1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * fs/f2fs/checkpoint.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
7 */
8 #include <linux/fs.h>
9 #include <linux/bio.h>
10 #include <linux/mpage.h>
11 #include <linux/writeback.h>
12 #include <linux/blkdev.h>
13 #include <linux/f2fs_fs.h>
14 #include <linux/pagevec.h>
15 #include <linux/swap.h>
16 #include <linux/kthread.h>
17
18 #include "f2fs.h"
19 #include "node.h"
20 #include "segment.h"
21 #include "iostat.h"
22 #include <trace/events/f2fs.h>
23
24 #define DEFAULT_CHECKPOINT_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
25
26 static struct kmem_cache *ino_entry_slab;
27 struct kmem_cache *f2fs_inode_entry_slab;
28
f2fs_stop_checkpoint(struct f2fs_sb_info * sbi,bool end_io,unsigned char reason)29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io,
30 unsigned char reason)
31 {
32 f2fs_build_fault_attr(sbi, 0, 0);
33 set_ckpt_flags(sbi, CP_ERROR_FLAG);
34 if (!end_io) {
35 f2fs_flush_merged_writes(sbi);
36
37 f2fs_handle_stop(sbi, reason);
38 }
39 }
40
41 /*
42 * We guarantee no failure on the returned page.
43 */
f2fs_grab_meta_page(struct f2fs_sb_info * sbi,pgoff_t index)44 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
45 {
46 struct address_space *mapping = META_MAPPING(sbi);
47 struct page *page;
48 repeat:
49 page = f2fs_grab_cache_page(mapping, index, false);
50 if (!page) {
51 cond_resched();
52 goto repeat;
53 }
54 f2fs_wait_on_page_writeback(page, META, true, true);
55 if (!PageUptodate(page))
56 SetPageUptodate(page);
57 return page;
58 }
59
__get_meta_page(struct f2fs_sb_info * sbi,pgoff_t index,bool is_meta)60 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
61 bool is_meta)
62 {
63 struct address_space *mapping = META_MAPPING(sbi);
64 struct page *page;
65 struct f2fs_io_info fio = {
66 .sbi = sbi,
67 .type = META,
68 .op = REQ_OP_READ,
69 .op_flags = REQ_META | REQ_PRIO,
70 .old_blkaddr = index,
71 .new_blkaddr = index,
72 .encrypted_page = NULL,
73 .is_por = !is_meta,
74 };
75 int err;
76
77 if (unlikely(!is_meta))
78 fio.op_flags &= ~REQ_META;
79 repeat:
80 page = f2fs_grab_cache_page(mapping, index, false);
81 if (!page) {
82 cond_resched();
83 goto repeat;
84 }
85 if (PageUptodate(page))
86 goto out;
87
88 fio.page = page;
89
90 err = f2fs_submit_page_bio(&fio);
91 if (err) {
92 f2fs_put_page(page, 1);
93 return ERR_PTR(err);
94 }
95
96 f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
97
98 lock_page(page);
99 if (unlikely(page->mapping != mapping)) {
100 f2fs_put_page(page, 1);
101 goto repeat;
102 }
103
104 if (unlikely(!PageUptodate(page))) {
105 f2fs_handle_page_eio(sbi, page->index, META);
106 f2fs_put_page(page, 1);
107 return ERR_PTR(-EIO);
108 }
109 out:
110 return page;
111 }
112
f2fs_get_meta_page(struct f2fs_sb_info * sbi,pgoff_t index)113 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
114 {
115 return __get_meta_page(sbi, index, true);
116 }
117
f2fs_get_meta_page_retry(struct f2fs_sb_info * sbi,pgoff_t index)118 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
119 {
120 struct page *page;
121 int count = 0;
122
123 retry:
124 page = __get_meta_page(sbi, index, true);
125 if (IS_ERR(page)) {
126 if (PTR_ERR(page) == -EIO &&
127 ++count <= DEFAULT_RETRY_IO_COUNT)
128 goto retry;
129 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_META_PAGE);
130 }
131 return page;
132 }
133
134 /* for POR only */
f2fs_get_tmp_page(struct f2fs_sb_info * sbi,pgoff_t index)135 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
136 {
137 return __get_meta_page(sbi, index, false);
138 }
139
__is_bitmap_valid(struct f2fs_sb_info * sbi,block_t blkaddr,int type)140 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
141 int type)
142 {
143 struct seg_entry *se;
144 unsigned int segno, offset;
145 bool exist;
146
147 if (type == DATA_GENERIC)
148 return true;
149
150 segno = GET_SEGNO(sbi, blkaddr);
151 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
152 se = get_seg_entry(sbi, segno);
153
154 exist = f2fs_test_bit(offset, se->cur_valid_map);
155 if (exist && type == DATA_GENERIC_ENHANCE_UPDATE) {
156 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
157 blkaddr, exist);
158 set_sbi_flag(sbi, SBI_NEED_FSCK);
159 return exist;
160 }
161
162 if (!exist && type == DATA_GENERIC_ENHANCE) {
163 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
164 blkaddr, exist);
165 set_sbi_flag(sbi, SBI_NEED_FSCK);
166 dump_stack();
167 }
168 return exist;
169 }
170
f2fs_is_valid_blkaddr(struct f2fs_sb_info * sbi,block_t blkaddr,int type)171 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
172 block_t blkaddr, int type)
173 {
174 switch (type) {
175 case META_NAT:
176 break;
177 case META_SIT:
178 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
179 return false;
180 break;
181 case META_SSA:
182 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
183 blkaddr < SM_I(sbi)->ssa_blkaddr))
184 return false;
185 break;
186 case META_CP:
187 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
188 blkaddr < __start_cp_addr(sbi)))
189 return false;
190 break;
191 case META_POR:
192 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
193 blkaddr < MAIN_BLKADDR(sbi)))
194 return false;
195 break;
196 case DATA_GENERIC:
197 case DATA_GENERIC_ENHANCE:
198 case DATA_GENERIC_ENHANCE_READ:
199 case DATA_GENERIC_ENHANCE_UPDATE:
200 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
201 blkaddr < MAIN_BLKADDR(sbi))) {
202 f2fs_warn(sbi, "access invalid blkaddr:%u",
203 blkaddr);
204 set_sbi_flag(sbi, SBI_NEED_FSCK);
205 dump_stack();
206 return false;
207 } else {
208 return __is_bitmap_valid(sbi, blkaddr, type);
209 }
210 break;
211 case META_GENERIC:
212 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
213 blkaddr >= MAIN_BLKADDR(sbi)))
214 return false;
215 break;
216 default:
217 BUG();
218 }
219
220 return true;
221 }
222
223 /*
224 * Readahead CP/NAT/SIT/SSA/POR pages
225 */
f2fs_ra_meta_pages(struct f2fs_sb_info * sbi,block_t start,int nrpages,int type,bool sync)226 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
227 int type, bool sync)
228 {
229 struct page *page;
230 block_t blkno = start;
231 struct f2fs_io_info fio = {
232 .sbi = sbi,
233 .type = META,
234 .op = REQ_OP_READ,
235 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
236 .encrypted_page = NULL,
237 .in_list = false,
238 .is_por = (type == META_POR),
239 };
240 struct blk_plug plug;
241 int err;
242
243 if (unlikely(type == META_POR))
244 fio.op_flags &= ~REQ_META;
245
246 blk_start_plug(&plug);
247 for (; nrpages-- > 0; blkno++) {
248
249 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
250 goto out;
251
252 switch (type) {
253 case META_NAT:
254 if (unlikely(blkno >=
255 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
