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 | { | |
9df27d98 | 33 | struct address_space *mapping = META_MAPPING(sbi); |
127e670a JK |
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 | { | |
9df27d98 | 53 | struct address_space *mapping = META_MAPPING(sbi); |
127e670a JK |
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); |
6bacf52f | 69 | if (unlikely(page->mapping != mapping)) { |
afcb7ca0 JK |
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 */ |
cfb271d4 CY |
85 | if (unlikely(sbi->por_doing || |
86 | is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG))) | |
87 | goto redirty_out; | |
88 | ||
89 | if (wbc->for_reclaim) | |
90 | goto redirty_out; | |
127e670a | 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; | |
cfb271d4 CY |
98 | |
99 | redirty_out: | |
100 | dec_page_count(sbi, F2FS_DIRTY_META); | |
101 | wbc->pages_skipped++; | |
102 | set_page_dirty(page); | |
103 | return AOP_WRITEPAGE_ACTIVATE; | |
127e670a JK |
104 | } |
105 | ||
106 | static int f2fs_write_meta_pages(struct address_space *mapping, | |
107 | struct writeback_control *wbc) | |
108 | { | |
109 | struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); | |
5459aa97 | 110 | int nrpages = MAX_BIO_BLOCKS(max_hw_blocks(sbi)); |
127e670a JK |
111 | long written; |
112 | ||
113 | if (wbc->for_kupdate) | |
114 | return 0; | |
115 | ||
5459aa97 JK |
116 | /* collect a number of dirty meta pages and write together */ |
117 | if (get_pages(sbi, F2FS_DIRTY_META) < nrpages) | |
127e670a JK |
118 | return 0; |
119 | ||
120 | /* if mounting is failed, skip writing node pages */ | |
121 | mutex_lock(&sbi->cp_mutex); | |
5459aa97 | 122 | written = sync_meta_pages(sbi, META, nrpages); |
127e670a JK |
123 | mutex_unlock(&sbi->cp_mutex); |
124 | wbc->nr_to_write -= written; | |
125 | return 0; | |
126 | } | |
127 | ||
128 | long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type, | |
129 | long nr_to_write) | |
130 | { | |
9df27d98 | 131 | struct address_space *mapping = META_MAPPING(sbi); |
127e670a JK |
132 | pgoff_t index = 0, end = LONG_MAX; |
133 | struct pagevec pvec; | |
134 | long nwritten = 0; | |
135 | struct writeback_control wbc = { | |
136 | .for_reclaim = 0, | |
137 | }; | |
138 | ||
139 | pagevec_init(&pvec, 0); | |
140 | ||
141 | while (index <= end) { | |
142 | int i, nr_pages; | |
143 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
144 | PAGECACHE_TAG_DIRTY, | |
145 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); | |
cfb271d4 | 146 | if (unlikely(nr_pages == 0)) |
127e670a JK |
147 | break; |
148 | ||
149 | for (i = 0; i < nr_pages; i++) { | |
150 | struct page *page = pvec.pages[i]; | |
151 | lock_page(page); | |
5d56b671 JK |
152 | f2fs_bug_on(page->mapping != mapping); |
153 | f2fs_bug_on(!PageDirty(page)); | |
127e670a | 154 | clear_page_dirty_for_io(page); |
577e3495 JK |
155 | if (f2fs_write_meta_page(page, &wbc)) { |
156 | unlock_page(page); | |
157 | break; | |
158 | } | |
cfb271d4 CY |
159 | nwritten++; |
160 | if (unlikely(nwritten >= nr_to_write)) | |
127e670a JK |
161 | break; |
162 | } | |
163 | pagevec_release(&pvec); | |
164 | cond_resched(); | |
165 | } | |
166 | ||
167 | if (nwritten) | |
458e6197 | 168 | f2fs_submit_merged_bio(sbi, type, WRITE); |
127e670a JK |
169 | |
170 | return nwritten; | |
171 | } | |
172 | ||
173 | static int f2fs_set_meta_page_dirty(struct page *page) | |
174 | { | |
175 | struct address_space *mapping = page->mapping; | |
176 | struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); | |
177 | ||
26c6b887 JK |
178 | trace_f2fs_set_page_dirty(page, META); |
179 | ||
127e670a JK |
180 | SetPageUptodate(page); |
181 | if (!PageDirty(page)) { | |
182 | __set_page_dirty_nobuffers(page); | |
