]>
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 | ||
662befda CY |
78 | inline int get_max_meta_blks(struct f2fs_sb_info *sbi, int type) |
79 | { | |
80 | switch (type) { | |
81 | case META_NAT: | |
82 | return NM_I(sbi)->max_nid / NAT_ENTRY_PER_BLOCK; | |
83 | case META_SIT: | |
84 | return SIT_BLK_CNT(sbi); | |
85 | case META_CP: | |
86 | return 0; | |
87 | default: | |
88 | BUG(); | |
89 | } | |
90 | } | |
91 | ||
92 | /* | |
93 | * Readahead CP/NAT/SIT pages | |
94 | */ | |
95 | int ra_meta_pages(struct f2fs_sb_info *sbi, int start, int nrpages, int type) | |
96 | { | |
97 | block_t prev_blk_addr = 0; | |
98 | struct page *page; | |
99 | int blkno = start; | |
100 | int max_blks = get_max_meta_blks(sbi, type); | |
101 | ||
102 | struct f2fs_io_info fio = { | |
103 | .type = META, | |
104 | .rw = READ_SYNC | REQ_META | REQ_PRIO | |
105 | }; | |
106 | ||
107 | for (; nrpages-- > 0; blkno++) { | |
108 | block_t blk_addr; | |
109 | ||
110 | switch (type) { | |
111 | case META_NAT: | |
112 | /* get nat block addr */ | |
113 | if (unlikely(blkno >= max_blks)) | |
114 | blkno = 0; | |
115 | blk_addr = current_nat_addr(sbi, | |
116 | blkno * NAT_ENTRY_PER_BLOCK); | |
117 | break; | |
118 | case META_SIT: | |
119 | /* get sit block addr */ | |
120 | if (unlikely(blkno >= max_blks)) | |
121 | goto out; | |
122 | blk_addr = current_sit_addr(sbi, | |
123 | blkno * SIT_ENTRY_PER_BLOCK); | |
124 | if (blkno != start && prev_blk_addr + 1 != blk_addr) | |
125 | goto out; | |
126 | prev_blk_addr = blk_addr; | |
127 | break; | |
128 | case META_CP: | |
129 | /* get cp block addr */ | |
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 | |
203681f6 | 160 | if (unlikely(sbi->por_doing)) |
cfb271d4 | 161 | goto redirty_out; |
cfb271d4 CY |
162 | if (wbc->for_reclaim) |
163 | goto redirty_out; | |
127e670a | 164 | |
203681f6 JK |
165 | /* Should not write any meta pages, if any IO error was occurred */ |
166 | if (unlikely(is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG))) | |
167 | goto no_write; | |
127e670a | 168 | |
203681f6 | 169 | wait_on_page_writeback(page); |
577e3495 | 170 | write_meta_page(sbi, page); |
203681f6 | 171 | no_write: |
577e3495 JK |
172 | dec_page_count(sbi, F2FS_DIRTY_META); |
173 | unlock_page(page); | |
174 | return 0; | |
cfb271d4 CY |
175 | |
176 | redirty_out: | |
177 | dec_page_count(sbi, F2FS_DIRTY_META); | |
178 | wbc->pages_skipped++; | |
179 | set_page_dirty(page); | |
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); | |
5459aa97 | 187 | int nrpages = MAX_BIO_BLOCKS(max_hw_blocks(sbi)); |
127e670a JK |
188 | long written; |
189 | ||
190 | if (wbc->for_kupdate) | |
191 | return 0; | |
192 | ||
5459aa97 JK |
193 | /* collect a number of dirty meta pages and write together */ |
194 | if (get_pages(sbi, F2FS_DIRTY_META) < nrpages) | |
127e670a JK |
195 | return 0; |
196 | ||
197 | /* if mounting is failed, skip writing node pages */ | |
198 | mutex_lock(&sbi->cp_mutex); | |
5459aa97 | 199 | written = sync_meta_pages(sbi, META, nrpages); |
127e670a JK |
200 | mutex_unlock(&sbi->cp_mutex); |
201 | wbc->nr_to_write -= written; | |
202 | return 0; | |
203 | } | |
204 | ||
205 | long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type, | |
206 | long nr_to_write) | |
207 | { | |
9df27d98 | 208 | struct address_space *mapping = META_MAPPING(sbi); |
127e670a JK |
