2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_error.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_alloc.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_log_priv.h"
42 #include "xfs_buf_item.h"
43 #include "xfs_log_recover.h"
44 #include "xfs_extfree_item.h"
45 #include "xfs_trans_priv.h"
46 #include "xfs_quota.h"
48 #include "xfs_utils.h"
50 STATIC
int xlog_find_zeroed(xlog_t
*, xfs_daddr_t
*);
51 STATIC
int xlog_clear_stale_blocks(xlog_t
*, xfs_lsn_t
);
52 STATIC
void xlog_recover_insert_item_backq(xlog_recover_item_t
**q
,
53 xlog_recover_item_t
*item
);
55 STATIC
void xlog_recover_check_summary(xlog_t
*);
57 #define xlog_recover_check_summary(log)
62 * Sector aligned buffer routines for buffer create/read/write/access
65 #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
66 ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
67 ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
68 #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
75 if (nbblks
<= 0 || nbblks
> log
->l_logBBsize
) {
76 xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks
);
77 XFS_ERROR_REPORT("xlog_get_bp(1)",
78 XFS_ERRLEVEL_HIGH
, log
->l_mp
);
82 if (log
->l_sectbb_log
) {
84 nbblks
+= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, 1);
85 nbblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, nbblks
);
87 return xfs_buf_get_noaddr(BBTOB(nbblks
), log
->l_mp
->m_logdev_targp
);
106 if (!log
->l_sectbb_log
)
107 return XFS_BUF_PTR(bp
);
109 ptr
= XFS_BUF_PTR(bp
) + BBTOB((int)blk_no
& log
->l_sectbb_mask
);
110 ASSERT(XFS_BUF_SIZE(bp
) >=
111 BBTOB(nbblks
+ (blk_no
& log
->l_sectbb_mask
)));
117 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
128 if (nbblks
<= 0 || nbblks
> log
->l_logBBsize
) {
129 xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks
);
130 XFS_ERROR_REPORT("xlog_bread(1)",
131 XFS_ERRLEVEL_HIGH
, log
->l_mp
);
135 if (log
->l_sectbb_log
) {
136 blk_no
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, blk_no
);
137 nbblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, nbblks
);
141 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
144 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
147 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
148 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
150 xfsbdstrat(log
->l_mp
, bp
);
151 error
= xfs_iowait(bp
);
153 xfs_ioerror_alert("xlog_bread", log
->l_mp
,
154 bp
, XFS_BUF_ADDR(bp
));
168 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
172 *offset
= xlog_align(log
, blk_no
, nbblks
, bp
);
177 * Write out the buffer at the given block for the given number of blocks.
178 * The buffer is kept locked across the write and is returned locked.
179 * This can only be used for synchronous log writes.
190 if (nbblks
<= 0 || nbblks
> log
->l_logBBsize
) {
191 xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks
);
192 XFS_ERROR_REPORT("xlog_bwrite(1)",
193 XFS_ERRLEVEL_HIGH
, log
->l_mp
);
197 if (log
->l_sectbb_log
) {
198 blk_no
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, blk_no
);
199 nbblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, nbblks
);
203 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
205 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
206 XFS_BUF_ZEROFLAGS(bp
);
209 XFS_BUF_PSEMA(bp
, PRIBIO
);
210 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
211 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
213 if ((error
= xfs_bwrite(log
->l_mp
, bp
)))
214 xfs_ioerror_alert("xlog_bwrite", log
->l_mp
,
215 bp
, XFS_BUF_ADDR(bp
));
221 * dump debug superblock and log record information
224 xlog_header_check_dump(
226 xlog_rec_header_t
*head
)
228 cmn_err(CE_DEBUG
, "%s: SB : uuid = %pU, fmt = %d\n",
229 __func__
, &mp
->m_sb
.sb_uuid
, XLOG_FMT
);
230 cmn_err(CE_DEBUG
, " log : uuid = %pU, fmt = %d\n",
231 &head
->h_fs_uuid
, be32_to_cpu(head
->h_fmt
));
234 #define xlog_header_check_dump(mp, head)
238 * check log record header for recovery
241 xlog_header_check_recover(
243 xlog_rec_header_t
*head
)
245 ASSERT(be32_to_cpu(head
->h_magicno
) == XLOG_HEADER_MAGIC_NUM
);
248 * IRIX doesn't write the h_fmt field and leaves it zeroed
249 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
250 * a dirty log created in IRIX.
252 if (unlikely(be32_to_cpu(head
->h_fmt
) != XLOG_FMT
)) {
254 "XFS: dirty log written in incompatible format - can't recover");
255 xlog_header_check_dump(mp
, head
);
256 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
257 XFS_ERRLEVEL_HIGH
, mp
);
258 return XFS_ERROR(EFSCORRUPTED
);
259 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
261 "XFS: dirty log entry has mismatched uuid - can't recover");
262 xlog_header_check_dump(mp
, head
);
263 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
264 XFS_ERRLEVEL_HIGH
, mp
);
265 return XFS_ERROR(EFSCORRUPTED
);
271 * read the head block of the log and check the header
274 xlog_header_check_mount(
276 xlog_rec_header_t
*head
)
278 ASSERT(be32_to_cpu(head
->h_magicno
) == XLOG_HEADER_MAGIC_NUM
);
280 if (uuid_is_nil(&head
->h_fs_uuid
)) {
282 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
283 * h_fs_uuid is nil, we assume this log was last mounted
284 * by IRIX and continue.
286 xlog_warn("XFS: nil uuid in log - IRIX style log");
287 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
288 xlog_warn("XFS: log has mismatched uuid - can't recover");
289 xlog_header_check_dump(mp
, head
);
290 XFS_ERROR_REPORT("xlog_header_check_mount",
291 XFS_ERRLEVEL_HIGH
, mp
);
292 return XFS_ERROR(EFSCORRUPTED
);
301 if (XFS_BUF_GETERROR(bp
)) {
303 * We're not going to bother about retrying
304 * this during recovery. One strike!
306 xfs_ioerror_alert("xlog_recover_iodone",
307 bp
->b_mount
, bp
, XFS_BUF_ADDR(bp
));
308 xfs_force_shutdown(bp
->b_mount
, SHUTDOWN_META_IO_ERROR
);
311 XFS_BUF_CLR_IODONE_FUNC(bp
);
316 * This routine finds (to an approximation) the first block in the physical
317 * log which contains the given cycle. It uses a binary search algorithm.
318 * Note that the algorithm can not be perfect because the disk will not
319 * necessarily be perfect.
322 xlog_find_cycle_start(
325 xfs_daddr_t first_blk
,
326 xfs_daddr_t
*last_blk
,
334 mid_blk
= BLK_AVG(first_blk
, *last_blk
);
335 while (mid_blk
!= first_blk
&& mid_blk
!= *last_blk
) {
336 error
= xlog_bread(log
, mid_blk
, 1, bp
, &offset
);
339 mid_cycle
= xlog_get_cycle(offset
);
340 if (mid_cycle
== cycle
) {
342 /* last_half_cycle == mid_cycle */
345 /* first_half_cycle == mid_cycle */
347 mid_blk
= BLK_AVG(first_blk
, *last_blk
);
349 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == *last_blk
) ||
350 (mid_blk
== *last_blk
&& mid_blk
-1 == first_blk
));
356 * Check that the range of blocks does not contain the cycle number
357 * given. The scan needs to occur from front to back and the ptr into the
358 * region must be updated since a later routine will need to perform another
359 * test. If the region is completely good, we end up returning the same
362 * Set blkno to -1 if we encounter no errors. This is an invalid block number
363 * since we don't ever expect logs to get this large.
366 xlog_find_verify_cycle(
368 xfs_daddr_t start_blk
,
370 uint stop_on_cycle_no
,
371 xfs_daddr_t
*new_blk
)
377 xfs_caddr_t buf
= NULL
;
380 bufblks
= 1 << ffs(nbblks
);
382 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
383 /* can't get enough memory to do everything in one big buffer */
385 if (bufblks
<= log
->l_sectbb_log
)
389 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
392 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
394 error
= xlog_bread(log
, i
, bcount
, bp
, &buf
);
398 for (j
= 0; j
< bcount
; j
++) {
399 cycle
= xlog_get_cycle(buf
);
400 if (cycle
== stop_on_cycle_no
) {
417 * Potentially backup over partial log record write.
419 * In the typical case, last_blk is the number of the block directly after
420 * a good log record. Therefore, we subtract one to get the block number
421 * of the last block in the given buffer. extra_bblks contains the number
422 * of blocks we would have read on a previous read. This happens when the
423 * last log record is split over the end of the physical log.
425 * extra_bblks is the number of blocks potentially verified on a previous
426 * call to this routine.
429 xlog_find_verify_log_record(
431 xfs_daddr_t start_blk
,
432 xfs_daddr_t
*last_blk
,
437 xfs_caddr_t offset
= NULL
;
438 xlog_rec_header_t
*head
= NULL
;
441 int num_blks
= *last_blk
- start_blk
;
444 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
446 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
447 if (!(bp
= xlog_get_bp(log
, 1)))
451 error
= xlog_bread(log
, start_blk
, num_blks
, bp
, &offset
);
454 offset
+= ((num_blks
- 1) << BBSHIFT
);
457 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
459 /* valid log record not found */
461 "XFS: Log inconsistent (didn't find previous header)");
463 error
= XFS_ERROR(EIO
);
468 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
473 head
= (xlog_rec_header_t
*)offset
;
475 if (XLOG_HEADER_MAGIC_NUM
== be32_to_cpu(head
->h_magicno
))
483 * We hit the beginning of the physical log & still no header. Return
484 * to caller. If caller can handle a return of -1, then this routine
485 * will be called again for the end of the physical log.
493 * We have the final block of the good log (the first block
494 * of the log record _before_ the head. So we check the uuid.
496 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
500 * We may have found a log record header before we expected one.
501 * last_blk will be the 1st block # with a given cycle #. We may end
502 * up reading an entire log record. In this case, we don't want to
503 * reset last_blk. Only when last_blk points in the middle of a log
504 * record do we update last_blk.
506 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
507 uint h_size
= be32_to_cpu(head
->h_size
);
509 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
510 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
516 if (*last_blk
- i
+ extra_bblks
!=
517 BTOBB(be32_to_cpu(head
->h_len
)) + xhdrs
)
526 * Head is defined to be the point of the log where the next log write
527 * write could go. This means that incomplete LR writes at the end are
528 * eliminated when calculating the head. We aren't guaranteed that previous
529 * LR have complete transactions. We only know that a cycle number of
530 * current cycle number -1 won't be present in the log if we start writing
531 * from our current block number.