183 | inc_page_count(sbi, F2FS_DIRTY_META); | |
127e670a JK |
184 | return 1; |
185 | } | |
186 | return 0; | |
187 | } | |
188 | ||
189 | const struct address_space_operations f2fs_meta_aops = { | |
190 | .writepage = f2fs_write_meta_page, | |
191 | .writepages = f2fs_write_meta_pages, | |
192 | .set_page_dirty = f2fs_set_meta_page_dirty, | |
193 | }; | |
194 | ||
cbd56e7d | 195 | int acquire_orphan_inode(struct f2fs_sb_info *sbi) |
127e670a | 196 | { |
127e670a JK |
197 | int err = 0; |
198 | ||
17b692f6 | 199 | spin_lock(&sbi->orphan_inode_lock); |
0d47c1ad | 200 | if (unlikely(sbi->n_orphans >= sbi->max_orphans)) |
127e670a | 201 | err = -ENOSPC; |
cbd56e7d JK |
202 | else |
203 | sbi->n_orphans++; | |
17b692f6 | 204 | spin_unlock(&sbi->orphan_inode_lock); |
0d47c1ad | 205 | |
127e670a JK |
206 | return err; |
207 | } | |
208 | ||
cbd56e7d JK |
209 | void release_orphan_inode(struct f2fs_sb_info *sbi) |
210 | { | |
17b692f6 | 211 | spin_lock(&sbi->orphan_inode_lock); |
5d56b671 | 212 | f2fs_bug_on(sbi->n_orphans == 0); |
cbd56e7d | 213 | sbi->n_orphans--; |
17b692f6 | 214 | spin_unlock(&sbi->orphan_inode_lock); |
cbd56e7d JK |
215 | } |
216 | ||
127e670a JK |
217 | void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) |
218 | { | |
219 | struct list_head *head, *this; | |
220 | struct orphan_inode_entry *new = NULL, *orphan = NULL; | |
221 | ||
c1ef3725 GZ |
222 | new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC); |
223 | new->ino = ino; | |
224 | ||
17b692f6 | 225 | spin_lock(&sbi->orphan_inode_lock); |
127e670a JK |
226 | head = &sbi->orphan_inode_list; |
227 | list_for_each(this, head) { | |
228 | orphan = list_entry(this, struct orphan_inode_entry, list); | |
c1ef3725 | 229 | if (orphan->ino == ino) { |
17b692f6 | 230 | spin_unlock(&sbi->orphan_inode_lock); |
c1ef3725 GZ |
231 | kmem_cache_free(orphan_entry_slab, new); |
232 | return; | |
233 | } | |
234 | ||
127e670a JK |
235 | if (orphan->ino > ino) |
236 | break; | |
237 | orphan = NULL; | |
238 | } | |
7bd59381 | 239 | |
127e670a | 240 | /* add new_oentry into list which is sorted by inode number */ |
a2617dc6 | 241 | if (orphan) |
242 | list_add(&new->list, this->prev); | |
243 | else | |
127e670a | 244 | list_add_tail(&new->list, head); |
17b692f6 | 245 | spin_unlock(&sbi->orphan_inode_lock); |
127e670a JK |
246 | } |
247 | ||
248 | void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
249 | { | |
60ed9a0f | 250 | struct list_head *head; |
127e670a JK |
251 | struct orphan_inode_entry *orphan; |
252 | ||
17b692f6 | 253 | spin_lock(&sbi->orphan_inode_lock); |
127e670a | 254 | head = &sbi->orphan_inode_list; |
60ed9a0f | 255 | list_for_each_entry(orphan, head, list) { |
127e670a JK |
256 | if (orphan->ino == ino) { |
257 | list_del(&orphan->list); | |
258 | kmem_cache_free(orphan_entry_slab, orphan); | |
5d56b671 | 259 | f2fs_bug_on(sbi->n_orphans == 0); |
127e670a JK |
260 | sbi->n_orphans--; |
261 | break; | |
262 | } | |
263 | } | |
17b692f6 | 264 | spin_unlock(&sbi->orphan_inode_lock); |
127e670a JK |
265 | } |
266 | ||
267 | static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
268 | { | |
269 | struct inode *inode = f2fs_iget(sbi->sb, ino); | |
5d56b671 | 270 | f2fs_bug_on(IS_ERR(inode)); |
127e670a JK |
271 | clear_nlink(inode); |
272 | ||
273 | /* truncate all the data during iput */ | |
274 | iput(inode); | |
275 | } | |
276 | ||
8f99a946 | 277 | void recover_orphan_inodes(struct f2fs_sb_info *sbi) |
127e670a JK |
278 | { |
279 | block_t start_blk, orphan_blkaddr, i, j; | |
280 | ||
25ca923b | 281 | if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG)) |
8f99a946 | 282 | return; |
127e670a | 283 | |
aabe5136 | 284 | sbi->por_doing = true; |
127e670a JK |
