blob: 38f4a2245085ae5615918c240f39f0b8bd5e5acd [file] [log] [blame]
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +09001/*
Jaegeuk Kim127e6702012-11-02 17:08:18 +09002 * fs/f2fs/checkpoint.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/bio.h>
13#include <linux/mpage.h>
14#include <linux/writeback.h>
15#include <linux/blkdev.h>
16#include <linux/f2fs_fs.h>
17#include <linux/pagevec.h>
18#include <linux/swap.h>
19
20#include "f2fs.h"
21#include "node.h"
22#include "segment.h"
Namjae Jeon2af4bd62013-04-23 18:26:54 +090023#include <trace/events/f2fs.h>
Jaegeuk Kim127e6702012-11-02 17:08:18 +090024
25static struct kmem_cache *orphan_entry_slab;
26static struct kmem_cache *inode_entry_slab;
27
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +090028/*
Jaegeuk Kim127e6702012-11-02 17:08:18 +090029 * We guarantee no failure on the returned page.
30 */
31struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
32{
33 struct address_space *mapping = sbi->meta_inode->i_mapping;
34 struct page *page = NULL;
35repeat:
36 page = grab_cache_page(mapping, index);
37 if (!page) {
38 cond_resched();
39 goto repeat;
40 }
41
42 /* We wait writeback only inside grab_meta_page() */
43 wait_on_page_writeback(page);
44 SetPageUptodate(page);
45 return page;
46}
47
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +090048/*
Jaegeuk Kim127e6702012-11-02 17:08:18 +090049 * We guarantee no failure on the returned page.
50 */
51struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
52{
53 struct address_space *mapping = sbi->meta_inode->i_mapping;
54 struct page *page;
55repeat:
56 page = grab_cache_page(mapping, index);
57 if (!page) {
58 cond_resched();
59 goto repeat;
60 }
Jaegeuk Kim393ff912013-03-08 21:29:23 +090061 if (PageUptodate(page))
62 goto out;
Jaegeuk Kim127e6702012-11-02 17:08:18 +090063
Jaegeuk Kim93dfe2a2013-11-30 12:51:14 +090064 if (f2fs_submit_page_bio(sbi, page, index,
65 READ_SYNC | REQ_META | REQ_PRIO))
Jaegeuk Kim393ff912013-03-08 21:29:23 +090066 goto repeat;
67
68 lock_page(page);
Jaegeuk Kimafcb7ca02013-04-26 11:55:17 +090069 if (page->mapping != mapping) {
70 f2fs_put_page(page, 1);
71 goto repeat;
72 }
Jaegeuk Kim393ff912013-03-08 21:29:23 +090073out:
74 mark_page_accessed(page);
Jaegeuk Kim127e6702012-11-02 17:08:18 +090075 return page;
76}
77
78static int f2fs_write_meta_page(struct page *page,
79 struct writeback_control *wbc)
80{
81 struct inode *inode = page->mapping->host;
82 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
Jaegeuk Kim577e3492013-01-24 19:56:11 +090083
84 /* Should not write any meta pages, if any IO error was occurred */
Jaegeuk Kim87a9bd22013-10-16 15:09:26 +090085 if (wbc->for_reclaim || sbi->por_doing ||
Jaegeuk Kim577e3492013-01-24 19:56:11 +090086 is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)) {
87 dec_page_count(sbi, F2FS_DIRTY_META);
88 wbc->pages_skipped++;
89 set_page_dirty(page);
90 return AOP_WRITEPAGE_ACTIVATE;
91 }
Jaegeuk Kim127e6702012-11-02 17:08:18 +090092
93 wait_on_page_writeback(page);
94
Jaegeuk Kim577e3492013-01-24 19:56:11 +090095 write_meta_page(sbi, page);
Jaegeuk Kim127e6702012-11-02 17:08:18 +090096 dec_page_count(sbi, F2FS_DIRTY_META);
Jaegeuk Kim577e3492013-01-24 19:56:11 +090097 unlock_page(page);
98 return 0;
Jaegeuk Kim127e6702012-11-02 17:08:18 +090099}
100
101static int f2fs_write_meta_pages(struct address_space *mapping,
102 struct writeback_control *wbc)
103{
104 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
105 struct block_device *bdev = sbi->sb->s_bdev;
106 long written;
107
108 if (wbc->for_kupdate)
109 return 0;
110
111 if (get_pages(sbi, F2FS_DIRTY_META) == 0)
112 return 0;
113
114 /* if mounting is failed, skip writing node pages */
115 mutex_lock(&sbi->cp_mutex);
116 written = sync_meta_pages(sbi, META, bio_get_nr_vecs(bdev));
117 mutex_unlock(&sbi->cp_mutex);
118 wbc->nr_to_write -= written;
119 return 0;
120}
121
122long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
123 long nr_to_write)
124{
