projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: produce more nids and reduce readahead nats
[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 SetPageUptodate(page);
52 return page;
53 }
54
55 /*
56 * We guarantee no failure on the returned page.
57 */
58 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
59 bool is_meta)
60 {
61 struct address_space *mapping = META_MAPPING(sbi);
62 struct page *page;
63 struct f2fs_io_info fio = {
64 .sbi = sbi,
65 .type = META,
66 .rw = READ_SYNC | REQ_META | REQ_PRIO,
67 .old_blkaddr = index,
68 .new_blkaddr = index,
69 .encrypted_page = NULL,
70 };
71
72 if (unlikely(!is_meta))
73 fio.rw &= ~REQ_META;
74 repeat:
75 page = f2fs_grab_cache_page(mapping, index, false);
76 if (!page) {
77 cond_resched();
78 goto repeat;
79 }
80 if (PageUptodate(page))
81 goto out;
82
83 fio.page = page;
84
85 if (f2fs_submit_page_bio(&fio)) {
86 f2fs_put_page(page, 1);
87 goto repeat;
88 }
89
90 lock_page(page);
91 if (unlikely(page->mapping != mapping)) {
92 f2fs_put_page(page, 1);
93 goto repeat;
94 }
95
96 /*
97 * if there is any IO error when accessing device, make our filesystem
98 * readonly and make sure do not write checkpoint with non-uptodate
99 * meta page.
100 */
101 if (unlikely(!PageUptodate(page)))
102 f2fs_stop_checkpoint(sbi, false);
103 out:
104 return page;
105 }
106
107 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
108 {
109 return __get_meta_page(sbi, index, true);
110 }
111
112 /* for POR only */
113 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
114 {
115 return __get_meta_page(sbi, index, false);
116 }
117
118 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
119 {
120 switch (type) {
121 case META_NAT:
122 break;
123 case META_SIT:
124 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
125 return false;
126 break;
127 case META_SSA:
128 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
129 blkaddr < SM_I(sbi)->ssa_blkaddr))
130 return false;
131 break;
132 case META_CP:
133 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
134 blkaddr < __start_cp_addr(sbi)))
135 return false;
136 break;
137 case META_POR:
138 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
139 blkaddr < MAIN_BLKADDR(sbi)))
140 return false;
141 break;
142 default:
143 BUG();
144 }
145
146 return true;
147 }
148
149 /*
150 * Readahead CP/NAT/SIT/SSA pages
151 */
152 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
153 int type, bool sync)
154 {
155 struct page *page;
156 block_t blkno = start;
157 struct f2fs_io_info fio = {
158 .sbi = sbi,
159 .type = META,
160 .rw = sync ? (READ_SYNC | REQ_META | REQ_PRIO) : READA,
161 .encrypted_page = NULL,
162 };
163 struct blk_plug plug;
164
165 if (unlikely(type == META_POR))
166 fio.rw &= ~REQ_META;
167
168 blk_start_plug(&plug);
169 for (; nrpages-- > 0; blkno++) {
170
171 if (!is_valid_blkaddr(sbi, blkno, type))
172 goto out;
173
174 switch (type) {
175 case META_NAT:
176 if (unlikely(blkno >=
177 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
178 blkno = 0;
179 /* get nat block addr */
180 fio.new_blkaddr = current_nat_addr(sbi,
181 blkno * NAT_ENTRY_PER_BLOCK);
182 break;
183 case META_SIT:
184 /* get sit block addr */
185 fio.new_blkaddr = current_sit_addr(sbi,
186 blkno * SIT_ENTRY_PER_BLOCK);
187 break;
188 case META_SSA:
189 case META_CP:
190 case META_POR:
191 fio.new_blkaddr = blkno;
192 break;
193 default:
194 BUG();
195 }
196
197 page = f2fs_grab_cache_page(META_MAPPING(sbi),
198 fio.new_blkaddr, false);
199 if (!page)
200 continue;
201 if (PageUptodate(page)) {
202 f2fs_put_page(page, 1);
203 continue;
204 }
205
206 fio.page = page;
207 fio.old_blkaddr = fio.new_blkaddr;
208 f2fs_submit_page_mbio(&fio);
209 f2fs_put_page(page, 0);
210 }
211 out:
212 f2fs_submit_merged_bio(sbi, META, READ);
213 blk_finish_plug(&plug);
214 return blkno - start;
215 }
216
217 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
218 {
219 struct page *page;
220 bool readahead = false;
221
