Merge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc
[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_recovery_info *ri)
411 {
412         struct list_head *head = &ri->ri_used_segments;
413         struct nilfs_segment_entry *ent, *n;
414         struct inode *sufile = nilfs->ns_sufile;
415         __u64 segnum[4];
416         int err;
417         int i;
418
419         segnum[0] = nilfs->ns_segnum;
420         segnum[1] = nilfs->ns_nextnum;
421         segnum[2] = ri->ri_segnum;
422         segnum[3] = ri->ri_nextnum;
423
424         /*
425          * Releasing the next segment of the latest super root.
426          * The next segment is invalidated by this recovery.
427          */
428         err = nilfs_sufile_free(sufile, segnum[1]);
429         if (unlikely(err))
430                 goto failed;
431
432         err = -ENOMEM;
433         for (i = 1; i < 4; i++) {
434                 ent = nilfs_alloc_segment_entry(segnum[i]);
435                 if (unlikely(!ent))
436                         goto failed;
437                 list_add_tail(&ent->list, head);
438         }
439
440         /*
441          * Collecting segments written after the latest super root.
442          * These are marked dirty to avoid being reallocated in the next write.
443          */
444         list_for_each_entry_safe(ent, n, head, list) {
445                 if (ent->segnum != segnum[0]) {
446                         err = nilfs_sufile_scrap(sufile, ent->segnum);
447                         if (unlikely(err))
448                                 goto failed;
449                 }
450                 list_del(&ent->list);
451                 nilfs_free_segment_entry(ent);
452         }
453
454         /* Allocate new segments for recovery */
455         err = nilfs_sufile_alloc(sufile, &segnum[0]);
456         if (unlikely(err))
457                 goto failed;
458
459         nilfs->ns_pseg_offset = 0;
460         nilfs->ns_seg_seq = ri->ri_seq + 2;
461         nilfs->ns_nextnum = nilfs->ns_segnum = segnum[0];
462         return 0;
463
464  failed:
465         /* No need to recover sufile because it will be destroyed on error */
466         return err;
467 }
468
469 static int nilfs_recovery_copy_block(struct nilfs_sb_info *sbi,
470                                      struct nilfs_recovery_block *rb,
471                                      struct page *page)
472 {
473         struct buffer_head *bh_org;
474         void *kaddr;
475
476         bh_org = sb_bread(sbi->s_super, rb->blocknr);
477         if (unlikely(!bh_org))
478                 return -EIO;
479
480         kaddr = kmap_atomic(page, KM_USER0);
481         memcpy(kaddr + bh_offset(bh_org), bh_org->b_data, bh_org->b_size);
482         kunmap_atomic(kaddr, KM_USER0);
483         brelse(bh_org);
484         return 0;
485 }
486
487 static int recover_dsync_blocks(struct nilfs_sb_info *sbi,
488                                 struct list_head *head,
489                                 unsigned long *nr_salvaged_blocks)
490 {
491         struct inode *inode;
492         struct nilfs_recovery_block *rb, *n;
493         unsigned blocksize = sbi->s_super->s_blocksize;
494         struct page *page;
495         loff_t pos;
496         int err = 0, err2 = 0;
497
498         list_for_each_entry_safe(rb, n, head, list) {
499                 inode = nilfs_iget(sbi->s_super, rb->ino);
500                 if (IS_ERR(inode)) {
501                         err = PTR_ERR(inode);
502                         inode = NULL;
503                         goto failed_inode;
504                 }
505
506                 pos = rb->blkoff << inode->i_blkbits;
507                 page = NULL;
508                 err = block_write_begin(NULL, inode->i_mapping, pos, blocksize,
509                                         0, &page, NULL, nilfs_get_block);
510                 if (unlikely(err))
511                         goto failed_inode;
512
513                 err = nilfs_recovery_copy_block(sbi, rb, page);
514                 if (unlikely(err))
515                         goto failed_page;
516
517                 err = nilfs_set_file_dirty(sbi, inode, 1);
518                 if (unlikely(err))
519                         goto failed_page;
520
521                 block_write_end(NULL, inode->i_mapping, pos, blocksize,
522                                 blocksize, page, NULL);
523
524                 unlock_page(page);
525                 page_cache_release(page);
526
527                 (*nr_salvaged_blocks)++;
528                 goto next;
529
530  failed_page:
531                 unlock_page(page);
532                 page_cache_release(page);
533
534  failed_inode:
535                 printk(KERN_WARNING
536                        "NILFS warning: error recovering data block "
537                        "(err=%d, ino=%lu, block-offset=%llu)\n",
538                        err, rb->ino, (unsigned long long)rb->blkoff);
539                 if (!err2)
540                         err2 = err;
541  next:
542                 iput(inode); /* iput(NULL) is just ignored */
543                 list_del_init(&rb->list);
544                 kfree(rb);
545         }
546         return err2;
547 }
548
549 /**
550  * nilfs_do_roll_forward - salvage logical segments newer than the latest
551  * checkpoint
552  * @sbi: nilfs_sb_info
553  * @nilfs: the_nilfs
554  * @ri: pointer to a nilfs_recovery_info
555  */
556 static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
557                                  struct nilfs_sb_info *sbi,
558                                  struct nilfs_recovery_info *ri)
