]>
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: | |
bde44686 | 36 | page = grab_cache_page(mapping, index); |
127e670a JK |
37 | if (!page) { |
38 | cond_resched(); | |
39 | goto repeat; | |
40 | } | |
bde44686 | 41 | f2fs_wait_on_page_writeback(page, META); |
127e670a JK |
42 | SetPageUptodate(page); |
43 | return page; | |
44 | } | |
45 | ||
0a8165d7 | 46 | /* |
127e670a JK |
47 | * We guarantee no failure on the returned page. |
48 | */ | |
49 | struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index) | |
50 | { | |
9df27d98 | 51 | struct address_space *mapping = META_MAPPING(sbi); |
127e670a JK |
52 | struct page *page; |
53 | repeat: | |
54 | page = grab_cache_page(mapping, index); | |
55 | if (!page) { | |
56 | cond_resched(); | |
57 | goto repeat; | |
58 | } | |
393ff91f JK |
59 | if (PageUptodate(page)) |
60 | goto out; | |
61 | ||
93dfe2ac JK |
62 | if (f2fs_submit_page_bio(sbi, page, index, |
63 | READ_SYNC | REQ_META | REQ_PRIO)) | |
127e670a | 64 | goto repeat; |
127e670a | 65 | |
393ff91f | 66 | lock_page(page); |
6bacf52f | 67 | if (unlikely(page->mapping != mapping)) { |
afcb7ca0 JK |
68 | f2fs_put_page(page, 1); |
69 | goto repeat; | |
70 | } | |
393ff91f JK |
71 | out: |
72 | mark_page_accessed(page); | |
127e670a JK |
73 | return page; |
74 | } | |
75 | ||
b49ad51e | 76 | static inline int get_max_meta_blks(struct f2fs_sb_info *sbi, int type) |
662befda CY |
77 | { |
78 | switch (type) { | |
79 | case META_NAT: | |
80 | return NM_I(sbi)->max_nid / NAT_ENTRY_PER_BLOCK; | |
81 | case META_SIT: | |
82 | return SIT_BLK_CNT(sbi); | |
81c1a0f1 | 83 | case META_SSA: |
662befda CY |
84 | case META_CP: |
85 | return 0; | |
86 | default: | |
87 | BUG(); | |
88 | } | |
89 | } | |
90 | ||
91 | /* | |
81c1a0f1 | 92 | * Readahead CP/NAT/SIT/SSA pages |
662befda CY |
93 | */ |
94 | int ra_meta_pages(struct f2fs_sb_info *sbi, int start, int nrpages, int type) | |
95 | { | |
96 | block_t prev_blk_addr = 0; | |
97 | struct page *page; | |
98 | int blkno = start; | |
99 | int max_blks = get_max_meta_blks(sbi, type); | |
100 | ||
101 | struct f2fs_io_info fio = { | |
102 | .type = META, | |
103 | .rw = READ_SYNC | REQ_META | REQ_PRIO | |
104 | }; | |
105 | ||
106 | for (; nrpages-- > 0; blkno++) { | |
107 | block_t blk_addr; | |
108 | ||
109 | switch (type) { | |
110 | case META_NAT: | |
111 | /* get nat block addr */ | |
112 | if (unlikely(blkno >= max_blks)) | |
113 | blkno = 0; | |
114 | blk_addr = current_nat_addr(sbi, | |
115 | blkno * NAT_ENTRY_PER_BLOCK); | |
116 | break; | |
117 | case META_SIT: | |
118 | /* get sit block addr */ | |
119 | if (unlikely(blkno >= max_blks)) | |
120 | goto out; | |
121 | blk_addr = current_sit_addr(sbi, | |
122 | blkno * SIT_ENTRY_PER_BLOCK); | |
123 | if (blkno != start && prev_blk_addr + 1 != blk_addr) | |
124 | goto out; | |
125 | prev_blk_addr = blk_addr; | |
126 | break; | |
81c1a0f1 | 127 | case META_SSA: |
662befda | 128 | case META_CP: |
81c1a0f1 | 129 | /* get ssa/cp block addr */ |
662befda CY |
130 | blk_addr = blkno; |
131 | break; | |
132 | default: | |
133 | BUG(); | |
134 | } | |
135 | ||
136 | page = grab_cache_page(META_MAPPING(sbi), blk_addr); | |
137 | if (!page) | |
138 | continue; | |
139 | if (PageUptodate(page)) { | |
140 | mark_page_accessed(page); | |
141 | f2fs_put_page(page, 1); | |
142 | continue; | |
143 | } | |
144 | ||
145 | f2fs_submit_page_mbio(sbi, page, blk_addr, &fio); | |
146 | mark_page_accessed(page); | |
147 | f2fs_put_page(page, 0); | |
148 | } | |
149 | out: | |
150 | f2fs_submit_merged_bio(sbi, META, READ); | |
151 | return blkno - start; | |
152 | } | |
153 | ||
127e670a JK |
154 | static int f2fs_write_meta_page(struct page *page, |
155 | struct writeback_control *wbc) | |
156 | { | |
157 | struct inode *inode = page->mapping->host; | |
158 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
127e670a | 159 | |
ecda0de3 CY |
160 | trace_f2fs_writepage(page, META); |
161 | ||
203681f6 | 162 | if (unlikely(sbi->por_doing)) |
cfb271d4 | 163 | goto redirty_out; |
cfb271d4 CY |
164 | if (wbc->for_reclaim) |
165 | goto redirty_out; | |
127e670a | 166 | |
203681f6 JK |
167 | /* Should not write any meta pages, if any IO error was occurred */ |
