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
f2fs_stop_checkpoint(struct f2fs_sb_info * sbi,bool end_io)29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
30 {
31 set_ckpt_flags(sbi, CP_ERROR_FLAG);
32 if (!end_io)
33 f2fs_flush_merged_writes(sbi);
34 }
35
36 /*
37 * We guarantee no failure on the returned page.
38 */
grab_meta_page(struct f2fs_sb_info * sbi,pgoff_t index)39 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
40 {
41 struct address_space *mapping = META_MAPPING(sbi);
42 struct page *page = NULL;
43 repeat:
44 page = f2fs_grab_cache_page(mapping, index, false);
45 if (!page) {
46 cond_resched();
47 goto repeat;
48 }
49 f2fs_wait_on_page_writeback(page, META, true);
50 if (!PageUptodate(page))
51 SetPageUptodate(page);
52 return page;
53 }
54
55 /*
56 * We guarantee no failure on the returned page.
57 */
__get_meta_page(struct f2fs_sb_info * sbi,pgoff_t index,bool is_meta)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 .op = REQ_OP_READ,
67 .op_flags = 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.op_flags &= ~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 mark_page_accessed(page);
106 return page;
107 }
108
get_meta_page(struct f2fs_sb_info * sbi,pgoff_t index)109 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
110 {
111 return __get_meta_page(sbi, index, true);
112 }
113
114 /* for POR only */
get_tmp_page(struct f2fs_sb_info * sbi,pgoff_t index)115 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
116 {
117 return __get_meta_page(sbi, index, false);
118 }
119
is_valid_blkaddr(struct f2fs_sb_info * sbi,block_t blkaddr,int type)120 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
121 {
122 switch (type) {
123 case META_NAT:
124 break;
125 case META_SIT:
126 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
127 return false;
128 break;
129 case META_SSA:
130 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
131 blkaddr < SM_I(sbi)->ssa_blkaddr))
132 return false;
133 break;
134 case META_CP:
135 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
136 blkaddr < __start_cp_addr(sbi)))
137 return false;
138 break;
139 case META_POR:
140 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
141 blkaddr < MAIN_BLKADDR(sbi)))
142 return false;
143 break;
144 default:
145 BUG();
146 }
147
148 return true;
149 }
150
151 /*
152 * Readahead CP/NAT/SIT/SSA pages
153 */
ra_meta_pages(struct f2fs_sb_info * sbi,block_t start,int nrpages,int type,bool sync)154 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
155 int type, bool sync)
156 {
157 struct page *page;
158 block_t blkno = start;
159 struct f2fs_io_info fio = {
160 .sbi = sbi,
161 .type = META,
162 .op = REQ_OP_READ,
163 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
164 .encrypted_page = NULL,
165 .in_list = false,
166 };
167 struct blk_plug plug;
168
169 if (unlikely(type == META_POR))
170 fio.op_flags &= ~REQ_META;
171
172 blk_start_plug(&plug);
173 for (; nrpages-- > 0; blkno++) {
174
175 if (!is_valid_blkaddr(sbi, blkno, type))
176 goto out;
177
178 switch (type) {
179 case META_NAT:
180 if (unlikely(blkno >=
181 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
182 blkno = 0;
183 /* get nat block addr */
184 fio.new_blkaddr = current_nat_addr(sbi,
185 blkno * NAT_ENTRY_PER_BLOCK);
186 break;
187 case META_SIT:
188 /* get sit block addr */
189 fio.new_blkaddr = current_sit_addr(sbi,
190 blkno * SIT_ENTRY_PER_BLOCK);
191 break;
192 case META_SSA:
193 case META_CP:
194 case META_POR:
195 fio.new_blkaddr = blkno;
196 break;
197 default:
198 BUG();
199 }
200
201 page = f2fs_grab_cache_page(META_MAPPING(sbi),
202 fio.new_blkaddr, false);
203 if (!page)
204 continue;
205 if (PageUptodate(page)) {
206 f2fs_put_page(page, 1);
207 continue;
208 }
209
210 fio.page = page;
211 f2fs_submit_page_bio(&fio);
212 f2fs_put_page(page, 0);
213 }
214 out:
215 blk_finish_plug(&plug);
216 return blkno - start;
217 }
218
ra_meta_pages_cond(struct f2fs_sb_info * sbi,pgoff_t index)219 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
220 {
221 struct page *page;
222 bool readahead = false;
223
224 page = find_get_page(META_MAPPING(sbi), index);
225 if (!page || !PageUptodate(page))
226 readahead = true;
227 f2fs_put_page(page, 0);
228
229 if (readahead)
230 ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
231 }
232
__f2fs_write_meta_page(struct page * page,struct writeback_control * wbc,enum iostat_type io_type)233 static int __f2fs_write_meta_page(struct page *page,
234 struct writeback_control *wbc,
235 enum iostat_type io_type)
236 {
237 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
238
239 trace_f2fs_writepage(page, META);
240
241 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
242 goto redirty_out;
243 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
244 goto redirty_out;
245 if (unlikely(f2fs_cp_error(sbi)))
246 goto redirty_out;
247
248 write_meta_page(sbi, page, io_type);
