nilfs2: remove header file for segment list operations
[linux-2.6.git] / fs / nilfs2 / recovery.c
1 /*
2  * recovery.c - NILFS recovery logic
3  *
4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
19  *
20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
21  */
22
23 #include <linux/buffer_head.h>
24 #include <linux/blkdev.h>
25 #include <linux/swap.h>
26 #include <linux/crc32.h>
27 #include "nilfs.h"
28 #include "segment.h"
29 #include "sufile.h"
30 #include "page.h"
31 #include "segbuf.h"
32
33 /*
34  * Segment check result
35  */
36 enum {
37         NILFS_SEG_VALID,
38         NILFS_SEG_NO_SUPER_ROOT,
39         NILFS_SEG_FAIL_IO,
40         NILFS_SEG_FAIL_MAGIC,
41         NILFS_SEG_FAIL_SEQ,
42         NILFS_SEG_FAIL_CHECKSUM_SEGSUM,
43         NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT,
44         NILFS_SEG_FAIL_CHECKSUM_FULL,
45         NILFS_SEG_FAIL_CONSISTENCY,
46 };
47
48 /* work structure for recovery */
49 struct nilfs_recovery_block {
50         ino_t ino;              /* Inode number of the file that this block
51                                    belongs to */
52         sector_t blocknr;       /* block number */
53         __u64 vblocknr;         /* virtual block number */
54         unsigned long blkoff;   /* File offset of the data block (per block) */
55         struct list_head list;
56 };
57
58
59 static int nilfs_warn_segment_error(int err)
60 {
61         switch (err) {
62         case NILFS_SEG_FAIL_IO:
63                 printk(KERN_WARNING
64                        "NILFS warning: I/O error on loading last segment\n");
65                 return -EIO;
66         case NILFS_SEG_FAIL_MAGIC:
67                 printk(KERN_WARNING
68                        "NILFS warning: Segment magic number invalid\n");
69                 break;
70         case NILFS_SEG_FAIL_SEQ:
71                 printk(KERN_WARNING
72                        "NILFS warning: Sequence number mismatch\n");
73                 break;
74         case NILFS_SEG_FAIL_CHECKSUM_SEGSUM:
75                 printk(KERN_WARNING
76                        "NILFS warning: Checksum error in segment summary\n");
77                 break;
78         case NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT:
79                 printk(KERN_WARNING
80                        "NILFS warning: Checksum error in super root\n");
81                 break;
82         case NILFS_SEG_FAIL_CHECKSUM_FULL:
83                 printk(KERN_WARNING
84                        "NILFS warning: Checksum error in segment payload\n");
85                 break;
86         case NILFS_SEG_FAIL_CONSISTENCY:
87                 printk(KERN_WARNING
88                        "NILFS warning: Inconsistent segment\n");
89                 break;
90         case NILFS_SEG_NO_SUPER_ROOT:
91                 printk(KERN_WARNING
92                        "NILFS warning: No super root in the last segment\n");
93                 break;
94         }
95         return -EINVAL;
96 }
97
98 static void store_segsum_info(struct nilfs_segsum_info *ssi,
99                               struct nilfs_segment_summary *sum,
100                               unsigned int blocksize)
101 {
102         ssi->flags = le16_to_cpu(sum->ss_flags);
103         ssi->seg_seq = le64_to_cpu(sum->ss_seq);
104         ssi->ctime = le64_to_cpu(sum->ss_create);
105         ssi->next = le64_to_cpu(sum->ss_next);
106         ssi->nblocks = le32_to_cpu(sum->ss_nblocks);
107         ssi->nfinfo = le32_to_cpu(sum->ss_nfinfo);
108         ssi->sumbytes = le32_to_cpu(sum->ss_sumbytes);
109
110         ssi->nsumblk = DIV_ROUND_UP(ssi->sumbytes, blocksize);
111         ssi->nfileblk = ssi->nblocks - ssi->nsumblk - !!NILFS_SEG_HAS_SR(ssi);
112 }
113
114 /**
115  * calc_crc_cont - check CRC of blocks continuously
116  * @sbi: nilfs_sb_info
117  * @bhs: buffer head of start block
118  * @sum: place to store result
119  * @offset: offset bytes in the first block
120  * @check_bytes: number of bytes to be checked
121  * @start: DBN of start block
122  * @nblock: number of blocks to be checked
123  */
124 static int calc_crc_cont(struct nilfs_sb_info *sbi, struct buffer_head *bhs,
125                          u32 *sum, unsigned long offset, u64 check_bytes,