209 | pgoff_t index = 0, end = LONG_MAX; |
210 | struct pagevec pvec; | |
211 | long nwritten = 0; | |
212 | struct writeback_control wbc = { | |
213 | .for_reclaim = 0, | |
214 | }; | |
215 | ||
216 | pagevec_init(&pvec, 0); | |
217 | ||
218 | while (index <= end) { | |
219 | int i, nr_pages; | |
220 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
221 | PAGECACHE_TAG_DIRTY, | |
222 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); | |
cfb271d4 | 223 | if (unlikely(nr_pages == 0)) |
127e670a JK |
224 | break; |
225 | ||
226 | for (i = 0; i < nr_pages; i++) { | |
227 | struct page *page = pvec.pages[i]; | |
203681f6 | 228 | |
127e670a | 229 | lock_page(page); |
203681f6 JK |
230 | |
231 | if (unlikely(page->mapping != mapping)) { | |
232 | continue_unlock: | |
233 | unlock_page(page); | |
234 | continue; | |
235 | } | |
236 | if (!PageDirty(page)) { | |
237 | /* someone wrote it for us */ | |
238 | goto continue_unlock; | |
239 | } | |
240 | ||
241 | if (!clear_page_dirty_for_io(page)) | |
242 | goto continue_unlock; | |
243 | ||
577e3495 JK |
244 | if (f2fs_write_meta_page(page, &wbc)) { |
245 | unlock_page(page); | |
246 | break; | |
247 | } | |
cfb271d4 CY |
248 | nwritten++; |
249 | if (unlikely(nwritten >= nr_to_write)) | |
127e670a JK |
250 | break; |
251 | } | |
252 | pagevec_release(&pvec); | |
253 | cond_resched(); | |
254 | } | |
255 | ||
256 | if (nwritten) | |
458e6197 | 257 | f2fs_submit_merged_bio(sbi, type, WRITE); |
127e670a JK |
258 | |
259 | return nwritten; | |
260 | } | |
261 | ||
262 | static int f2fs_set_meta_page_dirty(struct page *page) | |
263 | { | |
264 | struct address_space *mapping = page->mapping; | |
265 | struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb); | |
266 | ||
26c6b887 JK |
267 | trace_f2fs_set_page_dirty(page, META); |
268 | ||
127e670a JK |
269 | SetPageUptodate(page); |
270 | if (!PageDirty(page)) { | |
271 | __set_page_dirty_nobuffers(page); | |
272 | inc_page_count(sbi, F2FS_DIRTY_META); | |
127e670a JK |
273 | return 1; |
274 | } | |
275 | return 0; | |
276 | } | |
277 | ||
278 | const struct address_space_operations f2fs_meta_aops = { | |
279 | .writepage = f2fs_write_meta_page, | |
280 | .writepages = f2fs_write_meta_pages, | |
281 | .set_page_dirty = f2fs_set_meta_page_dirty, | |
282 | }; | |
283 | ||
cbd56e7d | 284 | int acquire_orphan_inode(struct f2fs_sb_info *sbi) |
127e670a | 285 | { |
127e670a JK |
286 | int err = 0; |
287 | ||
17b692f6 | 288 | spin_lock(&sbi->orphan_inode_lock); |
0d47c1ad | 289 | if (unlikely(sbi->n_orphans >= sbi->max_orphans)) |
127e670a | 290 | err = -ENOSPC; |
cbd56e7d JK |
291 | else |
292 | sbi->n_orphans++; | |
17b692f6 | 293 | spin_unlock(&sbi->orphan_inode_lock); |
0d47c1ad | 294 | |
127e670a JK |
295 | return err; |
296 | } | |
297 | ||
cbd56e7d JK |
298 | void release_orphan_inode(struct f2fs_sb_info *sbi) |
299 | { | |
17b692f6 | 300 | spin_lock(&sbi->orphan_inode_lock); |
5d56b671 | 301 | f2fs_bug_on(sbi->n_orphans == 0); |
cbd56e7d | 302 | sbi->n_orphans--; |
17b692f6 | 303 | spin_unlock(&sbi->orphan_inode_lock); |
cbd56e7d JK |
304 | } |
305 | ||
127e670a JK |
306 | void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) |
307 | { | |
308 | struct list_head *head, *this; | |
309 | struct orphan_inode_entry *new = NULL, *orphan = NULL; | |
310 | ||
c1ef3725 GZ |