533 * last_blk contains the block number of the first block with a given
536 * Return: zero if normal, non-zero if error.
541 xfs_daddr_t
*return_head_blk
)
545 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
547 uint first_half_cycle
, last_half_cycle
;
549 int error
, log_bbnum
= log
->l_logBBsize
;
551 /* Is the end of the log device zeroed? */
552 if ((error
= xlog_find_zeroed(log
, &first_blk
)) == -1) {
553 *return_head_blk
= first_blk
;
555 /* Is the whole lot zeroed? */
557 /* Linux XFS shouldn't generate totally zeroed logs -
558 * mkfs etc write a dummy unmount record to a fresh
559 * log so we can store the uuid in there
561 xlog_warn("XFS: totally zeroed log");
566 xlog_warn("XFS: empty log check failed");
570 first_blk
= 0; /* get cycle # of 1st block */
571 bp
= xlog_get_bp(log
, 1);
575 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
579 first_half_cycle
= xlog_get_cycle(offset
);
581 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
582 error
= xlog_bread(log
, last_blk
, 1, bp
, &offset
);
586 last_half_cycle
= xlog_get_cycle(offset
);
587 ASSERT(last_half_cycle
!= 0);
590 * If the 1st half cycle number is equal to the last half cycle number,
591 * then the entire log is stamped with the same cycle number. In this
592 * case, head_blk can't be set to zero (which makes sense). The below
593 * math doesn't work out properly with head_blk equal to zero. Instead,
594 * we set it to log_bbnum which is an invalid block number, but this
595 * value makes the math correct. If head_blk doesn't changed through
596 * all the tests below, *head_blk is set to zero at the very end rather
597 * than log_bbnum. In a sense, log_bbnum and zero are the same block
598 * in a circular file.
600 if (first_half_cycle
== last_half_cycle
) {
602 * In this case we believe that the entire log should have
603 * cycle number last_half_cycle. We need to scan backwards
604 * from the end verifying that there are no holes still
605 * containing last_half_cycle - 1. If we find such a hole,
606 * then the start of that hole will be the new head. The
607 * simple case looks like
608 * x | x ... | x - 1 | x
609 * Another case that fits this picture would be
610 * x | x + 1 | x ... | x
611 * In this case the head really is somewhere at the end of the
612 * log, as one of the latest writes at the beginning was
615 * x | x + 1 | x ... | x - 1 | x
616 * This is really the combination of the above two cases, and
617 * the head has to end up at the start of the x-1 hole at the
620 * In the 256k log case, we will read from the beginning to the
621 * end of the log and search for cycle numbers equal to x-1.
622 * We don't worry about the x+1 blocks that we encounter,
623 * because we know that they cannot be the head since the log
626 head_blk
= log_bbnum
;
627 stop_on_cycle
= last_half_cycle
- 1;
630 * In this case we want to find the first block with cycle
631 * number matching last_half_cycle. We expect the log to be
634 * The first block with cycle number x (last_half_cycle) will
635 * be where the new head belongs. First we do a binary search
636 * for the first occurrence of last_half_cycle. The binary
637 * search may not be totally accurate, so then we scan back
638 * from there looking for occurrences of last_half_cycle before
639 * us. If that backwards scan wraps around the beginning of
640 * the log, then we look for occurrences of last_half_cycle - 1
641 * at the end of the log. The cases we're looking for look
643 * x + 1 ... | x | x + 1 | x ...
644 * ^ binary search stopped here
646 * x + 1 ... | x ... | x - 1 | x
647 * <---------> less than scan distance
649 stop_on_cycle
= last_half_cycle
;
650 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
651 &head_blk
, last_half_cycle
)))
656 * Now validate the answer. Scan back some number of maximum possible
657 * blocks and make sure each one has the expected cycle number. The
658 * maximum is determined by the total possible amount of buffering
659 * in the in-core log. The following number can be made tighter if
660 * we actually look at the block size of the filesystem.
662 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
663 if (head_blk
>= num_scan_bblks
) {
665 * We are guaranteed that the entire check can be performed
668 start_blk
= head_blk
- num_scan_bblks
;
669 if ((error
= xlog_find_verify_cycle(log
,
670 start_blk
, num_scan_bblks
,
671 stop_on_cycle
, &new_blk
)))
675 } else { /* need to read 2 parts of log */
677 * We are going to scan backwards in the log in two parts.
678 * First we scan the physical end of the log. In this part
679 * of the log, we are looking for blocks with cycle number
680 * last_half_cycle - 1.
681 * If we find one, then we know that the log starts there, as
682 * we've found a hole that didn't get written in going around
683 * the end of the physical log. The simple case for this is
684 * x + 1 ... | x ... | x - 1 | x
685 * <---------> less than scan distance
686 * If all of the blocks at the end of the log have cycle number
687 * last_half_cycle, then we check the blocks at the start of
688 * the log looking for occurrences of last_half_cycle. If we
689 * find one, then our current estimate for the location of the
690 * first occurrence of last_half_cycle is wrong and we move
691 * back to the hole we've found. This case looks like
692 * x + 1 ... | x | x + 1 | x ...
693 * ^ binary search stopped here
694 * Another case we need to handle that only occurs in 256k
696 * x + 1 ... | x ... | x+1 | x ...
697 * ^ binary search stops here
698 * In a 256k log, the scan at the end of the log will see the
699 * x + 1 blocks. We need to skip past those since that is
700 * certainly not the head of the log. By searching for
701 * last_half_cycle-1 we accomplish that.
703 start_blk
= log_bbnum
- num_scan_bblks
+ head_blk
;
704 ASSERT(head_blk
<= INT_MAX
&&
705 (xfs_daddr_t
) num_scan_bblks
- head_blk
>= 0);
706 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
707 num_scan_bblks
- (int)head_blk
,
708 (stop_on_cycle
- 1), &new_blk
)))
716 * Scan beginning of log now. The last part of the physical
717 * log is good. This scan needs to verify that it doesn't find
718 * the last_half_cycle.
721 ASSERT(head_blk
<= INT_MAX
);
722 if ((error
= xlog_find_verify_cycle(log
,
723 start_blk
, (int)head_blk
,
724 stop_on_cycle
, &new_blk
)))
732 * Now we need to make sure head_blk is not pointing to a block in
733 * the middle of a log record.
735 num_scan_bblks
= XLOG_REC_SHIFT(log
);
736 if (head_blk
>= num_scan_bblks
) {
737 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
739 /* start ptr at last block ptr before head_blk */
740 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
741 &head_blk
, 0)) == -1) {
742 error
= XFS_ERROR(EIO
);
748 ASSERT(head_blk
<= INT_MAX
);
749 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
750 &head_blk
, 0)) == -1) {
751 /* We hit the beginning of the log during our search */
752 start_blk
= log_bbnum
- num_scan_bblks
+ head_blk
;
754 ASSERT(start_blk
<= INT_MAX
&&
755 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
756 ASSERT(head_blk
<= INT_MAX
);
757 if ((error
= xlog_find_verify_log_record(log
,
759 (int)head_blk
)) == -1) {
760 error
= XFS_ERROR(EIO
);
764 if (new_blk
!= log_bbnum
)
771 if (head_blk
== log_bbnum
)
772 *return_head_blk
= 0;
774 *return_head_blk
= head_blk
;
776 * When returning here, we have a good block number. Bad block
777 * means that during a previous crash, we didn't have a clean break
778 * from cycle number N to cycle number N-1. In this case, we need
779 * to find the first block with cycle number N-1.
787 xlog_warn("XFS: failed to find log head");
792 * Find the sync block number or the tail of the log.
794 * This will be the block number of the last record to have its
795 * associated buffers synced to disk. Every log record header has
796 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
797 * to get a sync block number. The only concern is to figure out which
798 * log record header to believe.
800 * The following algorithm uses the log record header with the largest
801 * lsn. The entire log record does not need to be valid. We only care
802 * that the header is valid.
804 * We could speed up search by using current head_blk buffer, but it is not
810 xfs_daddr_t
*head_blk
,
811 xfs_daddr_t
*tail_blk
)
813 xlog_rec_header_t
*rhead
;
814 xlog_op_header_t
*op_head
;
815 xfs_caddr_t offset
= NULL
;
818 xfs_daddr_t umount_data_blk
;
819 xfs_daddr_t after_umount_blk
;
826 * Find previous log record
828 if ((error
= xlog_find_head(log
, head_blk
)))
831 bp
= xlog_get_bp(log
, 1);
834 if (*head_blk
== 0) { /* special case */
835 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
839 if (xlog_get_cycle(offset
) == 0) {
841 /* leave all other log inited values alone */
847 * Search backwards looking for log record header block
849 ASSERT(*head_blk
< INT_MAX
);
850 for (i
= (int)(*head_blk
) - 1; i
>= 0; i
--) {
851 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
855 if (XLOG_HEADER_MAGIC_NUM
== be32_to_cpu(*(__be32
*)offset
)) {
861 * If we haven't found the log record header block, start looking
862 * again from the end of the physical log. XXXmiken: There should be
863 * a check here to make sure we didn't search more than N blocks in
867 for (i
= log
->l_logBBsize
- 1; i
>= (int)(*head_blk
); i
--) {
868 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
872 if (XLOG_HEADER_MAGIC_NUM
==
873 be32_to_cpu(*(__be32
*)offset
)) {
880 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
882 return XFS_ERROR(EIO
);
885 /* find blk_no of tail of log */
886 rhead
= (xlog_rec_header_t
*)offset
;
887 *tail_blk
= BLOCK_LSN(be64_to_cpu(rhead
->h_tail_lsn
));
890 * Reset log values according to the state of the log when we
891 * crashed. In the case where head_blk == 0, we bump curr_cycle
892 * one because the next write starts a new cycle rather than
893 * continuing the cycle of the last good log record. At this
894 * point we have guaranteed that all partial log records have been
895 * accounted for. Therefore, we know that the last good log record
896 * written was complete and ended exactly on the end boundary
897 * of the physical log.
899 log
->l_prev_block
= i
;
900 log
->l_curr_block
= (int)*head_blk
;
901 log
->l_curr_cycle
= be32_to_cpu(rhead
->h_cycle
);
904 log
->l_tail_lsn
= be64_to_cpu(rhead
->h_tail_lsn
);
905 log
->l_last_sync_lsn
= be64_to_cpu(rhead
->h_lsn
);
906 log
->l_grant_reserve_cycle
= log
->l_curr_cycle
;
907 log
->l_grant_reserve_bytes
= BBTOB(log
->l_curr_block
);
908 log
->l_grant_write_cycle
= log
->l_curr_cycle
;
909 log
->l_grant_write_bytes
= BBTOB(log
->l_curr_block
);
912 * Look for unmount record. If we find it, then we know there
913 * was a clean unmount. Since 'i' could be the last block in
914 * the physical log, we convert to a log block before comparing
917 * Save the current tail lsn to use to pass to
918 * xlog_clear_stale_blocks() below. We won't want to clear the
919 * unmount record if there is one, so we pass the lsn of the
920 * unmount record rather than the block after it.
