recovery.c 24.4 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
/*
 * recovery.c - NILFS recovery logic
 *
 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 *
 * Written by Ryusuke Konishi <ryusuke@osrg.net>
 */

#include <linux/buffer_head.h>
#include <linux/blkdev.h>
#include <linux/swap.h>
26
#include <linux/slab.h>
27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94
#include <linux/crc32.h>
#include "nilfs.h"
#include "segment.h"
#include "sufile.h"
#include "page.h"
#include "segbuf.h"

/*
 * Segment check result
 */
enum {
	NILFS_SEG_VALID,
	NILFS_SEG_NO_SUPER_ROOT,
	NILFS_SEG_FAIL_IO,
	NILFS_SEG_FAIL_MAGIC,
	NILFS_SEG_FAIL_SEQ,
	NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT,
	NILFS_SEG_FAIL_CHECKSUM_FULL,
	NILFS_SEG_FAIL_CONSISTENCY,
};

/* work structure for recovery */
struct nilfs_recovery_block {
	ino_t ino;		/* Inode number of the file that this block
				   belongs to */
	sector_t blocknr;	/* block number */
	__u64 vblocknr;		/* virtual block number */
	unsigned long blkoff;	/* File offset of the data block (per block) */
	struct list_head list;
};


static int nilfs_warn_segment_error(int err)
{
	switch (err) {
	case NILFS_SEG_FAIL_IO:
		printk(KERN_WARNING
		       "NILFS warning: I/O error on loading last segment\n");
		return -EIO;
	case NILFS_SEG_FAIL_MAGIC:
		printk(KERN_WARNING
		       "NILFS warning: Segment magic number invalid\n");
		break;
	case NILFS_SEG_FAIL_SEQ:
		printk(KERN_WARNING
		       "NILFS warning: Sequence number mismatch\n");
		break;
	case NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT:
		printk(KERN_WARNING
		       "NILFS warning: Checksum error in super root\n");
		break;
	case NILFS_SEG_FAIL_CHECKSUM_FULL:
		printk(KERN_WARNING
		       "NILFS warning: Checksum error in segment payload\n");
		break;
	case NILFS_SEG_FAIL_CONSISTENCY:
		printk(KERN_WARNING
		       "NILFS warning: Inconsistent segment\n");
		break;
	case NILFS_SEG_NO_SUPER_ROOT:
		printk(KERN_WARNING
		       "NILFS warning: No super root in the last segment\n");
		break;
	}
	return -EINVAL;
}

/**
95 96
 * nilfs_compute_checksum - compute checksum of blocks continuously
 * @nilfs: nilfs object
97 98 99 100 101 102 103
 * @bhs: buffer head of start block
 * @sum: place to store result
 * @offset: offset bytes in the first block
 * @check_bytes: number of bytes to be checked
 * @start: DBN of start block
 * @nblock: number of blocks to be checked
 */
104 105 106 107
static int nilfs_compute_checksum(struct the_nilfs *nilfs,
				  struct buffer_head *bhs, u32 *sum,
				  unsigned long offset, u64 check_bytes,
				  sector_t start, unsigned long nblock)
108
{
109
	unsigned int blocksize = nilfs->ns_blocksize;
110 111 112 113 114 115
	unsigned long size;
	u32 crc;

	BUG_ON(offset >= blocksize);
	check_bytes -= offset;
	size = min_t(u64, check_bytes, blocksize - offset);
116
	crc = crc32_le(nilfs->ns_crc_seed,
117 118 119
		       (unsigned char *)bhs->b_data + offset, size);
	if (--nblock > 0) {
		do {
120 121 122
			struct buffer_head *bh;

			bh = __bread(nilfs->ns_bdev, ++start, blocksize);
123 124 125 126 127 128 129 130 131 132 133 134 135 136
			if (!bh)
				return -EIO;
			check_bytes -= size;
			size = min_t(u64, check_bytes, blocksize);
			crc = crc32_le(crc, bh->b_data, size);
			brelse(bh);
		} while (--nblock > 0);
	}
	*sum = crc;
	return 0;
}

/**
 * nilfs_read_super_root_block - read super root block
137
 * @nilfs: nilfs object
138 139 140 141
 * @sr_block: disk block number of the super root block
 * @pbh: address of a buffer_head pointer to return super root buffer
 * @check: CRC check flag
 */
142
int nilfs_read_super_root_block(struct the_nilfs *nilfs, sector_t sr_block,
143 144 145 146 147 148 149 150
				struct buffer_head **pbh, int check)
{
	struct buffer_head *bh_sr;
	struct nilfs_super_root *sr;
	u32 crc;
	int ret;