311 | new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC); |
312 | new->ino = ino; | |
313 | ||
17b692f6 | 314 | spin_lock(&sbi->orphan_inode_lock); |
127e670a JK |
315 | head = &sbi->orphan_inode_list; |
316 | list_for_each(this, head) { | |
317 | orphan = list_entry(this, struct orphan_inode_entry, list); | |
c1ef3725 | 318 | if (orphan->ino == ino) { |
17b692f6 | 319 | spin_unlock(&sbi->orphan_inode_lock); |
c1ef3725 GZ |
320 | kmem_cache_free(orphan_entry_slab, new); |
321 | return; | |
322 | } | |
323 | ||
127e670a JK |
324 | if (orphan->ino > ino) |
325 | break; | |
326 | orphan = NULL; | |
327 | } | |
7bd59381 | 328 | |
127e670a | 329 | /* add new_oentry into list which is sorted by inode number */ |
a2617dc6 | 330 | if (orphan) |
331 | list_add(&new->list, this->prev); | |
332 | else | |
127e670a | 333 | list_add_tail(&new->list, head); |
17b692f6 | 334 | spin_unlock(&sbi->orphan_inode_lock); |
127e670a JK |
335 | } |
336 | ||
337 | void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
338 | { | |
60ed9a0f | 339 | struct list_head *head; |
127e670a JK |
340 | struct orphan_inode_entry *orphan; |
341 | ||
17b692f6 | 342 | spin_lock(&sbi->orphan_inode_lock); |
127e670a | 343 | head = &sbi->orphan_inode_list; |
60ed9a0f | 344 | list_for_each_entry(orphan, head, list) { |
127e670a JK |
345 | if (orphan->ino == ino) { |
346 | list_del(&orphan->list); | |
347 | kmem_cache_free(orphan_entry_slab, orphan); | |
5d56b671 | 348 | f2fs_bug_on(sbi->n_orphans == 0); |
127e670a JK |
349 | sbi->n_orphans--; |
350 | break; | |
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); | |
559 | struct list_head *head = &sbi->dir_inode_list; | |
127e670a JK |
560 | struct list_head *this; |
561 | ||
5deb8267 JK |
562 | list_for_each(this, head) { |
563 | struct dir_inode_entry *entry; | |
564 | entry = list_entry(this, struct dir_inode_entry, list); | |
6bacf52f | 565 | if (unlikely(entry->inode == inode)) |
5deb8267 JK |
566 | return -EEXIST; |
567 | } | |
568 | list_add_tail(&new->list, head); | |
dcdfff65 | 569 | stat_inc_dirty_dir(sbi); |
5deb8267 JK |
570 | return 0; |
571 | } | |
572 | ||
573 | void set_dirty_dir_page(struct inode *inode, struct page *page) | |
574 | { | |
575 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
576 | struct dir_inode_entry *new; | |
577 | ||
127e670a JK |
578 | if (!S_ISDIR(inode->i_mode)) |
579 | return; | |
7bd59381 GZ |
580 | |
581 | new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); | |
127e670a JK |
582 | new->inode = inode; |
583 | INIT_LIST_HEAD(&new->list); | |
584 | ||
585 | spin_lock(&sbi->dir_inode_lock); | |
5deb8267 JK |
586 | if (__add_dirty_inode(inode, new)) |
587 | kmem_cache_free(inode_entry_slab, new); | |
127e670a | 588 | |
127e670a JK |
589 | inode_inc_dirty_dents(inode); |
590 | SetPagePrivate(page); | |
5deb8267 JK |
591 | spin_unlock(&sbi->dir_inode_lock); |
592 | } | |
593 | ||
594 | void add_dirty_dir_inode(struct inode *inode) | |
595 | { | |
596 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
7bd59381 GZ |
597 | struct dir_inode_entry *new = |
598 | f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS); | |
599 | ||
5deb8267 JK |
600 | new->inode = inode; |
601 | INIT_LIST_HEAD(&new->list); | |
127e670a | 602 | |
5deb8267 JK |
603 | spin_lock(&sbi->dir_inode_lock); |
604 | if (__add_dirty_inode(inode, new)) | |
605 | kmem_cache_free(inode_entry_slab, new); | |