285 | start_blk = __start_cp_addr(sbi) + 1; |
286 | orphan_blkaddr = __start_sum_addr(sbi) - 1; | |
287 | ||
288 | for (i = 0; i < orphan_blkaddr; i++) { | |
289 | struct page *page = get_meta_page(sbi, start_blk + i); | |
290 | struct f2fs_orphan_block *orphan_blk; | |
291 | ||
292 | orphan_blk = (struct f2fs_orphan_block *)page_address(page); | |
293 | for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) { | |
294 | nid_t ino = le32_to_cpu(orphan_blk->ino[j]); | |
295 | recover_orphan_inode(sbi, ino); | |
296 | } | |
297 | f2fs_put_page(page, 1); | |
298 | } | |
299 | /* clear Orphan Flag */ | |
25ca923b | 300 | clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG); |
aabe5136 | 301 | sbi->por_doing = false; |
8f99a946 | 302 | return; |
127e670a JK |
303 | } |
304 | ||
305 | static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk) | |
306 | { | |
502c6e0b | 307 | struct list_head *head; |
127e670a | 308 | struct f2fs_orphan_block *orphan_blk = NULL; |
127e670a | 309 | unsigned int nentries = 0; |
4531929e GZ |
310 | unsigned short index; |
311 | unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans + | |
312 | (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK); | |
313 | struct page *page = NULL; | |
502c6e0b | 314 | struct orphan_inode_entry *orphan = NULL; |
127e670a | 315 | |
4531929e | 316 | for (index = 0; index < orphan_blocks; index++) |
63f5384c | 317 | grab_meta_page(sbi, start_blk + index); |
127e670a | 318 | |
4531929e | 319 | index = 1; |
17b692f6 | 320 | spin_lock(&sbi->orphan_inode_lock); |
127e670a JK |
321 | head = &sbi->orphan_inode_list; |
322 | ||
323 | /* loop for each orphan inode entry and write them in Jornal block */ | |
502c6e0b GZ |
324 | list_for_each_entry(orphan, head, list) { |
325 | if (!page) { | |
63f5384c GZ |
326 | page = find_get_page(META_MAPPING(sbi), start_blk++); |
327 | f2fs_bug_on(!page); | |
502c6e0b GZ |
328 | orphan_blk = |
329 | (struct f2fs_orphan_block *)page_address(page); | |
330 | memset(orphan_blk, 0, sizeof(*orphan_blk)); | |
63f5384c | 331 | f2fs_put_page(page, 0); |
502c6e0b | 332 | } |
127e670a | 333 | |
36795567 | 334 | orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino); |
127e670a | 335 | |
36795567 | 336 | if (nentries == F2FS_ORPHANS_PER_BLOCK) { |
127e670a JK |
337 | /* |
338 | * an orphan block is full of 1020 entries, | |
339 | * then we need to flush current orphan blocks | |
340 | * and bring another one in memory | |
341 | */ | |
342 | orphan_blk->blk_addr = cpu_to_le16(index); | |
343 | orphan_blk->blk_count = cpu_to_le16(orphan_blocks); | |
344 | orphan_blk->entry_count = cpu_to_le32(nentries); | |
345 | set_page_dirty(page); | |
346 | f2fs_put_page(page, 1); | |
347 | index++; | |
127e670a JK |
348 | nentries = 0; |
349 | page = NULL; | |
350 | } | |
502c6e0b | 351 | } |
127e670a | 352 | |
502c6e0b GZ |
353 | if (page) { |
354 | orphan_blk->blk_addr = cpu_to_le16(index); | |
355 | orphan_blk->blk_count = cpu_to_le16(orphan_blocks); | |
356 | orphan_blk->entry_count = cpu_to_le32(nentries); | |
357 | set_page_dirty(page); | |
358 | f2fs_put_page(page, 1); | |
127e670a | 359 | } |
502c6e0b | 360 | |
17b692f6 | 361 | spin_unlock(&sbi->orphan_inode_lock); |
127e670a JK |
362 | } |
363 | ||
364 | static struct page *validate_checkpoint(struct f2fs_sb_info *sbi, | |
365 | block_t cp_addr, unsigned long long *version) | |
366 | { | |
367 | struct page *cp_page_1, *cp_page_2 = NULL; | |
368 | unsigned long blk_size = sbi->blocksize; | |
369 | struct f2fs_checkpoint *cp_block; | |
370 | unsigned long long cur_version = 0, pre_version = 0; | |
127e670a | 371 | size_t crc_offset; |
7e586fa0 | 372 | __u32 crc = 0; |
127e670a JK |
373 | |
374 | /* Read the 1st cp block in this CP pack */ | |