168 | if (unlikely(is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG))) | |
169 | goto no_write; | |
127e670a | 170 | |
3cb5ad15 | 171 | f2fs_wait_on_page_writeback(page, META); |
577e3495 | 172 | write_meta_page(sbi, page); |
203681f6 | 173 | no_write: |
577e3495 JK |
174 | dec_page_count(sbi, F2FS_DIRTY_META); |
175 | unlock_page(page); | |
176 | return 0; | |
cfb271d4 CY |
177 | |
178 | redirty_out: | |
76f60268 | 179 | redirty_page_for_writepage(wbc, page); |
cfb271d4 | 180 | return AOP_WRITEPAGE_ACTIVATE; |
127e670a JK |
181 | } |
182 | ||
183 | static int f2fs_write_meta_pages(struct address_space *mapping, | |
184 | struct writeback_control *wbc) | |
185 | { | |
186 | struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); | |
50c8cdb3 | 187 | long diff, written; |
127e670a | 188 | |
5459aa97 | 189 | /* collect a number of dirty meta pages and write together */ |
50c8cdb3 JK |
190 | if (wbc->for_kupdate || |
191 | get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META)) | |
d3baf95d | 192 | goto skip_write; |
127e670a JK |
193 | |
194 | /* if mounting is failed, skip writing node pages */ | |
195 | mutex_lock(&sbi->cp_mutex); | |
50c8cdb3 JK |
196 | diff = nr_pages_to_write(sbi, META, wbc); |
197 | written = sync_meta_pages(sbi, META, wbc->nr_to_write); | |
127e670a | 198 | mutex_unlock(&sbi->cp_mutex); |
50c8cdb3 | 199 | wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff); |
127e670a | 200 | return 0; |
d3baf95d JK |
201 | |
202 | skip_write: | |
203 | wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META); | |
204 | return 0; | |
127e670a JK |
205 | } |
206 | ||
207 | long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type, | |
208 | long nr_to_write) | |
209 | { | |
9df27d98 | 210 | struct address_space *mapping = META_MAPPING(sbi); |
127e670a JK |
211 | pgoff_t index = 0, end = LONG_MAX; |
212 | struct pagevec pvec; | |
213 | long nwritten = 0; | |
214 | struct writeback_control wbc = { | |
215 | .for_reclaim = 0, | |
216 | }; | |
217 | ||
218 | pagevec_init(&pvec, 0); | |
219 | ||
220 | while (index <= end) { | |
221 | int i, nr_pages; | |
222 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
223 | PAGECACHE_TAG_DIRTY, | |
224 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); | |
cfb271d4 | 225 | if (unlikely(nr_pages == 0)) |
127e670a JK |
226 | break; |
227 | ||
228 | for (i = 0; i < nr_pages; i++) { | |
229 | struct page *page = pvec.pages[i]; | |
203681f6 | 230 | |
127e670a | 231 | lock_page(page); |
203681f6 JK |
232 | |
233 | if (unlikely(page->mapping != mapping)) { | |
234 | continue_unlock: | |
235 | unlock_page(page); | |
236 | continue; | |
237 | } | |
238 | if (!PageDirty(page)) { | |
239 | /* someone wrote it for us */ | |
240 | goto continue_unlock; | |
241 | } | |
242 | ||
243 | if (!clear_page_dirty_for_io(page)) | |
244 | goto continue_unlock; | |
245 | ||
577e3495 JK |
246 | if (f2fs_write_meta_page(page, &wbc)) { |
247 | unlock_page(page); | |
248 | break; | |
249 | } | |
cfb271d4 CY |
250 | nwritten++; |
251 | if (unlikely(nwritten >= nr_to_write)) | |
127e670a JK |
252 | break; |
253 | } | |
254 | pagevec_release(&pvec); | |
255 | cond_resched(); | |
256 | } | |
257 | ||
258 | if (nwritten) | |
458e6197 | 259 | f2fs_submit_merged_bio(sbi, type, WRITE); |
127e670a JK |
260 | |
261 | return nwritten; | |
262 | } | |
263 | ||
264 | static int f2fs_set_meta_page_dirty(struct page *page) | |
265 | { | |
266 | struct address_space *mapping = page->mapping; | |
267 | struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); | |
268 | ||
26c6b887 JK |
269 | trace_f2fs_set_page_dirty(page, META); |
270 | ||
127e670a JK |
271 | SetPageUptodate(page); |
272 | if (!PageDirty(page)) { | |
273 | __set_page_dirty_nobuffers(page); | |
274 | inc_page_count(sbi, F2FS_DIRTY_META); | |
127e670a JK |
275 | return 1; |
276 | } | |
277 | return 0; | |
278 | } | |
279 | ||
280 | const struct address_space_operations f2fs_meta_aops = { | |
281 | .writepage = f2fs_write_meta_page, | |
282 | .writepages = f2fs_write_meta_pages, | |
283 | .set_page_dirty = f2fs_set_meta_page_dirty, | |
284 | }; | |
285 | ||
cbd56e7d | 286 | int acquire_orphan_inode(struct f2fs_sb_info *sbi) |
127e670a | 287 | { |