222 page = find_get_page(META_MAPPING(sbi), index);
223 if (!page || !PageUptodate(page))
224 readahead = true;
225 f2fs_put_page(page, 0);
226
227 if (readahead)
228 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
229 }
230
231 static int f2fs_write_meta_page(struct page *page,
232 struct writeback_control *wbc)
233 {
234 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
235
236 trace_f2fs_writepage(page, META);
237
238 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
239 goto redirty_out;
240 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
241 goto redirty_out;
242 if (unlikely(f2fs_cp_error(sbi)))
243 goto redirty_out;
244
245 write_meta_page(sbi, page);
246 dec_page_count(sbi, F2FS_DIRTY_META);
247
248 if (wbc->for_reclaim)
249 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
250
251 unlock_page(page);
252
253 if (unlikely(f2fs_cp_error(sbi)))
254 f2fs_submit_merged_bio(sbi, META, WRITE);
255
256 return 0;
257
258 redirty_out:
259 redirty_page_for_writepage(wbc, page);
260 return AOP_WRITEPAGE_ACTIVATE;
261 }
262
263 static int f2fs_write_meta_pages(struct address_space *mapping,
264 struct writeback_control *wbc)
265 {
266 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
267 long diff, written;
268
269 /* collect a number of dirty meta pages and write together */
270 if (wbc->for_kupdate ||
271 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
272 goto skip_write;
273
274 trace_f2fs_writepages(mapping->host, wbc, META);
275
276 /* if mounting is failed, skip writing node pages */
277 mutex_lock(&sbi->cp_mutex);
278 diff = nr_pages_to_write(sbi, META, wbc);
279 written = sync_meta_pages(sbi, META, wbc->nr_to_write);
280 mutex_unlock(&sbi->cp_mutex);
281 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
282 return 0;
283
284 skip_write:
285 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
286 trace_f2fs_writepages(mapping->host, wbc, META);
287 return 0;
288 }
289
290 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
291 long nr_to_write)
292 {
293 struct address_space *mapping = META_MAPPING(sbi);
294 pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
295 struct pagevec pvec;
296 long nwritten = 0;
297 struct writeback_control wbc = {
298 .for_reclaim = 0,
299 };
300 struct blk_plug plug;
301
302 pagevec_init(&pvec, 0);
303
304 blk_start_plug(&plug);
305
306 while (index <= end) {
307 int i, nr_pages;
308 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
309 PAGECACHE_TAG_DIRTY,
310 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
311 if (unlikely(nr_pages == 0))
312 break;
313
314 for (i = 0; i < nr_pages; i++) {
315 struct page *page = pvec.pages[i];
316
317 if (prev == ULONG_MAX)
318 prev = page->index - 1;
319 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
320 pagevec_release(&pvec);
321 goto stop;
322 }
323
324 lock_page(page);
325
326 if (unlikely(page->mapping != mapping)) {
327 continue_unlock:
328 unlock_page(page);
329 continue;
330 }
331 if (!PageDirty(page)) {
332 /* someone wrote it for us */
333 goto continue_unlock;
334 }
335
336 f2fs_wait_on_page_writeback(page, META, true);
337
338 BUG_ON(PageWriteback(page));
339 if (!clear_page_dirty_for_io(page))
340 goto continue_unlock;
341
342 if (mapping->a_ops->writepage(page, &wbc)) {
343 unlock_page(page);
344 break;
345 }
346 nwritten++;
347 prev = page->index;
348 if (unlikely(nwritten >= nr_to_write))
349 break;
350 }
351 pagevec_release(&pvec);
352 cond_resched();
353 }
354 stop:
355 if (nwritten)
356 f2fs_submit_merged_bio(sbi, type, WRITE);
357
358 blk_finish_plug(&plug);
359
360 return nwritten;
361 }
362
363 static int f2fs_set_meta_page_dirty(struct page *page)
364 {
365 trace_f2fs_set_page_dirty(page, META);
366
367 SetPageUptodate(page);
368 if (!PageDirty(page)) {
369 __set_page_dirty_nobuffers(page);
370 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
371 SetPagePrivate(page);
372 f2fs_trace_pid(page);
373 return 1;
374 }
375 return 0;
376 }
377
378 const struct address_space_operations f2fs_meta_aops = {
