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 JK |
83 | /* Should not write any meta pages, if any IO error was occurred */ |
84 | if (wbc->for_reclaim || | |
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); | |
145 | BUG_ON(page->mapping != mapping); | |
146 | BUG_ON(!PageDirty(page)); | |
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 | ||
170 | SetPageUptodate(page); | |
171 | if (!PageDirty(page)) { | |
172 | __set_page_dirty_nobuffers(page); | |
173 | inc_page_count(sbi, F2FS_DIRTY_META); | |
127e670a JK |
174 | return 1; |
175 | } | |
176 | return 0; | |
177 | } | |
178 | ||
179 | const struct address_space_operations f2fs_meta_aops = { | |
180 | .writepage = f2fs_write_meta_page, | |
181 | .writepages = f2fs_write_meta_pages, | |
182 | .set_page_dirty = f2fs_set_meta_page_dirty, | |
183 | }; | |
184 | ||
185 | int check_orphan_space(struct f2fs_sb_info *sbi) | |
186 | { | |
187 | unsigned int max_orphans; | |
188 | int err = 0; | |
189 | ||
190 | /* | |
191 | * considering 512 blocks in a segment 5 blocks are needed for cp | |
192 | * and log segment summaries. Remaining blocks are used to keep | |
193 | * orphan entries with the limitation one reserved segment | |
194 | * for cp pack we can have max 1020*507 orphan entries | |
195 | */ | |
196 | max_orphans = (sbi->blocks_per_seg - 5) * F2FS_ORPHANS_PER_BLOCK; | |
197 | mutex_lock(&sbi->orphan_inode_mutex); | |
198 | if (sbi->n_orphans >= max_orphans) | |
199 | err = -ENOSPC; | |
200 | mutex_unlock(&sbi->orphan_inode_mutex); | |
201 | return err; | |
202 | } | |
203 | ||
204 | void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
205 | { | |
206 | struct list_head *head, *this; | |
207 | struct orphan_inode_entry *new = NULL, *orphan = NULL; | |
208 | ||
209 | mutex_lock(&sbi->orphan_inode_mutex); | |
210 | head = &sbi->orphan_inode_list; | |
211 | list_for_each(this, head) { | |
212 | orphan = list_entry(this, struct orphan_inode_entry, list); | |
213 | if (orphan->ino == ino) | |
214 | goto out; | |
215 | if (orphan->ino > ino) | |
216 | break; | |
217 | orphan = NULL; | |
218 | } | |
219 | retry: | |
220 | new = kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC); | |
221 | if (!new) { | |
222 | cond_resched(); | |
223 | goto retry; | |
224 | } | |
225 | new->ino = ino; | |
127e670a JK |
226 | |
227 | /* add new_oentry into list which is sorted by inode number */ | |
a2617dc6 | 228 | if (orphan) |
229 | list_add(&new->list, this->prev); | |
230 | else | |
127e670a | 231 | list_add_tail(&new->list, head); |
a2617dc6 | 232 | |
127e670a JK |
233 | sbi->n_orphans++; |
234 | out: | |
235 | mutex_unlock(&sbi->orphan_inode_mutex); | |
236 | } | |
237 | ||
238 | void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
239 | { | |
60ed9a0f | 240 | struct list_head *head; |
127e670a JK |
241 | struct orphan_inode_entry *orphan; |
242 | ||
243 | mutex_lock(&sbi->orphan_inode_mutex); | |
244 | head = &sbi->orphan_inode_list; | |
60ed9a0f | 245 | list_for_each_entry(orphan, head, list) { |
127e670a JK |
246 | if (orphan->ino == ino) { |
247 | list_del(&orphan->list); | |
248 | kmem_cache_free(orphan_entry_slab, orphan); | |
249 | sbi->n_orphans--; | |
250 | break; | |
251 | } | |
252 | } | |
253 | mutex_unlock(&sbi->orphan_inode_mutex); | |
254 | } | |
255 | ||
256 | static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
257 | { | |
258 | struct inode *inode = f2fs_iget(sbi->sb, ino); | |
259 | BUG_ON(IS_ERR(inode)); | |