249 dec_page_count(sbi, F2FS_DIRTY_META);
250
251 if (wbc->for_reclaim)
252 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
253 0, page->index, META);
254
255 unlock_page(page);
256
257 if (unlikely(f2fs_cp_error(sbi)))
258 f2fs_submit_merged_write(sbi, META);
259
260 return 0;
261
262 redirty_out:
263 redirty_page_for_writepage(wbc, page);
264 return AOP_WRITEPAGE_ACTIVATE;
265 }
266
f2fs_write_meta_page(struct page * page,struct writeback_control * wbc)267 static int f2fs_write_meta_page(struct page *page,
268 struct writeback_control *wbc)
269 {
270 return __f2fs_write_meta_page(page, wbc, FS_META_IO);
271 }
272
f2fs_write_meta_pages(struct address_space * mapping,struct writeback_control * wbc)273 static int f2fs_write_meta_pages(struct address_space *mapping,
274 struct writeback_control *wbc)
275 {
276 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
277 long diff, written;
278
279 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
280 goto skip_write;
281
282 /* collect a number of dirty meta pages and write together */
283 if (wbc->for_kupdate ||
284 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
285 goto skip_write;
286
287 /* if locked failed, cp will flush dirty pages instead */
288 if (!mutex_trylock(&sbi->cp_mutex))
289 goto skip_write;
290
291 trace_f2fs_writepages(mapping->host, wbc, META);
292 diff = nr_pages_to_write(sbi, META, wbc);
293 written = sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
294 mutex_unlock(&sbi->cp_mutex);
295 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
296 return 0;
297
298 skip_write:
299 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
300 trace_f2fs_writepages(mapping->host, wbc, META);
301 return 0;
302 }
303
sync_meta_pages(struct f2fs_sb_info * sbi,enum page_type type,long nr_to_write,enum iostat_type io_type)304 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
305 long nr_to_write, enum iostat_type io_type)
306 {
307 struct address_space *mapping = META_MAPPING(sbi);
308 pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
309 struct pagevec pvec;
310 long nwritten = 0;
311 struct writeback_control wbc = {
312 .for_reclaim = 0,
313 };
314 struct blk_plug plug;
315
316 pagevec_init(&pvec, 0);
317
318 blk_start_plug(&plug);
319
320 while (index <= end) {
321 int i, nr_pages;
322 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
323 PAGECACHE_TAG_DIRTY,
324 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
325 if (unlikely(nr_pages == 0))
326 break;
327
328 for (i = 0; i < nr_pages; i++) {
329 struct page *page = pvec.pages[i];
330
331 if (prev == ULONG_MAX)
332 prev = page->index - 1;
333 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
334 pagevec_release(&pvec);
335 goto stop;
336 }
337
338 lock_page(page);
339
340 if (unlikely(page->mapping != mapping)) {
341 continue_unlock:
342 unlock_page(page);
343 continue;
344 }
345 if (!PageDirty(page)) {
346 /* someone wrote it for us */
347 goto continue_unlock;
348 }
349
350 f2fs_wait_on_page_writeback(page, META, true);
351
352 BUG_ON(PageWriteback(page));
353 if (!clear_page_dirty_for_io(page))
354 goto continue_unlock;
355
356 if (__f2fs_write_meta_page(page, &wbc, io_type)) {
357 unlock_page(page);
358 break;
359 }
360 nwritten++;
361 prev = page->index;
362 if (unlikely(nwritten >= nr_to_write))
363 break;
364 }
365 pagevec_release(&pvec);
366 cond_resched();
367 }
368 stop:
369 if (nwritten)
370 f2fs_submit_merged_write(sbi, type);
371
372 blk_finish_plug(&plug);
373
374 return nwritten;
375 }
376
f2fs_set_meta_page_dirty(struct page * page)377 static int f2fs_set_meta_page_dirty(struct page *page)
378 {
379 trace_f2fs_set_page_dirty(page, META);
380
381 if (!PageUptodate(page))
382 SetPageUptodate(page);
383 if (!PageDirty(page)) {
384 f2fs_set_page_dirty_nobuffers(page);
385 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
386 SetPagePrivate(page);
387 f2fs_trace_pid(page);
388 return 1;
389 }
390 return 0;
391 }
392
393 const struct address_space_operations f2fs_meta_aops = {
394 .writepage = f2fs_write_meta_page,
395 .writepages = f2fs_write_meta_pages,
396 .set_page_dirty = f2fs_set_meta_page_dirty,
397 .invalidatepage = f2fs_invalidate_page,
398 .releasepage = f2fs_release_page,
399 #ifdef CONFIG_MIGRATION
400 .migratepage = f2fs_migrate_page,
401 #endif
402 };
403
__add_ino_entry(struct f2fs_sb_info * sbi,nid_t ino,unsigned int devidx,int type)404 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
405 unsigned int devidx, int type)
406 {
407 struct inode_management *im = &sbi->im[type];
408 struct ino_entry *e, *tmp;
409
410 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
411
412 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
413
414 spin_lock(&im->ino_lock);
415 e = radix_tree_lookup(&im->ino_root, ino);
416 if (!e) {
417 e = tmp;
418 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
419 f2fs_bug_on(sbi, 1);
420
421 memset(e, 0, sizeof(struct ino_entry));
422 e->ino = ino;