559 {
560         struct nilfs_segsum_info ssi;
561         sector_t pseg_start;
562         sector_t seg_start, seg_end;  /* Starting/ending DBN of full segment */
563         unsigned long nsalvaged_blocks = 0;
564         u64 seg_seq;
565         __u64 segnum, nextnum = 0;
566         int empty_seg = 0;
567         int err = 0, ret;
568         LIST_HEAD(dsync_blocks);  /* list of data blocks to be recovered */
569         enum {
570                 RF_INIT_ST,
571                 RF_DSYNC_ST,   /* scanning data-sync segments */
572         };
573         int state = RF_INIT_ST;
574
575         nilfs_attach_writer(nilfs, sbi);
576         pseg_start = ri->ri_lsegs_start;
577         seg_seq = ri->ri_lsegs_start_seq;
578         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
579         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
580
581         while (segnum != ri->ri_segnum || pseg_start <= ri->ri_pseg_start) {
582
583                 ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi, 1);
584                 if (ret) {
585                         if (ret == NILFS_SEG_FAIL_IO) {
586                                 err = -EIO;
587                                 goto failed;
588                         }
589                         goto strayed;
590                 }
591                 if (unlikely(NILFS_SEG_HAS_SR(&ssi)))
592                         goto confused;
593
594                 /* Found a valid partial segment; do recovery actions */
595                 nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next);
596                 empty_seg = 0;
597                 nilfs->ns_ctime = ssi.ctime;
598                 if (!(ssi.flags & NILFS_SS_GC))
599                         nilfs->ns_nongc_ctime = ssi.ctime;
600
601                 switch (state) {
602                 case RF_INIT_ST:
603                         if (!NILFS_SEG_LOGBGN(&ssi) || !NILFS_SEG_DSYNC(&ssi))
604                                 goto try_next_pseg;
605                         state = RF_DSYNC_ST;
606                         /* Fall through */
607                 case RF_DSYNC_ST:
608                         if (!NILFS_SEG_DSYNC(&ssi))
609                                 goto confused;
610
611                         err = collect_blocks_from_segsum(
612                                 sbi, pseg_start, &ssi, &dsync_blocks);
613                         if (unlikely(err))
614                                 goto failed;
615                         if (NILFS_SEG_LOGEND(&ssi)) {
616                                 err = recover_dsync_blocks(
617                                         sbi, &dsync_blocks, &nsalvaged_blocks);
618                                 if (unlikely(err))
619                                         goto failed;
620                                 state = RF_INIT_ST;
621                         }
622                         break; /* Fall through to try_next_pseg */
623                 }
624
625  try_next_pseg:
626                 if (pseg_start == ri->ri_lsegs_end)
627                         break;
628                 pseg_start += ssi.nblocks;
629                 if (pseg_start < seg_end)
630                         continue;
631                 goto feed_segment;
632
633  strayed:
634                 if (pseg_start == ri->ri_lsegs_end)
635                         break;
636
637  feed_segment:
638                 /* Looking to the next full segment */
639                 if (empty_seg++)
640                         break;
641                 seg_seq++;
642                 segnum = nextnum;
643                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
644                 pseg_start = seg_start;
645         }
646
647         if (nsalvaged_blocks) {
648                 printk(KERN_INFO "NILFS (device %s): salvaged %lu blocks\n",
649                        sbi->s_super->s_id, nsalvaged_blocks);
650                 ri->ri_need_recovery = NILFS_RECOVERY_ROLLFORWARD_DONE;
651         }
652  out:
653         dispose_recovery_list(&dsync_blocks);
654         nilfs_detach_writer(sbi->s_nilfs, sbi);
655         return err;
656
657  confused:
658         err = -EINVAL;
659  failed:
660         printk(KERN_ERR
661                "NILFS (device %s): Error roll-forwarding "
662                "(err=%d, pseg block=%llu). ",
663                sbi->s_super->s_id, err, (unsigned long long)pseg_start);
664         goto out;
665 }
666
667 static void nilfs_finish_roll_forward(struct the_nilfs *nilfs,
668                                       struct nilfs_sb_info *sbi,
669                                       struct nilfs_recovery_info *ri)
670 {
671         struct buffer_head *bh;
672         int err;
673
674         if (nilfs_get_segnum_of_block(nilfs, ri->ri_lsegs_start) !=
675             nilfs_get_segnum_of_block(nilfs, ri->ri_super_root))
676                 return;
677
678         bh = sb_getblk(sbi->s_super, ri->ri_lsegs_start);
679         BUG_ON(!bh);
680         memset(bh->b_data, 0, bh->b_size);
681         set_buffer_dirty(bh);
682         err = sync_dirty_buffer(bh);
683         if (unlikely(err))
684                 printk(KERN_WARNING
685                        "NILFS warning: buffer sync write failed during "
686                        "post-cleaning of recovery.\n");
687         brelse(bh);
688 }
689
690 /**
691  * nilfs_recover_logical_segments - salvage logical segments written after
692  * the latest super root
693  * @nilfs: the_nilfs
694  * @sbi: nilfs_sb_info
695  * @ri: pointer to a nilfs_recovery_info struct to store search results.