127e670a JK |
606 | spin_unlock(&sbi->dir_inode_lock); |
607 | } | |
608 | ||
609 | void remove_dirty_dir_inode(struct inode *inode) | |
610 | { | |
611 | struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); | |
ce3b7d80 GZ |
612 | |
613 | struct list_head *this, *head; | |
127e670a JK |
614 | |
615 | if (!S_ISDIR(inode->i_mode)) | |
616 | return; | |
617 | ||
618 | spin_lock(&sbi->dir_inode_lock); | |
3b10b1fd JK |
619 | if (atomic_read(&F2FS_I(inode)->dirty_dents)) { |
620 | spin_unlock(&sbi->dir_inode_lock); | |
621 | return; | |
622 | } | |
127e670a | 623 | |
ce3b7d80 | 624 | head = &sbi->dir_inode_list; |
127e670a JK |
625 | list_for_each(this, head) { |
626 | struct dir_inode_entry *entry; | |
627 | entry = list_entry(this, struct dir_inode_entry, list); | |
628 | if (entry->inode == inode) { | |
629 | list_del(&entry->list); | |
630 | kmem_cache_free(inode_entry_slab, entry); | |
dcdfff65 | 631 | stat_dec_dirty_dir(sbi); |
127e670a JK |
632 | break; |
633 | } | |
634 | } | |
127e670a | 635 | spin_unlock(&sbi->dir_inode_lock); |
74d0b917 JK |
636 | |
637 | /* Only from the recovery routine */ | |
afc3eda2 JK |
638 | if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) { |
639 | clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT); | |
74d0b917 | 640 | iput(inode); |
afc3eda2 | 641 | } |
74d0b917 JK |
642 | } |
643 | ||
644 | struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino) | |
645 | { | |
ce3b7d80 GZ |
646 | |
647 | struct list_head *this, *head; | |
74d0b917 JK |
648 | struct inode *inode = NULL; |
649 | ||
650 | spin_lock(&sbi->dir_inode_lock); | |
ce3b7d80 GZ |
651 | |
652 | head = &sbi->dir_inode_list; | |
74d0b917 JK |
653 | list_for_each(this, head) { |
654 | struct dir_inode_entry *entry; | |
655 | entry = list_entry(this, struct dir_inode_entry, list); | |
656 | if (entry->inode->i_ino == ino) { | |
657 | inode = entry->inode; | |
658 | break; | |
659 | } | |
660 | } | |
661 | spin_unlock(&sbi->dir_inode_lock); | |
662 | return inode; | |
127e670a JK |
663 | } |
664 | ||
665 | void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi) | |
666 | { | |
ce3b7d80 | 667 | struct list_head *head; |
127e670a JK |
668 | struct dir_inode_entry *entry; |
669 | struct inode *inode; | |
670 | retry: | |
671 | spin_lock(&sbi->dir_inode_lock); | |
ce3b7d80 GZ |
672 | |
673 | head = &sbi->dir_inode_list; | |
127e670a JK |
674 | if (list_empty(head)) { |
675 | spin_unlock(&sbi->dir_inode_lock); | |
676 | return; | |
677 | } | |
678 | entry = list_entry(head->next, struct dir_inode_entry, list); | |
679 | inode = igrab(entry->inode); | |
680 | spin_unlock(&sbi->dir_inode_lock); | |
681 | if (inode) { | |
682 | filemap_flush(inode->i_mapping); | |
683 | iput(inode); | |
684 | } else { | |
685 | /* | |
686 | * We should submit bio, since it exists several | |
687 | * wribacking dentry pages in the freeing inode. | |
688 | */ | |
458e6197 | 689 | f2fs_submit_merged_bio(sbi, DATA, WRITE); |
127e670a JK |
690 | } |
691 | goto retry; | |
692 | } | |
693 | ||
0a8165d7 | 694 | /* |
127e670a JK |
695 | * Freeze all the FS-operations for checkpoint. |
696 | */ | |
43727527 | 697 | static void block_operations(struct f2fs_sb_info *sbi) |
127e670a | 698 | { |
127e670a JK |
699 | struct writeback_control wbc = { |
700 | .sync_mode = WB_SYNC_ALL, | |
701 | .nr_to_write = LONG_MAX, | |
