 dbh->b_blocknr = sbh->b_blocknr;
217 dbh->b_bdev = sbh->b_bdev;
218 sbh = sbh->b_this_page;
219 dbh = dbh->b_this_page;
220 } while (dbh != dbufs);
221
222 copy_highpage(dst, src);
223
224 if (PageUptodate(src) && !PageUptodate(dst))
225 SetPageUptodate(dst);
226 else if (!PageUptodate(src) && PageUptodate(dst))
227 ClearPageUptodate(dst);
228 if (PageMappedToDisk(src) && !PageMappedToDisk(dst))
229 SetPageMappedToDisk(dst);
230 else if (!PageMappedToDisk(src) && PageMappedToDisk(dst))
231 ClearPageMappedToDisk(dst);
232
233 do {
234 unlock_buffer(sbh);
235 unlock_buffer(dbh);
236 sbh = sbh->b_this_page;
237 dbh = dbh->b_this_page;
238 } while (dbh != dbufs);
239 }
240
nilfs_copy_dirty_pages(struct address_space * dmap,struct address_space * smap)241 int nilfs_copy_dirty_pages(struct address_space *dmap,
242 struct address_space *smap)
243 {
244 struct pagevec pvec;
245 unsigned int i;
246 pgoff_t index = 0;
247 int err = 0;
248
249 pagevec_init(&pvec);
250 repeat:
251 if (!pagevec_lookup_tag(&pvec, smap, &index, PAGECACHE_TAG_DIRTY))
252 return 0;
253
254 for (i = 0; i < pagevec_count(&pvec); i++) {
255 struct page *page = pvec.pages[i], *dpage;
256
257 lock_page(page);
258 if (unlikely(!PageDirty(page)))
259 NILFS_PAGE_BUG(page, "inconsistent dirty state");
260
261 dpage = grab_cache_page(dmap, page->index);
262 if (unlikely(!dpage)) {
263 /* No empty page is added to the page cache */
264 err = -ENOMEM;
265 unlock_page(page);
266 break;
267 }
268 if (unlikely(!page_has_buffers(page)))
269 NILFS_PAGE_BUG(page,
270 "found empty page in dat page cache");
271
272 nilfs_copy_page(dpage, page, 1);
273 __set_page_dirty_nobuffers(dpage);
274
275 unlock_page(dpage);
276 put_page(dpage);
277 unlock_page(page);
278 }
279 pagevec_release(&pvec);
280 cond_resched();
281
282 if (likely(!err))
283 goto repeat;
284 return err;
285 }
286
287 /**
288 * nilfs_copy_back_pages -- copy back pages to original cache from shadow cache
289 * @dmap: destination page cache
290 * @smap: source page cache
291 *
292 * No pages must no be added to the cache during this process.
293 * This must be ensured by the caller.
294 */
nilfs_copy_back_pages(struct address_space * dmap,struct address_space * smap)295 void nilfs_copy_back_pages(struct address_space *dmap,
296 struct address_space *smap)
297 {
298 struct pagevec pvec;
299 unsigned int i, n;
300 pgoff_t index = 0;
301 int err;
302
303 pagevec_init(&pvec);
304 repeat:
305 n = pagevec_lookup(&pvec, smap, &index);
306 if (!n)
307 return;
308
309 for (i = 0; i < pagevec_count(&pvec); i++) {
310 struct page *page = pvec.pages[i], *dpage;
311 pgoff_t offset = page->index;
312
313 lock_page(page);
314 dpage = find_lock_page(dmap, offset);
315 if (dpage) {
316 /* override existing page on the destination cache */
317 WARN_ON(PageDirty(dpage));
318 nilfs_copy_page(dpage, page, 0);
319 unlock_page(dpage);
320 put_page(dpage);
321 } else {
322 struct page *page2;
323
324 /* move the page to the destination cache */
325 xa_lock_irq(&smap->i_pages);
326 page2 = radix_tree_delete(&smap->i_pages, offset);
327 WARN_ON(page2 != page);
328
329 smap->nrpages--;
330 xa_unlock_irq(&smap->i_pages);
331
332 xa_lock_irq(&dmap->i_pages);
333 err = radix_tree_insert(&dmap->i_pages, offset, page);
334 if (unlikely(err < 0)) {
335 WARN_ON(err == -EEXIST);
336 page->mapping = NULL;
337 put_page(page); /* for cache */
338 } else {
339 page->mapping = dmap;
340 dmap->nrpages++;
341 if (PageDirty(page))
342 radix_tree_tag_set(&dmap->i_pages,
343 offset,
344 PAGECACHE_TAG_DIRTY);
345 }
346 xa_unlock_irq(&dmap->i_pages);
347 }