922 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
923 int h_size
= be32_to_cpu(rhead
->h_size
);
924 int h_version
= be32_to_cpu(rhead
->h_version
);
926 if ((h_version
& XLOG_VERSION_2
) &&
927 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
928 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
929 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
937 after_umount_blk
= (i
+ hblks
+ (int)
938 BTOBB(be32_to_cpu(rhead
->h_len
))) % log
->l_logBBsize
;
939 tail_lsn
= log
->l_tail_lsn
;
940 if (*head_blk
== after_umount_blk
&&
941 be32_to_cpu(rhead
->h_num_logops
) == 1) {
942 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
943 error
= xlog_bread(log
, umount_data_blk
, 1, bp
, &offset
);
947 op_head
= (xlog_op_header_t
*)offset
;
948 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
950 * Set tail and last sync so that newly written
951 * log records will point recovery to after the
952 * current unmount record.
955 xlog_assign_lsn(log
->l_curr_cycle
,
957 log
->l_last_sync_lsn
=
958 xlog_assign_lsn(log
->l_curr_cycle
,
960 *tail_blk
= after_umount_blk
;
963 * Note that the unmount was clean. If the unmount
964 * was not clean, we need to know this to rebuild the
965 * superblock counters from the perag headers if we
966 * have a filesystem using non-persistent counters.
968 log
->l_mp
->m_flags
|= XFS_MOUNT_WAS_CLEAN
;
973 * Make sure that there are no blocks in front of the head
974 * with the same cycle number as the head. This can happen
975 * because we allow multiple outstanding log writes concurrently,
976 * and the later writes might make it out before earlier ones.
978 * We use the lsn from before modifying it so that we'll never
979 * overwrite the unmount record after a clean unmount.
981 * Do this only if we are going to recover the filesystem
983 * NOTE: This used to say "if (!readonly)"
984 * However on Linux, we can & do recover a read-only filesystem.
985 * We only skip recovery if NORECOVERY is specified on mount,
986 * in which case we would not be here.
988 * But... if the -device- itself is readonly, just skip this.
989 * We can't recover this device anyway, so it won't matter.
991 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
)) {
992 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
1000 xlog_warn("XFS: failed to locate log tail");
1005 * Is the log zeroed at all?
1007 * The last binary search should be changed to perform an X block read
1008 * once X becomes small enough. You can then search linearly through
1009 * the X blocks. This will cut down on the number of reads we need to do.
1011 * If the log is partially zeroed, this routine will pass back the blkno
1012 * of the first block with cycle number 0. It won't have a complete LR
1016 * 0 => the log is completely written to
1017 * -1 => use *blk_no as the first block of the log
1018 * >0 => error has occurred
1023 xfs_daddr_t
*blk_no
)
1027 uint first_cycle
, last_cycle
;
1028 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1029 xfs_daddr_t num_scan_bblks
;
1030 int error
, log_bbnum
= log
->l_logBBsize
;
1034 /* check totally zeroed log */
1035 bp
= xlog_get_bp(log
, 1);
1038 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1042 first_cycle
= xlog_get_cycle(offset
);
1043 if (first_cycle
== 0) { /* completely zeroed log */
1049 /* check partially zeroed log */
1050 error
= xlog_bread(log
, log_bbnum
-1, 1, bp
, &offset
);
1054 last_cycle
= xlog_get_cycle(offset
);
1055 if (last_cycle
!= 0) { /* log completely written to */
1058 } else if (first_cycle
!= 1) {
1060 * If the cycle of the last block is zero, the cycle of
1061 * the first block must be 1. If it's not, maybe we're
1062 * not looking at a log... Bail out.
1064 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1065 return XFS_ERROR(EINVAL
);
1068 /* we have a partially zeroed log */
1069 last_blk
= log_bbnum
-1;
1070 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1074 * Validate the answer. Because there is no way to guarantee that
1075 * the entire log is made up of log records which are the same size,
1076 * we scan over the defined maximum blocks. At this point, the maximum
1077 * is not chosen to mean anything special. XXXmiken
1079 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1080 ASSERT(num_scan_bblks
<= INT_MAX
);
1082 if (last_blk
< num_scan_bblks
)
1083 num_scan_bblks
= last_blk
;
1084 start_blk
= last_blk
- num_scan_bblks
;
1087 * We search for any instances of cycle number 0 that occur before
1088 * our current estimate of the head. What we're trying to detect is
1089 * 1 ... | 0 | 1 | 0...
1090 * ^ binary search ends here
1092 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1093 (int)num_scan_bblks
, 0, &new_blk
)))
1099 * Potentially backup over partial log record write. We don't need
1100 * to search the end of the log because we know it is zero.
1102 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
1103 &last_blk
, 0)) == -1) {
1104 error
= XFS_ERROR(EIO
);
1118 * These are simple subroutines used by xlog_clear_stale_blocks() below
1119 * to initialize a buffer full of empty log record headers and write
1120 * them into the log.
1131 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1133 memset(buf
, 0, BBSIZE
);
1134 recp
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1135 recp
->h_cycle
= cpu_to_be32(cycle
);
1136 recp
->h_version
= cpu_to_be32(
1137 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1138 recp
->h_lsn
= cpu_to_be64(xlog_assign_lsn(cycle
, block
));
1139 recp
->h_tail_lsn
= cpu_to_be64(xlog_assign_lsn(tail_cycle
, tail_block
));
1140 recp
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1141 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1145 xlog_write_log_records(
1156 int sectbb
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, 1);
1157 int end_block
= start_block
+ blocks
;
1162 bufblks
= 1 << ffs(blocks
);
1163 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1165 if (bufblks
<= log
->l_sectbb_log
)
1169 /* We may need to do a read at the start to fill in part of
1170 * the buffer in the starting sector not covered by the first
1173 balign
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, start_block
);
1174 if (balign
!= start_block
) {
1175 error
= xlog_bread_noalign(log
, start_block
, 1, bp
);
1179 j
= start_block
- balign
;
1182 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1183 int bcount
, endcount
;
1185 bcount
= min(bufblks
, end_block
- start_block
);
1186 endcount
= bcount
- j
;
1188 /* We may need to do a read at the end to fill in part of
1189 * the buffer in the final sector not covered by the write.
1190 * If this is the same sector as the above read, skip it.
1192 ealign
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, end_block
);
1193 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1194 offset
= XFS_BUF_PTR(bp
);
1195 balign
= BBTOB(ealign
- start_block
);
1196 error
= XFS_BUF_SET_PTR(bp
, offset
+ balign
,
1201 error
= xlog_bread_noalign(log
, ealign
, sectbb
, bp
);
1205 error
= XFS_BUF_SET_PTR(bp
, offset
, bufblks
);
1210 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1211 for (; j
< endcount
; j
++) {
1212 xlog_add_record(log
, offset
, cycle
, i
+j
,
1213 tail_cycle
, tail_block
);
1216 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1219 start_block
+= endcount
;
1229 * This routine is called to blow away any incomplete log writes out
1230 * in front of the log head. We do this so that we won't become confused
1231 * if we come up, write only a little bit more, and then crash again.
1232 * If we leave the partial log records out there, this situation could
1233 * cause us to think those partial writes are valid blocks since they
1234 * have the current cycle number. We get rid of them by overwriting them
1235 * with empty log records with the old cycle number rather than the
1238 * The tail lsn is passed in rather than taken from
1239 * the log so that we will not write over the unmount record after a
1240 * clean unmount in a 512 block log. Doing so would leave the log without
1241 * any valid log records in it until a new one was written. If we crashed
1242 * during that time we would not be able to recover.
1245 xlog_clear_stale_blocks(
1249 int tail_cycle
, head_cycle
;
1250 int tail_block
, head_block
;
1251 int tail_distance
, max_distance
;
1255 tail_cycle
= CYCLE_LSN(tail_lsn
);
1256 tail_block
= BLOCK_LSN(tail_lsn
);
1257 head_cycle
= log
->l_curr_cycle
;
1258 head_block
= log
->l_curr_block
;
1261 * Figure out the distance between the new head of the log
1262 * and the tail. We want to write over any blocks beyond the
1263 * head that we may have written just before the crash, but
1264 * we don't want to overwrite the tail of the log.
1266 if (head_cycle
== tail_cycle
) {
1268 * The tail is behind the head in the physical log,
1269 * so the distance from the head to the tail is the
1270 * distance from the head to the end of the log plus
1271 * the distance from the beginning of the log to the
1274 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1275 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1276 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1277 return XFS_ERROR(EFSCORRUPTED
);
1279 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1282 * The head is behind the tail in the physical log,
1283 * so the distance from the head to the tail is just
1284 * the tail block minus the head block.
1286 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1287 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1288 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1289 return XFS_ERROR(EFSCORRUPTED
);
1291 tail_distance
= tail_block
- head_block
;
1295 * If the head is right up against the tail, we can't clear
1298 if (tail_distance
<= 0) {
1299 ASSERT(tail_distance
== 0);
1303 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1305 * Take the smaller of the maximum amount of outstanding I/O
1306 * we could have and the distance to the tail to clear out.
1307 * We take the smaller so that we don't overwrite the tail and
1308 * we don't waste all day writing from the head to the tail
1311 max_distance
= MIN(max_distance
, tail_distance
);
1313 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1315 * We can stomp all the blocks we need to without
1316 * wrapping around the end of the log. Just do it
1317 * in a single write. Use the cycle number of the
1318 * current cycle minus one so that the log will look like:
1321 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1322 head_block
, max_distance
, tail_cycle
,
1328 * We need to wrap around the end of the physical log in
1329 * order to clear all the blocks. Do it in two separate
1330 * I/Os. The first write should be from the head to the
1331 * end of the physical log, and it should use the current
1332 * cycle number minus one just like above.
1334 distance
= log
->l_logBBsize
- head_block
;
1335 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1336 head_block
, distance
, tail_cycle
,
1343 * Now write the blocks at the start of the physical log.
1344 * This writes the remainder of the blocks we want to clear.
1345 * It uses the current cycle number since we're now on the
1346 * same cycle as the head so that we get:
1347 * n ... n ... | n - 1 ...