126                          sector_t start, unsigned long nblock)
127 {
128         unsigned long blocksize = sbi->s_super->s_blocksize;
129         unsigned long size;
130         u32 crc;
131
132         BUG_ON(offset >= blocksize);
133         check_bytes -= offset;
134         size = min_t(u64, check_bytes, blocksize - offset);
135         crc = crc32_le(sbi->s_nilfs->ns_crc_seed,
136                        (unsigned char *)bhs->b_data + offset, size);
137         if (--nblock > 0) {
138                 do {
139                         struct buffer_head *bh
140                                 = sb_bread(sbi->s_super, ++start);
141                         if (!bh)
142                                 return -EIO;
143                         check_bytes -= size;
144                         size = min_t(u64, check_bytes, blocksize);
145                         crc = crc32_le(crc, bh->b_data, size);
146                         brelse(bh);
147                 } while (--nblock > 0);
148         }
149         *sum = crc;
150         return 0;
151 }
152
153 /**
154  * nilfs_read_super_root_block - read super root block
155  * @sb: super_block
156  * @sr_block: disk block number of the super root block
157  * @pbh: address of a buffer_head pointer to return super root buffer
158  * @check: CRC check flag
159  */
160 int nilfs_read_super_root_block(struct super_block *sb, sector_t sr_block,
161                                 struct buffer_head **pbh, int check)
162 {
163         struct buffer_head *bh_sr;
164         struct nilfs_super_root *sr;
165         u32 crc;
166         int ret;
167
168         *pbh = NULL;
169         bh_sr = sb_bread(sb, sr_block);
170         if (unlikely(!bh_sr)) {
171                 ret = NILFS_SEG_FAIL_IO;
172                 goto failed;
173         }
174
175         sr = (struct nilfs_super_root *)bh_sr->b_data;
176         if (check) {
177                 unsigned bytes = le16_to_cpu(sr->sr_bytes);
178
179                 if (bytes == 0 || bytes > sb->s_blocksize) {
180                         ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
181                         goto failed_bh;
182                 }
183                 if (calc_crc_cont(NILFS_SB(sb), bh_sr, &crc,
184                                   sizeof(sr->sr_sum), bytes, sr_block, 1)) {
185                         ret = NILFS_SEG_FAIL_IO;
186                         goto failed_bh;
187                 }
188                 if (crc != le32_to_cpu(sr->sr_sum)) {
189                         ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
190                         goto failed_bh;
191                 }
192         }
193         *pbh = bh_sr;
194         return 0;
195
196  failed_bh:
197         brelse(bh_sr);
198
199  failed:
200         return nilfs_warn_segment_error(ret);
201 }
202
203 /**
204  * load_segment_summary - read segment summary of the specified partial segment
205  * @sbi: nilfs_sb_info
206  * @pseg_start: start disk block number of partial segment
207  * @seg_seq: sequence number requested
208  * @ssi: pointer to nilfs_segsum_info struct to store information
209  * @full_check: full check flag
210  *              (0: only checks segment summary CRC, 1: data CRC)
211  */
212 static int
213 load_segment_summary(struct nilfs_sb_info *sbi, sector_t pseg_start,
214                      u64 seg_seq, struct nilfs_segsum_info *ssi,
215                      int full_check)
216 {
217         struct buffer_head *bh_sum;
218         struct nilfs_segment_summary *sum;
219         unsigned long offset, nblock;
220         u64 check_bytes;
221         u32 crc, crc_sum;
222         int ret = NILFS_SEG_FAIL_IO;
223
224         bh_sum = sb_bread(sbi->s_super, pseg_start);
225         if (!bh_sum)
226                 goto out;
227
228         sum = (struct nilfs_segment_summary *)bh_sum->b_data;
229
230         /* Check consistency of segment summary */
231         if (le32_to_cpu(sum->ss_magic) != NILFS_SEGSUM_MAGIC) {
232                 ret = NILFS_SEG_FAIL_MAGIC;
233                 goto failed;
234         }
235         store_segsum_info(ssi, sum, sbi->s_super->s_blocksize);
236         if (seg_seq != ssi->seg_seq) {
237                 ret = NILFS_SEG_FAIL_SEQ;
