projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: call SetPageUptodate if needed
[deliverable/linux.git] / fs / f2fs / checkpoint.c
1 /*
2 * fs/f2fs/checkpoint.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11 #include <linux/fs.h>
12 #include <linux/bio.h>
13 #include <linux/mpage.h>
14 #include <linux/writeback.h>
15 #include <linux/blkdev.h>
16 #include <linux/f2fs_fs.h>
17 #include <linux/pagevec.h>
18 #include <linux/swap.h>
19
20 #include "f2fs.h"
21 #include "node.h"
22 #include "segment.h"
23 #include "trace.h"
24 #include <trace/events/f2fs.h>
25
26 static struct kmem_cache *ino_entry_slab;
27 struct kmem_cache *inode_entry_slab;
28
29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
30 {
31 set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
32 sbi->sb->s_flags |= MS_RDONLY;
33 if (!end_io)
34 f2fs_flush_merged_bios(sbi);
35 }
36
37 /*
38 * We guarantee no failure on the returned page.
39 */
40 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
41 {
42 struct address_space *mapping = META_MAPPING(sbi);
43 struct page *page = NULL;
44 repeat:
45 page = f2fs_grab_cache_page(mapping, index, false);
46 if (!page) {
47 cond_resched();
48 goto repeat;
49 }
50 f2fs_wait_on_page_writeback(page, META, true);
51 if (!PageUptodate(page))
52 SetPageUptodate(page);
53 return page;
54 }
55
56 /*
57 * We guarantee no failure on the returned page.
58 */
59 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
60 bool is_meta)
61 {
62 struct address_space *mapping = META_MAPPING(sbi);
63 struct page *page;
64 struct f2fs_io_info fio = {
65 .sbi = sbi,
66 .type = META,
67 .rw = READ_SYNC | REQ_META | REQ_PRIO,
68 .old_blkaddr = index,
69 .new_blkaddr = index,
70 .encrypted_page = NULL,
71 };
72
73 if (unlikely(!is_meta))
74 fio.rw &= ~REQ_META;
75 repeat:
76 page = f2fs_grab_cache_page(mapping, index, false);
77 if (!page) {
78 cond_resched();
79 goto repeat;
80 }
81 if (PageUptodate(page))
82 goto out;
83
84 fio.page = page;
85
86 if (f2fs_submit_page_bio(&fio)) {
87 f2fs_put_page(page, 1);
88 goto repeat;
89 }
90
91 lock_page(page);
92 if (unlikely(page->mapping != mapping)) {
93 f2fs_put_page(page, 1);
94 goto repeat;
95 }
96
97 /*
98 * if there is any IO error when accessing device, make our filesystem
99 * readonly and make sure do not write checkpoint with non-uptodate
100 * meta page.
101 */
102 if (unlikely(!PageUptodate(page)))
103 f2fs_stop_checkpoint(sbi, false);
104 out:
105 return page;
106 }
107
108 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
109 {
110 return __get_meta_page(sbi, index, true);
111 }
112
113 /* for POR only */
114 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
115 {
116 return __get_meta_page(sbi, index, false);
117 }
118
119 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
120 {
121 switch (type) {
122 case META_NAT:
123 break;
124 case META_SIT:
125 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
126 return false;
127 break;
128 case META_SSA:
129 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
130 blkaddr < SM_I(sbi)->ssa_blkaddr))
131 return false;
132 break;
133 case META_CP:
134 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
135 blkaddr < __start_cp_addr(sbi)))
136 return false;
137 break;
138 case META_POR:
139 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
140 blkaddr < MAIN_BLKADDR(sbi)))
141 return false;
142 break;
143 default:
144 BUG();
145 }
146
147 return true;
148 }
149
150 /*
151 * Readahead CP/NAT/SIT/SSA pages
152 */
153 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
154 int type, bool sync)
155 {
156 struct page *page;
157 block_t blkno = start;
158 struct f2fs_io_info fio = {
159 .sbi = sbi,
160 .type = META,
161 .rw = sync ? (READ_SYNC | REQ_META | REQ_PRIO) : READA,
