RISC-V: Fix a race condition during kernel stack overflow
[platform/kernel/linux-starfive.git] / fs / nilfs2 / recovery.c
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * NILFS recovery logic
4  *
5  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6  *
7  * Written by Ryusuke Konishi.
8  */
9
10 #include <linux/buffer_head.h>
11 #include <linux/blkdev.h>
12 #include <linux/swap.h>
13 #include <linux/slab.h>
14 #include <linux/crc32.h>
15 #include "nilfs.h"
16 #include "segment.h"
17 #include "sufile.h"
18 #include "page.h"
19 #include "segbuf.h"
20
21 /*
22  * Segment check result
23  */
24 enum {
25         NILFS_SEG_VALID,
26         NILFS_SEG_NO_SUPER_ROOT,
27         NILFS_SEG_FAIL_IO,
28         NILFS_SEG_FAIL_MAGIC,
29         NILFS_SEG_FAIL_SEQ,
30         NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT,
31         NILFS_SEG_FAIL_CHECKSUM_FULL,
32         NILFS_SEG_FAIL_CONSISTENCY,
33 };
34
35 /* work structure for recovery */
36 struct nilfs_recovery_block {
37         ino_t ino;              /*
38                                  * Inode number of the file that this block
39                                  * belongs to
40                                  */
41         sector_t blocknr;       /* block number */
42         __u64 vblocknr;         /* virtual block number */
43         unsigned long blkoff;   /* File offset of the data block (per block) */
44         struct list_head list;
45 };
46
47
48 static int nilfs_warn_segment_error(struct super_block *sb, int err)
49 {
50         const char *msg = NULL;
51
52         switch (err) {
53         case NILFS_SEG_FAIL_IO:
54                 nilfs_err(sb, "I/O error reading segment");
55                 return -EIO;
56         case NILFS_SEG_FAIL_MAGIC:
57                 msg = "Magic number mismatch";
58                 break;
59         case NILFS_SEG_FAIL_SEQ:
60                 msg = "Sequence number mismatch";
61                 break;
62         case NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT:
63                 msg = "Checksum error in super root";
64                 break;
65         case NILFS_SEG_FAIL_CHECKSUM_FULL:
66                 msg = "Checksum error in segment payload";
67                 break;
68         case NILFS_SEG_FAIL_CONSISTENCY:
69                 msg = "Inconsistency found";
70                 break;
71         case NILFS_SEG_NO_SUPER_ROOT:
72                 msg = "No super root in the last segment";
73                 break;
74         default:
75                 nilfs_err(sb, "unrecognized segment error %d", err);
76                 return -EINVAL;
77         }
78         nilfs_warn(sb, "invalid segment: %s", msg);
79         return -EINVAL;
80 }
81
82 /**
83  * nilfs_compute_checksum - compute checksum of blocks continuously
84  * @nilfs: nilfs object
85  * @bhs: buffer head of start block
86  * @sum: place to store result
87  * @offset: offset bytes in the first block
88  * @check_bytes: number of bytes to be checked
89  * @start: DBN of start block
90  * @nblock: number of blocks to be checked
91  */
92 static int nilfs_compute_checksum(struct the_nilfs *nilfs,
93                                   struct buffer_head *bhs, u32 *sum,
94                                   unsigned long offset, u64 check_bytes,
95                                   sector_t start, unsigned long nblock)
96 {
97         unsigned int blocksize = nilfs->ns_blocksize;
98         unsigned long size;
99         u32 crc;
100
101         BUG_ON(offset >= blocksize);
102         check_bytes -= offset;
103         size = min_t(u64, check_bytes, blocksize - offset);
104         crc = crc32_le(nilfs->ns_crc_seed,
105                        (unsigned char *)bhs->b_data + offset, size);
106         if (--nblock > 0) {
107                 do {
108                         struct buffer_head *bh;
109
110                         bh = __bread(nilfs->ns_bdev, ++start, blocksize);
111                         if (!bh)
112                                 return -EIO;
113                         check_bytes -= size;
114                         size = min_t(u64, check_bytes, blocksize);
115                         crc = crc32_le(crc, bh->b_data, size);