260 | clear_nlink(inode); | |
261 | ||
262 | /* truncate all the data during iput */ | |
263 | iput(inode); | |
264 | } | |
265 | ||
266 | int recover_orphan_inodes(struct f2fs_sb_info *sbi) | |
267 | { | |
268 | block_t start_blk, orphan_blkaddr, i, j; | |
269 | ||
25ca923b | 270 | if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG)) |
127e670a JK |
271 | return 0; |
272 | ||
273 | sbi->por_doing = 1; | |
274 | start_blk = __start_cp_addr(sbi) + 1; | |
275 | orphan_blkaddr = __start_sum_addr(sbi) - 1; | |
276 | ||
277 | for (i = 0; i < orphan_blkaddr; i++) { | |
278 | struct page *page = get_meta_page(sbi, start_blk + i); | |
279 | struct f2fs_orphan_block *orphan_blk; | |
280 | ||
281 | orphan_blk = (struct f2fs_orphan_block *)page_address(page); | |
282 | for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) { | |
283 | nid_t ino = le32_to_cpu(orphan_blk->ino[j]); | |
284 | recover_orphan_inode(sbi, ino); | |
285 | } | |
286 | f2fs_put_page(page, 1); | |
287 | } | |
288 | /* clear Orphan Flag */ | |
25ca923b | 289 | clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG); |
127e670a JK |
290 | sbi->por_doing = 0; |
291 | return 0; | |
292 | } | |
293 | ||
294 | static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk) | |
295 | { | |
296 | struct list_head *head, *this, *next; | |
297 | struct f2fs_orphan_block *orphan_blk = NULL; | |
298 | struct page *page = NULL; | |
299 | unsigned int nentries = 0; | |
300 | unsigned short index = 1; | |
301 | unsigned short orphan_blocks; | |
302 | ||
303 | orphan_blocks = (unsigned short)((sbi->n_orphans + | |
304 | (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK); | |
305 | ||
306 | mutex_lock(&sbi->orphan_inode_mutex); | |
307 | head = &sbi->orphan_inode_list; | |
308 | ||
309 | /* loop for each orphan inode entry and write them in Jornal block */ | |
310 | list_for_each_safe(this, next, head) { | |
311 | struct orphan_inode_entry *orphan; | |
312 | ||
313 | orphan = list_entry(this, struct orphan_inode_entry, list); | |
314 | ||
315 | if (nentries == F2FS_ORPHANS_PER_BLOCK) { | |
316 | /* | |
317 | * an orphan block is full of 1020 entries, | |
318 | * then we need to flush current orphan blocks | |
319 | * and bring another one in memory | |
320 | */ | |
321 | orphan_blk->blk_addr = cpu_to_le16(index); | |
322 | orphan_blk->blk_count = cpu_to_le16(orphan_blocks); | |
323 | orphan_blk->entry_count = cpu_to_le32(nentries); | |
324 | set_page_dirty(page); | |
325 | f2fs_put_page(page, 1); | |
326 | index++; | |
327 | start_blk++; | |
328 | nentries = 0; | |
329 | page = NULL; | |
330 | } | |
331 | if (page) | |
332 | goto page_exist; | |
333 | ||
334 | page = grab_meta_page(sbi, start_blk); | |
335 | orphan_blk = (struct f2fs_orphan_block *)page_address(page); | |
336 | memset(orphan_blk, 0, sizeof(*orphan_blk)); | |
337 | page_exist: | |
338 | orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino); | |
339 | } | |
340 | if (!page) | |
341 | goto end; | |
342 | ||
343 | orphan_blk->blk_addr = cpu_to_le16(index); | |
344 | orphan_blk->blk_count = cpu_to_le16(orphan_blocks); | |
345 | orphan_blk->entry_count = cpu_to_le32(nentries); | |
346 | set_page_dirty(page); | |
347 | f2fs_put_page(page, 1); | |
348 | end: | |
349 | mutex_unlock(&sbi->orphan_inode_mutex); | |
350 | } | |
351 | ||
352 | static struct page *validate_checkpoint(struct f2fs_sb_info *sbi, | |
353 | block_t cp_addr, unsigned long long *version) | |
354 | { | |