162 .encrypted_page = NULL,
163 };
164 struct blk_plug plug;
165
166 if (unlikely(type == META_POR))
167 fio.rw &= ~REQ_META;
168
169 blk_start_plug(&plug);
170 for (; nrpages-- > 0; blkno++) {
171
172 if (!is_valid_blkaddr(sbi, blkno, type))
173 goto out;
174
175 switch (type) {
176 case META_NAT:
177 if (unlikely(blkno >=
178 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
179 blkno = 0;
180 /* get nat block addr */
181 fio.new_blkaddr = current_nat_addr(sbi,
182 blkno * NAT_ENTRY_PER_BLOCK);
183 break;
184 case META_SIT:
185 /* get sit block addr */
186 fio.new_blkaddr = current_sit_addr(sbi,
187 blkno * SIT_ENTRY_PER_BLOCK);
188 break;
189 case META_SSA:
190 case META_CP:
191 case META_POR:
192 fio.new_blkaddr = blkno;
193 break;
194 default:
195 BUG();
196 }
197
198 page = f2fs_grab_cache_page(META_MAPPING(sbi),
199 fio.new_blkaddr, false);
200 if (!page)
201 continue;
202 if (PageUptodate(page)) {
203 f2fs_put_page(page, 1);
204 continue;
205 }
206
207 fio.page = page;
208 fio.old_blkaddr = fio.new_blkaddr;
209 f2fs_submit_page_mbio(&fio);
210 f2fs_put_page(page, 0);
211 }
212 out:
213 f2fs_submit_merged_bio(sbi, META, READ);
214 blk_finish_plug(&plug);
215 return blkno - start;
216 }
217
218 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
219 {
220 struct page *page;
221 bool readahead = false;
222
223 page = find_get_page(META_MAPPING(sbi), index);
224 if (!page || !PageUptodate(page))
225 readahead = true;
226 f2fs_put_page(page, 0);
227
228 if (readahead)
229 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
230 }
231
232 static int f2fs_write_meta_page(struct page *page,
233 struct writeback_control *wbc)
234 {
235 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
236
237 trace_f2fs_writepage(page, META);
238
239 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
240 goto redirty_out;
241 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
242 goto redirty_out;
243 if (unlikely(f2fs_cp_error(sbi)))
244 goto redirty_out;
245
246 write_meta_page(sbi, page);
247 dec_page_count(sbi, F2FS_DIRTY_META);
248
249 if (wbc->for_reclaim)
250 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
251
252 unlock_page(page);
253
254 if (unlikely(f2fs_cp_error(sbi)))
255 f2fs_submit_merged_bio(sbi, META, WRITE);
256
257 return 0;
258
259 redirty_out:
260 redirty_page_for_writepage(wbc, page);
261 return AOP_WRITEPAGE_ACTIVATE;
262 }
263
264 static int f2fs_write_meta_pages(struct address_space *mapping,
265 struct writeback_control *wbc)
266 {
267 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
268 long diff, written;
269
270 /* collect a number of dirty meta pages and write together */
271 if (wbc->for_kupdate ||
272 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
273 goto skip_write;
274
275 trace_f2fs_writepages(mapping->host, wbc, META);
276
277 /* if mounting is failed, skip writing node pages */
278 mutex_lock(&sbi->cp_mutex);
279 diff = nr_pages_to_write(sbi, META, wbc);
280 written = sync_meta_pages(sbi, META, wbc->nr_to_write);
281 mutex_unlock(&sbi->cp_mutex);
282 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
283 return 0;
284
285 skip_write:
286 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
287 trace_f2fs_writepages(mapping->host, wbc, META);
288 return 0;
289 }
290
291 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
292 long nr_to_write)
293 {
294 struct address_space *mapping = META_MAPPING(sbi);
295 pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
296 struct pagevec pvec;
297 long nwritten = 0;
298 struct writeback_control wbc = {
299 .for_reclaim = 0,
300 };
301 struct blk_plug plug;
302
303 pagevec_init(&pvec, 0);
304
305 blk_start_plug(&plug);
306
307 while (index <= end) {