	*pbh = NULL;
151
	bh_sr = __bread(nilfs->ns_bdev, sr_block, nilfs->ns_blocksize);
152 153 154 155 156 157 158 159 160
	if (unlikely(!bh_sr)) {
		ret = NILFS_SEG_FAIL_IO;
		goto failed;
	}

	sr = (struct nilfs_super_root *)bh_sr->b_data;
	if (check) {
		unsigned bytes = le16_to_cpu(sr->sr_bytes);

161
		if (bytes == 0 || bytes > nilfs->ns_blocksize) {
162 163 164
			ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
			goto failed_bh;
		}
165 166 167
		if (nilfs_compute_checksum(
			    nilfs, bh_sr, &crc, sizeof(sr->sr_sum), bytes,
			    sr_block, 1)) {
168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186
			ret = NILFS_SEG_FAIL_IO;
			goto failed_bh;
		}
		if (crc != le32_to_cpu(sr->sr_sum)) {
			ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
			goto failed_bh;
		}
	}
	*pbh = bh_sr;
	return 0;

 failed_bh:
	brelse(bh_sr);

 failed:
	return nilfs_warn_segment_error(ret);
}

/**
187
 * nilfs_read_log_header - read summary header of the specified log
188
 * @nilfs: nilfs object
189 190
 * @start_blocknr: start block number of the log
 * @sum: pointer to return segment summary structure
191
 */
192 193 194
static struct buffer_head *
nilfs_read_log_header(struct the_nilfs *nilfs, sector_t start_blocknr,
		      struct nilfs_segment_summary **sum)
195 196
{
	struct buffer_head *bh_sum;
197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214

	bh_sum = __bread(nilfs->ns_bdev, start_blocknr, nilfs->ns_blocksize);
	if (bh_sum)
		*sum = (struct nilfs_segment_summary *)bh_sum->b_data;
	return bh_sum;
}

/**
 * nilfs_validate_log - verify consistency of log
 * @nilfs: nilfs object
 * @seg_seq: sequence number of segment
 * @bh_sum: buffer head of summary block
 * @sum: segment summary struct
 */
static int nilfs_validate_log(struct the_nilfs *nilfs, u64 seg_seq,
			      struct buffer_head *bh_sum,
			      struct nilfs_segment_summary *sum)
{
215 216
	unsigned long nblock;
	u32 crc;
217
	int ret;
218

219 220
	ret = NILFS_SEG_FAIL_MAGIC;
	if (le32_to_cpu(sum->ss_magic) != NILFS_SEGSUM_MAGIC)
221 222
		goto out;

223 224 225
	ret = NILFS_SEG_FAIL_SEQ;
	if (le64_to_cpu(sum->ss_seq) != seg_seq)
		goto out;
226

227 228 229
	nblock = le32_to_cpu(sum->ss_nblocks);
	ret = NILFS_SEG_FAIL_CONSISTENCY;
	if (unlikely(nblock == 0 || nblock > nilfs->ns_blocks_per_segment))
230
		/* This limits the number of blocks read in the CRC check */
231 232 233
		goto out;

	ret = NILFS_SEG_FAIL_IO;
234 235
	if (nilfs_compute_checksum(nilfs, bh_sum, &crc, sizeof(sum->ss_datasum),
				   ((u64)nblock << nilfs->ns_blocksize_bits),
236 237 238 239 240 241 242 243
				   bh_sum->b_blocknr, nblock))
		goto out;

	ret = NILFS_SEG_FAIL_CHECKSUM_FULL;
	if (crc != le32_to_cpu(sum->ss_datasum))
		goto out;
	ret = 0;
out:
244 245 246
	return ret;
}

247 248 249 250 251 252 253 254 255 256
/**
 * nilfs_read_summary_info - read an item on summary blocks of a log
 * @nilfs: nilfs object
 * @pbh: the current buffer head on summary blocks [in, out]
 * @offset: the current byte offset on summary blocks [in, out]
 * @bytes: byte size of the item to be read
 */
static void *nilfs_read_summary_info(struct the_nilfs *nilfs,
				     struct buffer_head **pbh,
				     unsigned int *offset, unsigned int bytes)
257 258 259 260 261 262 263 264
{
	void *ptr;
	sector_t blocknr;

