nilfs2: set pointer to root object in inodes
[linux-2.6.git] / fs / nilfs2 / recovery.c
1 /*
2  * recovery.c - NILFS recovery logic
3  *
4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
19  *
20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
21  */
22
23 #include <linux/buffer_head.h>
24 #include <linux/blkdev.h>
25 #include <linux/swap.h>
26 #include <linux/slab.h>
27 #include <linux/crc32.h>
28 #include "nilfs.h"
29 #include "segment.h"
30 #include "sufile.h"
31 #include "page.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_SUPER_ROOT,
44         NILFS_SEG_FAIL_CHECKSUM_FULL,
45         NILFS_SEG_FAIL_CONSISTENCY,
46 };
47
48 /* work structure for recovery */
49 struct nilfs_recovery_block {
50         ino_t ino;              /* Inode number of the file that this block
51                                    belongs to */
52         sector_t blocknr;       /* block number */
53         __u64 vblocknr;         /* virtual block number */
54         unsigned long blkoff;   /* File offset of the data block (per block) */
55         struct list_head list;
56 };
57
58
59 static int nilfs_warn_segment_error(int err)
60 {
61         switch (err) {
62         case NILFS_SEG_FAIL_IO:
63                 printk(KERN_WARNING
64                        "NILFS warning: I/O error on loading last segment\n");
65                 return -EIO;
66         case NILFS_SEG_FAIL_MAGIC:
67                 printk(KERN_WARNING
68                        "NILFS warning: Segment magic number invalid\n");
69                 break;
70         case NILFS_SEG_FAIL_SEQ:
71                 printk(KERN_WARNING
72                        "NILFS warning: Sequence number mismatch\n");
73                 break;
74         case NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT:
75                 printk(KERN_WARNING
76                        "NILFS warning: Checksum error in super root\n");
77                 break;
78         case NILFS_SEG_FAIL_CHECKSUM_FULL:
79                 printk(KERN_WARNING
80                        "NILFS warning: Checksum error in segment payload\n");
81                 break;
82         case NILFS_SEG_FAIL_CONSISTENCY:
83                 printk(KERN_WARNING
84                        "NILFS warning: Inconsistent segment\n");
85                 break;
86         case NILFS_SEG_NO_SUPER_ROOT:
87                 printk(KERN_WARNING
88                        "NILFS warning: No super root in the last segment\n");
89                 break;
90         }
91         return -EINVAL;
92 }
93
94 /**
95  * nilfs_compute_checksum - compute checksum of blocks continuously
96  * @nilfs: nilfs object
97  * @bhs: buffer head of start block
98  * @sum: place to store result
99  * @offset: offset bytes in the first block
100  * @check_bytes: number of bytes to be checked
101  * @start: DBN of start block
102  * @nblock: number of blocks to be checked
103  */
104 static int nilfs_compute_checksum(struct the_nilfs *nilfs,
105                                   struct buffer_head *bhs, u32 *sum,
106                                   unsigned long offset, u64 check_bytes,
107                                   sector_t start, unsigned long nblock)
108 {
109         unsigned int blocksize = nilfs->ns_blocksize;
110         unsigned long size;
111         u32 crc;
112
113         BUG_ON(offset >= blocksize);
114         check_bytes -= offset;
115         size = min_t(u64, check_bytes, blocksize - offset);
116         crc = crc32_le(nilfs->ns_crc_seed,
117                        (unsigned char *)bhs->b_data + offset, size);
118         if (--nblock > 0) {
119                 do {
120                         struct buffer_head *bh;
121
122                         bh = __bread(nilfs->ns_bdev, ++start, blocksize);
123                         if (!bh)
124                                 return -EIO;
125                         check_bytes -= size;
126                         size = min_t(u64, check_bytes, blocksize);
127                         crc = crc32_le(crc, bh->b_data, size);
128                         brelse(bh);
129                 } while (--nblock > 0);
130         }
131         *sum = crc;
132         return 0;
133 }
134
135 /**
136  * nilfs_read_super_root_block - read super root block
137  * @nilfs: nilfs object
138  * @sr_block: disk block number of the super root block