355 | struct page *cp_page_1, *cp_page_2 = NULL; | |
356 | unsigned long blk_size = sbi->blocksize; | |
357 | struct f2fs_checkpoint *cp_block; | |
358 | unsigned long long cur_version = 0, pre_version = 0; | |
127e670a | 359 | size_t crc_offset; |
7e586fa0 | 360 | __u32 crc = 0; |
127e670a JK |
361 | |
362 | /* Read the 1st cp block in this CP pack */ | |
363 | cp_page_1 = get_meta_page(sbi, cp_addr); | |
364 | ||
365 | /* get the version number */ | |
366 | cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1); | |
367 | crc_offset = le32_to_cpu(cp_block->checksum_offset); | |
368 | if (crc_offset >= blk_size) | |
369 | goto invalid_cp1; | |
370 | ||
7e586fa0 | 371 | crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset))); |
127e670a JK |
372 | if (!f2fs_crc_valid(crc, cp_block, crc_offset)) |
373 | goto invalid_cp1; | |
374 | ||
375 | pre_version = le64_to_cpu(cp_block->checkpoint_ver); | |
376 | ||
377 | /* Read the 2nd cp block in this CP pack */ | |
25ca923b | 378 | cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1; |
127e670a JK |
379 | cp_page_2 = get_meta_page(sbi, cp_addr); |
380 | ||
381 | cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2); | |
382 | crc_offset = le32_to_cpu(cp_block->checksum_offset); | |
383 | if (crc_offset >= blk_size) | |
384 | goto invalid_cp2; | |
385 | ||
7e586fa0 | 386 | crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset))); |
127e670a JK |
387 | if (!f2fs_crc_valid(crc, cp_block, crc_offset)) |
388 | goto invalid_cp2; | |
389 | ||
390 | cur_version = le64_to_cpu(cp_block->checkpoint_ver); | |
391 | ||
392 | if (cur_version == pre_version) { | |
393 | *version = cur_version; | |
394 | f2fs_put_page(cp_page_2, 1); | |
395 | return cp_page_1; | |
396 | } | |
397 | invalid_cp2: | |
398 | f2fs_put_page(cp_page_2, 1); | |
399 | invalid_cp1: | |
400 | f2fs_put_page(cp_page_1, 1); | |
401 | return NULL; | |
402 | } | |
403 | ||
404 | int get_valid_checkpoint(struct f2fs_sb_info *sbi) | |
405 | { | |
406 | struct f2fs_checkpoint *cp_block; | |
407 | struct f2fs_super_block *fsb = sbi->raw_super; | |
408 | struct page *cp1, *cp2, *cur_page; | |
409 | unsigned long blk_size = sbi->blocksize; | |
410 | unsigned long long cp1_version = 0, cp2_version = 0; | |
411 | unsigned long long cp_start_blk_no; | |
412 | ||
413 | sbi->ckpt = kzalloc(blk_size, GFP_KERNEL); | |
414 | if (!sbi->ckpt) | |
415 | return -ENOMEM; | |
416 | /* | |
417 | * Finding out valid cp block involves read both | |
418 | * sets( cp pack1 and cp pack 2) | |
419 | */ | |
420 | cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr); | |
421 | cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version); | |
422 | ||
423 | /* The second checkpoint pack should start at the next segment */ | |
424 | cp_start_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg); | |
425 | cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version); | |
426 | ||
427 | if (cp1 && cp2) { | |
428 | if (ver_after(cp2_version, cp1_version)) | |
429 | cur_page = cp2; | |
430 | else | |
431 | cur_page = cp1; | |
432 | } else if (cp1) { | |
433 | cur_page = cp1; | |
434 | } else if (cp2) { | |
435 | cur_page = cp2; | |
436 | } else { | |
437 | goto fail_no_cp; | |
438 | } | |
439 | ||
440 | cp_block = (struct f2fs_checkpoint *)page_address(cur_page); | |
441 | memcpy(sbi->ckpt, cp_block, blk_size); | |
442 | ||
443 | f2fs_put_page(cp1, 1); | |
444 | f2fs_put_page(cp2, 1); | |
445 | return 0; | |
446 | ||
447 | fail_no_cp: | |