379 .writepage = f2fs_write_meta_page,
380 .writepages = f2fs_write_meta_pages,
381 .set_page_dirty = f2fs_set_meta_page_dirty,
382 .invalidatepage = f2fs_invalidate_page,
383 .releasepage = f2fs_release_page,
384 };
385
386 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
387 {
388 struct inode_management *im = &sbi->im[type];
389 struct ino_entry *e, *tmp;
390
391 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
392 retry:
393 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
394
395 spin_lock(&im->ino_lock);
396 e = radix_tree_lookup(&im->ino_root, ino);
397 if (!e) {
398 e = tmp;
399 if (radix_tree_insert(&im->ino_root, ino, e)) {
400 spin_unlock(&im->ino_lock);
401 radix_tree_preload_end();
402 goto retry;
403 }
404 memset(e, 0, sizeof(struct ino_entry));
405 e->ino = ino;
406
407 list_add_tail(&e->list, &im->ino_list);
408 if (type != ORPHAN_INO)
409 im->ino_num++;
410 }
411 spin_unlock(&im->ino_lock);
412 radix_tree_preload_end();
413
414 if (e != tmp)
415 kmem_cache_free(ino_entry_slab, tmp);
416 }
417
418 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
419 {
420 struct inode_management *im = &sbi->im[type];
421 struct ino_entry *e;
422
423 spin_lock(&im->ino_lock);
424 e = radix_tree_lookup(&im->ino_root, ino);
425 if (e) {
426 list_del(&e->list);
427 radix_tree_delete(&im->ino_root, ino);
428 im->ino_num--;
429 spin_unlock(&im->ino_lock);
430 kmem_cache_free(ino_entry_slab, e);
431 return;
432 }
433 spin_unlock(&im->ino_lock);
434 }
435
436 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
437 {
438 /* add new dirty ino entry into list */
439 __add_ino_entry(sbi, ino, type);
440 }
441
442 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
443 {
444 /* remove dirty ino entry from list */
445 __remove_ino_entry(sbi, ino, type);
446 }
447
448 /* mode should be APPEND_INO or UPDATE_INO */
449 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
450 {
451 struct inode_management *im = &sbi->im[mode];
452 struct ino_entry *e;
453
454 spin_lock(&im->ino_lock);
455 e = radix_tree_lookup(&im->ino_root, ino);
456 spin_unlock(&im->ino_lock);
457 return e ? true : false;
458 }
459
460 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
461 {
462 struct ino_entry *e, *tmp;
463 int i;
464
465 for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
466 struct inode_management *im = &sbi->im[i];
467
468 spin_lock(&im->ino_lock);
469 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
470 list_del(&e->list);
471 radix_tree_delete(&im->ino_root, e->ino);
472 kmem_cache_free(ino_entry_slab, e);
473 im->ino_num--;
474 }
475 spin_unlock(&im->ino_lock);
476 }
477 }
478
479 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
480 {
481 struct inode_management *im = &sbi->im[ORPHAN_INO];
482 int err = 0;
483
484 spin_lock(&im->ino_lock);
485
486 #ifdef CONFIG_F2FS_FAULT_INJECTION
487 if (time_to_inject(FAULT_ORPHAN)) {
488 spin_unlock(&im->ino_lock);
489 return -ENOSPC;
490 }
491 #endif
492 if (unlikely(im->ino_num >= sbi->max_orphans))
493 err = -ENOSPC;
494 else
495 im->ino_num++;
496 spin_unlock(&im->ino_lock);
497
498 return err;
499 }
500
501 void release_orphan_inode(struct f2fs_sb_info *sbi)
502 {
503 struct inode_management *im = &sbi->im[ORPHAN_INO];
504
505 spin_lock(&im->ino_lock);
506 f2fs_bug_on(sbi, im->ino_num == 0);
507 im->ino_num--;
508 spin_unlock(&im->ino_lock);
509 }
510
511 void add_orphan_inode(struct inode *inode)
512 {
513 /* add new orphan ino entry into list */
514 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO);
515 update_inode_page(inode);
516 }
517
518 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
519 {
520 /* remove orphan entry from orphan list */
521 __remove_ino_entry(sbi, ino, ORPHAN_INO);
522 }
523
524 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
525 {
526 struct inode *inode;
527
528 inode = f2fs_iget(sbi->sb, ino);
529 if (IS_ERR(inode)) {
530 /*
531 * there should be a bug that we can't find the entry
532 * to orphan inode.