308 int i, nr_pages;
309 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
310 PAGECACHE_TAG_DIRTY,
311 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
312 if (unlikely(nr_pages == 0))
313 break;
314
315 for (i = 0; i < nr_pages; i++) {
316 struct page *page = pvec.pages[i];
317
318 if (prev == ULONG_MAX)
319 prev = page->index - 1;
320 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
321 pagevec_release(&pvec);
322 goto stop;
323 }
324
325 lock_page(page);
326
327 if (unlikely(page->mapping != mapping)) {
328 continue_unlock:
329 unlock_page(page);
330 continue;
331 }
332 if (!PageDirty(page)) {
333 /* someone wrote it for us */
334 goto continue_unlock;
335 }
336
337 f2fs_wait_on_page_writeback(page, META, true);
338
339 BUG_ON(PageWriteback(page));
340 if (!clear_page_dirty_for_io(page))
341 goto continue_unlock;
342
343 if (mapping->a_ops->writepage(page, &wbc)) {
344 unlock_page(page);
345 break;
346 }
347 nwritten++;
348 prev = page->index;
349 if (unlikely(nwritten >= nr_to_write))
350 break;
351 }
352 pagevec_release(&pvec);
353 cond_resched();
354 }
355 stop:
356 if (nwritten)
357 f2fs_submit_merged_bio(sbi, type, WRITE);
358
359 blk_finish_plug(&plug);
360
361 return nwritten;
362 }
363
364 static int f2fs_set_meta_page_dirty(struct page *page)
365 {
366 trace_f2fs_set_page_dirty(page, META);
367
368 if (!PageUptodate(page))
369 SetPageUptodate(page);
370 if (!PageDirty(page)) {
371 f2fs_set_page_dirty_nobuffers(page);
372 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
373 SetPagePrivate(page);
374 f2fs_trace_pid(page);
375 return 1;
376 }
377 return 0;
378 }
379
380 const struct address_space_operations f2fs_meta_aops = {
381 .writepage = f2fs_write_meta_page,
382 .writepages = f2fs_write_meta_pages,
383 .set_page_dirty = f2fs_set_meta_page_dirty,
384 .invalidatepage = f2fs_invalidate_page,
385 .releasepage = f2fs_release_page,
386 };
387
388 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
389 {
390 struct inode_management *im = &sbi->im[type];
391 struct ino_entry *e, *tmp;
392
393 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
394 retry:
395 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
396
397 spin_lock(&im->ino_lock);
398 e = radix_tree_lookup(&im->ino_root, ino);
399 if (!e) {
400 e = tmp;
401 if (radix_tree_insert(&im->ino_root, ino, e)) {
402 spin_unlock(&im->ino_lock);
403 radix_tree_preload_end();
404 goto retry;
405 }
406 memset(e, 0, sizeof(struct ino_entry));
407 e->ino = ino;
408
409 list_add_tail(&e->list, &im->ino_list);
410 if (type != ORPHAN_INO)
411 im->ino_num++;
412 }
413 spin_unlock(&im->ino_lock);
414 radix_tree_preload_end();
415
416 if (e != tmp)
417 kmem_cache_free(ino_entry_slab, tmp);
418 }
419
420 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
421 {
422 struct inode_management *im = &sbi->im[type];
423 struct ino_entry *e;
424
425 spin_lock(&im->ino_lock);
426 e = radix_tree_lookup(&im->ino_root, ino);
427 if (e) {
428 list_del(&e->list);
429 radix_tree_delete(&im->ino_root, ino);
430 im->ino_num--;
431 spin_unlock(&im->ino_lock);
432 kmem_cache_free(ino_entry_slab, e);
433 return;
434 }
435 spin_unlock(&im->ino_lock);
436 }
437
438 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
439 {
440 /* add new dirty ino entry into list */
441 __add_ino_entry(sbi, ino, type);
442 }
443
444 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
445 {
446 /* remove dirty ino entry from list */
447 __remove_ino_entry(sbi, ino, type);
448 }
449
450 /* mode should be APPEND_INO or UPDATE_INO */
451 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
452 {
453 struct inode_management *im = &sbi->im[mode];
454 struct ino_entry *e;
455
456 spin_lock(&im->ino_lock);
457 e = radix_tree_lookup(&im->ino_root, ino);