	BUG_ON((*pbh)->b_size < *offset);
	if (bytes > (*pbh)->b_size - *offset) {
		blocknr = (*pbh)->b_blocknr;
		brelse(*pbh);
265 266
		*pbh = __bread(nilfs->ns_bdev, blocknr + 1,
			       nilfs->ns_blocksize);
267 268 269 270 271 272 273 274 275
		if (unlikely(!*pbh))
			return NULL;
		*offset = 0;
	}
	ptr = (*pbh)->b_data + *offset;
	*offset += bytes;
	return ptr;
}

276 277 278 279 280 281 282 283 284 285 286 287
/**
 * nilfs_skip_summary_info - skip items on summary blocks of a log
 * @nilfs: nilfs object
 * @pbh: the current buffer head on summary blocks [in, out]
 * @offset: the current byte offset on summary blocks [in, out]
 * @bytes: byte size of the item to be skipped
 * @count: number of items to be skipped
 */
static void nilfs_skip_summary_info(struct the_nilfs *nilfs,
				    struct buffer_head **pbh,
				    unsigned int *offset, unsigned int bytes,
				    unsigned long count)
288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303
{
	unsigned int rest_item_in_current_block
		= ((*pbh)->b_size - *offset) / bytes;

	if (count <= rest_item_in_current_block) {
		*offset += bytes * count;
	} else {
		sector_t blocknr = (*pbh)->b_blocknr;
		unsigned int nitem_per_block = (*pbh)->b_size / bytes;
		unsigned int bcnt;

		count -= rest_item_in_current_block;
		bcnt = DIV_ROUND_UP(count, nitem_per_block);
		*offset = bytes * (count - (bcnt - 1) * nitem_per_block);

		brelse(*pbh);
304 305
		*pbh = __bread(nilfs->ns_bdev, blocknr + bcnt,
			       nilfs->ns_blocksize);
306 307 308
	}
}

309 310 311 312
/**
 * nilfs_scan_dsync_log - get block information of a log written for data sync
 * @nilfs: nilfs object
 * @start_blocknr: start block number of the log
313
 * @sum: log summary information
314 315 316
 * @head: list head to add nilfs_recovery_block struct
 */
static int nilfs_scan_dsync_log(struct the_nilfs *nilfs, sector_t start_blocknr,
317
				struct nilfs_segment_summary *sum,
318
				struct list_head *head)
319 320 321
{
	struct buffer_head *bh;
	unsigned int offset;
322 323
	u32 nfinfo, sumbytes;
	sector_t blocknr;
324 325 326
	ino_t ino;
	int err = -EIO;

327
	nfinfo = le32_to_cpu(sum->ss_nfinfo);
328 329 330
	if (!nfinfo)
		return 0;

331 332
	sumbytes = le32_to_cpu(sum->ss_sumbytes);
	blocknr = start_blocknr + DIV_ROUND_UP(sumbytes, nilfs->ns_blocksize);
333
	bh = __bread(nilfs->ns_bdev, start_blocknr, nilfs->ns_blocksize);
334 335 336
	if (unlikely(!bh))
		goto out;

337
	offset = le16_to_cpu(sum->ss_bytes);
338 339 340 341
	for (;;) {
		unsigned long nblocks, ndatablk, nnodeblk;
		struct nilfs_finfo *finfo;

342 343
		finfo = nilfs_read_summary_info(nilfs, &bh, &offset,
						sizeof(*finfo));
344 345 346 347 348 349 350 351 352 353 354 355
		if (unlikely(!finfo))
			goto out;

		ino = le64_to_cpu(finfo->fi_ino);
		nblocks = le32_to_cpu(finfo->fi_nblocks);
		ndatablk = le32_to_cpu(finfo->fi_ndatablk);
		nnodeblk = nblocks - ndatablk;

		while (ndatablk-- > 0) {
			struct nilfs_recovery_block *rb;
			struct nilfs_binfo_v *binfo;

