4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
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.
12 #include <linux/f2fs_fs.h>
18 * Roll forward recovery scenarios.
20 * [Term] F: fsync_mark, D: dentry_mark
22 * 1. inode(x) | CP | inode(x) | dnode(F)
23 * -> Update the latest inode(x).
25 * 2. inode(x) | CP | inode(F) | dnode(F)
28 * 3. inode(x) | CP | dnode(F) | inode(x)
29 * -> Recover to the latest dnode(F), and drop the last inode(x)
31 * 4. inode(x) | CP | dnode(F) | inode(F)
34 * 5. CP | inode(x) | dnode(F)
35 * -> The inode(DF) was missing. Should drop this dnode(F).
37 * 6. CP | inode(DF) | dnode(F)
40 * 7. CP | dnode(F) | inode(DF)
41 * -> If f2fs_iget fails, then goto next to find inode(DF).
43 * 8. CP | dnode(F) | inode(x)
44 * -> If f2fs_iget fails, then goto next to find inode(DF).
45 * But it will fail due to no inode(DF).
48 static struct kmem_cache
*fsync_entry_slab
;
50 bool space_for_roll_forward(struct f2fs_sb_info
*sbi
)
52 if (sbi
->last_valid_block_count
+ sbi
->alloc_valid_block_count
53 > sbi
->user_block_count
)
58 static struct fsync_inode_entry
*get_fsync_inode(struct list_head
*head
,
61 struct fsync_inode_entry
*entry
;
63 list_for_each_entry(entry
, head
, list
)
64 if (entry
->inode
->i_ino
== ino
)
70 static int recover_dentry(struct inode
*inode
, struct page
*ipage
)
72 struct f2fs_inode
*raw_inode
= F2FS_INODE(ipage
);
73 nid_t pino
= le32_to_cpu(raw_inode
->i_pino
);
74 struct f2fs_dir_entry
*de
;
77 struct inode
*dir
, *einode
;
80 dir
= f2fs_iget(inode
->i_sb
, pino
);
86 name
.len
= le32_to_cpu(raw_inode
->i_namelen
);
87 name
.name
= raw_inode
->i_name
;
89 if (unlikely(name
.len
> F2FS_NAME_LEN
)) {
95 de
= f2fs_find_entry(dir
, &name
, &page
);
96 if (de
&& inode
->i_ino
== le32_to_cpu(de
->ino
)) {
97 clear_inode_flag(F2FS_I(inode
), FI_INC_LINK
);
101 einode
= f2fs_iget(inode
->i_sb
, le32_to_cpu(de
->ino
));
102 if (IS_ERR(einode
)) {
104 err
= PTR_ERR(einode
);
109 err
= acquire_orphan_inode(F2FS_I_SB(inode
));
114 f2fs_delete_entry(de
, page
, einode
);
118 err
= __f2fs_add_link(dir
, &name
, inode
);
122 if (is_inode_flag_set(F2FS_I(dir
), FI_DELAY_IPUT
)) {
125 add_dirty_dir_inode(dir
);
126 set_inode_flag(F2FS_I(dir
), FI_DELAY_IPUT
);
133 f2fs_put_page(page
, 0);
137 f2fs_msg(inode
->i_sb
, KERN_NOTICE
,
138 "%s: ino = %x, name = %s, dir = %lx, err = %d",
139 __func__
, ino_of_node(ipage
), raw_inode
->i_name
,
140 IS_ERR(dir
) ? 0 : dir
->i_ino
, err
);
144 static void recover_inode(struct inode
*inode
, struct page
*page
)
146 struct f2fs_inode
*raw
= F2FS_INODE(page
);
148 inode
->i_mode
= le16_to_cpu(raw
->i_mode
);
149 i_size_write(inode
, le64_to_cpu(raw
->i_size
));
150 inode
->i_atime
.tv_sec
= le64_to_cpu(raw
->i_mtime
);
151 inode
->i_ctime
.tv_sec
= le64_to_cpu(raw
->i_ctime
);
152 inode
->i_mtime
.tv_sec
= le64_to_cpu(raw
->i_mtime
);
153 inode
->i_atime
.tv_nsec
= le32_to_cpu(raw
->i_mtime_nsec
);
154 inode
->i_ctime
.tv_nsec
= le32_to_cpu(raw
->i_ctime_nsec
);
155 inode
->i_mtime
.tv_nsec
= le32_to_cpu(raw
->i_mtime_nsec
);
157 f2fs_msg(inode
->i_sb
, KERN_NOTICE
, "recover_inode: ino = %x, name = %s",
158 ino_of_node(page
), F2FS_INODE(page
)->i_name
);
161 static int find_fsync_dnodes(struct f2fs_sb_info
*sbi
, struct list_head
*head
)
163 unsigned long long cp_ver
= cur_cp_version(F2FS_CKPT(sbi
));
164 struct curseg_info
