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 loff_t pos, struct page *page)
476 {
477 struct buffer_head *bh_org;
478 size_t from = pos & ~PAGE_MASK;
479 void *kaddr;
480
481 bh_org = __bread(nilfs->ns_bdev, rb->blocknr, nilfs->ns_blocksize);
482 if (unlikely(!bh_org))
483 return -EIO;
484
485 kaddr = kmap_local_page(page);
486 memcpy(kaddr + from, bh_org->b_data, bh_org->b_size);
487 kunmap_local(kaddr);
488 brelse(bh_org);
489 return 0;
490 }
491
492 static int nilfs_recover_dsync_blocks(struct the_nilfs *nilfs,
493 struct super_block *sb,
494 struct nilfs_root *root,
495 struct list_head *head,
496 unsigned long *nr_salvaged_blocks)
497 {
498 struct inode *inode;
499 struct nilfs_recovery_block *rb, *n;
500 unsigned int blocksize = nilfs->ns_blocksize;
501 struct page *page;
502 loff_t pos;
503 int err = 0, err2 = 0;
504
505 list_for_each_entry_safe(rb, n, head, list) {
506 inode = nilfs_iget(sb, root, rb->ino);
507 if (IS_ERR(inode)) {
508 err = PTR_ERR(inode);
509 inode = NULL;
510 goto failed_inode;
511 }
512
513 pos = rb->blkoff << inode->i_blkbits;
514 err = block_write_begin(inode->i_mapping, pos, blocksize,
515 &page, nilfs_get_block);
516 if (unlikely(err)) {
517 loff_t isize = inode->i_size;
518
519 if (pos + blocksize > isize)
520 nilfs_write_failed(inode->i_mapping,
521 pos + blocksize);
522 goto failed_inode;
523 }
524
525 err = nilfs_recovery_copy_block(nilfs, rb, pos, page);
526 if (unlikely(err))
527 goto failed_page;
528
529 err = nilfs_set_file_dirty(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 put_page(page);
538
539 (*nr_salvaged_blocks)++;
540 goto next;
541
542 failed_page:
543 unlock_page(page);
544 put_page(page);
545
546 failed_inode:
547 nilfs_warn(sb,
548 "error %d recovering data block (ino=%lu, block-offset=%llu)",
549 err, (unsigned long)rb->ino,
550 (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 * @nilfs: nilfs object
565 * @sb: super block instance
566 * @root: NILFS root instance
567 * @ri: pointer to a nilfs_recovery_info
568 */
569 static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
570 struct super_block *sb,
571 struct nilfs_root *root,
572 struct nilfs_recovery_info *ri)
573 {
574 struct buffer_head *bh_sum = NULL;
575 struct nilfs_segment_summary *sum = NULL;
576 sector_t pseg_start;
577 sector_t seg_start, seg_end; /* Starting/ending DBN of full segment */
578 unsigned long nsalvaged_blocks = 0;
579 unsigned int flags;
580 u64 seg_seq;
581 __u64 segnum, nextnum = 0;
582 int empty_seg = 0;
583 int err = 0, ret;
584 LIST_HEAD(dsync_blocks); /* list of data blocks to be recovered */
585 enum {
586 RF_INIT_ST,
587 RF_DSYNC_ST, /* scanning data-sync segments */
588 };
589 int state = RF_INIT_ST;
590
591 pseg_start = ri->ri_lsegs_start;
592 seg_seq = ri->ri_lsegs_start_seq;
593 segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
594 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
595
596 while (segnum != ri->ri_segnum || pseg_start <= ri->ri_pseg_start) {
597 brelse(bh_sum);
598 bh_sum = nilfs_read_log_header(nilfs, pseg_start, &sum);
599 if (!bh_sum) {
600 err = -EIO;
601 goto failed;
602 }
603
604 ret = nilfs_validate_log(nilfs, seg_seq, bh_sum, sum);
605 if (ret) {
606 if (ret == NILFS_SEG_FAIL_IO) {
607 err = -EIO;
608 goto failed;
609 }
610 goto strayed;
611 }
612
613 flags = le16_to_cpu(sum->ss_flags);
614 if (flags & NILFS_SS_SR)
615 goto confused;
616
617 /* Found a valid partial segment; do recovery actions */
618 nextnum = nilfs_get_segnum_of_block(nilfs,