125 struct address_space *mapping = sbi->meta_inode->i_mapping;
126 pgoff_t index = 0, end = LONG_MAX;
127 struct pagevec pvec;
128 long nwritten = 0;
129 struct writeback_control wbc = {
130 .for_reclaim = 0,
131 };
132
133 pagevec_init(&pvec, 0);
134
135 while (index <= end) {
136 int i, nr_pages;
137 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
138 PAGECACHE_TAG_DIRTY,
139 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
140 if (nr_pages == 0)
141 break;
142
143 for (i = 0; i < nr_pages; i++) {
144 struct page *page = pvec.pages[i];
145 lock_page(page);
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900146 f2fs_bug_on(page->mapping != mapping);
147 f2fs_bug_on(!PageDirty(page));
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900148 clear_page_dirty_for_io(page);
Jaegeuk Kim577e3492013-01-24 19:56:11 +0900149 if (f2fs_write_meta_page(page, &wbc)) {
150 unlock_page(page);
151 break;
152 }
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900153 if (nwritten++ >= nr_to_write)
154 break;
155 }
156 pagevec_release(&pvec);
157 cond_resched();
158 }
159
160 if (nwritten)
Jaegeuk Kim93dfe2a2013-11-30 12:51:14 +0900161 f2fs_submit_merged_bio(sbi, type, nr_to_write == LONG_MAX,
162 WRITE);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900163
164 return nwritten;
165}
166
167static int f2fs_set_meta_page_dirty(struct page *page)
168{
169 struct address_space *mapping = page->mapping;
170 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
171
Jaegeuk Kim26c6b882013-10-24 17:53:29 +0900172 trace_f2fs_set_page_dirty(page, META);
173
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900174 SetPageUptodate(page);
175 if (!PageDirty(page)) {
176 __set_page_dirty_nobuffers(page);
177 inc_page_count(sbi, F2FS_DIRTY_META);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900178 return 1;
179 }
180 return 0;
181}
182
183const struct address_space_operations f2fs_meta_aops = {
184 .writepage = f2fs_write_meta_page,
185 .writepages = f2fs_write_meta_pages,
186 .set_page_dirty = f2fs_set_meta_page_dirty,
187};
188
Jaegeuk Kimcbd56e72013-07-30 11:36:53 +0900189int acquire_orphan_inode(struct f2fs_sb_info *sbi)
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900190{
191 unsigned int max_orphans;
192 int err = 0;
193
194 /*
Chao Yu6947eea2013-11-29 16:37:00 +0800195 * considering 512 blocks in a segment 8 blocks are needed for cp
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900196 * and log segment summaries. Remaining blocks are used to keep
197 * orphan entries with the limitation one reserved segment
Chao Yu6947eea2013-11-29 16:37:00 +0800198 * for cp pack we can have max 1020*504 orphan entries
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900199 */
Chao Yu6947eea2013-11-29 16:37:00 +0800200 max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
201 * F2FS_ORPHANS_PER_BLOCK;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900202 mutex_lock(&sbi->orphan_inode_mutex);
203 if (sbi->n_orphans >= max_orphans)
204 err = -ENOSPC;
Jaegeuk Kimcbd56e72013-07-30 11:36:53 +0900205 else
206 sbi->n_orphans++;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900207 mutex_unlock(&sbi->orphan_inode_mutex);
208 return err;
209}
210
Jaegeuk Kimcbd56e72013-07-30 11:36:53 +0900211void release_orphan_inode(struct f2fs_sb_info *sbi)
212{
213 mutex_lock(&sbi->orphan_inode_mutex);
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900214 f2fs_bug_on(sbi->n_orphans == 0);
Jaegeuk Kimcbd56e72013-07-30 11:36:53 +0900215 sbi->n_orphans--;
216 mutex_unlock(&sbi->orphan_inode_mutex);
217}
218
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900219void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
220{
221 struct list_head *head, *this;
222 struct orphan_inode_entry *new = NULL, *orphan = NULL;
223
224 mutex_lock(&sbi->orphan_inode_mutex);
225 head = &sbi->orphan_inode_list;
226 list_for_each(this, head) {
227 orphan = list_entry(this, struct orphan_inode_entry, list);