256 blkno = 0;
257 /* get nat block addr */
258 fio.new_blkaddr = current_nat_addr(sbi,
259 blkno * NAT_ENTRY_PER_BLOCK);
260 break;
261 case META_SIT:
262 if (unlikely(blkno >= TOTAL_SEGS(sbi)))
263 goto out;
264 /* get sit block addr */
265 fio.new_blkaddr = current_sit_addr(sbi,
266 blkno * SIT_ENTRY_PER_BLOCK);
267 break;
268 case META_SSA:
269 case META_CP:
270 case META_POR:
271 fio.new_blkaddr = blkno;
272 break;
273 default:
274 BUG();
275 }
276
277 page = f2fs_grab_cache_page(META_MAPPING(sbi),
278 fio.new_blkaddr, false);
279 if (!page)
280 continue;
281 if (PageUptodate(page)) {
282 f2fs_put_page(page, 1);
283 continue;
284 }
285
286 fio.page = page;
287 err = f2fs_submit_page_bio(&fio);
288 f2fs_put_page(page, err ? 1 : 0);
289
290 if (!err)
291 f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
292 }
293 out:
294 blk_finish_plug(&plug);
295 return blkno - start;
296 }
297
f2fs_ra_meta_pages_cond(struct f2fs_sb_info * sbi,pgoff_t index,unsigned int ra_blocks)298 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
299 unsigned int ra_blocks)
300 {
301 struct page *page;
302 bool readahead = false;
303
304 if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
305 return;
306
307 page = find_get_page(META_MAPPING(sbi), index);
308 if (!page || !PageUptodate(page))
309 readahead = true;
310 f2fs_put_page(page, 0);
311
312 if (readahead)
313 f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
314 }
315
__f2fs_write_meta_page(struct page * page,struct writeback_control * wbc,enum iostat_type io_type)316 static int __f2fs_write_meta_page(struct page *page,
317 struct writeback_control *wbc,
318 enum iostat_type io_type)
319 {
320 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
321
322 trace_f2fs_writepage(page, META);
323
324 if (unlikely(f2fs_cp_error(sbi))) {
325 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
326 ClearPageUptodate(page);
327 dec_page_count(sbi, F2FS_DIRTY_META);
328 unlock_page(page);
329 return 0;
330 }
331 goto redirty_out;
332 }
333 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
334 goto redirty_out;
335 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
336 goto redirty_out;
337
338 f2fs_do_write_meta_page(sbi, page, io_type);
339 dec_page_count(sbi, F2FS_DIRTY_META);
340
341 if (wbc->for_reclaim)
342 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
343
344 unlock_page(page);
345
346 if (unlikely(f2fs_cp_error(sbi)))
347 f2fs_submit_merged_write(sbi, META);
348
349 return 0;
350
351 redirty_out:
352 redirty_page_for_writepage(wbc, page);
353 return AOP_WRITEPAGE_ACTIVATE;
354 }
355
f2fs_write_meta_page(struct page * page,struct writeback_control * wbc)356 static int f2fs_write_meta_page(struct page *page,
357 struct writeback_control *wbc)
358 {
359 return __f2fs_write_meta_page(page, wbc, FS_META_IO);
360 }
361
f2fs_write_meta_pages(struct address_space * mapping,struct writeback_control * wbc)362 static int f2fs_write_meta_pages(struct address_space *mapping,
363 struct writeback_control *wbc)
364 {
365 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
366 long diff, written;
367
368 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
369 goto skip_write;
370
371 /* collect a number of dirty meta pages and write together */
372 if (wbc->sync_mode != WB_SYNC_ALL &&
373 get_pages(sbi, F2FS_DIRTY_META) <
374 nr_pages_to_skip(sbi, META))
375 goto skip_write;
376
377 /* if locked failed, cp will flush dirty pages instead */
378 if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
379 goto skip_write;
380
381 trace_f2fs_writepages(mapping->host, wbc, META);
382 diff = nr_pages_to_write(sbi, META, wbc);
383 written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
384 f2fs_up_write(&sbi->cp_global_sem);
385 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
386 return 0;
387
388 skip_write:
389 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
390 trace_f2fs_writepages(mapping->host, wbc, META);
391 return 0;
392 }
393
f2fs_sync_meta_pages(struct f2fs_sb_info * sbi,enum page_type type,long nr_to_write,enum iostat_type io_type)394 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
395 long nr_to_write, enum iostat_type io_type)
396 {
397 struct address_space *mapping = META_MAPPING(sbi);
398 pgoff_t index = 0, prev = ULONG_MAX;
399 struct pagevec pvec;
400 long nwritten = 0;
401 int nr_pages;
402 struct writeback_control wbc = {
403 .for_reclaim = 0,
404 };
405 struct blk_plug plug;
406
407 pagevec_init(&pvec);
408
409 blk_start_plug(&plug);
410
411 while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
412 PAGECACHE_TAG_DIRTY))) {
413 int i;
414
415 for (i = 0; i < nr_pages; i++) {
416 struct page *page = pvec.pages[i];
417
418 if (prev == ULONG_MAX)
419 prev = page->index - 1;
420 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
421 pagevec_release(&pvec);
422 goto stop;
423 }
424
425 lock_page(page);
426
427 if (unlikely(page->mapping != mapping)) {
428 continue_unlock:
429 unlock_page(page);
430 continue;
431 }
432 if (!PageDirty(page)) {
433 /* someone wrote it for us */
434 goto continue_unlock;
435 }
436
437 f2fs_wait_on_page_writeback(page, META, true, true);
438
439 if (!clear_page_dirty_for_io(page))
440 goto continue_unlock;
441
442 if (__f2fs_write_meta_page(page, &wbc, io_type)) {
443 unlock_page(page);
444 break;
445 }
446 nwritten++;
447 prev = page->index;
448 if (unlikely(nwritten >= nr_to_write))
449 break;
450 }
451 pagevec_release(&pvec);
452 cond_resched();
453 }
454 stop:
455 if (nwritten)
456 f2fs_submit_merged_write(sbi, type);
457
458 blk_finish_plug(&plug);
459
460 return nwritten;
461 }
462
f2fs_set_meta_page_dirty(struct page * page)463 static int f2fs_set_meta_page_dirty(struct page *page)
464 {
465 trace_f2fs_set_page_dirty(page, META);
466
467 if (!PageUptodate(page))
468 SetPageUptodate(page);
469 if (!PageDirty(page)) {
470 __set_page_dirty_nobuffers(page);
471 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
472 set_page_private_reference(page);
473 return 1;
474 }
475 return 0;
476 }
477
478 const struct address_space_operations f2fs_meta_aops = {
479 .writepage = f2fs_write_meta_page,
480 .writepages = f2fs_write_meta_pages,
481 .set_page_dirty = f2fs_set_meta_page_dirty,
482 .invalidatepage = f2fs_invalidate_page,
483 .releasepage = f2fs_release_page,
484 #ifdef CONFIG_MIGRATION
485 .migratepage = f2fs_migrate_page,
486 #endif
487 };
488
__add_ino_entry(struct f2fs_sb_info * sbi,nid_t ino,unsigned int devidx,int type)489 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
490 unsigned int devidx, int type)
491 {
492 struct inode_management *im = &sbi->im[type];
493 struct ino_entry *e = NULL, *new = NULL;
494
495 if (type == FLUSH_INO) {