458 spin_unlock(&im->ino_lock);
459 return e ? true : false;
460 }
461
462 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
463 {
464 struct ino_entry *e, *tmp;
465 int i;
466
467 for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
468 struct inode_management *im = &sbi->im[i];
469
470 spin_lock(&im->ino_lock);
471 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
472 list_del(&e->list);
473 radix_tree_delete(&im->ino_root, e->ino);
474 kmem_cache_free(ino_entry_slab, e);
475 im->ino_num--;
476 }
477 spin_unlock(&im->ino_lock);
478 }
479 }
480
481 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
482 {
483 struct inode_management *im = &sbi->im[ORPHAN_INO];
484 int err = 0;
485
486 spin_lock(&im->ino_lock);
487
488 #ifdef CONFIG_F2FS_FAULT_INJECTION
489 if (time_to_inject(FAULT_ORPHAN)) {
490 spin_unlock(&im->ino_lock);
491 return -ENOSPC;
492 }
493 #endif
494 if (unlikely(im->ino_num >= sbi->max_orphans))
495 err = -ENOSPC;
496 else
497 im->ino_num++;
498 spin_unlock(&im->ino_lock);
499
500 return err;
501 }
502
503 void release_orphan_inode(struct f2fs_sb_info *sbi)
504 {
505 struct inode_management *im = &sbi->im[ORPHAN_INO];
506
507 spin_lock(&im->ino_lock);
508 f2fs_bug_on(sbi, im->ino_num == 0);
509 im->ino_num--;
510 spin_unlock(&im->ino_lock);
511 }
512
513 void add_orphan_inode(struct inode *inode)
514 {
515 /* add new orphan ino entry into list */
516 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO);
517 update_inode_page(inode);
518 }
519
520 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
521 {
522 /* remove orphan entry from orphan list */
523 __remove_ino_entry(sbi, ino, ORPHAN_INO);
524 }
525
526 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
527 {
528 struct inode *inode;
529
530 inode = f2fs_iget(sbi->sb, ino);
531 if (IS_ERR(inode)) {
532 /*
533 * there should be a bug that we can't find the entry
534 * to orphan inode.
535 */
536 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
537 return PTR_ERR(inode);
538 }
539
540 clear_nlink(inode);
541
542 /* truncate all the data during iput */
543 iput(inode);
544 return 0;
545 }
546
547 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
548 {
549 block_t start_blk, orphan_blocks, i, j;
550 int err;
551
552 if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
553 return 0;
554
555 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
556 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
557
558 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
559
560 for (i = 0; i < orphan_blocks; i++) {
561 struct page *page = get_meta_page(sbi, start_blk + i);
562 struct f2fs_orphan_block *orphan_blk;
563
564 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
565 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
566 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
567 err = recover_orphan_inode(sbi, ino);
568 if (err) {
569 f2fs_put_page(page, 1);
570 return err;
571 }
572 }
573 f2fs_put_page(page, 1);
574 }
575 /* clear Orphan Flag */
576 clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
577 return 0;
578 }
579
580 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
581 {
582 struct list_head *head;
583 struct f2fs_orphan_block *orphan_blk = NULL;
584 unsigned int nentries = 0;
585 unsigned short index = 1;
586 unsigned short orphan_blocks;
587 struct page *page = NULL;
588 struct ino_entry *orphan = NULL;
589 struct inode_management *im = &sbi->im[ORPHAN_INO];
590
591 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
592
593 /*
594 * we don't need to do spin_lock(&im->ino_lock) here, since all the
595 * orphan inode operations are covered under f2fs_lock_op().
596 * And, spin_lock should be avoided due to page operations below.