228 if (orphan->ino == ino)
229 goto out;
230 if (orphan->ino > ino)
231 break;
232 orphan = NULL;
233 }
Gu Zheng7bd59382013-10-22 14:52:26 +0800234
235 new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900236 new->ino = ino;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900237
238 /* add new_oentry into list which is sorted by inode number */
majianpenga2617dc2013-01-29 16:19:02 +0800239 if (orphan)
240 list_add(&new->list, this->prev);
241 else
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900242 list_add_tail(&new->list, head);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900243out:
244 mutex_unlock(&sbi->orphan_inode_mutex);
245}
246
247void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
248{
Gu Zheng60ed9a02013-07-19 16:24:06 +0800249 struct list_head *head;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900250 struct orphan_inode_entry *orphan;
251
252 mutex_lock(&sbi->orphan_inode_mutex);
253 head = &sbi->orphan_inode_list;
Gu Zheng60ed9a02013-07-19 16:24:06 +0800254 list_for_each_entry(orphan, head, list) {
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900255 if (orphan->ino == ino) {
256 list_del(&orphan->list);
257 kmem_cache_free(orphan_entry_slab, orphan);
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900258 f2fs_bug_on(sbi->n_orphans == 0);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900259 sbi->n_orphans--;
260 break;
261 }
262 }
263 mutex_unlock(&sbi->orphan_inode_mutex);
264}
265
266static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
267{
268 struct inode *inode = f2fs_iget(sbi->sb, ino);
Jaegeuk Kim5d56b672013-10-29 15:14:54 +0900269 f2fs_bug_on(IS_ERR(inode));
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900270 clear_nlink(inode);
271
272 /* truncate all the data during iput */
273 iput(inode);
274}
275
Chao Yu8f99a942013-11-28 15:43:43 +0800276void recover_orphan_inodes(struct f2fs_sb_info *sbi)
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900277{
278 block_t start_blk, orphan_blkaddr, i, j;
279
Jaegeuk Kim25ca9232012-11-28 16:12:41 +0900280 if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
Chao Yu8f99a942013-11-28 15:43:43 +0800281 return;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900282
Haicheng Liaabe5132013-10-23 12:39:32 +0800283 sbi->por_doing = true;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900284 start_blk = __start_cp_addr(sbi) + 1;
285 orphan_blkaddr = __start_sum_addr(sbi) - 1;
286
287 for (i = 0; i < orphan_blkaddr; i++) {
288 struct page *page = get_meta_page(sbi, start_blk + i);
289 struct f2fs_orphan_block *orphan_blk;
290
291 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
292 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
293 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
294 recover_orphan_inode(sbi, ino);
295 }
296 f2fs_put_page(page, 1);
297 }
298 /* clear Orphan Flag */
Jaegeuk Kim25ca9232012-11-28 16:12:41 +0900299 clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
Haicheng Liaabe5132013-10-23 12:39:32 +0800300 sbi->por_doing = false;
Chao Yu8f99a942013-11-28 15:43:43 +0800301 return;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900302}
303
304static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
305{
Gu Zheng502c6e02013-11-19 18:03:58 +0800306 struct list_head *head;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900307 struct f2fs_orphan_block *orphan_blk = NULL;
308 struct page *page = NULL;
309 unsigned int nentries = 0;
310 unsigned short index = 1;
311 unsigned short orphan_blocks;
Gu Zheng502c6e02013-11-19 18:03:58 +0800312 struct orphan_inode_entry *orphan = NULL;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900313
314 orphan_blocks = (unsigned short)((sbi->n_orphans +
315 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
316
317 mutex_lock(&sbi->orphan_inode_mutex);
318 head = &sbi->orphan_inode_list;
319
320 /* loop for each orphan inode entry and write them in Jornal block */
