x86/bugs, KVM: Extend speculation control for VIRT_SPEC_CTRL
[linux/fpc-iii.git] / fs / f2fs / checkpoint.c
blobb4dbc2f59656c18fa6c2e488e11a680c834378d7
1 /*
2 * fs/f2fs/checkpoint.c
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
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.
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>
20 #include "f2fs.h"
21 #include "node.h"
22 #include "segment.h"
23 #include "trace.h"
24 #include <trace/events/f2fs.h>
26 static struct kmem_cache *ino_entry_slab;
27 struct kmem_cache *inode_entry_slab;
29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
31 set_ckpt_flags(sbi, CP_ERROR_FLAG);
32 sbi->sb->s_flags |= MS_RDONLY;
33 if (!end_io)
34 f2fs_flush_merged_bios(sbi);
38 * We guarantee no failure on the returned page.
40 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
42 struct address_space *mapping = META_MAPPING(sbi);
43 struct page *page = NULL;
44 repeat:
45 page = f2fs_grab_cache_page(mapping, index, false);
46 if (!page) {
47 cond_resched();
48 goto repeat;
50 f2fs_wait_on_page_writeback(page, META, true);
51 if (!PageUptodate(page))
52 SetPageUptodate(page);
53 return page;
57 * We guarantee no failure on the returned page.
59 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
60 bool is_meta)
62 struct address_space *mapping = META_MAPPING(sbi);
63 struct page *page;
64 struct f2fs_io_info fio = {
65 .sbi = sbi,
66 .type = META,
67 .op = REQ_OP_READ,
68 .op_flags = READ_SYNC | REQ_META | REQ_PRIO,
69 .old_blkaddr = index,
70 .new_blkaddr = index,
71 .encrypted_page = NULL,
74 if (unlikely(!is_meta))
75 fio.op_flags &= ~REQ_META;
76 repeat:
77 page = f2fs_grab_cache_page(mapping, index, false);
78 if (!page) {
79 cond_resched();
80 goto repeat;
82 if (PageUptodate(page))
83 goto out;
85 fio.page = page;
87 if (f2fs_submit_page_bio(&fio)) {
88 f2fs_put_page(page, 1);
89 goto repeat;
92 lock_page(page);
93 if (unlikely(page->mapping != mapping)) {
94 f2fs_put_page(page, 1);
95 goto repeat;
99 * if there is any IO error when accessing device, make our filesystem
100 * readonly and make sure do not write checkpoint with non-uptodate
101 * meta page.
103 if (unlikely(!PageUptodate(page)))
104 f2fs_stop_checkpoint(sbi, false);
105 out:
106 return page;
109 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
111 return __get_meta_page(sbi, index, true);
114 /* for POR only */
115 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
117 return __get_meta_page(sbi, index, false);
120 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
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();
148 return true;
152 * Readahead CP/NAT/SIT/SSA pages
154 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
155 int type, bool sync)
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 ? (READ_SYNC | REQ_META | REQ_PRIO) : REQ_RAHEAD,
164 .encrypted_page = NULL,
166 struct blk_plug plug;
168 if (unlikely(type == META_POR))
169 fio.op_flags &= ~REQ_META;
171 blk_start_plug(&plug);
172 for (; nrpages-- > 0; blkno++) {
174 if (!is_valid_blkaddr(sbi, blkno, type))
175 goto out;
177 switch (type) {
178 case META_NAT:
179 if (unlikely(blkno >=
180 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
181 blkno = 0;
182 /* get nat block addr */
183 fio.new_blkaddr = current_nat_addr(sbi,
184 blkno * NAT_ENTRY_PER_BLOCK);
185 break;
186 case META_SIT:
187 /* get sit block addr */
188 fio.new_blkaddr = current_sit_addr(sbi,
189 blkno * SIT_ENTRY_PER_BLOCK);
190 break;
191 case META_SSA:
192 case META_CP:
193 case META_POR:
194 fio.new_blkaddr = blkno;
195 break;
196 default:
197 BUG();
200 page = f2fs_grab_cache_page(META_MAPPING(sbi),
201 fio.new_blkaddr, false);
202 if (!page)
203 continue;
204 if (PageUptodate(page)) {
205 f2fs_put_page(page, 1);
206 continue;
209 fio.page = page;
210 fio.old_blkaddr = fio.new_blkaddr;
211 f2fs_submit_page_mbio(&fio);
212 f2fs_put_page(page, 0);
214 out:
215 f2fs_submit_merged_bio(sbi, META, READ);