1348 * ^^^^^ blocks we're writing
1350 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1351 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1352 tail_cycle
, tail_block
);
1360 /******************************************************************************
1362 * Log recover routines
1364 ******************************************************************************
1367 STATIC xlog_recover_t
*
1368 xlog_recover_find_tid(
1372 xlog_recover_t
*p
= q
;
1375 if (p
->r_log_tid
== tid
)
1383 xlog_recover_put_hashq(
1385 xlog_recover_t
*trans
)
1392 xlog_recover_add_item(
1393 xlog_recover_item_t
**itemq
)
1395 xlog_recover_item_t
*item
;
1397 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
1398 xlog_recover_insert_item_backq(itemq
, item
);
1402 xlog_recover_add_to_cont_trans(
1403 xlog_recover_t
*trans
,
1407 xlog_recover_item_t
*item
;
1408 xfs_caddr_t ptr
, old_ptr
;
1411 item
= trans
->r_itemq
;
1413 /* finish copying rest of trans header */
1414 xlog_recover_add_item(&trans
->r_itemq
);
1415 ptr
= (xfs_caddr_t
) &trans
->r_theader
+
1416 sizeof(xfs_trans_header_t
) - len
;
1417 memcpy(ptr
, dp
, len
); /* d, s, l */
1420 item
= item
->ri_prev
;
1422 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
1423 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
1425 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, 0u);
1426 memcpy(&ptr
[old_len
], dp
, len
); /* d, s, l */
1427 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
1428 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
1433 * The next region to add is the start of a new region. It could be
1434 * a whole region or it could be the first part of a new region. Because
1435 * of this, the assumption here is that the type and size fields of all
1436 * format structures fit into the first 32 bits of the structure.
1438 * This works because all regions must be 32 bit aligned. Therefore, we
1439 * either have both fields or we have neither field. In the case we have
1440 * neither field, the data part of the region is zero length. We only have
1441 * a log_op_header and can throw away the header since a new one will appear
1442 * later. If we have at least 4 bytes, then we can determine how many regions
1443 * will appear in the current log item.
1446 xlog_recover_add_to_trans(
1447 xlog_recover_t
*trans
,
1451 xfs_inode_log_format_t
*in_f
; /* any will do */
1452 xlog_recover_item_t
*item
;
1457 item
= trans
->r_itemq
;
1459 /* we need to catch log corruptions here */
1460 if (*(uint
*)dp
!= XFS_TRANS_HEADER_MAGIC
) {
1461 xlog_warn("XFS: xlog_recover_add_to_trans: "
1462 "bad header magic number");
1464 return XFS_ERROR(EIO
);
1466 if (len
== sizeof(xfs_trans_header_t
))
1467 xlog_recover_add_item(&trans
->r_itemq
);
1468 memcpy(&trans
->r_theader
, dp
, len
); /* d, s, l */
1472 ptr
= kmem_alloc(len
, KM_SLEEP
);
1473 memcpy(ptr
, dp
, len
);
1474 in_f
= (xfs_inode_log_format_t
*)ptr
;
1476 if (item
->ri_prev
->ri_total
!= 0 &&
1477 item
->ri_prev
->ri_total
== item
->ri_prev
->ri_cnt
) {
1478 xlog_recover_add_item(&trans
->r_itemq
);
1480 item
= trans
->r_itemq
;
1481 item
= item
->ri_prev
;
1483 if (item
->ri_total
== 0) { /* first region to be added */
1484 if (in_f
->ilf_size
== 0 ||
1485 in_f
->ilf_size
> XLOG_MAX_REGIONS_IN_ITEM
) {
1487 "XFS: bad number of regions (%d) in inode log format",
1490 return XFS_ERROR(EIO
);
1493 item
->ri_total
= in_f
->ilf_size
;
1495 kmem_zalloc(item
->ri_total
* sizeof(xfs_log_iovec_t
),
1498 ASSERT(item
->ri_total
> item
->ri_cnt
);
1499 /* Description region is ri_buf[0] */
1500 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
1501 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
1507 xlog_recover_new_tid(
1512 xlog_recover_t
*trans
;
1514 trans
= kmem_zalloc(sizeof(xlog_recover_t
), KM_SLEEP
);
1515 trans
->r_log_tid
= tid
;
1517 xlog_recover_put_hashq(q
, trans
);
1521 xlog_recover_unlink_tid(
1523 xlog_recover_t
*trans
)
1528 ASSERT(trans
!= NULL
);
1534 if (tp
->r_next
== trans
) {
1542 "XFS: xlog_recover_unlink_tid: trans not found");
1544 return XFS_ERROR(EIO
);
1546 tp
->r_next
= tp
->r_next
->r_next
;
1552 xlog_recover_insert_item_backq(
1553 xlog_recover_item_t
**q
,
1554 xlog_recover_item_t
*item
)
1557 item
->ri_prev
= item
->ri_next
= item
;
1561 item
->ri_prev
= (*q
)->ri_prev
;
1562 (*q
)->ri_prev
= item
;
1563 item
->ri_prev
->ri_next
= item
;
1568 xlog_recover_insert_item_frontq(
1569 xlog_recover_item_t
**q
,
1570 xlog_recover_item_t
*item
)
1572 xlog_recover_insert_item_backq(q
, item
);
1577 xlog_recover_reorder_trans(
1578 xlog_recover_t
*trans
)
1580 xlog_recover_item_t
*first_item
, *itemq
, *itemq_next
;
1581 xfs_buf_log_format_t
*buf_f
;
1584 first_item
= itemq
= trans
->r_itemq
;
1585 trans
->r_itemq
= NULL
;
1587 itemq_next
= itemq
->ri_next
;
1588 buf_f
= (xfs_buf_log_format_t
*)itemq
->ri_buf
[0].i_addr
;
1590 switch (ITEM_TYPE(itemq
)) {
1592 flags
= buf_f
->blf_flags
;
1593 if (!(flags
& XFS_BLI_CANCEL
)) {
1594 xlog_recover_insert_item_frontq(&trans
->r_itemq
,
1600 case XFS_LI_QUOTAOFF
:
1603 xlog_recover_insert_item_backq(&trans
->r_itemq
, itemq
);
1607 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1609 return XFS_ERROR(EIO
);
1612 } while (first_item
!= itemq
);
1617 * Build up the table of buf cancel records so that we don't replay
1618 * cancelled data in the second pass. For buffer records that are
1619 * not cancel records, there is nothing to do here so we just return.
1621 * If we get a cancel record which is already in the table, this indicates
1622 * that the buffer was cancelled multiple times. In order to ensure
1623 * that during pass 2 we keep the record in the table until we reach its
1624 * last occurrence in the log, we keep a reference count in the cancel
1625 * record in the table to tell us how many times we expect to see this
1626 * record during the second pass.
1629 xlog_recover_do_buffer_pass1(
1631 xfs_buf_log_format_t
*buf_f
)
1633 xfs_buf_cancel_t
*bcp
;
1634 xfs_buf_cancel_t
*nextp
;
1635 xfs_buf_cancel_t
*prevp
;
1636 xfs_buf_cancel_t
**bucket
;
1637 xfs_daddr_t blkno
= 0;
1641 switch (buf_f
->blf_type
) {
1643 blkno
= buf_f
->blf_blkno
;
1644 len
= buf_f
->blf_len
;
1645 flags
= buf_f
->blf_flags
;
1650 * If this isn't a cancel buffer item, then just return.
1652 if (!(flags
& XFS_BLI_CANCEL
))
1656 * Insert an xfs_buf_cancel record into the hash table of
1657 * them. If there is already an identical record, bump
1658 * its reference count.
1660 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1661 XLOG_BC_TABLE_SIZE
];
1663 * If the hash bucket is empty then just insert a new record into
1666 if (*bucket
== NULL
) {
1667 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1669 bcp
->bc_blkno
= blkno
;
1671 bcp
->bc_refcount
= 1;
1672 bcp
->bc_next
= NULL
;
1678 * The hash bucket is not empty, so search for duplicates of our
1679 * record. If we find one them just bump its refcount. If not
1680 * then add us at the end of the list.
1684 while (nextp
!= NULL
) {
1685 if (nextp
->bc_blkno
== blkno
&& nextp
->bc_len
== len
) {
1686 nextp
->bc_refcount
++;
1690 nextp
= nextp
->bc_next
;
1692 ASSERT(prevp
!= NULL
);
1693 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1695 bcp
->bc_blkno
= blkno
;
1697 bcp
->bc_refcount
= 1;
1698 bcp
->bc_next
= NULL
;
1699 prevp
->bc_next
= bcp
;
1703 * Check to see whether the buffer being recovered has a corresponding
1704 * entry in the buffer cancel record table. If it does then return 1
1705 * so that it will be cancelled, otherwise return 0. If the buffer is
1706 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1707 * the refcount on the entry in the table and remove it from the table
1708 * if this is the last reference.
1710 * We remove the cancel record from the table when we encounter its
1711 * last occurrence in the log so that if the same buffer is re-used
1712 * again after its last cancellation we actually replay the changes
1713 * made at that point.
1716 xlog_check_buffer_cancelled(
1722 xfs_buf_cancel_t
*bcp
;
1723 xfs_buf_cancel_t
*prevp
;
1724 xfs_buf_cancel_t
**bucket
;
1726 if (log
->l_buf_cancel_table
== NULL
) {
1728 * There is nothing in the table built in pass one,
1729 * so this buffer must not be cancelled.
1731 ASSERT(!(flags
& XFS_BLI_CANCEL
));
1735 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1736 XLOG_BC_TABLE_SIZE
];
1740 * There is no corresponding entry in the table built
1741 * in pass one, so this buffer has not been cancelled.
1743 ASSERT(!(flags
& XFS_BLI_CANCEL
));
1748 * Search for an entry in the buffer cancel table that
1749 * matches our buffer.
1752 while (bcp
!= NULL
) {
1753 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
) {
1755 * We've go a match, so return 1 so that the
1756 * recovery of this buffer is cancelled.
1757 * If this buffer is actually a buffer cancel
1758 * log item, then decrement the refcount on the
1759 * one in the table and remove it if this is the
1762 if (flags
& XFS_BLI_CANCEL
) {
1764 if (bcp
->bc_refcount
== 0) {
1765 if (prevp
== NULL
) {
1766 *bucket
= bcp
->bc_next
;
1768 prevp
->bc_next
= bcp
->bc_next
;
1779 * We didn't find a corresponding entry in the table, so
1780 * return 0 so that the buffer is NOT cancelled.