116                         brelse(bh);
117                 } while (--nblock > 0);
118         }
119         *sum = crc;
120         return 0;
121 }
122
123 /**
124  * nilfs_read_super_root_block - read super root block
125  * @nilfs: nilfs object
126  * @sr_block: disk block number of the super root block
127  * @pbh: address of a buffer_head pointer to return super root buffer
128  * @check: CRC check flag
129  */
130 int nilfs_read_super_root_block(struct the_nilfs *nilfs, sector_t sr_block,
131                                 struct buffer_head **pbh, int check)
132 {
133         struct buffer_head *bh_sr;
134         struct nilfs_super_root *sr;
135         u32 crc;
136         int ret;
137
138         *pbh = NULL;
139         bh_sr = __bread(nilfs->ns_bdev, sr_block, nilfs->ns_blocksize);
140         if (unlikely(!bh_sr)) {
141                 ret = NILFS_SEG_FAIL_IO;
142                 goto failed;
143         }
144
145         sr = (struct nilfs_super_root *)bh_sr->b_data;
146         if (check) {
147                 unsigned int bytes = le16_to_cpu(sr->sr_bytes);
148
149                 if (bytes == 0 || bytes > nilfs->ns_blocksize) {
150                         ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
151                         goto failed_bh;
152                 }
153                 if (nilfs_compute_checksum(
154                             nilfs, bh_sr, &crc, sizeof(sr->sr_sum), bytes,
155                             sr_block, 1)) {
156                         ret = NILFS_SEG_FAIL_IO;
157                         goto failed_bh;
158                 }
159                 if (crc != le32_to_cpu(sr->sr_sum)) {
160                         ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
161                         goto failed_bh;
162                 }
163         }
164         *pbh = bh_sr;
165         return 0;
166
167  failed_bh:
168         brelse(bh_sr);
169
170  failed:
171         return nilfs_warn_segment_error(nilfs->ns_sb, ret);
172 }
173
174 /**
175  * nilfs_read_log_header - read summary header of the specified log
176  * @nilfs: nilfs object
177  * @start_blocknr: start block number of the log
178  * @sum: pointer to return segment summary structure
179  */
180 static struct buffer_head *
181 nilfs_read_log_header(struct the_nilfs *nilfs, sector_t start_blocknr,
182                       struct nilfs_segment_summary **sum)
183 {
184         struct buffer_head *bh_sum;
185
186         bh_sum = __bread(nilfs->ns_bdev, start_blocknr, nilfs->ns_blocksize);
187         if (bh_sum)
188                 *sum = (struct nilfs_segment_summary *)bh_sum->b_data;
189         return bh_sum;
190 }
191
192 /**
193  * nilfs_validate_log - verify consistency of log
194  * @nilfs: nilfs object
195  * @seg_seq: sequence number of segment
196  * @bh_sum: buffer head of summary block
197  * @sum: segment summary struct
198  */
199 static int nilfs_validate_log(struct the_nilfs *nilfs, u64 seg_seq,
200                               struct buffer_head *bh_sum,
201                               struct nilfs_segment_summary *sum)
202 {
203         unsigned long nblock;
204         u32 crc;
205         int ret;
206
207         ret = NILFS_SEG_FAIL_MAGIC;
208         if (le32_to_cpu(sum->ss_magic) != NILFS_SEGSUM_MAGIC)
209                 goto out;
210
211         ret = NILFS_SEG_FAIL_SEQ;
212         if (le64_to_cpu(sum->ss_seq) != seg_seq)
213                 goto out;
214
215         nblock = le32_to_cpu(sum->ss_nblocks);
216         ret = NILFS_SEG_FAIL_CONSISTENCY;
217         if (unlikely(nblock == 0 || nblock > nilfs->ns_blocks_per_segment))
218                 /* This limits the number of blocks read in the CRC check */
219                 goto out;
220
221         ret = NILFS_SEG_FAIL_IO;
222         if (nilfs_compute_checksum(nilfs, bh_sum, &crc, sizeof(sum->ss_datasum),
223                                    ((u64)nblock << nilfs->ns_blocksize_bits),
224                                    bh_sum->b_blocknr, nblock))
225                 goto out;
226
227         ret = NILFS_SEG_FAIL_CHECKSUM_FULL;
228         if (crc != le32_to_cpu(sum->ss_datasum))
229                 goto out;