216 blk_finish_plug(&plug);
217 return blkno - start;
220 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
222 struct page *page;
223 bool readahead = false;
225 page = find_get_page(META_MAPPING(sbi), index);
226 if (!page || !PageUptodate(page))
227 readahead = true;
228 f2fs_put_page(page, 0);
230 if (readahead)
231 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
234 static int f2fs_write_meta_page(struct page *page,
235 struct writeback_control *wbc)
237 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
239 trace_f2fs_writepage(page, META);
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;
248 write_meta_page(sbi, page);
249 dec_page_count(sbi, F2FS_DIRTY_META);
251 if (wbc->for_reclaim)
252 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
254 unlock_page(page);
256 if (unlikely(f2fs_cp_error(sbi)))
257 f2fs_submit_merged_bio(sbi, META, WRITE);
259 return 0;
261 redirty_out:
262 redirty_page_for_writepage(wbc, page);
263 return AOP_WRITEPAGE_ACTIVATE;
266 static int f2fs_write_meta_pages(struct address_space *mapping,
267 struct writeback_control *wbc)
269 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
270 long diff, written;
272 /* collect a number of dirty meta pages and write together */
273 if (wbc->for_kupdate ||
274 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
275 goto skip_write;
277 trace_f2fs_writepages(mapping->host, wbc, META);
279 /* if mounting is failed, skip writing node pages */
280 mutex_lock(&sbi->cp_mutex);
281 diff = nr_pages_to_write(sbi, META, wbc);
282 written = sync_meta_pages(sbi, META, wbc->nr_to_write);
283 mutex_unlock(&sbi->cp_mutex);
284 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
285 return 0;
287 skip_write:
288 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
289 trace_f2fs_writepages(mapping->host, wbc, META);
290 return 0;
293 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
294 long nr_to_write)
296 struct address_space *mapping = META_MAPPING(sbi);
297 pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
298 struct pagevec pvec;
299 long nwritten = 0;
300 struct writeback_control wbc = {
301 .for_reclaim = 0,
303 struct blk_plug plug;
305 pagevec_init(&pvec, 0);
307 blk_start_plug(&plug);
309 while (index <= end) {
310 int i, nr_pages;
311 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
312 PAGECACHE_TAG_DIRTY,
313 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
314 if (unlikely(nr_pages == 0))
315 break;
317 for (i = 0; i < nr_pages; i++) {
318 struct page *page = pvec.pages[i];
320 if (prev == ULONG_MAX)
321 prev = page->index - 1;
322 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
323 pagevec_release(&pvec);
324 goto stop;
327 lock_page(page);
329 if (unlikely(page->mapping != mapping)) {
330 continue_unlock:
331 unlock_page(page);
332 continue;
334 if (!PageDirty(page)) {
335 /* someone wrote it for us */
336 goto continue_unlock;
339 f2fs_wait_on_page_writeback(page, META, true);
341 BUG_ON(PageWriteback(page));
342 if (!clear_page_dirty_for_io(page))
343 goto continue_unlock;
345 if (mapping->a_ops->writepage(page, &wbc)) {
346 unlock_page(page);
347 break;
349 nwritten++;
350 prev = page->index;
351 if (unlikely(nwritten >= nr_to_write))
352 break;
354 pagevec_release(&pvec);
355 cond_resched();
357 stop:
358 if (nwritten)
359 f2fs_submit_merged_bio(sbi, type, WRITE);
361 blk_finish_plug(&plug);
363 return nwritten;
366 static int f2fs_set_meta_page_dirty(struct page *page)
368 trace_f2fs_set_page_dirty(page, META);
370 if (!PageUptodate(page))
371 SetPageUptodate(page);
372 if (!PageDirty(page)) {
373 f2fs_set_page_dirty_nobuffers(page);
374 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
375 SetPagePrivate(page);
376 f2fs_trace_pid(page);
377 return 1;
379 return 0;
382 const struct address_space_operations f2fs_meta_aops = {
383 .writepage = f2fs_write_meta_page,
384 .writepages = f2fs_write_meta_pages,
385 .set_page_dirty = f2fs_set_meta_page_dirty,
