projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: inject ENOSPC failures
[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 /*
30 * We guarantee no failure on the returned page.
31 */
32 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
33 {
34 struct address_space *mapping = META_MAPPING(sbi);
35 struct page *page = NULL;
36 repeat:
37 page = f2fs_grab_cache_page(mapping, index, false);
38 if (!page) {
39 cond_resched();
40 goto repeat;
41 }
42 f2fs_wait_on_page_writeback(page, META, true);
43 SetPageUptodate(page);
44 return page;
45 }
46
47 /*
48 * We guarantee no failure on the returned page.
49 */
50 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
51 bool is_meta)
52 {
53 struct address_space *mapping = META_MAPPING(sbi);
54 struct page *page;
55 struct f2fs_io_info fio = {
56 .sbi = sbi,
57 .type = META,
58 .rw = READ_SYNC | REQ_META | REQ_PRIO,
59 .old_blkaddr = index,
60 .new_blkaddr = index,
61 .encrypted_page = NULL,
62 };
63
64 if (unlikely(!is_meta))
65 fio.rw &= ~REQ_META;
66 repeat:
67 page = f2fs_grab_cache_page(mapping, index, false);
68 if (!page) {
69 cond_resched();
70 goto repeat;
71 }
72 if (PageUptodate(page))
73 goto out;
74
75 fio.page = page;
76
77 if (f2fs_submit_page_bio(&fio)) {
78 f2fs_put_page(page, 1);
79 goto repeat;
80 }
81
82 lock_page(page);
83 if (unlikely(page->mapping != mapping)) {
84 f2fs_put_page(page, 1);
85 goto repeat;
86 }
87
88 /*
89 * if there is any IO error when accessing device, make our filesystem
90 * readonly and make sure do not write checkpoint with non-uptodate
91 * meta page.
92 */
93 if (unlikely(!PageUptodate(page)))
94 f2fs_stop_checkpoint(sbi);
95 out:
96 return page;
97 }
98
99 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
100 {
101 return __get_meta_page(sbi, index, true);
102 }
103
104 /* for POR only */
105 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
106 {
107 return __get_meta_page(sbi, index, false);
108 }
109
110 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
111 {
112 switch (type) {
113 case META_NAT:
114 break;
115 case META_SIT:
116 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
117 return false;
118 break;
119 case META_SSA:
120 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
121 blkaddr < SM_I(sbi)->ssa_blkaddr))
122 return false;
123 break;
124 case META_CP:
125 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
126 blkaddr < __start_cp_addr(sbi)))
127 return false;
128 break;
129 case META_POR:
130 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
131 blkaddr < MAIN_BLKADDR(sbi)))
132 return false;
133 break;
134 default:
135 BUG();
136 }
137
138 return true;
139 }
140
141 /*
142 * Readahead CP/NAT/SIT/SSA pages
143 */
144 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
145 int type, bool sync)
146 {
147 struct page *page;
148 block_t blkno = start;
149 struct f2fs_io_info fio = {
150 .sbi = sbi,
151 .type = META,
152 .rw = sync ? (READ_SYNC | REQ_META | REQ_PRIO) : READA,
153 .encrypted_page = NULL,
154 };
155 struct blk_plug plug;
156
157 if (unlikely(type == META_POR))
158 fio.rw &= ~REQ_META;
159
160 blk_start_plug(&plug);
161 for (; nrpages-- > 0; blkno++) {
162
163 if (!is_valid_blkaddr(sbi, blkno, type))
164 goto out;
165
166 switch (type) {
167 case META_NAT:
168 if (unlikely(blkno >=
169 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
170 blkno = 0;
171 /* get nat block addr */
172 fio.new_blkaddr = current_nat_addr(sbi,
173 blkno * NAT_ENTRY_PER_BLOCK);
174 break;
175 case META_SIT:
176 /* get sit block addr */
177 fio.new_blkaddr = current_sit_addr(sbi,
178 blkno * SIT_ENTRY_PER_BLOCK);
179 break;
180 case META_SSA:
181 case META_CP:
182 case META_POR:
183 fio.new_blkaddr = blkno;
184 break;
185 default:
186 BUG();
187 }
188
189 page = f2fs_grab_cache_page(META_MAPPING(sbi),
190 fio.new_blkaddr, false);
191 if (!page)
192 continue;
193 if (PageUptodate(page)) {
194 f2fs_put_page(page, 1);
195 continue;
196 }
197
198 fio.page = page;
199 fio.old_blkaddr = fio.new_blkaddr;
200 f2fs_submit_page_mbio(&fio);
201 f2fs_put_page(page, 0);
202 }
203 out:
204 f2fs_submit_merged_bio(sbi, META, READ);
205 blk_finish_plug(&plug);
206 return blkno - start;
207 }
208
209 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
210 {
211 struct page *page;
212 bool readahead = false;
213
214 page = find_get_page(META_MAPPING(sbi), index);
215 if (!page || !PageUptodate(page))
216 readahead = true;
217 f2fs_put_page(page, 0);
218
219 if (readahead)
220 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
221 }
222
223 static int f2fs_write_meta_page(struct page *page,
224 struct writeback_control *wbc)
225 {
226 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
227
228 trace_f2fs_writepage(page, META);
229
230 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
231 goto redirty_out;
232 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
233 goto redirty_out;
234 if (unlikely(f2fs_cp_error(sbi)))
235 goto redirty_out;
236
237 write_meta_page(sbi, page);
238 dec_page_count(sbi, F2FS_DIRTY_META);
239
240 if (wbc->for_reclaim)
241 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
242
243 unlock_page(page);
244
245 if (unlikely(f2fs_cp_error(sbi)))
246 f2fs_submit_merged_bio(sbi, META, WRITE);
247
248 return 0;
249
250 redirty_out:
251 redirty_page_for_writepage(wbc, page);
252 return AOP_WRITEPAGE_ACTIVATE;
253 }
254
255 static int f2fs_write_meta_pages(struct address_space *mapping,
256 struct writeback_control *wbc)
257 {
258 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
259 long diff, written;
260
261 /* collect a number of dirty meta pages and write together */
262 if (wbc->for_kupdate ||
263 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
264 goto skip_write;
265
266 trace_f2fs_writepages(mapping->host, wbc, META);
267
268 /* if mounting is failed, skip writing node pages */
269 mutex_lock(&sbi->cp_mutex);
270 diff = nr_pages_to_write(sbi, META, wbc);
271 written = sync_meta_pages(sbi, META, wbc->nr_to_write);
272 mutex_unlock(&sbi->cp_mutex);
273 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
274 return 0;
275
276 skip_write:
277 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
278 trace_f2fs_writepages(mapping->host, wbc, META);
279 return 0;
280 }
281
282 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
283 long nr_to_write)
284 {
285 struct address_space *mapping = META_MAPPING(sbi);
286 pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
287 struct pagevec pvec;
288 long nwritten = 0;
289 struct writeback_control wbc = {
290 .for_reclaim = 0,
291 };
292 struct blk_plug plug;
293
294 pagevec_init(&pvec, 0);
295
296 blk_start_plug(&plug);
297
298 while (index <= end) {
299 int i, nr_pages;
300 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
301 PAGECACHE_TAG_DIRTY,
302 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
303 if (unlikely(nr_pages == 0))
304 break;
305
306 for (i = 0; i < nr_pages; i++) {
307 struct page *page = pvec.pages[i];
308
309 if (prev == ULONG_MAX)
310 prev = page->index - 1;
311 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
312 pagevec_release(&pvec);
313 goto stop;
314 }
315
316 lock_page(page);
317
318 if (unlikely(page->mapping != mapping)) {
319 continue_unlock:
320 unlock_page(page);
321 continue;
322 }
323 if (!PageDirty(page)) {
324 /* someone wrote it for us */
325 goto continue_unlock;
326 }
327
328 f2fs_wait_on_page_writeback(page, META, true);
329
330 BUG_ON(PageWriteback(page));
331 if (!clear_page_dirty_for_io(page))
332 goto continue_unlock;
333
334 if (mapping->a_ops->writepage(page, &wbc)) {
335 unlock_page(page);
336 break;
337 }
338 nwritten++;
339 prev = page->index;
340 if (unlikely(nwritten >= nr_to_write))
341 break;
342 }
343 pagevec_release(&pvec);
344 cond_resched();
345 }
346 stop:
347 if (nwritten)
348 f2fs_submit_merged_bio(sbi, type, WRITE);
349
350 blk_finish_plug(&plug);
351
352 return nwritten;
353 }
354
355 static int f2fs_set_meta_page_dirty(struct page *page)
356 {
357 trace_f2fs_set_page_dirty(page, META);
358
359 SetPageUptodate(page);
360 if (!PageDirty(page)) {
361 __set_page_dirty_nobuffers(page);
362 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
