Commit | Line | Data |
---|---|---|
0a8165d7 | 1 | /* |
127e670a JK |
2 | * fs/f2fs/checkpoint.c |
3 | * | |
4 | * Copyright (c) 2012 Samsung Electronics Co., Ltd. | |
5 | * http://www.samsung.com/ | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License version 2 as | |
9 | * published by the Free Software Foundation. | |
10 | */ | |
11 | #include <linux/fs.h> | |
12 | #include <linux/bio.h> | |
13 | #include <linux/mpage.h> | |
14 | #include <linux/writeback.h> | |
15 | #include <linux/blkdev.h> | |
16 | #include <linux/f2fs_fs.h> | |
17 | #include <linux/pagevec.h> | |
18 | #include <linux/swap.h> | |
19 | ||
20 | #include "f2fs.h" | |
21 | #include "node.h" | |
22 | #include "segment.h" | |
2af4bd6c | 23 | #include <trace/events/f2fs.h> |
127e670a JK |
24 | |
25 | static struct kmem_cache *orphan_entry_slab; | |
26 | static struct kmem_cache *inode_entry_slab; | |
27 | ||
0a8165d7 | 28 | /* |
127e670a JK |
29 | * We guarantee no failure on the returned page. |
30 | */ | |
31 | struct 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; | |
35 | repeat: | |
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 | ||
0a8165d7 | 48 | /* |
127e670a JK |
49 | * We guarantee no failure on the returned page. |
50 | */ | |
51 | struct 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; | |
55 | repeat: | |
56 | page = grab_cache_page(mapping, index); | |
57 | if (!page) { | |
58 | cond_resched(); | |
59 | goto repeat; | |
60 | } | |
393ff91f JK |
61 | if (PageUptodate(page)) |
62 | goto out; | |
63 | ||
93dfe2ac JK |
64 | if (f2fs_submit_page_bio(sbi, page, index, |
65 | READ_SYNC | REQ_META | REQ_PRIO)) | |
127e670a | 66 | goto repeat; |
127e670a | 67 | |
393ff91f | 68 | lock_page(page); |
afcb7ca0 JK |
69 | if (page->mapping != mapping) { |
70 | f2fs_put_page(page, 1); | |
71 | goto repeat; | |
72 | } | |
393ff91f JK |
73 | out: |
74 | mark_page_accessed(page); | |
127e670a JK |
75 | return page; |
76 | } | |
77 | ||
78 | static 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); | |
127e670a | 83 | |
577e3495 | 84 | /* Should not write any meta pages, if any IO error was occurred */ |
87a9bd26 | 85 | if (wbc->for_reclaim || sbi->por_doing || |
577e3495 JK |
86 | is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)) { |
87 | dec_page_count(sbi, F2FS_DIRTY_META); | |
127e670a JK |
88 | wbc->pages_skipped++; |
89 | set_page_dirty(page); | |
577e3495 | 90 | return AOP_WRITEPAGE_ACTIVATE; |
127e670a JK |
91 | } |
92 | ||
577e3495 | 93 | wait_on_page_writeback(page); |
127e670a | 94 | |
577e3495 JK |
95 | write_meta_page(sbi, page); |
96 | dec_page_count(sbi, F2FS_DIRTY_META); | |
97 | unlock_page(page); | |
98 | return 0; | |
127e670a JK |
99 | } |
100 | ||
101 | static 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 | ||
122 | long 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); | |
5d56b671 JK |
146 | f2fs_bug_on(page->mapping != mapping); |
147 | f2fs_bug_on(!PageDirty(page)); | |
127e670a | 148 | clear_page_dirty_for_io(page); |
577e3495 JK |
149 | if (f2fs_write_meta_page(page, &wbc)) { |
150 | unlock_page(page); | |
151 | break; | |
152 | } | |
127e670a JK |
153 | if (nwritten++ >= nr_to_write) |
154 | break; | |
155 | } | |
156 | pagevec_release(&pvec); | |
157 | cond_resched(); | |
158 | } | |
159 | ||
160 | if (nwritten) | |
93dfe2ac JK |
161 | f2fs_submit_merged_bio(sbi, type, nr_to_write == LONG_MAX, |
162 | WRITE); | |
127e670a JK |
163 | |
164 | return nwritten; | |
165 | } | |
166 | ||
167 | static 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 | ||
26c6b887 JK |
172 | trace_f2fs_set_page_dirty(page, META); |
173 | ||
127e670a JK |
174 | SetPageUptodate(page); |
175 | if (!PageDirty(page)) { | |
176 | __set_page_dirty_nobuffers(page); | |