127e670a JK |
288 | int err = 0; |
289 | ||
17b692f6 | 290 | spin_lock(&sbi->orphan_inode_lock); |
0d47c1ad | 291 | if (unlikely(sbi->n_orphans >= sbi->max_orphans)) |
127e670a | 292 | err = -ENOSPC; |
cbd56e7d JK |
293 | else |
294 | sbi->n_orphans++; | |
17b692f6 | 295 | spin_unlock(&sbi->orphan_inode_lock); |
0d47c1ad | 296 | |
127e670a JK |
297 | return err; |
298 | } | |
299 | ||
cbd56e7d JK |
300 | void release_orphan_inode(struct f2fs_sb_info *sbi) |
301 | { | |
17b692f6 | 302 | spin_lock(&sbi->orphan_inode_lock); |
5d56b671 | 303 | f2fs_bug_on(sbi->n_orphans == 0); |
cbd56e7d | 304 | sbi->n_orphans--; |
17b692f6 | 305 | spin_unlock(&sbi->orphan_inode_lock); |
cbd56e7d JK |
306 | } |
307 | ||
127e670a JK |
308 | void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) |
309 | { | |
2d7b822a CY |
310 | struct list_head *head; |
311 | struct orphan_inode_entry *new, *orphan; | |
127e670a | 312 | |
c1ef3725 GZ |
313 | new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC); |
314 | new->ino = ino; | |
315 | ||
17b692f6 | 316 | spin_lock(&sbi->orphan_inode_lock); |
127e670a | 317 | head = &sbi->orphan_inode_list; |
2d7b822a | 318 | list_for_each_entry(orphan, head, list) { |
c1ef3725 | 319 | if (orphan->ino == ino) { |
17b692f6 | 320 | spin_unlock(&sbi->orphan_inode_lock); |
c1ef3725 GZ |
321 | kmem_cache_free(orphan_entry_slab, new); |
322 | return; | |
323 | } | |
324 | ||
127e670a JK |
325 | if (orphan->ino > ino) |
326 | break; | |
127e670a | 327 | } |
7bd59381 | 328 | |
2d7b822a CY |
329 | /* add new orphan entry into list which is sorted by inode number */ |
330 | list_add_tail(&new->list, &orphan->list); | |
17b692f6 | 331 | spin_unlock(&sbi->orphan_inode_lock); |
127e670a JK |
332 | } |
333 | ||
334 | void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
335 | { | |
60ed9a0f | 336 | struct list_head *head; |
127e670a JK |
337 | struct orphan_inode_entry *orphan; |
338 | ||
17b692f6 | 339 | spin_lock(&sbi->orphan_inode_lock); |
127e670a | 340 | head = &sbi->orphan_inode_list; |
60ed9a0f | 341 | list_for_each_entry(orphan, head, list) { |
127e670a JK |
342 | if (orphan->ino == ino) { |
343 | list_del(&orphan->list); | |
5d56b671 | 344 | f2fs_bug_on(sbi->n_orphans == 0); |
127e670a | 345 | sbi->n_orphans--; |
cf0ee0f0 CY |
346 | spin_unlock(&sbi->orphan_inode_lock); |
347 | kmem_cache_free(orphan_entry_slab, orphan); | |
348 | return; | |
127e670a JK |
349 | } |
350 | } | |
17b692f6 | 351 | spin_unlock(&sbi->orphan_inode_lock); |
127e670a JK |
352 | } |
353 | ||
354 | static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
355 | { | |
356 | struct inode *inode = f2fs_iget(sbi->sb, ino); | |
5d56b671 | 357 | f2fs_bug_on(IS_ERR(inode)); |
127e670a JK |
358 | clear_nlink(inode); |
359 | ||
360 | /* truncate all the data during iput */ | |
361 | iput(inode); | |
362 | } | |
363 | ||
8f99a946 | 364 | void recover_orphan_inodes(struct f2fs_sb_info *sbi) |
127e670a JK |
365 | { |
366 | block_t start_blk, orphan_blkaddr, i, j; | |
367 | ||
25ca923b | 368 | if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG)) |
8f99a946 | 369 | return; |
127e670a | 370 | |
aabe5136 | 371 | sbi->por_doing = true; |
127e670a JK |
372 | start_blk = __start_cp_addr(sbi) + 1; |
373 | orphan_blkaddr = __start_sum_addr(sbi) - 1; | |
374 | ||
662befda CY |
375 | ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP); |
376 | ||
127e670a JK |
377 | for (i = 0; i < orphan_blkaddr; i++) { |
378 | struct page *page = get_meta_page(sbi, start_blk + i); | |
379 | struct f2fs_orphan_block *orphan_blk; | |
380 | ||
381 | orphan_blk = (struct f2fs_orphan_block *)page_address(page); | |
382 | for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) { | |
383 | nid_t ino = le32_to_cpu(orphan_blk->ino[j]); | |
384 | recover_orphan_inode(sbi, ino); | |
385 | } | |
386 | f2fs_put_page(page, 1); | |
387 | } | |
388 | /* clear Orphan Flag */ | |
25ca923b | 389 | clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG); |
aabe5136 | 390 | sbi->por_doing = false; |
8f99a946 | 391 | return; |
127e670a JK |
392 | } |
393 | ||
394 | static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk) | |
395 | { | |
502c6e0b | 396 | struct list_head *head; |
