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 | |
e5748434 CY |
189 | trace_f2fs_writepages(mapping->host, wbc, META); |
190 | ||
5459aa97 | 191 | /* collect a number of dirty meta pages and write together */ |
50c8cdb3 JK |
192 | if (wbc->for_kupdate || |
193 | get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META)) | |
d3baf95d | 194 | goto skip_write; |
127e670a JK |
195 | |
196 | /* if mounting is failed, skip writing node pages */ | |
197 | mutex_lock(&sbi->cp_mutex); | |
50c8cdb3 JK |
198 | diff = nr_pages_to_write(sbi, META, wbc); |
199 | written = sync_meta_pages(sbi, META, wbc->nr_to_write); | |
127e670a | 200 | mutex_unlock(&sbi->cp_mutex); |
50c8cdb3 | 201 | wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff); |
127e670a | 202 | return 0; |
d3baf95d JK |
203 | |
204 | skip_write: | |
205 | wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META); | |
206 | return 0; | |
127e670a JK |
207 | } |
208 | ||
209 | long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type, | |
210 | long nr_to_write) | |
211 | { | |
9df27d98 | 212 | struct address_space *mapping = META_MAPPING(sbi); |
127e670a JK |
213 | pgoff_t index = 0, end = LONG_MAX; |
214 | struct pagevec pvec; | |
215 | long nwritten = 0; | |
216 | struct writeback_control wbc = { | |
217 | .for_reclaim = 0, | |
218 | }; | |
219 | ||
220 | pagevec_init(&pvec, 0); | |
221 | ||
222 | while (index <= end) { | |
223 | int i, nr_pages; | |
224 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
225 | PAGECACHE_TAG_DIRTY, | |
226 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); | |
cfb271d4 | 227 | if (unlikely(nr_pages == 0)) |
127e670a JK |
228 | break; |
229 | ||
230 | for (i = 0; i < nr_pages; i++) { | |
231 | struct page *page = pvec.pages[i]; | |
203681f6 | 232 | |
127e670a | 233 | lock_page(page); |
203681f6 JK |
234 | |
235 | if (unlikely(page->mapping != mapping)) { | |
236 | continue_unlock: | |
237 | unlock_page(page); | |
238 | continue; | |
239 | } | |
240 | if (!PageDirty(page)) { | |
241 | /* someone wrote it for us */ | |
242 | goto continue_unlock; | |
243 | } | |
244 | ||
245 | if (!clear_page_dirty_for_io(page)) | |
246 | goto continue_unlock; | |
247 | ||
577e3495 JK |
248 | if (f2fs_write_meta_page(page, &wbc)) { |
249 | unlock_page(page); | |
250 | break; | |
251 | } | |
cfb271d4 CY |
252 | nwritten++; |
253 | if (unlikely(nwritten >= nr_to_write)) | |
127e670a JK |
254 | break; |
255 | } | |
256 | pagevec_release(&pvec); | |
257 | cond_resched(); | |
258 | } | |
259 | ||
260 | if (nwritten) | |
458e6197 | 261 | f2fs_submit_merged_bio(sbi, type, WRITE); |
127e670a JK |
262 | |
263 | return nwritten; | |
264 | } | |
265 | ||
266 | static int f2fs_set_meta_page_dirty(struct page *page) | |
267 | { | |
268 | struct address_space *mapping = page->mapping; | |
269 | struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); | |
270 | ||
26c6b887 JK |
271 | trace_f2fs_set_page_dirty(page, META); |
272 | ||
127e670a JK |
273 | SetPageUptodate(page); |
274 | if (!PageDirty(page)) { | |
275 | __set_page_dirty_nobuffers(page); | |
276 | inc_page_count(sbi, F2FS_DIRTY_META); | |
127e670a JK |
277 | return 1; |
278 | } | |
279 | return 0; | |
280 | } | |
281 | ||
282 | const struct address_space_operations f2fs_meta_aops = { | |
283 | .writepage = f2fs_write_meta_page, | |
284 | .writepages = f2fs_write_meta_pages, | |
285 | .set_page_dirty = f2fs_set_meta_page_dirty, | |
286 | }; | |
287 | ||
cbd56e7d | 288 | int acquire_orphan_inode(struct f2fs_sb_info *sbi) |
127e670a | 289 | { |
127e670a JK |
290 | int err = 0; |
291 | ||
