xref: /linux/fs/nilfs2/recovery.c (revision e58e871becec2d3b04ed91c0c16fe8deac9c9dfa)
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  * Written by Ryusuke Konishi.
17  */
18 
19 #include <linux/buffer_head.h>
20 #include <linux/blkdev.h>
21 #include <linux/swap.h>
22 #include <linux/slab.h>
23 #include <linux/crc32.h>
24 #include "nilfs.h"
25 #include "segment.h"
26 #include "sufile.h"
27 #include "page.h"
28 #include "segbuf.h"
29 
30 /*
31  * Segment check result
32  */
33 enum {
34 	NILFS_SEG_VALID,
35 	NILFS_SEG_NO_SUPER_ROOT,
36 	NILFS_SEG_FAIL_IO,
37 	NILFS_SEG_FAIL_MAGIC,
38 	NILFS_SEG_FAIL_SEQ,
39 	NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT,
40 	NILFS_SEG_FAIL_CHECKSUM_FULL,
41 	NILFS_SEG_FAIL_CONSISTENCY,
42 };
43 
44 /* work structure for recovery */
45 struct nilfs_recovery_block {
46 	ino_t ino;		/*
47 				 * Inode number of the file that this block
48 				 * belongs to
49 				 */
50 	sector_t blocknr;	/* block number */
51 	__u64 vblocknr;		/* virtual block number */
52 	unsigned long blkoff;	/* File offset of the data block (per block) */
53 	struct list_head list;
54 };
55 
56 
57 static int nilfs_warn_segment_error(struct super_block *sb, int err)
58 {
59 	const char *msg = NULL;
60 
61 	switch (err) {
62 	case NILFS_SEG_FAIL_IO:
63 		nilfs_msg(sb, KERN_ERR, "I/O error reading segment");
64 		return -EIO;
65 	case NILFS_SEG_FAIL_MAGIC:
66 		msg = "Magic number mismatch";
67 		break;
68 	case NILFS_SEG_FAIL_SEQ:
69 		msg = "Sequence number mismatch";
70 		break;
71 	case NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT:
72 		msg = "Checksum error in super root";
73 		break;
74 	case NILFS_SEG_FAIL_CHECKSUM_FULL:
75 		msg = "Checksum error in segment payload";
76 		break;
77 	case NILFS_SEG_FAIL_CONSISTENCY:
78 		msg = "Inconsistency found";
79 		break;
80 	case NILFS_SEG_NO_SUPER_ROOT:
81 		msg = "No super root in the last segment";
82 		break;
83 	default:
84 		nilfs_msg(sb, KERN_ERR, "unrecognized segment error %d", err);
85 		return -EINVAL;
86 	}
87 	nilfs_msg(sb, KERN_WARNING, "invalid segment: %s", msg);
88 	return -EINVAL;
89 }
90 
91 /**
92  * nilfs_compute_checksum - compute checksum of blocks continuously
93  * @nilfs: nilfs object
94  * @bhs: buffer head of start block
95  * @sum: place to store result
96  * @offset: offset bytes in the first block
97  * @check_bytes: number of bytes to be checked
98  * @start: DBN of start block
99  * @nblock: number of blocks to be checked
100  */
101 static int nilfs_compute_checksum(struct the_nilfs *nilfs,
102 				  struct buffer_head *bhs, u32 *sum,
103 				  unsigned long offset, u64 check_bytes,
104 				  sector_t start, unsigned long nblock)
105 {
106 	unsigned int blocksize = nilfs->ns_blocksize;
107 	unsigned long size;
108 	u32 crc;
109 
110 	BUG_ON(offset >= blocksize);
111 	check_bytes -= offset;
112 	size = min_t(u64, check_bytes, blocksize - offset);
113 	crc = crc32_le(nilfs->ns_crc_seed,
114 		       (unsigned char *)bhs->b_data + offset, size);
115 	if (--nblock > 0) {
116 		do {
117 			struct buffer_head *bh;
118 
119 			bh = __bread(nilfs->ns_bdev, ++start, blocksize);
120 			if (!bh)
121 				return -EIO;
122 			check_bytes -= size;
123 			size = min_t(u64, check_bytes, blocksize);
124 			crc = crc32_le(crc, bh->b_data, size);