1782 ASSERT(!(flags
& XFS_BLI_CANCEL
));
1787 xlog_recover_do_buffer_pass2(
1789 xfs_buf_log_format_t
*buf_f
)
1791 xfs_daddr_t blkno
= 0;
1795 switch (buf_f
->blf_type
) {
1797 blkno
= buf_f
->blf_blkno
;
1798 flags
= buf_f
->blf_flags
;
1799 len
= buf_f
->blf_len
;
1803 return xlog_check_buffer_cancelled(log
, blkno
, len
, flags
);
1807 * Perform recovery for a buffer full of inodes. In these buffers,
1808 * the only data which should be recovered is that which corresponds
1809 * to the di_next_unlinked pointers in the on disk inode structures.
1810 * The rest of the data for the inodes is always logged through the
1811 * inodes themselves rather than the inode buffer and is recovered
1812 * in xlog_recover_do_inode_trans().
1814 * The only time when buffers full of inodes are fully recovered is
1815 * when the buffer is full of newly allocated inodes. In this case
1816 * the buffer will not be marked as an inode buffer and so will be
1817 * sent to xlog_recover_do_reg_buffer() below during recovery.
1820 xlog_recover_do_inode_buffer(
1822 xlog_recover_item_t
*item
,
1824 xfs_buf_log_format_t
*buf_f
)
1832 int next_unlinked_offset
;
1834 xfs_agino_t
*logged_nextp
;
1835 xfs_agino_t
*buffer_nextp
;
1836 unsigned int *data_map
= NULL
;
1837 unsigned int map_size
= 0;
1839 switch (buf_f
->blf_type
) {
1841 data_map
= buf_f
->blf_data_map
;
1842 map_size
= buf_f
->blf_map_size
;
1846 * Set the variables corresponding to the current region to
1847 * 0 so that we'll initialize them on the first pass through
1855 inodes_per_buf
= XFS_BUF_COUNT(bp
) >> mp
->m_sb
.sb_inodelog
;
1856 for (i
= 0; i
< inodes_per_buf
; i
++) {
1857 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1858 offsetof(xfs_dinode_t
, di_next_unlinked
);
1860 while (next_unlinked_offset
>=
1861 (reg_buf_offset
+ reg_buf_bytes
)) {
1863 * The next di_next_unlinked field is beyond
1864 * the current logged region. Find the next
1865 * logged region that contains or is beyond
1866 * the current di_next_unlinked field.
1869 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1872 * If there are no more logged regions in the
1873 * buffer, then we're done.
1879 nbits
= xfs_contig_bits(data_map
, map_size
,
1882 reg_buf_offset
= bit
<< XFS_BLI_SHIFT
;
1883 reg_buf_bytes
= nbits
<< XFS_BLI_SHIFT
;
1888 * If the current logged region starts after the current
1889 * di_next_unlinked field, then move on to the next
1890 * di_next_unlinked field.
1892 if (next_unlinked_offset
< reg_buf_offset
) {
1896 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1897 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLI_CHUNK
) == 0);
1898 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <= XFS_BUF_COUNT(bp
));
1901 * The current logged region contains a copy of the
1902 * current di_next_unlinked field. Extract its value
1903 * and copy it to the buffer copy.
1905 logged_nextp
= (xfs_agino_t
*)
1906 ((char *)(item
->ri_buf
[item_index
].i_addr
) +
1907 (next_unlinked_offset
- reg_buf_offset
));
1908 if (unlikely(*logged_nextp
== 0)) {
1909 xfs_fs_cmn_err(CE_ALERT
, mp
,
1910 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1912 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1913 XFS_ERRLEVEL_LOW
, mp
);
1914 return XFS_ERROR(EFSCORRUPTED
);
1917 buffer_nextp
= (xfs_agino_t
*)xfs_buf_offset(bp
,
1918 next_unlinked_offset
);
1919 *buffer_nextp
= *logged_nextp
;
1926 * Perform a 'normal' buffer recovery. Each logged region of the
1927 * buffer should be copied over the corresponding region in the
1928 * given buffer. The bitmap in the buf log format structure indicates
1929 * where to place the logged data.
1933 xlog_recover_do_reg_buffer(
1934 xlog_recover_item_t
*item
,
1936 xfs_buf_log_format_t
*buf_f
)
1941 unsigned int *data_map
= NULL
;
1942 unsigned int map_size
= 0;
1945 switch (buf_f
->blf_type
) {
1947 data_map
= buf_f
->blf_data_map
;
1948 map_size
= buf_f
->blf_map_size
;
1952 i
= 1; /* 0 is the buf format structure */
1954 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1957 nbits
= xfs_contig_bits(data_map
, map_size
, bit
);
1959 ASSERT(item
->ri_buf
[i
].i_addr
!= NULL
);
1960 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLI_CHUNK
== 0);
1961 ASSERT(XFS_BUF_COUNT(bp
) >=
1962 ((uint
)bit
<< XFS_BLI_SHIFT
)+(nbits
<<XFS_BLI_SHIFT
));
1965 * Do a sanity check if this is a dquot buffer. Just checking
1966 * the first dquot in the buffer should do. XXXThis is
1967 * probably a good thing to do for other buf types also.
1970 if (buf_f
->blf_flags
&
1971 (XFS_BLI_UDQUOT_BUF
|XFS_BLI_PDQUOT_BUF
|XFS_BLI_GDQUOT_BUF
)) {
1972 if (item
->ri_buf
[i
].i_addr
== NULL
) {
1974 "XFS: NULL dquot in %s.", __func__
);
1977 if (item
->ri_buf
[i
].i_len
< sizeof(xfs_dqblk_t
)) {
1979 "XFS: dquot too small (%d) in %s.",
1980 item
->ri_buf
[i
].i_len
, __func__
);
1983 error
= xfs_qm_dqcheck((xfs_disk_dquot_t
*)
1984 item
->ri_buf
[i
].i_addr
,
1985 -1, 0, XFS_QMOPT_DOWARN
,
1986 "dquot_buf_recover");
1991 memcpy(xfs_buf_offset(bp
,
1992 (uint
)bit
<< XFS_BLI_SHIFT
), /* dest */
1993 item
->ri_buf
[i
].i_addr
, /* source */
1994 nbits
<<XFS_BLI_SHIFT
); /* length */
2000 /* Shouldn't be any more regions */
2001 ASSERT(i
== item
->ri_total
);
2005 * Do some primitive error checking on ondisk dquot data structures.
2009 xfs_disk_dquot_t
*ddq
,
2011 uint type
, /* used only when IO_dorepair is true */
2015 xfs_dqblk_t
*d
= (xfs_dqblk_t
*)ddq
;
2019 * We can encounter an uninitialized dquot buffer for 2 reasons:
2020 * 1. If we crash while deleting the quotainode(s), and those blks got
2021 * used for user data. This is because we take the path of regular
2022 * file deletion; however, the size field of quotainodes is never
2023 * updated, so all the tricks that we play in itruncate_finish
2024 * don't quite matter.
2026 * 2. We don't play the quota buffers when there's a quotaoff logitem.
2027 * But the allocation will be replayed so we'll end up with an
2028 * uninitialized quota block.
2030 * This is all fine; things are still consistent, and we haven't lost
2031 * any quota information. Just don't complain about bad dquot blks.
2033 if (be16_to_cpu(ddq
->d_magic
) != XFS_DQUOT_MAGIC
) {
2034 if (flags
& XFS_QMOPT_DOWARN
)
2036 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
2037 str
, id
, be16_to_cpu(ddq
->d_magic
), XFS_DQUOT_MAGIC
);
2040 if (ddq
->d_version
!= XFS_DQUOT_VERSION
) {
2041 if (flags
& XFS_QMOPT_DOWARN
)
2043 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2044 str
, id
, ddq
->d_version
, XFS_DQUOT_VERSION
);
2048 if (ddq
->d_flags
!= XFS_DQ_USER
&&
2049 ddq
->d_flags
!= XFS_DQ_PROJ
&&
2050 ddq
->d_flags
!= XFS_DQ_GROUP
) {
2051 if (flags
& XFS_QMOPT_DOWARN
)
2053 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2054 str
, id
, ddq
->d_flags
);
2058 if (id
!= -1 && id
!= be32_to_cpu(ddq
->d_id
)) {
2059 if (flags
& XFS_QMOPT_DOWARN
)
2061 "%s : ondisk-dquot 0x%p, ID mismatch: "
2062 "0x%x expected, found id 0x%x",
2063 str
, ddq
, id
, be32_to_cpu(ddq
->d_id
));
2067 if (!errs
&& ddq
->d_id
) {
2068 if (ddq
->d_blk_softlimit
&&
2069 be64_to_cpu(ddq
->d_bcount
) >=
2070 be64_to_cpu(ddq
->d_blk_softlimit
)) {
2071 if (!ddq
->d_btimer
) {
2072 if (flags
& XFS_QMOPT_DOWARN
)
2074 "%s : Dquot ID 0x%x (0x%p) "
2075 "BLK TIMER NOT STARTED",
2076 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2080 if (ddq
->d_ino_softlimit
&&
2081 be64_to_cpu(ddq
->d_icount
) >=
2082 be64_to_cpu(ddq
->d_ino_softlimit
)) {
2083 if (!ddq
->d_itimer
) {
2084 if (flags
& XFS_QMOPT_DOWARN
)
2086 "%s : Dquot ID 0x%x (0x%p) "
2087 "INODE TIMER NOT STARTED",
2088 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2092 if (ddq
->d_rtb_softlimit
&&
2093 be64_to_cpu(ddq
->d_rtbcount
) >=
2094 be64_to_cpu(ddq
->d_rtb_softlimit
)) {
2095 if (!ddq
->d_rtbtimer
) {
2096 if (flags
& XFS_QMOPT_DOWARN
)
2098 "%s : Dquot ID 0x%x (0x%p) "
2099 "RTBLK TIMER NOT STARTED",
2100 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2106 if (!errs
|| !(flags
& XFS_QMOPT_DQREPAIR
))
2109 if (flags
& XFS_QMOPT_DOWARN
)
2110 cmn_err(CE_NOTE
, "Re-initializing dquot ID 0x%x", id
);
2113 * Typically, a repair is only requested by quotacheck.
2116 ASSERT(flags
& XFS_QMOPT_DQREPAIR
);
2117 memset(d
, 0, sizeof(xfs_dqblk_t
));
2119 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
2120 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
2121 d
->dd_diskdq
.d_flags
= type
;
2122 d
->dd_diskdq
.d_id
= cpu_to_be32(id
);
2128 * Perform a dquot buffer recovery.
2129 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2130 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2131 * Else, treat it as a regular buffer and do recovery.
2134 xlog_recover_do_dquot_buffer(
2137 xlog_recover_item_t
*item
,
2139 xfs_buf_log_format_t
*buf_f
)
2144 * Filesystems are required to send in quota flags at mount time.