696  *
697  * Return Value: On success, 0 is returned.  On error, one of the following
698  * negative error code is returned.
699  *
700  * %-EINVAL - Inconsistent filesystem state.
701  *
702  * %-EIO - I/O error
703  *
704  * %-ENOSPC - No space left on device (only in a panic state).
705  *
706  * %-ERESTARTSYS - Interrupted.
707  *
708  * %-ENOMEM - Insufficient memory available.
709  */
710 int nilfs_recover_logical_segments(struct the_nilfs *nilfs,
711                                    struct nilfs_sb_info *sbi,
712                                    struct nilfs_recovery_info *ri)
713 {
714         int err;
715
716         if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0)
717                 return 0;
718
719         err = nilfs_attach_checkpoint(sbi, ri->ri_cno);
720         if (unlikely(err)) {
721                 printk(KERN_ERR
722                        "NILFS: error loading the latest checkpoint.\n");
723                 return err;
724         }
725
726         err = nilfs_do_roll_forward(nilfs, sbi, ri);
727         if (unlikely(err))
728                 goto failed;
729
730         if (ri->ri_need_recovery == NILFS_RECOVERY_ROLLFORWARD_DONE) {
731                 err = nilfs_prepare_segment_for_recovery(nilfs, ri);
732                 if (unlikely(err)) {
733                         printk(KERN_ERR "NILFS: Error preparing segments for "
734                                "recovery.\n");
735                         goto failed;
736                 }
737
738                 err = nilfs_attach_segment_constructor(sbi);
739                 if (unlikely(err))
740                         goto failed;
741
742                 set_nilfs_discontinued(nilfs);
743                 err = nilfs_construct_segment(sbi->s_super);
744                 nilfs_detach_segment_constructor(sbi);
745
746                 if (unlikely(err)) {
747                         printk(KERN_ERR "NILFS: Oops! recovery failed. "
748                                "(err=%d)\n", err);
749                         goto failed;
750                 }
751
752                 nilfs_finish_roll_forward(nilfs, sbi, ri);
753         }
754
755         nilfs_detach_checkpoint(sbi);
756         return 0;
757
758  failed:
759         nilfs_detach_checkpoint(sbi);
760         nilfs_mdt_clear(nilfs->ns_cpfile);
761         nilfs_mdt_clear(nilfs->ns_sufile);
762         nilfs_mdt_clear(nilfs->ns_dat);
763         return err;
764 }
765
766 /**
767  * nilfs_search_super_root - search the latest valid super root
768  * @nilfs: the_nilfs
769  * @sbi: nilfs_sb_info
770  * @ri: pointer to a nilfs_recovery_info struct to store search results.
771  *
772  * nilfs_search_super_root() looks for the latest super-root from a partial
773  * segment pointed by the superblock.  It sets up struct the_nilfs through
774  * this search. It fills nilfs_recovery_info (ri) required for recovery.
775  *
776  * Return Value: On success, 0 is returned.  On error, one of the following
777  * negative error code is returned.