702 | .for_reclaim = 0, | |
703 | }; | |
c718379b JK |
704 | struct blk_plug plug; |
705 | ||
706 | blk_start_plug(&plug); | |
707 | ||
39936837 | 708 | retry_flush_dents: |
e479556b | 709 | f2fs_lock_all(sbi); |
127e670a | 710 | /* write all the dirty dentry pages */ |
127e670a | 711 | if (get_pages(sbi, F2FS_DIRTY_DENTS)) { |
e479556b | 712 | f2fs_unlock_all(sbi); |
39936837 JK |
713 | sync_dirty_dir_inodes(sbi); |
714 | goto retry_flush_dents; | |
127e670a JK |
715 | } |
716 | ||
127e670a JK |
717 | /* |
718 | * POR: we should ensure that there is no dirty node pages | |
719 | * until finishing nat/sit flush. | |
720 | */ | |
39936837 JK |
721 | retry_flush_nodes: |
722 | mutex_lock(&sbi->node_write); | |
127e670a JK |
723 | |
724 | if (get_pages(sbi, F2FS_DIRTY_NODES)) { | |
39936837 JK |
725 | mutex_unlock(&sbi->node_write); |
726 | sync_node_pages(sbi, 0, &wbc); | |
727 | goto retry_flush_nodes; | |
127e670a | 728 | } |
c718379b | 729 | blk_finish_plug(&plug); |
127e670a JK |
730 | } |
731 | ||
732 | static void unblock_operations(struct f2fs_sb_info *sbi) | |
733 | { | |
39936837 | 734 | mutex_unlock(&sbi->node_write); |
e479556b | 735 | f2fs_unlock_all(sbi); |
127e670a JK |
736 | } |
737 | ||
fb51b5ef CL |
738 | static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi) |
739 | { | |
740 | DEFINE_WAIT(wait); | |
741 | ||
742 | for (;;) { | |
743 | prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE); | |
744 | ||
745 | if (!get_pages(sbi, F2FS_WRITEBACK)) | |
746 | break; | |
747 | ||
748 | io_schedule(); | |
749 | } | |
750 | finish_wait(&sbi->cp_wait, &wait); | |
751 | } | |
752 | ||
127e670a JK |
753 | static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount) |
754 | { | |
755 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
756 | nid_t last_nid = 0; | |
757 | block_t start_blk; | |
758 | struct page *cp_page; | |
759 | unsigned int data_sum_blocks, orphan_blocks; | |
7e586fa0 | 760 | __u32 crc32 = 0; |
127e670a | 761 | void *kaddr; |
127e670a JK |
762 | int i; |
763 | ||
764 | /* Flush all the NAT/SIT pages */ | |
765 | while (get_pages(sbi, F2FS_DIRTY_META)) | |
766 | sync_meta_pages(sbi, META, LONG_MAX); | |
767 | ||
768 | next_free_nid(sbi, &last_nid); | |
769 | ||
770 | /* | |
771 | * modify checkpoint | |
772 | * version number is already updated | |
773 | */ | |
774 | ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi)); | |
775 | ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi)); | |
776 | ckpt->free_segment_count = cpu_to_le32(free_segments(sbi)); | |
777 | for (i = 0; i < 3; i++) { | |
778 | ckpt->cur_node_segno[i] = | |
779 | cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE)); | |
780 | ckpt->cur_node_blkoff[i] = | |
781 | cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE)); | |
782 | ckpt->alloc_type[i + CURSEG_HOT_NODE] = | |
783 | curseg_alloc_type(sbi, i + CURSEG_HOT_NODE); | |
784 | } | |
785 | for (i = 0; i < 3; i++) { | |
786 | ckpt->cur_data_segno[i] = | |
787 | cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA)); | |
788 | ckpt->cur_data_blkoff[i] = | |
789 | cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA)); | |
790 | ckpt->alloc_type[i + CURSEG_HOT_DATA] = | |
791 | curseg_alloc_type(sbi, i + CURSEG_HOT_DATA); | |
792 | } | |
793 | ||
794 | ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi)); | |