125 			brelse(bh);
126 		} while (--nblock > 0);
127 	}
128 	*sum = crc;
129 	return 0;
130 }
131 
132 /**
133  * nilfs_read_super_root_block - read super root block
134  * @nilfs: nilfs object
135  * @sr_block: disk block number of the super root block
136  * @pbh: address of a buffer_head pointer to return super root buffer
137  * @check: CRC check flag
138  */
139 int nilfs_read_super_root_block(struct the_nilfs *nilfs, sector_t sr_block,
140 				struct buffer_head **pbh, int check)
141 {
142 	struct buffer_head *bh_sr;
143 	struct nilfs_super_root *sr;
144 	u32 crc;
145 	int ret;
146 
147 	*pbh = NULL;
148 	bh_sr = __bread(nilfs->ns_bdev, sr_block, nilfs->ns_blocksize);
149 	if (unlikely(!bh_sr)) {
150 		ret = NILFS_SEG_FAIL_IO;
151 		goto failed;
152 	}
153 
154 	sr = (struct nilfs_super_root *)bh_sr->b_data;
155 	if (check) {
156 		unsigned int bytes = le16_to_cpu(sr->sr_bytes);
157 
158 		if (bytes == 0 || bytes > nilfs->ns_blocksize) {
159 			ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
160 			goto failed_bh;
161 		}
162 		if (nilfs_compute_checksum(
163 			    nilfs, bh_sr, &crc, sizeof(sr->sr_sum), bytes,
164 			    sr_block, 1)) {
165 			ret = NILFS_SEG_FAIL_IO;
166 			goto failed_bh;
167 		}
168 		if (crc != le32_to_cpu(sr->sr_sum)) {
169 			ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
170 			goto failed_bh;
171 		}
172 	}
173 	*pbh = bh_sr;
174 	return 0;
175 
176  failed_bh:
177 	brelse(bh_sr);
178 
179  failed:
180 	return nilfs_warn_segment_error(nilfs->ns_sb, ret);
181 }
182 
183 /**
184  * nilfs_read_log_header - read summary header of the specified log
185  * @nilfs: nilfs object
186  * @start_blocknr: start block number of the log
187  * @sum: pointer to return segment summary structure
188  */
189 static struct buffer_head *
190 nilfs_read_log_header(struct the_nilfs *nilfs, sector_t start_blocknr,
191 		      struct nilfs_segment_summary **sum)
192 {
193 	struct buffer_head *bh_sum;
194 
195 	bh_sum = __bread(nilfs->ns_bdev, start_blocknr, nilfs->ns_blocksize);
196 	if (bh_sum)
197 		*sum = (struct nilfs_segment_summary *)bh_sum->b_data;
198 	return bh_sum;
199 }
200 
201 /**
202  * nilfs_validate_log - verify consistency of log
203  * @nilfs: nilfs object
204  * @seg_seq: sequence number of segment
205  * @bh_sum: buffer head of summary block
206  * @sum: segment summary struct
207  */
208 static int nilfs_validate_log(struct the_nilfs *nilfs, u64 seg_seq,
209 			      struct buffer_head *bh_sum,
210 			      struct nilfs_segment_summary *sum)
211 {
212 	unsigned long nblock;
213 	u32 crc;
214 	int ret;
215 
216 	ret = NILFS_SEG_FAIL_MAGIC;
217 	if (le32_to_cpu(sum->ss_magic) != NILFS_SEGSUM_MAGIC)
218 		goto out;
219 
220 	ret = NILFS_SEG_FAIL_SEQ;
221 	if (le64_to_cpu(sum->ss_seq) != seg_seq)
222 		goto out;
223 
224 	nblock = le32_to_cpu(sum->ss_nblocks);
225 	ret = NILFS_SEG_FAIL_CONSISTENCY;
226 	if (unlikely(nblock == 0 || nblock > nilfs->ns_blocks_per_segment))
227 		/* This limits the number of blocks read in the CRC check */
228 		goto out;
229 
230 	ret = NILFS_SEG_FAIL_IO;
231 	if (nilfs_compute_checksum(nilfs, bh_sum, &crc, sizeof(sum->ss_datasum),
232 				   ((u64)nblock << nilfs->ns_blocksize_bits),
233 				   bh_sum->b_blocknr, nblock))
234 		goto out;