375 | cp_page_1 = get_meta_page(sbi, cp_addr); | |
376 | ||
377 | /* get the version number */ | |
378 | cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1); | |
379 | crc_offset = le32_to_cpu(cp_block->checksum_offset); | |
380 | if (crc_offset >= blk_size) | |
381 | goto invalid_cp1; | |
382 | ||
7e586fa0 | 383 | crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset))); |
127e670a JK |
384 | if (!f2fs_crc_valid(crc, cp_block, crc_offset)) |
385 | goto invalid_cp1; | |
386 | ||
d71b5564 | 387 | pre_version = cur_cp_version(cp_block); |
127e670a JK |
388 | |
389 | /* Read the 2nd cp block in this CP pack */ | |
25ca923b | 390 | cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1; |
127e670a JK |
391 | cp_page_2 = get_meta_page(sbi, cp_addr); |
392 | ||
393 | cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2); | |
394 | crc_offset = le32_to_cpu(cp_block->checksum_offset); | |
395 | if (crc_offset >= blk_size) | |
396 | goto invalid_cp2; | |
397 | ||
7e586fa0 | 398 | crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset))); |
127e670a JK |
399 | if (!f2fs_crc_valid(crc, cp_block, crc_offset)) |
400 | goto invalid_cp2; | |
401 | ||
d71b5564 | 402 | cur_version = cur_cp_version(cp_block); |
127e670a JK |
403 | |
404 | if (cur_version == pre_version) { | |
405 | *version = cur_version; | |
406 | f2fs_put_page(cp_page_2, 1); | |
407 | return cp_page_1; | |
408 | } | |
409 | invalid_cp2: | |
410 | f2fs_put_page(cp_page_2, 1); | |
411 | invalid_cp1: | |
412 | f2fs_put_page(cp_page_1, 1); | |
413 | return NULL; | |
414 | } | |
415 | ||
416 | int get_valid_checkpoint(struct f2fs_sb_info *sbi) | |
417 | { | |
418 | struct f2fs_checkpoint *cp_block; | |
419 | struct f2fs_super_block *fsb = sbi->raw_super; | |
420 | struct page *cp1, *cp2, *cur_page; | |
421 | unsigned long blk_size = sbi->blocksize; | |
422 | unsigned long long cp1_version = 0, cp2_version = 0; | |
423 | unsigned long long cp_start_blk_no; | |
424 | ||
425 | sbi->ckpt = kzalloc(blk_size, GFP_KERNEL); | |
426 | if (!sbi->ckpt) | |
427 | return -ENOMEM; | |
428 | /* | |
429 | * Finding out valid cp block involves read both | |
430 | * sets( cp pack1 and cp pack 2) | |
431 | */ | |
432 | cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr); | |
433 | cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version); | |
434 | ||
435 | /* The second checkpoint pack should start at the next segment */ | |
f9a4e6df JK |
436 | cp_start_blk_no += ((unsigned long long)1) << |
437 | le32_to_cpu(fsb->log_blocks_per_seg); | |
127e670a JK |
438 | cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version); |
439 | ||
440 | if (cp1 && cp2) { | |
441 | if (ver_after(cp2_version, cp1_version)) | |
442 | cur_page = cp2; | |
443 | else | |
444 | cur_page = cp1; | |
445 | } else if (cp1) { | |
446 | cur_page = cp1; | |
447 | } else if (cp2) { | |
448 | cur_page = cp2; | |
449 | } else { | |
450 | goto fail_no_cp; | |
451 | } | |
452 | ||
453 | cp_block = (struct f2fs_checkpoint *)page_address(cur_page); | |
454 | memcpy(sbi->ckpt, cp_block, blk_size); | |
455 | ||
456 | f2fs_put_page(cp1, 1); | |
457 | f2fs_put_page(cp2, 1); | |
458 | return 0; | |
459 | ||
460 | fail_no_cp: | |
461 | kfree(sbi->ckpt); | |
462 | return -EINVAL; | |
463 | } | |
464 | ||
5deb8267 | 465 | static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new) |
127e670a JK |
466 | { |
467 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
468 | struct list_head *head = &sbi->dir_inode_list; | |
127e670a JK |
469 | struct list_head *this; |
470 | ||
5deb8267 JK |
471 | list_for_each(this, head) { |
472 | struct dir_inode_entry *entry; | |
473 | entry = list_entry(this, struct dir_inode_entry, list); | |