139  * @pbh: address of a buffer_head pointer to return super root buffer
140  * @check: CRC check flag
141  */
142 int nilfs_read_super_root_block(struct the_nilfs *nilfs, sector_t sr_block,
143                                 struct buffer_head **pbh, int check)
144 {
145         struct buffer_head *bh_sr;
146         struct nilfs_super_root *sr;
147         u32 crc;
148         int ret;
149
150         *pbh = NULL;
151         bh_sr = __bread(nilfs->ns_bdev, sr_block, nilfs->ns_blocksize);
152         if (unlikely(!bh_sr)) {
153                 ret = NILFS_SEG_FAIL_IO;
154                 goto failed;
155         }
156
157         sr = (struct nilfs_super_root *)bh_sr->b_data;
158         if (check) {
159                 unsigned bytes = le16_to_cpu(sr->sr_bytes);
160
161                 if (bytes == 0 || bytes > nilfs->ns_blocksize) {
162                         ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
163                         goto failed_bh;
164                 }
165                 if (nilfs_compute_checksum(
166                             nilfs, bh_sr, &crc, sizeof(sr->sr_sum), bytes,
167                             sr_block, 1)) {
168                         ret = NILFS_SEG_FAIL_IO;
169                         goto failed_bh;
170                 }
171                 if (crc != le32_to_cpu(sr->sr_sum)) {
172                         ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
173                         goto failed_bh;
174                 }
175         }
176         *pbh = bh_sr;
177         return 0;
178
179  failed_bh:
180         brelse(bh_sr);
181
182  failed:
183         return nilfs_warn_segment_error(ret);
184 }
185
186 /**
187  * nilfs_read_log_header - read summary header of the specified log
188  * @nilfs: nilfs object
189  * @start_blocknr: start block number of the log
190  * @sum: pointer to return segment summary structure
191  */
192 static struct buffer_head *
193 nilfs_read_log_header(struct the_nilfs *nilfs, sector_t start_blocknr,
194                       struct nilfs_segment_summary **sum)
195 {
196         struct buffer_head *bh_sum;
197
198         bh_sum = __bread(nilfs->ns_bdev, start_blocknr, nilfs->ns_blocksize);
199         if (bh_sum)
200                 *sum = (struct nilfs_segment_summary *)bh_sum->b_data;
201         return bh_sum;
202 }
203
204 /**
205  * nilfs_validate_log - verify consistency of log
206  * @nilfs: nilfs object
207  * @seg_seq: sequence number of segment
208  * @bh_sum: buffer head of summary block
209  * @sum: segment summary struct
210  */
211 static int nilfs_validate_log(struct the_nilfs *nilfs, u64 seg_seq,
212                               struct buffer_head *bh_sum,
213                               struct nilfs_segment_summary *sum)
214 {
215         unsigned long nblock;
216         u32 crc;
217         int ret;
218
219         ret = NILFS_SEG_FAIL_MAGIC;
220         if (le32_to_cpu(sum->ss_magic) != NILFS_SEGSUM_MAGIC)
221                 goto out;
222
223         ret = NILFS_SEG_FAIL_SEQ;
224         if (le64_to_cpu(sum->ss_seq) != seg_seq)
225                 goto out;
226
227         nblock = le32_to_cpu(sum->ss_nblocks);
228         ret = NILFS_SEG_FAIL_CONSISTENCY;
229         if (unlikely(nblock == 0 || nblock > nilfs->ns_blocks_per_segment))
230                 /* This limits the number of blocks read in the CRC check */
231                 goto out;
232
233         ret = NILFS_SEG_FAIL_IO;
234         if (nilfs_compute_checksum(nilfs, bh_sum, &crc, sizeof(sum->ss_datasum),
235                                    ((u64)nblock << nilfs->ns_blocksize_bits),
236                                    bh_sum->b_blocknr, nblock))
237                 goto out;
238
239         ret = NILFS_SEG_FAIL_CHECKSUM_FULL;
240         if (crc != le32_to_cpu(sum->ss_datasum))
241                 goto out;
242         ret = 0;
243 out:
244         return ret;
245 }
246
247 /**
248  * nilfs_read_summary_info - read an item on summary blocks of a log
249  * @nilfs: nilfs object
250  * @pbh: the current buffer head on summary blocks [in, out]
251  * @offset: the current byte offset on summary blocks [in, out]
252  * @bytes: byte size of the item to be read
253  */
254 static void *nilfs_read_summary_info(struct the_nilfs *nilfs,
255                                      struct buffer_head **pbh,