177 | inc_page_count(sbi, F2FS_DIRTY_META); | |
127e670a JK |
178 | return 1; |
179 | } | |
180 | return 0; | |
181 | } | |
182 | ||
183 | const 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 | ||
cbd56e7d | 189 | int acquire_orphan_inode(struct f2fs_sb_info *sbi) |
127e670a JK |
190 | { |
191 | unsigned int max_orphans; | |
192 | int err = 0; | |
193 | ||
194 | /* | |
6947eea9 | 195 | * considering 512 blocks in a segment 8 blocks are needed for cp |
127e670a JK |
196 | * and log segment summaries. Remaining blocks are used to keep |
197 | * orphan entries with the limitation one reserved segment | |
6947eea9 | 198 | * for cp pack we can have max 1020*504 orphan entries |
127e670a | 199 | */ |
6947eea9 CY |
200 | max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE) |
201 | * F2FS_ORPHANS_PER_BLOCK; | |
127e670a JK |
202 | mutex_lock(&sbi->orphan_inode_mutex); |
203 | if (sbi->n_orphans >= max_orphans) | |
204 | err = -ENOSPC; | |
cbd56e7d JK |
205 | else |
206 | sbi->n_orphans++; | |
127e670a JK |
207 | mutex_unlock(&sbi->orphan_inode_mutex); |
208 | return err; | |
209 | } | |
210 | ||
cbd56e7d JK |
211 | void release_orphan_inode(struct f2fs_sb_info *sbi) |
212 | { | |
213 | mutex_lock(&sbi->orphan_inode_mutex); | |
5d56b671 | 214 | f2fs_bug_on(sbi->n_orphans == 0); |
cbd56e7d JK |
215 | sbi->n_orphans--; |
216 | mutex_unlock(&sbi->orphan_inode_mutex); | |
217 | } | |
218 | ||
127e670a JK |
219 | void 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 | } | |
7bd59381 GZ |
234 | |
235 | new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC); | |
127e670a | 236 | new->ino = ino; |
127e670a JK |
237 | |
238 | /* add new_oentry into list which is sorted by inode number */ | |
a2617dc6 | 239 | if (orphan) |
240 | list_add(&new->list, this->prev); | |
241 | else | |
127e670a | 242 | list_add_tail(&new->list, head); |
127e670a JK |
243 | out: |
244 | mutex_unlock(&sbi->orphan_inode_mutex); | |
245 | } | |
246 | ||
247 | void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
248 | { | |
60ed9a0f | 249 | struct list_head *head; |
127e670a JK |
250 | struct orphan_inode_entry *orphan; |
251 | ||
252 | mutex_lock(&sbi->orphan_inode_mutex); | |
253 | head = &sbi->orphan_inode_list; | |
60ed9a0f | 254 | list_for_each_entry(orphan, head, list) { |
127e670a JK |
255 | if (orphan->ino == ino) { |
256 | list_del(&orphan->list); | |
257 | kmem_cache_free(orphan_entry_slab, orphan); | |
5d56b671 | 258 | f2fs_bug_on(sbi->n_orphans == 0); |
127e670a JK |
259 | sbi->n_orphans--; |
260 | break; | |
261 | } | |
262 | } | |
263 | mutex_unlock(&sbi->orphan_inode_mutex); | |
264 | } | |
265 | ||
266 | static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
267 | { | |
268 | struct inode *inode = f2fs_iget(sbi->sb, ino); | |
5d56b671 | 269 | f2fs_bug_on(IS_ERR(inode)); |
127e670a JK |
270 | clear_nlink(inode); |
271 | ||
272 | /* truncate all the data during iput */ | |
273 | iput(inode); | |
274 | } | |
275 | ||
8f99a946 | 276 | void recover_orphan_inodes(struct f2fs_sb_info *sbi) |
127e670a JK |
277 | { |
278 | block_t start_blk, orphan_blkaddr, i, j; | |
279 | ||
25ca923b | 280 | if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG)) |
8f99a946 | 281 | return; |
127e670a | 282 | |
aabe5136 | 283 | sbi->por_doing = true; |
127e670a JK |
284 | 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 */ | |
25ca923b | 299 | clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG); |
aabe5136 | 300 | sbi->por_doing = false; |
8f99a946 | 301 | return; |
127e670a JK |
302 | } |
303 | ||
304 | static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk) | |
305 | { | |
502c6e0b | 306 | struct list_head *head; |
127e670a JK |