230         ret = 0;
231 out:
232         return ret;
233 }
234
235 /**
236  * nilfs_read_summary_info - read an item on summary blocks of a log
237  * @nilfs: nilfs object
238  * @pbh: the current buffer head on summary blocks [in, out]
239  * @offset: the current byte offset on summary blocks [in, out]
240  * @bytes: byte size of the item to be read
241  */
242 static void *nilfs_read_summary_info(struct the_nilfs *nilfs,
243                                      struct buffer_head **pbh,
244                                      unsigned int *offset, unsigned int bytes)
245 {
246         void *ptr;
247         sector_t blocknr;
248
249         BUG_ON((*pbh)->b_size < *offset);
250         if (bytes > (*pbh)->b_size - *offset) {
251                 blocknr = (*pbh)->b_blocknr;
252                 brelse(*pbh);
253                 *pbh = __bread(nilfs->ns_bdev, blocknr + 1,
254                                nilfs->ns_blocksize);
255                 if (unlikely(!*pbh))
256                         return NULL;
257                 *offset = 0;
258         }
259         ptr = (*pbh)->b_data + *offset;
260         *offset += bytes;
261         return ptr;
262 }
263
264 /**
265  * nilfs_skip_summary_info - skip items on summary blocks of a log
266  * @nilfs: nilfs object
267  * @pbh: the current buffer head on summary blocks [in, out]
268  * @offset: the current byte offset on summary blocks [in, out]
269  * @bytes: byte size of the item to be skipped
270  * @count: number of items to be skipped
271  */
272 static void nilfs_skip_summary_info(struct the_nilfs *nilfs,
273                                     struct buffer_head **pbh,
274                                     unsigned int *offset, unsigned int bytes,
275                                     unsigned long count)
276 {
277         unsigned int rest_item_in_current_block
278                 = ((*pbh)->b_size - *offset) / bytes;
279
280         if (count <= rest_item_in_current_block) {
281                 *offset += bytes * count;
282         } else {
283                 sector_t blocknr = (*pbh)->b_blocknr;
284                 unsigned int nitem_per_block = (*pbh)->b_size / bytes;
285                 unsigned int bcnt;
286
287                 count -= rest_item_in_current_block;
288                 bcnt = DIV_ROUND_UP(count, nitem_per_block);
289                 *offset = bytes * (count - (bcnt - 1) * nitem_per_block);
290
291                 brelse(*pbh);
292                 *pbh = __bread(nilfs->ns_bdev, blocknr + bcnt,
293                                nilfs->ns_blocksize);
294         }
295 }
296
297 /**
298  * nilfs_scan_dsync_log - get block information of a log written for data sync
299  * @nilfs: nilfs object
300  * @start_blocknr: start block number of the log
301  * @sum: log summary information
302  * @head: list head to add nilfs_recovery_block struct
303  */
304 static int nilfs_scan_dsync_log(struct the_nilfs *nilfs, sector_t start_blocknr,
305                                 struct nilfs_segment_summary *sum,
306                                 struct list_head *head)
307 {
308         struct buffer_head *bh;
309         unsigned int offset;
310         u32 nfinfo, sumbytes;
311         sector_t blocknr;
312         ino_t ino;
313         int err = -EIO;
314
315         nfinfo = le32_to_cpu(sum->ss_nfinfo);
316         if (!nfinfo)
317                 return 0;
318
319         sumbytes = le32_to_cpu(sum->ss_sumbytes);
320         blocknr = start_blocknr + DIV_ROUND_UP(sumbytes, nilfs->ns_blocksize);
321         bh = __bread(nilfs->ns_bdev, start_blocknr, nilfs->ns_blocksize);
322         if (unlikely(!bh))
323                 goto out;
324
325         offset = le16_to_cpu(sum->ss_bytes);
326         for (;;) {
327                 unsigned long nblocks, ndatablk, nnodeblk;
328                 struct nilfs_finfo *finfo;
329
330                 finfo = nilfs_read_summary_info(nilfs, &bh, &offset,
331                                                 sizeof(*finfo));
332                 if (unlikely(!finfo))
333                         goto out;
334
335                 ino = le64_to_cpu(finfo->fi_ino);
336                 nblocks = le32_to_cpu(finfo->fi_nblocks);