348 unlock_page(page);
349 }
350 pagevec_release(&pvec);
351 cond_resched();
352
353 goto repeat;
354 }
355
356 /**
357 * nilfs_clear_dirty_pages - discard dirty pages in address space
358 * @mapping: address space with dirty pages for discarding
359 * @silent: suppress [true] or print [false] warning messages
360 */
nilfs_clear_dirty_pages(struct address_space * mapping,bool silent)361 void nilfs_clear_dirty_pages(struct address_space *mapping, bool silent)
362 {
363 struct pagevec pvec;
364 unsigned int i;
365 pgoff_t index = 0;
366
367 pagevec_init(&pvec);
368
369 while (pagevec_lookup_tag(&pvec, mapping, &index,
370 PAGECACHE_TAG_DIRTY)) {
371 for (i = 0; i < pagevec_count(&pvec); i++) {
372 struct page *page = pvec.pages[i];
373
374 lock_page(page);
375
376 /*
377 * This page may have been removed from the address
378 * space by truncation or invalidation when the lock
379 * was acquired. Skip processing in that case.
380 */
381 if (likely(page->mapping == mapping))
382 nilfs_clear_dirty_page(page, silent);
383
384 unlock_page(page);
385 }
386 pagevec_release(&pvec);
387 cond_resched();
388 }
389 }
390
391 /**
392 * nilfs_clear_dirty_page - discard dirty page
393 * @page: dirty page that will be discarded
394 * @silent: suppress [true] or print [false] warning messages
395 */
nilfs_clear_dirty_page(struct page * page,bool silent)396 void nilfs_clear_dirty_page(struct page *page, bool silent)
397 {
398 struct inode *inode = page->mapping->host;
399 struct super_block *sb = inode->i_sb;
400
401 BUG_ON(!PageLocked(page));
402
403 if (!silent)
404 nilfs_msg(sb, KERN_WARNING,
405 "discard dirty page: offset=%lld, ino=%lu",
406 page_offset(page), inode->i_ino);
407
408 ClearPageUptodate(page);
409 ClearPageMappedToDisk(page);
410
411 if (page_has_buffers(page)) {
412 struct buffer_head *bh, *head;
413 const unsigned long clear_bits =
414 (BIT(BH_Uptodate) | BIT(BH_Dirty) | BIT(BH_Mapped) |
415 BIT(BH_Async_Write) | BIT(BH_NILFS_Volatile) |
416 BIT(BH_NILFS_Checked) | BIT(BH_NILFS_Redirected));
417
418 bh = head = page_buffers(page);
419 do {
420 lock_buffer(bh);
421 if (!silent)
422 nilfs_msg(sb, KERN_WARNING,
423 "discard dirty block: blocknr=%llu, size=%zu",
424 (u64)bh->b_blocknr, bh->b_size);
425
426 set_mask_bits(&bh->b_state, clear_bits, 0);
427 unlock_buffer(bh);
428 } while (bh = bh->b_this_page, bh != head);
429 }
430
431 __nilfs_clear_page_dirty(page);
432 }
433
nilfs_page_count_clean_buffers(struct page * page,unsigned int from,unsigned int to)434 unsigned int nilfs_page_count_clean_buffers(struct page *page,
435 unsigned int from, unsigned int to)
436 {
437 unsigned int block_start, block_end;
438 struct buffer_head *bh, *head;
439 unsigned int nc = 0;
440
441 for (bh = head = page_buffers(page), block_start = 0;
442 bh != head || !block_start;
443 block_start = block_end, bh = bh->b_this_page) {
444 block_end = block_start + bh->b_size;
445 if (block_end > from && block_start < to && !buffer_dirty(bh))
446 nc++;
447 }
448 return nc;
449 }
450
nilfs_mapping_init(struct address_space * mapping,struct inode * inode)451 void nilfs_mapping_init(struct address_space *mapping, struct inode *inode)
452 {
453 mapping->host = inode;
454 mapping->flags = 0;
455 mapping_set_gfp_mask(mapping, GFP_NOFS);
456 mapping->private_data = NULL;
457 mapping->a_ops = &empty_aops;
458 }
459
460 /*
461 * NILFS2 needs clear_page_dirty() in the following two cases:
462 *
463 * 1) For B-tree node pages and data pages of DAT file, NILFS2 clears dirty
464 * flag of pages when it copies back pages from shadow cache to the
465 * original cache.