*curseg
;
165 struct page
*page
= NULL
;
169 /* get node pages in the current segment */
170 curseg
= CURSEG_I(sbi
, CURSEG_WARM_NODE
);
171 blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
174 struct fsync_inode_entry
*entry
;
176 if (blkaddr
< MAIN_BLKADDR(sbi
) || blkaddr
>= MAX_BLKADDR(sbi
))
179 page
= get_meta_page_ra(sbi
, blkaddr
);
181 if (cp_ver
!= cpver_of_node(page
))
184 if (!is_fsync_dnode(page
))
187 entry
= get_fsync_inode(head
, ino_of_node(page
));
189 if (IS_INODE(page
) && is_dent_dnode(page
))
190 set_inode_flag(F2FS_I(entry
->inode
),
193 if (IS_INODE(page
) && is_dent_dnode(page
)) {
194 err
= recover_inode_page(sbi
, page
);
199 /* add this fsync inode to the list */
200 entry
= kmem_cache_alloc(fsync_entry_slab
, GFP_F2FS_ZERO
);
206 * CP | dnode(F) | inode(DF)
207 * For this case, we should not give up now.
209 entry
->inode
= f2fs_iget(sbi
->sb
, ino_of_node(page
));
210 if (IS_ERR(entry
->inode
)) {
211 err
= PTR_ERR(entry
->inode
);
212 kmem_cache_free(fsync_entry_slab
, entry
);
217 list_add_tail(&entry
->list
, head
);
219 entry
->blkaddr
= blkaddr
;
221 if (IS_INODE(page
)) {
222 entry
->last_inode
= blkaddr
;
223 if (is_dent_dnode(page
))
224 entry
->last_dentry
= blkaddr
;
227 /* check next segment */
228 blkaddr
= next_blkaddr_of_node(page
);
229 f2fs_put_page(page
, 1);
231 f2fs_put_page(page
, 1);
235 static void destroy_fsync_dnodes(struct list_head
*head
)
237 struct fsync_inode_entry
*entry
, *tmp
;
239 list_for_each_entry_safe(entry
, tmp
, head
, list
) {
241 list_del(&entry
->list
);
242 kmem_cache_free(fsync_entry_slab
, entry
);
246 static int check_index_in_prev_nodes(struct f2fs_sb_info
*sbi
,
247 block_t blkaddr
, struct dnode_of_data
*dn
)
249 struct seg_entry
*sentry
;
250 unsigned int segno
= GET_SEGNO(sbi
, blkaddr
);
251 unsigned short blkoff
= GET_BLKOFF_FROM_SEG0(sbi
, blkaddr
);
252 struct f2fs_summary_block
*sum_node
;
253 struct f2fs_summary sum
;
254 struct page
*sum_page
, *node_page
;
261 sentry
= get_seg_entry(sbi
, segno
);
262 if (!f2fs_test_bit(blkoff
, sentry
->cur_valid_map
))
265 /* Get the previous summary */
266 for (i
= CURSEG_WARM_DATA
; i
<= CURSEG_COLD_DATA
; i
++) {
267 struct curseg_info
*curseg
= CURSEG_I(sbi
, i
);
268 if (curseg
->segno
== segno
) {
269 sum
= curseg
->sum_blk
->entries
[blkoff
];
274 sum_page
= get_sum_page(sbi
, segno
);
275 sum_node
= (struct f2fs_summary_block
*)page_address(sum_page
);
276 sum
= sum_node
->entries
[blkoff
];
277 f2fs_put_page(sum_page
, 1);
279 /* Use the locked dnode page and inode */
280 nid
= le32_to_cpu(sum
.nid
);
281 if (dn
->inode
->i_ino
== nid
) {
282 struct dnode_of_data tdn
= *dn
;
284 tdn
.node_page
= dn
->inode_page
;
285 tdn
.ofs_in_node
= le16_to_cpu(sum
.ofs_in_node
);
286 truncate_data_blocks_range(&tdn
, 1);
288 } else if (dn
->nid
== nid
) {
289 struct dnode_of_data tdn
= *dn
;
290 tdn
.ofs_in_node
= le16_to_cpu(sum
.ofs_in_node
);
291 truncate_data_blocks_range(&tdn
, 1);
295 /* Get the node page */
296 node_page
= get_node_page(sbi
, nid
);
297 if (IS_ERR(node_page
))
298 return PTR_ERR(node_page
);
300 offset
= ofs_of_node(node_page
);
301 ino
= ino_of_node(node_page
);
302 f2fs_put_page(node_page
, 1);
304 if (ino
!= dn
->inode
->i_ino
) {
305 /* Deallocate previous index in the node page */
306 inode
= f2fs_iget(sbi
->sb
, ino
);
308 return PTR_ERR(inode