337                 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
338                 nnodeblk = nblocks - ndatablk;
339
340                 while (ndatablk-- > 0) {
341                         struct nilfs_recovery_block *rb;
342                         struct nilfs_binfo_v *binfo;
343
344                         binfo = nilfs_read_summary_info(nilfs, &bh, &offset,
345                                                         sizeof(*binfo));
346                         if (unlikely(!binfo))
347                                 goto out;
348
349                         rb = kmalloc(sizeof(*rb), GFP_NOFS);
350                         if (unlikely(!rb)) {
351                                 err = -ENOMEM;
352                                 goto out;
353                         }
354                         rb->ino = ino;
355                         rb->blocknr = blocknr++;
356                         rb->vblocknr = le64_to_cpu(binfo->bi_vblocknr);
357                         rb->blkoff = le64_to_cpu(binfo->bi_blkoff);
358                         /* INIT_LIST_HEAD(&rb->list); */
359                         list_add_tail(&rb->list, head);
360                 }
361                 if (--nfinfo == 0)
362                         break;
363                 blocknr += nnodeblk; /* always 0 for data sync logs */
364                 nilfs_skip_summary_info(nilfs, &bh, &offset, sizeof(__le64),
365                                         nnodeblk);
366                 if (unlikely(!bh))
367                         goto out;
368         }
369         err = 0;
370  out:
371         brelse(bh);   /* brelse(NULL) is just ignored */
372         return err;
373 }
374
375 static void dispose_recovery_list(struct list_head *head)
376 {
377         while (!list_empty(head)) {
378                 struct nilfs_recovery_block *rb;
379
380                 rb = list_first_entry(head, struct nilfs_recovery_block, list);
381                 list_del(&rb->list);
382                 kfree(rb);
383         }
384 }
385
386 struct nilfs_segment_entry {
387         struct list_head        list;
388         __u64                   segnum;
389 };
390
391 static int nilfs_segment_list_add(struct list_head *head, __u64 segnum)
392 {
393         struct nilfs_segment_entry *ent = kmalloc(sizeof(*ent), GFP_NOFS);
394
395         if (unlikely(!ent))
396                 return -ENOMEM;
397
398         ent->segnum = segnum;
399         INIT_LIST_HEAD(&ent->list);
400         list_add_tail(&ent->list, head);
401         return 0;
402 }
403
404 void nilfs_dispose_segment_list(struct list_head *head)
405 {
406         while (!list_empty(head)) {
407                 struct nilfs_segment_entry *ent;
408
409                 ent = list_first_entry(head, struct nilfs_segment_entry, list);
410                 list_del(&ent->list);
411                 kfree(ent);
412         }
413 }
414
415 static int nilfs_prepare_segment_for_recovery(struct the_nilfs *nilfs,
416                                               struct super_block *sb,
417                                               struct nilfs_recovery_info *ri)
418 {
419         struct list_head *head = &ri->ri_used_segments;
420         struct nilfs_segment_entry *ent, *n;
421         struct inode *sufile = nilfs->ns_sufile;
422         __u64 segnum[4];
423         int err;
424         int i;
425
426         segnum[0] = nilfs->ns_segnum;
427         segnum[1] = nilfs->ns_nextnum;
428         segnum[2] = ri->ri_segnum;
429         segnum[3] = ri->ri_nextnum;
430
431         /*
432          * Releasing the next segment of the latest super root.
433          * The next segment is invalidated by this recovery.
434          */
435         err = nilfs_sufile_free(sufile, segnum[1]);
436         if (unlikely(err))
437                 goto failed;
438
439         for (i = 1; i < 4; i++) {
440                 err = nilfs_segment_list_add(head, segnum[i]);
441                 if (unlikely(err))
442                         goto failed;
443         }
444
445         /*
446          * Collecting segments written after the latest super root.
447          * These are marked dirty to avoid being reallocated in the next write.