466 *
467 * 2) Some B-tree operations like insertion or deletion may dispose buffers
468 * in dirty state, and this needs to cancel the dirty state of their pages.
469 */
__nilfs_clear_page_dirty(struct page * page)470 int __nilfs_clear_page_dirty(struct page *page)
471 {
472 struct address_space *mapping = page->mapping;
473
474 if (mapping) {
475 xa_lock_irq(&mapping->i_pages);
476 if (test_bit(PG_dirty, &page->flags)) {
477 radix_tree_tag_clear(&mapping->i_pages,
478 page_index(page),
479 PAGECACHE_TAG_DIRTY);
480 xa_unlock_irq(&mapping->i_pages);
481 return clear_page_dirty_for_io(page);
482 }
483 xa_unlock_irq(&mapping->i_pages);
484 return 0;
485 }
486 return TestClearPageDirty(page);
487 }
488
489 /**
490 * nilfs_find_uncommitted_extent - find extent of uncommitted data
491 * @inode: inode
492 * @start_blk: start block offset (in)
493 * @blkoff: start offset of the found extent (out)
494 *
495 * This function searches an extent of buffers marked "delayed" which
496 * starts from a block offset equal to or larger than @start_blk. If
497 * such an extent was found, this will store the start offset in
498 * @blkoff and return its length in blocks. Otherwise, zero is
499 * returned.
500 */
nilfs_find_uncommitted_extent(struct inode * inode,sector_t start_blk,sector_t * blkoff)501 unsigned long nilfs_find_uncommitted_extent(struct inode *inode,
502 sector_t start_blk,
503 sector_t *blkoff)
504 {
505 unsigned int i;
506 pgoff_t index;
507 unsigned int nblocks_in_page;
508 unsigned long length = 0;
509 sector_t b;
510 struct pagevec pvec;
511 struct page *page;
512
513 if (inode->i_mapping->nrpages == 0)
514 return 0;
515
516 index = start_blk >> (PAGE_SHIFT - inode->i_blkbits);
517 nblocks_in_page = 1U << (PAGE_SHIFT - inode->i_blkbits);
518
519 pagevec_init(&pvec);
520
521 repeat:
522 pvec.nr = find_get_pages_contig(inode->i_mapping, index, PAGEVEC_SIZE,
523 pvec.pages);
524 if (pvec.nr == 0)
525 return length;
526
527 if (length > 0 && pvec.pages[0]->index > index)
528 goto out;
529
530 b = pvec.pages[0]->index << (PAGE_SHIFT - inode->i_blkbits);
531 i = 0;
532 do {
533 page = pvec.pages[i];
534
535 lock_page(page);
536 if (page_has_buffers(page)) {
537 struct buffer_head *bh, *head;
538
539 bh = head = page_buffers(page);
540 do {
541 if (b < start_blk)
542 continue;
543 if (buffer_delay(bh)) {
544 if (length == 0)
545 *blkoff = b;
546 length++;
547 } else if (length > 0) {
548 goto out_locked;
549 }
550 } while (++b, bh = bh->b_this_page, bh != head);
551 } else {
552 if (length > 0)
553 goto out_locked;
554
555 b += nblocks_in_page;
556 }
557 unlock_page(page);
558
559 } while (++i < pagevec_count(&pvec));
560
561 index = page->index + 1;
562 pagevec_release(&pvec);
563 cond_resched();
564 goto repeat;
565
566 out_locked:
567 unlock_page(page);
568 out:
569 pagevec_release(&pvec);
570 return length;
571 }
572