Gu Zheng502c6e02013-11-19 18:03:58 +0800321 list_for_each_entry(orphan, head, list) {
322 if (!page) {
323 page = grab_meta_page(sbi, start_blk);
324 orphan_blk =
325 (struct f2fs_orphan_block *)page_address(page);
326 memset(orphan_blk, 0, sizeof(*orphan_blk));
327 }
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900328
Gu Zheng36795562013-11-26 16:44:16 +0800329 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900330
Gu Zheng36795562013-11-26 16:44:16 +0800331 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900332 /*
333 * an orphan block is full of 1020 entries,
334 * then we need to flush current orphan blocks
335 * and bring another one in memory
336 */
337 orphan_blk->blk_addr = cpu_to_le16(index);
338 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
339 orphan_blk->entry_count = cpu_to_le32(nentries);
340 set_page_dirty(page);
341 f2fs_put_page(page, 1);
342 index++;
343 start_blk++;
344 nentries = 0;
345 page = NULL;
346 }
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900347 }
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900348
Gu Zheng502c6e02013-11-19 18:03:58 +0800349 if (page) {
350 orphan_blk->blk_addr = cpu_to_le16(index);
351 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
352 orphan_blk->entry_count = cpu_to_le32(nentries);
353 set_page_dirty(page);
354 f2fs_put_page(page, 1);
355 }
356
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900357 mutex_unlock(&sbi->orphan_inode_mutex);
358}
359
360static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
361 block_t cp_addr, unsigned long long *version)
362{
363 struct page *cp_page_1, *cp_page_2 = NULL;
364 unsigned long blk_size = sbi->blocksize;
365 struct f2fs_checkpoint *cp_block;
366 unsigned long long cur_version = 0, pre_version = 0;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900367 size_t crc_offset;
Jaegeuk Kim7e586fa2013-06-19 20:47:19 +0900368 __u32 crc = 0;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900369
370 /* Read the 1st cp block in this CP pack */
371 cp_page_1 = get_meta_page(sbi, cp_addr);
372
373 /* get the version number */
374 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
375 crc_offset = le32_to_cpu(cp_block->checksum_offset);
376 if (crc_offset >= blk_size)
377 goto invalid_cp1;
378
Jaegeuk Kim7e586fa2013-06-19 20:47:19 +0900379 crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900380 if (!f2fs_crc_valid(crc, cp_block, crc_offset))
381 goto invalid_cp1;
382
Jaegeuk Kimd71b5562013-08-09 15:03:21 +0900383 pre_version = cur_cp_version(cp_block);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900384
385 /* Read the 2nd cp block in this CP pack */
Jaegeuk Kim25ca9232012-11-28 16:12:41 +0900386 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900387 cp_page_2 = get_meta_page(sbi, cp_addr);
388
389 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
390 crc_offset = le32_to_cpu(cp_block->checksum_offset);
391 if (crc_offset >= blk_size)
392 goto invalid_cp2;
393
Jaegeuk Kim7e586fa2013-06-19 20:47:19 +0900394 crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900395 if (!f2fs_crc_valid(crc, cp_block, crc_offset))
396 goto invalid_cp2;
397
Jaegeuk Kimd71b5562013-08-09 15:03:21 +0900398 cur_version = cur_cp_version(cp_block);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900399
400 if (cur_version == pre_version) {
401 *version = cur_version;
402 f2fs_put_page(cp_page_2, 1);
403 return cp_page_1;
404 }
405invalid_cp2:
406 f2fs_put_page(cp_page_2, 1);
407invalid_cp1:
408 f2fs_put_page(cp_page_1, 1);
409 return NULL;
410}
411
412int get_valid_checkpoint(struct f2fs_sb_info *sbi)
413{
414 struct f2fs_checkpoint *cp_block;
415 struct f2fs_super_block *fsb = sbi->raw_super;
416 struct page *cp1, *cp2, *cur_page;
417 unsigned long blk_size = sbi->blocksize;
418 unsigned long long cp1_version = 0, cp2_version = 0;
419 unsigned long long cp_start_blk_no;
420
421 sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
422 if (!sbi->ckpt)
423 return -ENOMEM;
424 /*
425 * Finding out valid cp block involves read both
