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 | ||
64 | if (f2fs_readpage(sbi, page, index, READ_SYNC)) | |
127e670a | 65 | goto repeat; |
127e670a | 66 | |
393ff91f | 67 | lock_page(page); |
afcb7ca0 JK |
68 | if (page->mapping != mapping) { |
69 | f2fs_put_page(page, 1); | |
70 | goto repeat; | |
71 | } | |
393ff91f JK |
72 | out: |
73 | mark_page_accessed(page); | |
127e670a JK |
74 | return page; |
75 | } | |
76 | ||
77 | static int f2fs_write_meta_page(struct page *page, | |
78 | struct writeback_control *wbc) | |
79 | { | |
80 | struct inode *inode = page->mapping->host; | |
81 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
127e670a | 82 | |
577e3495 | 83 | /* Should not write any meta pages, if any IO error was occurred */ |
87a9bd26 | 84 | if (wbc->for_reclaim || sbi->por_doing || |
577e3495 JK |
85 | is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)) { |
86 | dec_page_count(sbi, F2FS_DIRTY_META); | |
127e670a JK |
87 | wbc->pages_skipped++; |
88 | set_page_dirty(page); | |
577e3495 | 89 | return AOP_WRITEPAGE_ACTIVATE; |
127e670a JK |
90 | } |
91 | ||
577e3495 | 92 | wait_on_page_writeback(page); |
127e670a | 93 | |
577e3495 JK |
94 | write_meta_page(sbi, page); |
95 | dec_page_count(sbi, F2FS_DIRTY_META); | |
96 | unlock_page(page); | |
97 | return 0; | |
127e670a JK |
98 | } |
99 | ||
100 | static int f2fs_write_meta_pages(struct address_space *mapping, | |
101 | struct writeback_control *wbc) | |
102 | { | |
103 | struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); | |
104 | struct block_device *bdev = sbi->sb->s_bdev; | |
105 | long written; | |
106 | ||
107 | if (wbc->for_kupdate) | |
108 | return 0; | |
109 | ||
110 | if (get_pages(sbi, F2FS_DIRTY_META) == 0) | |
111 | return 0; | |
112 | ||
113 | /* if mounting is failed, skip writing node pages */ | |
114 | mutex_lock(&sbi->cp_mutex); | |
115 | written = sync_meta_pages(sbi, META, bio_get_nr_vecs(bdev)); | |
116 | mutex_unlock(&sbi->cp_mutex); | |
117 | wbc->nr_to_write -= written; | |
118 | return 0; | |
119 | } | |
120 | ||
121 | long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type, | |
122 | long nr_to_write) | |
123 | { | |
124 | struct address_space *mapping = sbi->meta_inode->i_mapping; | |
125 | pgoff_t index = 0, end = LONG_MAX; | |
126 | struct pagevec pvec; | |
127 | long nwritten = 0; | |
128 | struct writeback_control wbc = { | |
129 | .for_reclaim = 0, | |
130 | }; | |
131 | ||
132 | pagevec_init(&pvec, 0); | |
133 | ||
134 | while (index <= end) { | |
135 | int i, nr_pages; | |
136 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
137 | PAGECACHE_TAG_DIRTY, | |
138 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); | |
139 | if (nr_pages == 0) | |
140 | break; | |
141 | ||
142 | for (i = 0; i < nr_pages; i++) { | |
143 | struct page *page = pvec.pages[i]; | |
144 | lock_page(page); | |
5d56b671 JK |
145 | f2fs_bug_on(page->mapping != mapping); |
146 | f2fs_bug_on(!PageDirty(page)); | |
127e670a | 147 | clear_page_dirty_for_io(page); |
577e3495 JK |
148 | if (f2fs_write_meta_page(page, &wbc)) { |
149 | unlock_page(page); | |
150 | break; | |
151 | } | |
127e670a JK |
152 | if (nwritten++ >= nr_to_write) |
153 | break; | |
154 | } | |
155 | pagevec_release(&pvec); | |
156 | cond_resched(); | |
157 | } | |
158 | ||
159 | if (nwritten) | |
160 | f2fs_submit_bio(sbi, type, nr_to_write == LONG_MAX); | |
161 | ||
162 | return nwritten; | |
163 | } | |
164 | ||
165 | static int f2fs_set_meta_page_dirty(struct page *page) | |
166 | { | |
167 | struct address_space *mapping = page->mapping; | |
168 | struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); | |