17b692f6 | 292 | spin_lock(&sbi->orphan_inode_lock); |
0d47c1ad | 293 | if (unlikely(sbi->n_orphans >= sbi->max_orphans)) |
127e670a | 294 | err = -ENOSPC; |
cbd56e7d JK |
295 | else |
296 | sbi->n_orphans++; | |
17b692f6 | 297 | spin_unlock(&sbi->orphan_inode_lock); |
0d47c1ad | 298 | |
127e670a JK |
299 | return err; |
300 | } | |
301 | ||
cbd56e7d JK |
302 | void release_orphan_inode(struct f2fs_sb_info *sbi) |
303 | { | |
17b692f6 | 304 | spin_lock(&sbi->orphan_inode_lock); |
5d56b671 | 305 | f2fs_bug_on(sbi->n_orphans == 0); |
cbd56e7d | 306 | sbi->n_orphans--; |
17b692f6 | 307 | spin_unlock(&sbi->orphan_inode_lock); |
cbd56e7d JK |
308 | } |
309 | ||
127e670a JK |
310 | void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) |
311 | { | |
2d7b822a CY |
312 | struct list_head *head; |
313 | struct orphan_inode_entry *new, *orphan; | |
127e670a | 314 | |
c1ef3725 GZ |
315 | new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC); |
316 | new->ino = ino; | |
317 | ||
17b692f6 | 318 | spin_lock(&sbi->orphan_inode_lock); |
127e670a | 319 | head = &sbi->orphan_inode_list; |
2d7b822a | 320 | list_for_each_entry(orphan, head, list) { |
c1ef3725 | 321 | if (orphan->ino == ino) { |
17b692f6 | 322 | spin_unlock(&sbi->orphan_inode_lock); |
c1ef3725 GZ |
323 | kmem_cache_free(orphan_entry_slab, new); |
324 | return; | |
325 | } | |
326 | ||
127e670a JK |
327 | if (orphan->ino > ino) |
328 | break; | |
127e670a | 329 | } |
7bd59381 | 330 | |
2d7b822a CY |
331 | /* add new orphan entry into list which is sorted by inode number */ |
332 | list_add_tail(&new->list, &orphan->list); | |
17b692f6 | 333 | spin_unlock(&sbi->orphan_inode_lock); |
127e670a JK |
334 | } |
335 | ||
336 | void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
337 | { | |
60ed9a0f | 338 | struct list_head *head; |
127e670a JK |
339 | struct orphan_inode_entry *orphan; |
340 | ||
17b692f6 | 341 | spin_lock(&sbi->orphan_inode_lock); |
127e670a | 342 | head = &sbi->orphan_inode_list; |
60ed9a0f | 343 | list_for_each_entry(orphan, head, list) { |
127e670a JK |
344 | if (orphan->ino == ino) { |
345 | list_del(&orphan->list); | |
5d56b671 | 346 | f2fs_bug_on(sbi->n_orphans == 0); |
127e670a | 347 | sbi->n_orphans--; |
cf0ee0f0 CY |
348 | spin_unlock(&sbi->orphan_inode_lock); |
349 | kmem_cache_free(orphan_entry_slab, orphan); | |
350 | return; | |
127e670a JK |
351 | } |
352 | } | |
17b692f6 | 353 | spin_unlock(&sbi->orphan_inode_lock); |
127e670a JK |
354 | } |
355 | ||
356 | static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
357 | { | |
358 | struct inode *inode = f2fs_iget(sbi->sb, ino); | |
5d56b671 | 359 | f2fs_bug_on(IS_ERR(inode)); |
127e670a JK |
360 | clear_nlink(inode); |
361 | ||
362 | /* truncate all the data during iput */ | |
363 | iput(inode); | |
364 | } | |
365 | ||
8f99a946 | 366 | void recover_orphan_inodes(struct f2fs_sb_info *sbi) |
127e670a JK |
367 | { |
368 | block_t start_blk, orphan_blkaddr, i, j; | |
369 | ||
25ca923b | 370 | if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG)) |
8f99a946 | 371 | return; |
127e670a | 372 | |
aabe5136 | 373 | sbi->por_doing = true; |
127e670a JK |
374 | start_blk = __start_cp_addr(sbi) + 1; |
375 | orphan_blkaddr = __start_sum_addr(sbi) - 1; | |
376 | ||
662befda CY |
377 | ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP); |
378 | ||
127e670a JK |
379 | for (i = 0; i < orphan_blkaddr; i++) { |
380 | struct page *page = get_meta_page(sbi, start_blk + i); | |
381 | struct f2fs_orphan_block *orphan_blk; | |
382 | ||
383 | orphan_blk = (struct f2fs_orphan_block *)page_address(page); | |