423
424 list_add_tail(&e->list, &im->ino_list);
425 if (type != ORPHAN_INO)
426 im->ino_num++;
427 }
428
429 if (type == FLUSH_INO)
430 f2fs_set_bit(devidx, (char *)&e->dirty_device);
431
432 spin_unlock(&im->ino_lock);
433 radix_tree_preload_end();
434
435 if (e != tmp)
436 kmem_cache_free(ino_entry_slab, tmp);
437 }
438
__remove_ino_entry(struct f2fs_sb_info * sbi,nid_t ino,int type)439 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
440 {
441 struct inode_management *im = &sbi->im[type];
442 struct ino_entry *e;
443
444 spin_lock(&im->ino_lock);
445 e = radix_tree_lookup(&im->ino_root, ino);
446 if (e) {
447 list_del(&e->list);
448 radix_tree_delete(&im->ino_root, ino);
449 im->ino_num--;
450 spin_unlock(&im->ino_lock);
451 kmem_cache_free(ino_entry_slab, e);
452 return;
453 }
454 spin_unlock(&im->ino_lock);
455 }
456
add_ino_entry(struct f2fs_sb_info * sbi,nid_t ino,int type)457 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
458 {
459 /* add new dirty ino entry into list */
460 __add_ino_entry(sbi, ino, 0, type);
461 }
462
remove_ino_entry(struct f2fs_sb_info * sbi,nid_t ino,int type)463 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
464 {
465 /* remove dirty ino entry from list */
466 __remove_ino_entry(sbi, ino, type);
467 }
468
469 /* mode should be APPEND_INO or UPDATE_INO */
exist_written_data(struct f2fs_sb_info * sbi,nid_t ino,int mode)470 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
471 {
472 struct inode_management *im = &sbi->im[mode];
473 struct ino_entry *e;
474
475 spin_lock(&im->ino_lock);
476 e = radix_tree_lookup(&im->ino_root, ino);
477 spin_unlock(&im->ino_lock);
478 return e ? true : false;
479 }
480
release_ino_entry(struct f2fs_sb_info * sbi,bool all)481 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
482 {
483 struct ino_entry *e, *tmp;
484 int i;
485
486 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
487 struct inode_management *im = &sbi->im[i];
488
489 spin_lock(&im->ino_lock);
490 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
491 list_del(&e->list);
492 radix_tree_delete(&im->ino_root, e->ino);
493 kmem_cache_free(ino_entry_slab, e);
494 im->ino_num--;
495 }
496 spin_unlock(&im->ino_lock);
497 }
498 }
499
set_dirty_device(struct f2fs_sb_info * sbi,nid_t ino,unsigned int devidx,int type)500 void set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
501 unsigned int devidx, int type)
502 {
503 __add_ino_entry(sbi, ino, devidx, type);
504 }
505
is_dirty_device(struct f2fs_sb_info * sbi,nid_t ino,unsigned int devidx,int type)506 bool is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
507 unsigned int devidx, int type)
508 {
509 struct inode_management *im = &sbi->im[type];
510 struct ino_entry *e;
511 bool is_dirty = false;
512
513 spin_lock(&im->ino_lock);
514 e = radix_tree_lookup(&im->ino_root, ino);
515 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
516 is_dirty = true;
517 spin_unlock(&im->ino_lock);
518 return is_dirty;
519 }
520
acquire_orphan_inode(struct f2fs_sb_info * sbi)521 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
522 {
523 struct inode_management *im = &sbi->im[ORPHAN_INO];
524 int err = 0;
525
526 spin_lock(&im->ino_lock);
527
528 #ifdef CONFIG_F2FS_FAULT_INJECTION
529 if (time_to_inject(sbi, FAULT_ORPHAN)) {
530 spin_unlock(&im->ino_lock);
531 f2fs_show_injection_info(FAULT_ORPHAN);
532 return -ENOSPC;
533 }
534 #endif
535 if (unlikely(im->ino_num >= sbi->max_orphans))
536 err = -ENOSPC;
537 else
538 im->ino_num++;
539 spin_unlock(&im->ino_lock);
540
541 return err;
542 }
543
release_orphan_inode(struct f2fs_sb_info * sbi)544 void release_orphan_inode(struct f2fs_sb_info *sbi)
545 {
546 struct inode_management *im = &sbi->im[ORPHAN_INO];
547
548 spin_lock(&im->ino_lock);
549 f2fs_bug_on(sbi, im->ino_num == 0);
550 im->ino_num--;
551 spin_unlock(&im->ino_lock);
552 }
553
add_orphan_inode(struct inode * inode)554 void add_orphan_inode(struct inode *inode)
555 {
556 /* add new orphan ino entry into list */
557 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
558 update_inode_page(inode);
559 }
560
remove_orphan_inode(struct f2fs_sb_info * sbi,nid_t ino)561 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
562 {
563 /* remove orphan entry from orphan list */
564 __remove_ino_entry(sbi, ino, ORPHAN_INO);
565 }
566
recover_orphan_inode(struct f2fs_sb_info * sbi,nid_t ino)567 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
568 {
569 struct inode *inode;
570 struct node_info ni;
571 int err = acquire_orphan_inode(sbi);
572
573 if (err) {
574 set_sbi_flag(sbi, SBI_NEED_FSCK);
575 f2fs_msg(sbi->sb, KERN_WARNING,
576 "%s: orphan failed (ino=%x), run fsck to fix.",
577 __func__, ino);
578 return err;
579 }
580
581 __add_ino_entry(sbi, ino, 0, ORPHAN_INO);
582
583 inode = f2fs_iget_retry(sbi->sb, ino);
584 if (IS_ERR(inode)) {
585 /*
586 * there should be a bug that we can't find the entry
587 * to orphan inode.