778  *
779  * %-EINVAL - No valid segment found
780  *
781  * %-EIO - I/O error
782  */
783 int nilfs_search_super_root(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi,
784                             struct nilfs_recovery_info *ri)
785 {
786         struct nilfs_segsum_info ssi;
787         sector_t pseg_start, pseg_end, sr_pseg_start = 0;
788         sector_t seg_start, seg_end; /* range of full segment (block number) */
789         u64 seg_seq;
790         __u64 segnum, nextnum = 0;
791         __u64 cno;
792         struct nilfs_segment_entry *ent;
793         LIST_HEAD(segments);
794         int empty_seg = 0, scan_newer = 0;
795         int ret;
796
797         pseg_start = nilfs->ns_last_pseg;
798         seg_seq = nilfs->ns_last_seq;
799         cno = nilfs->ns_last_cno;
800         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
801
802         /* Calculate range of segment */
803         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
804
805         for (;;) {
806                 /* Load segment summary */
807                 ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi, 1);
808                 if (ret) {
809                         if (ret == NILFS_SEG_FAIL_IO)
810                                 goto failed;
811                         goto strayed;
812                 }
813                 pseg_end = pseg_start + ssi.nblocks - 1;
814                 if (unlikely(pseg_end > seg_end)) {
815                         ret = NILFS_SEG_FAIL_CONSISTENCY;
816                         goto strayed;
817                 }
818
819                 /* A valid partial segment */
820                 ri->ri_pseg_start = pseg_start;
821                 ri->ri_seq = seg_seq;
822                 ri->ri_segnum = segnum;
823                 nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next);
824                 ri->ri_nextnum = nextnum;
825                 empty_seg = 0;
826
827                 if (!NILFS_SEG_HAS_SR(&ssi)) {
828                         if (!scan_newer) {
829                                 /* This will never happen because a superblock
830                                    (last_segment) always points to a pseg
831                                    having a super root. */
832                                 ret = NILFS_SEG_FAIL_CONSISTENCY;
833                                 goto failed;
834                         }
835                         if (!ri->ri_lsegs_start && NILFS_SEG_LOGBGN(&ssi)) {
836                                 ri->ri_lsegs_start = pseg_start;
837                                 ri->ri_lsegs_start_seq = seg_seq;
838                         }
839                         if (NILFS_SEG_LOGEND(&ssi))
840                                 ri->ri_lsegs_end = pseg_start;
841                         goto try_next_pseg;
842                 }
843
844                 /* A valid super root was found. */
845                 ri->ri_cno = cno++;
846                 ri->ri_super_root = pseg_end;
847                 ri->ri_lsegs_start = ri->ri_lsegs_end = 0;
848
849                 nilfs_dispose_segment_list(&segments);
850                 nilfs->ns_pseg_offset = (sr_pseg_start = pseg_start)
851                         + ssi.nblocks - seg_start;
852                 nilfs->ns_seg_seq = seg_seq;
853                 nilfs->ns_segnum = segnum;
854                 nilfs->ns_cno = cno;  /* nilfs->ns_cno = ri->ri_cno + 1 */
855                 nilfs->ns_ctime = ssi.ctime;
856                 nilfs->ns_nextnum = nextnum;
857
858                 if (scan_newer)
859                         ri->ri_need_recovery = NILFS_RECOVERY_SR_UPDATED;
860                 else {
861                         if (nilfs->ns_mount_state & NILFS_VALID_FS)
862                                 goto super_root_found;
863                         scan_newer = 1;
864                 }
865
866                 /* reset region for roll-forward */
867                 pseg_start += ssi.nblocks;
868                 if (pseg_start < seg_end)
869                         continue;
870                 goto feed_segment;
871
872  try_next_pseg:
873                 /* Standing on a course, or met an inconsistent state */
874                 pseg_start += ssi.nblocks;
875                 if (pseg_start < seg_end)
876                         continue;
877                 goto feed_segment;
878
879  strayed:
880                 /* Off the trail */
881                 if (!scan_newer)
882                         /*
883                          * This can happen if a checkpoint was written without
884                          * barriers, or as a result of an I/O failure.
885                          */
886                         goto failed;
887
888  feed_segment:
889                 /* Looking to the next full segment */
890                 if (empty_seg++)
891                         goto super_root_found; /* found a valid super root */
892
893                 ent = nilfs_alloc_segment_entry(segnum);
894                 if (unlikely(!ent)) {
895                         ret = -ENOMEM;
896                         goto failed;
897                 }
898                 list_add_tail(&ent->list, &segments);
899
900                 seg_seq++;
901                 segnum = nextnum;
902                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
903                 pseg_start = seg_start;
904         }
905
906  super_root_found:
907         /* Updating pointers relating to the latest checkpoint */
908         list_splice(&segments, ri->ri_used_segments.prev);
909         nilfs->ns_last_pseg = sr_pseg_start;
910         nilfs->ns_last_seq = nilfs->ns_seg_seq;
911         nilfs->ns_last_cno = ri->ri_cno;
912         return 0;
913
914  failed:
915         nilfs_dispose_segment_list(&segments);
916         return (ret < 0) ? ret : nilfs_warn_segment_error(ret);
917 }