356 357
			binfo = nilfs_read_summary_info(nilfs, &bh, &offset,
							sizeof(*binfo));
358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374
			if (unlikely(!binfo))
				goto out;

			rb = kmalloc(sizeof(*rb), GFP_NOFS);
			if (unlikely(!rb)) {
				err = -ENOMEM;
				goto out;
			}
			rb->ino = ino;
			rb->blocknr = blocknr++;
			rb->vblocknr = le64_to_cpu(binfo->bi_vblocknr);
			rb->blkoff = le64_to_cpu(binfo->bi_blkoff);
			/* INIT_LIST_HEAD(&rb->list); */
			list_add_tail(&rb->list, head);
		}
		if (--nfinfo == 0)
			break;
375 376 377
		blocknr += nnodeblk; /* always 0 for data sync logs */
		nilfs_skip_summary_info(nilfs, &bh, &offset, sizeof(__le64),
					nnodeblk);
378 379 380 381 382 383 384 385 386 387 388 389
		if (unlikely(!bh))
			goto out;
	}
	err = 0;
 out:
	brelse(bh);   /* brelse(NULL) is just ignored */
	return err;
}

static void dispose_recovery_list(struct list_head *head)
{
	while (!list_empty(head)) {
390 391 392
		struct nilfs_recovery_block *rb;

		rb = list_first_entry(head, struct nilfs_recovery_block, list);
393 394 395 396 397
		list_del(&rb->list);
		kfree(rb);
	}
}

398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415
struct nilfs_segment_entry {
	struct list_head	list;
	__u64			segnum;
};

static int nilfs_segment_list_add(struct list_head *head, __u64 segnum)
{
	struct nilfs_segment_entry *ent = kmalloc(sizeof(*ent), GFP_NOFS);

	if (unlikely(!ent))
		return -ENOMEM;

	ent->segnum = segnum;
	INIT_LIST_HEAD(&ent->list);
	list_add_tail(&ent->list, head);
	return 0;
}

416 417 418
void nilfs_dispose_segment_list(struct list_head *head)
{
	while (!list_empty(head)) {
419 420 421
		struct nilfs_segment_entry *ent;

		ent = list_first_entry(head, struct nilfs_segment_entry, list);
422
		list_del(&ent->list);
423
		kfree(ent);
424 425 426 427
	}
}

static int nilfs_prepare_segment_for_recovery(struct the_nilfs *nilfs,
428
					      struct super_block *sb,
429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451
					      struct nilfs_recovery_info *ri)
{
	struct list_head *head = &ri->ri_used_segments;
	struct nilfs_segment_entry *ent, *n;
	struct inode *sufile = nilfs->ns_sufile;
	__u64 segnum[4];
	int err;
	int i;

	segnum[0] = nilfs->ns_segnum;
	segnum[1] = nilfs->ns_nextnum;
	segnum[2] = ri->ri_segnum;
	segnum[3] = ri->ri_nextnum;

	/*
	 * Releasing the next segment of the latest super root.
	 * The next segment is invalidated by this recovery.
	 */
	err = nilfs_sufile_free(sufile, segnum[1]);
	if (unlikely(err))
		goto failed;

	for (i = 1; i < 4; i++) {
452 453
		err = nilfs_segment_list_add(head, segnum[i]);
		if (unlikely(err))
454 455 456 457 458
			goto failed;
	}

	/*
	 * Collecting segments written after the latest super root.
459
	 * These are marked dirty to avoid being reallocated in the next write.
460 461
	 */
	list_for_each_entry_safe(ent, n, head, list) {
462 463 464 465
		if (ent->segnum != segnum[0]) {
			err = nilfs_sufile_scrap(sufile, ent->segnum);
			if (unlikely(err))
				goto failed;
466
		}
467
		list_del(&ent->list);
468
		kfree(ent);
469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484
	}

	/* Allocate new segments for recovery */
	err = nilfs_sufile_alloc(sufile, &segnum[0]);
	if (unlikely(err))
		goto failed;

	nilfs->ns_pseg_offset = 0;
	nilfs->ns_seg_seq = ri->ri_seq + 2;
	nilfs->ns_nextnum = nilfs->ns_segnum = segnum[0];

 failed:
	/* No need to recover sufile because it will be destroyed on error */
	return err;
}

485
static int nilfs_recovery_copy_block(struct the_nilfs *nilfs,
486 487 488 489 490 491
				     struct nilfs_recovery_block *rb,
				     struct page *page)
{
	struct buffer_head *bh_org;
	void *kaddr;

492
	bh_org = __bread(nilfs->ns_bdev, rb->blocknr, nilfs->ns_blocksize);
493 494 495
	if (unlikely(!bh_org))
		return -EIO;