533 */
534 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
535 return PTR_ERR(inode);
536 }
537
538 clear_nlink(inode);
539
540 /* truncate all the data during iput */
541 iput(inode);
542 return 0;
543 }
544
545 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
546 {
547 block_t start_blk, orphan_blocks, i, j;
548 int err;
549
550 if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
551 return 0;
552
553 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
554 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
555
556 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
557
558 for (i = 0; i < orphan_blocks; i++) {
559 struct page *page = get_meta_page(sbi, start_blk + i);
560 struct f2fs_orphan_block *orphan_blk;
561
562 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
563 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
564 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
565 err = recover_orphan_inode(sbi, ino);
566 if (err) {
567 f2fs_put_page(page, 1);
568 return err;
569 }
570 }
571 f2fs_put_page(page, 1);
572 }
573 /* clear Orphan Flag */
574 clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
575 return 0;
576 }
577
578 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
579 {
580 struct list_head *head;
581 struct f2fs_orphan_block *orphan_blk = NULL;
582 unsigned int nentries = 0;
583 unsigned short index = 1;
584 unsigned short orphan_blocks;
585 struct page *page = NULL;
586 struct ino_entry *orphan = NULL;
587 struct inode_management *im = &sbi->im[ORPHAN_INO];
588
589 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
590
591 /*
592 * we don't need to do spin_lock(&im->ino_lock) here, since all the
593 * orphan inode operations are covered under f2fs_lock_op().
594 * And, spin_lock should be avoided due to page operations below.
595 */
596 head = &im->ino_list;
597
598 /* loop for each orphan inode entry and write them in Jornal block */
599 list_for_each_entry(orphan, head, list) {
600 if (!page) {
601 page = grab_meta_page(sbi, start_blk++);
602 orphan_blk =
603 (struct f2fs_orphan_block *)page_address(page);
604 memset(orphan_blk, 0, sizeof(*orphan_blk));
605 }
606
607 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
608
609 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
610 /*
611 * an orphan block is full of 1020 entries,
612 * then we need to flush current orphan blocks
613 * and bring another one in memory
614 */
615 orphan_blk->blk_addr = cpu_to_le16(index);
616 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
617 orphan_blk->entry_count = cpu_to_le32(nentries);
618 set_page_dirty(page);
619 f2fs_put_page(page, 1);
620 index++;
621 nentries = 0;
622 page = NULL;
623 }
624 }
625
626 if (page) {
627 orphan_blk->blk_addr = cpu_to_le16(index);
628 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
629 orphan_blk->entry_count = cpu_to_le32(nentries);
630 set_page_dirty(page);
631 f2fs_put_page(page, 1);
632 }
633 }
634
635 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
636 block_t cp_addr, unsigned long long *version)
637 {
638 struct page *cp_page_1, *cp_page_2 = NULL;
639 unsigned long blk_size = sbi->blocksize;
640 struct f2fs_checkpoint *cp_block;
641 unsigned long long cur_version = 0, pre_version = 0;
642 size_t crc_offset;
643 __u32 crc = 0;
644
645 /* Read the 1st cp block in this CP pack */
646 cp_page_1 = get_meta_page(sbi, cp_addr);
647
648 /* get the version number */
649 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
650 crc_offset = le32_to_cpu(cp_block->checksum_offset);
651 if (crc_offset >= blk_size)
652 goto invalid_cp1;
653
654 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
655 if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
656 goto invalid_cp1;
657
658 pre_version = cur_cp_version(cp_block);
659
660 /* Read the 2nd cp block in this CP pack */
661 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
662 cp_page_2 = get_meta_page(sbi, cp_addr);
663
664 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
665 crc_offset = le32_to_cpu(cp_block->checksum_offset);
666 if (crc_offset >= blk_size)
667 goto invalid_cp2;
668
669 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
670 if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
671 goto invalid_cp2;
672
673 cur_version = cur_cp_version(cp_block);
