Merge branch 'core/urgent' into core/rcu
[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         time_t mtime;
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         /*
426          * Releasing the next segment of the latest super root.
427          * The next segment is invalidated by this recovery.
428          */
429         err = nilfs_sufile_free(sufile, segnum[1]);
430         if (unlikely(err))
431                 goto failed;
432
433         err = -ENOMEM;
434         for (i = 1; i < 4; i++) {
435                 ent = nilfs_alloc_segment_entry(segnum[i]);
436                 if (unlikely(!ent))
437                         goto failed;
438                 list_add_tail(&ent->list, head);
439         }
440
441         /*
442          * Collecting segments written after the latest super root.
443          * These are marked dirty to avoid being reallocated in the next write.
444          */
445         mtime = get_seconds();
446         list_for_each_entry_safe(ent, n, head, list) {
447                 if (ent->segnum == segnum[0]) {
448                         list_del(&ent->list);
449                         nilfs_free_segment_entry(ent);
450                         continue;
451                 }
452                 err = nilfs_open_segment_entry(ent, sufile);
453                 if (unlikely(err))
454                         goto failed;
455                 if (!nilfs_segment_usage_dirty(ent->raw_su)) {
456                         /* make the segment garbage */
457                         ent->raw_su->su_nblocks = cpu_to_le32(0);
458                         ent->raw_su->su_lastmod = cpu_to_le32(mtime);
459                         nilfs_segment_usage_set_dirty(ent->raw_su);
460                 }
461                 list_del(&ent->list);
462                 nilfs_close_segment_entry(ent, sufile);
463                 nilfs_free_segment_entry(ent);
464         }
465
466         /* Allocate new segments for recovery */
467         err = nilfs_sufile_alloc(sufile, &segnum[0]);
468         if (unlikely(err))
469                 goto failed;
470
471         nilfs->ns_pseg_offset = 0;
472         nilfs->ns_seg_seq = ri->ri_seq + 2;
473         nilfs->ns_nextnum = nilfs->ns_segnum = segnum[0];
474         return 0;
475
476  failed:
477         /* No need to recover sufile because it will be destroyed on error */
478         return err;
479 }
480
481 static int nilfs_recovery_copy_block(struct nilfs_sb_info *sbi,
482                                      struct nilfs_recovery_block *rb,
483                                      struct page *page)
484 {
485         struct buffer_head *bh_org;
486         void *kaddr;
487
488         bh_org = sb_bread(sbi->s_super, rb->blocknr);
489         if (unlikely(!bh_org))
490                 return -EIO;
491
492         kaddr = kmap_atomic(page, KM_USER0);
493         memcpy(kaddr + bh_offset(bh_org), bh_org->b_data, bh_org->b_size);
494         kunmap_atomic(kaddr, KM_USER0);
495         brelse(bh_org);
496         return 0;
497 }
498
499 static int recover_dsync_blocks(struct nilfs_sb_info *sbi,
500                                 struct list_head *head,
501                                 unsigned long *nr_salvaged_blocks)
502 {
503         struct inode *inode;
504         struct nilfs_recovery_block *rb, *n;
505         unsigned blocksize = sbi->s_super->s_blocksize;
506         struct page *page;
507         loff_t pos;
508         int err = 0, err2 = 0;
509
510         list_for_each_entry_safe(rb, n, head, list) {
511                 inode = nilfs_iget(sbi->s_super, rb->ino);
512                 if (IS_ERR(inode)) {
513                         err = PTR_ERR(inode);
514                         inode = NULL;
515                         goto failed_inode;
516                 }
517
518                 pos = rb->blkoff << inode->i_blkbits;
519                 page = NULL;
520                 err = block_write_begin(NULL, inode->i_mapping, pos, blocksize,
521                                         0, &page, NULL, nilfs_get_block);
522                 if (unlikely(err))
523                         goto failed_inode;
524
525                 err = nilfs_recovery_copy_block(sbi, rb, page);
526                 if (unlikely(err))
527                         goto failed_page;
528
529                 err = nilfs_set_file_dirty(sbi, inode, 1);
530                 if (unlikely(err))
531                         goto failed_page;
532
533                 block_write_end(NULL, inode->i_mapping, pos, blocksize,
534                                 blocksize, page, NULL);
535
536                 unlock_page(page);
537                 page_cache_release(page);
538
539                 (*nr_salvaged_blocks)++;
540                 goto next;
541
542  failed_page:
543                 unlock_page(page);
544                 page_cache_release(page);
545
546  failed_inode:
547                 printk(KERN_WARNING
548                        "NILFS warning: error recovering data block "
549                        "(err=%d, ino=%lu, block-offset=%llu)\n",
550                        err, rb->ino, (unsigned long long)rb->blkoff);
551                 if (!err2)
552                         err2 = err;
553  next:
554                 iput(inode); /* iput(NULL) is just ignored */
555                 list_del_init(&rb->list);
556                 kfree(rb);
557         }
558         return err2;
559 }
560
561 /**
562  * nilfs_do_roll_forward - salvage logical segments newer than the latest
563  * checkpoint
564  * @sbi: nilfs_sb_info
565  * @nilfs: the_nilfs
566  * @ri: pointer to a nilfs_recovery_info
567  */
568 static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