6bacf52f | 474 | if (unlikely(entry->inode == inode)) |
5deb8267 JK |
475 | return -EEXIST; |
476 | } | |
477 | list_add_tail(&new->list, head); | |
dcdfff65 | 478 | stat_inc_dirty_dir(sbi); |
5deb8267 JK |
479 | return 0; |
480 | } | |
481 | ||
482 | void set_dirty_dir_page(struct inode *inode, struct page *page) | |
483 | { | |
484 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
485 | struct dir_inode_entry *new; | |
486 | ||
127e670a JK |
487 | if (!S_ISDIR(inode->i_mode)) |
488 | return; | |
7bd59381 GZ |
489 | |
490 | new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); | |
127e670a JK |
491 | new->inode = inode; |
492 | INIT_LIST_HEAD(&new->list); | |
493 | ||
494 | spin_lock(&sbi->dir_inode_lock); | |
5deb8267 JK |
495 | if (__add_dirty_inode(inode, new)) |
496 | kmem_cache_free(inode_entry_slab, new); | |
127e670a | 497 | |
127e670a JK |
498 | inc_page_count(sbi, F2FS_DIRTY_DENTS); |
499 | inode_inc_dirty_dents(inode); | |
500 | SetPagePrivate(page); | |
5deb8267 JK |
501 | spin_unlock(&sbi->dir_inode_lock); |
502 | } | |
503 | ||
504 | void add_dirty_dir_inode(struct inode *inode) | |
505 | { | |
506 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
7bd59381 GZ |
507 | struct dir_inode_entry *new = |
508 | f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); | |
509 | ||
5deb8267 JK |
510 | new->inode = inode; |
511 | INIT_LIST_HEAD(&new->list); | |
127e670a | 512 | |
5deb8267 JK |
513 | spin_lock(&sbi->dir_inode_lock); |
514 | if (__add_dirty_inode(inode, new)) | |
515 | kmem_cache_free(inode_entry_slab, new); | |
127e670a JK |
516 | spin_unlock(&sbi->dir_inode_lock); |
517 | } | |
518 | ||
519 | void remove_dirty_dir_inode(struct inode *inode) | |
520 | { | |
521 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
ce3b7d80 GZ |
522 | |
523 | struct list_head *this, *head; | |
127e670a JK |
524 | |
525 | if (!S_ISDIR(inode->i_mode)) | |
526 | return; | |
527 | ||
528 | spin_lock(&sbi->dir_inode_lock); | |
3b10b1fd JK |
529 | if (atomic_read(&F2FS_I(inode)->dirty_dents)) { |
530 | spin_unlock(&sbi->dir_inode_lock); | |
531 | return; | |
532 | } | |
127e670a | 533 | |
ce3b7d80 | 534 | head = &sbi->dir_inode_list; |
127e670a JK |
535 | list_for_each(this, head) { |
536 | struct dir_inode_entry *entry; | |
537 | entry = list_entry(this, struct dir_inode_entry, list); | |
538 | if (entry->inode == inode) { | |
539 | list_del(&entry->list); | |
540 | kmem_cache_free(inode_entry_slab, entry); | |
dcdfff65 | 541 | stat_dec_dirty_dir(sbi); |
127e670a JK |
542 | break; |
543 | } | |
544 | } | |
127e670a | 545 | spin_unlock(&sbi->dir_inode_lock); |
74d0b917 JK |
546 | |
547 | /* Only from the recovery routine */ | |
afc3eda2 JK |
548 | if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) { |
549 | clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT); | |
74d0b917 | 550 | iput(inode); |
afc3eda2 | 551 | } |
74d0b917 JK |
552 | } |
553 | ||
554 | struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
555 | { | |
ce3b7d80 GZ |
556 | |
557 | struct list_head *this, *head; | |
74d0b917 JK |
558 | struct inode *inode = NULL; |
559 | ||
560 | spin_lock(&sbi->dir_inode_lock); | |
ce3b7d80 GZ |
561 | |
562 | head = &sbi->dir_inode_list; | |
74d0b917 JK |
563 | list_for_each(this, head) { |
564 | struct dir_inode_entry *entry; | |
565 | entry = list_entry(this, struct dir_inode_entry, list); | |
566 | if (entry->inode->i_ino == ino) { | |
567 | inode = entry->inode; | |
568 | break; | |
569 | } | |
570 | } | |
571 | spin_unlock(&sbi->dir_inode_lock); | |
572 | return inode; | |
127e670a JK |
573 | } |
574 | ||
575 | void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi) | |
576 | { | |
ce3b7d80 | 577 | struct list_head *head; |