448 | kfree(sbi->ckpt); | |
449 | return -EINVAL; | |
450 | } | |
451 | ||
5deb8267 | 452 | static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new) |
127e670a JK |
453 | { |
454 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
455 | struct list_head *head = &sbi->dir_inode_list; | |
127e670a JK |
456 | struct list_head *this; |
457 | ||
5deb8267 JK |
458 | list_for_each(this, head) { |
459 | struct dir_inode_entry *entry; | |
460 | entry = list_entry(this, struct dir_inode_entry, list); | |
461 | if (entry->inode == inode) | |
462 | return -EEXIST; | |
463 | } | |
464 | list_add_tail(&new->list, head); | |
465 | #ifdef CONFIG_F2FS_STAT_FS | |
466 | sbi->n_dirty_dirs++; | |
467 | #endif | |
468 | return 0; | |
469 | } | |
470 | ||
471 | void set_dirty_dir_page(struct inode *inode, struct page *page) | |
472 | { | |
473 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
474 | struct dir_inode_entry *new; | |
475 | ||
127e670a JK |
476 | if (!S_ISDIR(inode->i_mode)) |
477 | return; | |
478 | retry: | |
479 | new = kmem_cache_alloc(inode_entry_slab, GFP_NOFS); | |
480 | if (!new) { | |
481 | cond_resched(); | |
482 | goto retry; | |
483 | } | |
484 | new->inode = inode; | |
485 | INIT_LIST_HEAD(&new->list); | |
486 | ||
487 | spin_lock(&sbi->dir_inode_lock); | |
5deb8267 JK |
488 | if (__add_dirty_inode(inode, new)) |
489 | kmem_cache_free(inode_entry_slab, new); | |
127e670a | 490 | |
127e670a JK |
491 | inc_page_count(sbi, F2FS_DIRTY_DENTS); |
492 | inode_inc_dirty_dents(inode); | |
493 | SetPagePrivate(page); | |
5deb8267 JK |
494 | spin_unlock(&sbi->dir_inode_lock); |
495 | } | |
496 | ||
497 | void add_dirty_dir_inode(struct inode *inode) | |
498 | { | |
499 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
500 | struct dir_inode_entry *new; | |
501 | retry: | |
502 | new = kmem_cache_alloc(inode_entry_slab, GFP_NOFS); | |
503 | if (!new) { | |
504 | cond_resched(); | |
505 | goto retry; | |
506 | } | |
507 | new->inode = inode; | |
508 | INIT_LIST_HEAD(&new->list); | |
127e670a | 509 | |
5deb8267 JK |
510 | spin_lock(&sbi->dir_inode_lock); |
511 | if (__add_dirty_inode(inode, new)) | |
512 | kmem_cache_free(inode_entry_slab, new); | |
127e670a JK |
513 | spin_unlock(&sbi->dir_inode_lock); |
514 | } | |
515 | ||
516 | void remove_dirty_dir_inode(struct inode *inode) | |
517 | { | |
518 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
519 | struct list_head *head = &sbi->dir_inode_list; | |
520 | struct list_head *this; | |
521 | ||
522 | if (!S_ISDIR(inode->i_mode)) | |
523 | return; | |
524 | ||
525 | spin_lock(&sbi->dir_inode_lock); | |
3b10b1fd JK |
526 | if (atomic_read(&F2FS_I(inode)->dirty_dents)) { |
527 | spin_unlock(&sbi->dir_inode_lock); | |
528 | return; | |
529 | } | |
127e670a JK |
530 | |
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); | |
35b09d82 | 537 | #ifdef CONFIG_F2FS_STAT_FS |
127e670a | 538 | sbi->n_dirty_dirs--; |
35b09d82 | 539 | #endif |
127e670a JK |
540 | break; |
541 | } | |
542 | } | |
127e670a | 543 | spin_unlock(&sbi->dir_inode_lock); |
74d0b917 JK |
544 | |
545 | /* Only from the recovery routine */ | |
afc3eda2 JK |
546 | if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) { |
547 | clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT); | |
74d0b917 | 548 | iput(inode); |
afc3eda2 | 549 | } |
74d0b917 JK |
550 | } |
551 | ||
552 | struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
553 | { | |