127e670a | 397 | struct f2fs_orphan_block *orphan_blk = NULL; |
127e670a | 398 | unsigned int nentries = 0; |
4531929e GZ |
399 | unsigned short index; |
400 | unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans + | |
401 | (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK); | |
402 | struct page *page = NULL; | |
502c6e0b | 403 | struct orphan_inode_entry *orphan = NULL; |
127e670a | 404 | |
4531929e | 405 | for (index = 0; index < orphan_blocks; index++) |
63f5384c | 406 | grab_meta_page(sbi, start_blk + index); |
127e670a | 407 | |
4531929e | 408 | index = 1; |
17b692f6 | 409 | spin_lock(&sbi->orphan_inode_lock); |
127e670a JK |
410 | head = &sbi->orphan_inode_list; |
411 | ||
412 | /* loop for each orphan inode entry and write them in Jornal block */ | |
502c6e0b GZ |
413 | list_for_each_entry(orphan, head, list) { |
414 | if (!page) { | |
63f5384c GZ |
415 | page = find_get_page(META_MAPPING(sbi), start_blk++); |
416 | f2fs_bug_on(!page); | |
502c6e0b GZ |
417 | orphan_blk = |
418 | (struct f2fs_orphan_block *)page_address(page); | |
419 | memset(orphan_blk, 0, sizeof(*orphan_blk)); | |
63f5384c | 420 | f2fs_put_page(page, 0); |
502c6e0b | 421 | } |
127e670a | 422 | |
36795567 | 423 | orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino); |
127e670a | 424 | |
36795567 | 425 | if (nentries == F2FS_ORPHANS_PER_BLOCK) { |
127e670a JK |
426 | /* |
427 | * an orphan block is full of 1020 entries, | |
428 | * then we need to flush current orphan blocks | |
429 | * and bring another one in memory | |
430 | */ | |
431 | orphan_blk->blk_addr = cpu_to_le16(index); | |
432 | orphan_blk->blk_count = cpu_to_le16(orphan_blocks); | |
433 | orphan_blk->entry_count = cpu_to_le32(nentries); | |
434 | set_page_dirty(page); | |
435 | f2fs_put_page(page, 1); | |
436 | index++; | |
127e670a JK |
437 | nentries = 0; |
438 | page = NULL; | |
439 | } | |
502c6e0b | 440 | } |
127e670a | 441 | |
502c6e0b GZ |
442 | if (page) { |
443 | orphan_blk->blk_addr = cpu_to_le16(index); | |
444 | orphan_blk->blk_count = cpu_to_le16(orphan_blocks); | |
445 | orphan_blk->entry_count = cpu_to_le32(nentries); | |
446 | set_page_dirty(page); | |
447 | f2fs_put_page(page, 1); | |
127e670a | 448 | } |
502c6e0b | 449 | |
17b692f6 | 450 | spin_unlock(&sbi->orphan_inode_lock); |
127e670a JK |
451 | } |
452 | ||
453 | static struct page *validate_checkpoint(struct f2fs_sb_info *sbi, | |
454 | block_t cp_addr, unsigned long long *version) | |
455 | { | |
456 | struct page *cp_page_1, *cp_page_2 = NULL; | |
457 | unsigned long blk_size = sbi->blocksize; | |
458 | struct f2fs_checkpoint *cp_block; | |
459 | unsigned long long cur_version = 0, pre_version = 0; | |
127e670a | 460 | size_t crc_offset; |
7e586fa0 | 461 | __u32 crc = 0; |
127e670a JK |
462 | |
463 | /* Read the 1st cp block in this CP pack */ | |
464 | cp_page_1 = get_meta_page(sbi, cp_addr); | |
465 | ||
466 | /* get the version number */ | |
467 | cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1); | |
468 | crc_offset = le32_to_cpu(cp_block->checksum_offset); | |
469 | if (crc_offset >= blk_size) | |
470 | goto invalid_cp1; | |
471 | ||
7e586fa0 | 472 | crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset))); |
127e670a JK |
473 | if (!f2fs_crc_valid(crc, cp_block, crc_offset)) |
474 | goto invalid_cp1; | |
475 | ||
d71b5564 | 476 | pre_version = cur_cp_version(cp_block); |
127e670a JK |
477 | |
478 | /* Read the 2nd cp block in this CP pack */ | |
25ca923b | 479 | cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1; |
127e670a JK |
480 | cp_page_2 = get_meta_page(sbi, cp_addr); |
481 | ||
482 | cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2); | |
483 | crc_offset = le32_to_cpu(cp_block->checksum_offset); | |
484 | if (crc_offset >= blk_size) | |
485 | goto invalid_cp2; | |
486 | ||
7e586fa0 | 487 | crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset))); |
127e670a JK |
488 | if (!f2fs_crc_valid(crc, cp_block, crc_offset)) |
489 | goto invalid_cp2; | |
490 | ||
d71b5564 | 491 | cur_version = cur_cp_version(cp_block); |
127e670a JK |
492 | |
493 | if (cur_version == pre_version) { | |
494 | *version = cur_version; | |
495 | f2fs_put_page(cp_page_2, 1); | |