169 | ||
26c6b887 JK |
170 | trace_f2fs_set_page_dirty(page, META); |
171 | ||
127e670a JK |
172 | SetPageUptodate(page); |
173 | if (!PageDirty(page)) { | |
174 | __set_page_dirty_nobuffers(page); | |
175 | inc_page_count(sbi, F2FS_DIRTY_META); | |
127e670a JK |
176 | return 1; |
177 | } | |
178 | return 0; | |
179 | } | |
180 | ||
181 | const struct address_space_operations f2fs_meta_aops = { | |
182 | .writepage = f2fs_write_meta_page, | |
183 | .writepages = f2fs_write_meta_pages, | |
184 | .set_page_dirty = f2fs_set_meta_page_dirty, | |
185 | }; | |
186 | ||
cbd56e7d | 187 | int acquire_orphan_inode(struct f2fs_sb_info *sbi) |
127e670a JK |
188 | { |
189 | unsigned int max_orphans; | |
190 | int err = 0; | |
191 | ||
192 | /* | |
193 | * considering 512 blocks in a segment 5 blocks are needed for cp | |
194 | * and log segment summaries. Remaining blocks are used to keep | |
195 | * orphan entries with the limitation one reserved segment | |
196 | * for cp pack we can have max 1020*507 orphan entries | |
197 | */ | |
198 | max_orphans = (sbi->blocks_per_seg - 5) * F2FS_ORPHANS_PER_BLOCK; | |
199 | mutex_lock(&sbi->orphan_inode_mutex); | |
200 | if (sbi->n_orphans >= max_orphans) | |
201 | err = -ENOSPC; | |
cbd56e7d JK |
202 | else |
203 | sbi->n_orphans++; | |
127e670a JK |
204 | mutex_unlock(&sbi->orphan_inode_mutex); |
205 | return err; | |
206 | } | |
207 | ||
cbd56e7d JK |
208 | void release_orphan_inode(struct f2fs_sb_info *sbi) |
209 | { | |
210 | mutex_lock(&sbi->orphan_inode_mutex); | |
5d56b671 | 211 | f2fs_bug_on(sbi->n_orphans == 0); |
cbd56e7d JK |
212 | sbi->n_orphans--; |
213 | mutex_unlock(&sbi->orphan_inode_mutex); | |
214 | } | |
215 | ||
127e670a JK |
216 | void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) |
217 | { | |
218 | struct list_head *head, *this; | |
219 | struct orphan_inode_entry *new = NULL, *orphan = NULL; | |
220 | ||
221 | mutex_lock(&sbi->orphan_inode_mutex); | |
222 | head = &sbi->orphan_inode_list; | |
223 | list_for_each(this, head) { | |
224 | orphan = list_entry(this, struct orphan_inode_entry, list); | |
225 | if (orphan->ino == ino) | |
226 | goto out; | |
227 | if (orphan->ino > ino) | |
228 | break; | |
229 | orphan = NULL; | |
230 | } | |
7bd59381 GZ |
231 | |
232 | new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC); | |
127e670a | 233 | new->ino = ino; |
127e670a JK |
234 | |
235 | /* add new_oentry into list which is sorted by inode number */ | |
a2617dc6 | 236 | if (orphan) |
237 | list_add(&new->list, this->prev); | |
238 | else | |
127e670a | 239 | list_add_tail(&new->list, head); |
127e670a JK |
240 | out: |
241 | mutex_unlock(&sbi->orphan_inode_mutex); | |
242 | } | |
243 | ||
244 | void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
245 | { | |
60ed9a0f | 246 | struct list_head *head; |
127e670a JK |
247 | struct orphan_inode_entry *orphan; |
248 | ||
249 | mutex_lock(&sbi->orphan_inode_mutex); | |
250 | head = &sbi->orphan_inode_list; | |
60ed9a0f | 251 | list_for_each_entry(orphan, head, list) { |
127e670a JK |
252 | if (orphan->ino == ino) { |
253 | list_del(&orphan->list); | |
254 | kmem_cache_free(orphan_entry_slab, orphan); | |
5d56b671 | 255 | f2fs_bug_on(sbi->n_orphans == 0); |
127e670a JK |
256 | sbi->n_orphans--; |
257 | break; | |
258 | } | |
259 | } | |
260 | mutex_unlock(&sbi->orphan_inode_mutex); | |
261 | } | |
262 | ||
263 | static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
264 | { | |
265 | struct inode *inode = f2fs_iget(sbi->sb, ino); | |
5d56b671 | 266 | f2fs_bug_on(IS_ERR(inode)); |
127e670a JK |