496
	kaddr = kmap_atomic(page);
497
	memcpy(kaddr + bh_offset(bh_org), bh_org->b_data, bh_org->b_size);
498
	kunmap_atomic(kaddr);
499 500 501 502
	brelse(bh_org);
	return 0;
}

503
static int nilfs_recover_dsync_blocks(struct the_nilfs *nilfs,
504
				      struct super_block *sb,
505
				      struct nilfs_root *root,
506 507
				      struct list_head *head,
				      unsigned long *nr_salvaged_blocks)
508 509 510
{
	struct inode *inode;
	struct nilfs_recovery_block *rb, *n;
511
	unsigned blocksize = nilfs->ns_blocksize;
512 513 514 515 516
	struct page *page;
	loff_t pos;
	int err = 0, err2 = 0;

	list_for_each_entry_safe(rb, n, head, list) {
517
		inode = nilfs_iget(sb, root, rb->ino);
518 519 520 521 522 523 524
		if (IS_ERR(inode)) {
			err = PTR_ERR(inode);
			inode = NULL;
			goto failed_inode;
		}

		pos = rb->blkoff << inode->i_blkbits;
525 526 527 528 529 530
		err = block_write_begin(inode->i_mapping, pos, blocksize,
					0, &page, nilfs_get_block);
		if (unlikely(err)) {
			loff_t isize = inode->i_size;
			if (pos + blocksize > isize)
				vmtruncate(inode, isize);
531
			goto failed_inode;
532
		}
533

534
		err = nilfs_recovery_copy_block(nilfs, rb, page);
535 536 537
		if (unlikely(err))
			goto failed_page;

538
		err = nilfs_set_file_dirty(inode, 1);
539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558
		if (unlikely(err))
			goto failed_page;

		block_write_end(NULL, inode->i_mapping, pos, blocksize,
				blocksize, page, NULL);

		unlock_page(page);
		page_cache_release(page);

		(*nr_salvaged_blocks)++;
		goto next;

 failed_page:
		unlock_page(page);
		page_cache_release(page);

 failed_inode:
		printk(KERN_WARNING
		       "NILFS warning: error recovering data block "
		       "(err=%d, ino=%lu, block-offset=%llu)\n",
559 560
		       err, (unsigned long)rb->ino,
		       (unsigned long long)rb->blkoff);
561 562 563 564 565 566 567 568 569 570 571 572 573
		if (!err2)
			err2 = err;
 next:
		iput(inode); /* iput(NULL) is just ignored */
		list_del_init(&rb->list);
		kfree(rb);
	}
	return err2;
}

/**
 * nilfs_do_roll_forward - salvage logical segments newer than the latest
 * checkpoint
574
 * @nilfs: nilfs object
575
 * @sb: super block instance
576 577 578
 * @ri: pointer to a nilfs_recovery_info
 */
static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
579
				 struct super_block *sb,
580
				 struct nilfs_root *root,
581 582
				 struct nilfs_recovery_info *ri)
{
583 584
	struct buffer_head *bh_sum = NULL;
	struct nilfs_segment_summary *sum;
585 586 587
	sector_t pseg_start;
	sector_t seg_start, seg_end;  /* Starting/ending DBN of full segment */
	unsigned long nsalvaged_blocks = 0;
588
	unsigned int flags;
589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605
	u64 seg_seq;
	__u64 segnum, nextnum = 0;
	int empty_seg = 0;
	int err = 0, ret;
	LIST_HEAD(dsync_blocks);  /* list of data blocks to be recovered */
	enum {
		RF_INIT_ST,
		RF_DSYNC_ST,   /* scanning data-sync segments */
	};
	int state = RF_INIT_ST;

	pseg_start = ri->ri_lsegs_start;
	seg_seq = ri->ri_lsegs_start_seq;
	segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
	nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);

	while (segnum != ri->ri_segnum || pseg_start <= ri->ri_pseg_start) {
606 607 608 609 610 611
		brelse(bh_sum);
		bh_sum = nilfs_read_log_header(nilfs, pseg_start, &sum);
		if (!bh_sum) {
			err = -EIO;
			goto failed;
		}
612

613
		ret = nilfs_validate_log(nilfs, seg_seq, bh_sum, sum);
614 615 616 617 618 619 620
		if (ret) {
			if (ret == NILFS_SEG_FAIL_IO) {
				err = -EIO;
				goto failed;
			}
			goto strayed;
		}
621