496 | return cp_page_1; | |
497 | } | |
498 | invalid_cp2: | |
499 | f2fs_put_page(cp_page_2, 1); | |
500 | invalid_cp1: | |
501 | f2fs_put_page(cp_page_1, 1); | |
502 | return NULL; | |
503 | } | |
504 | ||
505 | int get_valid_checkpoint(struct f2fs_sb_info *sbi) | |
506 | { | |
507 | struct f2fs_checkpoint *cp_block; | |
508 | struct f2fs_super_block *fsb = sbi->raw_super; | |
509 | struct page *cp1, *cp2, *cur_page; | |
510 | unsigned long blk_size = sbi->blocksize; | |
511 | unsigned long long cp1_version = 0, cp2_version = 0; | |
512 | unsigned long long cp_start_blk_no; | |
513 | ||
514 | sbi->ckpt = kzalloc(blk_size, GFP_KERNEL); | |
515 | if (!sbi->ckpt) | |
516 | return -ENOMEM; | |
517 | /* | |
518 | * Finding out valid cp block involves read both | |
519 | * sets( cp pack1 and cp pack 2) | |
520 | */ | |
521 | cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr); | |
522 | cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version); | |
523 | ||
524 | /* The second checkpoint pack should start at the next segment */ | |
f9a4e6df JK |
525 | cp_start_blk_no += ((unsigned long long)1) << |
526 | le32_to_cpu(fsb->log_blocks_per_seg); | |
127e670a JK |
527 | cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version); |
528 | ||
529 | if (cp1 && cp2) { | |
530 | if (ver_after(cp2_version, cp1_version)) | |
531 | cur_page = cp2; | |
532 | else | |
533 | cur_page = cp1; | |
534 | } else if (cp1) { | |
535 | cur_page = cp1; | |
536 | } else if (cp2) { | |
537 | cur_page = cp2; | |
538 | } else { | |
539 | goto fail_no_cp; | |
540 | } | |
541 | ||
542 | cp_block = (struct f2fs_checkpoint *)page_address(cur_page); | |
543 | memcpy(sbi->ckpt, cp_block, blk_size); | |
544 | ||
545 | f2fs_put_page(cp1, 1); | |
546 | f2fs_put_page(cp2, 1); | |
547 | return 0; | |
548 | ||
549 | fail_no_cp: | |
550 | kfree(sbi->ckpt); | |
551 | return -EINVAL; | |
552 | } | |
553 | ||
5deb8267 | 554 | static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new) |
127e670a JK |
555 | { |
556 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
127e670a | 557 | |
ed57c27f JK |
558 | if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) |
559 | return -EEXIST; | |
2d7b822a | 560 | |
ed57c27f JK |
561 | set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR); |
562 | F2FS_I(inode)->dirty_dir = new; | |
563 | list_add_tail(&new->list, &sbi->dir_inode_list); | |
dcdfff65 | 564 | stat_inc_dirty_dir(sbi); |
5deb8267 JK |
565 | return 0; |
566 | } | |
567 | ||
568 | void set_dirty_dir_page(struct inode *inode, struct page *page) | |
569 | { | |
570 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
571 | struct dir_inode_entry *new; | |
cf0ee0f0 | 572 | int ret = 0; |
5deb8267 | 573 | |
127e670a JK |
574 | if (!S_ISDIR(inode->i_mode)) |
575 | return; | |
7bd59381 GZ |
576 | |
577 | new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); | |
127e670a JK |
578 | new->inode = inode; |
579 | INIT_LIST_HEAD(&new->list); | |
580 | ||
581 | spin_lock(&sbi->dir_inode_lock); | |
cf0ee0f0 | 582 | ret = __add_dirty_inode(inode, new); |
127e670a JK |
583 | inode_inc_dirty_dents(inode); |
584 | SetPagePrivate(page); | |
5deb8267 | 585 | spin_unlock(&sbi->dir_inode_lock); |
cf0ee0f0 CY |
586 | |
587 | if (ret) | |
588 | kmem_cache_free(inode_entry_slab, new); | |
5deb8267 JK |
589 | } |
590 | ||
591 | void add_dirty_dir_inode(struct inode *inode) | |
592 | { | |
593 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
7bd59381 GZ |
594 | struct dir_inode_entry *new = |
595 | f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); | |
cf0ee0f0 | 596 | int ret = 0; |
7bd59381 | 597 | |
5deb8267 JK |
598 | new->inode = inode; |
599 | INIT_LIST_HEAD(&new->list); | |
127e670a | 600 | |
5deb8267 | 601 | spin_lock(&sbi->dir_inode_lock); |
cf0ee0f0 | 602 | ret = __add_dirty_inode(inode, new); |
127e670a | 603 | spin_unlock(&sbi->dir_inode_lock); |
cf0ee0f0 CY |
604 | |
605 | if (ret) | |
606 | kmem_cache_free(inode_entry_slab, new); | |
127e670a JK |
607 | } |
608 | ||
609 | void remove_dirty_dir_inode(struct inode *inode) | |
610 | { | |
611 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
2d7b822a | 612 | struct dir_inode_entry *entry; |
127e670a JK |
613 | |