448          */
449         list_for_each_entry_safe(ent, n, head, list) {
450                 if (ent->segnum != segnum[0]) {
451                         err = nilfs_sufile_scrap(sufile, ent->segnum);
452                         if (unlikely(err))
453                                 goto failed;
454                 }
455                 list_del(&ent->list);
456                 kfree(ent);
457         }
458
459         /* Allocate new segments for recovery */
460         err = nilfs_sufile_alloc(sufile, &segnum[0]);
461         if (unlikely(err))
462                 goto failed;
463
464         nilfs->ns_pseg_offset = 0;
465         nilfs->ns_seg_seq = ri->ri_seq + 2;
466         nilfs->ns_nextnum = nilfs->ns_segnum = segnum[0];
467
468  failed:
469         /* No need to recover sufile because it will be destroyed on error */
470         return err;
471 }
472
473 static int nilfs_recovery_copy_block(struct the_nilfs *nilfs,
474                                      struct nilfs_recovery_block *rb,
475                                      struct page *page)
476 {
477         struct buffer_head *bh_org;
478         void *kaddr;
479
480         bh_org = __bread(nilfs->ns_bdev, rb->blocknr, nilfs->ns_blocksize);
481         if (unlikely(!bh_org))
482                 return -EIO;
483
484         kaddr = kmap_atomic(page);
485         memcpy(kaddr + bh_offset(bh_org), bh_org->b_data, bh_org->b_size);
486         kunmap_atomic(kaddr);
487         brelse(bh_org);
488         return 0;
489 }
490
491 static int nilfs_recover_dsync_blocks(struct the_nilfs *nilfs,
492                                       struct super_block *sb,
493                                       struct nilfs_root *root,
494                                       struct list_head *head,
495                                       unsigned long *nr_salvaged_blocks)
496 {
497         struct inode *inode;
498         struct nilfs_recovery_block *rb, *n;
499         unsigned int blocksize = nilfs->ns_blocksize;
500         struct page *page;
501         loff_t pos;
502         int err = 0, err2 = 0;
503
504         list_for_each_entry_safe(rb, n, head, list) {
505                 inode = nilfs_iget(sb, root, rb->ino);
506                 if (IS_ERR(inode)) {
507                         err = PTR_ERR(inode);
508                         inode = NULL;
509                         goto failed_inode;
510                 }
511
512                 pos = rb->blkoff << inode->i_blkbits;
513                 err = block_write_begin(inode->i_mapping, pos, blocksize,
514                                         &page, nilfs_get_block);
515                 if (unlikely(err)) {
516                         loff_t isize = inode->i_size;
517
518                         if (pos + blocksize > isize)
519                                 nilfs_write_failed(inode->i_mapping,
520                                                         pos + blocksize);
521                         goto failed_inode;
522                 }
523
524                 err = nilfs_recovery_copy_block(nilfs, rb, page);
525                 if (unlikely(err))
526                         goto failed_page;
527
528                 err = nilfs_set_file_dirty(inode, 1);
529                 if (unlikely(err))
530                         goto failed_page;
531
532                 block_write_end(NULL, inode->i_mapping, pos, blocksize,
533                                 blocksize, page, NULL);
534
535                 unlock_page(page);
536                 put_page(page);
537
538                 (*nr_salvaged_blocks)++;
539                 goto next;
540
541  failed_page:
542                 unlock_page(page);
543                 put_page(page);
544
545  failed_inode:
546                 nilfs_warn(sb,
547                            "error %d recovering data block (ino=%lu, block-offset=%llu)",
548                            err, (unsigned long)rb->ino,
549                            (unsigned long long)rb->blkoff);
550                 if (!err2)
551                         err2 = err;
552  next:
553                 iput(inode); /* iput(NULL) is just ignored */
554                 list_del_init(&rb->list);
555                 kfree(rb);
556         }
557         return err2;
558 }
559
560 /**
561  * nilfs_do_roll_forward - salvage logical segments newer than the latest
562  * checkpoint
563  * @nilfs: nilfs object
564  * @sb: super block instance
565  * @ri: pointer to a nilfs_recovery_info
566  */
567 static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
568                                  struct super_block *sb,
569                                  struct nilfs_root *root,
570                                  struct nilfs_recovery_info *ri)
571 {
572         struct buffer_head *bh_sum = NULL;
573         struct nilfs_segment_summary *sum = NULL;
574         sector_t pseg_start;
575         sector_t seg_start, seg_end;  /* Starting/ending DBN of full segment */
576         unsigned long nsalvaged_blocks = 0;
577         unsigned int flags;
578         u64 seg_seq;
579         __u64 segnum, nextnum = 0;
580         int empty_seg = 0;
581         int err = 0, ret;
582         LIST_HEAD(dsync_blocks);  /* list of data blocks to be recovered */
583         enum {
584                 RF_INIT_ST,