619 le64_to_cpu(sum->ss_next));
620 empty_seg = 0;
621 nilfs->ns_ctime = le64_to_cpu(sum->ss_create);
622 if (!(flags & NILFS_SS_GC))
623 nilfs->ns_nongc_ctime = nilfs->ns_ctime;
624
625 switch (state) {
626 case RF_INIT_ST:
627 if (!(flags & NILFS_SS_LOGBGN) ||
628 !(flags & NILFS_SS_SYNDT))
629 goto try_next_pseg;
630 state = RF_DSYNC_ST;
631 fallthrough;
632 case RF_DSYNC_ST:
633 if (!(flags & NILFS_SS_SYNDT))
634 goto confused;
635
636 err = nilfs_scan_dsync_log(nilfs, pseg_start, sum,
637 &dsync_blocks);
638 if (unlikely(err))
639 goto failed;
640 if (flags & NILFS_SS_LOGEND) {
641 err = nilfs_recover_dsync_blocks(
642 nilfs, sb, root, &dsync_blocks,
643 &nsalvaged_blocks);
644 if (unlikely(err))
645 goto failed;
646 state = RF_INIT_ST;
647 }
648 break; /* Fall through to try_next_pseg */
649 }
650
651 try_next_pseg:
652 if (pseg_start == ri->ri_lsegs_end)
653 break;
654 pseg_start += le32_to_cpu(sum->ss_nblocks);
655 if (pseg_start < seg_end)
656 continue;
657 goto feed_segment;
658
659 strayed:
660 if (pseg_start == ri->ri_lsegs_end)
661 break;
662
663 feed_segment:
664 /* Looking to the next full segment */
665 if (empty_seg++)
666 break;
667 seg_seq++;
668 segnum = nextnum;
669 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
670 pseg_start = seg_start;
671 }
672
673 if (nsalvaged_blocks) {
674 nilfs_info(sb, "salvaged %lu blocks", nsalvaged_blocks);
675 ri->ri_need_recovery = NILFS_RECOVERY_ROLLFORWARD_DONE;
676 }
677 out:
678 brelse(bh_sum);
679 dispose_recovery_list(&dsync_blocks);
680 return err;
681
682 confused:
683 err = -EINVAL;
684 failed:
685 nilfs_err(sb,
686 "error %d roll-forwarding partial segment at blocknr = %llu",
687 err, (unsigned long long)pseg_start);
688 goto out;
689 }
690
691 static void nilfs_finish_roll_forward(struct the_nilfs *nilfs,
692 struct nilfs_recovery_info *ri)
693 {
694 struct buffer_head *bh;
695 int err;
696
697 if (nilfs_get_segnum_of_block(nilfs, ri->ri_lsegs_start) !=
698 nilfs_get_segnum_of_block(nilfs, ri->ri_super_root))
699 return;
700
701 bh = __getblk(nilfs->ns_bdev, ri->ri_lsegs_start, nilfs->ns_blocksize);
702 if (WARN_ON(!bh))
703 return; /* should never happen */
704
705 lock_buffer(bh);
706 memset(bh->b_data, 0, bh->b_size);
707 set_buffer_uptodate(bh);
708 set_buffer_dirty(bh);
709 unlock_buffer(bh);
710
711 err = sync_dirty_buffer(bh);
712 if (unlikely(err))
713 nilfs_warn(nilfs->ns_sb,
714 "buffer sync write failed during post-cleaning of recovery.");
715 brelse(bh);
716 }
717
718 /**
719 * nilfs_abort_roll_forward - cleaning up after a failed rollforward recovery
720 * @nilfs: nilfs object
721 */
722 static void nilfs_abort_roll_forward(struct the_nilfs *nilfs)
723 {
724 struct nilfs_inode_info *ii, *n;
725 LIST_HEAD(head);
726
727 /* Abandon inodes that have read recovery data */
728 spin_lock(&nilfs->ns_inode_lock);
729 list_splice_init(&nilfs->ns_dirty_files, &head);
730 spin_unlock(&nilfs->ns_inode_lock);
731 if (list_empty(&head))
732 return;
733
734 set_nilfs_purging(nilfs);
735 list_for_each_entry_safe(ii, n, &head, i_dirty) {
736 spin_lock(&nilfs->ns_inode_lock);
737 list_del_init(&ii->i_dirty);
738 spin_unlock(&nilfs->ns_inode_lock);
739
740 iput(&ii->vfs_inode);
741 }
742 clear_nilfs_purging(nilfs);
743 }
744
745 /**
746 * nilfs_salvage_orphan_logs - salvage logs written after the latest checkpoint
747 * @nilfs: nilfs object
748 * @sb: super block instance
749 * @ri: pointer to a nilfs_recovery_info struct to store search results.