597 */
598 head = &im->ino_list;
599
600 /* loop for each orphan inode entry and write them in Jornal block */
601 list_for_each_entry(orphan, head, list) {
602 if (!page) {
603 page = grab_meta_page(sbi, start_blk++);
604 orphan_blk =
605 (struct f2fs_orphan_block *)page_address(page);
606 memset(orphan_blk, 0, sizeof(*orphan_blk));
607 }
608
609 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
610
611 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
612 /*
613 * an orphan block is full of 1020 entries,
614 * then we need to flush current orphan blocks
615 * and bring another one in memory
616 */
617 orphan_blk->blk_addr = cpu_to_le16(index);
618 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
619 orphan_blk->entry_count = cpu_to_le32(nentries);
620 set_page_dirty(page);
621 f2fs_put_page(page, 1);
622 index++;
623 nentries = 0;
624 page = NULL;
625 }
626 }
627
628 if (page) {
629 orphan_blk->blk_addr = cpu_to_le16(index);
630 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
631 orphan_blk->entry_count = cpu_to_le32(nentries);
632 set_page_dirty(page);
633 f2fs_put_page(page, 1);
634 }
635 }
636
637 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
638 block_t cp_addr, unsigned long long *version)
639 {
640 struct page *cp_page_1, *cp_page_2 = NULL;
641 unsigned long blk_size = sbi->blocksize;
642 struct f2fs_checkpoint *cp_block;
643 unsigned long long cur_version = 0, pre_version = 0;
644 size_t crc_offset;
645 __u32 crc = 0;
646
647 /* Read the 1st cp block in this CP pack */
648 cp_page_1 = get_meta_page(sbi, cp_addr);
649
650 /* get the version number */
651 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
652 crc_offset = le32_to_cpu(cp_block->checksum_offset);
653 if (crc_offset >= blk_size)
654 goto invalid_cp1;
655
656 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
657 if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
658 goto invalid_cp1;
659
660 pre_version = cur_cp_version(cp_block);
661
662 /* Read the 2nd cp block in this CP pack */
663 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
664 cp_page_2 = get_meta_page(sbi, cp_addr);
665
666 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
667 crc_offset = le32_to_cpu(cp_block->checksum_offset);
668 if (crc_offset >= blk_size)
669 goto invalid_cp2;
670
671 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
672 if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
673 goto invalid_cp2;
674
675 cur_version = cur_cp_version(cp_block);
676
677 if (cur_version == pre_version) {
678 *version = cur_version;
679 f2fs_put_page(cp_page_2, 1);
680 return cp_page_1;
681 }
682 invalid_cp2:
683 f2fs_put_page(cp_page_2, 1);
684 invalid_cp1:
685 f2fs_put_page(cp_page_1, 1);
686 return NULL;
687 }
688
689 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
690 {
691 struct f2fs_checkpoint *cp_block;
692 struct f2fs_super_block *fsb = sbi->raw_super;
693 struct page *cp1, *cp2, *cur_page;
694 unsigned long blk_size = sbi->blocksize;
695 unsigned long long cp1_version = 0, cp2_version = 0;
696 unsigned long long cp_start_blk_no;
697 unsigned int cp_blks = 1 + __cp_payload(sbi);
698 block_t cp_blk_no;
699 int i;
700
701 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
702 if (!sbi->ckpt)
703 return -ENOMEM;
704 /*
705 * Finding out valid cp block involves read both
706 * sets( cp pack1 and cp pack 2)
707 */
708 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
709 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
710
711 /* The second checkpoint pack should start at the next segment */
712 cp_start_blk_no += ((unsigned long long)1) <<
713 le32_to_cpu(fsb->log_blocks_per_seg);
714 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
715
716 if (cp1 && cp2) {
717 if (ver_after(cp2_version, cp1_version))
718 cur_page = cp2;
719 else
720 cur_page = cp1;
721 } else if (cp1) {
722 cur_page = cp1;
723 } else if (cp2) {
724 cur_page = cp2;
725 } else {
726 goto fail_no_cp;
727 }
728
729 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
730 memcpy(sbi->ckpt, cp_block, blk_size);
731
732 /* Sanity checking of checkpoint */
733 if (sanity_check_ckpt(sbi))
734 goto fail_no_cp;
735
736 if (cp_blks <= 1)
737 goto done;
738
739 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
740 if (cur_page == cp2)
741 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
742
743 for (i = 1; i < cp_blks; i++) {
744 void *sit_bitmap_ptr;
745 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
746
747 cur_page = get_meta_page(sbi, cp_blk_no + i);
748 sit_bitmap_ptr = page_address(cur_page);
749 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
750 f2fs_put_page(cur_page, 1);
751 }
752 done:
753 f2fs_put_page(cp1, 1);
754 f2fs_put_page(cp2, 1);
755 return 0;
756
757 fail_no_cp:
758 kfree(sbi->ckpt);
759 return -EINVAL;
760 }
761
762 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
763 {
764 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
765 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
766
767 if (is_inode_flag_set(inode, flag))
768 return;
769
770 set_inode_flag(inode, flag);
771 list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
772 stat_inc_dirty_inode(sbi, type);
773 }
774
775 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
776 {
777 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
778
779 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
780 return;
781
782 list_del_init(&F2FS_I(inode)->dirty_list);
783 clear_inode_flag(inode, flag);
784 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
785 }
786
787 void update_dirty_page(struct inode *inode, struct page *page)
788 {
789 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
790 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
791
792 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
793 !S_ISLNK(inode->i_mode))
794 return;
795
796 spin_lock(&sbi->inode_lock[type]);
797 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
798 __add_dirty_inode(inode, type);
799 inode_inc_dirty_pages(inode);
800 spin_unlock(&sbi->inode_lock[type]);
801
802 SetPagePrivate(page);
803 f2fs_trace_pid(page);
804 }
805
806 void remove_dirty_inode(struct inode *inode)
807 {
808 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
809 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
810
811 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
812 !S_ISLNK(inode->i_mode))
813 return;
814
815 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
816 return;
817
818 spin_lock(&sbi->inode_lock[type]);
819 __remove_dirty_inode(inode, type);
820 spin_unlock(&sbi->inode_lock[type]);
821 }
822
823 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
824 {
825 struct list_head *head;
826 struct inode *inode;
827 struct f2fs_inode_info *fi;
828 bool is_dir = (type == DIR_INODE);
829
830 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
831 get_pages(sbi, is_dir ?