384 | for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) { | |
385 | nid_t ino = le32_to_cpu(orphan_blk->ino[j]); | |
386 | recover_orphan_inode(sbi, ino); | |
387 | } | |
388 | f2fs_put_page(page, 1); | |
389 | } | |
390 | /* clear Orphan Flag */ | |
25ca923b | 391 | clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG); |
aabe5136 | 392 | sbi->por_doing = false; |
8f99a946 | 393 | return; |
127e670a JK |
394 | } |
395 | ||
396 | static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk) | |
397 | { | |
502c6e0b | 398 | struct list_head *head; |
127e670a | 399 | struct f2fs_orphan_block *orphan_blk = NULL; |
127e670a | 400 | unsigned int nentries = 0; |
4531929e GZ |
401 | unsigned short index; |
402 | unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans + | |
403 | (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK); | |
404 | struct page *page = NULL; | |
502c6e0b | 405 | struct orphan_inode_entry *orphan = NULL; |
127e670a | 406 | |
4531929e | 407 | for (index = 0; index < orphan_blocks; index++) |
63f5384c | 408 | grab_meta_page(sbi, start_blk + index); |
127e670a | 409 | |
4531929e | 410 | index = 1; |
17b692f6 | 411 | spin_lock(&sbi->orphan_inode_lock); |
127e670a JK |
412 | head = &sbi->orphan_inode_list; |
413 | ||
414 | /* loop for each orphan inode entry and write them in Jornal block */ | |
502c6e0b GZ |
415 | list_for_each_entry(orphan, head, list) { |
416 | if (!page) { | |
63f5384c GZ |
417 | page = find_get_page(META_MAPPING(sbi), start_blk++); |
418 | f2fs_bug_on(!page); | |
502c6e0b GZ |
419 | orphan_blk = |
420 | (struct f2fs_orphan_block *)page_address(page); | |
421 | memset(orphan_blk, 0, sizeof(*orphan_blk)); | |
63f5384c | 422 | f2fs_put_page(page, 0); |
502c6e0b | 423 | } |
127e670a | 424 | |
36795567 | 425 | orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino); |
127e670a | 426 | |
36795567 | 427 | if (nentries == F2FS_ORPHANS_PER_BLOCK) { |
127e670a JK |
428 | /* |
429 | * an orphan block is full of 1020 entries, | |
430 | * then we need to flush current orphan blocks | |
431 | * and bring another one in memory | |
432 | */ | |
433 | orphan_blk->blk_addr = cpu_to_le16(index); | |
434 | orphan_blk->blk_count = cpu_to_le16(orphan_blocks); | |
435 | orphan_blk->entry_count = cpu_to_le32(nentries); | |
436 | set_page_dirty(page); | |
437 | f2fs_put_page(page, 1); | |
438 | index++; | |
127e670a JK |
439 | nentries = 0; |
440 | page = NULL; | |
441 | } | |
502c6e0b | 442 | } |
127e670a | 443 | |
502c6e0b GZ |
444 | if (page) { |
445 | orphan_blk->blk_addr = cpu_to_le16(index); | |
446 | orphan_blk->blk_count = cpu_to_le16(orphan_blocks); | |
447 | orphan_blk->entry_count = cpu_to_le32(nentries); | |
448 | set_page_dirty(page); | |
449 | f2fs_put_page(page, 1); | |
127e670a | 450 | } |
502c6e0b | 451 | |
17b692f6 | 452 | spin_unlock(&sbi->orphan_inode_lock); |
127e670a JK |
453 | } |
454 | ||
455 | static struct page *validate_checkpoint(struct f2fs_sb_info *sbi, | |
456 | block_t cp_addr, unsigned long long *version) | |
457 | { | |
458 | struct page *cp_page_1, *cp_page_2 = NULL; | |
459 | unsigned long blk_size = sbi->blocksize; | |
460 | struct f2fs_checkpoint *cp_block; | |
461 | unsigned long long cur_version = 0, pre_version = 0; | |
127e670a | 462 | size_t crc_offset; |
7e586fa0 | 463 | __u32 crc = 0; |
127e670a JK |
464 | |
465 | /* Read the 1st cp block in this CP pack */ | |
466 | cp_page_1 = get_meta_page(sbi, cp_addr); | |
467 | ||
468 | /* get the version number */ | |
469 | cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1); | |
470 | crc_offset = le32_to_cpu(cp_block->checksum_offset); | |