674
675 if (cur_version == pre_version) {
676 *version = cur_version;
677 f2fs_put_page(cp_page_2, 1);
678 return cp_page_1;
679 }
680 invalid_cp2:
681 f2fs_put_page(cp_page_2, 1);
682 invalid_cp1:
683 f2fs_put_page(cp_page_1, 1);
684 return NULL;
685 }
686
687 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
688 {
689 struct f2fs_checkpoint *cp_block;
690 struct f2fs_super_block *fsb = sbi->raw_super;
691 struct page *cp1, *cp2, *cur_page;
692 unsigned long blk_size = sbi->blocksize;
693 unsigned long long cp1_version = 0, cp2_version = 0;
694 unsigned long long cp_start_blk_no;
695 unsigned int cp_blks = 1 + __cp_payload(sbi);
696 block_t cp_blk_no;
697 int i;
698
699 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
700 if (!sbi->ckpt)
701 return -ENOMEM;
702 /*
703 * Finding out valid cp block involves read both
704 * sets( cp pack1 and cp pack 2)
705 */
706 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
707 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
708
709 /* The second checkpoint pack should start at the next segment */
710 cp_start_blk_no += ((unsigned long long)1) <<
711 le32_to_cpu(fsb->log_blocks_per_seg);
712 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
713
714 if (cp1 && cp2) {
715 if (ver_after(cp2_version, cp1_version))
716 cur_page = cp2;
717 else
718 cur_page = cp1;
719 } else if (cp1) {
720 cur_page = cp1;
721 } else if (cp2) {
722 cur_page = cp2;
723 } else {
724 goto fail_no_cp;
725 }
726
727 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
728 memcpy(sbi->ckpt, cp_block, blk_size);
729
730 /* Sanity checking of checkpoint */
731 if (sanity_check_ckpt(sbi))
732 goto fail_no_cp;
733
734 if (cp_blks <= 1)
735 goto done;
736
737 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
738 if (cur_page == cp2)
739 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
740
741 for (i = 1; i < cp_blks; i++) {
742 void *sit_bitmap_ptr;
743 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
744
745 cur_page = get_meta_page(sbi, cp_blk_no + i);
746 sit_bitmap_ptr = page_address(cur_page);
747 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
748 f2fs_put_page(cur_page, 1);
749 }
750 done:
751 f2fs_put_page(cp1, 1);
752 f2fs_put_page(cp2, 1);
753 return 0;
754
755 fail_no_cp:
756 kfree(sbi->ckpt);
757 return -EINVAL;
758 }
759
760 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
761 {
762 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
763 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
764
765 if (is_inode_flag_set(inode, flag))
766 return;
767
768 set_inode_flag(inode, flag);
769 list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
770 stat_inc_dirty_inode(sbi, type);
771 }
772
773 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
774 {
775 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
776
777 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
778 return;
779
780 list_del_init(&F2FS_I(inode)->dirty_list);
781 clear_inode_flag(inode, flag);
782 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
783 }
784
785 void update_dirty_page(struct inode *inode, struct page *page)
786 {
787 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
788 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
789
790 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
791 !S_ISLNK(inode->i_mode))
792 return;
793
794 spin_lock(&sbi->inode_lock[type]);
795 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
796 __add_dirty_inode(inode, type);
797 inode_inc_dirty_pages(inode);
798 spin_unlock(&sbi->inode_lock[type]);
799
800 SetPagePrivate(page);
801 f2fs_trace_pid(page);
802 }
803
804 void remove_dirty_inode(struct inode *inode)
805 {
806 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
807 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
808
809 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
810 !S_ISLNK(inode->i_mode))
811 return;
812
813 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
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
821 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
822 {
823 struct list_head *head;
824 struct inode *inode;
825 struct f2fs_inode_info *fi;
826 bool is_dir = (type == DIR_INODE);
827
828 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
829 get_pages(sbi, is_dir ?