);
313 bidx
= start_bidx_of_node(offset
, F2FS_I(inode
)) +
314 le16_to_cpu(sum
.ofs_in_node
);
316 if (ino
!= dn
->inode
->i_ino
) {
317 truncate_hole(inode
, bidx
, bidx
+ 1);
320 struct dnode_of_data tdn
;
321 set_new_dnode(&tdn
, inode
, dn
->inode_page
, NULL
, 0);
322 if (get_dnode_of_data(&tdn
, bidx
, LOOKUP_NODE
))
324 if (tdn
.data_blkaddr
!= NULL_ADDR
)
325 truncate_data_blocks_range(&tdn
, 1);
326 f2fs_put_page(tdn
.node_page
, 1);
331 static int do_recover_data(struct f2fs_sb_info
*sbi
, struct inode
*inode
,
332 struct page
*page
, block_t blkaddr
)
334 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
335 unsigned int start
, end
;
336 struct dnode_of_data dn
;
337 struct f2fs_summary sum
;
339 int err
= 0, recovered
= 0;
341 /* step 1: recover xattr */
342 if (IS_INODE(page
)) {
343 recover_inline_xattr(inode
, page
);
344 } else if (f2fs_has_xattr_block(ofs_of_node(page
))) {
345 recover_xattr_data(inode
, page
, blkaddr
);
349 /* step 2: recover inline data */
350 if (recover_inline_data(inode
, page
))
353 /* step 3: recover data indices */
354 start
= start_bidx_of_node(ofs_of_node(page
), fi
);
355 end
= start
+ ADDRS_PER_PAGE(page
, fi
);
359 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
361 err
= get_dnode_of_data(&dn
, start
, ALLOC_NODE
);
367 f2fs_wait_on_page_writeback(dn
.node_page
, NODE
);
369 get_node_info(sbi
, dn
.nid
, &ni
);
370 f2fs_bug_on(sbi
, ni
.ino
!= ino_of_node(page
));
371 f2fs_bug_on(sbi
, ofs_of_node(dn
.node_page
) != ofs_of_node(page
));
373 for (; start
< end
; start
++) {
376 src
= datablock_addr(dn
.node_page
, dn
.ofs_in_node
);
377 dest
= datablock_addr(page
, dn
.ofs_in_node
);
379 if (src
!= dest
&& dest
!= NEW_ADDR
&& dest
!= NULL_ADDR
) {
380 if (src
== NULL_ADDR
) {
381 err
= reserve_new_block(&dn
);
382 /* We should not get -ENOSPC */
383 f2fs_bug_on(sbi
, err
);
386 /* Check the previous node page having this index */
387 err
= check_index_in_prev_nodes(sbi
, dest
, &dn
);
391 set_summary(&sum
, dn
.nid
, dn
.ofs_in_node
, ni
.version
);
393 /* write dummy data page */
394 recover_data_page(sbi
, NULL
, &sum
, src
, dest
);
395 update_extent_cache(dest
, &dn
);
401 /* write node page in place */
402 set_summary(&sum
, dn
.nid
, 0, 0);
403 if (IS_INODE(dn
.node_page
))
404 sync_inode_page(&dn
);
406 copy_node_footer(dn
.node_page
, page
);
407 fill_node_footer(dn
.node_page
, dn
.nid
, ni
.ino
,
408 ofs_of_node(page
), false);
409 set_page_dirty(dn
.node_page
);
414 f2fs_msg(sbi
->sb
, KERN_NOTICE
,
415 "recover_data: ino = %lx, recovered = %d blocks, err = %d",
416 inode
->i_ino
, recovered
, err
);
420 static int recover_data(struct f2fs_sb_info
*sbi
,
421 struct list_head
*head
, int type
)
423 unsigned long long cp_ver
= cur_cp_version(F2FS_CKPT(sbi
));
424 struct curseg_info
*curseg
;
425 struct page
*page
= NULL
;
429 /* get node pages in the current segment */
430 curseg
= CURSEG_I(sbi
, type
);
431 blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
434 struct fsync_inode_entry
*entry
;
436 if (blkaddr
< MAIN_BLKADDR(sbi
) || blkaddr
>= MAX_BLKADDR(sbi
))
439 page
= get_meta_page_ra(sbi
, blkaddr
);
441 if (cp_ver
!= cpver_of_node(page
)) {
442 f2fs_put_page(page
, 1);
446 entry
= get_fsync_inode(head
, ino_of_node(page
));
450 * inode(x) | CP | inode(x) | dnode(F)
451 * In this case, we can lose the latest inode(x).