307 | 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; | |
502c6e0b | 312 | struct orphan_inode_entry *orphan = NULL; |
127e670a JK |
313 | |
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 */ | |
502c6e0b GZ |
321 | 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 | } | |
127e670a | 328 | |
36795567 | 329 | orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino); |
127e670a | 330 | |
36795567 | 331 | if (nentries == F2FS_ORPHANS_PER_BLOCK) { |
127e670a JK |
332 | /* |
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 | } | |
502c6e0b | 347 | } |
127e670a | 348 | |
502c6e0b GZ |
349 | 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); | |
127e670a | 355 | } |
502c6e0b | 356 | |
127e670a JK |
357 | mutex_unlock(&sbi->orphan_inode_mutex); |
358 | } | |
359 | ||
360 | static 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; | |
127e670a | 367 | size_t crc_offset; |
7e586fa0 | 368 | __u32 crc = 0; |
127e670a JK |
369 | |
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 | ||
7e586fa0 | 379 | crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset))); |
127e670a JK |
380 | if (!f2fs_crc_valid(crc, cp_block, crc_offset)) |
381 | goto invalid_cp1; | |
382 | ||
d71b5564 | 383 | pre_version = cur_cp_version(cp_block); |
127e670a JK |
384 | |
385 | /* Read the 2nd cp block in this CP pack */ | |
25ca923b | 386 | cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1; |
127e670a JK |
387 | 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 | ||
7e586fa0 | 394 | crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset))); |
127e670a JK |
395 | if (!f2fs_crc_valid(crc, cp_block, crc_offset)) |
396 | goto invalid_cp2; | |
397 | ||
d71b5564 | 398 | cur_version = cur_cp_version(cp_block); |
127e670a JK |
399 | |
400 | if (cur_version == pre_version) { | |
401 | *version = cur_version; | |
402 | f2fs_put_page(cp_page_2, 1); | |
403 | return cp_page_1; | |
404 | } | |
405 | invalid_cp2: | |
406 | f2fs_put_page(cp_page_2, 1); | |
407 | invalid_cp1: | |
408 | f2fs_put_page(cp_page_1, 1); | |
409 | return NULL; | |
410 | } | |
411 | ||
412 | int 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 */ | |
f9a4e6df JK |
432 | cp_start_blk_no += ((unsigned long long)1) << |
433 | le32_to_cpu(fsb->log_blocks_per_seg); | |
127e670a JK |
434 | 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 | ||
456 | fail_no_cp: | |
457 | kfree(sbi->ckpt); | |
458 | return -EINVAL; | |
459 | } | |
460 | ||
5deb8267 | 461 | static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new) |
127e670a JK |
462 | { |
463 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
464 | struct list_head *head = &sbi->dir_inode_list; | |
127e670a JK |
465 | struct list_head *this; |
466 | ||
5deb8267 JK |
467 | 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); | |
dcdfff65 | 474 | stat_inc_dirty_dir(sbi); |
5deb8267 JK |
475 | return 0; |
476 | } | |
477 | ||
478 | void 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 | ||
127e670a JK |
483 | if (!S_ISDIR(inode->i_mode)) |
484 | return; | |
7bd59381 GZ |
485 | |
486 | new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); | |
127e670a JK |
487 | new->inode = inode; |
488 | INIT_LIST_HEAD(&new->list); | |
489 | ||
490 | spin_lock(&sbi->dir_inode_lock); | |
5deb8267 JK |
491 | if (__add_dirty_inode(inode, new)) |
492 | kmem_cache_free(inode_entry_slab, new); | |
127e670a | 493 | |
127e670a JK |
494 | inc_page_count(sbi, F2FS_DIRTY_DENTS); |
495 | inode_inc_dirty_dents(inode); | |
496 | SetPagePrivate(page); | |
5deb8267 JK |
497 | spin_unlock(&sbi->dir_inode_lock); |
498 | } | |
499 | ||
500 | void add_dirty_dir_inode(struct inode *inode) | |
501 | { | |
502 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
7bd59381 GZ |