614 | if (!S_ISDIR(inode->i_mode)) | |
615 | return; | |
616 | ||
617 | spin_lock(&sbi->dir_inode_lock); | |
ed57c27f JK |
618 | if (get_dirty_dents(inode) || |
619 | !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) { | |
3b10b1fd JK |
620 | spin_unlock(&sbi->dir_inode_lock); |
621 | return; | |
622 | } | |
127e670a | 623 | |
ed57c27f JK |
624 | entry = F2FS_I(inode)->dirty_dir; |
625 | list_del(&entry->list); | |
626 | F2FS_I(inode)->dirty_dir = NULL; | |
627 | clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR); | |
628 | stat_dec_dirty_dir(sbi); | |
127e670a | 629 | spin_unlock(&sbi->dir_inode_lock); |
ed57c27f | 630 | kmem_cache_free(inode_entry_slab, entry); |
74d0b917 JK |
631 | |
632 | /* Only from the recovery routine */ | |
afc3eda2 JK |
633 | if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) { |
634 | clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT); | |
74d0b917 | 635 | iput(inode); |
afc3eda2 | 636 | } |
74d0b917 JK |
637 | } |
638 | ||
127e670a JK |
639 | void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi) |
640 | { | |
ce3b7d80 | 641 | struct list_head *head; |
127e670a JK |
642 | struct dir_inode_entry *entry; |
643 | struct inode *inode; | |
644 | retry: | |
645 | spin_lock(&sbi->dir_inode_lock); | |
ce3b7d80 GZ |
646 | |
647 | head = &sbi->dir_inode_list; | |
127e670a JK |
648 | if (list_empty(head)) { |
649 | spin_unlock(&sbi->dir_inode_lock); | |
650 | return; | |
651 | } | |
652 | entry = list_entry(head->next, struct dir_inode_entry, list); | |
653 | inode = igrab(entry->inode); | |
654 | spin_unlock(&sbi->dir_inode_lock); | |
655 | if (inode) { | |
87d6f890 | 656 | filemap_fdatawrite(inode->i_mapping); |
127e670a JK |
657 | iput(inode); |
658 | } else { | |
659 | /* | |
660 | * We should submit bio, since it exists several | |
661 | * wribacking dentry pages in the freeing inode. | |
662 | */ | |
458e6197 | 663 | f2fs_submit_merged_bio(sbi, DATA, WRITE); |
127e670a JK |
664 | } |
665 | goto retry; | |
666 | } | |
667 | ||
0a8165d7 | 668 | /* |
127e670a JK |
669 | * Freeze all the FS-operations for checkpoint. |
670 | */ | |
43727527 | 671 | static void block_operations(struct f2fs_sb_info *sbi) |
127e670a | 672 | { |
127e670a JK |
673 | struct writeback_control wbc = { |
674 | .sync_mode = WB_SYNC_ALL, | |
675 | .nr_to_write = LONG_MAX, | |
676 | .for_reclaim = 0, | |
677 | }; | |
c718379b JK |
678 | struct blk_plug plug; |
679 | ||
680 | blk_start_plug(&plug); | |
681 | ||
39936837 | 682 | retry_flush_dents: |
e479556b | 683 | f2fs_lock_all(sbi); |
127e670a | 684 | /* write all the dirty dentry pages */ |
127e670a | 685 | if (get_pages(sbi, F2FS_DIRTY_DENTS)) { |
e479556b | 686 | f2fs_unlock_all(sbi); |
39936837 JK |
687 | sync_dirty_dir_inodes(sbi); |
688 | goto retry_flush_dents; | |
127e670a JK |
689 | } |
690 | ||
127e670a JK |
691 | /* |
692 | * POR: we should ensure that there is no dirty node pages | |
693 | * until finishing nat/sit flush. | |
694 | */ | |
39936837 JK |
695 | retry_flush_nodes: |
696 | mutex_lock(&sbi->node_write); | |
127e670a JK |
697 | |
698 | if (get_pages(sbi, F2FS_DIRTY_NODES)) { | |
39936837 JK |
699 | mutex_unlock(&sbi->node_write); |
700 | sync_node_pages(sbi, 0, &wbc); | |
701 | goto retry_flush_nodes; | |
127e670a | 702 | } |
c718379b | 703 | blk_finish_plug(&plug); |
127e670a JK |
704 | } |
705 | ||
706 | static void unblock_operations(struct f2fs_sb_info *sbi) | |
707 | { | |
39936837 | 708 | mutex_unlock(&sbi->node_write); |
e479556b | 709 | f2fs_unlock_all(sbi); |
127e670a JK |
710 | } |
711 | ||
fb51b5ef CL |
712 | static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi) |
713 | { | |
714 | DEFINE_WAIT(wait); | |
715 | ||
716 | for (;;) { | |
717 | prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE); | |
718 | ||
719 | if (!get_pages(sbi, F2FS_WRITEBACK)) | |
720 | break; | |
721 | ||
722 | io_schedule(); | |
723 | } | |
724 | finish_wait(&sbi->cp_wait, &wait); | |
725 | } | |
726 | ||
127e670a JK |
727 | static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount) |
728 | { | |
729 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
730 | nid_t last_nid = 0; | |
731 | block_t start_blk; | |
732 | struct page *cp_page; | |