588 */
589 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
590 return PTR_ERR(inode);
591 }
592
593 clear_nlink(inode);
594
595 /* truncate all the data during iput */
596 iput(inode);
597
598 get_node_info(sbi, ino, &ni);
599
600 /* ENOMEM was fully retried in f2fs_evict_inode. */
601 if (ni.blk_addr != NULL_ADDR) {
602 set_sbi_flag(sbi, SBI_NEED_FSCK);
603 f2fs_msg(sbi->sb, KERN_WARNING,
604 "%s: orphan failed (ino=%x) by kernel, retry mount.",
605 __func__, ino);
606 return -EIO;
607 }
608 __remove_ino_entry(sbi, ino, ORPHAN_INO);
609 return 0;
610 }
611
recover_orphan_inodes(struct f2fs_sb_info * sbi)612 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
613 {
614 block_t start_blk, orphan_blocks, i, j;
615 unsigned int s_flags = sbi->sb->s_flags;
616 int err = 0;
617 #ifdef CONFIG_QUOTA
618 int quota_enabled;
619 #endif
620
621 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
622 return 0;
623
624 if (s_flags & MS_RDONLY) {
625 f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
626 sbi->sb->s_flags &= ~MS_RDONLY;
627 }
628
629 #ifdef CONFIG_QUOTA
630 /* Needed for iput() to work correctly and not trash data */
631 sbi->sb->s_flags |= MS_ACTIVE;
632
633 /* Turn on quotas so that they are updated correctly */
634 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & MS_RDONLY);
635 #endif
636
637 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
638 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
639
640 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
641
642 for (i = 0; i < orphan_blocks; i++) {
643 struct page *page = get_meta_page(sbi, start_blk + i);
644 struct f2fs_orphan_block *orphan_blk;
645
646 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
647 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
648 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
649 err = recover_orphan_inode(sbi, ino);
650 if (err) {
651 f2fs_put_page(page, 1);
652 goto out;
653 }
654 }
655 f2fs_put_page(page, 1);
656 }
657 /* clear Orphan Flag */
658 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
659 out:
660 #ifdef CONFIG_QUOTA
661 /* Turn quotas off */
662 if (quota_enabled)
663 f2fs_quota_off_umount(sbi->sb);
664 #endif
665 sbi->sb->s_flags = s_flags; /* Restore MS_RDONLY status */
666
667 return err;
668 }
669
write_orphan_inodes(struct f2fs_sb_info * sbi,block_t start_blk)670 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
671 {
672 struct list_head *head;
673 struct f2fs_orphan_block *orphan_blk = NULL;
674 unsigned int nentries = 0;
675 unsigned short index = 1;
676 unsigned short orphan_blocks;
677 struct page *page = NULL;
678 struct ino_entry *orphan = NULL;
679 struct inode_management *im = &sbi->im[ORPHAN_INO];
680
681 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
682
683 /*
684 * we don't need to do spin_lock(&im->ino_lock) here, since all the
685 * orphan inode operations are covered under f2fs_lock_op().
686 * And, spin_lock should be avoided due to page operations below.
687 */
688 head = &im->ino_list;
689
690 /* loop for each orphan inode entry and write them in Jornal block */
691 list_for_each_entry(orphan, head, list) {
692 if (!page) {
693 page = grab_meta_page(sbi, start_blk++);
694 orphan_blk =
695 (struct f2fs_orphan_block *)page_address(page);
696 memset(orphan_blk, 0, sizeof(*orphan_blk));
697 }
698
699 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
700
701 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
702 /*
703 * an orphan block is full of 1020 entries,
704 * then we need to flush current orphan blocks
705 * and bring another one in memory
706 */
707 orphan_blk->blk_addr = cpu_to_le16(index);
708 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
709 orphan_blk->entry_count = cpu_to_le32(nentries);
710 set_page_dirty(page);
711 f2fs_put_page(page, 1);
712 index++;
713 nentries = 0;
714 page = NULL;
715 }
716 }
717
718 if (page) {
719 orphan_blk->blk_addr = cpu_to_le16(index);
720 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
721 orphan_blk->entry_count = cpu_to_le32(nentries);
722 set_page_dirty(page);
723 f2fs_put_page(page, 1);
724 }
725 }
726
get_checkpoint_version(struct f2fs_sb_info * sbi,block_t cp_addr,struct f2fs_checkpoint ** cp_block,struct page ** cp_page,unsigned long long * version)727 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
728 struct f2fs_checkpoint **cp_block, struct page **cp_page,
729 unsigned long long *version)
730 {
731 unsigned long blk_size = sbi->blocksize;
732 size_t crc_offset = 0;
733 __u32 crc = 0;
734
735 *cp_page = get_meta_page(sbi, cp_addr);
736 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
737
738 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
739 if (crc_offset > (blk_size - sizeof(__le32))) {
740 f2fs_msg(sbi->sb, KERN_WARNING,
741 "invalid crc_offset: %zu", crc_offset);
742 return -EINVAL;
743 }
744
745 crc = cur_cp_crc(*cp_block);
746 if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
747 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
748 return -EINVAL;
749 }
750
751 *version = cur_cp_version(*cp_block);
752 return 0;
753 }
754
validate_checkpoint(struct f2fs_sb_info * sbi,block_t cp_addr,unsigned long long * version)755 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
756 block_t cp_addr, unsigned long long *version)
757 {
758 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
759 struct f2fs_checkpoint *cp_block = NULL;