496 rcu_read_lock();
497 e = radix_tree_lookup(&im->ino_root, ino);
498 rcu_read_unlock();
499 }
500
501 retry:
502 if (!e)
503 new = f2fs_kmem_cache_alloc(ino_entry_slab,
504 GFP_NOFS, true, NULL);
505
506 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
507
508 spin_lock(&im->ino_lock);
509 e = radix_tree_lookup(&im->ino_root, ino);
510 if (!e) {
511 if (!new) {
512 spin_unlock(&im->ino_lock);
513 goto retry;
514 }
515 e = new;
516 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
517 f2fs_bug_on(sbi, 1);
518
519 memset(e, 0, sizeof(struct ino_entry));
520 e->ino = ino;
521
522 list_add_tail(&e->list, &im->ino_list);
523 if (type != ORPHAN_INO)
524 im->ino_num++;
525 }
526
527 if (type == FLUSH_INO)
528 f2fs_set_bit(devidx, (char *)&e->dirty_device);
529
530 spin_unlock(&im->ino_lock);
531 radix_tree_preload_end();
532
533 if (new && e != new)
534 kmem_cache_free(ino_entry_slab, new);
535 }
536
__remove_ino_entry(struct f2fs_sb_info * sbi,nid_t ino,int type)537 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
538 {
539 struct inode_management *im = &sbi->im[type];
540 struct ino_entry *e;
541
542 spin_lock(&im->ino_lock);
543 e = radix_tree_lookup(&im->ino_root, ino);
544 if (e) {
545 list_del(&e->list);
546 radix_tree_delete(&im->ino_root, ino);
547 im->ino_num--;
548 spin_unlock(&im->ino_lock);
549 kmem_cache_free(ino_entry_slab, e);
550 return;
551 }
552 spin_unlock(&im->ino_lock);
553 }
554
f2fs_add_ino_entry(struct f2fs_sb_info * sbi,nid_t ino,int type)555 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
556 {
557 /* add new dirty ino entry into list */
558 __add_ino_entry(sbi, ino, 0, type);
559 }
560
f2fs_remove_ino_entry(struct f2fs_sb_info * sbi,nid_t ino,int type)561 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
562 {
563 /* remove dirty ino entry from list */
564 __remove_ino_entry(sbi, ino, type);
565 }
566
567 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
f2fs_exist_written_data(struct f2fs_sb_info * sbi,nid_t ino,int mode)568 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
569 {
570 struct inode_management *im = &sbi->im[mode];
571 struct ino_entry *e;
572
573 spin_lock(&im->ino_lock);
574 e = radix_tree_lookup(&im->ino_root, ino);
575 spin_unlock(&im->ino_lock);
576 return e ? true : false;
577 }
578
f2fs_release_ino_entry(struct f2fs_sb_info * sbi,bool all)579 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
580 {
581 struct ino_entry *e, *tmp;
582 int i;
583
584 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
585 struct inode_management *im = &sbi->im[i];
586
587 spin_lock(&im->ino_lock);
588 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
589 list_del(&e->list);
590 radix_tree_delete(&im->ino_root, e->ino);
591 kmem_cache_free(ino_entry_slab, e);
592 im->ino_num--;
593 }
594 spin_unlock(&im->ino_lock);
595 }
596 }
597
f2fs_set_dirty_device(struct f2fs_sb_info * sbi,nid_t ino,unsigned int devidx,int type)598 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
599 unsigned int devidx, int type)
600 {
601 __add_ino_entry(sbi, ino, devidx, type);
602 }
603
f2fs_is_dirty_device(struct f2fs_sb_info * sbi,nid_t ino,unsigned int devidx,int type)604 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
605 unsigned int devidx, int type)
606 {
607 struct inode_management *im = &sbi->im[type];
608 struct ino_entry *e;
609 bool is_dirty = false;
610
611 spin_lock(&im->ino_lock);
612 e = radix_tree_lookup(&im->ino_root, ino);
613 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
614 is_dirty = true;
615 spin_unlock(&im->ino_lock);
616 return is_dirty;
617 }
618
f2fs_acquire_orphan_inode(struct f2fs_sb_info * sbi)619 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
620 {
621 struct inode_management *im = &sbi->im[ORPHAN_INO];
622 int err = 0;
623
624 spin_lock(&im->ino_lock);
625
626 if (time_to_inject(sbi, FAULT_ORPHAN)) {
627 spin_unlock(&im->ino_lock);
628 f2fs_show_injection_info(sbi, FAULT_ORPHAN);
629 return -ENOSPC;
630 }
631
632 if (unlikely(im->ino_num >= sbi->max_orphans))
633 err = -ENOSPC;
634 else
635 im->ino_num++;
636 spin_unlock(&im->ino_lock);
637
638 return err;
639 }
640
f2fs_release_orphan_inode(struct f2fs_sb_info * sbi)641 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
642 {
643 struct inode_management *im = &sbi->im[ORPHAN_INO];
644
645 spin_lock(&im->ino_lock);
646 f2fs_bug_on(sbi, im->ino_num == 0);
647 im->ino_num--;
648 spin_unlock(&im->ino_lock);
649 }
650
f2fs_add_orphan_inode(struct inode * inode)651 void f2fs_add_orphan_inode(struct inode *inode)
652 {
653 /* add new orphan ino entry into list */
654 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
655 f2fs_update_inode_page(inode);
656 }
657
f2fs_remove_orphan_inode(struct f2fs_sb_info * sbi,nid_t ino)658 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
659 {
660 /* remove orphan entry from orphan list */
661 __remove_ino_entry(sbi, ino, ORPHAN_INO);
662 }
663
recover_orphan_inode(struct f2fs_sb_info * sbi,nid_t ino)664 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
665 {
666 struct inode *inode;
667 struct node_info ni;
668 int err;
669
670 inode = f2fs_iget_retry(sbi->sb, ino);
671 if (IS_ERR(inode)) {
672 /*
673 * there should be a bug that we can't find the entry
674 * to orphan inode.
675 */
676 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
677 return PTR_ERR(inode);
678 }
679
680 err = f2fs_dquot_initialize(inode);
681 if (err) {
682 iput(inode);
683 goto err_out;
684 }
685
686 clear_nlink(inode);
687
688 /* truncate all the data during iput */
689 iput(inode);
690
691 err = f2fs_get_node_info(sbi, ino, &ni, false);
692 if (err)
693 goto err_out;
694
695 /* ENOMEM was fully retried in f2fs_evict_inode. */
696 if (ni.blk_addr != NULL_ADDR) {
697 err = -EIO;
698 goto err_out;
699 }
700 return 0;
701
702 err_out:
703 set_sbi_flag(sbi, SBI_NEED_FSCK);
704 f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
705 __func__, ino);
706 return err;
707 }
708
f2fs_recover_orphan_inodes(struct f2fs_sb_info * sbi)709 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
710 {
711 block_t start_blk, orphan_blocks, i, j;
712 unsigned int s_flags = sbi->sb->s_flags;
713 int err = 0;
714 #ifdef CONFIG_QUOTA
715 int quota_enabled;
716 #endif
717
718 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
719 return 0;
720
721 if (bdev_read_only(sbi->sb->s_bdev)) {
722 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
723 return 0;
724 }
725
726 if (s_flags & SB_RDONLY) {
727 f2fs_info(sbi, "orphan cleanup on readonly fs");
728 sbi->sb->s_flags &= ~SB_RDONLY;
729 }
730
731 #ifdef CONFIG_QUOTA
732 /*
733 * Turn on quotas which were not enabled for read-only mounts if
734 * filesystem has quota feature, so that they are updated correctly.