426 * sets( cp pack1 and cp pack 2)
427 */
428 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
429 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
430
431 /* The second checkpoint pack should start at the next segment */
Jaegeuk Kimf9a4e6d2013-11-28 12:44:05 +0900432 cp_start_blk_no += ((unsigned long long)1) <<
433 le32_to_cpu(fsb->log_blocks_per_seg);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900434 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
435
436 if (cp1 && cp2) {
437 if (ver_after(cp2_version, cp1_version))
438 cur_page = cp2;
439 else
440 cur_page = cp1;
441 } else if (cp1) {
442 cur_page = cp1;
443 } else if (cp2) {
444 cur_page = cp2;
445 } else {
446 goto fail_no_cp;
447 }
448
449 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
450 memcpy(sbi->ckpt, cp_block, blk_size);
451
452 f2fs_put_page(cp1, 1);
453 f2fs_put_page(cp2, 1);
454 return 0;
455
456fail_no_cp:
457 kfree(sbi->ckpt);
458 return -EINVAL;
459}
460
Jaegeuk Kim5deb8262013-06-05 17:42:45 +0900461static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900462{
463 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
464 struct list_head *head = &sbi->dir_inode_list;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900465 struct list_head *this;
466
Jaegeuk Kim5deb8262013-06-05 17:42:45 +0900467 list_for_each(this, head) {
468 struct dir_inode_entry *entry;
469 entry = list_entry(this, struct dir_inode_entry, list);
470 if (entry->inode == inode)
471 return -EEXIST;
472 }
473 list_add_tail(&new->list, head);
Jaegeuk Kimdcdfff62013-10-22 20:56:10 +0900474 stat_inc_dirty_dir(sbi);
Jaegeuk Kim5deb8262013-06-05 17:42:45 +0900475 return 0;
476}
477
478void set_dirty_dir_page(struct inode *inode, struct page *page)
479{
480 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
481 struct dir_inode_entry *new;
482
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900483 if (!S_ISDIR(inode->i_mode))
484 return;
Gu Zheng7bd59382013-10-22 14:52:26 +0800485
486 new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900487 new->inode = inode;
488 INIT_LIST_HEAD(&new->list);
489
490 spin_lock(&sbi->dir_inode_lock);
Jaegeuk Kim5deb8262013-06-05 17:42:45 +0900491 if (__add_dirty_inode(inode, new))
492 kmem_cache_free(inode_entry_slab, new);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900493
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900494 inc_page_count(sbi, F2FS_DIRTY_DENTS);
495 inode_inc_dirty_dents(inode);
496 SetPagePrivate(page);
Jaegeuk Kim5deb8262013-06-05 17:42:45 +0900497 spin_unlock(&sbi->dir_inode_lock);
498}
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900499
Jaegeuk Kim5deb8262013-06-05 17:42:45 +0900500void add_dirty_dir_inode(struct inode *inode)
501{
502 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
Gu Zheng7bd59382013-10-22 14:52:26 +0800503 struct dir_inode_entry *new =
504 f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
505
Jaegeuk Kim5deb8262013-06-05 17:42:45 +0900506 new->inode = inode;
507 INIT_LIST_HEAD(&new->list);
508
509 spin_lock(&sbi->dir_inode_lock);
510 if (__add_dirty_inode(inode, new))
511 kmem_cache_free(inode_entry_slab, new);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900512 spin_unlock(&sbi->dir_inode_lock);
513}
514
515void remove_dirty_dir_inode(struct inode *inode)
516{
517 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
Gu Zhengce3b7d82013-11-19 18:03:47 +0800518
519 struct list_head *this, *head;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900520
521 if (!S_ISDIR(inode->i_mode))
522 return;
523
524 spin_lock(&sbi->dir_inode_lock);
Jaegeuk Kim3b10b1f2013-05-27 10:32:01 +0900525 if (atomic_read(&F2FS_I(inode)->dirty_dents)) {
526 spin_unlock(&sbi->dir_inode_lock);
527 return;
528 }
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900529
Gu Zhengce3b7d82013-11-19 18:03:47 +0800530 head = &sbi->dir_inode_list;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900531 list_for_each(this, head) {