760 unsigned long long cur_version = 0, pre_version = 0;
761 int err;
762
763 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
764 &cp_page_1, version);
765 if (err)
766 goto invalid_cp1;
767 pre_version = *version;
768
769 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
770 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
771 &cp_page_2, version);
772 if (err)
773 goto invalid_cp2;
774 cur_version = *version;
775
776 if (cur_version == pre_version) {
777 *version = cur_version;
778 f2fs_put_page(cp_page_2, 1);
779 return cp_page_1;
780 }
781 invalid_cp2:
782 f2fs_put_page(cp_page_2, 1);
783 invalid_cp1:
784 f2fs_put_page(cp_page_1, 1);
785 return NULL;
786 }
787
get_valid_checkpoint(struct f2fs_sb_info * sbi)788 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
789 {
790 struct f2fs_checkpoint *cp_block;
791 struct f2fs_super_block *fsb = sbi->raw_super;
792 struct page *cp1, *cp2, *cur_page;
793 unsigned long blk_size = sbi->blocksize;
794 unsigned long long cp1_version = 0, cp2_version = 0;
795 unsigned long long cp_start_blk_no;
796 unsigned int cp_blks = 1 + __cp_payload(sbi);
797 block_t cp_blk_no;
798 int i;
799
800 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
801 if (!sbi->ckpt)
802 return -ENOMEM;
803 /*
804 * Finding out valid cp block involves read both
805 * sets( cp pack1 and cp pack 2)
806 */
807 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
808 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
809
810 /* The second checkpoint pack should start at the next segment */
811 cp_start_blk_no += ((unsigned long long)1) <<
812 le32_to_cpu(fsb->log_blocks_per_seg);
813 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
814
815 if (cp1 && cp2) {
816 if (ver_after(cp2_version, cp1_version))
817 cur_page = cp2;
818 else
819 cur_page = cp1;
820 } else if (cp1) {
821 cur_page = cp1;
822 } else if (cp2) {
823 cur_page = cp2;
824 } else {
825 goto fail_no_cp;
826 }
827
828 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
829 memcpy(sbi->ckpt, cp_block, blk_size);
830
831 /* Sanity checking of checkpoint */
832 if (sanity_check_ckpt(sbi))
833 goto free_fail_no_cp;
834
835 if (cur_page == cp1)
836 sbi->cur_cp_pack = 1;
837 else
838 sbi->cur_cp_pack = 2;
839
840 if (cp_blks <= 1)
841 goto done;
842
843 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
844 if (cur_page == cp2)
845 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
846
847 for (i = 1; i < cp_blks; i++) {
848 void *sit_bitmap_ptr;
849 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
850
851 cur_page = get_meta_page(sbi, cp_blk_no + i);
852 sit_bitmap_ptr = page_address(cur_page);
853 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
854 f2fs_put_page(cur_page, 1);
855 }
856 done:
857 f2fs_put_page(cp1, 1);
858 f2fs_put_page(cp2, 1);
859 return 0;
860
861 free_fail_no_cp:
862 f2fs_put_page(cp1, 1);
863 f2fs_put_page(cp2, 1);
864 fail_no_cp:
865 kfree(sbi->ckpt);
866 return -EINVAL;
867 }
868
__add_dirty_inode(struct inode * inode,enum inode_type type)869 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
870 {
871 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
872 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
873
874 if (is_inode_flag_set(inode, flag))
875 return;
876
877 set_inode_flag(inode, flag);
878 if (!f2fs_is_volatile_file(inode))
879 list_add_tail(&F2FS_I(inode)->dirty_list,
880 &sbi->inode_list[type]);
881 stat_inc_dirty_inode(sbi, type);
882 }
883
__remove_dirty_inode(struct inode * inode,enum inode_type type)884 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
885 {
886 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
887
888 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
889 return;
890
891 list_del_init(&F2FS_I(inode)->dirty_list);
892 clear_inode_flag(inode, flag);
893 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
894 }
895
update_dirty_page(struct inode * inode,struct page * page)896 void update_dirty_page(struct inode *inode, struct page *page)
897 {
898 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
899 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
900
901 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
902 !S_ISLNK(inode->i_mode))
903 return;
904
905 spin_lock(&sbi->inode_lock[type]);
906 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
907 __add_dirty_inode(inode, type);
908 inode_inc_dirty_pages(inode);
909 spin_unlock(&sbi->inode_lock[type]);
910
911 SetPagePrivate(page);
912 f2fs_trace_pid(page);
913 }
914
remove_dirty_inode(struct inode * inode)915 void remove_dirty_inode(struct inode *inode)
916 {
917 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
918 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
919
920 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
921 !S_ISLNK(inode->i_mode))
922 return;
923
924 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
925 return;
926
927 spin_lock(&sbi->inode_lock[type]);
928 __remove_dirty_inode(inode, type);
929 spin_unlock(&sbi->inode_lock[type]);
930 }
931
sync_dirty_inodes(struct f2fs_sb_info * sbi,enum inode_type type)932 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
933 {
934 struct list_head *head;
935 struct inode *inode;
936 struct f2fs_inode_info *fi;
937 bool is_dir = (type == DIR_INODE);
938 unsigned long ino = 0;
939
940 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
941 get_pages(sbi, is_dir ?