363 SetPagePrivate(page);
364 f2fs_trace_pid(page);
365 return 1;
366 }
367 return 0;
368 }
369
370 const struct address_space_operations f2fs_meta_aops = {
371 .writepage = f2fs_write_meta_page,
372 .writepages = f2fs_write_meta_pages,
373 .set_page_dirty = f2fs_set_meta_page_dirty,
374 .invalidatepage = f2fs_invalidate_page,
375 .releasepage = f2fs_release_page,
376 };
377
378 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
379 {
380 struct inode_management *im = &sbi->im[type];
381 struct ino_entry *e, *tmp;
382
383 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
384 retry:
385 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
386
387 spin_lock(&im->ino_lock);
388 e = radix_tree_lookup(&im->ino_root, ino);
389 if (!e) {
390 e = tmp;
391 if (radix_tree_insert(&im->ino_root, ino, e)) {
392 spin_unlock(&im->ino_lock);
393 radix_tree_preload_end();
394 goto retry;
395 }
396 memset(e, 0, sizeof(struct ino_entry));
397 e->ino = ino;
398
399 list_add_tail(&e->list, &im->ino_list);
400 if (type != ORPHAN_INO)
401 im->ino_num++;
402 }
403 spin_unlock(&im->ino_lock);
404 radix_tree_preload_end();
405
406 if (e != tmp)
407 kmem_cache_free(ino_entry_slab, tmp);
408 }
409
410 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
411 {
412 struct inode_management *im = &sbi->im[type];
413 struct ino_entry *e;
414
415 spin_lock(&im->ino_lock);
416 e = radix_tree_lookup(&im->ino_root, ino);
417 if (e) {
418 list_del(&e->list);
419 radix_tree_delete(&im->ino_root, ino);
420 im->ino_num--;
421 spin_unlock(&im->ino_lock);
422 kmem_cache_free(ino_entry_slab, e);
423 return;
424 }
425 spin_unlock(&im->ino_lock);
426 }
427
428 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
429 {
430 /* add new dirty ino entry into list */
431 __add_ino_entry(sbi, ino, type);
432 }
433
434 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
435 {
436 /* remove dirty ino entry from list */
437 __remove_ino_entry(sbi, ino, type);
438 }
439
440 /* mode should be APPEND_INO or UPDATE_INO */
441 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
442 {
443 struct inode_management *im = &sbi->im[mode];
444 struct ino_entry *e;
445
446 spin_lock(&im->ino_lock);
447 e = radix_tree_lookup(&im->ino_root, ino);
448 spin_unlock(&im->ino_lock);
449 return e ? true : false;
450 }
451
452 void release_ino_entry(struct f2fs_sb_info *sbi)
453 {
454 struct ino_entry *e, *tmp;
455 int i;
456
457 for (i = APPEND_INO; i <= UPDATE_INO; i++) {
458 struct inode_management *im = &sbi->im[i];
459
460 spin_lock(&im->ino_lock);
461 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
462 list_del(&e->list);
463 radix_tree_delete(&im->ino_root, e->ino);
464 kmem_cache_free(ino_entry_slab, e);
465 im->ino_num--;
466 }
467 spin_unlock(&im->ino_lock);
468 }
469 }
470
471 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
472 {
473 struct inode_management *im = &sbi->im[ORPHAN_INO];
474 int err = 0;
475
476 spin_lock(&im->ino_lock);
477
478 #ifdef CONFIG_F2FS_FAULT_INJECTION
479 if (time_to_inject(FAULT_ORPHAN)) {
480 spin_unlock(&im->ino_lock);
481 return -ENOSPC;
482 }
483 #endif
484 if (unlikely(im->ino_num >= sbi->max_orphans))
485 err = -ENOSPC;
486 else
487 im->ino_num++;
488 spin_unlock(&im->ino_lock);
489
490 return err;
491 }
492
493 void release_orphan_inode(struct f2fs_sb_info *sbi)
494 {
495 struct inode_management *im = &sbi->im[ORPHAN_INO];
496
497 spin_lock(&im->ino_lock);
498 f2fs_bug_on(sbi, im->ino_num == 0);
499 im->ino_num--;
500 spin_unlock(&im->ino_lock);
501 }
502
503 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
504 {
505 /* add new orphan ino entry into list */
506 __add_ino_entry(sbi, ino, ORPHAN_INO);
507 }
508
509 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
510 {
511 /* remove orphan entry from orphan list */
512 __remove_ino_entry(sbi, ino, ORPHAN_INO);
513 }
514
515 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
516 {
517 struct inode *inode;
518
519 inode = f2fs_iget(sbi->sb, ino);
520 if (IS_ERR(inode)) {
521 /*
522 * there should be a bug that we can't find the entry
523 * to orphan inode.