503 | struct dir_inode_entry *new = |
504 | f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); | |
505 | ||
5deb8267 JK |
506 | new->inode = inode; |
507 | INIT_LIST_HEAD(&new->list); | |
127e670a | 508 | |
5deb8267 JK |
509 | spin_lock(&sbi->dir_inode_lock); |
510 | if (__add_dirty_inode(inode, new)) | |
511 | kmem_cache_free(inode_entry_slab, new); | |
127e670a JK |
512 | spin_unlock(&sbi->dir_inode_lock); |
513 | } | |
514 | ||
515 | void remove_dirty_dir_inode(struct inode *inode) | |
516 | { | |
517 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
ce3b7d80 GZ |
518 | |
519 | struct list_head *this, *head; | |
127e670a JK |
520 | |
521 | if (!S_ISDIR(inode->i_mode)) | |
522 | return; | |
523 | ||
524 | spin_lock(&sbi->dir_inode_lock); | |
3b10b1fd JK |
525 | if (atomic_read(&F2FS_I(inode)->dirty_dents)) { |
526 | spin_unlock(&sbi->dir_inode_lock); | |
527 | return; | |
528 | } | |
127e670a | 529 | |
ce3b7d80 | 530 | head = &sbi->dir_inode_list; |
127e670a JK |
531 | 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); | |
dcdfff65 | 537 | stat_dec_dirty_dir(sbi); |
127e670a JK |
538 | break; |
539 | } | |
540 | } | |
127e670a | 541 | spin_unlock(&sbi->dir_inode_lock); |
74d0b917 JK |
542 | |
543 | /* Only from the recovery routine */ | |
afc3eda2 JK |
544 | if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) { |
545 | clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT); | |
74d0b917 | 546 | iput(inode); |
afc3eda2 | 547 | } |
74d0b917 JK |
548 | } |
549 | ||
550 | struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
551 | { | |
ce3b7d80 GZ |
552 | |
553 | struct list_head *this, *head; | |
74d0b917 JK |
554 | struct inode *inode = NULL; |
555 | ||
556 | spin_lock(&sbi->dir_inode_lock); | |
ce3b7d80 GZ |
557 | |
558 | head = &sbi->dir_inode_list; | |
74d0b917 JK |
559 | 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; | |
127e670a JK |
569 | } |
570 | ||
571 | void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi) | |
572 | { | |
ce3b7d80 | 573 | struct list_head *head; |
127e670a JK |
574 | struct dir_inode_entry *entry; |
575 | struct inode *inode; | |
576 | retry: | |
577 | spin_lock(&sbi->dir_inode_lock); | |
ce3b7d80 GZ |
578 | |
579 | head = &sbi->dir_inode_list; | |
127e670a JK |
580 | 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 | */ | |
93dfe2ac | 595 | f2fs_submit_merged_bio(sbi, DATA, true, WRITE); |
127e670a JK |
596 | } |
597 | goto retry; | |
598 | } | |
599 | ||
0a8165d7 | 600 | /* |
127e670a JK |
601 | * Freeze all the FS-operations for checkpoint. |
602 | */ | |
43727527 | 603 | static void block_operations(struct f2fs_sb_info *sbi) |
127e670a | 604 | { |
127e670a JK |
605 | struct writeback_control wbc = { |
606 | .sync_mode = WB_SYNC_ALL, | |
607 | .nr_to_write = LONG_MAX, | |
608 | .for_reclaim = 0, | |
609 | }; | |
c718379b JK |
610 | struct blk_plug plug; |
611 | ||
612 | blk_start_plug(&plug); | |
613 | ||
39936837 | 614 | retry_flush_dents: |
e479556b | 615 | f2fs_lock_all(sbi); |
127e670a | 616 | /* write all the dirty dentry pages */ |
127e670a | 617 | if (get_pages(sbi, F2FS_DIRTY_DENTS)) { |
e479556b | 618 | f2fs_unlock_all(sbi); |
39936837 JK |
619 | sync_dirty_dir_inodes(sbi); |
620 | goto retry_flush_dents; | |
127e670a JK |
621 | } |
622 | ||
127e670a JK |
623 | /* |
624 | * POR: we should ensure that there is no dirty node pages | |
625 | * until finishing nat/sit flush. | |
626 | */ | |
39936837 JK |
627 | retry_flush_nodes: |
628 | mutex_lock(&sbi->node_write); | |
127e670a JK |
629 | |
630 | if (get_pages(sbi, F2FS_DIRTY_NODES)) { | |
39936837 JK |
631 | mutex_unlock(&sbi->node_write); |
632 | sync_node_pages(sbi, 0, &wbc); | |
633 | goto retry_flush_nodes; | |
127e670a | 634 | } |