569                                  struct nilfs_sb_info *sbi,
570                                  struct nilfs_recovery_info *ri)
571 {
572         struct nilfs_segsum_info ssi;
573         sector_t pseg_start;
574         sector_t seg_start, seg_end;  /* Starting/ending DBN of full segment */
575         unsigned long nsalvaged_blocks = 0;
576         u64 seg_seq;
577         __u64 segnum, nextnum = 0;
578         int empty_seg = 0;
579         int err = 0, ret;
580         LIST_HEAD(dsync_blocks);  /* list of data blocks to be recovered */
581         enum {
582                 RF_INIT_ST,
583                 RF_DSYNC_ST,   /* scanning data-sync segments */
584         };
585         int state = RF_INIT_ST;
586
587         nilfs_attach_writer(nilfs, sbi);
588         pseg_start = ri->ri_lsegs_start;
589         seg_seq = ri->ri_lsegs_start_seq;
590         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
591         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
592
593         while (segnum != ri->ri_segnum || pseg_start <= ri->ri_pseg_start) {
594
595                 ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi, 1);
596                 if (ret) {
597                         if (ret == NILFS_SEG_FAIL_IO) {
598                                 err = -EIO;
599                                 goto failed;
600                         }
601                         goto strayed;
602                 }
603                 if (unlikely(NILFS_SEG_HAS_SR(&ssi)))
604                         goto confused;
605
606                 /* Found a valid partial segment; do recovery actions */
607                 nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next);
608                 empty_seg = 0;
609                 nilfs->ns_ctime = ssi.ctime;
610                 if (!(ssi.flags & NILFS_SS_GC))
611                         nilfs->ns_nongc_ctime = ssi.ctime;
612
613                 switch (state) {
614                 case RF_INIT_ST:
615                         if (!NILFS_SEG_LOGBGN(&ssi) || !NILFS_SEG_DSYNC(&ssi))
616                                 goto try_next_pseg;
617                         state = RF_DSYNC_ST;
618                         /* Fall through */
619                 case RF_DSYNC_ST:
620                         if (!NILFS_SEG_DSYNC(&ssi))
621                                 goto confused;
622
623                         err = collect_blocks_from_segsum(
624                                 sbi, pseg_start, &ssi, &dsync_blocks);
625                         if (unlikely(err))
626                                 goto failed;
627                         if (NILFS_SEG_LOGEND(&ssi)) {
628                                 err = recover_dsync_blocks(
629                                         sbi, &dsync_blocks, &nsalvaged_blocks);
630                                 if (unlikely(err))
631                                         goto failed;
632                                 state = RF_INIT_ST;
633                         }
634                         break; /* Fall through to try_next_pseg */
635                 }
636
637  try_next_pseg:
638                 if (pseg_start == ri->ri_lsegs_end)
639                         break;
640                 pseg_start += ssi.nblocks;
641                 if (pseg_start < seg_end)
642                         continue;
643                 goto feed_segment;
644
645  strayed:
646                 if (pseg_start == ri->ri_lsegs_end)
647                         break;
648
649  feed_segment:
650                 /* Looking to the next full segment */
651                 if (empty_seg++)
652                         break;
653                 seg_seq++;
654                 segnum = nextnum;
655                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
656                 pseg_start = seg_start;
657         }
658
659         if (nsalvaged_blocks) {
660                 printk(KERN_INFO "NILFS (device %s): salvaged %lu blocks\n",
661                        sbi->s_super->s_id, nsalvaged_blocks);
662                 ri->ri_need_recovery = NILFS_RECOVERY_ROLLFORWARD_DONE;
663         }
664  out:
665         dispose_recovery_list(&dsync_blocks);
666         nilfs_detach_writer(sbi->s_nilfs, sbi);
667         return err;
668
669  confused:
670         err = -EINVAL;
671  failed:
672         printk(KERN_ERR
673                "NILFS (device %s): Error roll-forwarding "
674                "(err=%d, pseg block=%llu). ",
675                sbi->s_super->s_id, err, (unsigned long long)pseg_start);
676         goto out;
677 }
678
679 static void nilfs_finish_roll_forward(struct the_nilfs *nilfs,
680                                       struct nilfs_sb_info *sbi,
681                                       struct nilfs_recovery_info *ri)
682 {
683         struct buffer_head *bh;
684         int err;
685
686         if (nilfs_get_segnum_of_block(nilfs, ri->ri_lsegs_start) !=
687             nilfs_get_segnum_of_block(nilfs, ri->ri_super_root))
688                 return;
689
690         bh = sb_getblk(sbi->s_super, ri->ri_lsegs_start);
691         BUG_ON(!bh);
692         memset(bh->b_data, 0, bh->b_size);
693         set_buffer_dirty(bh);
694         err = sync_dirty_buffer(bh);
695         if (unlikely(err))
696                 printk(KERN_WARNING
697                        "NILFS warning: buffer sync write failed during "
698                        "post-cleaning of recovery.\n");
699         brelse(bh);
700 }
701
702 /**
703  * nilfs_recover_logical_segments - salvage logical segments written after
704  * the latest super root
705  * @nilfs: the_nilfs
706  * @sbi: nilfs_sb_info
707  * @ri: pointer to a nilfs_recovery_info struct to store search results.