235 
236 	ret = NILFS_SEG_FAIL_CHECKSUM_FULL;
237 	if (crc != le32_to_cpu(sum->ss_datasum))
238 		goto out;
239 	ret = 0;
240 out:
241 	return ret;
242 }
243 
244 /**
245  * nilfs_read_summary_info - read an item on summary blocks of a log
246  * @nilfs: nilfs object
247  * @pbh: the current buffer head on summary blocks [in, out]
248  * @offset: the current byte offset on summary blocks [in, out]
249  * @bytes: byte size of the item to be read
250  */
251 static void *nilfs_read_summary_info(struct the_nilfs *nilfs,
252 				     struct buffer_head **pbh,
253 				     unsigned int *offset, unsigned int bytes)
254 {
255 	void *ptr;
256 	sector_t blocknr;
257 
258 	BUG_ON((*pbh)->b_size < *offset);
259 	if (bytes > (*pbh)->b_size - *offset) {
260 		blocknr = (*pbh)->b_blocknr;
261 		brelse(*pbh);
262 		*pbh = __bread(nilfs->ns_bdev, blocknr + 1,
263 			       nilfs->ns_blocksize);
264 		if (unlikely(!*pbh))
265 			return NULL;
266 		*offset = 0;
267 	}
268 	ptr = (*pbh)->b_data + *offset;
269 	*offset += bytes;
270 	return ptr;
271 }
272 
273 /**
274  * nilfs_skip_summary_info - skip items on summary blocks of a log
275  * @nilfs: nilfs object
276  * @pbh: the current buffer head on summary blocks [in, out]
277  * @offset: the current byte offset on summary blocks [in, out]
278  * @bytes: byte size of the item to be skipped
279  * @count: number of items to be skipped
280  */
281 static void nilfs_skip_summary_info(struct the_nilfs *nilfs,
282 				    struct buffer_head **pbh,
283 				    unsigned int *offset, unsigned int bytes,
284 				    unsigned long count)
285 {
286 	unsigned int rest_item_in_current_block
287 		= ((*pbh)->b_size - *offset) / bytes;
288 
289 	if (count <= rest_item_in_current_block) {
290 		*offset += bytes * count;
291 	} else {
292 		sector_t blocknr = (*pbh)->b_blocknr;
293 		unsigned int nitem_per_block = (*pbh)->b_size / bytes;
294 		unsigned int bcnt;
295 
296 		count -= rest_item_in_current_block;
297 		bcnt = DIV_ROUND_UP(count, nitem_per_block);
298 		*offset = bytes * (count - (bcnt - 1) * nitem_per_block);
299 
300 		brelse(*pbh);
301 		*pbh = __bread(nilfs->ns_bdev, blocknr + bcnt,
302 			       nilfs->ns_blocksize);
303 	}
304 }
305 
306 /**
307  * nilfs_scan_dsync_log - get block information of a log written for data sync
308  * @nilfs: nilfs object
309  * @start_blocknr: start block number of the log
310  * @sum: log summary information
311  * @head: list head to add nilfs_recovery_block struct
312  */
313 static int nilfs_scan_dsync_log(struct the_nilfs *nilfs, sector_t start_blocknr,
314 				struct nilfs_segment_summary *sum,
315 				struct list_head *head)
316 {
317 	struct buffer_head *bh;
318 	unsigned int offset;
319 	u32 nfinfo, sumbytes;
320 	sector_t blocknr;
321 	ino_t ino;
322 	int err = -EIO;
323 
324 	nfinfo = le32_to_cpu(sum->ss_nfinfo);
325 	if (!nfinfo)
326 		return 0;
327 
328 	sumbytes = le32_to_cpu(sum->ss_sumbytes);
329 	blocknr = start_blocknr + DIV_ROUND_UP(sumbytes, nilfs->ns_blocksize);
330 	bh = __bread(nilfs->ns_bdev, start_blocknr, nilfs->ns_blocksize);
331 	if (unlikely(!bh))
332 		goto out;
333 
334 	offset = le16_to_cpu(sum->ss_bytes);
335 	for (;;) {
336 		unsigned long nblocks, ndatablk, nnodeblk;
337 		struct nilfs_finfo *finfo;
338 