238                 goto failed;
239         }
240         if (full_check) {
241                 offset = sizeof(sum->ss_datasum);
242                 check_bytes =
243                         ((u64)ssi->nblocks << sbi->s_super->s_blocksize_bits);
244                 nblock = ssi->nblocks;
245                 crc_sum = le32_to_cpu(sum->ss_datasum);
246                 ret = NILFS_SEG_FAIL_CHECKSUM_FULL;
247         } else { /* only checks segment summary */
248                 offset = sizeof(sum->ss_datasum) + sizeof(sum->ss_sumsum);
249                 check_bytes = ssi->sumbytes;
250                 nblock = ssi->nsumblk;
251                 crc_sum = le32_to_cpu(sum->ss_sumsum);
252                 ret = NILFS_SEG_FAIL_CHECKSUM_SEGSUM;
253         }
254
255         if (unlikely(nblock == 0 ||
256                      nblock > sbi->s_nilfs->ns_blocks_per_segment)) {
257                 /* This limits the number of blocks read in the CRC check */
258                 ret = NILFS_SEG_FAIL_CONSISTENCY;
259                 goto failed;
260         }
261         if (calc_crc_cont(sbi, bh_sum, &crc, offset, check_bytes,
262                           pseg_start, nblock)) {
263                 ret = NILFS_SEG_FAIL_IO;
264                 goto failed;
265         }
266         if (crc == crc_sum)
267                 ret = 0;
268  failed:
269         brelse(bh_sum);
270  out:
271         return ret;
272 }
273
274 static void *segsum_get(struct super_block *sb, struct buffer_head **pbh,
275                         unsigned int *offset, unsigned int bytes)
276 {
277         void *ptr;
278         sector_t blocknr;
279
280         BUG_ON((*pbh)->b_size < *offset);
281         if (bytes > (*pbh)->b_size - *offset) {
282                 blocknr = (*pbh)->b_blocknr;
283                 brelse(*pbh);
284                 *pbh = sb_bread(sb, blocknr + 1);
285                 if (unlikely(!*pbh))
286                         return NULL;
287                 *offset = 0;
288         }
289         ptr = (*pbh)->b_data + *offset;
290         *offset += bytes;
291         return ptr;
292 }
293
294 static void segsum_skip(struct super_block *sb, struct buffer_head **pbh,
295                         unsigned int *offset, unsigned int bytes,
296                         unsigned long count)
297 {
298         unsigned int rest_item_in_current_block
299                 = ((*pbh)->b_size - *offset) / bytes;
300
301         if (count <= rest_item_in_current_block) {
302                 *offset += bytes * count;
303         } else {
304                 sector_t blocknr = (*pbh)->b_blocknr;
305                 unsigned int nitem_per_block = (*pbh)->b_size / bytes;
306                 unsigned int bcnt;
307
308                 count -= rest_item_in_current_block;
309                 bcnt = DIV_ROUND_UP(count, nitem_per_block);
310                 *offset = bytes * (count - (bcnt - 1) * nitem_per_block);
311
312                 brelse(*pbh);
313                 *pbh = sb_bread(sb, blocknr + bcnt);
314         }
315 }
316
317 static int
318 collect_blocks_from_segsum(struct nilfs_sb_info *sbi, sector_t sum_blocknr,
319                            struct nilfs_segsum_info *ssi,
320                            struct list_head *head)
321 {
322         struct buffer_head *bh;
323         unsigned int offset;
324         unsigned long nfinfo = ssi->nfinfo;
325         sector_t blocknr = sum_blocknr + ssi->nsumblk;
326         ino_t ino;
327         int err = -EIO;
328
329         if (!nfinfo)
330                 return 0;
331
332         bh = sb_bread(sbi->s_super, sum_blocknr);
333         if (unlikely(!bh))
334                 goto out;
335
336         offset = le16_to_cpu(
337                 ((struct nilfs_segment_summary *)bh->b_data)->ss_bytes);
338         for (;;) {
339                 unsigned long nblocks, ndatablk, nnodeblk;
340                 struct nilfs_finfo *finfo;
341
342                 finfo = segsum_get(sbi->s_super, &bh, &offset, sizeof(*finfo));
343                 if (unlikely(!finfo))
344                         goto out;
345
346                 ino = le64_to_cpu(finfo->fi_ino);
347                 nblocks = le32_to_cpu(finfo->fi_nblocks);
348                 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