708  *
709  * Return Value: On success, 0 is returned.  On error, one of the following
710  * negative error code is returned.
711  *
712  * %-EINVAL - Inconsistent filesystem state.
713  *
714  * %-EIO - I/O error
715  *
716  * %-ENOSPC - No space left on device (only in a panic state).
717  *
718  * %-ERESTARTSYS - Interrupted.
719  *
720  * %-ENOMEM - Insufficient memory available.
721  */
722 int nilfs_recover_logical_segments(struct the_nilfs *nilfs,
723                                    struct nilfs_sb_info *sbi,
724                                    struct nilfs_recovery_info *ri)
725 {
726         int err;
727
728         if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0)
729                 return 0;
730
731         err = nilfs_attach_checkpoint(sbi, ri->ri_cno);
732         if (unlikely(err)) {
733                 printk(KERN_ERR
734                        "NILFS: error loading the latest checkpoint.\n");
735                 return err;
736         }
737
738         err = nilfs_do_roll_forward(nilfs, sbi, ri);
739         if (unlikely(err))
740                 goto failed;
741
742         if (ri->ri_need_recovery == NILFS_RECOVERY_ROLLFORWARD_DONE) {
743                 err = nilfs_prepare_segment_for_recovery(nilfs, ri);
744                 if (unlikely(err)) {
745                         printk(KERN_ERR "NILFS: Error preparing segments for "
746                                "recovery.\n");
747                         goto failed;
748                 }
749
750                 err = nilfs_attach_segment_constructor(sbi);
751                 if (unlikely(err))
752                         goto failed;
753
754                 set_nilfs_discontinued(nilfs);
755                 err = nilfs_construct_segment(sbi->s_super);
756                 nilfs_detach_segment_constructor(sbi);
757
758                 if (unlikely(err)) {
759                         printk(KERN_ERR "NILFS: Oops! recovery failed. "
760                                "(err=%d)\n", err);
761                         goto failed;
762                 }
763
764                 nilfs_finish_roll_forward(nilfs, sbi, ri);
765         }
766
767         nilfs_detach_checkpoint(sbi);
768         return 0;
769
770  failed:
771         nilfs_detach_checkpoint(sbi);
772         nilfs_mdt_clear(nilfs->ns_cpfile);
773         nilfs_mdt_clear(nilfs->ns_sufile);
774         nilfs_mdt_clear(nilfs->ns_dat);
775         return err;
776 }
777
778 /**
779  * nilfs_search_super_root - search the latest valid super root
780  * @nilfs: the_nilfs
781  * @sbi: nilfs_sb_info
782  * @ri: pointer to a nilfs_recovery_info struct to store search results.
783  *
784  * nilfs_search_super_root() looks for the latest super-root from a partial
785  * segment pointed by the superblock.  It sets up struct the_nilfs through
786  * this search. It fills nilfs_recovery_info (ri) required for recovery.
787  *
788  * Return Value: On success, 0 is returned.  On error, one of the following
789  * negative error code is returned.