339 		finfo = nilfs_read_summary_info(nilfs, &bh, &offset,
340 						sizeof(*finfo));
341 		if (unlikely(!finfo))
342 			goto out;
343 
344 		ino = le64_to_cpu(finfo->fi_ino);
345 		nblocks = le32_to_cpu(finfo->fi_nblocks);
346 		ndatablk = le32_to_cpu(finfo->fi_ndatablk);
347 		nnodeblk = nblocks - ndatablk;
348 
349 		while (ndatablk-- > 0) {
350 			struct nilfs_recovery_block *rb;
351 			struct nilfs_binfo_v *binfo;
352 
353 			binfo = nilfs_read_summary_info(nilfs, &bh, &offset,
354 							sizeof(*binfo));
355 			if (unlikely(!binfo))
356 				goto out;
357 
358 			rb = kmalloc(sizeof(*rb), GFP_NOFS);
359 			if (unlikely(!rb)) {
360 				err = -ENOMEM;
361 				goto out;
362 			}
363 			rb->ino = ino;
364 			rb->blocknr = blocknr++;
365 			rb->vblocknr = le64_to_cpu(binfo->bi_vblocknr);
366 			rb->blkoff = le64_to_cpu(binfo->bi_blkoff);
367 			/* INIT_LIST_HEAD(&rb->list); */
368 			list_add_tail(&rb->list, head);
369 		}
370 		if (--nfinfo == 0)
371 			break;
372 		blocknr += nnodeblk; /* always 0 for data sync logs */
373 		nilfs_skip_summary_info(nilfs, &bh, &offset, sizeof(__le64),
374 					nnodeblk);
375 		if (unlikely(!bh))
376 			goto out;
377 	}
378 	err = 0;
379  out:
380 	brelse(bh);   /* brelse(NULL) is just ignored */
381 	return err;
382 }
383 
384 static void dispose_recovery_list(struct list_head *head)
385 {
386 	while (!list_empty(head)) {
387 		struct nilfs_recovery_block *rb;
388 
389 		rb = list_first_entry(head, struct nilfs_recovery_block, list);
390 		list_del(&rb->list);
391 		kfree(rb);
392 	}
393 }
394 
395 struct nilfs_segment_entry {
396 	struct list_head	list;
397 	__u64			segnum;
398 };
399 
400 static int nilfs_segment_list_add(struct list_head *head, __u64 segnum)
401 {
402 	struct nilfs_segment_entry *ent = kmalloc(sizeof(*ent), GFP_NOFS);
403 
404 	if (unlikely(!ent))
405 		return -ENOMEM;
406 
407 	ent->segnum = segnum;
408 	INIT_LIST_HEAD(&ent->list);
409 	list_add_tail(&ent->list, head);
410 	return 0;
411 }
412 
413 void nilfs_dispose_segment_list(struct list_head *head)
414 {
415 	while (!list_empty(head)) {
416 		struct nilfs_segment_entry *ent;
417 
418 		ent = list_first_entry(head, struct nilfs_segment_entry, list);
419 		list_del(&ent->list);
420 		kfree(ent);
421 	}
422 }
423 
424 static int nilfs_prepare_segment_for_recovery(struct the_nilfs *nilfs,
425 					      struct super_block *sb,
426 					      struct nilfs_recovery_info *ri)
427 {
428 	struct list_head *head = &ri->ri_used_segments;
429 	struct nilfs_segment_entry *ent, *n;
430 	struct inode *sufile = nilfs->ns_sufile;
431 	__u64 segnum[4];
432 	int err;
433 	int i;
434 
435 	segnum[0] = nilfs->ns_segnum;
436 	segnum[1] = nilfs->ns_nextnum;
437 	segnum[2] = ri->ri_segnum;
438 	segnum[3] = ri->ri_nextnum;
439 
440 	/*
441 	 * Releasing the next segment of the latest super root.
442 	 * The next segment is invalidated by this recovery.
443 	 */
444 	err = nilfs_sufile_free(sufile, segnum[1]);
445 	if (unlikely(err))
446 		goto failed;
447 
448 	for (i = 1; i < 4; i++) {
449 		err = nilfs_segment_list_add(head, segnum[i]);
450 		if (unlikely(err))
451 			goto failed;
452 	}
453 
454 	/*
455 	 * Collecting segments written after the latest super root.
456 	 * These are marked dirty to avoid being reallocated in the next write.