267 | clear_nlink(inode); |
268 | ||
269 | /* truncate all the data during iput */ | |
270 | iput(inode); | |
271 | } | |
272 | ||
273 | int recover_orphan_inodes(struct f2fs_sb_info *sbi) | |
274 | { | |
275 | block_t start_blk, orphan_blkaddr, i, j; | |
276 | ||
25ca923b | 277 | if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG)) |
127e670a JK |
278 | return 0; |
279 | ||
aabe5136 | 280 | sbi->por_doing = true; |
127e670a JK |
281 | start_blk = __start_cp_addr(sbi) + 1; |
282 | orphan_blkaddr = __start_sum_addr(sbi) - 1; | |
283 | ||
284 | for (i = 0; i < orphan_blkaddr; i++) { | |
285 | struct page *page = get_meta_page(sbi, start_blk + i); | |
286 | struct f2fs_orphan_block *orphan_blk; | |
287 | ||
288 | orphan_blk = (struct f2fs_orphan_block *)page_address(page); | |
289 | for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) { | |
290 | nid_t ino = le32_to_cpu(orphan_blk->ino[j]); | |
291 | recover_orphan_inode(sbi, ino); | |
292 | } | |
293 | f2fs_put_page(page, 1); | |
294 | } | |
295 | /* clear Orphan Flag */ | |
25ca923b | 296 | clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG); |
aabe5136 | 297 | sbi->por_doing = false; |
127e670a JK |
298 | return 0; |
299 | } | |
300 | ||
301 | static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk) | |
302 | { | |
502c6e0b | 303 | struct list_head *head; |
127e670a JK |
304 | struct f2fs_orphan_block *orphan_blk = NULL; |
305 | struct page *page = NULL; | |
306 | unsigned int nentries = 0; | |
307 | unsigned short index = 1; | |
308 | unsigned short orphan_blocks; | |
502c6e0b | 309 | struct orphan_inode_entry *orphan = NULL; |
127e670a JK |
310 | |
311 | orphan_blocks = (unsigned short)((sbi->n_orphans + | |
312 | (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK); | |
313 | ||
314 | mutex_lock(&sbi->orphan_inode_mutex); | |
315 | head = &sbi->orphan_inode_list; | |
316 | ||
317 | /* loop for each orphan inode entry and write them in Jornal block */ | |
502c6e0b GZ |
318 | list_for_each_entry(orphan, head, list) { |
319 | if (!page) { | |
320 | page = grab_meta_page(sbi, start_blk); | |
321 | orphan_blk = | |
322 | (struct f2fs_orphan_block *)page_address(page); | |
323 | memset(orphan_blk, 0, sizeof(*orphan_blk)); | |
324 | } | |
127e670a | 325 | |
502c6e0b | 326 | orphan_blk->ino[nentries] = cpu_to_le32(orphan->ino); |
127e670a | 327 | |
502c6e0b | 328 | if (nentries++ == F2FS_ORPHANS_PER_BLOCK) { |
127e670a JK |
329 | /* |
330 | * an orphan block is full of 1020 entries, | |
331 | * then we need to flush current orphan blocks | |
332 | * and bring another one in memory | |
333 | */ | |
334 | orphan_blk->blk_addr = cpu_to_le16(index); | |
335 | orphan_blk->blk_count = cpu_to_le16(orphan_blocks); | |
336 | orphan_blk->entry_count = cpu_to_le32(nentries); | |
337 | set_page_dirty(page); | |
338 | f2fs_put_page(page, 1); | |
339 | index++; | |
340 | start_blk++; | |
341 | nentries = 0; | |
342 | page = NULL; | |
343 | } | |
502c6e0b | 344 | } |
127e670a | 345 | |
502c6e0b GZ |
346 | if (page) { |
347 | orphan_blk->blk_addr = cpu_to_le16(index); | |
348 | orphan_blk->blk_count = cpu_to_le16(orphan_blocks); | |
349 | orphan_blk->entry_count = cpu_to_le32(nentries); | |
350 | set_page_dirty(page); | |
351 | f2fs_put_page(page, 1); | |
127e670a | 352 | } |
502c6e0b | 353 | |
127e670a JK |
354 | mutex_unlock(&sbi->orphan_inode_mutex); |
355 | } | |
356 | ||
357 | static struct page *validate_checkpoint(struct f2fs_sb_info *sbi, | |
358 | block_t cp_addr, unsigned long long *version) | |
359 | { | |
360 | struct page *cp_page_1, *cp_page_2 = NULL; | |
361 | unsigned long blk_size = sbi->blocksize; | |