735 */
736 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
737 #endif
738
739 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
740 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
741
742 f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
743
744 for (i = 0; i < orphan_blocks; i++) {
745 struct page *page;
746 struct f2fs_orphan_block *orphan_blk;
747
748 page = f2fs_get_meta_page(sbi, start_blk + i);
749 if (IS_ERR(page)) {
750 err = PTR_ERR(page);
751 goto out;
752 }
753
754 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
755 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
756 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
757
758 err = recover_orphan_inode(sbi, ino);
759 if (err) {
760 f2fs_put_page(page, 1);
761 goto out;
762 }
763 }
764 f2fs_put_page(page, 1);
765 }
766 /* clear Orphan Flag */
767 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
768 out:
769 set_sbi_flag(sbi, SBI_IS_RECOVERED);
770
771 #ifdef CONFIG_QUOTA
772 /* Turn quotas off */
773 if (quota_enabled)
774 f2fs_quota_off_umount(sbi->sb);
775 #endif
776 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
777
778 return err;
779 }
780
write_orphan_inodes(struct f2fs_sb_info * sbi,block_t start_blk)781 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
782 {
783 struct list_head *head;
784 struct f2fs_orphan_block *orphan_blk = NULL;
785 unsigned int nentries = 0;
786 unsigned short index = 1;
787 unsigned short orphan_blocks;
788 struct page *page = NULL;
789 struct ino_entry *orphan = NULL;
790 struct inode_management *im = &sbi->im[ORPHAN_INO];
791
792 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
793
794 /*
795 * we don't need to do spin_lock(&im->ino_lock) here, since all the
796 * orphan inode operations are covered under f2fs_lock_op().
797 * And, spin_lock should be avoided due to page operations below.
798 */
799 head = &im->ino_list;
800
801 /* loop for each orphan inode entry and write them in Jornal block */
802 list_for_each_entry(orphan, head, list) {
803 if (!page) {
804 page = f2fs_grab_meta_page(sbi, start_blk++);
805 orphan_blk =
806 (struct f2fs_orphan_block *)page_address(page);
807 memset(orphan_blk, 0, sizeof(*orphan_blk));
808 }
809
810 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
811
812 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
813 /*
814 * an orphan block is full of 1020 entries,
815 * then we need to flush current orphan blocks
816 * and bring another one in memory
817 */
818 orphan_blk->blk_addr = cpu_to_le16(index);
819 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
820 orphan_blk->entry_count = cpu_to_le32(nentries);
821 set_page_dirty(page);
822 f2fs_put_page(page, 1);
823 index++;
824 nentries = 0;
825 page = NULL;
826 }
827 }
828
829 if (page) {
830 orphan_blk->blk_addr = cpu_to_le16(index);
831 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
832 orphan_blk->entry_count = cpu_to_le32(nentries);
833 set_page_dirty(page);
834 f2fs_put_page(page, 1);
835 }
836 }
837
f2fs_checkpoint_chksum(struct f2fs_sb_info * sbi,struct f2fs_checkpoint * ckpt)838 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
839 struct f2fs_checkpoint *ckpt)
840 {
841 unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
842 __u32 chksum;
843
844 chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
845 if (chksum_ofs < CP_CHKSUM_OFFSET) {
846 chksum_ofs += sizeof(chksum);
847 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
848 F2FS_BLKSIZE - chksum_ofs);
849 }
850 return chksum;
851 }
852
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)853 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
854 struct f2fs_checkpoint **cp_block, struct page **cp_page,
855 unsigned long long *version)
856 {
857 size_t crc_offset = 0;
858 __u32 crc;
859
860 *cp_page = f2fs_get_meta_page(sbi, cp_addr);
861 if (IS_ERR(*cp_page))
862 return PTR_ERR(*cp_page);
863
864 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
865
866 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
867 if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
868 crc_offset > CP_CHKSUM_OFFSET) {
869 f2fs_put_page(*cp_page, 1);
870 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
871 return -EINVAL;
872 }
873
874 crc = f2fs_checkpoint_chksum(sbi, *cp_block);
875 if (crc != cur_cp_crc(*cp_block)) {
876 f2fs_put_page(*cp_page, 1);
877 f2fs_warn(sbi, "invalid crc value");
878 return -EINVAL;
879 }
880
881 *version = cur_cp_version(*cp_block);
882 return 0;
883 }
884
validate_checkpoint(struct f2fs_sb_info * sbi,block_t cp_addr,unsigned long long * version)885 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
886 block_t cp_addr, unsigned long long *version)
887 {
888 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
889 struct f2fs_checkpoint *cp_block = NULL;
890 unsigned long long cur_version = 0, pre_version = 0;
891 unsigned int cp_blocks;
892 int err;
893
894 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
895 &cp_page_1, version);
896 if (err)
897 return NULL;
898
899 cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
900
901 if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
902 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
903 le32_to_cpu(cp_block->cp_pack_total_block_count));
904 goto invalid_cp;
905 }
906 pre_version = *version;
907
908 cp_addr += cp_blocks - 1;
909 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
910 &cp_page_2, version);
911 if (err)
912 goto invalid_cp;
913 cur_version = *version;
914
915 if (cur_version == pre_version) {
916 *version = cur_version;
917 f2fs_put_page(cp_page_2, 1);
918 return cp_page_1;
919 }
920 f2fs_put_page(cp_page_2, 1);
921 invalid_cp:
922 f2fs_put_page(cp_page_1, 1);
923 return NULL;
924 }
925
f2fs_get_valid_checkpoint(struct f2fs_sb_info * sbi)926 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
927 {
928 struct f2fs_checkpoint *cp_block;
929 struct f2fs_super_block *fsb = sbi->raw_super;
930 struct page *cp1, *cp2, *cur_page;
931 unsigned long blk_size = sbi->blocksize;
932 unsigned long long cp1_version = 0, cp2_version = 0;
933 unsigned long long cp_start_blk_no;
934 unsigned int cp_blks = 1 + __cp_payload(sbi);
935 block_t cp_blk_no;
936 int i;
937 int err;
938
939 sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
940 GFP_KERNEL);
941 if (!sbi->ckpt)
942 return -ENOMEM;
943 /*
944 * Finding out valid cp block involves read both
945 * sets( cp pack 1 and cp pack 2)
946 */
947 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
948 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
949
950 /* The second checkpoint pack should start at the next segment */
951 cp_start_blk_no += ((unsigned long long)1) <<
952 le32_to_cpu(fsb->log_blocks_per_seg);
953 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
954
955 if (cp1 && cp2) {
956 if (ver_after(cp2_version, cp1_version))
957 cur_page = cp2;
958 else
959 cur_page = cp1;
960 } else if (cp1) {
961 cur_page = cp1;
962 } else if (cp2) {
963 cur_page = cp2;
964 } else {
965 err = -EFSCORRUPTED;
966 goto fail_no_cp;
967 }
968
969 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
970 memcpy(sbi->ckpt, cp_block, blk_size);
971
972 if (cur_page == cp1)
973 sbi->cur_cp_pack = 1;
974 else
975 sbi->cur_cp_pack = 2;
976
977 /* Sanity checking of checkpoint */
978 if (f2fs_sanity_check_ckpt(sbi)) {
979 err = -EFSCORRUPTED;
980 goto free_fail_no_cp;
981 }
982
983 if (cp_blks <= 1)
984 goto done;
985
986 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
987 if (cur_page == cp2)
988 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
989
990 for (i = 1; i < cp_blks; i++) {
991 void *sit_bitmap_ptr;
992 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
993
994 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
995 if (IS_ERR(cur_page)) {
996 err = PTR_ERR(cur_page);
997 goto free_fail_no_cp;
998 }
999 sit_bitmap_ptr = page_address(cur_page);