554 | struct list_head *head = &sbi->dir_inode_list; | |
555 | struct list_head *this; | |
556 | struct inode *inode = NULL; | |
557 | ||
558 | spin_lock(&sbi->dir_inode_lock); | |
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 | { | |
573 | struct list_head *head = &sbi->dir_inode_list; | |
574 | struct dir_inode_entry *entry; | |
575 | struct inode *inode; | |
576 | retry: | |
577 | spin_lock(&sbi->dir_inode_lock); | |
578 | if (list_empty(head)) { | |
579 | spin_unlock(&sbi->dir_inode_lock); | |
580 | return; | |
581 | } | |
582 | entry = list_entry(head->next, struct dir_inode_entry, list); | |
583 | inode = igrab(entry->inode); | |
584 | spin_unlock(&sbi->dir_inode_lock); | |
585 | if (inode) { | |
586 | filemap_flush(inode->i_mapping); | |
587 | iput(inode); | |
588 | } else { | |
589 | /* | |
590 | * We should submit bio, since it exists several | |
591 | * wribacking dentry pages in the freeing inode. | |
592 | */ | |
593 | f2fs_submit_bio(sbi, DATA, true); | |
594 | } | |
595 | goto retry; | |
596 | } | |
597 | ||
0a8165d7 | 598 | /* |
127e670a JK |
599 | * Freeze all the FS-operations for checkpoint. |
600 | */ | |
43727527 | 601 | static void block_operations(struct f2fs_sb_info *sbi) |
127e670a | 602 | { |
127e670a JK |
603 | struct writeback_control wbc = { |
604 | .sync_mode = WB_SYNC_ALL, | |
605 | .nr_to_write = LONG_MAX, | |
606 | .for_reclaim = 0, | |
607 | }; | |
c718379b JK |
608 | struct blk_plug plug; |
609 | ||
610 | blk_start_plug(&plug); | |
611 | ||
39936837 JK |
612 | retry_flush_dents: |
613 | mutex_lock_all(sbi); | |
127e670a | 614 | |
127e670a | 615 | /* write all the dirty dentry pages */ |
127e670a | 616 | if (get_pages(sbi, F2FS_DIRTY_DENTS)) { |
39936837 JK |
617 | mutex_unlock_all(sbi); |
618 | sync_dirty_dir_inodes(sbi); | |
619 | goto retry_flush_dents; | |
127e670a JK |
620 | } |
621 | ||
127e670a JK |
622 | /* |
623 | * POR: we should ensure that there is no dirty node pages | |
624 | * until finishing nat/sit flush. | |
625 | */ | |
39936837 JK |
626 | retry_flush_nodes: |
627 | mutex_lock(&sbi->node_write); | |
127e670a JK |
628 | |
629 | if (get_pages(sbi, F2FS_DIRTY_NODES)) { | |
39936837 JK |
630 | mutex_unlock(&sbi->node_write); |
631 | sync_node_pages(sbi, 0, &wbc); | |
632 | goto retry_flush_nodes; | |
127e670a | 633 | } |
c718379b | 634 | blk_finish_plug(&plug); |
127e670a JK |
635 | } |
636 | ||
637 | static void unblock_operations(struct f2fs_sb_info *sbi) | |
638 | { | |
39936837 JK |
639 | mutex_unlock(&sbi->node_write); |
640 | mutex_unlock_all(sbi); | |
127e670a JK |
641 | } |
642 | ||
643 | static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount) | |
644 | { | |
645 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
646 | nid_t last_nid = 0; | |
647 | block_t start_blk; | |
648 | struct page *cp_page; | |
649 | unsigned int data_sum_blocks, orphan_blocks; | |
7e586fa0 | 650 | __u32 crc32 = 0; |
127e670a | 651 | void *kaddr; |
127e670a JK |
652 | int i; |
653 | ||
654 | /* Flush all the NAT/SIT pages */ | |
655 | while (get_pages(sbi, F2FS_DIRTY_META)) | |
656 | sync_meta_pages(sbi, META, LONG_MAX); | |
657 | ||
658 | next_free_nid(sbi, &last_nid); | |
659 | ||
660 | /* | |
661 | * modify checkpoint | |
662 | * version number is already updated | |
663 | */ | |
664 | ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi)); | |
665 | ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi)); | |
666 | ckpt->free_segment_count = cpu_to_le32(free_segments(sbi)); | |