524 */
525 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
526 return PTR_ERR(inode);
527 }
528
529 clear_nlink(inode);
530
531 /* truncate all the data during iput */
532 iput(inode);
533 return 0;
534 }
535
536 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
537 {
538 block_t start_blk, orphan_blocks, i, j;
539 int err;
540
541 if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
542 return 0;
543
544 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
545 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
546
547 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
548
549 for (i = 0; i < orphan_blocks; i++) {
550 struct page *page = get_meta_page(sbi, start_blk + i);
551 struct f2fs_orphan_block *orphan_blk;
552
553 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
554 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
555 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
556 err = recover_orphan_inode(sbi, ino);
557 if (err) {
558 f2fs_put_page(page, 1);
559 return err;
560 }
561 }
562 f2fs_put_page(page, 1);
563 }
564 /* clear Orphan Flag */
565 clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
566 return 0;
567 }
568
569 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
570 {
571 struct list_head *head;
572 struct f2fs_orphan_block *orphan_blk = NULL;
573 unsigned int nentries = 0;
574 unsigned short index = 1;
575 unsigned short orphan_blocks;
576 struct page *page = NULL;
577 struct ino_entry *orphan = NULL;
578 struct inode_management *im = &sbi->im[ORPHAN_INO];
579
580 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
581
582 /*
583 * we don't need to do spin_lock(&im->ino_lock) here, since all the
584 * orphan inode operations are covered under f2fs_lock_op().
585 * And, spin_lock should be avoided due to page operations below.
586 */
587 head = &im->ino_list;
588
589 /* loop for each orphan inode entry and write them in Jornal block */
590 list_for_each_entry(orphan, head, list) {
591 if (!page) {
592 page = grab_meta_page(sbi, start_blk++);
593 orphan_blk =
594 (struct f2fs_orphan_block *)page_address(page);
595 memset(orphan_blk, 0, sizeof(*orphan_blk));
596 }
597
598 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
599
600 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
601 /*
602 * an orphan block is full of 1020 entries,
603 * then we need to flush current orphan blocks
604 * and bring another one in memory
605 */
606 orphan_blk->blk_addr = cpu_to_le16(index);
607 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
608 orphan_blk->entry_count = cpu_to_le32(nentries);
609 set_page_dirty(page);
610 f2fs_put_page(page, 1);
611 index++;
612 nentries = 0;
613 page = NULL;
614 }
615 }
616
617 if (page) {
618 orphan_blk->blk_addr = cpu_to_le16(index);
619 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
620 orphan_blk->entry_count = cpu_to_le32(nentries);
621 set_page_dirty(page);
622 f2fs_put_page(page, 1);
623 }
624 }
625
626 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
627 block_t cp_addr, unsigned long long *version)
628 {
629 struct page *cp_page_1, *cp_page_2 = NULL;
630 unsigned long blk_size = sbi->blocksize;
631 struct f2fs_checkpoint *cp_block;
632 unsigned long long cur_version = 0, pre_version = 0;
633 size_t crc_offset;
634 __u32 crc = 0;
635
636 /* Read the 1st cp block in this CP pack */
637 cp_page_1 = get_meta_page(sbi, cp_addr);
638
639 /* get the version number */
640 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
641 crc_offset = le32_to_cpu(cp_block->checksum_offset);
642 if (crc_offset >= blk_size)