832 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
833 retry:
834 if (unlikely(f2fs_cp_error(sbi)))
835 return -EIO;
836
837 spin_lock(&sbi->inode_lock[type]);
838
839 head = &sbi->inode_list[type];
840 if (list_empty(head)) {
841 spin_unlock(&sbi->inode_lock[type]);
842 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
843 get_pages(sbi, is_dir ?
844 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
845 return 0;
846 }
847 fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
848 inode = igrab(&fi->vfs_inode);
849 spin_unlock(&sbi->inode_lock[type]);
850 if (inode) {
851 filemap_fdatawrite(inode->i_mapping);
852 iput(inode);
853 } else {
854 /*
855 * We should submit bio, since it exists several
856 * wribacking dentry pages in the freeing inode.
857 */
858 f2fs_submit_merged_bio(sbi, DATA, WRITE);
859 cond_resched();
860 }
861 goto retry;
862 }
863
864 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
865 {
866 struct list_head *head = &sbi->inode_list[DIRTY_META];
867 struct inode *inode;
868 struct f2fs_inode_info *fi;
869 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
870
871 while (total--) {
872 if (unlikely(f2fs_cp_error(sbi)))
873 return -EIO;
874
875 spin_lock(&sbi->inode_lock[DIRTY_META]);
876 if (list_empty(head)) {
877 spin_unlock(&sbi->inode_lock[DIRTY_META]);
878 return 0;
879 }
880 fi = list_entry(head->next, struct f2fs_inode_info,
881 gdirty_list);
882 inode = igrab(&fi->vfs_inode);
883 spin_unlock(&sbi->inode_lock[DIRTY_META]);
884 if (inode) {
885 update_inode_page(inode);
886 iput(inode);
887 }
888 };
889 return 0;
890 }
891
892 /*
893 * Freeze all the FS-operations for checkpoint.
894 */
895 static int block_operations(struct f2fs_sb_info *sbi)
896 {
897 struct writeback_control wbc = {
898 .sync_mode = WB_SYNC_ALL,
899 .nr_to_write = LONG_MAX,
900 .for_reclaim = 0,
901 };
902 int err = 0;
903
904 retry_flush_dents:
905 f2fs_lock_all(sbi);
906 /* write all the dirty dentry pages */
907 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
908 f2fs_unlock_all(sbi);
909 err = sync_dirty_inodes(sbi, DIR_INODE);
910 if (err)
911 goto out;
912 goto retry_flush_dents;
913 }
914
915 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
916 f2fs_unlock_all(sbi);
917 err = f2fs_sync_inode_meta(sbi);
918 if (err)
919 goto out;
920 goto retry_flush_dents;
921 }
922
923 /*
924 * POR: we should ensure that there are no dirty node pages
925 * until finishing nat/sit flush.