750 *
751 * Return Value: On success, 0 is returned. On error, one of the following
752 * negative error code is returned.
753 *
754 * %-EINVAL - Inconsistent filesystem state.
755 *
756 * %-EIO - I/O error
757 *
758 * %-ENOSPC - No space left on device (only in a panic state).
759 *
760 * %-ERESTARTSYS - Interrupted.
761 *
762 * %-ENOMEM - Insufficient memory available.
763 */
764 int nilfs_salvage_orphan_logs(struct the_nilfs *nilfs,
765 struct super_block *sb,
766 struct nilfs_recovery_info *ri)
767 {
768 struct nilfs_root *root;
769 int err;
770
771 if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0)
772 return 0;
773
774 err = nilfs_attach_checkpoint(sb, ri->ri_cno, true, &root);
775 if (unlikely(err)) {
776 nilfs_err(sb, "error %d loading the latest checkpoint", err);
777 return err;
778 }
779
780 err = nilfs_do_roll_forward(nilfs, sb, root, ri);
781 if (unlikely(err))
782 goto failed;
783
784 if (ri->ri_need_recovery == NILFS_RECOVERY_ROLLFORWARD_DONE) {
785 err = nilfs_prepare_segment_for_recovery(nilfs, sb, ri);
786 if (unlikely(err)) {
787 nilfs_err(sb, "error %d preparing segment for recovery",
788 err);
789 goto failed;
790 }
791
792 err = nilfs_attach_log_writer(sb, root);
793 if (unlikely(err))
794 goto failed;
795
796 set_nilfs_discontinued(nilfs);
797 err = nilfs_construct_segment(sb);
798 nilfs_detach_log_writer(sb);
799
800 if (unlikely(err)) {
801 nilfs_err(sb, "error %d writing segment for recovery",
802 err);
803 goto put_root;
804 }
805
806 nilfs_finish_roll_forward(nilfs, ri);
807 }
808
809 put_root:
810 nilfs_put_root(root);
811 return err;
812
813 failed:
814 nilfs_abort_roll_forward(nilfs);
815 goto put_root;
816 }
817
818 /**
819 * nilfs_search_super_root - search the latest valid super root
820 * @nilfs: the_nilfs
821 * @ri: pointer to a nilfs_recovery_info struct to store search results.
822 *
823 * nilfs_search_super_root() looks for the latest super-root from a partial
824 * segment pointed by the superblock. It sets up struct the_nilfs through
825 * this search. It fills nilfs_recovery_info (ri) required for recovery.
826 *
827 * Return Value: On success, 0 is returned. On error, one of the following
828 * negative error code is returned.
829 *
830 * %-EINVAL - No valid segment found
831 *
832 * %-EIO - I/O error
833 *
834 * %-ENOMEM - Insufficient memory available.
835 */
836 int nilfs_search_super_root(struct the_nilfs *nilfs,
837 struct nilfs_recovery_info *ri)
838 {
839 struct buffer_head *bh_sum = NULL;
840 struct nilfs_segment_summary *sum = NULL;
841 sector_t pseg_start, pseg_end, sr_pseg_start = 0;
842 sector_t seg_start, seg_end; /* range of full segment (block number) */
843 sector_t b, end;
844 unsigned long nblocks;
845 unsigned int flags;
846 u64 seg_seq;
847 __u64 segnum, nextnum = 0;
848 __u64 cno;
849 LIST_HEAD(segments);
850 int empty_seg = 0, scan_newer = 0;
851 int ret;
852
853 pseg_start = nilfs->ns_last_pseg;
854 seg_seq = nilfs->ns_last_seq;
855 cno = nilfs->ns_last_cno;
856 segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
857
858 /* Calculate range of segment */
859 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
860
861 /* Read ahead segment */
862 b = seg_start;
863 while (b <= seg_end)
864 __breadahead(nilfs->ns_bdev, b++, nilfs->ns_blocksize);
865
866 for (;;) {
867 brelse(bh_sum);
868 ret = NILFS_SEG_FAIL_IO;
869 bh_sum = nilfs_read_log_header(nilfs, pseg_start, &sum);
870 if (!bh_sum)
871 goto failed;
872
873 ret = nilfs_validate_log(nilfs, seg_seq, bh_sum, sum);
874 if (ret) {
875 if (ret == NILFS_SEG_FAIL_IO)
876 goto failed;
877 goto strayed;
878 }
879
880 nblocks = le32_to_cpu(sum->ss_nblocks);
881 pseg_end = pseg_start + nblocks - 1;
882 if (unlikely(pseg_end > seg_end)) {
883 ret = NILFS_SEG_FAIL_CONSISTENCY;
884 goto strayed;
885 }
886
887 /* A valid partial segment */
888 ri->ri_pseg_start = pseg_start;
889 ri->ri_seq = seg_seq;
890 ri->ri_segnum = segnum;
891 nextnum = nilfs_get_segnum_of_block(nilfs,
892 le64_to_cpu(sum->ss_next));
893 ri->ri_nextnum = nextnum;
894 empty_seg = 0;
895
896 flags = le16_to_cpu(sum->ss_flags);
897 if (!(flags & NILFS_SS_SR) && !scan_newer) {
898 /*
899 * This will never happen because a superblock
900 * (last_segment) always points to a pseg with
901 * a super root.