457 	 */
458 	list_for_each_entry_safe(ent, n, head, list) {
459 		if (ent->segnum != segnum[0]) {
460 			err = nilfs_sufile_scrap(sufile, ent->segnum);
461 			if (unlikely(err))
462 				goto failed;
463 		}
464 		list_del(&ent->list);
465 		kfree(ent);
466 	}
467 
468 	/* Allocate new segments for recovery */
469 	err = nilfs_sufile_alloc(sufile, &segnum[0]);
470 	if (unlikely(err))
471 		goto failed;
472 
473 	nilfs->ns_pseg_offset = 0;
474 	nilfs->ns_seg_seq = ri->ri_seq + 2;
475 	nilfs->ns_nextnum = nilfs->ns_segnum = segnum[0];
476 
477  failed:
478 	/* No need to recover sufile because it will be destroyed on error */
479 	return err;
480 }
481 
482 static int nilfs_recovery_copy_block(struct the_nilfs *nilfs,
483 				     struct nilfs_recovery_block *rb,
484 				     struct page *page)
485 {
486 	struct buffer_head *bh_org;
487 	void *kaddr;
488 
489 	bh_org = __bread(nilfs->ns_bdev, rb->blocknr, nilfs->ns_blocksize);
490 	if (unlikely(!bh_org))
491 		return -EIO;
492 
493 	kaddr = kmap_atomic(page);
494 	memcpy(kaddr + bh_offset(bh_org), bh_org->b_data, bh_org->b_size);
495 	kunmap_atomic(kaddr);
496 	brelse(bh_org);
497 	return 0;
498 }
499 
500 static int nilfs_recover_dsync_blocks(struct the_nilfs *nilfs,
501 				      struct super_block *sb,
502 				      struct nilfs_root *root,
503 				      struct list_head *head,
504 				      unsigned long *nr_salvaged_blocks)
505 {
506 	struct inode *inode;
507 	struct nilfs_recovery_block *rb, *n;
508 	unsigned int blocksize = nilfs->ns_blocksize;
509 	struct page *page;
510 	loff_t pos;
511 	int err = 0, err2 = 0;
512 
513 	list_for_each_entry_safe(rb, n, head, list) {
514 		inode = nilfs_iget(sb, root, rb->ino);
515 		if (IS_ERR(inode)) {
516 			err = PTR_ERR(inode);
517 			inode = NULL;
518 			goto failed_inode;
519 		}
520 
521 		pos = rb->blkoff << inode->i_blkbits;
522 		err = block_write_begin(inode->i_mapping, pos, blocksize,
523 					0, &page, nilfs_get_block);
524 		if (unlikely(err)) {
525 			loff_t isize = inode->i_size;
526 
527 			if (pos + blocksize > isize)
528 				nilfs_write_failed(inode->i_mapping,
529 							pos + blocksize);
530 			goto failed_inode;
531 		}
532 
533 		err = nilfs_recovery_copy_block(nilfs, rb, page);
534 		if (unlikely(err))
535 			goto failed_page;
536 
537 		err = nilfs_set_file_dirty(inode, 1);
538 		if (unlikely(err))
539 			goto failed_page;
540 
541 		block_write_end(NULL, inode->i_mapping, pos, blocksize,
542 				blocksize, page, NULL);
543 
544 		unlock_page(page);
545 		put_page(page);
546 
547 		(*nr_salvaged_blocks)++;
548 		goto next;
549 
550  failed_page:
551 		unlock_page(page);
552 		put_page(page);
553 
554  failed_inode:
555 		nilfs_msg(sb, KERN_WARNING,
556 			  "error %d recovering data block (ino=%lu, block-offset=%llu)",
557 			  err, (unsigned long)rb->ino,
558 			  (unsigned long long)rb->blkoff);
559 		if (!err2)
560 			err2 = err;
561  next:
562 		iput(inode); /* iput(NULL) is just ignored */
563 		list_del_init(&rb->list);
564 		kfree(rb);
565 	}
566 	return err2;
567 }
568 
569 /**
570  * nilfs_do_roll_forward - salvage logical segments newer than the latest
571  * checkpoint
572  * @nilfs: nilfs object
573  * @sb: super block instance
574  * @ri: pointer to a nilfs_recovery_info
575  */
576 static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
577 				 struct super_block *sb,
578 				 struct nilfs_root *root,
579 				 struct nilfs_recovery_info *ri)
580 {
581 	struct buffer_head *bh_sum = NULL;
582 	struct nilfs_segment_summary *sum = NULL;
583 	sector_t pseg_start;
584 	sector_t seg_start, seg_end;  /* Starting/ending DBN of full segment */
585 	unsigned long nsalvaged_blocks = 0;
586 	unsigned int flags;
587 	u64 seg_seq;
588 	__u64 segnum, nextnum = 0;
589 	int empty_seg = 0;
590 	int err = 0, ret;
591 	LIST_HEAD(dsync_blocks);  /* list of data blocks to be recovered */