790  *
791  * %-EINVAL - No valid segment found
792  *
793  * %-EIO - I/O error
794  */
795 int nilfs_search_super_root(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi,
796                             struct nilfs_recovery_info *ri)
797 {
798         struct nilfs_segsum_info ssi;
799         sector_t pseg_start, pseg_end, sr_pseg_start = 0;
800         sector_t seg_start, seg_end; /* range of full segment (block number) */
801         u64 seg_seq;
802         __u64 segnum, nextnum = 0;
803         __u64 cno;
804         struct nilfs_segment_entry *ent;
805         LIST_HEAD(segments);
806         int empty_seg = 0, scan_newer = 0;
807         int ret;
808
809         pseg_start = nilfs->ns_last_pseg;
810         seg_seq = nilfs->ns_last_seq;
811         cno = nilfs->ns_last_cno;
812         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
813
814         /* Calculate range of segment */
815         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
816
817         for (;;) {
818                 /* Load segment summary */
819                 ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi, 1);
820                 if (ret) {
821                         if (ret == NILFS_SEG_FAIL_IO)
822                                 goto failed;
823                         goto strayed;
824                 }
825                 pseg_end = pseg_start + ssi.nblocks - 1;
826                 if (unlikely(pseg_end > seg_end)) {
827                         ret = NILFS_SEG_FAIL_CONSISTENCY;
828                         goto strayed;
829                 }
830
831                 /* A valid partial segment */
832                 ri->ri_pseg_start = pseg_start;
833                 ri->ri_seq = seg_seq;
834                 ri->ri_segnum = segnum;
835                 nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next);
836                 ri->ri_nextnum = nextnum;
837                 empty_seg = 0;
838
839                 if (!NILFS_SEG_HAS_SR(&ssi)) {
840                         if (!scan_newer) {
841                                 /* This will never happen because a superblock
842                                    (last_segment) always points to a pseg
843                                    having a super root. */
844                                 ret = NILFS_SEG_FAIL_CONSISTENCY;
845                                 goto failed;
846                         }
847                         if (!ri->ri_lsegs_start && NILFS_SEG_LOGBGN(&ssi)) {
848                                 ri->ri_lsegs_start = pseg_start;
849                                 ri->ri_lsegs_start_seq = seg_seq;
850                         }
851                         if (NILFS_SEG_LOGEND(&ssi))
852                                 ri->ri_lsegs_end = pseg_start;
853                         goto try_next_pseg;
854                 }
855
856                 /* A valid super root was found. */
857                 ri->ri_cno = cno++;
858                 ri->ri_super_root = pseg_end;
859                 ri->ri_lsegs_start = ri->ri_lsegs_end = 0;
860
861                 nilfs_dispose_segment_list(&segments);
862                 nilfs->ns_pseg_offset = (sr_pseg_start = pseg_start)
863                         + ssi.nblocks - seg_start;
864                 nilfs->ns_seg_seq = seg_seq;
865                 nilfs->ns_segnum = segnum;
866                 nilfs->ns_cno = cno;  /* nilfs->ns_cno = ri->ri_cno + 1 */
867                 nilfs->ns_ctime = ssi.ctime;
868                 nilfs->ns_nextnum = nextnum;
869
870                 if (scan_newer)
871                         ri->ri_need_recovery = NILFS_RECOVERY_SR_UPDATED;
872                 else {
873                         if (nilfs->ns_mount_state & NILFS_VALID_FS)
874                                 goto super_root_found;
875                         scan_newer = 1;
876                 }
877
878                 /* reset region for roll-forward */
879                 pseg_start += ssi.nblocks;
880                 if (pseg_start < seg_end)
881                         continue;
882                 goto feed_segment;
883
884  try_next_pseg:
885                 /* Standing on a course, or met an inconsistent state */
886                 pseg_start += ssi.nblocks;
887                 if (pseg_start < seg_end)
888                         continue;
889                 goto feed_segment;
890
891  strayed:
892                 /* Off the trail */
893                 if (!scan_newer)
894                         /*
895                          * This can happen if a checkpoint was written without
896                          * barriers, or as a result of an I/O failure.
897                          */
898                         goto failed;
899
900  feed_segment:
901                 /* Looking to the next full segment */
902                 if (empty_seg++)
903                         goto super_root_found; /* found a valid super root */
904
905                 ent = nilfs_alloc_segment_entry(segnum);
906                 if (unlikely(!ent)) {
907                         ret = -ENOMEM;
908                         goto failed;
909                 }
910                 list_add_tail(&ent->list, &segments);
911
912                 seg_seq++;
913                 segnum = nextnum;
914                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
915                 pseg_start = seg_start;
916         }
917
918  super_root_found:
919         /* Updating pointers relating to the latest checkpoint */
920         list_splice(&segments, ri->ri_used_segments.prev);
921         nilfs->ns_last_pseg = sr_pseg_start;
922         nilfs->ns_last_seq = nilfs->ns_seg_seq;
923         nilfs->ns_last_cno = ri->ri_cno;
924         return 0;
925
926  failed:
927         nilfs_dispose_segment_list(&segments);
928         return (ret < 0) ? ret : nilfs_warn_segment_error(ret);
929 }