926 */
927 retry_flush_nodes:
928 down_write(&sbi->node_write);
929
930 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
931 up_write(&sbi->node_write);
932 err = sync_node_pages(sbi, &wbc);
933 if (err) {
934 f2fs_unlock_all(sbi);
935 goto out;
936 }
937 goto retry_flush_nodes;
938 }
939 out:
940 return err;
941 }
942
943 static void unblock_operations(struct f2fs_sb_info *sbi)
944 {
945 up_write(&sbi->node_write);
946
947 build_free_nids(sbi);
948 f2fs_unlock_all(sbi);
949 }
950
951 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
952 {
953 DEFINE_WAIT(wait);
954
955 for (;;) {
956 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
957
958 if (!atomic_read(&sbi->nr_wb_bios))
959 break;
960
961 io_schedule_timeout(5*HZ);
962 }
963 finish_wait(&sbi->cp_wait, &wait);
964 }
965
966 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
967 {
968 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
969 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
970 struct f2fs_nm_info *nm_i = NM_I(sbi);
971 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
972 nid_t last_nid = nm_i->next_scan_nid;
973 block_t start_blk;
974 unsigned int data_sum_blocks, orphan_blocks;
975 __u32 crc32 = 0;
976 int i;
977 int cp_payload_blks = __cp_payload(sbi);
978 block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
979 bool invalidate = false;
980 struct super_block *sb = sbi->sb;
981 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
982 u64 kbytes_written;
983
984 /*
985 * This avoids to conduct wrong roll-forward operations and uses
986 * metapages, so should be called prior to sync_meta_pages below.
987 */
988 if (!test_opt(sbi, LFS) && discard_next_dnode(sbi, discard_blk))
989 invalidate = true;
990
991 /* Flush all the NAT/SIT pages */
992 while (get_pages(sbi, F2FS_DIRTY_META)) {
993 sync_meta_pages(sbi, META, LONG_MAX);
994 if (unlikely(f2fs_cp_error(sbi)))
995 return -EIO;
996 }
997
998 next_free_nid(sbi, &last_nid);
999
1000 /*
1001 * modify checkpoint
1002 * version number is already updated
1003 */
1004 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1005 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1006 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1007 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1008 ckpt->cur_node_segno[i] =
1009 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1010 ckpt->cur_node_blkoff[i] =
1011 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1012 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1013 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1014 }
1015 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1016 ckpt->cur_data_segno[i] =
1017 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1018 ckpt->cur_data_blkoff[i] =
1019 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1020 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1021 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1022 }
1023
1024 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1025 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1026 ckpt->next_free_nid = cpu_to_le32(last_nid);
1027
1028 /* 2 cp + n data seg summary + orphan inode blocks */
1029 data_sum_blocks = npages_for_summary_flush(sbi, false);
1030 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1031 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1032 else
1033 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1034
1035 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1036 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1037 orphan_blocks);
1038
1039 if (__remain_node_summaries(cpc->reason))
1040 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1041 cp_payload_blks + data_sum_blocks +
1042 orphan_blocks + NR_CURSEG_NODE_TYPE);
1043 else
1044 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1045 cp_payload_blks + data_sum_blocks +
1046 orphan_blocks);
1047
1048 if (cpc->reason == CP_UMOUNT)
1049 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1050 else
1051 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1052
1053 if (cpc->reason == CP_FASTBOOT)
1054 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1055 else
1056 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1057
1058 if (orphan_num)
1059 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1060 else
1061 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1062
1063 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1064 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1065
1066 /* update SIT/NAT bitmap */
1067 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1068 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1069
1070 crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1071 *((__le32 *)((unsigned char *)ckpt +
1072 le32_to_cpu(ckpt->checksum_offset)))
1073 = cpu_to_le32(crc32);
1074
1075 start_blk = __start_cp_addr(sbi);
1076
1077 /* need to wait for end_io results */
1078 wait_on_all_pages_writeback(sbi);
1079 if (unlikely(f2fs_cp_error(sbi)))
1080 return -EIO;
1081
1082 /* write out checkpoint buffer at block 0 */
1083 update_meta_page(sbi, ckpt, start_blk++);
1084
1085 for (i = 1; i < 1 + cp_payload_blks; i++)
1086 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1087 start_blk++);
1088
1089 if (orphan_num) {
1090 write_orphan_inodes(sbi, start_blk);
1091 start_blk += orphan_blocks;
1092 }
1093
1094 write_data_summaries(sbi, start_blk);
1095 start_blk += data_sum_blocks;
1096
1097 /* Record write statistics in the hot node summary */
1098 kbytes_written = sbi->kbytes_written;
1099 if (sb->s_bdev->bd_part)
1100 kbytes_written += BD_PART_WRITTEN(sbi);
1101
1102 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1103
1104 if (__remain_node_summaries(cpc->reason)) {
1105 write_node_summaries(sbi, start_blk);
1106 start_blk += NR_CURSEG_NODE_TYPE;
1107 }
1108
1109 /* writeout checkpoint block */
1110 update_meta_page(sbi, ckpt, start_blk);
1111
1112 /* wait for previous submitted node/meta pages writeback */
1113 wait_on_all_pages_writeback(sbi);
1114
1115 if (unlikely(f2fs_cp_error(sbi)))
1116 return -EIO;
1117
1118 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1119 filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1120
1121 /* update user_block_counts */
1122 sbi->last_valid_block_count = sbi->total_valid_block_count;
1123 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1124
1125 /* Here, we only have one bio having CP pack */
1126 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1127
1128 /* wait for previous submitted meta pages writeback */
1129 wait_on_all_pages_writeback(sbi);
1130
1131 /*
1132 * invalidate meta page which is used temporarily for zeroing out
1133 * block at the end of warm node chain.