830 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
831 retry:
832 if (unlikely(f2fs_cp_error(sbi)))
833 return -EIO;
834
835 spin_lock(&sbi->inode_lock[type]);
836
837 head = &sbi->inode_list[type];
838 if (list_empty(head)) {
839 spin_unlock(&sbi->inode_lock[type]);
840 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
841 get_pages(sbi, is_dir ?
842 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
843 return 0;
844 }
845 fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
846 inode = igrab(&fi->vfs_inode);
847 spin_unlock(&sbi->inode_lock[type]);
848 if (inode) {
849 filemap_fdatawrite(inode->i_mapping);
850 iput(inode);
851 } else {
852 /*
853 * We should submit bio, since it exists several
854 * wribacking dentry pages in the freeing inode.
855 */
856 f2fs_submit_merged_bio(sbi, DATA, WRITE);
857 cond_resched();
858 }
859 goto retry;
860 }
861
862 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
863 {
864 struct list_head *head = &sbi->inode_list[DIRTY_META];
865 struct inode *inode;
866 struct f2fs_inode_info *fi;
867 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
868
869 while (total--) {
870 if (unlikely(f2fs_cp_error(sbi)))
871 return -EIO;
872
873 spin_lock(&sbi->inode_lock[DIRTY_META]);
874 if (list_empty(head)) {
875 spin_unlock(&sbi->inode_lock[DIRTY_META]);
876 return 0;
877 }
878 fi = list_entry(head->next, struct f2fs_inode_info,
879 gdirty_list);
880 inode = igrab(&fi->vfs_inode);
881 spin_unlock(&sbi->inode_lock[DIRTY_META]);
882 if (inode) {
883 update_inode_page(inode);
884 iput(inode);
885 }
886 };
887 return 0;
888 }
889
890 /*
891 * Freeze all the FS-operations for checkpoint.
892 */
893 static int block_operations(struct f2fs_sb_info *sbi)
894 {
895 struct writeback_control wbc = {
896 .sync_mode = WB_SYNC_ALL,
897 .nr_to_write = LONG_MAX,
898 .for_reclaim = 0,
899 };
900 int err = 0;
901
902 retry_flush_dents:
903 f2fs_lock_all(sbi);
904 /* write all the dirty dentry pages */
905 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
906 f2fs_unlock_all(sbi);
907 err = sync_dirty_inodes(sbi, DIR_INODE);
908 if (err)
909 goto out;
910 goto retry_flush_dents;
911 }
912
913 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
914 f2fs_unlock_all(sbi);
915 err = f2fs_sync_inode_meta(sbi);
916 if (err)
917 goto out;
918 goto retry_flush_dents;
919 }
920
921 /*
922 * POR: we should ensure that there are no dirty node pages
923 * until finishing nat/sit flush.
924 */
925 retry_flush_nodes:
926 down_write(&sbi->node_write);
927
928 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
929 up_write(&sbi->node_write);
930 err = sync_node_pages(sbi, &wbc);
931 if (err) {
932 f2fs_unlock_all(sbi);
933 goto out;
934 }
935 goto retry_flush_nodes;
936 }
937 out:
938 return err;
939 }
940
941 static void unblock_operations(struct f2fs_sb_info *sbi)
942 {
943 up_write(&sbi->node_write);
944
945 build_free_nids(sbi);
946 f2fs_unlock_all(sbi);
947 }
948
949 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
950 {
951 DEFINE_WAIT(wait);
952
953 for (;;) {
954 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
955
956 if (!atomic_read(&sbi->nr_wb_bios))
957 break;
958
959 io_schedule_timeout(5*HZ);
960 }
961 finish_wait(&sbi->cp_wait, &wait);
962 }
963
964 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
965 {
966 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
967 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
968 struct f2fs_nm_info *nm_i = NM_I(sbi);
969 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
970 nid_t last_nid = nm_i->next_scan_nid;
971 block_t start_blk;
972 unsigned int data_sum_blocks, orphan_blocks;
973 __u32 crc32 = 0;
974 int i;
975 int cp_payload_blks = __cp_payload(sbi);
976 block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
977 bool invalidate = false;
978 struct super_block *sb = sbi->sb;
979 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
980 u64 kbytes_written;
981
982 /*
983 * This avoids to conduct wrong roll-forward operations and uses
984 * metapages, so should be called prior to sync_meta_pages below.