471 | if (crc_offset >= blk_size) | |
472 | goto invalid_cp1; | |
473 | ||
7e586fa0 | 474 | crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset))); |
127e670a JK |
475 | if (!f2fs_crc_valid(crc, cp_block, crc_offset)) |
476 | goto invalid_cp1; | |
477 | ||
d71b5564 | 478 | pre_version = cur_cp_version(cp_block); |
127e670a JK |
479 | |
480 | /* Read the 2nd cp block in this CP pack */ | |
25ca923b | 481 | cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1; |
127e670a JK |
482 | cp_page_2 = get_meta_page(sbi, cp_addr); |
483 | ||
484 | cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2); | |
485 | crc_offset = le32_to_cpu(cp_block->checksum_offset); | |
486 | if (crc_offset >= blk_size) | |
487 | goto invalid_cp2; | |
488 | ||
7e586fa0 | 489 | crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset))); |
127e670a JK |
490 | if (!f2fs_crc_valid(crc, cp_block, crc_offset)) |
491 | goto invalid_cp2; | |
492 | ||
d71b5564 | 493 | cur_version = cur_cp_version(cp_block); |
127e670a JK |
494 | |
495 | if (cur_version == pre_version) { | |
496 | *version = cur_version; | |
497 | f2fs_put_page(cp_page_2, 1); | |
498 | return cp_page_1; | |
499 | } | |
500 | invalid_cp2: | |
501 | f2fs_put_page(cp_page_2, 1); | |
502 | invalid_cp1: | |
503 | f2fs_put_page(cp_page_1, 1); | |
504 | return NULL; | |
505 | } | |
506 | ||
507 | int get_valid_checkpoint(struct f2fs_sb_info *sbi) | |
508 | { | |
509 | struct f2fs_checkpoint *cp_block; | |
510 | struct f2fs_super_block *fsb = sbi->raw_super; | |
511 | struct page *cp1, *cp2, *cur_page; | |
512 | unsigned long blk_size = sbi->blocksize; | |
513 | unsigned long long cp1_version = 0, cp2_version = 0; | |
514 | unsigned long long cp_start_blk_no; | |
515 | ||
516 | sbi->ckpt = kzalloc(blk_size, GFP_KERNEL); | |
517 | if (!sbi->ckpt) | |
518 | return -ENOMEM; | |
519 | /* | |
520 | * Finding out valid cp block involves read both | |
521 | * sets( cp pack1 and cp pack 2) | |
522 | */ | |
523 | cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr); | |
524 | cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version); | |
525 | ||
526 | /* The second checkpoint pack should start at the next segment */ | |
f9a4e6df JK |
527 | cp_start_blk_no += ((unsigned long long)1) << |
528 | le32_to_cpu(fsb->log_blocks_per_seg); | |
127e670a JK |
529 | cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version); |
530 | ||
531 | if (cp1 && cp2) { | |
532 | if (ver_after(cp2_version, cp1_version)) | |
533 | cur_page = cp2; | |
534 | else | |
535 | cur_page = cp1; | |
536 | } else if (cp1) { | |
537 | cur_page = cp1; | |
538 | } else if (cp2) { | |
539 | cur_page = cp2; | |
540 | } else { | |
541 | goto fail_no_cp; | |
542 | } | |
543 | ||
544 | cp_block = (struct f2fs_checkpoint *)page_address(cur_page); | |
545 | memcpy(sbi->ckpt, cp_block, blk_size); | |
546 | ||
547 | f2fs_put_page(cp1, 1); | |
548 | f2fs_put_page(cp2, 1); | |
549 | return 0; | |
550 | ||
551 | fail_no_cp: | |
552 | kfree(sbi->ckpt); | |
553 | return -EINVAL; | |
554 | } | |
555 | ||
5deb8267 | 556 | static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new) |
127e670a JK |
557 | { |
558 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
127e670a | 559 | |
ed57c27f JK |
560 | if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) |
561 | return -EEXIST; | |
2d7b822a | 562 | |
ed57c27f JK |
563 | set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR); |
564 | F2FS_I(inode)->dirty_dir = new; | |
565 | list_add_tail(&new->list, &sbi->dir_inode_list); | |
dcdfff65 | 566 | stat_inc_dirty_dir(sbi); |
5deb8267 JK |