386 .invalidatepage = f2fs_invalidate_page,
387 .releasepage = f2fs_release_page,
388 #ifdef CONFIG_MIGRATION
389 .migratepage = f2fs_migrate_page,
390 #endif
393 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
395 struct inode_management *im = &sbi->im[type];
396 struct ino_entry *e, *tmp;
398 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
399 retry:
400 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
402 spin_lock(&im->ino_lock);
403 e = radix_tree_lookup(&im->ino_root, ino);
404 if (!e) {
405 e = tmp;
406 if (radix_tree_insert(&im->ino_root, ino, e)) {
407 spin_unlock(&im->ino_lock);
408 radix_tree_preload_end();
409 goto retry;
411 memset(e, 0, sizeof(struct ino_entry));
412 e->ino = ino;
414 list_add_tail(&e->list, &im->ino_list);
415 if (type != ORPHAN_INO)
416 im->ino_num++;
418 spin_unlock(&im->ino_lock);
419 radix_tree_preload_end();
421 if (e != tmp)
422 kmem_cache_free(ino_entry_slab, tmp);
425 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
427 struct inode_management *im = &sbi->im[type];
428 struct ino_entry *e;
430 spin_lock(&im->ino_lock);
431 e = radix_tree_lookup(&im->ino_root, ino);
432 if (e) {
433 list_del(&e->list);
434 radix_tree_delete(&im->ino_root, ino);
435 im->ino_num--;
436 spin_unlock(&im->ino_lock);
437 kmem_cache_free(ino_entry_slab, e);
438 return;
440 spin_unlock(&im->ino_lock);
443 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
445 /* add new dirty ino entry into list */
446 __add_ino_entry(sbi, ino, type);
449 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
451 /* remove dirty ino entry from list */
452 __remove_ino_entry(sbi, ino, type);
455 /* mode should be APPEND_INO or UPDATE_INO */
456 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
458 struct inode_management *im = &sbi->im[mode];
459 struct ino_entry *e;
461 spin_lock(&im->ino_lock);
462 e = radix_tree_lookup(&im->ino_root, ino);
463 spin_unlock(&im->ino_lock);
464 return e ? true : false;
467 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
469 struct ino_entry *e, *tmp;
470 int i;
472 for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
473 struct inode_management *im = &sbi->im[i];
475 spin_lock(&im->ino_lock);
476 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
477 list_del(&e->list);
478 radix_tree_delete(&im->ino_root, e->ino);
479 kmem_cache_free(ino_entry_slab, e);
480 im->ino_num--;
482 spin_unlock(&im->ino_lock);
486 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
488 struct inode_management *im = &sbi->im[ORPHAN_INO];
489 int err = 0;
491 spin_lock(&im->ino_lock);
493 #ifdef CONFIG_F2FS_FAULT_INJECTION
494 if (time_to_inject(sbi, FAULT_ORPHAN)) {
495 spin_unlock(&im->ino_lock);
496 return -ENOSPC;
498 #endif
499 if (unlikely(im->ino_num >= sbi->max_orphans))
500 err = -ENOSPC;
501 else
502 im->ino_num++;
503 spin_unlock(&im->ino_lock);
505 return err;
508 void release_orphan_inode(struct f2fs_sb_info *sbi)
510 struct inode_management *im = &sbi->im[ORPHAN_INO];
512 spin_lock(&im->ino_lock);
513 f2fs_bug_on(sbi, im->ino_num == 0);
514 im->ino_num--;
515 spin_unlock(&im->ino_lock);
518 void add_orphan_inode(struct inode *inode)
520 /* add new orphan ino entry into list */
521 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO);
522 update_inode_page(inode);
525 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
527 /* remove orphan entry from orphan list */
528 __remove_ino_entry(sbi, ino, ORPHAN_INO);
531 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
533 struct inode *inode;
534 struct node_info ni;
535 int err = acquire_orphan_inode(sbi);
537 if (err) {
538 set_sbi_flag(sbi, SBI_NEED_FSCK);
539 f2fs_msg(sbi->sb, KERN_WARNING,
540 "%s: orphan failed (ino=%x), run fsck to fix.",
541 __func__, ino);
542 return err;
545 __add_ino_entry(sbi, ino, ORPHAN_INO);
547 inode = f2fs_iget_retry(sbi->sb, ino);
548 if (IS_ERR(inode)) {
550 * there should be a bug that we can't find the entry
551 * to orphan inode.