985 */
986 if (!test_opt(sbi, LFS) && discard_next_dnode(sbi, discard_blk))
987 invalidate = true;
988
989 /* Flush all the NAT/SIT pages */
990 while (get_pages(sbi, F2FS_DIRTY_META)) {
991 sync_meta_pages(sbi, META, LONG_MAX);
992 if (unlikely(f2fs_cp_error(sbi)))
993 return -EIO;
994 }
995
996 next_free_nid(sbi, &last_nid);
997
998 /*
999 * modify checkpoint
1000 * version number is already updated
1001 */
1002 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1003 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1004 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1005 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1006 ckpt->cur_node_segno[i] =
1007 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1008 ckpt->cur_node_blkoff[i] =
1009 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1010 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1011 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1012 }
1013 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1014 ckpt->cur_data_segno[i] =
1015 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1016 ckpt->cur_data_blkoff[i] =
1017 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1018 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1019 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1020 }
1021
1022 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1023 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1024 ckpt->next_free_nid = cpu_to_le32(last_nid);
1025
1026 /* 2 cp + n data seg summary + orphan inode blocks */
1027 data_sum_blocks = npages_for_summary_flush(sbi, false);
1028 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1029 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1030 else
1031 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1032
1033 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1034 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1035 orphan_blocks);
1036
1037 if (__remain_node_summaries(cpc->reason))
1038 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1039 cp_payload_blks + data_sum_blocks +
1040 orphan_blocks + NR_CURSEG_NODE_TYPE);
1041 else
1042 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1043 cp_payload_blks + data_sum_blocks +
1044 orphan_blocks);
1045
1046 if (cpc->reason == CP_UMOUNT)
1047 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1048 else
1049 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1050
1051 if (cpc->reason == CP_FASTBOOT)
1052 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1053 else
1054 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1055
1056 if (orphan_num)
1057 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1058 else
1059 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1060
1061 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1062 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1063
1064 /* update SIT/NAT bitmap */
1065 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1066 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1067
1068 crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1069 *((__le32 *)((unsigned char *)ckpt +
1070 le32_to_cpu(ckpt->checksum_offset)))
1071 = cpu_to_le32(crc32);
1072
1073 start_blk = __start_cp_addr(sbi);
1074
1075 /* need to wait for end_io results */
1076 wait_on_all_pages_writeback(sbi);
1077 if (unlikely(f2fs_cp_error(sbi)))
1078 return -EIO;
1079
1080 /* write out checkpoint buffer at block 0 */
1081 update_meta_page(sbi, ckpt, start_blk++);
1082
1083 for (i = 1; i < 1 + cp_payload_blks; i++)
1084 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1085 start_blk++);
1086
1087 if (orphan_num) {
1088 write_orphan_inodes(sbi, start_blk);
1089 start_blk += orphan_blocks;
1090 }
1091
1092 write_data_summaries(sbi, start_blk);
1093 start_blk += data_sum_blocks;
1094
1095 /* Record write statistics in the hot node summary */
1096 kbytes_written = sbi->kbytes_written;
1097 if (sb->s_bdev->bd_part)
1098 kbytes_written += BD_PART_WRITTEN(sbi);
1099
1100 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1101
1102 if (__remain_node_summaries(cpc->reason)) {
1103 write_node_summaries(sbi, start_blk);
1104 start_blk += NR_CURSEG_NODE_TYPE;
1105 }
1106
1107 /* writeout checkpoint block */
1108 update_meta_page(sbi, ckpt, start_blk);
1109
1110 /* wait for previous submitted node/meta pages writeback */
1111 wait_on_all_pages_writeback(sbi);
1112
1113 if (unlikely(f2fs_cp_error(sbi)))
1114 return -EIO;
1115
1116 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1117 filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1118
1119 /* update user_block_counts */
1120 sbi->last_valid_block_count = sbi->total_valid_block_count;
1121 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1122
1123 /* Here, we only have one bio having CP pack */
1124 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1125
1126 /* wait for previous submitted meta pages writeback */
1127 wait_on_all_pages_writeback(sbi);
1128
1129 /*
1130 * invalidate meta page which is used temporarily for zeroing out
1131 * block at the end of warm node chain.