733 | unsigned int data_sum_blocks, orphan_blocks; | |
7e586fa0 | 734 | __u32 crc32 = 0; |
127e670a | 735 | void *kaddr; |
127e670a JK |
736 | int i; |
737 | ||
1e87a78d JK |
738 | /* |
739 | * This avoids to conduct wrong roll-forward operations and uses | |
740 | * metapages, so should be called prior to sync_meta_pages below. | |
741 | */ | |
742 | discard_next_dnode(sbi); | |
743 | ||
127e670a JK |
744 | /* Flush all the NAT/SIT pages */ |
745 | while (get_pages(sbi, F2FS_DIRTY_META)) | |
746 | sync_meta_pages(sbi, META, LONG_MAX); | |
747 | ||
748 | next_free_nid(sbi, &last_nid); | |
749 | ||
750 | /* | |
751 | * modify checkpoint | |
752 | * version number is already updated | |
753 | */ | |
754 | ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi)); | |
755 | ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi)); | |
756 | ckpt->free_segment_count = cpu_to_le32(free_segments(sbi)); | |
757 | for (i = 0; i < 3; i++) { | |
758 | ckpt->cur_node_segno[i] = | |
759 | cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE)); | |
760 | ckpt->cur_node_blkoff[i] = | |
761 | cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE)); | |
762 | ckpt->alloc_type[i + CURSEG_HOT_NODE] = | |
763 | curseg_alloc_type(sbi, i + CURSEG_HOT_NODE); | |
764 | } | |
765 | for (i = 0; i < 3; i++) { | |
766 | ckpt->cur_data_segno[i] = | |
767 | cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA)); | |
768 | ckpt->cur_data_blkoff[i] = | |
769 | cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA)); | |
770 | ckpt->alloc_type[i + CURSEG_HOT_DATA] = | |
771 | curseg_alloc_type(sbi, i + CURSEG_HOT_DATA); | |
772 | } | |
773 | ||
774 | ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi)); | |
775 | ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi)); | |
776 | ckpt->next_free_nid = cpu_to_le32(last_nid); | |
777 | ||
778 | /* 2 cp + n data seg summary + orphan inode blocks */ | |
779 | data_sum_blocks = npages_for_summary_flush(sbi); | |
780 | if (data_sum_blocks < 3) | |
25ca923b | 781 | set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); |
127e670a | 782 | else |
25ca923b | 783 | clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); |
127e670a JK |
784 | |
785 | orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1) | |
786 | / F2FS_ORPHANS_PER_BLOCK; | |
25ca923b | 787 | ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks); |
127e670a JK |
788 | |
789 | if (is_umount) { | |
25ca923b JK |
790 | set_ckpt_flags(ckpt, CP_UMOUNT_FLAG); |
791 | ckpt->cp_pack_total_block_count = cpu_to_le32(2 + | |
792 | data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE); | |
127e670a | 793 | } else { |
25ca923b JK |
794 | clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG); |
795 | ckpt->cp_pack_total_block_count = cpu_to_le32(2 + | |
796 | data_sum_blocks + orphan_blocks); | |
127e670a JK |
797 | } |
798 | ||
799 | if (sbi->n_orphans) | |
25ca923b | 800 | set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); |
127e670a | 801 | else |
25ca923b | 802 | clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); |
127e670a JK |
803 | |
804 | /* update SIT/NAT bitmap */ | |
805 | get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP)); | |
806 | get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP)); | |
807 | ||
808 | crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset)); | |
7e586fa0 JK |
809 | *((__le32 *)((unsigned char *)ckpt + |
810 | le32_to_cpu(ckpt->checksum_offset))) | |
127e670a JK |
811 | = cpu_to_le32(crc32); |
812 | ||
813 | start_blk = __start_cp_addr(sbi); | |
814 | ||
815 | /* write out checkpoint buffer at block 0 */ | |
816 | cp_page = grab_meta_page(sbi, start_blk++); | |
817 | kaddr = page_address(cp_page); | |
818 | memcpy(kaddr, ckpt, (1 << sbi->log_blocksize)); | |
819 | set_page_dirty(cp_page); | |
820 | f2fs_put_page(cp_page, 1); | |
821 | ||
822 | if (sbi->n_orphans) { | |
823 | write_orphan_inodes(sbi, start_blk); | |
824 | start_blk += orphan_blocks; | |
825 | } | |
826 | ||
827 | write_data_summaries(sbi, start_blk); | |
828 | start_blk += data_sum_blocks; | |
829 | if (is_umount) { | |
830 | write_node_summaries(sbi, start_blk); | |
831 | start_blk += NR_CURSEG_NODE_TYPE; | |