567 | return 0; |
568 | } | |
569 | ||
570 | void set_dirty_dir_page(struct inode *inode, struct page *page) | |
571 | { | |
572 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
573 | struct dir_inode_entry *new; | |
cf0ee0f0 | 574 | int ret = 0; |
5deb8267 | 575 | |
127e670a JK |
576 | if (!S_ISDIR(inode->i_mode)) |
577 | return; | |
7bd59381 GZ |
578 | |
579 | new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); | |
127e670a JK |
580 | new->inode = inode; |
581 | INIT_LIST_HEAD(&new->list); | |
582 | ||
583 | spin_lock(&sbi->dir_inode_lock); | |
cf0ee0f0 | 584 | ret = __add_dirty_inode(inode, new); |
127e670a JK |
585 | inode_inc_dirty_dents(inode); |
586 | SetPagePrivate(page); | |
5deb8267 | 587 | spin_unlock(&sbi->dir_inode_lock); |
cf0ee0f0 CY |
588 | |
589 | if (ret) | |
590 | kmem_cache_free(inode_entry_slab, new); | |
5deb8267 JK |
591 | } |
592 | ||
593 | void add_dirty_dir_inode(struct inode *inode) | |
594 | { | |
595 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
7bd59381 GZ |
596 | struct dir_inode_entry *new = |
597 | f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); | |
cf0ee0f0 | 598 | int ret = 0; |
7bd59381 | 599 | |
5deb8267 JK |
600 | new->inode = inode; |
601 | INIT_LIST_HEAD(&new->list); | |
127e670a | 602 | |
5deb8267 | 603 | spin_lock(&sbi->dir_inode_lock); |
cf0ee0f0 | 604 | ret = __add_dirty_inode(inode, new); |
127e670a | 605 | spin_unlock(&sbi->dir_inode_lock); |
cf0ee0f0 CY |
606 | |
607 | if (ret) | |
608 | kmem_cache_free(inode_entry_slab, new); | |
127e670a JK |
609 | } |
610 | ||
611 | void remove_dirty_dir_inode(struct inode *inode) | |
612 | { | |
613 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
2d7b822a | 614 | struct dir_inode_entry *entry; |
127e670a JK |
615 | |
616 | if (!S_ISDIR(inode->i_mode)) | |
617 | return; | |
618 | ||
619 | spin_lock(&sbi->dir_inode_lock); | |
ed57c27f JK |
620 | if (get_dirty_dents(inode) || |
621 | !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) { | |
3b10b1fd JK |
622 | spin_unlock(&sbi->dir_inode_lock); |
623 | return; | |
624 | } | |
127e670a | 625 | |
ed57c27f JK |
626 | entry = F2FS_I(inode)->dirty_dir; |
627 | list_del(&entry->list); | |
628 | F2FS_I(inode)->dirty_dir = NULL; | |
629 | clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR); | |
630 | stat_dec_dirty_dir(sbi); | |
127e670a | 631 | spin_unlock(&sbi->dir_inode_lock); |
ed57c27f | 632 | kmem_cache_free(inode_entry_slab, entry); |
74d0b917 JK |
633 | |
634 | /* Only from the recovery routine */ | |
afc3eda2 JK |
635 | if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) { |
636 | clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT); | |
74d0b917 | 637 | iput(inode); |
afc3eda2 | 638 | } |
74d0b917 JK |
639 | } |
640 | ||
127e670a JK |
641 | void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi) |
642 | { | |
ce3b7d80 | 643 | struct list_head *head; |
127e670a JK |
644 | struct dir_inode_entry *entry; |
645 | struct inode *inode; | |
646 | retry: | |
647 | spin_lock(&sbi->dir_inode_lock); | |
ce3b7d80 GZ |
648 | |
649 | head = &sbi->dir_inode_list; | |
127e670a JK |
650 | if (list_empty(head)) { |
651 | spin_unlock(&sbi->dir_inode_lock); | |
652 | return; | |
653 | } | |
654 | entry = list_entry(head->next, struct dir_inode_entry, list); | |
655 | inode = igrab(entry->inode); | |
656 | spin_unlock(&sbi->dir_inode_lock); | |
657 | if (inode) { | |
87d6f890 | 658 | filemap_fdatawrite(inode->i_mapping); |
127e670a JK |
659 | iput(inode); |
660 | } else { | |
661 | /* | |
662 | * We should submit bio, since it exists several | |