362 | struct f2fs_checkpoint *cp_block; | |
363 | unsigned long long cur_version = 0, pre_version = 0; | |
127e670a | 364 | size_t crc_offset; |
7e586fa0 | 365 | __u32 crc = 0; |
127e670a JK |
366 | |
367 | /* Read the 1st cp block in this CP pack */ | |
368 | cp_page_1 = get_meta_page(sbi, cp_addr); | |
369 | ||
370 | /* get the version number */ | |
371 | cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1); | |
372 | crc_offset = le32_to_cpu(cp_block->checksum_offset); | |
373 | if (crc_offset >= blk_size) | |
374 | goto invalid_cp1; | |
375 | ||
7e586fa0 | 376 | crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset))); |
127e670a JK |
377 | if (!f2fs_crc_valid(crc, cp_block, crc_offset)) |
378 | goto invalid_cp1; | |
379 | ||
d71b5564 | 380 | pre_version = cur_cp_version(cp_block); |
127e670a JK |
381 | |
382 | /* Read the 2nd cp block in this CP pack */ | |
25ca923b | 383 | cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1; |
127e670a JK |
384 | cp_page_2 = get_meta_page(sbi, cp_addr); |
385 | ||
386 | cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2); | |
387 | crc_offset = le32_to_cpu(cp_block->checksum_offset); | |
388 | if (crc_offset >= blk_size) | |
389 | goto invalid_cp2; | |
390 | ||
7e586fa0 | 391 | crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset))); |
127e670a JK |
392 | if (!f2fs_crc_valid(crc, cp_block, crc_offset)) |
393 | goto invalid_cp2; | |
394 | ||
d71b5564 | 395 | cur_version = cur_cp_version(cp_block); |
127e670a JK |
396 | |
397 | if (cur_version == pre_version) { | |
398 | *version = cur_version; | |
399 | f2fs_put_page(cp_page_2, 1); | |
400 | return cp_page_1; | |
401 | } | |
402 | invalid_cp2: | |
403 | f2fs_put_page(cp_page_2, 1); | |
404 | invalid_cp1: | |
405 | f2fs_put_page(cp_page_1, 1); | |
406 | return NULL; | |
407 | } | |
408 | ||
409 | int get_valid_checkpoint(struct f2fs_sb_info *sbi) | |
410 | { | |
411 | struct f2fs_checkpoint *cp_block; | |
412 | struct f2fs_super_block *fsb = sbi->raw_super; | |
413 | struct page *cp1, *cp2, *cur_page; | |
414 | unsigned long blk_size = sbi->blocksize; | |
415 | unsigned long long cp1_version = 0, cp2_version = 0; | |
416 | unsigned long long cp_start_blk_no; | |
417 | ||
418 | sbi->ckpt = kzalloc(blk_size, GFP_KERNEL); | |
419 | if (!sbi->ckpt) | |
420 | return -ENOMEM; | |
421 | /* | |
422 | * Finding out valid cp block involves read both | |
423 | * sets( cp pack1 and cp pack 2) | |
424 | */ | |
425 | cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr); | |
426 | cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version); | |
427 | ||
428 | /* The second checkpoint pack should start at the next segment */ | |
429 | cp_start_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg); | |
430 | cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version); | |
431 | ||
432 | if (cp1 && cp2) { | |
433 | if (ver_after(cp2_version, cp1_version)) | |
434 | cur_page = cp2; | |
435 | else | |
436 | cur_page = cp1; | |
437 | } else if (cp1) { | |
438 | cur_page = cp1; | |
439 | } else if (cp2) { | |
440 | cur_page = cp2; | |
441 | } else { | |
442 | goto fail_no_cp; | |
443 | } | |
444 | ||
445 | cp_block = (struct f2fs_checkpoint *)page_address(cur_page); | |
446 | memcpy(sbi->ckpt, cp_block, blk_size); | |
447 | ||
448 | f2fs_put_page(cp1, 1); | |
449 | f2fs_put_page(cp2, 1); | |
450 | return 0; | |
451 | ||
452 | fail_no_cp: | |
453 | kfree(sbi->ckpt); | |
454 | return -EINVAL; | |
455 | } | |
456 | ||
5deb8267 | 457 | static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new) |
127e670a JK |
458 | { |