942 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
943 retry:
944 if (unlikely(f2fs_cp_error(sbi)))
945 return -EIO;
946
947 spin_lock(&sbi->inode_lock[type]);
948
949 head = &sbi->inode_list[type];
950 if (list_empty(head)) {
951 spin_unlock(&sbi->inode_lock[type]);
952 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
953 get_pages(sbi, is_dir ?
954 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
955 return 0;
956 }
957 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
958 inode = igrab(&fi->vfs_inode);
959 spin_unlock(&sbi->inode_lock[type]);
960 if (inode) {
961 unsigned long cur_ino = inode->i_ino;
962
963 if (is_dir)
964 F2FS_I(inode)->cp_task = current;
965
966 filemap_fdatawrite(inode->i_mapping);
967
968 if (is_dir)
969 F2FS_I(inode)->cp_task = NULL;
970
971 iput(inode);
972 /* We need to give cpu to another writers. */
973 if (ino == cur_ino) {
974 congestion_wait(BLK_RW_ASYNC, HZ/50);
975 cond_resched();
976 } else {
977 ino = cur_ino;
978 }
979 } else {
980 /*
981 * We should submit bio, since it exists several
982 * wribacking dentry pages in the freeing inode.
983 */
984 f2fs_submit_merged_write(sbi, DATA);
985 cond_resched();
986 }
987 goto retry;
988 }
989
f2fs_sync_inode_meta(struct f2fs_sb_info * sbi)990 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
991 {
992 struct list_head *head = &sbi->inode_list[DIRTY_META];
993 struct inode *inode;
994 struct f2fs_inode_info *fi;
995 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
996
997 while (total--) {
998 if (unlikely(f2fs_cp_error(sbi)))
999 return -EIO;
1000
1001 spin_lock(&sbi->inode_lock[DIRTY_META]);
1002 if (list_empty(head)) {
1003 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1004 return 0;
1005 }
1006 fi = list_first_entry(head, struct f2fs_inode_info,
1007 gdirty_list);
1008 inode = igrab(&fi->vfs_inode);
1009 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1010 if (inode) {
1011 sync_inode_metadata(inode, 0);
1012
1013 /* it's on eviction */
1014 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1015 update_inode_page(inode);
1016 iput(inode);
1017 }
1018 }
1019 return 0;
1020 }
1021
__prepare_cp_block(struct f2fs_sb_info * sbi)1022 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1023 {
1024 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1025 struct f2fs_nm_info *nm_i = NM_I(sbi);
1026 nid_t last_nid = nm_i->next_scan_nid;
1027
1028 next_free_nid(sbi, &last_nid);
1029 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1030 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1031 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1032 ckpt->next_free_nid = cpu_to_le32(last_nid);
1033 }
1034
1035 /*
1036 * Freeze all the FS-operations for checkpoint.
1037 */
block_operations(struct f2fs_sb_info * sbi)1038 static int block_operations(struct f2fs_sb_info *sbi)
1039 {
1040 struct writeback_control wbc = {
1041 .sync_mode = WB_SYNC_ALL,
1042 .nr_to_write = LONG_MAX,
1043 .for_reclaim = 0,
1044 };
1045 struct blk_plug plug;
1046 int err = 0;
1047
1048 blk_start_plug(&plug);
1049
1050 retry_flush_dents:
1051 f2fs_lock_all(sbi);
1052 /* write all the dirty dentry pages */
1053 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1054 f2fs_unlock_all(sbi);
1055 err = sync_dirty_inodes(sbi, DIR_INODE);
1056 if (err)
1057 goto out;
1058 cond_resched();
1059 goto retry_flush_dents;
1060 }
1061
1062 /*
1063 * POR: we should ensure that there are no dirty node pages
1064 * until finishing nat/sit flush. inode->i_blocks can be updated.
1065 */
1066 down_write(&sbi->node_change);
1067
1068 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1069 up_write(&sbi->node_change);
1070 f2fs_unlock_all(sbi);
1071 err = f2fs_sync_inode_meta(sbi);
1072 if (err)
1073 goto out;
1074 cond_resched();
1075 goto retry_flush_dents;
1076 }
1077
1078 retry_flush_nodes:
1079 down_write(&sbi->node_write);
1080
1081 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1082 up_write(&sbi->node_write);
1083 err = sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1084 if (err) {
1085 up_write(&sbi->node_change);
1086 f2fs_unlock_all(sbi);
1087 goto out;
1088 }
1089 cond_resched();
1090 goto retry_flush_nodes;
1091 }
1092
1093 /*
1094 * sbi->node_change is used only for AIO write_begin path which produces
1095 * dirty node blocks and some checkpoint values by block allocation.