349                 nnodeblk = nblocks - ndatablk;
350
351                 while (ndatablk-- > 0) {
352                         struct nilfs_recovery_block *rb;
353                         struct nilfs_binfo_v *binfo;
354
355                         binfo = segsum_get(sbi->s_super, &bh, &offset,
356                                            sizeof(*binfo));
357                         if (unlikely(!binfo))
358                                 goto out;
359
360                         rb = kmalloc(sizeof(*rb), GFP_NOFS);
361                         if (unlikely(!rb)) {
362                                 err = -ENOMEM;
363                                 goto out;
364                         }
365                         rb->ino = ino;
366                         rb->blocknr = blocknr++;
367                         rb->vblocknr = le64_to_cpu(binfo->bi_vblocknr);
368                         rb->blkoff = le64_to_cpu(binfo->bi_blkoff);
369                         /* INIT_LIST_HEAD(&rb->list); */
370                         list_add_tail(&rb->list, head);
371                 }
372                 if (--nfinfo == 0)
373                         break;
374                 blocknr += nnodeblk; /* always 0 for the data sync segments */
375                 segsum_skip(sbi->s_super, &bh, &offset, sizeof(__le64),
376                             nnodeblk);
377                 if (unlikely(!bh))
378                         goto out;
379         }
380         err = 0;
381  out:
382         brelse(bh);   /* brelse(NULL) is just ignored */
383         return err;
384 }
385
386 static void dispose_recovery_list(struct list_head *head)
387 {
388         while (!list_empty(head)) {
389                 struct nilfs_recovery_block *rb
390                         = list_entry(head->next,
391                                      struct nilfs_recovery_block, list);
392                 list_del(&rb->list);
393                 kfree(rb);
394         }
395 }
396
397 struct nilfs_segment_entry {
398         struct list_head        list;
399         __u64                   segnum;
400 };
401
402 static int nilfs_segment_list_add(struct list_head *head, __u64 segnum)
403 {
404         struct nilfs_segment_entry *ent = kmalloc(sizeof(*ent), GFP_NOFS);
405
406         if (unlikely(!ent))
407                 return -ENOMEM;
408
409         ent->segnum = segnum;
410         INIT_LIST_HEAD(&ent->list);
411         list_add_tail(&ent->list, head);
412         return 0;
413 }
414
415 void nilfs_dispose_segment_list(struct list_head *head)
416 {
417         while (!list_empty(head)) {
418                 struct nilfs_segment_entry *ent
419                         = list_entry(head->next,
420                                      struct nilfs_segment_entry, list);
421                 list_del(&ent->list);
422                 kfree(ent);
423         }
424 }
425
426 static int nilfs_prepare_segment_for_recovery(struct the_nilfs *nilfs,
427                                               struct nilfs_sb_info *sbi,
428                                               struct nilfs_recovery_info *ri)
429 {
430         struct list_head *head = &ri->ri_used_segments;
431         struct nilfs_segment_entry *ent, *n;
432         struct inode *sufile = nilfs->ns_sufile;
433         __u64 segnum[4];
434         int err;
435         int i;
436
437         segnum[0] = nilfs->ns_segnum;
438         segnum[1] = nilfs->ns_nextnum;
439         segnum[2] = ri->ri_segnum;
440         segnum[3] = ri->ri_nextnum;
441
442         nilfs_attach_writer(nilfs, sbi);
443         /*
444          * Releasing the next segment of the latest super root.
445          * The next segment is invalidated by this recovery.
446          */
447         err = nilfs_sufile_free(sufile, segnum[1]);
448         if (unlikely(err))
449                 goto failed;
450
451         for (i = 1; i < 4; i++) {
452                 err = nilfs_segment_list_add(head, segnum[i]);
453                 if (unlikely(err))
454                         goto failed;
455         }
456
457         /*
458          * Collecting segments written after the latest super root.
459          * These are marked dirty to avoid being reallocated in the next write.