c718379b | 635 | blk_finish_plug(&plug); |
127e670a JK |
636 | } |
637 | ||
638 | static void unblock_operations(struct f2fs_sb_info *sbi) | |
639 | { | |
39936837 | 640 | mutex_unlock(&sbi->node_write); |
e479556b | 641 | f2fs_unlock_all(sbi); |
127e670a JK |
642 | } |
643 | ||
fb51b5ef CL |
644 | static 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 | ||
127e670a JK |
659 | static 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; | |
7e586fa0 | 666 | __u32 crc32 = 0; |
127e670a | 667 | void *kaddr; |
127e670a JK |
668 | 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) | |
25ca923b | 707 | set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); |
127e670a | 708 | else |
25ca923b | 709 | clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); |
127e670a JK |
710 | |
711 | orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1) | |
712 | / F2FS_ORPHANS_PER_BLOCK; | |
25ca923b | 713 | ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks); |
127e670a JK |
714 | |
715 | if (is_umount) { | |
25ca923b JK |
716 | 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); | |
127e670a | 719 | } else { |
25ca923b JK |
720 | clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG); |
721 | ckpt->cp_pack_total_block_count = cpu_to_le32(2 + | |
722 | data_sum_blocks + orphan_blocks); | |
127e670a JK |
723 | } |
724 | ||
725 | if (sbi->n_orphans) | |
25ca923b | 726 | set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); |
127e670a | 727 | else |
25ca923b | 728 | clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); |
127e670a JK |
729 | |
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)); | |
7e586fa0 JK |
735 | *((__le32 *)((unsigned char *)ckpt + |
736 | le32_to_cpu(ckpt->checksum_offset))) | |
127e670a JK |
737 | = 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 */ | |
fb51b5ef | 768 | wait_on_all_pages_writeback(sbi); |
127e670a JK |
769 | |
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 */ | |
577e3495 | 778 | sync_meta_pages(sbi, META_FLUSH, LONG_MAX); |
127e670a | 779 | |
577e3495 JK |
780 | if (!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG)) { |
781 | clear_prefree_segments(sbi); | |
782 | F2FS_RESET_SB_DIRT(sbi); | |
783 | } | |
127e670a JK |
784 | } |
785 | ||
0a8165d7 | 786 | /* |
127e670a JK |
787 | * We guarantee that this checkpoint procedure should not fail. |
788 | */ | |
43727527 | 789 | void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount) |
127e670a JK |
790 | { |
791 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
792 | unsigned long long ckpt_ver; | |
793 | ||
2af4bd6c NJ |
794 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops"); |
795 | ||
43727527 JK |
796 | mutex_lock(&sbi->cp_mutex); |
797 | block_operations(sbi); | |
127e670a | 798 | |
2af4bd6c NJ |
799 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops"); |
800 | ||
93dfe2ac JK |
801 | 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); | |
127e670a JK |
804 | |
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 | */ | |
d71b5564 | 810 | ckpt_ver = cur_cp_version(ckpt); |
127e670a JK |
811 | 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 | ||
127e670a JK |
817 | /* 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); | |
2af4bd6c NJ |
822 | |
823 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint"); | |
127e670a JK |
824 | } |
825 | ||
826 | void 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 | ||
6e6093a8 | 833 | int __init create_checkpoint_caches(void) |
127e670a JK |
834 | { |
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 | ||
848 | void destroy_checkpoint_caches(void) | |
849 | { | |
850 | kmem_cache_destroy(orphan_entry_slab); | |
851 | kmem_cache_destroy(inode_entry_slab); | |
852 | } |