643 goto invalid_cp1;
644
645 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
646 if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
647 goto invalid_cp1;
648
649 pre_version = cur_cp_version(cp_block);
650
651 /* Read the 2nd cp block in this CP pack */
652 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
653 cp_page_2 = get_meta_page(sbi, cp_addr);
654
655 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
656 crc_offset = le32_to_cpu(cp_block->checksum_offset);
657 if (crc_offset >= blk_size)
658 goto invalid_cp2;
659
660 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
661 if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
662 goto invalid_cp2;
663
664 cur_version = cur_cp_version(cp_block);
665
666 if (cur_version == pre_version) {
667 *version = cur_version;
668 f2fs_put_page(cp_page_2, 1);
669 return cp_page_1;
670 }
671 invalid_cp2:
672 f2fs_put_page(cp_page_2, 1);
673 invalid_cp1:
674 f2fs_put_page(cp_page_1, 1);
675 return NULL;
676 }
677
678 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
679 {
680 struct f2fs_checkpoint *cp_block;
681 struct f2fs_super_block *fsb = sbi->raw_super;
682 struct page *cp1, *cp2, *cur_page;
683 unsigned long blk_size = sbi->blocksize;
684 unsigned long long cp1_version = 0, cp2_version = 0;
685 unsigned long long cp_start_blk_no;
686 unsigned int cp_blks = 1 + __cp_payload(sbi);
687 block_t cp_blk_no;
688 int i;
689
690 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
691 if (!sbi->ckpt)
692 return -ENOMEM;
693 /*
694 * Finding out valid cp block involves read both
695 * sets( cp pack1 and cp pack 2)
696 */
697 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
698 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
699
700 /* The second checkpoint pack should start at the next segment */
701 cp_start_blk_no += ((unsigned long long)1) <<
702 le32_to_cpu(fsb->log_blocks_per_seg);
703 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
704
705 if (cp1 && cp2) {
706 if (ver_after(cp2_version, cp1_version))
707 cur_page = cp2;
708 else
709 cur_page = cp1;
710 } else if (cp1) {
711 cur_page = cp1;
712 } else if (cp2) {
713 cur_page = cp2;
714 } else {
715 goto fail_no_cp;
716 }
717
718 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
719 memcpy(sbi->ckpt, cp_block, blk_size);
720
721 /* Sanity checking of checkpoint */
722 if (sanity_check_ckpt(sbi))
723 goto fail_no_cp;
724
725 if (cp_blks <= 1)
726 goto done;
727
728 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
729 if (cur_page == cp2)
730 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
731
732 for (i = 1; i < cp_blks; i++) {
733 void *sit_bitmap_ptr;
734 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
735
736 cur_page = get_meta_page(sbi, cp_blk_no + i);
737 sit_bitmap_ptr = page_address(cur_page);
738 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
739 f2fs_put_page(cur_page, 1);
740 }
741 done:
742 f2fs_put_page(cp1, 1);
743 f2fs_put_page(cp2, 1);
744 return 0;
745
746 fail_no_cp:
747 kfree(sbi->ckpt);
748 return -EINVAL;
749 }
750
751 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
752 {
753 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
754 struct f2fs_inode_info *fi = F2FS_I(inode);
755 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
756
757 if (is_inode_flag_set(fi, flag))
758 return;
759
760 set_inode_flag(fi, flag);
761 list_add_tail(&fi->dirty_list, &sbi->inode_list[type]);
762 stat_inc_dirty_inode(sbi, type);
763 }
764
765 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
766 {
767 struct f2fs_inode_info *fi = F2FS_I(inode);