1000 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
1001 f2fs_put_page(cur_page, 1);
1002 }
1003 done:
1004 f2fs_put_page(cp1, 1);
1005 f2fs_put_page(cp2, 1);
1006 return 0;
1007
1008 free_fail_no_cp:
1009 f2fs_put_page(cp1, 1);
1010 f2fs_put_page(cp2, 1);
1011 fail_no_cp:
1012 kvfree(sbi->ckpt);
1013 return err;
1014 }
1015
__add_dirty_inode(struct inode * inode,enum inode_type type)1016 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1017 {
1018 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1019 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1020
1021 if (is_inode_flag_set(inode, flag))
1022 return;
1023
1024 set_inode_flag(inode, flag);
1025 list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1026 stat_inc_dirty_inode(sbi, type);
1027 }
1028
__remove_dirty_inode(struct inode * inode,enum inode_type type)1029 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1030 {
1031 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1032
1033 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1034 return;
1035
1036 list_del_init(&F2FS_I(inode)->dirty_list);
1037 clear_inode_flag(inode, flag);
1038 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1039 }
1040
f2fs_update_dirty_page(struct inode * inode,struct page * page)1041 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
1042 {
1043 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1044 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1045
1046 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1047 !S_ISLNK(inode->i_mode))
1048 return;
1049
1050 spin_lock(&sbi->inode_lock[type]);
1051 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1052 __add_dirty_inode(inode, type);
1053 inode_inc_dirty_pages(inode);
1054 spin_unlock(&sbi->inode_lock[type]);
1055
1056 set_page_private_reference(page);
1057 }
1058
f2fs_remove_dirty_inode(struct inode * inode)1059 void f2fs_remove_dirty_inode(struct inode *inode)
1060 {
1061 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1062 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1063
1064 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1065 !S_ISLNK(inode->i_mode))
1066 return;
1067
1068 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1069 return;
1070
1071 spin_lock(&sbi->inode_lock[type]);
1072 __remove_dirty_inode(inode, type);
1073 spin_unlock(&sbi->inode_lock[type]);
1074 }
1075
f2fs_sync_dirty_inodes(struct f2fs_sb_info * sbi,enum inode_type type,bool from_cp)1076 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1077 bool from_cp)
1078 {
1079 struct list_head *head;
1080 struct inode *inode;
1081 struct f2fs_inode_info *fi;
1082 bool is_dir = (type == DIR_INODE);
1083 unsigned long ino = 0;
1084
1085 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1086 get_pages(sbi, is_dir ?
1087 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1088 retry:
1089 if (unlikely(f2fs_cp_error(sbi))) {
1090 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1091 get_pages(sbi, is_dir ?
1092 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1093 return -EIO;
1094 }
1095
1096 spin_lock(&sbi->inode_lock[type]);
1097
1098 head = &sbi->inode_list[type];
1099 if (list_empty(head)) {
1100 spin_unlock(&sbi->inode_lock[type]);
1101 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1102 get_pages(sbi, is_dir ?
1103 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1104 return 0;
1105 }
1106 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1107 inode = igrab(&fi->vfs_inode);
1108 spin_unlock(&sbi->inode_lock[type]);
1109 if (inode) {
1110 unsigned long cur_ino = inode->i_ino;
1111
1112 if (from_cp)
1113 F2FS_I(inode)->cp_task = current;
1114 F2FS_I(inode)->wb_task = current;
1115
1116 filemap_fdatawrite(inode->i_mapping);
1117
1118 F2FS_I(inode)->wb_task = NULL;
1119 if (from_cp)
1120 F2FS_I(inode)->cp_task = NULL;
1121
1122 iput(inode);
1123 /* We need to give cpu to another writers. */
1124 if (ino == cur_ino)
1125 cond_resched();
1126 else
1127 ino = cur_ino;
1128 } else {
1129 /*
1130 * We should submit bio, since it exists several
1131 * wribacking dentry pages in the freeing inode.
1132 */
1133 f2fs_submit_merged_write(sbi, DATA);
1134 cond_resched();
1135 }
1136 goto retry;
1137 }
1138
f2fs_sync_inode_meta(struct f2fs_sb_info * sbi)1139 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1140 {
1141 struct list_head *head = &sbi->inode_list[DIRTY_META];
1142 struct inode *inode;
1143 struct f2fs_inode_info *fi;
1144 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1145
1146 while (total--) {
1147 if (unlikely(f2fs_cp_error(sbi)))
1148 return -EIO;
1149
1150 spin_lock(&sbi->inode_lock[DIRTY_META]);
1151 if (list_empty(head)) {
1152 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1153 return 0;
1154 }
1155 fi = list_first_entry(head, struct f2fs_inode_info,
1156 gdirty_list);
1157 inode = igrab(&fi->vfs_inode);
1158 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1159 if (inode) {
1160 sync_inode_metadata(inode, 0);
1161
1162 /* it's on eviction */
1163 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1164 f2fs_update_inode_page(inode);
1165 iput(inode);
1166 }
1167 }
1168 return 0;
1169 }
1170
__prepare_cp_block(struct f2fs_sb_info * sbi)1171 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1172 {
1173 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1174 struct f2fs_nm_info *nm_i = NM_I(sbi);
1175 nid_t last_nid = nm_i->next_scan_nid;
1176
1177 next_free_nid(sbi, &last_nid);
1178 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1179 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1180 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1181 ckpt->next_free_nid = cpu_to_le32(last_nid);
1182 }
1183
__need_flush_quota(struct f2fs_sb_info * sbi)1184 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1185 {
1186 bool ret = false;
1187
1188 if (!is_journalled_quota(sbi))
1189 return false;
1190
1191 if (!f2fs_down_write_trylock(&sbi->quota_sem))
1192 return true;
1193 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1194 ret = false;
1195 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1196 ret = false;
1197 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1198 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1199 ret = true;
1200 } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1201 ret = true;
1202 }
1203 f2fs_up_write(&sbi->quota_sem);
1204 return ret;
1205 }
1206
1207 /*
1208 * Freeze all the FS-operations for checkpoint.
1209 */
block_operations(struct f2fs_sb_info * sbi)1210 static int block_operations(struct f2fs_sb_info *sbi)
1211 {
1212 struct writeback_control wbc = {
1213 .sync_mode = WB_SYNC_ALL,
1214 .nr_to_write = LONG_MAX,
1215 .for_reclaim = 0,
1216 };
1217 int err = 0, cnt = 0;
1218
1219 /*
1220 * Let's flush inline_data in dirty node pages.
1221 */
1222 f2fs_flush_inline_data(sbi);
1223
1224 retry_flush_quotas:
1225 f2fs_lock_all(sbi);
1226 if (__need_flush_quota(sbi)) {
1227 int locked;
1228
1229 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1230 set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1231 set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1232 goto retry_flush_dents;
1233 }
1234 f2fs_unlock_all(sbi);
1235
1236 /* only failed during mount/umount/freeze/quotactl */
1237 locked = down_read_trylock(&sbi->sb->s_umount);
1238 f2fs_quota_sync(sbi->sb, -1);
1239 if (locked)
1240 up_read(&sbi->sb->s_umount);
1241 cond_resched();
1242 goto retry_flush_quotas;
1243 }
1244
1245 retry_flush_dents:
1246 /* write all the dirty dentry pages */
1247 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1248 f2fs_unlock_all(sbi);
1249 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1250 if (err)
1251 return err;
1252 cond_resched();
1253 goto retry_flush_quotas;
1254 }
1255
1256 /*
1257 * POR: we should ensure that there are no dirty node pages
1258 * until finishing nat/sit flush. inode->i_blocks can be updated.