459 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
460 | struct list_head *head = &sbi->dir_inode_list; | |
127e670a JK |
461 | struct list_head *this; |
462 | ||
5deb8267 JK |
463 | list_for_each(this, head) { |
464 | struct dir_inode_entry *entry; | |
465 | entry = list_entry(this, struct dir_inode_entry, list); | |
466 | if (entry->inode == inode) | |
467 | return -EEXIST; | |
468 | } | |
469 | list_add_tail(&new->list, head); | |
dcdfff65 | 470 | stat_inc_dirty_dir(sbi); |
5deb8267 JK |
471 | return 0; |
472 | } | |
473 | ||
474 | void set_dirty_dir_page(struct inode *inode, struct page *page) | |
475 | { | |
476 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
477 | struct dir_inode_entry *new; | |
478 | ||
127e670a JK |
479 | if (!S_ISDIR(inode->i_mode)) |
480 | return; | |
7bd59381 GZ |
481 | |
482 | new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); | |
127e670a JK |
483 | new->inode = inode; |
484 | INIT_LIST_HEAD(&new->list); | |
485 | ||
486 | spin_lock(&sbi->dir_inode_lock); | |
5deb8267 JK |
487 | if (__add_dirty_inode(inode, new)) |
488 | kmem_cache_free(inode_entry_slab, new); | |
127e670a | 489 | |
127e670a JK |
490 | inc_page_count(sbi, F2FS_DIRTY_DENTS); |
491 | inode_inc_dirty_dents(inode); | |
492 | SetPagePrivate(page); | |
5deb8267 JK |
493 | spin_unlock(&sbi->dir_inode_lock); |
494 | } | |
495 | ||
496 | void add_dirty_dir_inode(struct inode *inode) | |
497 | { | |
498 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
7bd59381 GZ |
499 | struct dir_inode_entry *new = |
500 | f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); | |
501 | ||
5deb8267 JK |
502 | new->inode = inode; |
503 | INIT_LIST_HEAD(&new->list); | |
127e670a | 504 | |
5deb8267 JK |
505 | spin_lock(&sbi->dir_inode_lock); |
506 | if (__add_dirty_inode(inode, new)) | |
507 | kmem_cache_free(inode_entry_slab, new); | |
127e670a JK |
508 | spin_unlock(&sbi->dir_inode_lock); |
509 | } | |
510 | ||
511 | void remove_dirty_dir_inode(struct inode *inode) | |
512 | { | |
513 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
514 | struct list_head *head = &sbi->dir_inode_list; | |
515 | struct list_head *this; | |
516 | ||
517 | if (!S_ISDIR(inode->i_mode)) | |
518 | return; | |
519 | ||
520 | spin_lock(&sbi->dir_inode_lock); | |
3b10b1fd JK |
521 | if (atomic_read(&F2FS_I(inode)->dirty_dents)) { |
522 | spin_unlock(&sbi->dir_inode_lock); | |
523 | return; | |
524 | } | |
127e670a JK |
525 | |
526 | list_for_each(this, head) { | |
527 | struct dir_inode_entry *entry; | |
528 | entry = list_entry(this, struct dir_inode_entry, list); | |
529 | if (entry->inode == inode) { | |
530 | list_del(&entry->list); | |
531 | kmem_cache_free(inode_entry_slab, entry); | |
dcdfff65 | 532 | stat_dec_dirty_dir(sbi); |
127e670a JK |
533 | break; |
534 | } | |
535 | } | |
127e670a | 536 | spin_unlock(&sbi->dir_inode_lock); |
74d0b917 JK |
537 | |
538 | /* Only from the recovery routine */ | |
afc3eda2 JK |
539 | if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) { |
540 | clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT); | |
74d0b917 | 541 | iput(inode); |
afc3eda2 | 542 | } |
74d0b917 JK |
543 | } |
544 | ||
545 | struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
546 | { | |
547 | struct list_head *head = &sbi->dir_inode_list; | |
548 | struct list_head *this; | |
549 | struct inode *inode = NULL; | |
550 | ||
551 | spin_lock(&sbi->dir_inode_lock); | |
552 | list_for_each(this, head) { | |
553 | struct dir_inode_entry *entry; | |
554 | entry = list_entry(this, struct dir_inode_entry, list); | |
555 | if (entry->inode->i_ino == ino) { | |