663 | * wribacking dentry pages in the freeing inode. | |
664 | */ | |
458e6197 | 665 | f2fs_submit_merged_bio(sbi, DATA, WRITE); |
127e670a JK |
666 | } |
667 | goto retry; | |
668 | } | |
669 | ||
0a8165d7 | 670 | /* |
127e670a JK |
671 | * Freeze all the FS-operations for checkpoint. |
672 | */ | |
43727527 | 673 | static void block_operations(struct f2fs_sb_info *sbi) |
127e670a | 674 | { |
127e670a JK |
675 | struct writeback_control wbc = { |
676 | .sync_mode = WB_SYNC_ALL, | |
677 | .nr_to_write = LONG_MAX, | |
678 | .for_reclaim = 0, | |
679 | }; | |
c718379b JK |
680 | struct blk_plug plug; |
681 | ||
682 | blk_start_plug(&plug); | |
683 | ||
39936837 | 684 | retry_flush_dents: |
e479556b | 685 | f2fs_lock_all(sbi); |
127e670a | 686 | /* write all the dirty dentry pages */ |
127e670a | 687 | if (get_pages(sbi, F2FS_DIRTY_DENTS)) { |
e479556b | 688 | f2fs_unlock_all(sbi); |
39936837 JK |
689 | sync_dirty_dir_inodes(sbi); |
690 | goto retry_flush_dents; | |
127e670a JK |
691 | } |
692 | ||
127e670a JK |
693 | /* |
694 | * POR: we should ensure that there is no dirty node pages | |
695 | * until finishing nat/sit flush. | |
696 | */ | |
39936837 JK |
697 | retry_flush_nodes: |
698 | mutex_lock(&sbi->node_write); | |
127e670a JK |
699 | |
700 | if (get_pages(sbi, F2FS_DIRTY_NODES)) { | |
39936837 JK |
701 | mutex_unlock(&sbi->node_write); |
702 | sync_node_pages(sbi, 0, &wbc); | |
703 | goto retry_flush_nodes; | |
127e670a | 704 | } |
c718379b | 705 | blk_finish_plug(&plug); |
127e670a JK |
706 | } |
707 | ||
708 | static void unblock_operations(struct f2fs_sb_info *sbi) | |
709 | { | |
39936837 | 710 | mutex_unlock(&sbi->node_write); |
e479556b | 711 | f2fs_unlock_all(sbi); |
127e670a JK |
712 | } |
713 | ||
fb51b5ef CL |
714 | static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi) |
715 | { | |
716 | DEFINE_WAIT(wait); | |
717 | ||
718 | for (;;) { | |
719 | prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE); | |
720 | ||
721 | if (!get_pages(sbi, F2FS_WRITEBACK)) | |
722 | break; | |
723 | ||
724 | io_schedule(); | |
725 | } | |
726 | finish_wait(&sbi->cp_wait, &wait); | |
727 | } | |
728 | ||
127e670a JK |
729 | static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount) |
730 | { | |
731 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
732 | nid_t last_nid = 0; | |
733 | block_t start_blk; | |
734 | struct page *cp_page; | |
735 | unsigned int data_sum_blocks, orphan_blocks; | |
7e586fa0 | 736 | __u32 crc32 = 0; |
127e670a | 737 | void *kaddr; |
127e670a JK |
738 | int i; |
739 | ||
1e87a78d JK |
740 | /* |
741 | * This avoids to conduct wrong roll-forward operations and uses | |
742 | * metapages, so should be called prior to sync_meta_pages below. | |
743 | */ | |
744 | discard_next_dnode(sbi); | |
745 | ||
127e670a JK |
746 | /* Flush all the NAT/SIT pages */ |
747 | while (get_pages(sbi, F2FS_DIRTY_META)) | |
748 | sync_meta_pages(sbi, META, LONG_MAX); | |
749 | ||
750 | next_free_nid(sbi, &last_nid); | |
751 | ||
752 | /* | |
753 | * modify checkpoint | |
754 | * version number is already updated | |
755 | */ | |
756 | ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi)); | |
757 | ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi)); | |
758 | ckpt->free_segment_count = cpu_to_le32(free_segments(sbi)); | |
759 | for (i = 0; i < 3; i++) { | |
760 | ckpt->cur_node_segno[i] = | |
761 | cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE)); | |
762 | ckpt->cur_node_blkoff[i] = | |
763 | cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE)); | |
764 | ckpt->alloc_type[i + CURSEG_HOT_NODE] = | |