1259 */
1260 f2fs_down_write(&sbi->node_change);
1261
1262 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1263 f2fs_up_write(&sbi->node_change);
1264 f2fs_unlock_all(sbi);
1265 err = f2fs_sync_inode_meta(sbi);
1266 if (err)
1267 return err;
1268 cond_resched();
1269 goto retry_flush_quotas;
1270 }
1271
1272 retry_flush_nodes:
1273 f2fs_down_write(&sbi->node_write);
1274
1275 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1276 f2fs_up_write(&sbi->node_write);
1277 atomic_inc(&sbi->wb_sync_req[NODE]);
1278 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1279 atomic_dec(&sbi->wb_sync_req[NODE]);
1280 if (err) {
1281 f2fs_up_write(&sbi->node_change);
1282 f2fs_unlock_all(sbi);
1283 return err;
1284 }
1285 cond_resched();
1286 goto retry_flush_nodes;
1287 }
1288
1289 /*
1290 * sbi->node_change is used only for AIO write_begin path which produces
1291 * dirty node blocks and some checkpoint values by block allocation.
1292 */
1293 __prepare_cp_block(sbi);
1294 f2fs_up_write(&sbi->node_change);
1295 return err;
1296 }
1297
unblock_operations(struct f2fs_sb_info * sbi)1298 static void unblock_operations(struct f2fs_sb_info *sbi)
1299 {
1300 f2fs_up_write(&sbi->node_write);
1301 f2fs_unlock_all(sbi);
1302 }
1303
f2fs_wait_on_all_pages(struct f2fs_sb_info * sbi,int type)1304 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1305 {
1306 DEFINE_WAIT(wait);
1307
1308 for (;;) {
1309 if (!get_pages(sbi, type))
1310 break;
1311
1312 if (unlikely(f2fs_cp_error(sbi) &&
1313 !is_sbi_flag_set(sbi, SBI_IS_CLOSE)))
1314 break;
1315
1316 if (type == F2FS_DIRTY_META)
1317 f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1318 FS_CP_META_IO);
1319 else if (type == F2FS_WB_CP_DATA)
1320 f2fs_submit_merged_write(sbi, DATA);
1321
1322 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1323 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1324 }
1325 finish_wait(&sbi->cp_wait, &wait);
1326 }
1327
update_ckpt_flags(struct f2fs_sb_info * sbi,struct cp_control * cpc)1328 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1329 {
1330 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1331 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1332 unsigned long flags;
1333
1334 if (cpc->reason & CP_UMOUNT) {
1335 if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1336 NM_I(sbi)->nat_bits_blocks > sbi->blocks_per_seg) {
1337 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1338 f2fs_notice(sbi, "Disable nat_bits due to no space");
1339 } else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1340 f2fs_nat_bitmap_enabled(sbi)) {
1341 f2fs_enable_nat_bits(sbi);
1342 set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1343 f2fs_notice(sbi, "Rebuild and enable nat_bits");
1344 }
1345 }
1346
1347 spin_lock_irqsave(&sbi->cp_lock, flags);
1348
1349 if (cpc->reason & CP_TRIMMED)
1350 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1351 else
1352 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1353
1354 if (cpc->reason & CP_UMOUNT)
1355 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1356 else
1357 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1358
1359 if (cpc->reason & CP_FASTBOOT)
1360 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1361 else
1362 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1363
1364 if (orphan_num)
1365 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1366 else
1367 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1368
1369 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1370 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1371
1372 if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1373 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1374 else
1375 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1376
1377 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1378 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1379 else
1380 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1381
1382 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1383 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1384 else
1385 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1386
1387 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1388 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1389 else
1390 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1391
1392 if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1393 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1394
1395 /* set this flag to activate crc|cp_ver for recovery */
1396 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1397 __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1398
1399 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1400 }
1401
commit_checkpoint(struct f2fs_sb_info * sbi,void * src,block_t blk_addr)1402 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1403 void *src, block_t blk_addr)
1404 {
1405 struct writeback_control wbc = {
1406 .for_reclaim = 0,
1407 };
1408
1409 /*
1410 * pagevec_lookup_tag and lock_page again will take
1411 * some extra time. Therefore, f2fs_update_meta_pages and
1412 * f2fs_sync_meta_pages are combined in this function.
1413 */
1414 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1415 int err;
1416
1417 f2fs_wait_on_page_writeback(page, META, true, true);
1418
1419 memcpy(page_address(page), src, PAGE_SIZE);
1420
1421 set_page_dirty(page);
1422 if (unlikely(!clear_page_dirty_for_io(page)))
1423 f2fs_bug_on(sbi, 1);
1424
1425 /* writeout cp pack 2 page */
1426 err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1427 if (unlikely(err && f2fs_cp_error(sbi))) {
1428 f2fs_put_page(page, 1);
1429 return;
1430 }
1431
1432 f2fs_bug_on(sbi, err);
1433 f2fs_put_page(page, 0);
1434
1435 /* submit checkpoint (with barrier if NOBARRIER is not set) */
1436 f2fs_submit_merged_write(sbi, META_FLUSH);
1437 }
1438
get_sectors_written(struct block_device * bdev)1439 static inline u64 get_sectors_written(struct block_device *bdev)
1440 {
1441 return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1442 }
1443
f2fs_get_sectors_written(struct f2fs_sb_info * sbi)1444 u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1445 {
1446 if (f2fs_is_multi_device(sbi)) {
1447 u64 sectors = 0;
1448 int i;
1449
1450 for (i = 0; i < sbi->s_ndevs; i++)
1451 sectors += get_sectors_written(FDEV(i).bdev);
1452
1453 return sectors;
1454 }
1455
1456 return get_sectors_written(sbi->sb->s_bdev);
1457 }
1458
do_checkpoint(struct f2fs_sb_info * sbi,struct cp_control * cpc)1459 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1460 {
1461 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1462 struct f2fs_nm_info *nm_i = NM_I(sbi);
1463 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1464 block_t start_blk;
1465 unsigned int data_sum_blocks, orphan_blocks;
1466 __u32 crc32 = 0;
1467 int i;
1468 int cp_payload_blks = __cp_payload(sbi);
1469 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1470 u64 kbytes_written;
1471 int err;
1472
1473 /* Flush all the NAT/SIT pages */
1474 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1475
1476 /* start to update checkpoint, cp ver is already updated previously */
1477 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1478 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1479 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1480 ckpt->cur_node_segno[i] =
1481 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1482 ckpt->cur_node_blkoff[i] =
1483 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1484 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1485 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1486 }