127e670a JK |
578 | struct dir_inode_entry *entry; |
579 | struct inode *inode; | |
580 | retry: | |
581 | spin_lock(&sbi->dir_inode_lock); | |
ce3b7d80 GZ |
582 | |
583 | head = &sbi->dir_inode_list; | |
127e670a JK |
584 | if (list_empty(head)) { |
585 | spin_unlock(&sbi->dir_inode_lock); | |
586 | return; | |
587 | } | |
588 | entry = list_entry(head->next, struct dir_inode_entry, list); | |
589 | inode = igrab(entry->inode); | |
590 | spin_unlock(&sbi->dir_inode_lock); | |
591 | if (inode) { | |
592 | filemap_flush(inode->i_mapping); | |
593 | iput(inode); | |
594 | } else { | |
595 | /* | |
596 | * We should submit bio, since it exists several | |
597 | * wribacking dentry pages in the freeing inode. | |
598 | */ | |
458e6197 | 599 | f2fs_submit_merged_bio(sbi, DATA, WRITE); |
127e670a JK |
600 | } |
601 | goto retry; | |
602 | } | |
603 | ||
0a8165d7 | 604 | /* |
127e670a JK |
605 | * Freeze all the FS-operations for checkpoint. |
606 | */ | |
43727527 | 607 | static void block_operations(struct f2fs_sb_info *sbi) |
127e670a | 608 | { |
127e670a JK |
609 | struct writeback_control wbc = { |
610 | .sync_mode = WB_SYNC_ALL, | |
611 | .nr_to_write = LONG_MAX, | |
612 | .for_reclaim = 0, | |
613 | }; | |
c718379b JK |
614 | struct blk_plug plug; |
615 | ||
616 | blk_start_plug(&plug); | |
617 | ||
39936837 | 618 | retry_flush_dents: |
e479556b | 619 | f2fs_lock_all(sbi); |
127e670a | 620 | /* write all the dirty dentry pages */ |
127e670a | 621 | if (get_pages(sbi, F2FS_DIRTY_DENTS)) { |
e479556b | 622 | f2fs_unlock_all(sbi); |
39936837 JK |
623 | sync_dirty_dir_inodes(sbi); |
624 | goto retry_flush_dents; | |
127e670a JK |
625 | } |
626 | ||
127e670a JK |
627 | /* |
628 | * POR: we should ensure that there is no dirty node pages | |
629 | * until finishing nat/sit flush. | |
630 | */ | |
39936837 JK |
631 | retry_flush_nodes: |
632 | mutex_lock(&sbi->node_write); | |
127e670a JK |
633 | |
634 | if (get_pages(sbi, F2FS_DIRTY_NODES)) { | |
39936837 JK |
635 | mutex_unlock(&sbi->node_write); |
636 | sync_node_pages(sbi, 0, &wbc); | |
637 | goto retry_flush_nodes; | |
127e670a | 638 | } |
c718379b | 639 | blk_finish_plug(&plug); |
127e670a JK |
640 | } |
641 | ||
642 | static void unblock_operations(struct f2fs_sb_info *sbi) | |
643 | { | |
39936837 | 644 | mutex_unlock(&sbi->node_write); |
e479556b | 645 | f2fs_unlock_all(sbi); |
127e670a JK |
646 | } |
647 | ||
fb51b5ef CL |
648 | static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi) |
649 | { | |
650 | DEFINE_WAIT(wait); | |
651 | ||
652 | for (;;) { | |
653 | prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE); | |
654 | ||
655 | if (!get_pages(sbi, F2FS_WRITEBACK)) | |
656 | break; | |
657 | ||
658 | io_schedule(); | |
659 | } | |
660 | finish_wait(&sbi->cp_wait, &wait); | |
661 | } | |
662 | ||
127e670a JK |
663 | static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount) |
664 | { | |
665 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
666 | nid_t last_nid = 0; | |
667 | block_t start_blk; | |
668 | struct page *cp_page; | |
669 | unsigned int data_sum_blocks, orphan_blocks; | |
7e586fa0 | 670 | __u32 crc32 = 0; |
127e670a | 671 | void *kaddr; |
127e670a JK |
672 | int i; |
673 | ||
674 | /* Flush all the NAT/SIT pages */ | |
675 | while (get_pages(sbi, F2FS_DIRTY_META)) | |
676 | sync_meta_pages(sbi, META, LONG_MAX); | |
677 | ||
678 | next_free_nid(sbi, &last_nid); | |
679 | ||
680 | /* | |
681 | * modify checkpoint | |
682 | * version number is already updated | |
683 | */ | |
684 | ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi)); | |
685 | ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi)); | |
686 | ckpt->free_segment_count = cpu_to_le32(free_segments(sbi)); | |