622 623
		flags = le16_to_cpu(sum->ss_flags);
		if (flags & NILFS_SS_SR)
624 625 626
			goto confused;

		/* Found a valid partial segment; do recovery actions */
627 628
		nextnum = nilfs_get_segnum_of_block(nilfs,
						    le64_to_cpu(sum->ss_next));
629
		empty_seg = 0;
630 631 632
		nilfs->ns_ctime = le64_to_cpu(sum->ss_create);
		if (!(flags & NILFS_SS_GC))
			nilfs->ns_nongc_ctime = nilfs->ns_ctime;
633 634 635

		switch (state) {
		case RF_INIT_ST:
636 637
			if (!(flags & NILFS_SS_LOGBGN) ||
			    !(flags & NILFS_SS_SYNDT))
638 639 640 641
				goto try_next_pseg;
			state = RF_DSYNC_ST;
			/* Fall through */
		case RF_DSYNC_ST:
642
			if (!(flags & NILFS_SS_SYNDT))
643 644
				goto confused;

645
			err = nilfs_scan_dsync_log(nilfs, pseg_start, sum,
646
						   &dsync_blocks);
647 648
			if (unlikely(err))
				goto failed;
649
			if (flags & NILFS_SS_LOGEND) {
650
				err = nilfs_recover_dsync_blocks(
651
					nilfs, sb, root, &dsync_blocks,
652
					&nsalvaged_blocks);
653 654 655 656 657 658 659 660 661 662
				if (unlikely(err))
					goto failed;
				state = RF_INIT_ST;
			}
			break; /* Fall through to try_next_pseg */
		}

 try_next_pseg:
		if (pseg_start == ri->ri_lsegs_end)
			break;
663
		pseg_start += le32_to_cpu(sum->ss_nblocks);
664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683
		if (pseg_start < seg_end)
			continue;
		goto feed_segment;

 strayed:
		if (pseg_start == ri->ri_lsegs_end)
			break;

 feed_segment:
		/* Looking to the next full segment */
		if (empty_seg++)
			break;
		seg_seq++;
		segnum = nextnum;
		nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
		pseg_start = seg_start;
	}

	if (nsalvaged_blocks) {
		printk(KERN_INFO "NILFS (device %s): salvaged %lu blocks\n",
684
		       sb->s_id, nsalvaged_blocks);
685 686 687
		ri->ri_need_recovery = NILFS_RECOVERY_ROLLFORWARD_DONE;
	}
 out:
688
	brelse(bh_sum);
689 690 691 692 693 694 695 696 697
	dispose_recovery_list(&dsync_blocks);
	return err;

 confused:
	err = -EINVAL;
 failed:
	printk(KERN_ERR
	       "NILFS (device %s): Error roll-forwarding "
	       "(err=%d, pseg block=%llu). ",
698
	       sb->s_id, err, (unsigned long long)pseg_start);
699 700 701 702 703 704 705 706 707 708 709 710 711
	goto out;
}

static void nilfs_finish_roll_forward(struct the_nilfs *nilfs,
				      struct nilfs_recovery_info *ri)
{
	struct buffer_head *bh;
	int err;

	if (nilfs_get_segnum_of_block(nilfs, ri->ri_lsegs_start) !=
	    nilfs_get_segnum_of_block(nilfs, ri->ri_super_root))
		return;

712
	bh = __getblk(nilfs->ns_bdev, ri->ri_lsegs_start, nilfs->ns_blocksize);
713 714 715 716 717 718 719 720 721 722 723 724
	BUG_ON(!bh);
	memset(bh->b_data, 0, bh->b_size);
	set_buffer_dirty(bh);
	err = sync_dirty_buffer(bh);
	if (unlikely(err))
		printk(KERN_WARNING
		       "NILFS warning: buffer sync write failed during "
		       "post-cleaning of recovery.\n");
	brelse(bh);
}