1487 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1488 ckpt->cur_data_segno[i] =
1489 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1490 ckpt->cur_data_blkoff[i] =
1491 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1492 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1493 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1494 }
1495
1496 /* 2 cp + n data seg summary + orphan inode blocks */
1497 data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1498 spin_lock_irqsave(&sbi->cp_lock, flags);
1499 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1500 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1501 else
1502 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1503 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1504
1505 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1506 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1507 orphan_blocks);
1508
1509 if (__remain_node_summaries(cpc->reason))
1510 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1511 cp_payload_blks + data_sum_blocks +
1512 orphan_blocks + NR_CURSEG_NODE_TYPE);
1513 else
1514 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1515 cp_payload_blks + data_sum_blocks +
1516 orphan_blocks);
1517
1518 /* update ckpt flag for checkpoint */
1519 update_ckpt_flags(sbi, cpc);
1520
1521 /* update SIT/NAT bitmap */
1522 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1523 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1524
1525 crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1526 *((__le32 *)((unsigned char *)ckpt +
1527 le32_to_cpu(ckpt->checksum_offset)))
1528 = cpu_to_le32(crc32);
1529
1530 start_blk = __start_cp_next_addr(sbi);
1531
1532 /* write nat bits */
1533 if ((cpc->reason & CP_UMOUNT) &&
1534 is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1535 __u64 cp_ver = cur_cp_version(ckpt);
1536 block_t blk;
1537
1538 cp_ver |= ((__u64)crc32 << 32);
1539 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1540
1541 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1542 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1543 f2fs_update_meta_page(sbi, nm_i->nat_bits +
1544 (i << F2FS_BLKSIZE_BITS), blk + i);
1545 }
1546
1547 /* write out checkpoint buffer at block 0 */
1548 f2fs_update_meta_page(sbi, ckpt, start_blk++);
1549
1550 for (i = 1; i < 1 + cp_payload_blks; i++)
1551 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1552 start_blk++);
1553
1554 if (orphan_num) {
1555 write_orphan_inodes(sbi, start_blk);
1556 start_blk += orphan_blocks;
1557 }
1558
1559 f2fs_write_data_summaries(sbi, start_blk);
1560 start_blk += data_sum_blocks;
1561
1562 /* Record write statistics in the hot node summary */
1563 kbytes_written = sbi->kbytes_written;
1564 kbytes_written += (f2fs_get_sectors_written(sbi) -
1565 sbi->sectors_written_start) >> 1;
1566 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1567
1568 if (__remain_node_summaries(cpc->reason)) {
1569 f2fs_write_node_summaries(sbi, start_blk);
1570 start_blk += NR_CURSEG_NODE_TYPE;
1571 }
1572
1573 /* update user_block_counts */
1574 sbi->last_valid_block_count = sbi->total_valid_block_count;
1575 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1576 percpu_counter_set(&sbi->rf_node_block_count, 0);
1577
1578 /* Here, we have one bio having CP pack except cp pack 2 page */
1579 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1580 /* Wait for all dirty meta pages to be submitted for IO */
1581 f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1582
1583 /* wait for previous submitted meta pages writeback */
1584 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1585
1586 /* flush all device cache */
1587 err = f2fs_flush_device_cache(sbi);
1588 if (err)
1589 return err;
1590
1591 /* barrier and flush checkpoint cp pack 2 page if it can */
1592 commit_checkpoint(sbi, ckpt, start_blk);
1593 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1594
1595 /*
1596 * invalidate intermediate page cache borrowed from meta inode which are
1597 * used for migration of encrypted, verity or compressed inode's blocks.
1598 */
1599 if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1600 f2fs_sb_has_compression(sbi))
1601 invalidate_mapping_pages(META_MAPPING(sbi),
1602 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1603
1604 f2fs_release_ino_entry(sbi, false);
1605
1606 f2fs_reset_fsync_node_info(sbi);
1607
1608 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1609 clear_sbi_flag(sbi, SBI_NEED_CP);
1610 clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1611
1612 spin_lock(&sbi->stat_lock);
1613 sbi->unusable_block_count = 0;
1614 spin_unlock(&sbi->stat_lock);
1615
1616 __set_cp_next_pack(sbi);
1617
1618 /*
1619 * redirty superblock if metadata like node page or inode cache is
1620 * updated during writing checkpoint.
1621 */
1622 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1623 get_pages(sbi, F2FS_DIRTY_IMETA))
1624 set_sbi_flag(sbi, SBI_IS_DIRTY);
1625
1626 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1627
1628 return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1629 }
1630
f2fs_write_checkpoint(struct f2fs_sb_info * sbi,struct cp_control * cpc)1631 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1632 {
1633 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1634 unsigned long long ckpt_ver;
1635 int err = 0;
1636
1637 if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1638 return -EROFS;
1639
1640 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1641 if (cpc->reason != CP_PAUSE)
1642 return 0;
1643 f2fs_warn(sbi, "Start checkpoint disabled!");
1644 }
1645 if (cpc->reason != CP_RESIZE)
1646 f2fs_down_write(&sbi->cp_global_sem);
1647
1648 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1649 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1650 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1651 goto out;
1652 if (unlikely(f2fs_cp_error(sbi))) {
1653 err = -EIO;
1654 goto out;
1655 }
1656
1657 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1658
1659 err = block_operations(sbi);
1660 if (err)
1661 goto out;
1662
1663 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1664
1665 f2fs_flush_merged_writes(sbi);
1666
1667 /* this is the case of multiple fstrims without any changes */
1668 if (cpc->reason & CP_DISCARD) {
1669 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1670 unblock_operations(sbi);
1671 goto out;
1672 }
1673
1674 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1675 SIT_I(sbi)->dirty_sentries == 0 &&
1676 prefree_segments(sbi) == 0) {
1677 f2fs_flush_sit_entries(sbi, cpc);
1678 f2fs_clear_prefree_segments(sbi, cpc);
1679 unblock_operations(sbi);
1680 goto out;
1681 }
1682 }
1683
1684 /*
1685 * update checkpoint pack index
1686 * Increase the version number so that
1687 * SIT entries and seg summaries are written at correct place
1688 */
1689 ckpt_ver = cur_cp_version(ckpt);
1690 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1691
1692 /* write cached NAT/SIT entries to NAT/SIT area */
1693 err = f2fs_flush_nat_entries(sbi, cpc);
1694 if (err) {
1695 f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1696 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1697 goto stop;
1698 }
1699
1700 f2fs_flush_sit_entries(sbi, cpc);
1701
1702 /* save inmem log status */
1703 f2fs_save_inmem_curseg(sbi);
1704
1705 err = do_checkpoint(sbi, cpc);
1706 if (err) {
1707 f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1708 f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1709 f2fs_release_discard_addrs(sbi);