667 | for (i = 0; i < 3; i++) { | |
668 | ckpt->cur_node_segno[i] = | |
669 | cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE)); | |
670 | ckpt->cur_node_blkoff[i] = | |
671 | cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE)); | |
672 | ckpt->alloc_type[i + CURSEG_HOT_NODE] = | |
673 | curseg_alloc_type(sbi, i + CURSEG_HOT_NODE); | |
674 | } | |
675 | for (i = 0; i < 3; i++) { | |
676 | ckpt->cur_data_segno[i] = | |
677 | cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA)); | |
678 | ckpt->cur_data_blkoff[i] = | |
679 | cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA)); | |
680 | ckpt->alloc_type[i + CURSEG_HOT_DATA] = | |
681 | curseg_alloc_type(sbi, i + CURSEG_HOT_DATA); | |
682 | } | |
683 | ||
684 | ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi)); | |
685 | ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi)); | |
686 | ckpt->next_free_nid = cpu_to_le32(last_nid); | |
687 | ||
688 | /* 2 cp + n data seg summary + orphan inode blocks */ | |
689 | data_sum_blocks = npages_for_summary_flush(sbi); | |
690 | if (data_sum_blocks < 3) | |
25ca923b | 691 | set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); |
127e670a | 692 | else |
25ca923b | 693 | clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); |
127e670a JK |
694 | |
695 | orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1) | |
696 | / F2FS_ORPHANS_PER_BLOCK; | |
25ca923b | 697 | ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks); |
127e670a JK |
698 | |
699 | if (is_umount) { | |
25ca923b JK |
700 | set_ckpt_flags(ckpt, CP_UMOUNT_FLAG); |
701 | ckpt->cp_pack_total_block_count = cpu_to_le32(2 + | |
702 | data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE); | |
127e670a | 703 | } else { |
25ca923b JK |
704 | clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG); |
705 | ckpt->cp_pack_total_block_count = cpu_to_le32(2 + | |
706 | data_sum_blocks + orphan_blocks); | |
127e670a JK |
707 | } |
708 | ||
709 | if (sbi->n_orphans) | |
25ca923b | 710 | set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); |
127e670a | 711 | else |
25ca923b | 712 | clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); |
127e670a JK |
713 | |
714 | /* update SIT/NAT bitmap */ | |
715 | get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP)); | |
716 | get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP)); | |
717 | ||
718 | crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset)); | |
7e586fa0 JK |
719 | *((__le32 *)((unsigned char *)ckpt + |
720 | le32_to_cpu(ckpt->checksum_offset))) | |
127e670a JK |
721 | = cpu_to_le32(crc32); |
722 | ||
723 | start_blk = __start_cp_addr(sbi); | |
724 | ||
725 | /* write out checkpoint buffer at block 0 */ | |
726 | cp_page = grab_meta_page(sbi, start_blk++); | |
727 | kaddr = page_address(cp_page); | |
728 | memcpy(kaddr, ckpt, (1 << sbi->log_blocksize)); | |
729 | set_page_dirty(cp_page); | |
730 | f2fs_put_page(cp_page, 1); | |
731 | ||
732 | if (sbi->n_orphans) { | |
733 | write_orphan_inodes(sbi, start_blk); | |
734 | start_blk += orphan_blocks; | |
735 | } | |
736 | ||
737 | write_data_summaries(sbi, start_blk); | |
738 | start_blk += data_sum_blocks; | |
739 | if (is_umount) { | |
740 | write_node_summaries(sbi, start_blk); | |
741 | start_blk += NR_CURSEG_NODE_TYPE; | |
742 | } | |
743 | ||
744 | /* writeout checkpoint block */ | |
745 | cp_page = grab_meta_page(sbi, start_blk); | |
746 | kaddr = page_address(cp_page); | |