256                                      unsigned int *offset, unsigned int bytes)
257 {
258         void *ptr;
259         sector_t blocknr;
260
261         BUG_ON((*pbh)->b_size < *offset);
262         if (bytes > (*pbh)->b_size - *offset) {
263                 blocknr = (*pbh)->b_blocknr;
264                 brelse(*pbh);
265                 *pbh = __bread(nilfs->ns_bdev, blocknr + 1,
266                                nilfs->ns_blocksize);
267                 if (unlikely(!*pbh))
268                         return NULL;
269                 *offset = 0;
270         }
271         ptr = (*pbh)->b_data + *offset;
272         *offset += bytes;
273         return ptr;
274 }
275
276 /**
277  * nilfs_skip_summary_info - skip items on summary blocks of a log
278  * @nilfs: nilfs object
279  * @pbh: the current buffer head on summary blocks [in, out]
280  * @offset: the current byte offset on summary blocks [in, out]
281  * @bytes: byte size of the item to be skipped
282  * @count: number of items to be skipped
283  */
284 static void nilfs_skip_summary_info(struct the_nilfs *nilfs,
285                                     struct buffer_head **pbh,
286                                     unsigned int *offset, unsigned int bytes,
287                                     unsigned long count)
288 {
289         unsigned int rest_item_in_current_block
290                 = ((*pbh)->b_size - *offset) / bytes;
291
292         if (count <= rest_item_in_current_block) {
293                 *offset += bytes * count;
294         } else {
295                 sector_t blocknr = (*pbh)->b_blocknr;
296                 unsigned int nitem_per_block = (*pbh)->b_size / bytes;
297                 unsigned int bcnt;
298
299                 count -= rest_item_in_current_block;
300                 bcnt = DIV_ROUND_UP(count, nitem_per_block);
301                 *offset = bytes * (count - (bcnt - 1) * nitem_per_block);
302
303                 brelse(*pbh);
304                 *pbh = __bread(nilfs->ns_bdev, blocknr + bcnt,
305                                nilfs->ns_blocksize);
306         }
307 }
308
309 /**
310  * nilfs_scan_dsync_log - get block information of a log written for data sync
311  * @nilfs: nilfs object
312  * @start_blocknr: start block number of the log
313  * @sum: log summary information
314  * @head: list head to add nilfs_recovery_block struct
315  */
316 static int nilfs_scan_dsync_log(struct the_nilfs *nilfs, sector_t start_blocknr,
317                                 struct nilfs_segment_summary *sum,
318                                 struct list_head *head)
319 {
320         struct buffer_head *bh;
321         unsigned int offset;
322         u32 nfinfo, sumbytes;
323         sector_t blocknr;
324         ino_t ino;
325         int err = -EIO;
326
327         nfinfo = le32_to_cpu(sum->ss_nfinfo);
328         if (!nfinfo)
329                 return 0;
330
331         sumbytes = le32_to_cpu(sum->ss_sumbytes);
332         blocknr = start_blocknr + DIV_ROUND_UP(sumbytes, nilfs->ns_blocksize);
333         bh = __bread(nilfs->ns_bdev, start_blocknr, nilfs->ns_blocksize);
334         if (unlikely(!bh))
335                 goto out;
336
337         offset = le16_to_cpu(sum->ss_bytes);
338         for (;;) {
339                 unsigned long nblocks, ndatablk, nnodeblk;
340                 struct nilfs_finfo *finfo;
341
342                 finfo = nilfs_read_summary_info(nilfs, &bh, &offset,
343                                                 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 = nilfs_read_summary_info(nilfs, &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 data sync logs */
376                 nilfs_skip_summary_info(nilfs, &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 struct nilfs_segment_entry {
399         struct list_head        list;
400         __u64                   segnum;
401 };
402
403 static int nilfs_segment_list_add(struct list_head *head, __u64 segnum)
404 {
405         struct nilfs_segment_entry *ent = kmalloc(sizeof(*ent), GFP_NOFS);
406
407         if (unlikely(!ent))
408                 return -ENOMEM;
409
410         ent->segnum = segnum;
411         INIT_LIST_HEAD(&ent->list);
412         list_add_tail(&ent->list, head);
413         return 0;
414 }
415
416 void nilfs_dispose_segment_list(struct list_head *head)