768 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
769
770 if (get_dirty_pages(inode) ||
771 !is_inode_flag_set(F2FS_I(inode), flag))
772 return;
773
774 list_del_init(&fi->dirty_list);
775 clear_inode_flag(fi, flag);
776 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
777 }
778
779 void update_dirty_page(struct inode *inode, struct page *page)
780 {
781 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
782 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
783
784 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
785 !S_ISLNK(inode->i_mode))
786 return;
787
788 spin_lock(&sbi->inode_lock[type]);
789 __add_dirty_inode(inode, type);
790 inode_inc_dirty_pages(inode);
791 spin_unlock(&sbi->inode_lock[type]);
792
793 SetPagePrivate(page);
794 f2fs_trace_pid(page);
795 }
796
797 void add_dirty_dir_inode(struct inode *inode)
798 {
799 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
800
801 spin_lock(&sbi->inode_lock[DIR_INODE]);
802 __add_dirty_inode(inode, DIR_INODE);
803 spin_unlock(&sbi->inode_lock[DIR_INODE]);
804 }
805
806 void remove_dirty_inode(struct inode *inode)
807 {
808 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
809 struct f2fs_inode_info *fi = F2FS_I(inode);
810 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
811
812 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
813 !S_ISLNK(inode->i_mode))
814 return;
815
816 spin_lock(&sbi->inode_lock[type]);
817 __remove_dirty_inode(inode, type);
818 spin_unlock(&sbi->inode_lock[type]);
819
820 /* Only from the recovery routine */
821 if (is_inode_flag_set(fi, FI_DELAY_IPUT)) {
822 clear_inode_flag(fi, FI_DELAY_IPUT);
823 iput(inode);
824 }
825 }
826
827 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
828 {
829 struct list_head *head;
830 struct inode *inode;
831 struct f2fs_inode_info *fi;
832 bool is_dir = (type == DIR_INODE);
833
834 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
835 get_pages(sbi, is_dir ?
836 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
837 retry:
838 if (unlikely(f2fs_cp_error(sbi)))
839 return -EIO;
840
841 spin_lock(&sbi->inode_lock[type]);
842
843 head = &sbi->inode_list[type];
844 if (list_empty(head)) {
845 spin_unlock(&sbi->inode_lock[type]);
846 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
847 get_pages(sbi, is_dir ?
848 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
849 return 0;
850 }
851 fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
852 inode = igrab(&fi->vfs_inode);
853 spin_unlock(&sbi->inode_lock[type]);
854 if (inode) {
855 filemap_fdatawrite(inode->i_mapping);
856 iput(inode);
857 } else {
858 /*
859 * We should submit bio, since it exists several
860 * wribacking dentry pages in the freeing inode.
861 */
862 f2fs_submit_merged_bio(sbi, DATA, WRITE);
863 cond_resched();
864 }
865 goto retry;
866 }
867
868 /*
869 * Freeze all the FS-operations for checkpoint.
870 */
871 static int block_operations(struct f2fs_sb_info *sbi)
872 {
873 struct writeback_control wbc = {
874 .sync_mode = WB_SYNC_ALL,
875 .nr_to_write = LONG_MAX,
876 .for_reclaim = 0,
877 };
878 struct blk_plug plug;
879 int err = 0;
880
881 blk_start_plug(&plug);
882
883 retry_flush_dents:
884 f2fs_lock_all(sbi);
885 /* write all the dirty dentry pages */
886 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
887 f2fs_unlock_all(sbi);
888 err = sync_dirty_inodes(sbi, DIR_INODE);
889 if (err)
890 goto out;
891 goto retry_flush_dents;
892 }
893
894 /*
895 * POR: we should ensure that there are no dirty node pages
896 * until finishing nat/sit flush.