832 | } | |
833 | ||
834 | /* writeout checkpoint block */ | |
835 | cp_page = grab_meta_page(sbi, start_blk); | |
836 | kaddr = page_address(cp_page); | |
837 | memcpy(kaddr, ckpt, (1 << sbi->log_blocksize)); | |
838 | set_page_dirty(cp_page); | |
839 | f2fs_put_page(cp_page, 1); | |
840 | ||
841 | /* wait for previous submitted node/meta pages writeback */ | |
fb51b5ef | 842 | wait_on_all_pages_writeback(sbi); |
127e670a | 843 | |
4ef51a8f | 844 | filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX); |
9df27d98 | 845 | filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX); |
127e670a JK |
846 | |
847 | /* update user_block_counts */ | |
848 | sbi->last_valid_block_count = sbi->total_valid_block_count; | |
849 | sbi->alloc_valid_block_count = 0; | |
850 | ||
851 | /* Here, we only have one bio having CP pack */ | |
577e3495 | 852 | sync_meta_pages(sbi, META_FLUSH, LONG_MAX); |
127e670a | 853 | |
6bacf52f | 854 | if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) { |
577e3495 JK |
855 | clear_prefree_segments(sbi); |
856 | F2FS_RESET_SB_DIRT(sbi); | |
857 | } | |
127e670a JK |
858 | } |
859 | ||
0a8165d7 | 860 | /* |
127e670a JK |
861 | * We guarantee that this checkpoint procedure should not fail. |
862 | */ | |
43727527 | 863 | void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount) |
127e670a JK |
864 | { |
865 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
866 | unsigned long long ckpt_ver; | |
867 | ||
2af4bd6c NJ |
868 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops"); |
869 | ||
43727527 JK |
870 | mutex_lock(&sbi->cp_mutex); |
871 | block_operations(sbi); | |
127e670a | 872 | |
2af4bd6c NJ |
873 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops"); |
874 | ||
458e6197 JK |
875 | f2fs_submit_merged_bio(sbi, DATA, WRITE); |
876 | f2fs_submit_merged_bio(sbi, NODE, WRITE); | |
877 | f2fs_submit_merged_bio(sbi, META, WRITE); | |
127e670a JK |
878 | |
879 | /* | |
880 | * update checkpoint pack index | |
881 | * Increase the version number so that | |
882 | * SIT entries and seg summaries are written at correct place | |
883 | */ | |
d71b5564 | 884 | ckpt_ver = cur_cp_version(ckpt); |
127e670a JK |
885 | ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver); |
886 | ||
887 | /* write cached NAT/SIT entries to NAT/SIT area */ | |
888 | flush_nat_entries(sbi); | |
889 | flush_sit_entries(sbi); | |
890 | ||
127e670a JK |
891 | /* unlock all the fs_lock[] in do_checkpoint() */ |
892 | do_checkpoint(sbi, is_umount); | |
893 | ||
894 | unblock_operations(sbi); | |
895 | mutex_unlock(&sbi->cp_mutex); | |
2af4bd6c | 896 | |
942e0be6 | 897 | stat_inc_cp_count(sbi->stat_info); |
2af4bd6c | 898 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint"); |
127e670a JK |
899 | } |
900 | ||
901 | void init_orphan_info(struct f2fs_sb_info *sbi) | |
902 | { | |
17b692f6 | 903 | spin_lock_init(&sbi->orphan_inode_lock); |
127e670a JK |
904 | INIT_LIST_HEAD(&sbi->orphan_inode_list); |
905 | sbi->n_orphans = 0; | |
0d47c1ad GZ |
906 | /* |
907 | * considering 512 blocks in a segment 8 blocks are needed for cp | |
908 | * and log segment summaries. Remaining blocks are used to keep | |
909 | * orphan entries with the limitation one reserved segment | |
910 | * for cp pack we can have max 1020*504 orphan entries | |
911 | */ | |
912 | sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE) | |
913 | * F2FS_ORPHANS_PER_BLOCK; | |
127e670a JK |
914 | } |
915 | ||
6e6093a8 | 916 | int __init create_checkpoint_caches(void) |
127e670a JK |
917 | { |
918 | orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry", | |
e8512d2e | 919 | sizeof(struct orphan_inode_entry)); |
6bacf52f | 920 | if (!orphan_entry_slab) |
127e670a JK |
921 | return -ENOMEM; |
922 | inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry", | |
e8512d2e | 923 | sizeof(struct dir_inode_entry)); |
6bacf52f | 924 | if (!inode_entry_slab) { |
127e670a JK |
925 | kmem_cache_destroy(orphan_entry_slab); |
926 | return -ENOMEM; | |
927 | } | |
928 | return 0; | |
929 | } | |
930 | ||
931 | void destroy_checkpoint_caches(void) | |
932 | { | |
933 | kmem_cache_destroy(orphan_entry_slab); | |
934 | kmem_cache_destroy(inode_entry_slab); | |
935 | } |