532 struct dir_inode_entry *entry;
533 entry = list_entry(this, struct dir_inode_entry, list);
534 if (entry->inode == inode) {
535 list_del(&entry->list);
536 kmem_cache_free(inode_entry_slab, entry);
Jaegeuk Kimdcdfff62013-10-22 20:56:10 +0900537 stat_dec_dirty_dir(sbi);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900538 break;
539 }
540 }
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900541 spin_unlock(&sbi->dir_inode_lock);
Jaegeuk Kim74d0b912013-05-15 16:40:02 +0900542
543 /* Only from the recovery routine */
Jaegeuk Kimafc3eda2013-05-28 09:59:27 +0900544 if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
545 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
Jaegeuk Kim74d0b912013-05-15 16:40:02 +0900546 iput(inode);
Jaegeuk Kimafc3eda2013-05-28 09:59:27 +0900547 }
Jaegeuk Kim74d0b912013-05-15 16:40:02 +0900548}
549
550struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino)
551{
Gu Zhengce3b7d82013-11-19 18:03:47 +0800552
553 struct list_head *this, *head;
Jaegeuk Kim74d0b912013-05-15 16:40:02 +0900554 struct inode *inode = NULL;
555
556 spin_lock(&sbi->dir_inode_lock);
Gu Zhengce3b7d82013-11-19 18:03:47 +0800557
558 head = &sbi->dir_inode_list;
Jaegeuk Kim74d0b912013-05-15 16:40:02 +0900559 list_for_each(this, head) {
560 struct dir_inode_entry *entry;
561 entry = list_entry(this, struct dir_inode_entry, list);
562 if (entry->inode->i_ino == ino) {
563 inode = entry->inode;
564 break;
565 }
566 }
567 spin_unlock(&sbi->dir_inode_lock);
568 return inode;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900569}
570
571void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
572{
Gu Zhengce3b7d82013-11-19 18:03:47 +0800573 struct list_head *head;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900574 struct dir_inode_entry *entry;
575 struct inode *inode;
576retry:
577 spin_lock(&sbi->dir_inode_lock);
Gu Zhengce3b7d82013-11-19 18:03:47 +0800578
579 head = &sbi->dir_inode_list;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900580 if (list_empty(head)) {
581 spin_unlock(&sbi->dir_inode_lock);
582 return;
583 }
584 entry = list_entry(head->next, struct dir_inode_entry, list);
585 inode = igrab(entry->inode);
586 spin_unlock(&sbi->dir_inode_lock);
587 if (inode) {
588 filemap_flush(inode->i_mapping);
589 iput(inode);
590 } else {
591 /*
592 * We should submit bio, since it exists several
593 * wribacking dentry pages in the freeing inode.
594 */
Jaegeuk Kim93dfe2a2013-11-30 12:51:14 +0900595 f2fs_submit_merged_bio(sbi, DATA, true, WRITE);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900596 }
597 goto retry;
598}
599
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900600/*
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900601 * Freeze all the FS-operations for checkpoint.
602 */
Jaegeuk Kim43727522013-02-04 15:11:17 +0900603static void block_operations(struct f2fs_sb_info *sbi)
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900604{
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900605 struct writeback_control wbc = {
606 .sync_mode = WB_SYNC_ALL,
607 .nr_to_write = LONG_MAX,
608 .for_reclaim = 0,
609 };
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900610 struct blk_plug plug;
611
612 blk_start_plug(&plug);
613
Jaegeuk Kim39936832012-11-22 16:21:29 +0900614retry_flush_dents:
Gu Zhenge4795562013-09-27 18:08:30 +0800615 f2fs_lock_all(sbi);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900616 /* write all the dirty dentry pages */
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900617 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
Gu Zhenge4795562013-09-27 18:08:30 +0800618 f2fs_unlock_all(sbi);
Jaegeuk Kim39936832012-11-22 16:21:29 +0900619 sync_dirty_dir_inodes(sbi);
620 goto retry_flush_dents;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900621 }
622
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900623 /*
624 * POR: we should ensure that there is no dirty node pages
625 * until finishing nat/sit flush.