897 */
898 retry_flush_nodes:
899 down_write(&sbi->node_write);
900
901 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
902 up_write(&sbi->node_write);
903 err = sync_node_pages(sbi, &wbc);
904 if (err) {
905 f2fs_unlock_all(sbi);
906 goto out;
907 }
908 goto retry_flush_nodes;
909 }
910 out:
911 blk_finish_plug(&plug);
912 return err;
913 }
914
915 static void unblock_operations(struct f2fs_sb_info *sbi)
916 {
917 up_write(&sbi->node_write);
918 f2fs_unlock_all(sbi);
919 }
920
921 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
922 {
923 DEFINE_WAIT(wait);
924
925 for (;;) {
926 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
927
928 if (!get_pages(sbi, F2FS_WRITEBACK))
929 break;
930
931 io_schedule_timeout(5*HZ);
932 }
933 finish_wait(&sbi->cp_wait, &wait);
934 }
935
936 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
937 {
938 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
939 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
940 struct f2fs_nm_info *nm_i = NM_I(sbi);
941 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
942 nid_t last_nid = nm_i->next_scan_nid;
943 block_t start_blk;
944 unsigned int data_sum_blocks, orphan_blocks;
945 __u32 crc32 = 0;
946 int i;
947 int cp_payload_blks = __cp_payload(sbi);
948 block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
949 bool invalidate = false;
950 struct super_block *sb = sbi->sb;
951 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
952 u64 kbytes_written;
953
954 /*
955 * This avoids to conduct wrong roll-forward operations and uses
956 * metapages, so should be called prior to sync_meta_pages below.
957 */
958 if (discard_next_dnode(sbi, discard_blk))
959 invalidate = true;
960
961 /* Flush all the NAT/SIT pages */
962 while (get_pages(sbi, F2FS_DIRTY_META)) {
963 sync_meta_pages(sbi, META, LONG_MAX);
964 if (unlikely(f2fs_cp_error(sbi)))
965 return -EIO;
966 }
967
968 next_free_nid(sbi, &last_nid);
969
970 /*
971 * modify checkpoint
972 * version number is already updated
973 */
974 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
975 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
976 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
977 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
978 ckpt->cur_node_segno[i] =
979 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
980 ckpt->cur_node_blkoff[i] =
981 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
982 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
983 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
984 }
985 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
986 ckpt->cur_data_segno[i] =
987 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
988 ckpt->cur_data_blkoff[i] =
989 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
990 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
991 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
992 }
993
994 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
995 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
996 ckpt->next_free_nid = cpu_to_le32(last_nid);
997
998 /* 2 cp + n data seg summary + orphan inode blocks */
999 data_sum_blocks = npages_for_summary_flush(sbi, false);
1000 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1001 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1002 else
1003 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1004
1005 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1006 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1007 orphan_blocks);
1008
1009 if (__remain_node_summaries(cpc->reason))
1010 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1011 cp_payload_blks + data_sum_blocks +
1012 orphan_blocks + NR_CURSEG_NODE_TYPE);
1013 else
1014 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1015 cp_payload_blks + data_sum_blocks +
1016 orphan_blocks);
1017
1018 if (cpc->reason == CP_UMOUNT)
1019 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1020 else
1021 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1022
1023 if (cpc->reason == CP_FASTBOOT)
1024 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1025 else
1026 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1027
1028 if (orphan_num)
1029 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1030 else
1031 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1032
1033 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1034 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1035
1036 /* update SIT/NAT bitmap */
1037 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1038 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1039
1040 crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1041 *((__le32 *)((unsigned char *)ckpt +
1042 le32_to_cpu(ckpt->checksum_offset)))
1043 = cpu_to_le32(crc32);
1044
1045 start_blk = __start_cp_addr(sbi);
1046
1047 /* need to wait for end_io results */
1048 wait_on_all_pages_writeback(sbi);
1049 if (unlikely(f2fs_cp_error(sbi)))
1050 return -EIO;
1051
1052 /* write out checkpoint buffer at block 0 */
1053 update_meta_page(sbi, ckpt, start_blk++);
1054
1055 for (i = 1; i < 1 + cp_payload_blks; i++)
1056 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1057 start_blk++);
1058
1059 if (orphan_num) {
1060 write_orphan_inodes(sbi, start_blk);
1061 start_blk += orphan_blocks;
1062 }
1063
1064 write_data_summaries(sbi, start_blk);
1065 start_blk += data_sum_blocks;
1066
1067 /* Record write statistics in the hot node summary */
1068 kbytes_written = sbi->kbytes_written;
1069 if (sb->s_bdev->bd_part)
1070 kbytes_written += BD_PART_WRITTEN(sbi);
1071
1072 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1073
1074 if (__remain_node_summaries(cpc->reason)) {
1075 write_node_summaries(sbi, start_blk);
1076 start_blk += NR_CURSEG_NODE_TYPE;
1077 }
1078
1079 /* writeout checkpoint block */
1080 update_meta_page(sbi, ckpt, start_blk);
1081
1082 /* wait for previous submitted node/meta pages writeback */
1083 wait_on_all_pages_writeback(sbi);
1084
1085 if (unlikely(f2fs_cp_error(sbi)))
1086 return -EIO;
1087
1088 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1089 filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1090
1091 /* update user_block_counts */
1092 sbi->last_valid_block_count = sbi->total_valid_block_count;
1093 sbi->alloc_valid_block_count = 0;
1094
1095 /* Here, we only have one bio having CP pack */
1096 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1097
1098 /* wait for previous submitted meta pages writeback */
1099 wait_on_all_pages_writeback(sbi);
1100
1101 /*
1102 * invalidate meta page which is used temporarily for zeroing out
1103 * block at the end of warm node chain.