1134 */
1135 if (invalidate)
1136 invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
1137 discard_blk);
1138
1139 release_ino_entry(sbi, false);
1140
1141 if (unlikely(f2fs_cp_error(sbi)))
1142 return -EIO;
1143
1144 clear_prefree_segments(sbi, cpc);
1145 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1146
1147 return 0;
1148 }
1149
1150 /*
1151 * We guarantee that this checkpoint procedure will not fail.
1152 */
1153 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1154 {
1155 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1156 unsigned long long ckpt_ver;
1157 int err = 0;
1158
1159 mutex_lock(&sbi->cp_mutex);
1160
1161 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1162 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1163 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1164 goto out;
1165 if (unlikely(f2fs_cp_error(sbi))) {
1166 err = -EIO;
1167 goto out;
1168 }
1169 if (f2fs_readonly(sbi->sb)) {
1170 err = -EROFS;
1171 goto out;
1172 }
1173
1174 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1175
1176 err = block_operations(sbi);
1177 if (err)
1178 goto out;
1179
1180 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1181
1182 f2fs_flush_merged_bios(sbi);
1183
1184 /*
1185 * update checkpoint pack index
1186 * Increase the version number so that
1187 * SIT entries and seg summaries are written at correct place
1188 */
1189 ckpt_ver = cur_cp_version(ckpt);
1190 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1191
1192 /* write cached NAT/SIT entries to NAT/SIT area */
1193 flush_nat_entries(sbi);
1194 flush_sit_entries(sbi, cpc);
1195
1196 /* unlock all the fs_lock[] in do_checkpoint() */
1197 err = do_checkpoint(sbi, cpc);
1198
1199 unblock_operations(sbi);
1200 stat_inc_cp_count(sbi->stat_info);
1201
1202 if (cpc->reason == CP_RECOVERY)
1203 f2fs_msg(sbi->sb, KERN_NOTICE,
1204 "checkpoint: version = %llx", ckpt_ver);
1205
1206 /* do checkpoint periodically */
1207 f2fs_update_time(sbi, CP_TIME);
1208 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1209 out:
1210 mutex_unlock(&sbi->cp_mutex);
1211 return err;
1212 }
1213
1214 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1215 {
1216 int i;
1217
1218 for (i = 0; i < MAX_INO_ENTRY; i++) {
1219 struct inode_management *im = &sbi->im[i];
1220
1221 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1222 spin_lock_init(&im->ino_lock);
1223 INIT_LIST_HEAD(&im->ino_list);
1224 im->ino_num = 0;
1225 }
1226
1227 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1228 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1229 F2FS_ORPHANS_PER_BLOCK;
1230 }
1231
1232 int __init create_checkpoint_caches(void)
1233 {
1234 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1235 sizeof(struct ino_entry));
1236 if (!ino_entry_slab)
1237 return -ENOMEM;
1238 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1239 sizeof(struct inode_entry));
1240 if (!inode_entry_slab) {
1241 kmem_cache_destroy(ino_entry_slab);
1242 return -ENOMEM;
1243 }
1244 return 0;
1245 }
1246
1247 void destroy_checkpoint_caches(void)
1248 {
1249 kmem_cache_destroy(ino_entry_slab);
1250 kmem_cache_destroy(inode_entry_slab);
1251 }
Linux kernel - Deliverables
This page took 0.082552 seconds and 5 git commands to generate.