585                 RF_DSYNC_ST,   /* scanning data-sync segments */
586         };
587         int state = RF_INIT_ST;
588
589         pseg_start = ri->ri_lsegs_start;
590         seg_seq = ri->ri_lsegs_start_seq;
591         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
592         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
593
594         while (segnum != ri->ri_segnum || pseg_start <= ri->ri_pseg_start) {
595                 brelse(bh_sum);
596                 bh_sum = nilfs_read_log_header(nilfs, pseg_start, &sum);
597                 if (!bh_sum) {
598                         err = -EIO;
599                         goto failed;
600                 }
601
602                 ret = nilfs_validate_log(nilfs, seg_seq, bh_sum, sum);
603                 if (ret) {
604                         if (ret == NILFS_SEG_FAIL_IO) {
605                                 err = -EIO;
606                                 goto failed;
607                         }
608                         goto strayed;
609                 }
610
611                 flags = le16_to_cpu(sum->ss_flags);
612                 if (flags & NILFS_SS_SR)
613                         goto confused;
614
615                 /* Found a valid partial segment; do recovery actions */
616                 nextnum = nilfs_get_segnum_of_block(nilfs,
617                                                     le64_to_cpu(sum->ss_next));
618                 empty_seg = 0;
619                 nilfs->ns_ctime = le64_to_cpu(sum->ss_create);
620                 if (!(flags & NILFS_SS_GC))
621                         nilfs->ns_nongc_ctime = nilfs->ns_ctime;
622
623                 switch (state) {
624                 case RF_INIT_ST:
625                         if (!(flags & NILFS_SS_LOGBGN) ||
626                             !(flags & NILFS_SS_SYNDT))
627                                 goto try_next_pseg;
628                         state = RF_DSYNC_ST;
629                         fallthrough;
630                 case RF_DSYNC_ST:
631                         if (!(flags & NILFS_SS_SYNDT))
632                                 goto confused;
633
634                         err = nilfs_scan_dsync_log(nilfs, pseg_start, sum,
635                                                    &dsync_blocks);
636                         if (unlikely(err))
637                                 goto failed;
638                         if (flags & NILFS_SS_LOGEND) {
639                                 err = nilfs_recover_dsync_blocks(
640                                         nilfs, sb, root, &dsync_blocks,
641                                         &nsalvaged_blocks);
642                                 if (unlikely(err))
643                                         goto failed;
644                                 state = RF_INIT_ST;
645                         }
646                         break; /* Fall through to try_next_pseg */
647                 }
648
649  try_next_pseg:
650                 if (pseg_start == ri->ri_lsegs_end)
651                         break;
652                 pseg_start += le32_to_cpu(sum->ss_nblocks);
653                 if (pseg_start < seg_end)
654                         continue;
655                 goto feed_segment;
656
657  strayed:
658                 if (pseg_start == ri->ri_lsegs_end)
659                         break;
660
661  feed_segment:
662                 /* Looking to the next full segment */
663                 if (empty_seg++)
664                         break;
665                 seg_seq++;
666                 segnum = nextnum;
667                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
668                 pseg_start = seg_start;
669         }
670
671         if (nsalvaged_blocks) {
672                 nilfs_info(sb, "salvaged %lu blocks", nsalvaged_blocks);
673                 ri->ri_need_recovery = NILFS_RECOVERY_ROLLFORWARD_DONE;
674         }
675  out:
676         brelse(bh_sum);
677         dispose_recovery_list(&dsync_blocks);
678         return err;
679
680  confused:
681         err = -EINVAL;
682  failed:
683         nilfs_err(sb,
684                   "error %d roll-forwarding partial segment at blocknr = %llu",
685                   err, (unsigned long long)pseg_start);
686         goto out;
687 }
688
689 static void nilfs_finish_roll_forward(struct the_nilfs *nilfs,
690                                       struct nilfs_recovery_info *ri)
691 {
692         struct buffer_head *bh;
693         int err;
694
695         if (nilfs_get_segnum_of_block(nilfs, ri->ri_lsegs_start) !=
696             nilfs_get_segnum_of_block(nilfs, ri->ri_super_root))
697                 return;
698
699         bh = __getblk(nilfs->ns_bdev, ri->ri_lsegs_start, nilfs->ns_blocksize);
700         BUG_ON(!bh);
701         memset(bh->b_data, 0, bh->b_size);
702         set_buffer_dirty(bh);
703         err = sync_dirty_buffer(bh);
704         if (unlikely(err))
705                 nilfs_warn(nilfs->ns_sb,
706                            "buffer sync write failed during post-cleaning of recovery.");
707         brelse(bh);
708 }
709
710 /**
711  * nilfs_salvage_orphan_logs - salvage logs written after the latest checkpoint
712  * @nilfs: nilfs object
713  * @sb: super block instance
714  * @ri: pointer to a nilfs_recovery_info struct to store search results.