1096 */
1097 __prepare_cp_block(sbi);
1098 up_write(&sbi->node_change);
1099 out:
1100 blk_finish_plug(&plug);
1101 return err;
1102 }
1103
unblock_operations(struct f2fs_sb_info * sbi)1104 static void unblock_operations(struct f2fs_sb_info *sbi)
1105 {
1106 up_write(&sbi->node_write);
1107 f2fs_unlock_all(sbi);
1108 }
1109
wait_on_all_pages_writeback(struct f2fs_sb_info * sbi)1110 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1111 {
1112 DEFINE_WAIT(wait);
1113
1114 for (;;) {
1115 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1116
1117 if (!get_pages(sbi, F2FS_WB_CP_DATA))
1118 break;
1119
1120 io_schedule_timeout(5*HZ);
1121 }
1122 finish_wait(&sbi->cp_wait, &wait);
1123 }
1124
update_ckpt_flags(struct f2fs_sb_info * sbi,struct cp_control * cpc)1125 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1126 {
1127 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1128 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1129 unsigned long flags;
1130
1131 spin_lock_irqsave(&sbi->cp_lock, flags);
1132
1133 if ((cpc->reason & CP_UMOUNT) &&
1134 le32_to_cpu(ckpt->cp_pack_total_block_count) >
1135 sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1136 disable_nat_bits(sbi, false);
1137
1138 if (cpc->reason & CP_TRIMMED)
1139 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1140
1141 if (cpc->reason & CP_UMOUNT)
1142 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1143 else
1144 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1145
1146 if (cpc->reason & CP_FASTBOOT)
1147 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1148 else
1149 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1150
1151 if (orphan_num)
1152 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1153 else
1154 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1155
1156 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1157 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1158
1159 /* set this flag to activate crc|cp_ver for recovery */
1160 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1161
1162 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1163 }
1164
do_checkpoint(struct f2fs_sb_info * sbi,struct cp_control * cpc)1165 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1166 {
1167 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1168 struct f2fs_nm_info *nm_i = NM_I(sbi);
1169 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1170 block_t start_blk;
1171 unsigned int data_sum_blocks, orphan_blocks;
1172 __u32 crc32 = 0;
1173 int i;
1174 int cp_payload_blks = __cp_payload(sbi);
1175 struct super_block *sb = sbi->sb;
1176 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1177 u64 kbytes_written;
1178 int err;
1179
1180 /* Flush all the NAT/SIT pages */
1181 while (get_pages(sbi, F2FS_DIRTY_META)) {
1182 sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1183 if (unlikely(f2fs_cp_error(sbi)))
1184 return -EIO;
1185 }
1186
1187 /*
1188 * modify checkpoint
1189 * version number is already updated
1190 */
1191 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1192 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1193 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1194 ckpt->cur_node_segno[i] =
1195 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1196 ckpt->cur_node_blkoff[i] =
1197 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1198 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1199 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1200 }
1201 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1202 ckpt->cur_data_segno[i] =
1203 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1204 ckpt->cur_data_blkoff[i] =
1205 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1206 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1207 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1208 }
1209
1210 /* 2 cp + n data seg summary + orphan inode blocks */
1211 data_sum_blocks = npages_for_summary_flush(sbi, false);
1212 spin_lock_irqsave(&sbi->cp_lock, flags);
1213 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1214 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1215 else
1216 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1217 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1218
1219 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1220 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1221 orphan_blocks);
1222
1223 if (__remain_node_summaries(cpc->reason))
1224 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1225 cp_payload_blks + data_sum_blocks +
1226 orphan_blocks + NR_CURSEG_NODE_TYPE);
1227 else
1228 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1229 cp_payload_blks + data_sum_blocks +
1230 orphan_blocks);
1231
1232 /* update ckpt flag for checkpoint */
1233 update_ckpt_flags(sbi, cpc);
1234
1235 /* update SIT/NAT bitmap */
1236 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1237 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1238
1239 crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1240 *((__le32 *)((unsigned char *)ckpt +
1241 le32_to_cpu(ckpt->checksum_offset)))
1242 = cpu_to_le32(crc32);
1243
1244 start_blk = __start_cp_next_addr(sbi);
1245
1246 /* write nat bits */
1247 if (enabled_nat_bits(sbi, cpc)) {
1248 __u64 cp_ver = cur_cp_version(ckpt);
1249 block_t blk;
1250
1251 cp_ver |= ((__u64)crc32 << 32);
1252 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1253
1254 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1255 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1256 update_meta_page(sbi, nm_i->nat_bits +
1257 (i << F2FS_BLKSIZE_BITS), blk + i);
1258
1259 /* Flush all the NAT BITS pages */
1260 while (get_pages(sbi, F2FS_DIRTY_META)) {
1261 sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1262 if (unlikely(f2fs_cp_error(sbi)))
1263 return -EIO;
1264 }
1265 }
1266
1267 /* need to wait for end_io results */
1268 wait_on_all_pages_writeback(sbi);
1269 if (unlikely(f2fs_cp_error(sbi)))
1270 return -EIO;
1271
1272 /* flush all device cache */
1273 err = f2fs_flush_device_cache(sbi);
1274 if (err)
1275 return err;
1276
1277 /* write out checkpoint buffer at block 0 */
1278 update_meta_page(sbi, ckpt, start_blk++);
1279
1280 for (i = 1; i < 1 + cp_payload_blks; i++)
1281 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1282 start_blk++);
1283
1284 if (orphan_num) {
1285 write_orphan_inodes(sbi, start_blk);
1286 start_blk += orphan_blocks;
1287 }
1288
1289 write_data_summaries(sbi, start_blk);
1290 start_blk += data_sum_blocks;
1291
1292 /* Record write statistics in the hot node summary */
1293 kbytes_written = sbi->kbytes_written;
1294 if (sb->s_bdev->bd_part)
1295 kbytes_written += BD_PART_WRITTEN(sbi);
1296
1297 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1298
1299 if (__remain_node_summaries(cpc->reason)) {
1300 write_node_summaries(sbi, start_blk);
1301 start_blk += NR_CURSEG_NODE_TYPE;
1302 }
1303
1304 /* writeout checkpoint block */
1305 update_meta_page(sbi, ckpt, start_blk);
1306
1307 /* wait for previous submitted node/meta pages writeback */
1308 wait_on_all_pages_writeback(sbi);
1309
1310 if (unlikely(f2fs_cp_error(sbi)))
1311 return -EIO;
1312
1313 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1314 filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1315
1316 /* update user_block_counts */
1317 sbi->last_valid_block_count = sbi->total_valid_block_count;
1318 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1319
1320 /* Here, we only have one bio having CP pack */
1321 sync_meta_pages(sbi, META_FLUSH, LONG_MAX, FS_CP_META_IO);
1322
1323 /* wait for previous submitted meta pages writeback */
1324 wait_on_all_pages_writeback(sbi);
1325
1326 release_ino_entry(sbi, false);
1327
1328 if (unlikely(f2fs_cp_error(sbi)))
1329 return -EIO;
1330
1331 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1332 clear_sbi_flag(sbi, SBI_NEED_CP);
1333 __set_cp_next_pack(sbi);
1334
1335 /*
1336 * redirty superblock if metadata like node page or inode cache is
1337 * updated during writing checkpoint.
