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