626 */
Jaegeuk Kim39936832012-11-22 16:21:29 +0900627retry_flush_nodes:
628 mutex_lock(&sbi->node_write);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900629
630 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
Jaegeuk Kim39936832012-11-22 16:21:29 +0900631 mutex_unlock(&sbi->node_write);
632 sync_node_pages(sbi, 0, &wbc);
633 goto retry_flush_nodes;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900634 }
Jaegeuk Kimc718379b2013-04-24 13:19:56 +0900635 blk_finish_plug(&plug);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900636}
637
638static void unblock_operations(struct f2fs_sb_info *sbi)
639{
Jaegeuk Kim39936832012-11-22 16:21:29 +0900640 mutex_unlock(&sbi->node_write);
Gu Zhenge4795562013-09-27 18:08:30 +0800641 f2fs_unlock_all(sbi);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900642}
643
Changman Leefb51b5e2013-11-07 12:48:25 +0900644static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
645{
646 DEFINE_WAIT(wait);
647
648 for (;;) {
649 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
650
651 if (!get_pages(sbi, F2FS_WRITEBACK))
652 break;
653
654 io_schedule();
655 }
656 finish_wait(&sbi->cp_wait, &wait);
657}
658
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900659static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
660{
661 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
662 nid_t last_nid = 0;
663 block_t start_blk;
664 struct page *cp_page;
665 unsigned int data_sum_blocks, orphan_blocks;
Jaegeuk Kim7e586fa2013-06-19 20:47:19 +0900666 __u32 crc32 = 0;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900667 void *kaddr;
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900668 int i;
669
670 /* Flush all the NAT/SIT pages */
671 while (get_pages(sbi, F2FS_DIRTY_META))
672 sync_meta_pages(sbi, META, LONG_MAX);
673
674 next_free_nid(sbi, &last_nid);
675
676 /*
677 * modify checkpoint
678 * version number is already updated
679 */
680 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
681 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
682 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
683 for (i = 0; i < 3; i++) {
684 ckpt->cur_node_segno[i] =
685 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
686 ckpt->cur_node_blkoff[i] =
687 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
688 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
689 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
690 }
691 for (i = 0; i < 3; i++) {
692 ckpt->cur_data_segno[i] =
693 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
694 ckpt->cur_data_blkoff[i] =
695 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
696 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
697 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
698 }
699
700 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
701 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
702 ckpt->next_free_nid = cpu_to_le32(last_nid);
703
704 /* 2 cp + n data seg summary + orphan inode blocks */
705 data_sum_blocks = npages_for_summary_flush(sbi);
706 if (data_sum_blocks < 3)
Jaegeuk Kim25ca9232012-11-28 16:12:41 +0900707 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900708 else
Jaegeuk Kim25ca9232012-11-28 16:12:41 +0900709 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900710
711 orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
712 / F2FS_ORPHANS_PER_BLOCK;
Jaegeuk Kim25ca9232012-11-28 16:12:41 +0900713 ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900714
715 if (is_umount) {
Jaegeuk Kim25ca9232012-11-28 16:12:41 +0900716 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
717 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
718 data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900719 } else {
Jaegeuk Kim25ca9232012-11-28 16:12:41 +0900720 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
721 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
722 data_sum_blocks + orphan_blocks);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900723 }
724
725 if (sbi->n_orphans)
Jaegeuk Kim25ca9232012-11-28 16:12:41 +0900726 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900727 else
Jaegeuk Kim25ca9232012-11-28 16:12:41 +0900728 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900729
730 /* update SIT/NAT bitmap */
731 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
732 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
733
734 crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
Jaegeuk Kim7e586fa2013-06-19 20:47:19 +0900735 *((__le32 *)((unsigned char *)ckpt +