2146 if (mp
->m_qflags
== 0) {
2151 if (buf_f
->blf_flags
& XFS_BLI_UDQUOT_BUF
)
2152 type
|= XFS_DQ_USER
;
2153 if (buf_f
->blf_flags
& XFS_BLI_PDQUOT_BUF
)
2154 type
|= XFS_DQ_PROJ
;
2155 if (buf_f
->blf_flags
& XFS_BLI_GDQUOT_BUF
)
2156 type
|= XFS_DQ_GROUP
;
2158 * This type of quotas was turned off, so ignore this buffer
2160 if (log
->l_quotaoffs_flag
& type
)
2163 xlog_recover_do_reg_buffer(item
, bp
, buf_f
);
2167 * This routine replays a modification made to a buffer at runtime.
2168 * There are actually two types of buffer, regular and inode, which
2169 * are handled differently. Inode buffers are handled differently
2170 * in that we only recover a specific set of data from them, namely
2171 * the inode di_next_unlinked fields. This is because all other inode
2172 * data is actually logged via inode records and any data we replay
2173 * here which overlaps that may be stale.
2175 * When meta-data buffers are freed at run time we log a buffer item
2176 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2177 * of the buffer in the log should not be replayed at recovery time.
2178 * This is so that if the blocks covered by the buffer are reused for
2179 * file data before we crash we don't end up replaying old, freed
2180 * meta-data into a user's file.
2182 * To handle the cancellation of buffer log items, we make two passes
2183 * over the log during recovery. During the first we build a table of
2184 * those buffers which have been cancelled, and during the second we
2185 * only replay those buffers which do not have corresponding cancel
2186 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2187 * for more details on the implementation of the table of cancel records.
2190 xlog_recover_do_buffer_trans(
2192 xlog_recover_item_t
*item
,
2195 xfs_buf_log_format_t
*buf_f
;
2204 buf_f
= (xfs_buf_log_format_t
*)item
->ri_buf
[0].i_addr
;
2206 if (pass
== XLOG_RECOVER_PASS1
) {
2208 * In this pass we're only looking for buf items
2209 * with the XFS_BLI_CANCEL bit set.
2211 xlog_recover_do_buffer_pass1(log
, buf_f
);
2215 * In this pass we want to recover all the buffers
2216 * which have not been cancelled and are not
2217 * cancellation buffers themselves. The routine
2218 * we call here will tell us whether or not to
2219 * continue with the replay of this buffer.
2221 cancel
= xlog_recover_do_buffer_pass2(log
, buf_f
);
2226 switch (buf_f
->blf_type
) {
2228 blkno
= buf_f
->blf_blkno
;
2229 len
= buf_f
->blf_len
;
2230 flags
= buf_f
->blf_flags
;
2233 xfs_fs_cmn_err(CE_ALERT
, log
->l_mp
,
2234 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2235 buf_f
->blf_type
, log
->l_mp
->m_logname
?
2236 log
->l_mp
->m_logname
: "internal");
2237 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2238 XFS_ERRLEVEL_LOW
, log
->l_mp
);
2239 return XFS_ERROR(EFSCORRUPTED
);
2243 if (flags
& XFS_BLI_INODE_BUF
) {
2244 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, blkno
, len
,
2247 bp
= xfs_buf_read(mp
->m_ddev_targp
, blkno
, len
, 0);
2249 if (XFS_BUF_ISERROR(bp
)) {
2250 xfs_ioerror_alert("xlog_recover_do..(read#1)", log
->l_mp
,
2252 error
= XFS_BUF_GETERROR(bp
);
2258 if (flags
& XFS_BLI_INODE_BUF
) {
2259 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2261 (XFS_BLI_UDQUOT_BUF
|XFS_BLI_PDQUOT_BUF
|XFS_BLI_GDQUOT_BUF
)) {
2262 xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2264 xlog_recover_do_reg_buffer(item
, bp
, buf_f
);
2267 return XFS_ERROR(error
);
2270 * Perform delayed write on the buffer. Asynchronous writes will be
2271 * slower when taking into account all the buffers to be flushed.
2273 * Also make sure that only inode buffers with good sizes stay in
2274 * the buffer cache. The kernel moves inodes in buffers of 1 block
2275 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2276 * buffers in the log can be a different size if the log was generated
2277 * by an older kernel using unclustered inode buffers or a newer kernel
2278 * running with a different inode cluster size. Regardless, if the
2279 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2280 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2281 * the buffer out of the buffer cache so that the buffer won't
2282 * overlap with future reads of those inodes.
2284 if (XFS_DINODE_MAGIC
==
2285 be16_to_cpu(*((__be16
*)xfs_buf_offset(bp
, 0))) &&
2286 (XFS_BUF_COUNT(bp
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2287 (__uint32_t
)XFS_INODE_CLUSTER_SIZE(log
->l_mp
)))) {
2289 error
= xfs_bwrite(mp
, bp
);
2291 ASSERT(bp
->b_mount
== NULL
|| bp
->b_mount
== mp
);
2293 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2294 xfs_bdwrite(mp
, bp
);
2301 xlog_recover_do_inode_trans(
2303 xlog_recover_item_t
*item
,
2306 xfs_inode_log_format_t
*in_f
;
2317 xfs_icdinode_t
*dicp
;
2320 if (pass
== XLOG_RECOVER_PASS1
) {
2324 if (item
->ri_buf
[0].i_len
== sizeof(xfs_inode_log_format_t
)) {
2325 in_f
= (xfs_inode_log_format_t
*)item
->ri_buf
[0].i_addr
;
2327 in_f
= (xfs_inode_log_format_t
*)kmem_alloc(
2328 sizeof(xfs_inode_log_format_t
), KM_SLEEP
);
2330 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
2334 ino
= in_f
->ilf_ino
;
2338 * Inode buffers can be freed, look out for it,
2339 * and do not replay the inode.
2341 if (xlog_check_buffer_cancelled(log
, in_f
->ilf_blkno
,
2342 in_f
->ilf_len
, 0)) {
2347 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, in_f
->ilf_blkno
,
2348 in_f
->ilf_len
, XFS_BUF_LOCK
);
2349 if (XFS_BUF_ISERROR(bp
)) {
2350 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp
,
2351 bp
, in_f
->ilf_blkno
);
2352 error
= XFS_BUF_GETERROR(bp
);
2357 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2358 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, in_f
->ilf_boffset
);
2361 * Make sure the place we're flushing out to really looks
2364 if (unlikely(be16_to_cpu(dip
->di_magic
) != XFS_DINODE_MAGIC
)) {
2366 xfs_fs_cmn_err(CE_ALERT
, mp
,
2367 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2369 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2370 XFS_ERRLEVEL_LOW
, mp
);
2371 error
= EFSCORRUPTED
;
2374 dicp
= (xfs_icdinode_t
*)(item
->ri_buf
[1].i_addr
);
2375 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2377 xfs_fs_cmn_err(CE_ALERT
, mp
,
2378 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2380 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2381 XFS_ERRLEVEL_LOW
, mp
);
2382 error
= EFSCORRUPTED
;
2386 /* Skip replay when the on disk inode is newer than the log one */
2387 if (dicp
->di_flushiter
< be16_to_cpu(dip
->di_flushiter
)) {
2389 * Deal with the wrap case, DI_MAX_FLUSH is less
2390 * than smaller numbers
2392 if (be16_to_cpu(dip
->di_flushiter
) == DI_MAX_FLUSH
&&
2393 dicp
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
2401 /* Take the opportunity to reset the flush iteration count */
2402 dicp
->di_flushiter
= 0;
2404 if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFREG
)) {
2405 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2406 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2407 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2408 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2410 xfs_fs_cmn_err(CE_ALERT
, mp
,
2411 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2412 item
, dip
, bp
, ino
);
2413 error
= EFSCORRUPTED
;
2416 } else if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFDIR
)) {
2417 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2418 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2419 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2420 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2421 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2423 xfs_fs_cmn_err(CE_ALERT
, mp
,
2424 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2425 item
, dip
, bp
, ino
);
2426 error
= EFSCORRUPTED
;
2430 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2431 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2432 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2434 xfs_fs_cmn_err(CE_ALERT
, mp
,
2435 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2437 dicp
->di_nextents
+ dicp
->di_anextents
,
2439 error
= EFSCORRUPTED
;
2442 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2443 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2444 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2446 xfs_fs_cmn_err(CE_ALERT
, mp
,
2447 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2448 item
, dip
, bp
, ino
, dicp
->di_forkoff
);
2449 error
= EFSCORRUPTED
;
2452 if (unlikely(item
->ri_buf
[1].i_len
> sizeof(struct xfs_icdinode
))) {
2453 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2454 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2456 xfs_fs_cmn_err(CE_ALERT
, mp
,
2457 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2458 item
->ri_buf
[1].i_len
, item
);
2459 error
= EFSCORRUPTED
;
2463 /* The core is in in-core format */
2464 xfs_dinode_to_disk(dip
, (xfs_icdinode_t
*)item
->ri_buf
[1].i_addr
);
2466 /* the rest is in on-disk format */
2467 if (item
->ri_buf
[1].i_len
> sizeof(struct xfs_icdinode
)) {
2468 memcpy((xfs_caddr_t
) dip
+ sizeof(struct xfs_icdinode
),
2469 item
->ri_buf
[1].i_addr
+ sizeof(struct xfs_icdinode
),
2470 item
->ri_buf
[1].i_len
- sizeof(struct xfs_icdinode
));
2473 fields
= in_f
->ilf_fields
;
2474 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
2476 xfs_dinode_put_rdev(dip
, in_f
->ilf_u
.ilfu_rdev
);
2479 memcpy(XFS_DFORK_DPTR(dip
),
2480 &in_f
->ilf_u
.ilfu_uuid
,
2485 if (in_f
->ilf_size
== 2)
2486 goto write_inode_buffer
;
2487 len
= item
->ri_buf
[2].i_len
;
2488 src
= item
->ri_buf
[2].i_addr
;
2489 ASSERT(in_f
->ilf_size
<= 4);
2490 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
2491 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
2492 (len
== in_f
->ilf_dsize
));
2494 switch (fields
& XFS_ILOG_DFORK
) {
2495 case XFS_ILOG_DDATA
:
2497 memcpy(XFS_DFORK_DPTR(dip
), src
, len
);
2500 case XFS_ILOG_DBROOT
:
2501 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
, len
,
2502 (xfs_bmdr_block_t
*)XFS_DFORK_DPTR(dip
),
2503 XFS_DFORK_DSIZE(dip
, mp
));
2508 * There are no data fork flags set.