460          */
461         list_for_each_entry_safe(ent, n, head, list) {
462                 if (ent->segnum != segnum[0]) {
463                         err = nilfs_sufile_scrap(sufile, ent->segnum);
464                         if (unlikely(err))
465                                 goto failed;
466                 }
467                 list_del(&ent->list);
468                 kfree(ent);
469         }
470
471         /* Allocate new segments for recovery */
472         err = nilfs_sufile_alloc(sufile, &segnum[0]);
473         if (unlikely(err))
474                 goto failed;
475
476         nilfs->ns_pseg_offset = 0;
477         nilfs->ns_seg_seq = ri->ri_seq + 2;
478         nilfs->ns_nextnum = nilfs->ns_segnum = segnum[0];
479
480  failed:
481         /* No need to recover sufile because it will be destroyed on error */
482         nilfs_detach_writer(nilfs, sbi);
483         return err;
484 }
485
486 static int nilfs_recovery_copy_block(struct nilfs_sb_info *sbi,
487                                      struct nilfs_recovery_block *rb,
488                                      struct page *page)
489 {
490         struct buffer_head *bh_org;
491         void *kaddr;
492
493         bh_org = sb_bread(sbi->s_super, rb->blocknr);
494         if (unlikely(!bh_org))
495                 return -EIO;
496
497         kaddr = kmap_atomic(page, KM_USER0);
498         memcpy(kaddr + bh_offset(bh_org), bh_org->b_data, bh_org->b_size);
499         kunmap_atomic(kaddr, KM_USER0);
500         brelse(bh_org);
501         return 0;
502 }
503
504 static int recover_dsync_blocks(struct nilfs_sb_info *sbi,
505                                 struct list_head *head,
506                                 unsigned long *nr_salvaged_blocks)
507 {
508         struct inode *inode;
509         struct nilfs_recovery_block *rb, *n;
510         unsigned blocksize = sbi->s_super->s_blocksize;
511         struct page *page;
512         loff_t pos;
513         int err = 0, err2 = 0;
514
515         list_for_each_entry_safe(rb, n, head, list) {
516                 inode = nilfs_iget(sbi->s_super, rb->ino);
517                 if (IS_ERR(inode)) {
518                         err = PTR_ERR(inode);
519                         inode = NULL;
520                         goto failed_inode;
521                 }
522
523                 pos = rb->blkoff << inode->i_blkbits;
524                 page = NULL;
525                 err = block_write_begin(NULL, inode->i_mapping, pos, blocksize,
526                                         0, &page, NULL, nilfs_get_block);
527                 if (unlikely(err))
528                         goto failed_inode;
529
530                 err = nilfs_recovery_copy_block(sbi, rb, page);
531                 if (unlikely(err))
532                         goto failed_page;
533
534                 err = nilfs_set_file_dirty(sbi, inode, 1);
535                 if (unlikely(err))
536                         goto failed_page;
537
538                 block_write_end(NULL, inode->i_mapping, pos, blocksize,
539                                 blocksize, page, NULL);
540
541                 unlock_page(page);
542                 page_cache_release(page);
543
544                 (*nr_salvaged_blocks)++;
545                 goto next;
546
547  failed_page:
548                 unlock_page(page);
549                 page_cache_release(page);
550
551  failed_inode:
552                 printk(KERN_WARNING
553                        "NILFS warning: error recovering data block "
554                        "(err=%d, ino=%lu, block-offset=%llu)\n",
555                        err, rb->ino, (unsigned long long)rb->blkoff);
556                 if (!err2)
557                         err2 = err;
558  next:
559                 iput(inode); /* iput(NULL) is just ignored */
560                 list_del_init(&rb->list);
561                 kfree(rb);
562         }
563         return err2;
564 }
565
566 /**
567  * nilfs_do_roll_forward - salvage logical segments newer than the latest
568  * checkpoint
569  * @sbi: nilfs_sb_info
570  * @nilfs: the_nilfs
571  * @ri: pointer to a nilfs_recovery_info
572  */
573 static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
574                                  struct nilfs_sb_info *sbi,
575                                  struct nilfs_recovery_info *ri)
576 {
577         struct nilfs_segsum_info ssi;
578         sector_t pseg_start;
579         sector_t seg_start, seg_end;  /* Starting/ending DBN of full segment */
580         unsigned long nsalvaged_blocks = 0;
581         u64 seg_seq;
582         __u64 segnum, nextnum = 0;
583         int empty_seg = 0;
584         int err = 0, ret;
585         LIST_HEAD(dsync_blocks);  /* list of data blocks to be recovered */
586         enum {
587                 RF_INIT_ST,
588                 RF_DSYNC_ST,   /* scanning data-sync segments */
589         };
590         int state = RF_INIT_ST;
591
592         nilfs_attach_writer(nilfs, sbi);