747 | memcpy(kaddr, ckpt, (1 << sbi->log_blocksize)); | |
748 | set_page_dirty(cp_page); | |
749 | f2fs_put_page(cp_page, 1); | |
750 | ||
751 | /* wait for previous submitted node/meta pages writeback */ | |
752 | while (get_pages(sbi, F2FS_WRITEBACK)) | |
753 | congestion_wait(BLK_RW_ASYNC, HZ / 50); | |
754 | ||
755 | filemap_fdatawait_range(sbi->node_inode->i_mapping, 0, LONG_MAX); | |
756 | filemap_fdatawait_range(sbi->meta_inode->i_mapping, 0, LONG_MAX); | |
757 | ||
758 | /* update user_block_counts */ | |
759 | sbi->last_valid_block_count = sbi->total_valid_block_count; | |
760 | sbi->alloc_valid_block_count = 0; | |
761 | ||
762 | /* Here, we only have one bio having CP pack */ | |
577e3495 | 763 | sync_meta_pages(sbi, META_FLUSH, LONG_MAX); |
127e670a | 764 | |
577e3495 JK |
765 | if (!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG)) { |
766 | clear_prefree_segments(sbi); | |
767 | F2FS_RESET_SB_DIRT(sbi); | |
768 | } | |
127e670a JK |
769 | } |
770 | ||
0a8165d7 | 771 | /* |
127e670a JK |
772 | * We guarantee that this checkpoint procedure should not fail. |
773 | */ | |
43727527 | 774 | void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount) |
127e670a JK |
775 | { |
776 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
777 | unsigned long long ckpt_ver; | |
778 | ||
2af4bd6c NJ |
779 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops"); |
780 | ||
43727527 JK |
781 | mutex_lock(&sbi->cp_mutex); |
782 | block_operations(sbi); | |
127e670a | 783 | |
2af4bd6c NJ |
784 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops"); |
785 | ||
127e670a JK |
786 | f2fs_submit_bio(sbi, DATA, true); |
787 | f2fs_submit_bio(sbi, NODE, true); | |
788 | f2fs_submit_bio(sbi, META, true); | |
789 | ||
790 | /* | |
791 | * update checkpoint pack index | |
792 | * Increase the version number so that | |
793 | * SIT entries and seg summaries are written at correct place | |
794 | */ | |
795 | ckpt_ver = le64_to_cpu(ckpt->checkpoint_ver); | |
796 | ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver); | |
797 | ||
798 | /* write cached NAT/SIT entries to NAT/SIT area */ | |
799 | flush_nat_entries(sbi); | |
800 | flush_sit_entries(sbi); | |
801 | ||
127e670a JK |
802 | /* unlock all the fs_lock[] in do_checkpoint() */ |
803 | do_checkpoint(sbi, is_umount); | |
804 | ||
805 | unblock_operations(sbi); | |
806 | mutex_unlock(&sbi->cp_mutex); | |
2af4bd6c NJ |
807 | |
808 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint"); | |
127e670a JK |
809 | } |
810 | ||
811 | void init_orphan_info(struct f2fs_sb_info *sbi) | |
812 | { | |
813 | mutex_init(&sbi->orphan_inode_mutex); | |
814 | INIT_LIST_HEAD(&sbi->orphan_inode_list); | |
815 | sbi->n_orphans = 0; | |
816 | } | |
817 | ||
6e6093a8 | 818 | int __init create_checkpoint_caches(void) |
127e670a JK |
819 | { |
820 | orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry", | |
821 | sizeof(struct orphan_inode_entry), NULL); | |
822 | if (unlikely(!orphan_entry_slab)) | |
823 | return -ENOMEM; | |
824 | inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry", | |
825 | sizeof(struct dir_inode_entry), NULL); | |
826 | if (unlikely(!inode_entry_slab)) { | |
827 | kmem_cache_destroy(orphan_entry_slab); | |
828 | return -ENOMEM; | |
829 | } | |
830 | return 0; | |
831 | } | |
832 | ||
833 | void destroy_checkpoint_caches(void) | |
834 | { | |
835 | kmem_cache_destroy(orphan_entry_slab); | |
836 | kmem_cache_destroy(inode_entry_slab); | |
837 | } |