/**
725 726
 * nilfs_salvage_orphan_logs - salvage logs written after the latest checkpoint
 * @nilfs: nilfs object
727
 * @sb: super block instance
728 729 730 731 732 733 734 735 736 737 738 739 740 741 742
 * @ri: pointer to a nilfs_recovery_info struct to store search results.
 *
 * Return Value: On success, 0 is returned.  On error, one of the following
 * negative error code is returned.
 *
 * %-EINVAL - Inconsistent filesystem state.
 *
 * %-EIO - I/O error
 *
 * %-ENOSPC - No space left on device (only in a panic state).
 *
 * %-ERESTARTSYS - Interrupted.
 *
 * %-ENOMEM - Insufficient memory available.
 */
743
int nilfs_salvage_orphan_logs(struct the_nilfs *nilfs,
744
			      struct super_block *sb,
745
			      struct nilfs_recovery_info *ri)
746
{
747
	struct nilfs_root *root;
748 749 750 751 752
	int err;

	if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0)
		return 0;

753
	err = nilfs_attach_checkpoint(sb, ri->ri_cno, true, &root);
754 755 756 757 758 759
	if (unlikely(err)) {
		printk(KERN_ERR
		       "NILFS: error loading the latest checkpoint.\n");
		return err;
	}

760
	err = nilfs_do_roll_forward(nilfs, sb, root, ri);
761 762 763 764
	if (unlikely(err))
		goto failed;

	if (ri->ri_need_recovery == NILFS_RECOVERY_ROLLFORWARD_DONE) {
765
		err = nilfs_prepare_segment_for_recovery(nilfs, sb, ri);
766 767 768 769 770 771
		if (unlikely(err)) {
			printk(KERN_ERR "NILFS: Error preparing segments for "
			       "recovery.\n");
			goto failed;
		}

772
		err = nilfs_attach_log_writer(sb, root);
773 774 775 776
		if (unlikely(err))
			goto failed;

		set_nilfs_discontinued(nilfs);
777 778
		err = nilfs_construct_segment(sb);
		nilfs_detach_log_writer(sb);
779 780 781 782 783 784 785

		if (unlikely(err)) {
			printk(KERN_ERR "NILFS: Oops! recovery failed. "
			       "(err=%d)\n", err);
			goto failed;
		}

786
		nilfs_finish_roll_forward(nilfs, ri);
787 788 789
	}

 failed:
790
	nilfs_put_root(root);
791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808
	return err;
}

/**
 * nilfs_search_super_root - search the latest valid super root
 * @nilfs: the_nilfs
 * @ri: pointer to a nilfs_recovery_info struct to store search results.
 *
 * nilfs_search_super_root() looks for the latest super-root from a partial
 * segment pointed by the superblock.  It sets up struct the_nilfs through
 * this search. It fills nilfs_recovery_info (ri) required for recovery.
 *
 * Return Value: On success, 0 is returned.  On error, one of the following
 * negative error code is returned.
 *
 * %-EINVAL - No valid segment found
 *
 * %-EIO - I/O error
809 810
 *
 * %-ENOMEM - Insufficient memory available.
811
 */
812
int nilfs_search_super_root(struct the_nilfs *nilfs,
813 814
			    struct nilfs_recovery_info *ri)
{
815 816
	struct buffer_head *bh_sum = NULL;
	struct nilfs_segment_summary *sum;
817 818
	sector_t pseg_start, pseg_end, sr_pseg_start = 0;
	sector_t seg_start, seg_end; /* range of full segment (block number) */
819
	sector_t b, end;
820 821
	unsigned long nblocks;
	unsigned int flags;
822 823 824 825 826 827 828 829 830 831 832 833 834 835 836
	u64 seg_seq;
	__u64 segnum, nextnum = 0;
	__u64 cno;
	LIST_HEAD(segments);
	int empty_seg = 0, scan_newer = 0;
	int ret;

	pseg_start = nilfs->ns_last_pseg;
	seg_seq = nilfs->ns_last_seq;
	cno = nilfs->ns_last_cno;
	segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);

	/* Calculate range of segment */
	nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);

837 838 839
	/* Read ahead segment */
	b = seg_start;
	while (b <= seg_end)
840
		__breadahead(nilfs->ns_bdev, b++, nilfs->ns_blocksize);
841

842
	for (;;) {
843 844 845 846 847 848 849
		brelse(bh_sum);
		ret = NILFS_SEG_FAIL_IO;
		bh_sum = nilfs_read_log_header(nilfs, pseg_start, &sum);
		if (!bh_sum)
			goto failed;

		ret = nilfs_validate_log(nilfs, seg_seq, bh_sum, sum);
850 851 852 853 854
		if (ret) {
			if (ret == NILFS_SEG_FAIL_IO)
				goto failed;
			goto strayed;
		}
855