553 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
554 return PTR_ERR(inode);
557 clear_nlink(inode);
559 /* truncate all the data during iput */
560 iput(inode);
562 get_node_info(sbi, ino, &ni);
564 /* ENOMEM was fully retried in f2fs_evict_inode. */
565 if (ni.blk_addr != NULL_ADDR) {
566 set_sbi_flag(sbi, SBI_NEED_FSCK);
567 f2fs_msg(sbi->sb, KERN_WARNING,
568 "%s: orphan failed (ino=%x), run fsck to fix.",
569 __func__, ino);
570 return -EIO;
572 __remove_ino_entry(sbi, ino, ORPHAN_INO);
573 return 0;
576 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
578 block_t start_blk, orphan_blocks, i, j;
579 int err;
581 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
582 return 0;
584 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
585 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
587 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
589 for (i = 0; i < orphan_blocks; i++) {
590 struct page *page = get_meta_page(sbi, start_blk + i);
591 struct f2fs_orphan_block *orphan_blk;
593 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
594 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
595 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
596 err = recover_orphan_inode(sbi, ino);
597 if (err) {
598 f2fs_put_page(page, 1);
599 return err;
602 f2fs_put_page(page, 1);
604 /* clear Orphan Flag */
605 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
606 return 0;
609 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
611 struct list_head *head;
612 struct f2fs_orphan_block *orphan_blk = NULL;
613 unsigned int nentries = 0;
614 unsigned short index = 1;
615 unsigned short orphan_blocks;
616 struct page *page = NULL;
617 struct ino_entry *orphan = NULL;
618 struct inode_management *im = &sbi->im[ORPHAN_INO];
620 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
623 * we don't need to do spin_lock(&im->ino_lock) here, since all the
624 * orphan inode operations are covered under f2fs_lock_op().
625 * And, spin_lock should be avoided due to page operations below.
627 head = &im->ino_list;
629 /* loop for each orphan inode entry and write them in Jornal block */
630 list_for_each_entry(orphan, head, list) {
631 if (!page) {
632 page = grab_meta_page(sbi, start_blk++);
633 orphan_blk =
634 (struct f2fs_orphan_block *)page_address(page);
635 memset(orphan_blk, 0, sizeof(*orphan_blk));
638 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
640 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
642 * an orphan block is full of 1020 entries,
643 * then we need to flush current orphan blocks
644 * and bring another one in memory
646 orphan_blk->blk_addr = cpu_to_le16(index);
647 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
648 orphan_blk->entry_count = cpu_to_le32(nentries);
649 set_page_dirty(page);
650 f2fs_put_page(page, 1);
651 index++;
652 nentries = 0;
653 page = NULL;
657 if (page) {
658 orphan_blk->blk_addr = cpu_to_le16(index);
659 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
660 orphan_blk->entry_count = cpu_to_le32(nentries);
661 set_page_dirty(page);
662 f2fs_put_page(page, 1);
666 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
667 struct f2fs_checkpoint **cp_block, struct page **cp_page,
668 unsigned long long *version)
670 unsigned long blk_size = sbi->blocksize;
671 size_t crc_offset = 0;
672 __u32 crc = 0;
674 *cp_page = get_meta_page(sbi, cp_addr);
675 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
677 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
678 if (crc_offset >= blk_size) {
679 f2fs_msg(sbi->sb, KERN_WARNING,
680 "invalid crc_offset: %zu", crc_offset);
681 return -EINVAL;
684 crc = le32_to_cpu(*((__le32 *)((unsigned char *)*cp_block
685 + crc_offset)));
686 if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
687 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
688 return -EINVAL;
691 *version = cur_cp_version(*cp_block);
692 return 0;
695 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
696 block_t cp_addr, unsigned long long *version)
698 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
699 struct f2fs_checkpoint *cp_block = NULL;
700 unsigned long long cur_version = 0, pre_version = 0;
701 int err;
703 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
704 &cp_page_1, version);
705 if (err)
706 goto invalid_cp1;
707 pre_version = *version;
709 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