452 * So, call recover_inode for the inode update.
454 if (entry
->last_inode
== blkaddr
)
455 recover_inode(entry
->inode
, page
);
456 if (entry
->last_dentry
== blkaddr
) {
457 err
= recover_dentry(entry
->inode
, page
);
459 f2fs_put_page(page
, 1);
463 err
= do_recover_data(sbi
, entry
->inode
, page
, blkaddr
);
465 f2fs_put_page(page
, 1);
469 if (entry
->blkaddr
== blkaddr
) {
471 list_del(&entry
->list
);
472 kmem_cache_free(fsync_entry_slab
, entry
);
475 /* check next segment */
476 blkaddr
= next_blkaddr_of_node(page
);
477 f2fs_put_page(page
, 1);
480 allocate_new_segments(sbi
);
484 int recover_fsync_data(struct f2fs_sb_info
*sbi
)
486 struct curseg_info
*curseg
= CURSEG_I(sbi
, CURSEG_WARM_NODE
);
487 struct list_head inode_list
;
490 bool need_writecp
= false;
492 fsync_entry_slab
= f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
493 sizeof(struct fsync_inode_entry
));
494 if (!fsync_entry_slab
)
497 INIT_LIST_HEAD(&inode_list
);
499 /* step #1: find fsynced inode numbers */
500 sbi
->por_doing
= true;
502 /* prevent checkpoint */
503 mutex_lock(&sbi
->cp_mutex
);
505 blkaddr
= NEXT_FREE_BLKADDR(sbi
, curseg
);
507 err
= find_fsync_dnodes(sbi
, &inode_list
);
511 if (list_empty(&inode_list
))
516 /* step #2: recover data */
517 err
= recover_data(sbi
, &inode_list
, CURSEG_WARM_NODE
);
519 f2fs_bug_on(sbi
, !list_empty(&inode_list
));
521 destroy_fsync_dnodes(&inode_list
);
522 kmem_cache_destroy(fsync_entry_slab
);
524 /* truncate meta pages to be used by the recovery */
525 truncate_inode_pages_range(META_MAPPING(sbi
),
526 MAIN_BLKADDR(sbi
) << PAGE_CACHE_SHIFT
, -1);
529 truncate_inode_pages_final(NODE_MAPPING(sbi
));
530 truncate_inode_pages_final(META_MAPPING(sbi
));
533 sbi
->por_doing
= false;
535 discard_next_dnode(sbi
, blkaddr
);
537 /* Flush all the NAT/SIT pages */
538 while (get_pages(sbi
, F2FS_DIRTY_META
))
539 sync_meta_pages(sbi
, META
, LONG_MAX
);
540 set_ckpt_flags(sbi
->ckpt
, CP_ERROR_FLAG
);
541 mutex_unlock(&sbi
->cp_mutex
);
542 } else if (need_writecp
) {
543 struct cp_control cpc
= {
546 mutex_unlock(&sbi
->cp_mutex
);
547 write_checkpoint(sbi
, &cpc
);
549 mutex_unlock(&sbi
->cp_mutex
);
This page took 0.045307 seconds and 6 git commands to generate.