417 {
418         while (!list_empty(head)) {
419                 struct nilfs_segment_entry *ent
420                         = list_entry(head->next,
421                                      struct nilfs_segment_entry, list);
422                 list_del(&ent->list);
423                 kfree(ent);
424         }
425 }
426
427 static int nilfs_prepare_segment_for_recovery(struct the_nilfs *nilfs,
428                                               struct nilfs_sb_info *sbi,
429                                               struct nilfs_recovery_info *ri)
430 {
431         struct list_head *head = &ri->ri_used_segments;
432         struct nilfs_segment_entry *ent, *n;
433         struct inode *sufile = nilfs->ns_sufile;
434         __u64 segnum[4];
435         int err;
436         int i;
437
438         segnum[0] = nilfs->ns_segnum;
439         segnum[1] = nilfs->ns_nextnum;
440         segnum[2] = ri->ri_segnum;
441         segnum[3] = ri->ri_nextnum;
442
443         nilfs_attach_writer(nilfs, sbi);
444         /*
445          * Releasing the next segment of the latest super root.
446          * The next segment is invalidated by this recovery.
447          */
448         err = nilfs_sufile_free(sufile, segnum[1]);
449         if (unlikely(err))
450                 goto failed;
451
452         for (i = 1; i < 4; i++) {
453                 err = nilfs_segment_list_add(head, segnum[i]);
454                 if (unlikely(err))
455                         goto failed;
456         }
457
458         /*
459          * Collecting segments written after the latest super root.
460          * These are marked dirty to avoid being reallocated in the next write.
461          */
462         list_for_each_entry_safe(ent, n, head, list) {
463                 if (ent->segnum != segnum[0]) {
464                         err = nilfs_sufile_scrap(sufile, ent->segnum);
465                         if (unlikely(err))
466                                 goto failed;
467                 }
468                 list_del(&ent->list);
469                 kfree(ent);
470         }
471
472         /* Allocate new segments for recovery */
473         err = nilfs_sufile_alloc(sufile, &segnum[0]);
474         if (unlikely(err))
475                 goto failed;
476
477         nilfs->ns_pseg_offset = 0;
478         nilfs->ns_seg_seq = ri->ri_seq + 2;
479         nilfs->ns_nextnum = nilfs->ns_segnum = segnum[0];
480
481  failed:
482         /* No need to recover sufile because it will be destroyed on error */
483         nilfs_detach_writer(nilfs, sbi);
484         return err;
485 }
486
487 static int nilfs_recovery_copy_block(struct the_nilfs *nilfs,
488                                      struct nilfs_recovery_block *rb,
489                                      struct page *page)
490 {
491         struct buffer_head *bh_org;
492         void *kaddr;
493
494         bh_org = __bread(nilfs->ns_bdev, rb->blocknr, nilfs->ns_blocksize);
495         if (unlikely(!bh_org))
496                 return -EIO;
497
498         kaddr = kmap_atomic(page, KM_USER0);
499         memcpy(kaddr + bh_offset(bh_org), bh_org->b_data, bh_org->b_size);
500         kunmap_atomic(kaddr, KM_USER0);
501         brelse(bh_org);
502         return 0;
503 }
504
505 static int nilfs_recover_dsync_blocks(struct the_nilfs *nilfs,
506                                       struct nilfs_sb_info *sbi,
507                                       struct nilfs_root *root,
508                                       struct list_head *head,
509                                       unsigned long *nr_salvaged_blocks)
510 {
511         struct inode *inode;
512         struct nilfs_recovery_block *rb, *n;
513         unsigned blocksize = nilfs->ns_blocksize;
514         struct page *page;
515         loff_t pos;
516         int err = 0, err2 = 0;
517
518         list_for_each_entry_safe(rb, n, head, list) {
519                 inode = nilfs_iget(sbi->s_super, root, rb->ino);
520                 if (IS_ERR(inode)) {
521                         err = PTR_ERR(inode);
522                         inode = NULL;
523                         goto failed_inode;
524                 }
525
526                 pos = rb->blkoff << inode->i_blkbits;
527                 err = block_write_begin(inode->i_mapping, pos, blocksize,