765 | curseg_alloc_type(sbi, i + CURSEG_HOT_NODE); | |
766 | } | |
767 | for (i = 0; i < 3; i++) { | |
768 | ckpt->cur_data_segno[i] = | |
769 | cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA)); | |
770 | ckpt->cur_data_blkoff[i] = | |
771 | cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA)); | |
772 | ckpt->alloc_type[i + CURSEG_HOT_DATA] = | |
773 | curseg_alloc_type(sbi, i + CURSEG_HOT_DATA); | |
774 | } | |
775 | ||
776 | ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi)); | |
777 | ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi)); | |
778 | ckpt->next_free_nid = cpu_to_le32(last_nid); | |
779 | ||
780 | /* 2 cp + n data seg summary + orphan inode blocks */ | |
781 | data_sum_blocks = npages_for_summary_flush(sbi); | |
782 | if (data_sum_blocks < 3) | |
25ca923b | 783 | set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); |
127e670a | 784 | else |
25ca923b | 785 | clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); |
127e670a JK |
786 | |
787 | orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1) | |
788 | / F2FS_ORPHANS_PER_BLOCK; | |
25ca923b | 789 | ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks); |
127e670a JK |
790 | |
791 | if (is_umount) { | |
25ca923b JK |
792 | set_ckpt_flags(ckpt, CP_UMOUNT_FLAG); |
793 | ckpt->cp_pack_total_block_count = cpu_to_le32(2 + | |
794 | data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE); | |
127e670a | 795 | } else { |
25ca923b JK |
796 | clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG); |
797 | ckpt->cp_pack_total_block_count = cpu_to_le32(2 + | |
798 | data_sum_blocks + orphan_blocks); | |
127e670a JK |
799 | } |
800 | ||
801 | if (sbi->n_orphans) | |
25ca923b | 802 | set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); |
127e670a | 803 | else |
25ca923b | 804 | clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); |
127e670a JK |
805 | |
806 | /* update SIT/NAT bitmap */ | |
807 | get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP)); | |
808 | get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP)); | |
809 | ||
810 | crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset)); | |
7e586fa0 JK |
811 | *((__le32 *)((unsigned char *)ckpt + |
812 | le32_to_cpu(ckpt->checksum_offset))) | |
127e670a JK |
813 | = cpu_to_le32(crc32); |
814 | ||
815 | start_blk = __start_cp_addr(sbi); | |
816 | ||
817 | /* write out checkpoint buffer at block 0 */ | |
818 | cp_page = grab_meta_page(sbi, start_blk++); | |
819 | kaddr = page_address(cp_page); | |
820 | memcpy(kaddr, ckpt, (1 << sbi->log_blocksize)); | |
821 | set_page_dirty(cp_page); | |
822 | f2fs_put_page(cp_page, 1); | |
823 | ||
824 | if (sbi->n_orphans) { | |
825 | write_orphan_inodes(sbi, start_blk); | |
826 | start_blk += orphan_blocks; | |
827 | } | |
828 | ||
829 | write_data_summaries(sbi, start_blk); | |
830 | start_blk += data_sum_blocks; | |
831 | if (is_umount) { | |
832 | write_node_summaries(sbi, start_blk); | |
833 | start_blk += NR_CURSEG_NODE_TYPE; | |
834 | } | |
835 | ||
836 | /* writeout checkpoint block */ | |
837 | cp_page = grab_meta_page(sbi, start_blk); | |
838 | kaddr = page_address(cp_page); | |
839 | memcpy(kaddr, ckpt, (1 << sbi->log_blocksize)); | |
840 | set_page_dirty(cp_page); | |
841 | f2fs_put_page(cp_page, 1); | |
842 | ||
843 | /* wait for previous submitted node/meta pages writeback */ | |
fb51b5ef | 844 | wait_on_all_pages_writeback(sbi); |
127e670a | 845 | |
4ef51a8f | 846 | filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX); |
9df27d98 | 847 | filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX); |
127e670a JK |
848 | |