687 | for (i = 0; i < 3; i++) { | |
688 | ckpt->cur_node_segno[i] = | |
689 | cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE)); | |
690 | ckpt->cur_node_blkoff[i] = | |
691 | cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE)); | |
692 | ckpt->alloc_type[i + CURSEG_HOT_NODE] = | |
693 | curseg_alloc_type(sbi, i + CURSEG_HOT_NODE); | |
694 | } | |
695 | for (i = 0; i < 3; i++) { | |
696 | ckpt->cur_data_segno[i] = | |
697 | cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA)); | |
698 | ckpt->cur_data_blkoff[i] = | |
699 | cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA)); | |
700 | ckpt->alloc_type[i + CURSEG_HOT_DATA] = | |
701 | curseg_alloc_type(sbi, i + CURSEG_HOT_DATA); | |
702 | } | |
703 | ||
704 | ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi)); | |
705 | ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi)); | |
706 | ckpt->next_free_nid = cpu_to_le32(last_nid); | |
707 | ||
708 | /* 2 cp + n data seg summary + orphan inode blocks */ | |
709 | data_sum_blocks = npages_for_summary_flush(sbi); | |
710 | if (data_sum_blocks < 3) | |
25ca923b | 711 | set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); |
127e670a | 712 | else |
25ca923b | 713 | clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); |
127e670a JK |
714 | |
715 | orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1) | |
716 | / F2FS_ORPHANS_PER_BLOCK; | |
25ca923b | 717 | ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks); |
127e670a JK |
718 | |
719 | if (is_umount) { | |
25ca923b JK |
720 | set_ckpt_flags(ckpt, CP_UMOUNT_FLAG); |
721 | ckpt->cp_pack_total_block_count = cpu_to_le32(2 + | |
722 | data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE); | |
127e670a | 723 | } else { |
25ca923b JK |
724 | clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG); |
725 | ckpt->cp_pack_total_block_count = cpu_to_le32(2 + | |
726 | data_sum_blocks + orphan_blocks); | |
127e670a JK |
727 | } |
728 | ||
729 | if (sbi->n_orphans) | |
25ca923b | 730 | set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); |
127e670a | 731 | else |
25ca923b | 732 | clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); |
127e670a JK |
733 | |
734 | /* update SIT/NAT bitmap */ | |
735 | get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP)); | |
736 | get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP)); | |
737 | ||
738 | crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset)); | |
7e586fa0 JK |
739 | *((__le32 *)((unsigned char *)ckpt + |
740 | le32_to_cpu(ckpt->checksum_offset))) | |
127e670a JK |
741 | = cpu_to_le32(crc32); |
742 | ||
743 | start_blk = __start_cp_addr(sbi); | |
744 | ||
745 | /* write out checkpoint buffer at block 0 */ | |
746 | cp_page = grab_meta_page(sbi, start_blk++); | |
747 | kaddr = page_address(cp_page); | |
748 | memcpy(kaddr, ckpt, (1 << sbi->log_blocksize)); | |
749 | set_page_dirty(cp_page); | |
750 | f2fs_put_page(cp_page, 1); | |
751 | ||
752 | if (sbi->n_orphans) { | |
753 | write_orphan_inodes(sbi, start_blk); | |
754 | start_blk += orphan_blocks; | |
755 | } | |
756 | ||
757 | write_data_summaries(sbi, start_blk); | |
758 | start_blk += data_sum_blocks; | |
759 | if (is_umount) { | |
760 | write_node_summaries(sbi, start_blk); | |
761 | start_blk += NR_CURSEG_NODE_TYPE; | |
762 | } | |
763 | ||
764 | /* writeout checkpoint block */ | |
765 | cp_page = grab_meta_page(sbi, start_blk); | |
766 | kaddr = page_address(cp_page); | |
767 | memcpy(kaddr, ckpt, (1 << sbi->log_blocksize)); | |
768 | set_page_dirty(cp_page); | |
769 | f2fs_put_page(cp_page, 1); | |
770 | ||
771 | /* wait for previous submitted node/meta pages writeback */ | |
fb51b5ef | 772 | wait_on_all_pages_writeback(sbi); |