528                                         0, &page, nilfs_get_block);
529                 if (unlikely(err)) {
530                         loff_t isize = inode->i_size;
531                         if (pos + blocksize > isize)
532                                 vmtruncate(inode, isize);
533                         goto failed_inode;
534                 }
535
536                 err = nilfs_recovery_copy_block(nilfs, rb, page);
537                 if (unlikely(err))
538                         goto failed_page;
539
540                 err = nilfs_set_file_dirty(sbi, inode, 1);
541                 if (unlikely(err))
542                         goto failed_page;
543
544                 block_write_end(NULL, inode->i_mapping, pos, blocksize,
545                                 blocksize, page, NULL);
546
547                 unlock_page(page);
548                 page_cache_release(page);
549
550                 (*nr_salvaged_blocks)++;
551                 goto next;
552
553  failed_page:
554                 unlock_page(page);
555                 page_cache_release(page);
556
557  failed_inode:
558                 printk(KERN_WARNING
559                        "NILFS warning: error recovering data block "
560                        "(err=%d, ino=%lu, block-offset=%llu)\n",
561                        err, (unsigned long)rb->ino,
562                        (unsigned long long)rb->blkoff);
563                 if (!err2)
564                         err2 = err;
565  next:
566                 iput(inode); /* iput(NULL) is just ignored */
567                 list_del_init(&rb->list);
568                 kfree(rb);
569         }
570         return err2;
571 }
572
573 /**
574  * nilfs_do_roll_forward - salvage logical segments newer than the latest
575  * checkpoint
576  * @nilfs: nilfs object
577  * @sbi: nilfs_sb_info
578  * @ri: pointer to a nilfs_recovery_info
579  */
580 static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
581                                  struct nilfs_sb_info *sbi,
582                                  struct nilfs_root *root,
583                                  struct nilfs_recovery_info *ri)
584 {
585         struct buffer_head *bh_sum = NULL;
586         struct nilfs_segment_summary *sum;
587         sector_t pseg_start;
588         sector_t seg_start, seg_end;  /* Starting/ending DBN of full segment */
589         unsigned long nsalvaged_blocks = 0;
590         unsigned int flags;
591         u64 seg_seq;
592         __u64 segnum, nextnum = 0;
593         int empty_seg = 0;
594         int err = 0, ret;
595         LIST_HEAD(dsync_blocks);  /* list of data blocks to be recovered */
596         enum {
597                 RF_INIT_ST,
598                 RF_DSYNC_ST,   /* scanning data-sync segments */
599         };
600         int state = RF_INIT_ST;
601
602         nilfs_attach_writer(nilfs, sbi);
603         pseg_start = ri->ri_lsegs_start;
604         seg_seq = ri->ri_lsegs_start_seq;
605         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
606         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
607
608         while (segnum != ri->ri_segnum || pseg_start <= ri->ri_pseg_start) {
609                 brelse(bh_sum);
610                 bh_sum = nilfs_read_log_header(nilfs, pseg_start, &sum);
611                 if (!bh_sum) {
612                         err = -EIO;
613                         goto failed;
614                 }
615
616                 ret = nilfs_validate_log(nilfs, seg_seq, bh_sum, sum);
617                 if (ret) {
618                         if (ret == NILFS_SEG_FAIL_IO) {
619                                 err = -EIO;
620                                 goto failed;
621                         }
622                         goto strayed;
623                 }
624
625                 flags = le16_to_cpu(sum->ss_flags);
626                 if (flags & NILFS_SS_SR)
627                         goto confused;
628
629                 /* Found a valid partial segment; do recovery actions */
630                 nextnum = nilfs_get_segnum_of_block(nilfs,
631                                                     le64_to_cpu(sum->ss_next));
632                 empty_seg = 0;
633                 nilfs->ns_ctime = le64_to_cpu(sum->ss_create);
634                 if (!(flags & NILFS_SS_GC))