710 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
711 &cp_page_2, version);
712 if (err)
713 goto invalid_cp2;
714 cur_version = *version;
716 if (cur_version == pre_version) {
717 *version = cur_version;
718 f2fs_put_page(cp_page_2, 1);
719 return cp_page_1;
721 invalid_cp2:
722 f2fs_put_page(cp_page_2, 1);
723 invalid_cp1:
724 f2fs_put_page(cp_page_1, 1);
725 return NULL;
728 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
730 struct f2fs_checkpoint *cp_block;
731 struct f2fs_super_block *fsb = sbi->raw_super;
732 struct page *cp1, *cp2, *cur_page;
733 unsigned long blk_size = sbi->blocksize;
734 unsigned long long cp1_version = 0, cp2_version = 0;
735 unsigned long long cp_start_blk_no;
736 unsigned int cp_blks = 1 + __cp_payload(sbi);
737 block_t cp_blk_no;
738 int i;
740 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
741 if (!sbi->ckpt)
742 return -ENOMEM;
744 * Finding out valid cp block involves read both
745 * sets( cp pack1 and cp pack 2)
747 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
748 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
750 /* The second checkpoint pack should start at the next segment */
751 cp_start_blk_no += ((unsigned long long)1) <<
752 le32_to_cpu(fsb->log_blocks_per_seg);
753 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
755 if (cp1 && cp2) {
756 if (ver_after(cp2_version, cp1_version))
757 cur_page = cp2;
758 else
759 cur_page = cp1;
760 } else if (cp1) {
761 cur_page = cp1;
762 } else if (cp2) {
763 cur_page = cp2;
764 } else {
765 goto fail_no_cp;
768 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
769 memcpy(sbi->ckpt, cp_block, blk_size);
771 /* Sanity checking of checkpoint */
772 if (sanity_check_ckpt(sbi))
773 goto fail_no_cp;
775 if (cur_page == cp1)
776 sbi->cur_cp_pack = 1;
777 else
778 sbi->cur_cp_pack = 2;
780 if (cp_blks <= 1)
781 goto done;
783 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
784 if (cur_page == cp2)
785 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
787 for (i = 1; i < cp_blks; i++) {
788 void *sit_bitmap_ptr;
789 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
791 cur_page = get_meta_page(sbi, cp_blk_no + i);
792 sit_bitmap_ptr = page_address(cur_page);
793 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
794 f2fs_put_page(cur_page, 1);
796 done:
797 f2fs_put_page(cp1, 1);
798 f2fs_put_page(cp2, 1);
799 return 0;
801 fail_no_cp:
802 kfree(sbi->ckpt);
803 return -EINVAL;
806 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
808 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
809 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
811 if (is_inode_flag_set(inode, flag))
812 return;
814 set_inode_flag(inode, flag);
815 list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
816 stat_inc_dirty_inode(sbi, type);
819 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
821 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
823 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
824 return;
826 list_del_init(&F2FS_I(inode)->dirty_list);
827 clear_inode_flag(inode, flag);
828 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
831 void update_dirty_page(struct inode *inode, struct page *page)
833 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
834 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
836 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
837 !S_ISLNK(inode->i_mode))
838 return;
840 spin_lock(&sbi->inode_lock[type]);
841 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
842 __add_dirty_inode(inode, type);
843 inode_inc_dirty_pages(inode);
844 spin_unlock(&sbi->inode_lock[type]);
846 SetPagePrivate(page);
847 f2fs_trace_pid(page);
850 void remove_dirty_inode(struct inode *inode)
852 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
853 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
855 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
856 !S_ISLNK(inode->i_mode))
857 return;
859 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
860 return;
862 spin_lock(&sbi->inode_lock[type]);
863 __remove_dirty_inode(inode, type);
864 spin_unlock(&sbi->inode_lock[type]);
867 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
869 struct list_head *head;
870 struct inode *inode;
871 struct f2fs_inode_info *fi;
872 bool is_dir = (type == DIR_INODE);
874 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
875 get_pages(sbi, is_dir ?