592 	enum {
593 		RF_INIT_ST,
594 		RF_DSYNC_ST,   /* scanning data-sync segments */
595 	};
596 	int state = RF_INIT_ST;
597 
598 	pseg_start = ri->ri_lsegs_start;
599 	seg_seq = ri->ri_lsegs_start_seq;
600 	segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
601 	nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
602 
603 	while (segnum != ri->ri_segnum || pseg_start <= ri->ri_pseg_start) {
604 		brelse(bh_sum);
605 		bh_sum = nilfs_read_log_header(nilfs, pseg_start, &sum);
606 		if (!bh_sum) {
607 			err = -EIO;
608 			goto failed;
609 		}
610 
611 		ret = nilfs_validate_log(nilfs, seg_seq, bh_sum, sum);
612 		if (ret) {
613 			if (ret == NILFS_SEG_FAIL_IO) {
614 				err = -EIO;
615 				goto failed;
616 			}
617 			goto strayed;
618 		}
619 
620 		flags = le16_to_cpu(sum->ss_flags);
621 		if (flags & NILFS_SS_SR)
622 			goto confused;
623 
624 		/* Found a valid partial segment; do recovery actions */
625 		nextnum = nilfs_get_segnum_of_block(nilfs,
626 						    le64_to_cpu(sum->ss_next));
627 		empty_seg = 0;
628 		nilfs->ns_ctime = le64_to_cpu(sum->ss_create);
629 		if (!(flags & NILFS_SS_GC))
630 			nilfs->ns_nongc_ctime = nilfs->ns_ctime;
631 
632 		switch (state) {
633 		case RF_INIT_ST:
634 			if (!(flags & NILFS_SS_LOGBGN) ||
635 			    !(flags & NILFS_SS_SYNDT))
636 				goto try_next_pseg;
637 			state = RF_DSYNC_ST;
638 			/* Fall through */
639 		case RF_DSYNC_ST:
640 			if (!(flags & NILFS_SS_SYNDT))
641 				goto confused;
642 
643 			err = nilfs_scan_dsync_log(nilfs, pseg_start, sum,
644 						   &dsync_blocks);
645 			if (unlikely(err))
646 				goto failed;
647 			if (flags & NILFS_SS_LOGEND) {
648 				err = nilfs_recover_dsync_blocks(
649 					nilfs, sb, root, &dsync_blocks,
650 					&nsalvaged_blocks);
651 				if (unlikely(err))
652 					goto failed;
653 				state = RF_INIT_ST;
654 			}
655 			break; /* Fall through to try_next_pseg */
656 		}
657 
658  try_next_pseg:
659 		if (pseg_start == ri->ri_lsegs_end)
660 			break;
661 		pseg_start += le32_to_cpu(sum->ss_nblocks);
662 		if (pseg_start < seg_end)
663 			continue;
664 		goto feed_segment;
665 
666  strayed:
667 		if (pseg_start == ri->ri_lsegs_end)
668 			break;
669 
670  feed_segment:
671 		/* Looking to the next full segment */
672 		if (empty_seg++)
673 			break;
674 		seg_seq++;
675 		segnum = nextnum;
676 		nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
677 		pseg_start = seg_start;
678 	}
679 
680 	if (nsalvaged_blocks) {
681 		nilfs_msg(sb, KERN_INFO, "salvaged %lu blocks",
682 			  nsalvaged_blocks);
683 		ri->ri_need_recovery = NILFS_RECOVERY_ROLLFORWARD_DONE;
684 	}
685  out:
686 	brelse(bh_sum);
687 	dispose_recovery_list(&dsync_blocks);
688 	return err;
689 
690  confused:
691 	err = -EINVAL;
692  failed:
693 	nilfs_msg(sb, KERN_ERR,
694 		  "error %d roll-forwarding partial segment at blocknr = %llu",
695 		  err, (unsigned long long)pseg_start);
696 	goto out;
697 }
698 
699 static void nilfs_finish_roll_forward(struct the_nilfs *nilfs,
700 				      struct nilfs_recovery_info *ri)
701 {
702 	struct buffer_head *bh;
703 	int err;
704 
705 	if (nilfs_get_segnum_of_block(nilfs, ri->ri_lsegs_start) !=
706 	    nilfs_get_segnum_of_block(nilfs, ri->ri_super_root))
707 		return;
708 
709 	bh = __getblk(nilfs->ns_bdev, ri->ri_lsegs_start, nilfs->ns_blocksize);
710 	BUG_ON(!bh);
711 	memset(bh->b_data, 0, bh->b_size);
712 	set_buffer_dirty(bh);
713 	err = sync_dirty_buffer(bh);
714 	if (unlikely(err))
715 		nilfs_msg(nilfs->ns_sb, KERN_WARNING,
716 			  "buffer sync write failed during post-cleaning of recovery.");
717 	brelse(bh);
718 }
719 
720 /**
721  * nilfs_salvage_orphan_logs - salvage logs written after the latest checkpoint
722  * @nilfs: nilfs object
723  * @sb: super block instance
724  * @ri: pointer to a nilfs_recovery_info struct to store search results.