593         pseg_start = ri->ri_lsegs_start;
594         seg_seq = ri->ri_lsegs_start_seq;
595         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
596         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
597
598         while (segnum != ri->ri_segnum || pseg_start <= ri->ri_pseg_start) {
599
600                 ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi, 1);
601                 if (ret) {
602                         if (ret == NILFS_SEG_FAIL_IO) {
603                                 err = -EIO;
604                                 goto failed;
605                         }
606                         goto strayed;
607                 }
608                 if (unlikely(NILFS_SEG_HAS_SR(&ssi)))
609                         goto confused;
610
611                 /* Found a valid partial segment; do recovery actions */
612                 nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next);
613                 empty_seg = 0;
614                 nilfs->ns_ctime = ssi.ctime;
615                 if (!(ssi.flags & NILFS_SS_GC))
616                         nilfs->ns_nongc_ctime = ssi.ctime;
617
618                 switch (state) {
619                 case RF_INIT_ST:
620                         if (!NILFS_SEG_LOGBGN(&ssi) || !NILFS_SEG_DSYNC(&ssi))
621                                 goto try_next_pseg;
622                         state = RF_DSYNC_ST;
623                         /* Fall through */
624                 case RF_DSYNC_ST:
625                         if (!NILFS_SEG_DSYNC(&ssi))
626                                 goto confused;
627
628                         err = collect_blocks_from_segsum(
629                                 sbi, pseg_start, &ssi, &dsync_blocks);
630                         if (unlikely(err))
631                                 goto failed;
632                         if (NILFS_SEG_LOGEND(&ssi)) {
633                                 err = recover_dsync_blocks(
634                                         sbi, &dsync_blocks, &nsalvaged_blocks);
635                                 if (unlikely(err))
636                                         goto failed;
637                                 state = RF_INIT_ST;
638                         }
639                         break; /* Fall through to try_next_pseg */
640                 }
641
642  try_next_pseg:
643                 if (pseg_start == ri->ri_lsegs_end)
644                         break;
645                 pseg_start += ssi.nblocks;
646                 if (pseg_start < seg_end)
647                         continue;
648                 goto feed_segment;
649
650  strayed:
651                 if (pseg_start == ri->ri_lsegs_end)
652                         break;
653
654  feed_segment:
655                 /* Looking to the next full segment */
656                 if (empty_seg++)
657                         break;
658                 seg_seq++;
659                 segnum = nextnum;
660                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
661                 pseg_start = seg_start;
662         }
663
664         if (nsalvaged_blocks) {
665                 printk(KERN_INFO "NILFS (device %s): salvaged %lu blocks\n",
666                        sbi->s_super->s_id, nsalvaged_blocks);
667                 ri->ri_need_recovery = NILFS_RECOVERY_ROLLFORWARD_DONE;
668         }
669  out:
670         dispose_recovery_list(&dsync_blocks);
671         nilfs_detach_writer(sbi->s_nilfs, sbi);
672         return err;
673
674  confused:
675         err = -EINVAL;
676  failed:
677         printk(KERN_ERR
678                "NILFS (device %s): Error roll-forwarding "
679                "(err=%d, pseg block=%llu). ",
680                sbi->s_super->s_id, err, (unsigned long long)pseg_start);
681         goto out;
682 }
683
684 static void nilfs_finish_roll_forward(struct the_nilfs *nilfs,
685                                       struct nilfs_sb_info *sbi,
686                                       struct nilfs_recovery_info *ri)
687 {
688         struct buffer_head *bh;
689         int err;
690
691         if (nilfs_get_segnum_of_block(nilfs, ri->ri_lsegs_start) !=
692             nilfs_get_segnum_of_block(nilfs, ri->ri_super_root))
693                 return;
694
695         bh = sb_getblk(sbi->s_super, ri->ri_lsegs_start);
696         BUG_ON(!bh);
697         memset(bh->b_data, 0, bh->b_size);
698         set_buffer_dirty(bh);
699         err = sync_dirty_buffer(bh);
700         if (unlikely(err))
701                 printk(KERN_WARNING
702                        "NILFS warning: buffer sync write failed during "
703                        "post-cleaning of recovery.\n");
704         brelse(bh);
705 }
706
707 /**
708  * nilfs_recover_logical_segments - salvage logical segments written after
709  * the latest super root
710  * @nilfs: the_nilfs
711  * @sbi: nilfs_sb_info
712  * @ri: pointer to a nilfs_recovery_info struct to store search results.