2510 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
2515 * If we logged any attribute data, recover it. There may or
2516 * may not have been any other non-core data logged in this
2519 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2520 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
2525 len
= item
->ri_buf
[attr_index
].i_len
;
2526 src
= item
->ri_buf
[attr_index
].i_addr
;
2527 ASSERT(len
== in_f
->ilf_asize
);
2529 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2530 case XFS_ILOG_ADATA
:
2532 dest
= XFS_DFORK_APTR(dip
);
2533 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
2534 memcpy(dest
, src
, len
);
2537 case XFS_ILOG_ABROOT
:
2538 dest
= XFS_DFORK_APTR(dip
);
2539 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
,
2540 len
, (xfs_bmdr_block_t
*)dest
,
2541 XFS_DFORK_ASIZE(dip
, mp
));
2545 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2554 ASSERT(bp
->b_mount
== NULL
|| bp
->b_mount
== mp
);
2556 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2557 xfs_bdwrite(mp
, bp
);
2561 return XFS_ERROR(error
);
2565 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2566 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2570 xlog_recover_do_quotaoff_trans(
2572 xlog_recover_item_t
*item
,
2575 xfs_qoff_logformat_t
*qoff_f
;
2577 if (pass
== XLOG_RECOVER_PASS2
) {
2581 qoff_f
= (xfs_qoff_logformat_t
*)item
->ri_buf
[0].i_addr
;
2585 * The logitem format's flag tells us if this was user quotaoff,
2586 * group/project quotaoff or both.
2588 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2589 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2590 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
2591 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
2592 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2593 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2599 * Recover a dquot record
2602 xlog_recover_do_dquot_trans(
2604 xlog_recover_item_t
*item
,
2609 struct xfs_disk_dquot
*ddq
, *recddq
;
2611 xfs_dq_logformat_t
*dq_f
;
2614 if (pass
== XLOG_RECOVER_PASS1
) {
2620 * Filesystems are required to send in quota flags at mount time.
2622 if (mp
->m_qflags
== 0)
2625 recddq
= (xfs_disk_dquot_t
*)item
->ri_buf
[1].i_addr
;
2627 if (item
->ri_buf
[1].i_addr
== NULL
) {
2629 "XFS: NULL dquot in %s.", __func__
);
2630 return XFS_ERROR(EIO
);
2632 if (item
->ri_buf
[1].i_len
< sizeof(xfs_dqblk_t
)) {
2634 "XFS: dquot too small (%d) in %s.",
2635 item
->ri_buf
[1].i_len
, __func__
);
2636 return XFS_ERROR(EIO
);
2640 * This type of quotas was turned off, so ignore this record.
2642 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
2644 if (log
->l_quotaoffs_flag
& type
)
2648 * At this point we know that quota was _not_ turned off.
2649 * Since the mount flags are not indicating to us otherwise, this
2650 * must mean that quota is on, and the dquot needs to be replayed.
2651 * Remember that we may not have fully recovered the superblock yet,
2652 * so we can't do the usual trick of looking at the SB quota bits.
2654 * The other possibility, of course, is that the quota subsystem was
2655 * removed since the last mount - ENOSYS.
2657 dq_f
= (xfs_dq_logformat_t
*)item
->ri_buf
[0].i_addr
;
2659 if ((error
= xfs_qm_dqcheck(recddq
,
2661 0, XFS_QMOPT_DOWARN
,
2662 "xlog_recover_do_dquot_trans (log copy)"))) {
2663 return XFS_ERROR(EIO
);
2665 ASSERT(dq_f
->qlf_len
== 1);
2667 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
2669 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
),
2672 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp
,
2673 bp
, dq_f
->qlf_blkno
);
2677 ddq
= (xfs_disk_dquot_t
*)xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
2680 * At least the magic num portion should be on disk because this
2681 * was among a chunk of dquots created earlier, and we did some
2682 * minimal initialization then.
2684 if (xfs_qm_dqcheck(ddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2685 "xlog_recover_do_dquot_trans")) {
2687 return XFS_ERROR(EIO
);
2690 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
2692 ASSERT(dq_f
->qlf_size
== 2);
2693 ASSERT(bp
->b_mount
== NULL
|| bp
->b_mount
== mp
);
2695 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2696 xfs_bdwrite(mp
, bp
);
2702 * This routine is called to create an in-core extent free intent
2703 * item from the efi format structure which was logged on disk.
2704 * It allocates an in-core efi, copies the extents from the format
2705 * structure into it, and adds the efi to the AIL with the given
2709 xlog_recover_do_efi_trans(
2711 xlog_recover_item_t
*item
,
2717 xfs_efi_log_item_t
*efip
;
2718 xfs_efi_log_format_t
*efi_formatp
;
2720 if (pass
== XLOG_RECOVER_PASS1
) {
2724 efi_formatp
= (xfs_efi_log_format_t
*)item
->ri_buf
[0].i_addr
;
2727 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
2728 if ((error
= xfs_efi_copy_format(&(item
->ri_buf
[0]),
2729 &(efip
->efi_format
)))) {
2730 xfs_efi_item_free(efip
);
2733 efip
->efi_next_extent
= efi_formatp
->efi_nextents
;
2734 efip
->efi_flags
|= XFS_EFI_COMMITTED
;
2736 spin_lock(&log
->l_ailp
->xa_lock
);
2738 * xfs_trans_ail_update() drops the AIL lock.
2740 xfs_trans_ail_update(log
->l_ailp
, (xfs_log_item_t
*)efip
, lsn
);
2746 * This routine is called when an efd format structure is found in
2747 * a committed transaction in the log. It's purpose is to cancel
2748 * the corresponding efi if it was still in the log. To do this
2749 * it searches the AIL for the efi with an id equal to that in the
2750 * efd format structure. If we find it, we remove the efi from the
2754 xlog_recover_do_efd_trans(
2756 xlog_recover_item_t
*item
,
2759 xfs_efd_log_format_t
*efd_formatp
;
2760 xfs_efi_log_item_t
*efip
= NULL
;
2761 xfs_log_item_t
*lip
;
2763 struct xfs_ail_cursor cur
;
2764 struct xfs_ail
*ailp
= log
->l_ailp
;
2766 if (pass
== XLOG_RECOVER_PASS1
) {
2770 efd_formatp
= (xfs_efd_log_format_t
*)item
->ri_buf
[0].i_addr
;
2771 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
2772 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
2773 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
2774 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
2775 efi_id
= efd_formatp
->efd_efi_id
;
2778 * Search for the efi with the id in the efd format structure
2781 spin_lock(&ailp
->xa_lock
);
2782 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
2783 while (lip
!= NULL
) {
2784 if (lip
->li_type
== XFS_LI_EFI
) {
2785 efip
= (xfs_efi_log_item_t
*)lip
;
2786 if (efip
->efi_format
.efi_id
== efi_id
) {
2788 * xfs_trans_ail_delete() drops the
2791 xfs_trans_ail_delete(ailp
, lip
);
2792 xfs_efi_item_free(efip
);
2793 spin_lock(&ailp
->xa_lock
);
2797 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
2799 xfs_trans_ail_cursor_done(ailp
, &cur
);
2800 spin_unlock(&ailp
->xa_lock
);
2804 * Perform the transaction
2806 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2807 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2810 xlog_recover_do_trans(
2812 xlog_recover_t
*trans
,
2816 xlog_recover_item_t
*item
, *first_item
;
2818 error
= xlog_recover_reorder_trans(trans
);
2822 first_item
= item
= trans
->r_itemq
;
2824 switch (ITEM_TYPE(item
)) {
2826 error
= xlog_recover_do_buffer_trans(log
, item
, pass
);
2829 error
= xlog_recover_do_inode_trans(log
, item
, pass
);
2832 error
= xlog_recover_do_efi_trans(log
, item
,
2833 trans
->r_lsn
, pass
);
2836 xlog_recover_do_efd_trans(log
, item
, pass
);
2840 error
= xlog_recover_do_dquot_trans(log
, item
, pass
);
2842 case XFS_LI_QUOTAOFF
:
2843 error
= xlog_recover_do_quotaoff_trans(log
, item
,
2848 "XFS: invalid item type (%d) xlog_recover_do_trans", ITEM_TYPE(item
));
2850 error
= XFS_ERROR(EIO
);
2856 item
= item
->ri_next
;
2857 } while (first_item
!= item
);
2863 * Free up any resources allocated by the transaction
2865 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2868 xlog_recover_free_trans(
2869 xlog_recover_t
*trans
)
2871 xlog_recover_item_t
*first_item
, *item
, *free_item
;
2874 item
= first_item
= trans
->r_itemq
;
2877 item
= item
->ri_next
;
2878 /* Free the regions in the item. */
2879 for (i
= 0; i
< free_item
->ri_cnt
; i
++) {
2880 kmem_free(free_item
->ri_buf
[i
].i_addr
);
2882 /* Free the item itself */
2883 kmem_free(free_item
->ri_buf
);
2884 kmem_free(free_item
);
2885 } while (first_item
!= item
);
2886 /* Free the transaction recover structure */
2891 xlog_recover_commit_trans(
2894 xlog_recover_t
*trans
,
2899 if ((error
= xlog_recover_unlink_tid(q
, trans
)))
2901 if ((error
= xlog_recover_do_trans(log
, trans
, pass
)))
2903 xlog_recover_free_trans(trans
); /* no error */
2908 xlog_recover_unmount_trans(
2909 xlog_recover_t
*trans
)
2911 /* Do nothing now */
2912 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2917 * There are two valid states of the r_state field. 0 indicates that the
2918 * transaction structure is in a normal state. We have either seen the
2919 * start of the transaction or the last operation we added was not a partial
2920 * operation. If the last operation we added to the transaction was a
2921 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2923 * NOTE: skip LRs with 0 data length.