635                         nilfs->ns_nongc_ctime = nilfs->ns_ctime;
636
637                 switch (state) {
638                 case RF_INIT_ST:
639                         if (!(flags & NILFS_SS_LOGBGN) ||
640                             !(flags & NILFS_SS_SYNDT))
641                                 goto try_next_pseg;
642                         state = RF_DSYNC_ST;
643                         /* Fall through */
644                 case RF_DSYNC_ST:
645                         if (!(flags & NILFS_SS_SYNDT))
646                                 goto confused;
647
648                         err = nilfs_scan_dsync_log(nilfs, pseg_start, sum,
649                                                    &dsync_blocks);
650                         if (unlikely(err))
651                                 goto failed;
652                         if (flags & NILFS_SS_LOGEND) {
653                                 err = nilfs_recover_dsync_blocks(
654                                         nilfs, sbi, root, &dsync_blocks,
655                                         &nsalvaged_blocks);
656                                 if (unlikely(err))
657                                         goto failed;
658                                 state = RF_INIT_ST;
659                         }
660                         break; /* Fall through to try_next_pseg */
661                 }
662
663  try_next_pseg:
664                 if (pseg_start == ri->ri_lsegs_end)
665                         break;
666                 pseg_start += le32_to_cpu(sum->ss_nblocks);
667                 if (pseg_start < seg_end)
668                         continue;
669                 goto feed_segment;
670
671  strayed:
672                 if (pseg_start == ri->ri_lsegs_end)
673                         break;
674
675  feed_segment:
676                 /* Looking to the next full segment */
677                 if (empty_seg++)
678                         break;
679                 seg_seq++;
680                 segnum = nextnum;
681                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
682                 pseg_start = seg_start;
683         }
684
685         if (nsalvaged_blocks) {
686                 printk(KERN_INFO "NILFS (device %s): salvaged %lu blocks\n",
687                        sbi->s_super->s_id, nsalvaged_blocks);
688                 ri->ri_need_recovery = NILFS_RECOVERY_ROLLFORWARD_DONE;
689         }
690  out:
691         brelse(bh_sum);
692         dispose_recovery_list(&dsync_blocks);
693         nilfs_detach_writer(nilfs, sbi);
694         return err;
695
696  confused:
697         err = -EINVAL;
698  failed:
699         printk(KERN_ERR
700                "NILFS (device %s): Error roll-forwarding "
701                "(err=%d, pseg block=%llu). ",
702                sbi->s_super->s_id, err, (unsigned long long)pseg_start);
703         goto out;
704 }
705
706 static void nilfs_finish_roll_forward(struct the_nilfs *nilfs,
707                                       struct nilfs_recovery_info *ri)
708 {
709         struct buffer_head *bh;
710         int err;
711
712         if (nilfs_get_segnum_of_block(nilfs, ri->ri_lsegs_start) !=
713             nilfs_get_segnum_of_block(nilfs, ri->ri_super_root))
714                 return;
715
716         bh = __getblk(nilfs->ns_bdev, ri->ri_lsegs_start, nilfs->ns_blocksize);
717         BUG_ON(!bh);
718         memset(bh->b_data, 0, bh->b_size);
719         set_buffer_dirty(bh);
720         err = sync_dirty_buffer(bh);
721         if (unlikely(err))
722                 printk(KERN_WARNING
723                        "NILFS warning: buffer sync write failed during "
724                        "post-cleaning of recovery.\n");
725         brelse(bh);
726 }
727
728 /**
729  * nilfs_salvage_orphan_logs - salvage logs written after the latest checkpoint
730  * @nilfs: nilfs object
731  * @sbi: nilfs_sb_info
732  * @ri: pointer to a nilfs_recovery_info struct to store search results.
733  *
734  * Return Value: On success, 0 is returned.  On error, one of the following
735  * negative error code is returned.
736  *
737  * %-EINVAL - Inconsistent filesystem state.
738  *
739  * %-EIO - I/O error
740  *
741  * %-ENOSPC - No space left on device (only in a panic state).
742  *
743  * %-ERESTARTSYS - Interrupted.
744  *
745  * %-ENOMEM - Insufficient memory available.