849 | /* update user_block_counts */ | |
850 | sbi->last_valid_block_count = sbi->total_valid_block_count; | |
851 | sbi->alloc_valid_block_count = 0; | |
852 | ||
853 | /* Here, we only have one bio having CP pack */ | |
577e3495 | 854 | sync_meta_pages(sbi, META_FLUSH, LONG_MAX); |
127e670a | 855 | |
6bacf52f | 856 | if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) { |
577e3495 JK |
857 | clear_prefree_segments(sbi); |
858 | F2FS_RESET_SB_DIRT(sbi); | |
859 | } | |
127e670a JK |
860 | } |
861 | ||
0a8165d7 | 862 | /* |
127e670a JK |
863 | * We guarantee that this checkpoint procedure should not fail. |
864 | */ | |
43727527 | 865 | void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount) |
127e670a JK |
866 | { |
867 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
868 | unsigned long long ckpt_ver; | |
869 | ||
2af4bd6c NJ |
870 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops"); |
871 | ||
43727527 JK |
872 | mutex_lock(&sbi->cp_mutex); |
873 | block_operations(sbi); | |
127e670a | 874 | |
2af4bd6c NJ |
875 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops"); |
876 | ||
458e6197 JK |
877 | f2fs_submit_merged_bio(sbi, DATA, WRITE); |
878 | f2fs_submit_merged_bio(sbi, NODE, WRITE); | |
879 | f2fs_submit_merged_bio(sbi, META, WRITE); | |
127e670a JK |
880 | |
881 | /* | |
882 | * update checkpoint pack index | |
883 | * Increase the version number so that | |
884 | * SIT entries and seg summaries are written at correct place | |
885 | */ | |
d71b5564 | 886 | ckpt_ver = cur_cp_version(ckpt); |
127e670a JK |
887 | ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver); |
888 | ||
889 | /* write cached NAT/SIT entries to NAT/SIT area */ | |
890 | flush_nat_entries(sbi); | |
891 | flush_sit_entries(sbi); | |
892 | ||
127e670a JK |
893 | /* unlock all the fs_lock[] in do_checkpoint() */ |
894 | do_checkpoint(sbi, is_umount); | |
895 | ||
896 | unblock_operations(sbi); | |
897 | mutex_unlock(&sbi->cp_mutex); | |
2af4bd6c | 898 | |
942e0be6 | 899 | stat_inc_cp_count(sbi->stat_info); |
2af4bd6c | 900 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint"); |
127e670a JK |
901 | } |
902 | ||
903 | void init_orphan_info(struct f2fs_sb_info *sbi) | |
904 | { | |
17b692f6 | 905 | spin_lock_init(&sbi->orphan_inode_lock); |
127e670a JK |
906 | INIT_LIST_HEAD(&sbi->orphan_inode_list); |
907 | sbi->n_orphans = 0; | |
0d47c1ad GZ |
908 | /* |
909 | * considering 512 blocks in a segment 8 blocks are needed for cp | |
910 | * and log segment summaries. Remaining blocks are used to keep | |
911 | * orphan entries with the limitation one reserved segment | |
912 | * for cp pack we can have max 1020*504 orphan entries | |
913 | */ | |
914 | sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE) | |
915 | * F2FS_ORPHANS_PER_BLOCK; | |
127e670a JK |
916 | } |
917 | ||
6e6093a8 | 918 | int __init create_checkpoint_caches(void) |
127e670a JK |
919 | { |
920 | orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry", | |
e8512d2e | 921 | sizeof(struct orphan_inode_entry)); |
6bacf52f | 922 | if (!orphan_entry_slab) |
127e670a JK |
923 | return -ENOMEM; |
924 | inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry", | |
e8512d2e | 925 | sizeof(struct dir_inode_entry)); |
6bacf52f | 926 | if (!inode_entry_slab) { |
127e670a JK |
927 | kmem_cache_destroy(orphan_entry_slab); |
928 | return -ENOMEM; | |
929 | } | |
930 | return 0; | |
931 | } | |
932 | ||
933 | void destroy_checkpoint_caches(void) | |
934 | { | |
935 | kmem_cache_destroy(orphan_entry_slab); | |
936 | kmem_cache_destroy(inode_entry_slab); | |
937 | } |