1132 */
1133 if (invalidate)
1134 invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
1135 discard_blk);
1136
1137 release_ino_entry(sbi, false);
1138
1139 if (unlikely(f2fs_cp_error(sbi)))
1140 return -EIO;
1141
1142 clear_prefree_segments(sbi, cpc);
1143 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1144
1145 return 0;
1146 }
1147
1148 /*
1149 * We guarantee that this checkpoint procedure will not fail.
1150 */
1151 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1152 {
1153 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1154 unsigned long long ckpt_ver;
1155 int err = 0;
1156
1157 mutex_lock(&sbi->cp_mutex);
1158
1159 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1160 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1161 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1162 goto out;
1163 if (unlikely(f2fs_cp_error(sbi))) {
1164 err = -EIO;
1165 goto out;
1166 }
1167 if (f2fs_readonly(sbi->sb)) {
1168 err = -EROFS;
1169 goto out;
1170 }
1171
1172 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1173
1174 err = block_operations(sbi);
1175 if (err)
1176 goto out;
1177
1178 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1179
1180 f2fs_flush_merged_bios(sbi);
1181
1182 /*
1183 * update checkpoint pack index
1184 * Increase the version number so that
1185 * SIT entries and seg summaries are written at correct place
1186 */
1187 ckpt_ver = cur_cp_version(ckpt);
1188 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1189
1190 /* write cached NAT/SIT entries to NAT/SIT area */
1191 flush_nat_entries(sbi);
1192 flush_sit_entries(sbi, cpc);
1193
1194 /* unlock all the fs_lock[] in do_checkpoint() */
1195 err = do_checkpoint(sbi, cpc);
1196
1197 unblock_operations(sbi);
1198 stat_inc_cp_count(sbi->stat_info);
1199
1200 if (cpc->reason == CP_RECOVERY)
1201 f2fs_msg(sbi->sb, KERN_NOTICE,
1202 "checkpoint: version = %llx", ckpt_ver);
1203
1204 /* do checkpoint periodically */
1205 f2fs_update_time(sbi, CP_TIME);
1206 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1207 out:
1208 mutex_unlock(&sbi->cp_mutex);
1209 return err;
1210 }
1211
1212 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1213 {
1214 int i;
1215
1216 for (i = 0; i < MAX_INO_ENTRY; i++) {
1217 struct inode_management *im = &sbi->im[i];
1218
1219 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1220 spin_lock_init(&im->ino_lock);
1221 INIT_LIST_HEAD(&im->ino_list);
1222 im->ino_num = 0;
1223 }
1224
1225 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1226 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1227 F2FS_ORPHANS_PER_BLOCK;
1228 }
1229
1230 int __init create_checkpoint_caches(void)
1231 {
1232 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1233 sizeof(struct ino_entry));
1234 if (!ino_entry_slab)
1235 return -ENOMEM;
1236 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1237 sizeof(struct inode_entry));
1238 if (!inode_entry_slab) {
1239 kmem_cache_destroy(ino_entry_slab);
1240 return -ENOMEM;
1241 }
1242 return 0;
1243 }
1244
1245 void destroy_checkpoint_caches(void)
1246 {
1247 kmem_cache_destroy(ino_entry_slab);
1248 kmem_cache_destroy(inode_entry_slab);
1249 }
Linux kernel - Deliverables
This page took 0.056828 seconds and 5 git commands to generate.