746  */
747 int nilfs_salvage_orphan_logs(struct the_nilfs *nilfs,
748                               struct nilfs_sb_info *sbi,
749                               struct nilfs_recovery_info *ri)
750 {
751         struct nilfs_root *root;
752         int err;
753
754         if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0)
755                 return 0;
756
757         err = nilfs_attach_checkpoint(sbi, ri->ri_cno, true, &root);
758         if (unlikely(err)) {
759                 printk(KERN_ERR
760                        "NILFS: error loading the latest checkpoint.\n");
761                 return err;
762         }
763
764         err = nilfs_do_roll_forward(nilfs, sbi, root, ri);
765         if (unlikely(err))
766                 goto failed;
767
768         if (ri->ri_need_recovery == NILFS_RECOVERY_ROLLFORWARD_DONE) {
769                 err = nilfs_prepare_segment_for_recovery(nilfs, sbi, ri);
770                 if (unlikely(err)) {
771                         printk(KERN_ERR "NILFS: Error preparing segments for "
772                                "recovery.\n");
773                         goto failed;
774                 }
775
776                 err = nilfs_attach_segment_constructor(sbi);
777                 if (unlikely(err))
778                         goto failed;
779
780                 set_nilfs_discontinued(nilfs);
781                 err = nilfs_construct_segment(sbi->s_super);
782                 nilfs_detach_segment_constructor(sbi);
783
784                 if (unlikely(err)) {
785                         printk(KERN_ERR "NILFS: Oops! recovery failed. "
786                                "(err=%d)\n", err);
787                         goto failed;
788                 }
789
790                 nilfs_finish_roll_forward(nilfs, ri);
791         }
792
793  failed:
794         nilfs_detach_checkpoint(sbi);
795         nilfs_put_root(root);
796         return err;
797 }
798
799 /**
800  * nilfs_search_super_root - search the latest valid super root
801  * @nilfs: the_nilfs
802  * @ri: pointer to a nilfs_recovery_info struct to store search results.
803  *
804  * nilfs_search_super_root() looks for the latest super-root from a partial
805  * segment pointed by the superblock.  It sets up struct the_nilfs through
806  * this search. It fills nilfs_recovery_info (ri) required for recovery.
807  *
808  * Return Value: On success, 0 is returned.  On error, one of the following
809  * negative error code is returned.
810  *
811  * %-EINVAL - No valid segment found
812  *
813  * %-EIO - I/O error
814  *
815  * %-ENOMEM - Insufficient memory available.
816  */
817 int nilfs_search_super_root(struct the_nilfs *nilfs,
818                             struct nilfs_recovery_info *ri)
819 {
820         struct buffer_head *bh_sum = NULL;
821         struct nilfs_segment_summary *sum;
822         sector_t pseg_start, pseg_end, sr_pseg_start = 0;
823         sector_t seg_start, seg_end; /* range of full segment (block number) */
824         sector_t b, end;
825         unsigned long nblocks;
826         unsigned int flags;
827         u64 seg_seq;
828         __u64 segnum, nextnum = 0;
829         __u64 cno;
830         LIST_HEAD(segments);
831         int empty_seg = 0, scan_newer = 0;
832         int ret;
833
834         pseg_start = nilfs->ns_last_pseg;
835         seg_seq = nilfs->ns_last_seq;
836         cno = nilfs->ns_last_cno;
837         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
838
839         /* Calculate range of segment */
840         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
841
842         /* Read ahead segment */
843         b = seg_start;
844         while (b <= seg_end)
845                 __breadahead(nilfs->ns_bdev, b++, nilfs->ns_blocksize);
846
847         for (;;) {
848                 brelse(bh_sum);
849                 ret = NILFS_SEG_FAIL_IO;
850                 bh_sum = nilfs_read_log_header(nilfs, pseg_start, &sum);
851                 if (!bh_sum)
852                         goto failed;
853
854                 ret = nilfs_validate_log(nilfs, seg_seq, bh_sum, sum);
855                 if (ret) {
856                         if (ret == NILFS_SEG_FAIL_IO)
857                                 goto failed;
858                         goto strayed;
859                 }
860
861                 nblocks = le32_to_cpu(sum->ss_nblocks);
862                 pseg_end = pseg_start + nblocks - 1;
863                 if (unlikely(pseg_end > seg_end)) {
864                         ret = NILFS_SEG_FAIL_CONSISTENCY;