902 */
903 ret = NILFS_SEG_FAIL_CONSISTENCY;
904 goto failed;
905 }
906
907 if (pseg_start == seg_start) {
908 nilfs_get_segment_range(nilfs, nextnum, &b, &end);
909 while (b <= end)
910 __breadahead(nilfs->ns_bdev, b++,
911 nilfs->ns_blocksize);
912 }
913 if (!(flags & NILFS_SS_SR)) {
914 if (!ri->ri_lsegs_start && (flags & NILFS_SS_LOGBGN)) {
915 ri->ri_lsegs_start = pseg_start;
916 ri->ri_lsegs_start_seq = seg_seq;
917 }
918 if (flags & NILFS_SS_LOGEND)
919 ri->ri_lsegs_end = pseg_start;
920 goto try_next_pseg;
921 }
922
923 /* A valid super root was found. */
924 ri->ri_cno = cno++;
925 ri->ri_super_root = pseg_end;
926 ri->ri_lsegs_start = ri->ri_lsegs_end = 0;
927
928 nilfs_dispose_segment_list(&segments);
929 sr_pseg_start = pseg_start;
930 nilfs->ns_pseg_offset = pseg_start + nblocks - seg_start;
931 nilfs->ns_seg_seq = seg_seq;
932 nilfs->ns_segnum = segnum;
933 nilfs->ns_cno = cno; /* nilfs->ns_cno = ri->ri_cno + 1 */
934 nilfs->ns_ctime = le64_to_cpu(sum->ss_create);
935 nilfs->ns_nextnum = nextnum;
936
937 if (scan_newer)
938 ri->ri_need_recovery = NILFS_RECOVERY_SR_UPDATED;
939 else {
940 if (nilfs->ns_mount_state & NILFS_VALID_FS)
941 goto super_root_found;
942 scan_newer = 1;
943 }
944
945 try_next_pseg:
946 /* Standing on a course, or met an inconsistent state */
947 pseg_start += nblocks;
948 if (pseg_start < seg_end)
949 continue;
950 goto feed_segment;
951
952 strayed:
953 /* Off the trail */
954 if (!scan_newer)
955 /*
956 * This can happen if a checkpoint was written without
957 * barriers, or as a result of an I/O failure.
958 */
959 goto failed;
960
961 feed_segment:
962 /* Looking to the next full segment */
963 if (empty_seg++)
964 goto super_root_found; /* found a valid super root */
965
966 ret = nilfs_segment_list_add(&segments, segnum);
967 if (unlikely(ret))
968 goto failed;
969
970 seg_seq++;
971 segnum = nextnum;
972 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
973 pseg_start = seg_start;
974 }
975
976 super_root_found:
977 /* Updating pointers relating to the latest checkpoint */
978 brelse(bh_sum);
979 list_splice_tail(&segments, &ri->ri_used_segments);
980 nilfs->ns_last_pseg = sr_pseg_start;
981 nilfs->ns_last_seq = nilfs->ns_seg_seq;
982 nilfs->ns_last_cno = ri->ri_cno;
983 return 0;
984
985 failed:
986 brelse(bh_sum);
987 nilfs_dispose_segment_list(&segments);
988 return ret < 0 ? ret : nilfs_warn_segment_error(nilfs->ns_sb, ret);
989 }
990
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.