1338 */
1339 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1340 get_pages(sbi, F2FS_DIRTY_IMETA))
1341 set_sbi_flag(sbi, SBI_IS_DIRTY);
1342
1343 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1344
1345 return 0;
1346 }
1347
1348 /*
1349 * We guarantee that this checkpoint procedure will not fail.
1350 */
write_checkpoint(struct f2fs_sb_info * sbi,struct cp_control * cpc)1351 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1352 {
1353 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1354 unsigned long long ckpt_ver;
1355 int err = 0;
1356
1357 mutex_lock(&sbi->cp_mutex);
1358
1359 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1360 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1361 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1362 goto out;
1363 if (unlikely(f2fs_cp_error(sbi))) {
1364 err = -EIO;
1365 goto out;
1366 }
1367 if (f2fs_readonly(sbi->sb)) {
1368 err = -EROFS;
1369 goto out;
1370 }
1371
1372 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1373
1374 err = block_operations(sbi);
1375 if (err)
1376 goto out;
1377
1378 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1379
1380 f2fs_flush_merged_writes(sbi);
1381
1382 /* this is the case of multiple fstrims without any changes */
1383 if (cpc->reason & CP_DISCARD) {
1384 if (!exist_trim_candidates(sbi, cpc)) {
1385 unblock_operations(sbi);
1386 goto out;
1387 }
1388
1389 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1390 SIT_I(sbi)->dirty_sentries == 0 &&
1391 prefree_segments(sbi) == 0) {
1392 flush_sit_entries(sbi, cpc);
1393 clear_prefree_segments(sbi, cpc);
1394 unblock_operations(sbi);
1395 goto out;
1396 }
1397 }
1398
1399 /*
1400 * update checkpoint pack index
1401 * Increase the version number so that
1402 * SIT entries and seg summaries are written at correct place
1403 */
1404 ckpt_ver = cur_cp_version(ckpt);
1405 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1406
1407 /* write cached NAT/SIT entries to NAT/SIT area */
1408 flush_nat_entries(sbi, cpc);
1409 flush_sit_entries(sbi, cpc);
1410
1411 /* unlock all the fs_lock[] in do_checkpoint() */
1412 err = do_checkpoint(sbi, cpc);
1413 if (err)
1414 release_discard_addrs(sbi);
1415 else
1416 clear_prefree_segments(sbi, cpc);
1417
1418 unblock_operations(sbi);
1419 stat_inc_cp_count(sbi->stat_info);
1420
1421 if (cpc->reason & CP_RECOVERY)
1422 f2fs_msg(sbi->sb, KERN_NOTICE,
1423 "checkpoint: version = %llx", ckpt_ver);
1424
1425 /* do checkpoint periodically */
1426 f2fs_update_time(sbi, CP_TIME);
1427 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1428 out:
1429 mutex_unlock(&sbi->cp_mutex);
1430 return err;
1431 }
1432
init_ino_entry_info(struct f2fs_sb_info * sbi)1433 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1434 {
1435 int i;
1436
1437 for (i = 0; i < MAX_INO_ENTRY; i++) {
1438 struct inode_management *im = &sbi->im[i];
1439
1440 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1441 spin_lock_init(&im->ino_lock);
1442 INIT_LIST_HEAD(&im->ino_list);
1443 im->ino_num = 0;
1444 }
1445
1446 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1447 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1448 F2FS_ORPHANS_PER_BLOCK;
1449 }
1450
create_checkpoint_caches(void)1451 int __init create_checkpoint_caches(void)
1452 {
1453 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1454 sizeof(struct ino_entry));
1455 if (!ino_entry_slab)
1456 return -ENOMEM;
1457 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1458 sizeof(struct inode_entry));
1459 if (!inode_entry_slab) {
1460 kmem_cache_destroy(ino_entry_slab);
1461 return -ENOMEM;
1462 }
1463 return 0;
1464 }
1465
destroy_checkpoint_caches(void)1466 void destroy_checkpoint_caches(void)
1467 {
1468 kmem_cache_destroy(ino_entry_slab);
1469 kmem_cache_destroy(inode_entry_slab);
1470 }
1471