795 | ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi)); | |
796 | ckpt->next_free_nid = cpu_to_le32(last_nid); | |
797 | ||
798 | /* 2 cp + n data seg summary + orphan inode blocks */ | |
799 | data_sum_blocks = npages_for_summary_flush(sbi); | |
800 | if (data_sum_blocks < 3) | |
25ca923b | 801 | set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); |
127e670a | 802 | else |
25ca923b | 803 | clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG); |
127e670a JK |
804 | |
805 | orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1) | |
806 | / F2FS_ORPHANS_PER_BLOCK; | |
25ca923b | 807 | ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks); |
127e670a JK |
808 | |
809 | if (is_umount) { | |
25ca923b JK |
810 | set_ckpt_flags(ckpt, CP_UMOUNT_FLAG); |
811 | ckpt->cp_pack_total_block_count = cpu_to_le32(2 + | |
812 | data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE); | |
127e670a | 813 | } else { |
25ca923b JK |
814 | clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG); |
815 | ckpt->cp_pack_total_block_count = cpu_to_le32(2 + | |
816 | data_sum_blocks + orphan_blocks); | |
127e670a JK |
817 | } |
818 | ||
819 | if (sbi->n_orphans) | |
25ca923b | 820 | set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); |
127e670a | 821 | else |
25ca923b | 822 | clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG); |
127e670a JK |
823 | |
824 | /* update SIT/NAT bitmap */ | |
825 | get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP)); | |
826 | get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP)); | |
827 | ||
828 | crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset)); | |
7e586fa0 JK |
829 | *((__le32 *)((unsigned char *)ckpt + |
830 | le32_to_cpu(ckpt->checksum_offset))) | |
127e670a JK |
831 | = cpu_to_le32(crc32); |
832 | ||
833 | start_blk = __start_cp_addr(sbi); | |
834 | ||
835 | /* write out checkpoint buffer at block 0 */ | |
836 | cp_page = grab_meta_page(sbi, start_blk++); | |
837 | kaddr = page_address(cp_page); | |
838 | memcpy(kaddr, ckpt, (1 << sbi->log_blocksize)); | |
839 | set_page_dirty(cp_page); | |
840 | f2fs_put_page(cp_page, 1); | |
841 | ||
842 | if (sbi->n_orphans) { | |
843 | write_orphan_inodes(sbi, start_blk); | |
844 | start_blk += orphan_blocks; | |
845 | } | |
846 | ||
847 | write_data_summaries(sbi, start_blk); | |
848 | start_blk += data_sum_blocks; | |
849 | if (is_umount) { | |
850 | write_node_summaries(sbi, start_blk); | |
851 | start_blk += NR_CURSEG_NODE_TYPE; | |
852 | } | |
853 | ||
854 | /* writeout checkpoint block */ | |
855 | cp_page = grab_meta_page(sbi, start_blk); | |
856 | kaddr = page_address(cp_page); | |
857 | memcpy(kaddr, ckpt, (1 << sbi->log_blocksize)); | |
858 | set_page_dirty(cp_page); | |
859 | f2fs_put_page(cp_page, 1); | |
860 | ||
861 | /* wait for previous submitted node/meta pages writeback */ | |
fb51b5ef | 862 | wait_on_all_pages_writeback(sbi); |
127e670a | 863 | |
4ef51a8f | 864 | filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX); |
9df27d98 | 865 | filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX); |
127e670a JK |
866 | |
867 | /* update user_block_counts */ | |
868 | sbi->last_valid_block_count = sbi->total_valid_block_count; | |
869 | sbi->alloc_valid_block_count = 0; | |
870 | ||
871 | /* Here, we only have one bio having CP pack */ | |