556 | inode = entry->inode; | |
557 | break; | |
558 | } | |
559 | } | |
560 | spin_unlock(&sbi->dir_inode_lock); | |
561 | return inode; | |
127e670a JK |
562 | } |
563 | ||
564 | void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi) | |
565 | { | |
566 | struct list_head *head = &sbi->dir_inode_list; | |
567 | struct dir_inode_entry *entry; | |
568 | struct inode *inode; | |
569 | retry: | |
570 | spin_lock(&sbi->dir_inode_lock); | |
571 | if (list_empty(head)) { | |
572 | spin_unlock(&sbi->dir_inode_lock); | |
573 | return; | |
574 | } | |
575 | entry = list_entry(head->next, struct dir_inode_entry, list); | |
576 | inode = igrab(entry->inode); | |
577 | spin_unlock(&sbi->dir_inode_lock); | |
578 | if (inode) { | |
579 | filemap_flush(inode->i_mapping); | |
580 | iput(inode); | |
581 | } else { | |
582 | /* | |
583 | * We should submit bio, since it exists several | |
584 | * wribacking dentry pages in the freeing inode. | |
585 | */ | |
586 | f2fs_submit_bio(sbi, DATA, true); | |
587 | } | |
588 | goto retry; | |
589 | } | |
590 | ||
0a8165d7 | 591 | /* |
127e670a JK |
592 | * Freeze all the FS-operations for checkpoint. |
593 | */ | |
43727527 | 594 | static void block_operations(struct f2fs_sb_info *sbi) |
127e670a | 595 | { |
127e670a JK |
596 | struct writeback_control wbc = { |
597 | .sync_mode = WB_SYNC_ALL, | |
598 | .nr_to_write = LONG_MAX, | |
599 | .for_reclaim = 0, | |
600 | }; | |
c718379b JK |
601 | struct blk_plug plug; |
602 | ||
603 | blk_start_plug(&plug); | |
604 | ||
39936837 | 605 | retry_flush_dents: |
e479556b | 606 | f2fs_lock_all(sbi); |
127e670a | 607 | /* write all the dirty dentry pages */ |
127e670a | 608 | if (get_pages(sbi, F2FS_DIRTY_DENTS)) { |
e479556b | 609 | f2fs_unlock_all(sbi); |
39936837 JK |
610 | sync_dirty_dir_inodes(sbi); |
611 | goto retry_flush_dents; | |
127e670a JK |
612 | } |
613 | ||
127e670a JK |
614 | /* |
615 | * POR: we should ensure that there is no dirty node pages | |
616 | * until finishing nat/sit flush. | |
617 | */ | |
39936837 JK |
618 | retry_flush_nodes: |
619 | mutex_lock(&sbi->node_write); | |
127e670a JK |
620 | |
621 | if (get_pages(sbi, F2FS_DIRTY_NODES)) { | |
39936837 JK |
622 | mutex_unlock(&sbi->node_write); |
623 | sync_node_pages(sbi, 0, &wbc); | |
624 | goto retry_flush_nodes; | |
127e670a | 625 | } |
c718379b | 626 | blk_finish_plug(&plug); |
127e670a JK |
627 | } |
628 | ||
629 | static void unblock_operations(struct f2fs_sb_info *sbi) | |
630 | { | |
39936837 | 631 | mutex_unlock(&sbi->node_write); |
e479556b | 632 | f2fs_unlock_all(sbi); |
127e670a JK |
633 | } |
634 | ||
fb51b5ef CL |
635 | static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi) |
636 | { | |
637 | DEFINE_WAIT(wait); | |
638 | ||
639 | for (;;) { | |
640 | prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE); | |
641 | ||
642 | if (!get_pages(sbi, F2FS_WRITEBACK)) | |
643 | break; | |
644 | ||
645 | io_schedule(); | |
646 | } | |
647 | finish_wait(&sbi->cp_wait, &wait); | |
648 | } | |
649 | ||
127e670a JK |
650 | static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount) |
651 | { | |
652 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
653 | nid_t last_nid = 0; | |
654 | block_t start_blk; | |
655 | struct page *cp_page; | |
656 | unsigned int data_sum_blocks, orphan_blocks; | |
7e586fa0 | 657 | __u32 crc32 = 0; |
127e670a | 658 | void *kaddr; |
127e670a JK |
659 | int i; |
660 | ||
661 | /* Flush all the NAT/SIT pages */ | |
662 | while (get_pages(sbi, F2FS_DIRTY_META)) | |
663 | sync_meta_pages(sbi, META, LONG_MAX); | |
664 | ||
665 | next_free_nid(sbi, &last_nid); | |