713  *
714  * Return Value: On success, 0 is returned.  On error, one of the following
715  * negative error code is returned.
716  *
717  * %-EINVAL - Inconsistent filesystem state.
718  *
719  * %-EIO - I/O error
720  *
721  * %-ENOSPC - No space left on device (only in a panic state).
722  *
723  * %-ERESTARTSYS - Interrupted.
724  *
725  * %-ENOMEM - Insufficient memory available.
726  */
727 int nilfs_recover_logical_segments(struct the_nilfs *nilfs,
728                                    struct nilfs_sb_info *sbi,
729                                    struct nilfs_recovery_info *ri)
730 {
731         int err;
732
733         if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0)
734                 return 0;
735
736         err = nilfs_attach_checkpoint(sbi, ri->ri_cno);
737         if (unlikely(err)) {
738                 printk(KERN_ERR
739                        "NILFS: error loading the latest checkpoint.\n");
740                 return err;
741         }
742
743         err = nilfs_do_roll_forward(nilfs, sbi, ri);
744         if (unlikely(err))
745                 goto failed;
746
747         if (ri->ri_need_recovery == NILFS_RECOVERY_ROLLFORWARD_DONE) {
748                 err = nilfs_prepare_segment_for_recovery(nilfs, sbi, ri);
749                 if (unlikely(err)) {
750                         printk(KERN_ERR "NILFS: Error preparing segments for "
751                                "recovery.\n");
752                         goto failed;
753                 }
754
755                 err = nilfs_attach_segment_constructor(sbi);
756                 if (unlikely(err))
757                         goto failed;
758
759                 set_nilfs_discontinued(nilfs);
760                 err = nilfs_construct_segment(sbi->s_super);
761                 nilfs_detach_segment_constructor(sbi);
762
763                 if (unlikely(err)) {
764                         printk(KERN_ERR "NILFS: Oops! recovery failed. "
765                                "(err=%d)\n", err);
766                         goto failed;
767                 }
768
769                 nilfs_finish_roll_forward(nilfs, sbi, ri);
770         }
771
772         nilfs_detach_checkpoint(sbi);
773         return 0;
774
775  failed:
776         nilfs_detach_checkpoint(sbi);
777         nilfs_mdt_clear(nilfs->ns_cpfile);
778         nilfs_mdt_clear(nilfs->ns_sufile);
779         nilfs_mdt_clear(nilfs->ns_dat);
780         return err;
781 }
782
783 /**
784  * nilfs_search_super_root - search the latest valid super root
785  * @nilfs: the_nilfs
786  * @sbi: nilfs_sb_info
787  * @ri: pointer to a nilfs_recovery_info struct to store search results.
788  *
789  * nilfs_search_super_root() looks for the latest super-root from a partial
790  * segment pointed by the superblock.  It sets up struct the_nilfs through
791  * this search. It fills nilfs_recovery_info (ri) required for recovery.
792  *
793  * Return Value: On success, 0 is returned.  On error, one of the following
794  * negative error code is returned.