876 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
877 retry:
878 if (unlikely(f2fs_cp_error(sbi)))
879 return -EIO;
881 spin_lock(&sbi->inode_lock[type]);
883 head = &sbi->inode_list[type];
884 if (list_empty(head)) {
885 spin_unlock(&sbi->inode_lock[type]);
886 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
887 get_pages(sbi, is_dir ?
888 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
889 return 0;
891 fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
892 inode = igrab(&fi->vfs_inode);
893 spin_unlock(&sbi->inode_lock[type]);
894 if (inode) {
895 filemap_fdatawrite(inode->i_mapping);
896 iput(inode);
897 } else {
899 * We should submit bio, since it exists several
900 * wribacking dentry pages in the freeing inode.
902 f2fs_submit_merged_bio(sbi, DATA, WRITE);
903 cond_resched();
905 goto retry;
908 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
910 struct list_head *head = &sbi->inode_list[DIRTY_META];
911 struct inode *inode;
912 struct f2fs_inode_info *fi;
913 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
915 while (total--) {
916 if (unlikely(f2fs_cp_error(sbi)))
917 return -EIO;
919 spin_lock(&sbi->inode_lock[DIRTY_META]);
920 if (list_empty(head)) {
921 spin_unlock(&sbi->inode_lock[DIRTY_META]);
922 return 0;
924 fi = list_entry(head->next, struct f2fs_inode_info,
925 gdirty_list);
926 inode = igrab(&fi->vfs_inode);
927 spin_unlock(&sbi->inode_lock[DIRTY_META]);
928 if (inode) {
929 update_inode_page(inode);
930 iput(inode);
933 return 0;
937 * Freeze all the FS-operations for checkpoint.
939 static int block_operations(struct f2fs_sb_info *sbi)
941 struct writeback_control wbc = {
942 .sync_mode = WB_SYNC_ALL,
943 .nr_to_write = LONG_MAX,
944 .for_reclaim = 0,
946 struct blk_plug plug;
947 int err = 0;
949 blk_start_plug(&plug);
951 retry_flush_dents:
952 f2fs_lock_all(sbi);
953 /* write all the dirty dentry pages */
954 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
955 f2fs_unlock_all(sbi);
956 err = sync_dirty_inodes(sbi, DIR_INODE);
957 if (err)
958 goto out;
959 goto retry_flush_dents;
962 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
963 f2fs_unlock_all(sbi);
964 err = f2fs_sync_inode_meta(sbi);
965 if (err)
966 goto out;
967 goto retry_flush_dents;
971 * POR: we should ensure that there are no dirty node pages
972 * until finishing nat/sit flush.