666 | ||
667 | /* | |
668 | * modify checkpoint | |
669 | * version number is already updated | |
670 | */ | |
671 | ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi)); | |
672 | ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi)); | |
673 | ckpt->free_segment_count = cpu_to_le32(free_segments(sbi)); | |
674 | for (i = 0; i < 3; i++) { | |
675 | ckpt->cur_node_segno[i] = | |
676 | cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE)); | |
677 | ckpt->cur_node_blkoff[i] = | |
678 | cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE)); | |
679 | ckpt->alloc_type[i + CURSEG_HOT_NODE] = | |
680 | curseg_alloc_type(sbi, i + CURSEG_HOT_NODE); | |
681 | } | |
682 | for (i = 0; i < 3; i++) { | |
683 | ckpt->cur_data_segno[i] = | |
684 | cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA)); | |
685 | ckpt->cur_data_blkoff[i] = | |
686 | cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA)); | |
687 | ckpt->alloc_type[i + CURSEG_HOT_DATA] = | |
688 | curseg_alloc_type(sbi, i + CURSEG_HOT_DATA); | |
689 | } | |
690 | ||
691 | ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi)); | |
692 | ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi)); | |
693 | ckpt->next_free_nid = cpu_to_le32(last_nid); | |
694 | ||
695 | /* 2 cp + n data seg summary + orphan inode blocks */ | |
696 | data_sum_blocks = npages_for_summary_flush(sbi); | |
697 | if (data_sum_blocks < 3) | |
25ca923b | 698 | set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); |
127e670a | 699 | else |
25ca923b | 700 | clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); |
127e670a JK |
701 | |
702 | orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1) | |
703 | / F2FS_ORPHANS_PER_BLOCK; | |
25ca923b | 704 | ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks); |
127e670a JK |
705 | |
706 | if (is_umount) { | |
25ca923b JK |
707 | set_ckpt_flags(ckpt, CP_UMOUNT_FLAG); |
708 | ckpt->cp_pack_total_block_count = cpu_to_le32(2 + | |
709 | data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE); | |
127e670a | 710 | } else { |
25ca923b JK |
711 | clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG); |
712 | ckpt->cp_pack_total_block_count = cpu_to_le32(2 + | |
713 | data_sum_blocks + orphan_blocks); | |
127e670a JK |
714 | } |
715 | ||
716 | if (sbi->n_orphans) | |
25ca923b | 717 | set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); |
127e670a | 718 | else |
25ca923b | 719 | clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); |
127e670a JK |
720 | |
721 | /* update SIT/NAT bitmap */ | |
722 | get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP)); | |
723 | get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP)); | |
724 | ||
725 | crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset)); | |
7e586fa0 JK |
726 | *((__le32 *)((unsigned char *)ckpt + |
727 | le32_to_cpu(ckpt->checksum_offset))) | |
127e670a JK |
728 | = cpu_to_le32(crc32); |
729 | ||
730 | start_blk = __start_cp_addr(sbi); | |
731 | ||
732 | /* write out checkpoint buffer at block 0 */ | |
733 | cp_page = grab_meta_page(sbi, start_blk++); | |
734 | kaddr = page_address(cp_page); | |
735 | memcpy(kaddr, ckpt, (1 << sbi->log_blocksize)); | |
736 | set_page_dirty(cp_page); | |
737 | f2fs_put_page(cp_page, 1); | |
738 | ||
739 | if (sbi->n_orphans) { | |
740 | write_orphan_inodes(sbi, start_blk); | |
741 | start_blk += orphan_blocks; | |
742 | } | |
743 | ||
744 | write_data_summaries(sbi, start_blk); | |
745 | start_blk += data_sum_blocks; | |
746 | if (is_umount) { | |
747 | write_node_summaries(sbi, start_blk); | |
748 | start_blk += NR_CURSEG_NODE_TYPE; | |