725  *
726  * Return Value: On success, 0 is returned.  On error, one of the following
727  * negative error code is returned.
728  *
729  * %-EINVAL - Inconsistent filesystem state.
730  *
731  * %-EIO - I/O error
732  *
733  * %-ENOSPC - No space left on device (only in a panic state).
734  *
735  * %-ERESTARTSYS - Interrupted.
736  *
737  * %-ENOMEM - Insufficient memory available.
738  */
739 int nilfs_salvage_orphan_logs(struct the_nilfs *nilfs,
740 			      struct super_block *sb,
741 			      struct nilfs_recovery_info *ri)
742 {
743 	struct nilfs_root *root;
744 	int err;
745 
746 	if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0)
747 		return 0;
748 
749 	err = nilfs_attach_checkpoint(sb, ri->ri_cno, true, &root);
750 	if (unlikely(err)) {
751 		nilfs_msg(sb, KERN_ERR,
752 			  "error %d loading the latest checkpoint", err);
753 		return err;
754 	}
755 
756 	err = nilfs_do_roll_forward(nilfs, sb, root, ri);
757 	if (unlikely(err))
758 		goto failed;
759 
760 	if (ri->ri_need_recovery == NILFS_RECOVERY_ROLLFORWARD_DONE) {
761 		err = nilfs_prepare_segment_for_recovery(nilfs, sb, ri);
762 		if (unlikely(err)) {
763 			nilfs_msg(sb, KERN_ERR,
764 				  "error %d preparing segment for recovery",
765 				  err);
766 			goto failed;
767 		}
768 
769 		err = nilfs_attach_log_writer(sb, root);
770 		if (unlikely(err))
771 			goto failed;
772 
773 		set_nilfs_discontinued(nilfs);
774 		err = nilfs_construct_segment(sb);
775 		nilfs_detach_log_writer(sb);
776 
777 		if (unlikely(err)) {
778 			nilfs_msg(sb, KERN_ERR,
779 				  "error %d writing segment for recovery",
780 				  err);
781 			goto failed;
782 		}
783 
784 		nilfs_finish_roll_forward(nilfs, ri);
785 	}
786 
787  failed:
788 	nilfs_put_root(root);
789 	return err;
790 }
791 
792 /**
793  * nilfs_search_super_root - search the latest valid super root
794  * @nilfs: the_nilfs
795  * @ri: pointer to a nilfs_recovery_info struct to store search results.
796  *
797  * nilfs_search_super_root() looks for the latest super-root from a partial
798  * segment pointed by the superblock.  It sets up struct the_nilfs through
799  * this search. It fills nilfs_recovery_info (ri) required for recovery.
800  *
801  * Return Value: On success, 0 is returned.  On error, one of the following
802  * negative error code is returned.
803  *
804  * %-EINVAL - No valid segment found
805  *
806  * %-EIO - I/O error
807  *
808  * %-ENOMEM - Insufficient memory available.
809  */
810 int nilfs_search_super_root(struct the_nilfs *nilfs,
811 			    struct nilfs_recovery_info *ri)
812 {
813 	struct buffer_head *bh_sum = NULL;
814 	struct nilfs_segment_summary *sum = NULL;
815 	sector_t pseg_start, pseg_end, sr_pseg_start = 0;
816 	sector_t seg_start, seg_end; /* range of full segment (block number) */
817 	sector_t b, end;
818 	unsigned long nblocks;
819 	unsigned int flags;
820 	u64 seg_seq;
821 	__u64 segnum, nextnum = 0;
822 	__u64 cno;
823 	LIST_HEAD(segments);
824 	int empty_seg = 0, scan_newer = 0;
825 	int ret;
826 
827 	pseg_start = nilfs->ns_last_pseg;
828 	seg_seq = nilfs->ns_last_seq;
829 	cno = nilfs->ns_last_cno;
830 	segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
831 
832 	/* Calculate range of segment */
833 	nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
834 
835 	/* Read ahead segment */
836 	b = seg_start;
837 	while (b <= seg_end)
838 		__breadahead(nilfs->ns_bdev, b++, nilfs->ns_blocksize);
839 
840 	for (;;) {
841 		brelse(bh_sum);
842 		ret = NILFS_SEG_FAIL_IO;
843 		bh_sum = nilfs_read_log_header(nilfs, pseg_start, &sum);
844 		if (!bh_sum)
845 			goto failed;
846 
847 		ret = nilfs_validate_log(nilfs, seg_seq, bh_sum, sum);
848 		if (ret) {
849 			if (ret == NILFS_SEG_FAIL_IO)
850 				goto failed;
851 			goto strayed;
852 		}
853 
854 		nblocks = le32_to_cpu(sum->ss_nblocks);
855 		pseg_end = pseg_start + nblocks - 1;
856 		if (unlikely(pseg_end > seg_end)) {
857 			ret = NILFS_SEG_FAIL_CONSISTENCY;
858 			goto strayed;
859 		}
860 
861 		/* A valid partial segment */
862 		ri->ri_pseg_start = pseg_start;
863 		ri->ri_seq = seg_seq;
864 		ri->ri_segnum = segnum;
865 		nextnum = nilfs_get_segnum_of_block(nilfs,
866 						    le64_to_cpu(sum->ss_next));
867 		ri->ri_nextnum = nextnum;
868 		empty_seg = 0;
869 
870 		flags = le16_to_cpu(sum->ss_flags);
871 		if (!(flags & NILFS_SS_SR) && !scan_newer) {
872 			/*
873 			 * This will never happen because a superblock
874 			 * (last_segment) always points to a pseg with
875 			 * a super root.