127e670a | 773 | |
4ef51a8f | 774 | filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX); |
9df27d98 | 775 | filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX); |
127e670a JK |
776 | |
777 | /* update user_block_counts */ | |
778 | sbi->last_valid_block_count = sbi->total_valid_block_count; | |
779 | sbi->alloc_valid_block_count = 0; | |
780 | ||
781 | /* Here, we only have one bio having CP pack */ | |
577e3495 | 782 | sync_meta_pages(sbi, META_FLUSH, LONG_MAX); |
127e670a | 783 | |
6bacf52f | 784 | if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) { |
577e3495 JK |
785 | clear_prefree_segments(sbi); |
786 | F2FS_RESET_SB_DIRT(sbi); | |
787 | } | |
127e670a JK |
788 | } |
789 | ||
0a8165d7 | 790 | /* |
127e670a JK |
791 | * We guarantee that this checkpoint procedure should not fail. |
792 | */ | |
43727527 | 793 | void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount) |
127e670a JK |
794 | { |
795 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
796 | unsigned long long ckpt_ver; | |
797 | ||
2af4bd6c NJ |
798 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops"); |
799 | ||
43727527 JK |
800 | mutex_lock(&sbi->cp_mutex); |
801 | block_operations(sbi); | |
127e670a | 802 | |
2af4bd6c NJ |
803 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops"); |
804 | ||
458e6197 JK |
805 | f2fs_submit_merged_bio(sbi, DATA, WRITE); |
806 | f2fs_submit_merged_bio(sbi, NODE, WRITE); | |
807 | f2fs_submit_merged_bio(sbi, META, WRITE); | |
127e670a JK |
808 | |
809 | /* | |
810 | * update checkpoint pack index | |
811 | * Increase the version number so that | |
812 | * SIT entries and seg summaries are written at correct place | |
813 | */ | |
d71b5564 | 814 | ckpt_ver = cur_cp_version(ckpt); |
127e670a JK |
815 | ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver); |
816 | ||
817 | /* write cached NAT/SIT entries to NAT/SIT area */ | |
818 | flush_nat_entries(sbi); | |
819 | flush_sit_entries(sbi); | |
820 | ||
127e670a JK |
821 | /* unlock all the fs_lock[] in do_checkpoint() */ |
822 | do_checkpoint(sbi, is_umount); | |
823 | ||
824 | unblock_operations(sbi); | |
825 | mutex_unlock(&sbi->cp_mutex); | |
2af4bd6c NJ |
826 | |
827 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint"); | |
127e670a JK |
828 | } |
829 | ||
830 | void init_orphan_info(struct f2fs_sb_info *sbi) | |
831 | { | |
17b692f6 | 832 | spin_lock_init(&sbi->orphan_inode_lock); |
127e670a JK |
833 | INIT_LIST_HEAD(&sbi->orphan_inode_list); |
834 | sbi->n_orphans = 0; | |
0d47c1ad GZ |
835 | /* |
836 | * considering 512 blocks in a segment 8 blocks are needed for cp | |
837 | * and log segment summaries. Remaining blocks are used to keep | |
838 | * orphan entries with the limitation one reserved segment | |
839 | * for cp pack we can have max 1020*504 orphan entries | |
840 | */ | |
841 | sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE) | |
842 | * F2FS_ORPHANS_PER_BLOCK; | |
127e670a JK |
843 | } |
844 | ||
6e6093a8 | 845 | int __init create_checkpoint_caches(void) |
127e670a JK |
846 | { |
847 | orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry", | |
848 | sizeof(struct orphan_inode_entry), NULL); | |
6bacf52f | 849 | if (!orphan_entry_slab) |
127e670a JK |
850 | return -ENOMEM; |
851 | inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry", | |
852 | sizeof(struct dir_inode_entry), NULL); | |
6bacf52f | 853 | if (!inode_entry_slab) { |
127e670a JK |
854 | kmem_cache_destroy(orphan_entry_slab); |
855 | return -ENOMEM; | |
856 | } | |
857 | return 0; | |
858 | } | |
859 | ||
860 | void destroy_checkpoint_caches(void) | |
861 | { | |
862 | kmem_cache_destroy(orphan_entry_slab); | |
863 | kmem_cache_destroy(inode_entry_slab); | |
864 | } |