749 | } | |
750 | ||
751 | /* writeout checkpoint block */ | |
752 | cp_page = grab_meta_page(sbi, start_blk); | |
753 | kaddr = page_address(cp_page); | |
754 | memcpy(kaddr, ckpt, (1 << sbi->log_blocksize)); | |
755 | set_page_dirty(cp_page); | |
756 | f2fs_put_page(cp_page, 1); | |
757 | ||
758 | /* wait for previous submitted node/meta pages writeback */ | |
fb51b5ef | 759 | wait_on_all_pages_writeback(sbi); |
127e670a JK |
760 | |
761 | filemap_fdatawait_range(sbi->node_inode->i_mapping, 0, LONG_MAX); | |
762 | filemap_fdatawait_range(sbi->meta_inode->i_mapping, 0, LONG_MAX); | |
763 | ||
764 | /* update user_block_counts */ | |
765 | sbi->last_valid_block_count = sbi->total_valid_block_count; | |
766 | sbi->alloc_valid_block_count = 0; | |
767 | ||
768 | /* Here, we only have one bio having CP pack */ | |
577e3495 | 769 | sync_meta_pages(sbi, META_FLUSH, LONG_MAX); |
127e670a | 770 | |
577e3495 JK |
771 | if (!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG)) { |
772 | clear_prefree_segments(sbi); | |
773 | F2FS_RESET_SB_DIRT(sbi); | |
774 | } | |
127e670a JK |
775 | } |
776 | ||
0a8165d7 | 777 | /* |
127e670a JK |
778 | * We guarantee that this checkpoint procedure should not fail. |
779 | */ | |
43727527 | 780 | void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount) |
127e670a JK |
781 | { |
782 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
783 | unsigned long long ckpt_ver; | |
784 | ||
2af4bd6c NJ |
785 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops"); |
786 | ||
43727527 JK |
787 | mutex_lock(&sbi->cp_mutex); |
788 | block_operations(sbi); | |
127e670a | 789 | |
2af4bd6c NJ |
790 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops"); |
791 | ||
127e670a JK |
792 | f2fs_submit_bio(sbi, DATA, true); |
793 | f2fs_submit_bio(sbi, NODE, true); | |
794 | f2fs_submit_bio(sbi, META, true); | |
795 | ||
796 | /* | |
797 | * update checkpoint pack index | |
798 | * Increase the version number so that | |
799 | * SIT entries and seg summaries are written at correct place | |
800 | */ | |
d71b5564 | 801 | ckpt_ver = cur_cp_version(ckpt); |
127e670a JK |
802 | ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver); |
803 | ||
804 | /* write cached NAT/SIT entries to NAT/SIT area */ | |
805 | flush_nat_entries(sbi); | |
806 | flush_sit_entries(sbi); | |
807 | ||
127e670a JK |
808 | /* unlock all the fs_lock[] in do_checkpoint() */ |
809 | do_checkpoint(sbi, is_umount); | |
810 | ||
811 | unblock_operations(sbi); | |
812 | mutex_unlock(&sbi->cp_mutex); | |
2af4bd6c NJ |
813 | |
814 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint"); | |
127e670a JK |
815 | } |
816 | ||
817 | void init_orphan_info(struct f2fs_sb_info *sbi) | |
818 | { | |
819 | mutex_init(&sbi->orphan_inode_mutex); | |
820 | INIT_LIST_HEAD(&sbi->orphan_inode_list); | |
821 | sbi->n_orphans = 0; | |
822 | } | |
823 | ||
6e6093a8 | 824 | int __init create_checkpoint_caches(void) |
127e670a JK |
825 | { |
826 | orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry", | |
827 | sizeof(struct orphan_inode_entry), NULL); | |
828 | if (unlikely(!orphan_entry_slab)) | |
829 | return -ENOMEM; | |
830 | inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry", | |
831 | sizeof(struct dir_inode_entry), NULL); | |
832 | if (unlikely(!inode_entry_slab)) { | |
833 | kmem_cache_destroy(orphan_entry_slab); | |
834 | return -ENOMEM; | |
835 | } | |
836 | return 0; | |
837 | } | |
838 | ||
839 | void destroy_checkpoint_caches(void) | |
840 | { | |
841 | kmem_cache_destroy(orphan_entry_slab); | |
842 | kmem_cache_destroy(inode_entry_slab); | |
843 | } |