856 857
		nblocks = le32_to_cpu(sum->ss_nblocks);
		pseg_end = pseg_start + nblocks - 1;
858 859 860 861 862 863 864 865 866
		if (unlikely(pseg_end > seg_end)) {
			ret = NILFS_SEG_FAIL_CONSISTENCY;
			goto strayed;
		}

		/* A valid partial segment */
		ri->ri_pseg_start = pseg_start;
		ri->ri_seq = seg_seq;
		ri->ri_segnum = segnum;
867 868
		nextnum = nilfs_get_segnum_of_block(nilfs,
						    le64_to_cpu(sum->ss_next));
869 870 871
		ri->ri_nextnum = nextnum;
		empty_seg = 0;

872 873
		flags = le16_to_cpu(sum->ss_flags);
		if (!(flags & NILFS_SS_SR) && !scan_newer) {
874 875 876 877 878 879 880 881 882 883
			/* This will never happen because a superblock
			   (last_segment) always points to a pseg
			   having a super root. */
			ret = NILFS_SEG_FAIL_CONSISTENCY;
			goto failed;
		}

		if (pseg_start == seg_start) {
			nilfs_get_segment_range(nilfs, nextnum, &b, &end);
			while (b <= end)
884 885
				__breadahead(nilfs->ns_bdev, b++,
					     nilfs->ns_blocksize);
886
		}
887 888
		if (!(flags & NILFS_SS_SR)) {
			if (!ri->ri_lsegs_start && (flags & NILFS_SS_LOGBGN)) {
889 890 891
				ri->ri_lsegs_start = pseg_start;
				ri->ri_lsegs_start_seq = seg_seq;
			}
892
			if (flags & NILFS_SS_LOGEND)
893 894 895 896 897 898 899 900 901 902
				ri->ri_lsegs_end = pseg_start;
			goto try_next_pseg;
		}

		/* A valid super root was found. */
		ri->ri_cno = cno++;
		ri->ri_super_root = pseg_end;
		ri->ri_lsegs_start = ri->ri_lsegs_end = 0;

		nilfs_dispose_segment_list(&segments);
903 904
		sr_pseg_start = pseg_start;
		nilfs->ns_pseg_offset = pseg_start + nblocks - seg_start;
905 906 907
		nilfs->ns_seg_seq = seg_seq;
		nilfs->ns_segnum = segnum;
		nilfs->ns_cno = cno;  /* nilfs->ns_cno = ri->ri_cno + 1 */
908
		nilfs->ns_ctime = le64_to_cpu(sum->ss_create);
909 910 911 912
		nilfs->ns_nextnum = nextnum;

		if (scan_newer)
			ri->ri_need_recovery = NILFS_RECOVERY_SR_UPDATED;
913 914 915 916 917
		else {
			if (nilfs->ns_mount_state & NILFS_VALID_FS)
				goto super_root_found;
			scan_newer = 1;
		}
918 919 920

 try_next_pseg:
		/* Standing on a course, or met an inconsistent state */
921
		pseg_start += nblocks;
922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939
		if (pseg_start < seg_end)
			continue;
		goto feed_segment;

 strayed:
		/* Off the trail */
		if (!scan_newer)
			/*
			 * This can happen if a checkpoint was written without
			 * barriers, or as a result of an I/O failure.
			 */
			goto failed;

 feed_segment:
		/* Looking to the next full segment */
		if (empty_seg++)
			goto super_root_found; /* found a valid super root */

940 941
		ret = nilfs_segment_list_add(&segments, segnum);
		if (unlikely(ret))
942 943 944 945 946 947 948 949 950 951
			goto failed;

		seg_seq++;
		segnum = nextnum;
		nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
		pseg_start = seg_start;
	}

 super_root_found:
	/* Updating pointers relating to the latest checkpoint */
952
	brelse(bh_sum);
953
	list_splice_tail(&segments, &ri->ri_used_segments);
954 955 956 957 958 959
	nilfs->ns_last_pseg = sr_pseg_start;
	nilfs->ns_last_seq = nilfs->ns_seg_seq;
	nilfs->ns_last_cno = ri->ri_cno;
	return 0;

 failed:
960
	brelse(bh_sum);
961 962 963
	nilfs_dispose_segment_list(&segments);
	return (ret < 0) ? ret : nilfs_warn_segment_error(ret);
}