1104 */
1105 if (invalidate)
1106 invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
1107 discard_blk);
1108
1109 release_ino_entry(sbi);
1110
1111 if (unlikely(f2fs_cp_error(sbi)))
1112 return -EIO;
1113
1114 clear_prefree_segments(sbi, cpc);
1115 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1116
1117 return 0;
1118 }
1119
1120 /*
1121 * We guarantee that this checkpoint procedure will not fail.
1122 */
1123 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1124 {
1125 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1126 unsigned long long ckpt_ver;
1127 int err = 0;
1128
1129 mutex_lock(&sbi->cp_mutex);
1130
1131 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1132 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1133 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1134 goto out;
1135 if (unlikely(f2fs_cp_error(sbi))) {
1136 err = -EIO;
1137 goto out;
1138 }
1139 if (f2fs_readonly(sbi->sb)) {
1140 err = -EROFS;
1141 goto out;
1142 }
1143
1144 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1145
1146 err = block_operations(sbi);
1147 if (err)
1148 goto out;
1149
1150 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1151
1152 f2fs_flush_merged_bios(sbi);
1153
1154 /*
1155 * update checkpoint pack index
1156 * Increase the version number so that
1157 * SIT entries and seg summaries are written at correct place
1158 */
1159 ckpt_ver = cur_cp_version(ckpt);
1160 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1161
1162 /* write cached NAT/SIT entries to NAT/SIT area */
1163 flush_nat_entries(sbi);
1164 flush_sit_entries(sbi, cpc);
1165
1166 /* unlock all the fs_lock[] in do_checkpoint() */
1167 err = do_checkpoint(sbi, cpc);
1168
1169 unblock_operations(sbi);
1170 stat_inc_cp_count(sbi->stat_info);
1171
1172 if (cpc->reason == CP_RECOVERY)
1173 f2fs_msg(sbi->sb, KERN_NOTICE,
1174 "checkpoint: version = %llx", ckpt_ver);
1175
1176 /* do checkpoint periodically */
1177 f2fs_update_time(sbi, CP_TIME);
1178 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1179 out:
1180 mutex_unlock(&sbi->cp_mutex);
1181 return err;
1182 }
1183
1184 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1185 {
1186 int i;
1187
1188 for (i = 0; i < MAX_INO_ENTRY; i++) {
1189 struct inode_management *im = &sbi->im[i];
1190
1191 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1192 spin_lock_init(&im->ino_lock);
1193 INIT_LIST_HEAD(&im->ino_list);
1194 im->ino_num = 0;
1195 }
1196
1197 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1198 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1199 F2FS_ORPHANS_PER_BLOCK;
1200 }
1201
1202 int __init create_checkpoint_caches(void)
1203 {
1204 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1205 sizeof(struct ino_entry));
1206 if (!ino_entry_slab)
1207 return -ENOMEM;
1208 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1209 sizeof(struct inode_entry));
1210 if (!inode_entry_slab) {
1211 kmem_cache_destroy(ino_entry_slab);
1212 return -ENOMEM;
1213 }
1214 return 0;
1215 }
1216
1217 void destroy_checkpoint_caches(void)
1218 {
1219 kmem_cache_destroy(ino_entry_slab);
1220 kmem_cache_destroy(inode_entry_slab);
1221 }
Linux kernel - Deliverables
This page took 0.058844 seconds and 5 git commands to generate.