795  *
796  * %-EINVAL - No valid segment found
797  *
798  * %-EIO - I/O error
799  */
800 int nilfs_search_super_root(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi,
801                             struct nilfs_recovery_info *ri)
802 {
803         struct nilfs_segsum_info ssi;
804         sector_t pseg_start, pseg_end, sr_pseg_start = 0;
805         sector_t seg_start, seg_end; /* range of full segment (block number) */
806         u64 seg_seq;
807         __u64 segnum, nextnum = 0;
808         __u64 cno;
809         LIST_HEAD(segments);
810         int empty_seg = 0, scan_newer = 0;
811         int ret;
812
813         pseg_start = nilfs->ns_last_pseg;
814         seg_seq = nilfs->ns_last_seq;
815         cno = nilfs->ns_last_cno;
816         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
817
818         /* Calculate range of segment */
819         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
820
821         for (;;) {
822                 /* Load segment summary */
823                 ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi, 1);
824                 if (ret) {
825                         if (ret == NILFS_SEG_FAIL_IO)
826                                 goto failed;
827                         goto strayed;
828                 }
829                 pseg_end = pseg_start + ssi.nblocks - 1;
830                 if (unlikely(pseg_end > seg_end)) {
831                         ret = NILFS_SEG_FAIL_CONSISTENCY;
832                         goto strayed;
833                 }
834
835                 /* A valid partial segment */
836                 ri->ri_pseg_start = pseg_start;
837                 ri->ri_seq = seg_seq;
838                 ri->ri_segnum = segnum;
839                 nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next);
840                 ri->ri_nextnum = nextnum;
841                 empty_seg = 0;
842
843                 if (!NILFS_SEG_HAS_SR(&ssi)) {
844                         if (!scan_newer) {
845                                 /* This will never happen because a superblock
846                                    (last_segment) always points to a pseg
847                                    having a super root. */
848                                 ret = NILFS_SEG_FAIL_CONSISTENCY;
849                                 goto failed;
850                         }
851                         if (!ri->ri_lsegs_start && NILFS_SEG_LOGBGN(&ssi)) {
852                                 ri->ri_lsegs_start = pseg_start;
853                                 ri->ri_lsegs_start_seq = seg_seq;
854                         }
855                         if (NILFS_SEG_LOGEND(&ssi))
856                                 ri->ri_lsegs_end = pseg_start;
857                         goto try_next_pseg;
858                 }
859
860                 /* A valid super root was found. */
861                 ri->ri_cno = cno++;
862                 ri->ri_super_root = pseg_end;
863                 ri->ri_lsegs_start = ri->ri_lsegs_end = 0;
864
865                 nilfs_dispose_segment_list(&segments);
866                 nilfs->ns_pseg_offset = (sr_pseg_start = pseg_start)
867                         + ssi.nblocks - seg_start;
868                 nilfs->ns_seg_seq = seg_seq;
869                 nilfs->ns_segnum = segnum;
870                 nilfs->ns_cno = cno;  /* nilfs->ns_cno = ri->ri_cno + 1 */
871                 nilfs->ns_ctime = ssi.ctime;
872                 nilfs->ns_nextnum = nextnum;
873
874                 if (scan_newer)
875                         ri->ri_need_recovery = NILFS_RECOVERY_SR_UPDATED;
876                 else {
877                         if (nilfs->ns_mount_state & NILFS_VALID_FS)
878                                 goto super_root_found;
879                         scan_newer = 1;
880                 }
881
882                 /* reset region for roll-forward */
883                 pseg_start += ssi.nblocks;
884                 if (pseg_start < seg_end)
885                         continue;
886                 goto feed_segment;
887
888  try_next_pseg:
889                 /* Standing on a course, or met an inconsistent state */
890                 pseg_start += ssi.nblocks;
891                 if (pseg_start < seg_end)
892                         continue;
893                 goto feed_segment;
894
895  strayed:
896                 /* Off the trail */
897                 if (!scan_newer)
898                         /*
899                          * This can happen if a checkpoint was written without
900                          * barriers, or as a result of an I/O failure.
901                          */
902                         goto failed;
903
904  feed_segment:
905                 /* Looking to the next full segment */
906                 if (empty_seg++)
907                         goto super_root_found; /* found a valid super root */
908
909                 ret = nilfs_segment_list_add(&segments, segnum);
910                 if (unlikely(ret))
911                         goto failed;
912
913                 seg_seq++;
914                 segnum = nextnum;
915                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
916                 pseg_start = seg_start;
917         }
918
919  super_root_found:
920         /* Updating pointers relating to the latest checkpoint */
921         list_splice(&segments, ri->ri_used_segments.prev);
922         nilfs->ns_last_pseg = sr_pseg_start;
923         nilfs->ns_last_seq = nilfs->ns_seg_seq;
924         nilfs->ns_last_cno = ri->ri_cno;
925         return 0;
926
927  failed:
928         nilfs_dispose_segment_list(&segments);
929         return (ret < 0) ? ret : nilfs_warn_segment_error(ret);
930 }