577e3495 | 872 | sync_meta_pages(sbi, META_FLUSH, LONG_MAX); |
127e670a | 873 | |
6bacf52f | 874 | if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) { |
577e3495 JK |
875 | clear_prefree_segments(sbi); |
876 | F2FS_RESET_SB_DIRT(sbi); | |
877 | } | |
127e670a JK |
878 | } |
879 | ||
0a8165d7 | 880 | /* |
127e670a JK |
881 | * We guarantee that this checkpoint procedure should not fail. |
882 | */ | |
43727527 | 883 | void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount) |
127e670a JK |
884 | { |
885 | struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); | |
886 | unsigned long long ckpt_ver; | |
887 | ||
2af4bd6c NJ |
888 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops"); |
889 | ||
43727527 JK |
890 | mutex_lock(&sbi->cp_mutex); |
891 | block_operations(sbi); | |
127e670a | 892 | |
2af4bd6c NJ |
893 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops"); |
894 | ||
458e6197 JK |
895 | f2fs_submit_merged_bio(sbi, DATA, WRITE); |
896 | f2fs_submit_merged_bio(sbi, NODE, WRITE); | |
897 | f2fs_submit_merged_bio(sbi, META, WRITE); | |
127e670a JK |
898 | |
899 | /* | |
900 | * update checkpoint pack index | |
901 | * Increase the version number so that | |
902 | * SIT entries and seg summaries are written at correct place | |
903 | */ | |
d71b5564 | 904 | ckpt_ver = cur_cp_version(ckpt); |
127e670a JK |
905 | ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver); |
906 | ||
907 | /* write cached NAT/SIT entries to NAT/SIT area */ | |
908 | flush_nat_entries(sbi); | |
909 | flush_sit_entries(sbi); | |
910 | ||
127e670a JK |
911 | /* unlock all the fs_lock[] in do_checkpoint() */ |
912 | do_checkpoint(sbi, is_umount); | |
913 | ||
914 | unblock_operations(sbi); | |
915 | mutex_unlock(&sbi->cp_mutex); | |
2af4bd6c | 916 | |
942e0be6 | 917 | stat_inc_cp_count(sbi->stat_info); |
2af4bd6c | 918 | trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint"); |
127e670a JK |
919 | } |
920 | ||
921 | void init_orphan_info(struct f2fs_sb_info *sbi) | |
922 | { | |
17b692f6 | 923 | spin_lock_init(&sbi->orphan_inode_lock); |
127e670a JK |
924 | INIT_LIST_HEAD(&sbi->orphan_inode_list); |
925 | sbi->n_orphans = 0; | |
0d47c1ad GZ |
926 | /* |
927 | * considering 512 blocks in a segment 8 blocks are needed for cp | |
928 | * and log segment summaries. Remaining blocks are used to keep | |
929 | * orphan entries with the limitation one reserved segment | |
930 | * for cp pack we can have max 1020*504 orphan entries | |
931 | */ | |
932 | sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE) | |
933 | * F2FS_ORPHANS_PER_BLOCK; | |
127e670a JK |
934 | } |
935 | ||
6e6093a8 | 936 | int __init create_checkpoint_caches(void) |
127e670a JK |
937 | { |
938 | orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry", | |
939 | sizeof(struct orphan_inode_entry), NULL); | |
6bacf52f | 940 | if (!orphan_entry_slab) |
127e670a JK |
941 | return -ENOMEM; |
942 | inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry", | |
943 | sizeof(struct dir_inode_entry), NULL); | |
6bacf52f | 944 | if (!inode_entry_slab) { |
127e670a JK |
945 | kmem_cache_destroy(orphan_entry_slab); |
946 | return -ENOMEM; | |
947 | } | |
948 | return 0; | |
949 | } | |
950 | ||
951 | void destroy_checkpoint_caches(void) | |
952 | { | |
953 | kmem_cache_destroy(orphan_entry_slab); | |
954 | kmem_cache_destroy(inode_entry_slab); | |
955 | } |