974 retry_flush_nodes:
975 down_write(&sbi->node_write);
977 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
978 up_write(&sbi->node_write);
979 err = sync_node_pages(sbi, &wbc);
980 if (err) {
981 f2fs_unlock_all(sbi);
982 goto out;
984 goto retry_flush_nodes;
986 out:
987 blk_finish_plug(&plug);
988 return err;
991 static void unblock_operations(struct f2fs_sb_info *sbi)
993 up_write(&sbi->node_write);
995 build_free_nids(sbi);
996 f2fs_unlock_all(sbi);
999 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1001 DEFINE_WAIT(wait);
1003 for (;;) {
1004 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1006 if (!atomic_read(&sbi->nr_wb_bios))
1007 break;
1009 io_schedule_timeout(5*HZ);
1011 finish_wait(&sbi->cp_wait, &wait);
1014 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1016 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1017 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1019 spin_lock(&sbi->cp_lock);
1021 if (cpc->reason == CP_UMOUNT)
1022 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1023 else
1024 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1026 if (cpc->reason == CP_FASTBOOT)
1027 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1028 else
1029 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1031 if (orphan_num)
1032 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1033 else
1034 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1036 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1037 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1039 /* set this flag to activate crc|cp_ver for recovery */
1040 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1042 spin_unlock(&sbi->cp_lock);
1045 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1047 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1048 struct f2fs_nm_info *nm_i = NM_I(sbi);
1049 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1050 nid_t last_nid = nm_i->next_scan_nid;
1051 block_t start_blk;
1052 unsigned int data_sum_blocks, orphan_blocks;
1053 __u32 crc32 = 0;
1054 int i;
1055 int cp_payload_blks = __cp_payload(sbi);
1056 struct super_block *sb = sbi->sb;
1057 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1058 u64 kbytes_written;
1060 /* Flush all the NAT/SIT pages */
1061 while (get_pages(sbi, F2FS_DIRTY_META)) {
1062 sync_meta_pages(sbi, META, LONG_MAX);
1063 if (unlikely(f2fs_cp_error(sbi)))
1064 return -EIO;
1067 next_free_nid(sbi, &last_nid);
1070 * modify checkpoint
1071 * version number is already updated
1073 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1074 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1075 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1076 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1077 ckpt->cur_node_segno[i] =
1078 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1079 ckpt->cur_node_blkoff[i] =
1080 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1081 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1082 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1084 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1085 ckpt->cur_data_segno[i] =
1086 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1087 ckpt->cur_data_blkoff[i] =
1088 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1089 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1090 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1093 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1094 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1095 ckpt->next_free_nid = cpu_to_le32(last_nid);
1097 /* 2 cp + n data seg summary + orphan inode blocks */
1098 data_sum_blocks = npages_for_summary_flush(sbi, false);
1099 spin_lock(&sbi->cp_lock);
1100 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1101 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1102 else
1103 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1104 spin_unlock(&sbi->cp_lock);
1106 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1107 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1108 orphan_blocks);
1110 if (__remain_node_summaries(cpc->reason))
1111 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1112 cp_payload_blks + data_sum_blocks +
1113 orphan_blocks + NR_CURSEG_NODE_TYPE);
1114 else
1115 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1116 cp_payload_blks + data_sum_blocks +
1117 orphan_blocks);
1119 /* update ckpt flag for checkpoint */
1120 update_ckpt_flags(sbi, cpc);
1122 /* update SIT/NAT bitmap */
1123 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1124 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1126 crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1127 *((__le32 *)((unsigned char *)ckpt +
1128 le32_to_cpu(ckpt->checksum_offset)))
1129 = cpu_to_le32(crc32);
1131 start_blk = __start_cp_next_addr(sbi);
1133 /* need to wait for end_io results */
1134 wait_on_all_pages_writeback(sbi);
1135 if (unlikely(f2fs_cp_error(sbi)))
1136 return -EIO;
1138 /* write out checkpoint buffer at block 0 */
1139 update_meta_page(sbi, ckpt, start_blk++);
1141 for (i = 1; i < 1 + cp_payload_blks; i++)
1142 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1143 start_blk++);
1145 if (orphan_num) {
1146 write_orphan_inodes(sbi, start_blk);
1147 start_blk += orphan_blocks;
1150 write_data_summaries(sbi, start_blk);
1151 start_blk += data_sum_blocks;
1153 /* Record write statistics in the hot node summary */
1154 kbytes_written = sbi->kbytes_written;
1155 if (sb->s_bdev->bd_part)
1156 kbytes_written += BD_PART_WRITTEN(sbi);
1158 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1160 if (__remain_node_summaries(cpc->reason)) {
1161 write_node_summaries(sbi, start_blk);
1162 start_blk += NR_CURSEG_NODE_TYPE;
1165 /* writeout checkpoint block */
1166 update_meta_page(sbi, ckpt, start_blk);
1168 /* wait for previous submitted node/meta pages writeback */
1169 wait_on_all_pages_writeback(sbi);
1171 if (unlikely(f2fs_cp_error(sbi)))
1172 return -EIO;
1174 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1175 filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1177 /* update user_block_counts */
1178 sbi->last_valid_block_count = sbi->total_valid_block_count;
1179 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1181 /* Here, we only have one bio having CP pack */
1182 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1184 /* wait for previous submitted meta pages writeback */
1185 wait_on_all_pages_writeback(sbi);
1187 release_ino_entry(sbi, false);
1189 if (unlikely(f2fs_cp_error(sbi)))
1190 return -EIO;
1192 clear_prefree_segments(sbi, cpc);
1193 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1194 clear_sbi_flag(sbi, SBI_NEED_CP);
1195 __set_cp_next_pack(sbi);
1198 * redirty superblock if metadata like node page or inode cache is
1199 * updated during writing checkpoint.