1710 } else {
1711 f2fs_clear_prefree_segments(sbi, cpc);
1712 }
1713
1714 f2fs_restore_inmem_curseg(sbi);
1715 stop:
1716 unblock_operations(sbi);
1717 stat_inc_cp_count(sbi->stat_info);
1718
1719 if (cpc->reason & CP_RECOVERY)
1720 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1721
1722 /* update CP_TIME to trigger checkpoint periodically */
1723 f2fs_update_time(sbi, CP_TIME);
1724 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1725 out:
1726 if (cpc->reason != CP_RESIZE)
1727 f2fs_up_write(&sbi->cp_global_sem);
1728 return err;
1729 }
1730
f2fs_init_ino_entry_info(struct f2fs_sb_info * sbi)1731 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1732 {
1733 int i;
1734
1735 for (i = 0; i < MAX_INO_ENTRY; i++) {
1736 struct inode_management *im = &sbi->im[i];
1737
1738 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1739 spin_lock_init(&im->ino_lock);
1740 INIT_LIST_HEAD(&im->ino_list);
1741 im->ino_num = 0;
1742 }
1743
1744 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1745 NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1746 F2FS_ORPHANS_PER_BLOCK;
1747 }
1748
f2fs_create_checkpoint_caches(void)1749 int __init f2fs_create_checkpoint_caches(void)
1750 {
1751 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1752 sizeof(struct ino_entry));
1753 if (!ino_entry_slab)
1754 return -ENOMEM;
1755 f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1756 sizeof(struct inode_entry));
1757 if (!f2fs_inode_entry_slab) {
1758 kmem_cache_destroy(ino_entry_slab);
1759 return -ENOMEM;
1760 }
1761 return 0;
1762 }
1763
f2fs_destroy_checkpoint_caches(void)1764 void f2fs_destroy_checkpoint_caches(void)
1765 {
1766 kmem_cache_destroy(ino_entry_slab);
1767 kmem_cache_destroy(f2fs_inode_entry_slab);
1768 }
1769
__write_checkpoint_sync(struct f2fs_sb_info * sbi)1770 static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1771 {
1772 struct cp_control cpc = { .reason = CP_SYNC, };
1773 int err;
1774
1775 f2fs_down_write(&sbi->gc_lock);
1776 err = f2fs_write_checkpoint(sbi, &cpc);
1777 f2fs_up_write(&sbi->gc_lock);
1778
1779 return err;
1780 }
1781
__checkpoint_and_complete_reqs(struct f2fs_sb_info * sbi)1782 static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1783 {
1784 struct ckpt_req_control *cprc = &sbi->cprc_info;
1785 struct ckpt_req *req, *next;
1786 struct llist_node *dispatch_list;
1787 u64 sum_diff = 0, diff, count = 0;
1788 int ret;
1789
1790 dispatch_list = llist_del_all(&cprc->issue_list);
1791 if (!dispatch_list)
1792 return;
1793 dispatch_list = llist_reverse_order(dispatch_list);
1794
1795 ret = __write_checkpoint_sync(sbi);
1796 atomic_inc(&cprc->issued_ckpt);
1797
1798 llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1799 diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1800 req->ret = ret;
1801 complete(&req->wait);
1802
1803 sum_diff += diff;
1804 count++;
1805 }
1806 atomic_sub(count, &cprc->queued_ckpt);
1807 atomic_add(count, &cprc->total_ckpt);
1808
1809 spin_lock(&cprc->stat_lock);
1810 cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1811 if (cprc->peak_time < cprc->cur_time)
1812 cprc->peak_time = cprc->cur_time;
1813 spin_unlock(&cprc->stat_lock);
1814 }
1815
issue_checkpoint_thread(void * data)1816 static int issue_checkpoint_thread(void *data)
1817 {
1818 struct f2fs_sb_info *sbi = data;
1819 struct ckpt_req_control *cprc = &sbi->cprc_info;
1820 wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1821 repeat:
1822 if (kthread_should_stop())
1823 return 0;
1824
1825 if (!llist_empty(&cprc->issue_list))
1826 __checkpoint_and_complete_reqs(sbi);
1827
1828 wait_event_interruptible(*q,
1829 kthread_should_stop() || !llist_empty(&cprc->issue_list));
1830 goto repeat;
1831 }
1832
flush_remained_ckpt_reqs(struct f2fs_sb_info * sbi,struct ckpt_req * wait_req)1833 static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1834 struct ckpt_req *wait_req)
1835 {
1836 struct ckpt_req_control *cprc = &sbi->cprc_info;
1837
1838 if (!llist_empty(&cprc->issue_list)) {
1839 __checkpoint_and_complete_reqs(sbi);
1840 } else {
1841 /* already dispatched by issue_checkpoint_thread */
1842 if (wait_req)
1843 wait_for_completion(&wait_req->wait);
1844 }
1845 }
1846
init_ckpt_req(struct ckpt_req * req)1847 static void init_ckpt_req(struct ckpt_req *req)
1848 {
1849 memset(req, 0, sizeof(struct ckpt_req));
1850
1851 init_completion(&req->wait);
1852 req->queue_time = ktime_get();
1853 }
1854
f2fs_issue_checkpoint(struct f2fs_sb_info * sbi)1855 int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1856 {
1857 struct ckpt_req_control *cprc = &sbi->cprc_info;
1858 struct ckpt_req req;
1859 struct cp_control cpc;
1860
1861 cpc.reason = __get_cp_reason(sbi);
1862 if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1863 int ret;
1864
1865 f2fs_down_write(&sbi->gc_lock);
1866 ret = f2fs_write_checkpoint(sbi, &cpc);
1867 f2fs_up_write(&sbi->gc_lock);
1868
1869 return ret;
1870 }
1871
1872 if (!cprc->f2fs_issue_ckpt)
1873 return __write_checkpoint_sync(sbi);
1874
1875 init_ckpt_req(&req);
1876
1877 llist_add(&req.llnode, &cprc->issue_list);
1878 atomic_inc(&cprc->queued_ckpt);
1879
1880 /*
1881 * update issue_list before we wake up issue_checkpoint thread,
1882 * this smp_mb() pairs with another barrier in ___wait_event(),
1883 * see more details in comments of waitqueue_active().
1884 */
1885 smp_mb();
1886
1887 if (waitqueue_active(&cprc->ckpt_wait_queue))
1888 wake_up(&cprc->ckpt_wait_queue);
1889
1890 if (cprc->f2fs_issue_ckpt)
1891 wait_for_completion(&req.wait);
1892 else
1893 flush_remained_ckpt_reqs(sbi, &req);
1894
1895 return req.ret;
1896 }
1897
f2fs_start_ckpt_thread(struct f2fs_sb_info * sbi)1898 int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1899 {
1900 dev_t dev = sbi->sb->s_bdev->bd_dev;
1901 struct ckpt_req_control *cprc = &sbi->cprc_info;
1902
1903 if (cprc->f2fs_issue_ckpt)
1904 return 0;
1905
1906 cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1907 "f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1908 if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1909 cprc->f2fs_issue_ckpt = NULL;
1910 return -ENOMEM;
1911 }
1912
1913 set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1914
1915 return 0;
1916 }
1917
f2fs_stop_ckpt_thread(struct f2fs_sb_info * sbi)1918 void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1919 {
1920 struct ckpt_req_control *cprc = &sbi->cprc_info;
1921 struct task_struct *ckpt_task;
1922
1923 if (!cprc->f2fs_issue_ckpt)
1924 return;
1925
1926 ckpt_task = cprc->f2fs_issue_ckpt;
1927 cprc->f2fs_issue_ckpt = NULL;
1928 kthread_stop(ckpt_task);
1929
1930 f2fs_flush_ckpt_thread(sbi);
1931 }
1932
f2fs_flush_ckpt_thread(struct f2fs_sb_info * sbi)1933 void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1934 {
1935 struct ckpt_req_control *cprc = &sbi->cprc_info;
1936
1937 flush_remained_ckpt_reqs(sbi, NULL);
1938
1939 /* Let's wait for the previous dispatched checkpoint. */
1940 while (atomic_read(&cprc->queued_ckpt))
1941 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1942 }
1943
f2fs_init_ckpt_req_control(struct f2fs_sb_info * sbi)1944 void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1945 {
1946 struct ckpt_req_control *cprc = &sbi->cprc_info;
1947
1948 atomic_set(&cprc->issued_ckpt, 0);
1949 atomic_set(&cprc->total_ckpt, 0);
1950 atomic_set(&cprc->queued_ckpt, 0);
1951 cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1952 init_waitqueue_head(&cprc->ckpt_wait_queue);
1953 init_llist_head(&cprc->issue_list);
1954 spin_lock_init(&cprc->stat_lock);
1955 }
1956