715  *
716  * Return Value: On success, 0 is returned.  On error, one of the following
717  * negative error code is returned.
718  *
719  * %-EINVAL - Inconsistent filesystem state.
720  *
721  * %-EIO - I/O error
722  *
723  * %-ENOSPC - No space left on device (only in a panic state).
724  *
725  * %-ERESTARTSYS - Interrupted.
726  *
727  * %-ENOMEM - Insufficient memory available.
728  */
729 int nilfs_salvage_orphan_logs(struct the_nilfs *nilfs,
730                               struct super_block *sb,
731                               struct nilfs_recovery_info *ri)
732 {
733         struct nilfs_root *root;
734         int err;
735
736         if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0)
737                 return 0;
738
739         err = nilfs_attach_checkpoint(sb, ri->ri_cno, true, &root);
740         if (unlikely(err)) {
741                 nilfs_err(sb, "error %d loading the latest checkpoint", err);
742                 return err;
743         }
744
745         err = nilfs_do_roll_forward(nilfs, sb, root, ri);
746         if (unlikely(err))
747                 goto failed;
748
749         if (ri->ri_need_recovery == NILFS_RECOVERY_ROLLFORWARD_DONE) {
750                 err = nilfs_prepare_segment_for_recovery(nilfs, sb, ri);
751                 if (unlikely(err)) {
752                         nilfs_err(sb, "error %d preparing segment for recovery",
753                                   err);
754                         goto failed;
755                 }
756
757                 err = nilfs_attach_log_writer(sb, root);
758                 if (unlikely(err))
759                         goto failed;
760
761                 set_nilfs_discontinued(nilfs);
762                 err = nilfs_construct_segment(sb);
763                 nilfs_detach_log_writer(sb);
764
765                 if (unlikely(err)) {
766                         nilfs_err(sb, "error %d writing segment for recovery",
767                                   err);
768                         goto failed;
769                 }
770
771                 nilfs_finish_roll_forward(nilfs, ri);
772         }
773
774  failed:
775         nilfs_put_root(root);
776         return err;
777 }
778
779 /**
780  * nilfs_search_super_root - search the latest valid super root
781  * @nilfs: the_nilfs
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  * %-ENOMEM - Insufficient memory available.
796  */
797 int nilfs_search_super_root(struct the_nilfs *nilfs,
798                             struct nilfs_recovery_info *ri)
799 {
800         struct buffer_head *bh_sum = NULL;
801         struct nilfs_segment_summary *sum = NULL;
802         sector_t pseg_start, pseg_end, sr_pseg_start = 0;
803         sector_t seg_start, seg_end; /* range of full segment (block number) */
804         sector_t b, end;
805         unsigned long nblocks;
806         unsigned int flags;
807         u64 seg_seq;
808         __u64 segnum, nextnum = 0;
809         __u64 cno;
810         LIST_HEAD(segments);
811         int empty_seg = 0, scan_newer = 0;
812         int ret;
813
814         pseg_start = nilfs->ns_last_pseg;
815         seg_seq = nilfs->ns_last_seq;
816         cno = nilfs->ns_last_cno;
817         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
818
819         /* Calculate range of segment */
820         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
821
822         /* Read ahead segment */
823         b = seg_start;
824         while (b <= seg_end)
825                 __breadahead(nilfs->ns_bdev, b++, nilfs->ns_blocksize);
826
827         for (;;) {
828                 brelse(bh_sum);
829                 ret = NILFS_SEG_FAIL_IO;
830                 bh_sum = nilfs_read_log_header(nilfs, pseg_start, &sum);
831                 if (!bh_sum)
832                         goto failed;
833
834                 ret = nilfs_validate_log(nilfs, seg_seq, bh_sum, sum);
835                 if (ret) {
836                         if (ret == NILFS_SEG_FAIL_IO)
837                                 goto failed;
838                         goto strayed;
839                 }
840
841                 nblocks = le32_to_cpu(sum->ss_nblocks);
842                 pseg_end = pseg_start + nblocks - 1;
843                 if (unlikely(pseg_end > seg_end)) {
844                         ret = NILFS_SEG_FAIL_CONSISTENCY;
845                         goto strayed;
846                 }
847
848                 /* A valid partial segment */
849                 ri->ri_pseg_start = pseg_start;
850                 ri->ri_seq = seg_seq;
851                 ri->ri_segnum = segnum;
852                 nextnum = nilfs_get_segnum_of_block(nilfs,
853                                                     le64_to_cpu(sum->ss_next));
854                 ri->ri_nextnum = nextnum;
855                 empty_seg = 0;
856
857                 flags = le16_to_cpu(sum->ss_flags);
858                 if (!(flags & NILFS_SS_SR) && !scan_newer) {
859                         /*
860                          * This will never happen because a superblock
861                          * (last_segment) always points to a pseg with
862                          * a super root.