736 le32_to_cpu(ckpt->checksum_offset)))
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900737 = cpu_to_le32(crc32);
738
739 start_blk = __start_cp_addr(sbi);
740
741 /* write out checkpoint buffer at block 0 */
742 cp_page = grab_meta_page(sbi, start_blk++);
743 kaddr = page_address(cp_page);
744 memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
745 set_page_dirty(cp_page);
746 f2fs_put_page(cp_page, 1);
747
748 if (sbi->n_orphans) {
749 write_orphan_inodes(sbi, start_blk);
750 start_blk += orphan_blocks;
751 }
752
753 write_data_summaries(sbi, start_blk);
754 start_blk += data_sum_blocks;
755 if (is_umount) {
756 write_node_summaries(sbi, start_blk);
757 start_blk += NR_CURSEG_NODE_TYPE;
758 }
759
760 /* writeout checkpoint block */
761 cp_page = grab_meta_page(sbi, start_blk);
762 kaddr = page_address(cp_page);
763 memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
764 set_page_dirty(cp_page);
765 f2fs_put_page(cp_page, 1);
766
767 /* wait for previous submitted node/meta pages writeback */
Changman Leefb51b5e2013-11-07 12:48:25 +0900768 wait_on_all_pages_writeback(sbi);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900769
770 filemap_fdatawait_range(sbi->node_inode->i_mapping, 0, LONG_MAX);
771 filemap_fdatawait_range(sbi->meta_inode->i_mapping, 0, LONG_MAX);
772
773 /* update user_block_counts */
774 sbi->last_valid_block_count = sbi->total_valid_block_count;
775 sbi->alloc_valid_block_count = 0;
776
777 /* Here, we only have one bio having CP pack */
Jaegeuk Kim577e3492013-01-24 19:56:11 +0900778 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900779
Jaegeuk Kim577e3492013-01-24 19:56:11 +0900780 if (!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG)) {
781 clear_prefree_segments(sbi);
782 F2FS_RESET_SB_DIRT(sbi);
783 }
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900784}
785
Jaegeuk Kim0a8165d2012-11-29 13:28:09 +0900786/*
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900787 * We guarantee that this checkpoint procedure should not fail.
788 */
Jaegeuk Kim43727522013-02-04 15:11:17 +0900789void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900790{
791 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
792 unsigned long long ckpt_ver;
793
Namjae Jeon2af4bd62013-04-23 18:26:54 +0900794 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
795
Jaegeuk Kim43727522013-02-04 15:11:17 +0900796 mutex_lock(&sbi->cp_mutex);
797 block_operations(sbi);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900798
Namjae Jeon2af4bd62013-04-23 18:26:54 +0900799 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
800
Jaegeuk Kim93dfe2a2013-11-30 12:51:14 +0900801 f2fs_submit_merged_bio(sbi, DATA, true, WRITE);
802 f2fs_submit_merged_bio(sbi, NODE, true, WRITE);
803 f2fs_submit_merged_bio(sbi, META, true, WRITE);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900804
805 /*
806 * update checkpoint pack index
807 * Increase the version number so that
808 * SIT entries and seg summaries are written at correct place
809 */
Jaegeuk Kimd71b5562013-08-09 15:03:21 +0900810 ckpt_ver = cur_cp_version(ckpt);
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900811 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
812
813 /* write cached NAT/SIT entries to NAT/SIT area */
814 flush_nat_entries(sbi);
815 flush_sit_entries(sbi);
816
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900817 /* unlock all the fs_lock[] in do_checkpoint() */
818 do_checkpoint(sbi, is_umount);
819
820 unblock_operations(sbi);
821 mutex_unlock(&sbi->cp_mutex);
Namjae Jeon2af4bd62013-04-23 18:26:54 +0900822
823 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900824}
825
826void init_orphan_info(struct f2fs_sb_info *sbi)
827{
828 mutex_init(&sbi->orphan_inode_mutex);
829 INIT_LIST_HEAD(&sbi->orphan_inode_list);
830 sbi->n_orphans = 0;
831}
832
Namjae Jeon6e6093a2013-01-17 00:08:30 +0900833int __init create_checkpoint_caches(void)
Jaegeuk Kim127e6702012-11-02 17:08:18 +0900834{
835 orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
836 sizeof(struct orphan_inode_entry), NULL);
837 if (unlikely(!orphan_entry_slab))
838 return -ENOMEM;
839 inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
840 sizeof(struct dir_inode_entry), NULL);
841 if (unlikely(!inode_entry_slab)) {
842 kmem_cache_destroy(orphan_entry_slab);
843 return -ENOMEM;
844 }
845 return 0;
846}
847
848void destroy_checkpoint_caches(void)
849{
850 kmem_cache_destroy(orphan_entry_slab);
851 kmem_cache_destroy(inode_entry_slab);
852}