876 			 */
877 			ret = NILFS_SEG_FAIL_CONSISTENCY;
878 			goto failed;
879 		}
880 
881 		if (pseg_start == seg_start) {
882 			nilfs_get_segment_range(nilfs, nextnum, &b, &end);
883 			while (b <= end)
884 				__breadahead(nilfs->ns_bdev, b++,
885 					     nilfs->ns_blocksize);
886 		}
887 		if (!(flags & NILFS_SS_SR)) {
888 			if (!ri->ri_lsegs_start && (flags & NILFS_SS_LOGBGN)) {
889 				ri->ri_lsegs_start = pseg_start;
890 				ri->ri_lsegs_start_seq = seg_seq;
891 			}
892 			if (flags & NILFS_SS_LOGEND)
893 				ri->ri_lsegs_end = pseg_start;
894 			goto try_next_pseg;
895 		}
896 
897 		/* A valid super root was found. */
898 		ri->ri_cno = cno++;
899 		ri->ri_super_root = pseg_end;
900 		ri->ri_lsegs_start = ri->ri_lsegs_end = 0;
901 
902 		nilfs_dispose_segment_list(&segments);
903 		sr_pseg_start = pseg_start;
904 		nilfs->ns_pseg_offset = pseg_start + nblocks - seg_start;
905 		nilfs->ns_seg_seq = seg_seq;
906 		nilfs->ns_segnum = segnum;
907 		nilfs->ns_cno = cno;  /* nilfs->ns_cno = ri->ri_cno + 1 */
908 		nilfs->ns_ctime = le64_to_cpu(sum->ss_create);
909 		nilfs->ns_nextnum = nextnum;
910 
911 		if (scan_newer)
912 			ri->ri_need_recovery = NILFS_RECOVERY_SR_UPDATED;
913 		else {
914 			if (nilfs->ns_mount_state & NILFS_VALID_FS)
915 				goto super_root_found;
916 			scan_newer = 1;
917 		}
918 
919  try_next_pseg:
920 		/* Standing on a course, or met an inconsistent state */
921 		pseg_start += nblocks;
922 		if (pseg_start < seg_end)
923 			continue;
924 		goto feed_segment;
925 
926  strayed:
927 		/* Off the trail */
928 		if (!scan_newer)
929 			/*
930 			 * This can happen if a checkpoint was written without
931 			 * barriers, or as a result of an I/O failure.
932 			 */
933 			goto failed;
934 
935  feed_segment:
936 		/* Looking to the next full segment */
937 		if (empty_seg++)
938 			goto super_root_found; /* found a valid super root */
939 
940 		ret = nilfs_segment_list_add(&segments, segnum);
941 		if (unlikely(ret))
942 			goto failed;
943 
944 		seg_seq++;
945 		segnum = nextnum;
946 		nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
947 		pseg_start = seg_start;
948 	}
949 
950  super_root_found:
951 	/* Updating pointers relating to the latest checkpoint */
952 	brelse(bh_sum);
953 	list_splice_tail(&segments, &ri->ri_used_segments);
954 	nilfs->ns_last_pseg = sr_pseg_start;
955 	nilfs->ns_last_seq = nilfs->ns_seg_seq;
956 	nilfs->ns_last_cno = ri->ri_cno;
957 	return 0;
958 
959  failed:
960 	brelse(bh_sum);
961 	nilfs_dispose_segment_list(&segments);
962 	return ret < 0 ? ret : nilfs_warn_segment_error(nilfs->ns_sb, ret);
963 }
964