863                          */
864                         ret = NILFS_SEG_FAIL_CONSISTENCY;
865                         goto failed;
866                 }
867
868                 if (pseg_start == seg_start) {
869                         nilfs_get_segment_range(nilfs, nextnum, &b, &end);
870                         while (b <= end)
871                                 __breadahead(nilfs->ns_bdev, b++,
872                                              nilfs->ns_blocksize);
873                 }
874                 if (!(flags & NILFS_SS_SR)) {
875                         if (!ri->ri_lsegs_start && (flags & NILFS_SS_LOGBGN)) {
876                                 ri->ri_lsegs_start = pseg_start;
877                                 ri->ri_lsegs_start_seq = seg_seq;
878                         }
879                         if (flags & NILFS_SS_LOGEND)
880                                 ri->ri_lsegs_end = pseg_start;
881                         goto try_next_pseg;
882                 }
883
884                 /* A valid super root was found. */
885                 ri->ri_cno = cno++;
886                 ri->ri_super_root = pseg_end;
887                 ri->ri_lsegs_start = ri->ri_lsegs_end = 0;
888
889                 nilfs_dispose_segment_list(&segments);
890                 sr_pseg_start = pseg_start;
891                 nilfs->ns_pseg_offset = pseg_start + nblocks - seg_start;
892                 nilfs->ns_seg_seq = seg_seq;
893                 nilfs->ns_segnum = segnum;
894                 nilfs->ns_cno = cno;  /* nilfs->ns_cno = ri->ri_cno + 1 */
895                 nilfs->ns_ctime = le64_to_cpu(sum->ss_create);
896                 nilfs->ns_nextnum = nextnum;
897
898                 if (scan_newer)
899                         ri->ri_need_recovery = NILFS_RECOVERY_SR_UPDATED;
900                 else {
901                         if (nilfs->ns_mount_state & NILFS_VALID_FS)
902                                 goto super_root_found;
903                         scan_newer = 1;
904                 }
905
906  try_next_pseg:
907                 /* Standing on a course, or met an inconsistent state */
908                 pseg_start += nblocks;
909                 if (pseg_start < seg_end)
910                         continue;
911                 goto feed_segment;
912
913  strayed:
914                 /* Off the trail */
915                 if (!scan_newer)
916                         /*
917                          * This can happen if a checkpoint was written without
918                          * barriers, or as a result of an I/O failure.
919                          */
920                         goto failed;
921
922  feed_segment:
923                 /* Looking to the next full segment */
924                 if (empty_seg++)
925                         goto super_root_found; /* found a valid super root */
926
927                 ret = nilfs_segment_list_add(&segments, segnum);
928                 if (unlikely(ret))
929                         goto failed;
930
931                 seg_seq++;
932                 segnum = nextnum;
933                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
934                 pseg_start = seg_start;
935         }
936
937  super_root_found:
938         /* Updating pointers relating to the latest checkpoint */
939         brelse(bh_sum);
940         list_splice_tail(&segments, &ri->ri_used_segments);
941         nilfs->ns_last_pseg = sr_pseg_start;
942         nilfs->ns_last_seq = nilfs->ns_seg_seq;
943         nilfs->ns_last_cno = ri->ri_cno;
944         return 0;
945
946  failed:
947         brelse(bh_sum);
948         nilfs_dispose_segment_list(&segments);
949         return ret < 0 ? ret : nilfs_warn_segment_error(nilfs->ns_sb, ret);
950 }