865                         goto strayed;
866                 }
867
868                 /* A valid partial segment */
869                 ri->ri_pseg_start = pseg_start;
870                 ri->ri_seq = seg_seq;
871                 ri->ri_segnum = segnum;
872                 nextnum = nilfs_get_segnum_of_block(nilfs,
873                                                     le64_to_cpu(sum->ss_next));
874                 ri->ri_nextnum = nextnum;
875                 empty_seg = 0;
876
877                 flags = le16_to_cpu(sum->ss_flags);
878                 if (!(flags & NILFS_SS_SR) && !scan_newer) {
879                         /* This will never happen because a superblock
880                            (last_segment) always points to a pseg
881                            having a super root. */
882                         ret = NILFS_SEG_FAIL_CONSISTENCY;
883                         goto failed;
884                 }
885
886                 if (pseg_start == seg_start) {
887                         nilfs_get_segment_range(nilfs, nextnum, &b, &end);
888                         while (b <= end)
889                                 __breadahead(nilfs->ns_bdev, b++,
890                                              nilfs->ns_blocksize);
891                 }
892                 if (!(flags & NILFS_SS_SR)) {
893                         if (!ri->ri_lsegs_start && (flags & NILFS_SS_LOGBGN)) {
894                                 ri->ri_lsegs_start = pseg_start;
895                                 ri->ri_lsegs_start_seq = seg_seq;
896                         }
897                         if (flags & NILFS_SS_LOGEND)
898                                 ri->ri_lsegs_end = pseg_start;
899                         goto try_next_pseg;
900                 }
901
902                 /* A valid super root was found. */
903                 ri->ri_cno = cno++;
904                 ri->ri_super_root = pseg_end;
905                 ri->ri_lsegs_start = ri->ri_lsegs_end = 0;
906
907                 nilfs_dispose_segment_list(&segments);
908                 sr_pseg_start = pseg_start;
909                 nilfs->ns_pseg_offset = pseg_start + nblocks - seg_start;
910                 nilfs->ns_seg_seq = seg_seq;
911                 nilfs->ns_segnum = segnum;
912                 nilfs->ns_cno = cno;  /* nilfs->ns_cno = ri->ri_cno + 1 */
913                 nilfs->ns_ctime = le64_to_cpu(sum->ss_create);
914                 nilfs->ns_nextnum = nextnum;
915
916                 if (scan_newer)
917                         ri->ri_need_recovery = NILFS_RECOVERY_SR_UPDATED;
918                 else {
919                         if (nilfs->ns_mount_state & NILFS_VALID_FS)
920                                 goto super_root_found;
921                         scan_newer = 1;
922                 }
923
924  try_next_pseg:
925                 /* Standing on a course, or met an inconsistent state */
926                 pseg_start += nblocks;
927                 if (pseg_start < seg_end)
928                         continue;
929                 goto feed_segment;
930
931  strayed:
932                 /* Off the trail */
933                 if (!scan_newer)
934                         /*
935                          * This can happen if a checkpoint was written without
936                          * barriers, or as a result of an I/O failure.
937                          */
938                         goto failed;
939
940  feed_segment:
941                 /* Looking to the next full segment */
942                 if (empty_seg++)
943                         goto super_root_found; /* found a valid super root */
944
945                 ret = nilfs_segment_list_add(&segments, segnum);
946                 if (unlikely(ret))
947                         goto failed;
948
949                 seg_seq++;
950                 segnum = nextnum;
951                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
952                 pseg_start = seg_start;
953         }
954
955  super_root_found:
956         /* Updating pointers relating to the latest checkpoint */
957         brelse(bh_sum);
958         list_splice_tail(&segments, &ri->ri_used_segments);
959         nilfs->ns_last_pseg = sr_pseg_start;
960         nilfs->ns_last_seq = nilfs->ns_seg_seq;
961         nilfs->ns_last_cno = ri->ri_cno;
962         return 0;
963
964  failed:
965         brelse(bh_sum);
966         nilfs_dispose_segment_list(&segments);
967         return (ret < 0) ? ret : nilfs_warn_segment_error(ret);
968 }