1201 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1202 get_pages(sbi, F2FS_DIRTY_IMETA))
1203 set_sbi_flag(sbi, SBI_IS_DIRTY);
1205 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1207 return 0;
1211 * We guarantee that this checkpoint procedure will not fail.
1213 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1215 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1216 unsigned long long ckpt_ver;
1217 int err = 0;
1219 mutex_lock(&sbi->cp_mutex);
1221 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1222 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1223 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1224 goto out;
1225 if (unlikely(f2fs_cp_error(sbi))) {
1226 err = -EIO;
1227 goto out;
1229 if (f2fs_readonly(sbi->sb)) {
1230 err = -EROFS;
1231 goto out;
1234 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1236 err = block_operations(sbi);
1237 if (err)
1238 goto out;
1240 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1242 f2fs_flush_merged_bios(sbi);
1244 /* this is the case of multiple fstrims without any changes */
1245 if (cpc->reason == CP_DISCARD && !is_sbi_flag_set(sbi, SBI_IS_DIRTY)) {
1246 f2fs_bug_on(sbi, NM_I(sbi)->dirty_nat_cnt);
1247 f2fs_bug_on(sbi, SIT_I(sbi)->dirty_sentries);
1248 f2fs_bug_on(sbi, prefree_segments(sbi));
1249 flush_sit_entries(sbi, cpc);
1250 clear_prefree_segments(sbi, cpc);
1251 f2fs_wait_all_discard_bio(sbi);
1252 unblock_operations(sbi);
1253 goto out;
1257 * update checkpoint pack index
1258 * Increase the version number so that
1259 * SIT entries and seg summaries are written at correct place
1261 ckpt_ver = cur_cp_version(ckpt);
1262 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1264 /* write cached NAT/SIT entries to NAT/SIT area */
1265 flush_nat_entries(sbi);
1266 flush_sit_entries(sbi, cpc);
1268 /* unlock all the fs_lock[] in do_checkpoint() */
1269 err = do_checkpoint(sbi, cpc);
1271 f2fs_wait_all_discard_bio(sbi);
1273 unblock_operations(sbi);
1274 stat_inc_cp_count(sbi->stat_info);
1276 if (cpc->reason == CP_RECOVERY)
1277 f2fs_msg(sbi->sb, KERN_NOTICE,
1278 "checkpoint: version = %llx", ckpt_ver);
1280 /* do checkpoint periodically */
1281 f2fs_update_time(sbi, CP_TIME);
1282 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1283 out:
1284 mutex_unlock(&sbi->cp_mutex);
1285 return err;
1288 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1290 int i;
1292 for (i = 0; i < MAX_INO_ENTRY; i++) {
1293 struct inode_management *im = &sbi->im[i];
1295 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1296 spin_lock_init(&im->ino_lock);
1297 INIT_LIST_HEAD(&im->ino_list);
1298 im->ino_num = 0;
1301 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1302 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1303 F2FS_ORPHANS_PER_BLOCK;
1306 int __init create_checkpoint_caches(void)
1308 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1309 sizeof(struct ino_entry));
1310 if (!ino_entry_slab)
1311 return -ENOMEM;
1312 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1313 sizeof(struct inode_entry));
1314 if (!inode_entry_slab) {
1315 kmem_cache_destroy(ino_entry_slab);
1316 return -ENOMEM;
1318 return 0;
1321 void destroy_checkpoint_caches(void)
1323 kmem_cache_destroy(ino_entry_slab);
1324 kmem_cache_destroy(inode_entry_slab);