2926 xlog_recover_process_data(
2928 xlog_recover_t
*rhash
[],
2929 xlog_rec_header_t
*rhead
,
2935 xlog_op_header_t
*ohead
;
2936 xlog_recover_t
*trans
;
2942 lp
= dp
+ be32_to_cpu(rhead
->h_len
);
2943 num_logops
= be32_to_cpu(rhead
->h_num_logops
);
2945 /* check the log format matches our own - else we can't recover */
2946 if (xlog_header_check_recover(log
->l_mp
, rhead
))
2947 return (XFS_ERROR(EIO
));
2949 while ((dp
< lp
) && num_logops
) {
2950 ASSERT(dp
+ sizeof(xlog_op_header_t
) <= lp
);
2951 ohead
= (xlog_op_header_t
*)dp
;
2952 dp
+= sizeof(xlog_op_header_t
);
2953 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
2954 ohead
->oh_clientid
!= XFS_LOG
) {
2956 "XFS: xlog_recover_process_data: bad clientid");
2958 return (XFS_ERROR(EIO
));
2960 tid
= be32_to_cpu(ohead
->oh_tid
);
2961 hash
= XLOG_RHASH(tid
);
2962 trans
= xlog_recover_find_tid(rhash
[hash
], tid
);
2963 if (trans
== NULL
) { /* not found; add new tid */
2964 if (ohead
->oh_flags
& XLOG_START_TRANS
)
2965 xlog_recover_new_tid(&rhash
[hash
], tid
,
2966 be64_to_cpu(rhead
->h_lsn
));
2968 if (dp
+ be32_to_cpu(ohead
->oh_len
) > lp
) {
2970 "XFS: xlog_recover_process_data: bad length");
2972 return (XFS_ERROR(EIO
));
2974 flags
= ohead
->oh_flags
& ~XLOG_END_TRANS
;
2975 if (flags
& XLOG_WAS_CONT_TRANS
)
2976 flags
&= ~XLOG_CONTINUE_TRANS
;
2978 case XLOG_COMMIT_TRANS
:
2979 error
= xlog_recover_commit_trans(log
,
2980 &rhash
[hash
], trans
, pass
);
2982 case XLOG_UNMOUNT_TRANS
:
2983 error
= xlog_recover_unmount_trans(trans
);
2985 case XLOG_WAS_CONT_TRANS
:
2986 error
= xlog_recover_add_to_cont_trans(trans
,
2987 dp
, be32_to_cpu(ohead
->oh_len
));
2989 case XLOG_START_TRANS
:
2991 "XFS: xlog_recover_process_data: bad transaction");
2993 error
= XFS_ERROR(EIO
);
2996 case XLOG_CONTINUE_TRANS
:
2997 error
= xlog_recover_add_to_trans(trans
,
2998 dp
, be32_to_cpu(ohead
->oh_len
));
3002 "XFS: xlog_recover_process_data: bad flag");
3004 error
= XFS_ERROR(EIO
);
3010 dp
+= be32_to_cpu(ohead
->oh_len
);
3017 * Process an extent free intent item that was recovered from
3018 * the log. We need to free the extents that it describes.
3021 xlog_recover_process_efi(
3023 xfs_efi_log_item_t
*efip
)
3025 xfs_efd_log_item_t
*efdp
;
3030 xfs_fsblock_t startblock_fsb
;
3032 ASSERT(!(efip
->efi_flags
& XFS_EFI_RECOVERED
));
3035 * First check the validity of the extents described by the
3036 * EFI. If any are bad, then assume that all are bad and
3037 * just toss the EFI.
3039 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3040 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3041 startblock_fsb
= XFS_BB_TO_FSB(mp
,
3042 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
3043 if ((startblock_fsb
== 0) ||
3044 (extp
->ext_len
== 0) ||
3045 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
3046 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
3048 * This will pull the EFI from the AIL and
3049 * free the memory associated with it.
3051 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
3052 return XFS_ERROR(EIO
);
3056 tp
= xfs_trans_alloc(mp
, 0);
3057 error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0, 0, 0);
3060 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
3062 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3063 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3064 error
= xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
3067 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
3071 efip
->efi_flags
|= XFS_EFI_RECOVERED
;
3072 error
= xfs_trans_commit(tp
, 0);
3076 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3081 * When this is called, all of the EFIs which did not have
3082 * corresponding EFDs should be in the AIL. What we do now
3083 * is free the extents associated with each one.
3085 * Since we process the EFIs in normal transactions, they
3086 * will be removed at some point after the commit. This prevents
3087 * us from just walking down the list processing each one.
3088 * We'll use a flag in the EFI to skip those that we've already
3089 * processed and use the AIL iteration mechanism's generation
3090 * count to try to speed this up at least a bit.
3092 * When we start, we know that the EFIs are the only things in
3093 * the AIL. As we process them, however, other items are added
3094 * to the AIL. Since everything added to the AIL must come after
3095 * everything already in the AIL, we stop processing as soon as
3096 * we see something other than an EFI in the AIL.
3099 xlog_recover_process_efis(
3102 xfs_log_item_t
*lip
;
3103 xfs_efi_log_item_t
*efip
;
3105 struct xfs_ail_cursor cur
;
3106 struct xfs_ail
*ailp
;
3109 spin_lock(&ailp
->xa_lock
);
3110 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3111 while (lip
!= NULL
) {
3113 * We're done when we see something other than an EFI.
3114 * There should be no EFIs left in the AIL now.
3116 if (lip
->li_type
!= XFS_LI_EFI
) {
3118 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
3119 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
3125 * Skip EFIs that we've already processed.
3127 efip
= (xfs_efi_log_item_t
*)lip
;
3128 if (efip
->efi_flags
& XFS_EFI_RECOVERED
) {
3129 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3133 spin_unlock(&ailp
->xa_lock
);
3134 error
= xlog_recover_process_efi(log
->l_mp
, efip
);
3135 spin_lock(&ailp
->xa_lock
);
3138 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3141 xfs_trans_ail_cursor_done(ailp
, &cur
);
3142 spin_unlock(&ailp
->xa_lock
);
3147 * This routine performs a transaction to null out a bad inode pointer
3148 * in an agi unlinked inode hash bucket.
3151 xlog_recover_clear_agi_bucket(
3153 xfs_agnumber_t agno
,
3162 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3163 error
= xfs_trans_reserve(tp
, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp
),
3168 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
3172 agi
= XFS_BUF_TO_AGI(agibp
);
3173 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
3174 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3175 (sizeof(xfs_agino_t
) * bucket
);
3176 xfs_trans_log_buf(tp
, agibp
, offset
,
3177 (offset
+ sizeof(xfs_agino_t
) - 1));
3179 error
= xfs_trans_commit(tp
, 0);
3185 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3187 xfs_fs_cmn_err(CE_WARN
, mp
, "xlog_recover_clear_agi_bucket: "
3188 "failed to clear agi %d. Continuing.", agno
);
3193 xlog_recover_process_one_iunlink(
3194 struct xfs_mount
*mp
,
3195 xfs_agnumber_t agno
,
3199 struct xfs_buf
*ibp
;
3200 struct xfs_dinode
*dip
;
3201 struct xfs_inode
*ip
;
3205 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3206 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
, 0);
3211 * Get the on disk inode to find the next inode in the bucket.
3213 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &ibp
, XFS_BUF_LOCK
);
3217 ASSERT(ip
->i_d
.di_nlink
== 0);
3218 ASSERT(ip
->i_d
.di_mode
!= 0);
3220 /* setup for the next pass */
3221 agino
= be32_to_cpu(dip
->di_next_unlinked
);
3225 * Prevent any DMAPI event from being sent when the reference on
3226 * the inode is dropped.
3228 ip
->i_d
.di_dmevmask
= 0;
3237 * We can't read in the inode this bucket points to, or this inode
3238 * is messed up. Just ditch this bucket of inodes. We will lose
3239 * some inodes and space, but at least we won't hang.
3241 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3242 * clear the inode pointer in the bucket.
3244 xlog_recover_clear_agi_bucket(mp
, agno
, bucket
);
3249 * xlog_iunlink_recover
3251 * This is called during recovery to process any inodes which
3252 * we unlinked but not freed when the system crashed. These
3253 * inodes will be on the lists in the AGI blocks. What we do
3254 * here is scan all the AGIs and fully truncate and free any
3255 * inodes found on the lists. Each inode is removed from the
3256 * lists when it has been fully truncated and is freed. The
3257 * freeing of the inode and its removal from the list must be
3261 xlog_recover_process_iunlinks(
3265 xfs_agnumber_t agno
;
3276 * Prevent any DMAPI event from being sent while in this function.
3278 mp_dmevmask
= mp
->m_dmevmask
;
3281 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3283 * Find the agi for this ag.
3285 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3288 * AGI is b0rked. Don't process it.
3290 * We should probably mark the filesystem as corrupt
3291 * after we've recovered all the ag's we can....
3295 agi
= XFS_BUF_TO_AGI(agibp
);
3297 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3298 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
3299 while (agino
!= NULLAGINO
) {
3301 * Release the agi buffer so that it can
3302 * be acquired in the normal course of the
3303 * transaction to truncate and free the inode.
3305 xfs_buf_relse(agibp
);
3307 agino
= xlog_recover_process_one_iunlink(mp
,
3308 agno
, agino
, bucket
);
3311 * Reacquire the agibuffer and continue around
3312 * the loop. This should never fail as we know
3313 * the buffer was good earlier on.
3315 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3317 agi
= XFS_BUF_TO_AGI(agibp
);
3322 * Release the buffer for the current agi so we can
3323 * go on to the next one.
3325 xfs_buf_relse(agibp
);
3328 mp
->m_dmevmask
= mp_dmevmask
;
3334 xlog_pack_data_checksum(
3336 xlog_in_core_t
*iclog
,
3343 up
= (__be32
*)iclog
->ic_datap
;
3344 /* divide length by 4 to get # words */
3345 for (i
= 0; i
< (size
>> 2); i
++) {
3346 chksum
^= be32_to_cpu(*up
);
3349 iclog
->ic_header
.h_chksum
= cpu_to_be32(chksum
);
3352 #define xlog_pack_data_checksum(log, iclog, size)
3356 * Stamp cycle number in every block
3361 xlog_in_core_t
*iclog
,
3365 int size
= iclog
->ic_offset
+ roundoff
;
3369 xlog_pack_data_checksum(log
, iclog
, size
);
3371 cycle_lsn
= CYCLE_LSN_DISK(iclog
->ic_header
.h_lsn
);
3373 dp
= iclog
->ic_datap
;
3374 for (i
= 0; i
< BTOBB(size
) &&
3375 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3376 iclog
->ic_header
.h_cycle_data
[i
] = *(__be32
*)dp
;
3377 *(__be32
*)dp
= cycle_lsn
;
3381 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3382 xlog_in_core_2_t
*xhdr
= iclog
->ic_data
;
3384 for ( ; i
< BTOBB(size
); i
++) {
3385 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3386 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3387 xhdr
[j
].hic_xheader
.xh_cycle_data
[k
] = *(__be32
*)dp
;
3388 *(__be32
*)dp
= cycle_lsn
;
3392 for (i
= 1; i
< log
->l_iclog_heads
; i
++) {
3393 xhdr
[i
].hic_xheader
.xh_cycle
= cycle_lsn
;
3398 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3400 xlog_unpack_data_checksum(
3401 xlog_rec_header_t
*rhead
,
3405 __be32
*up
= (__be32
*)dp
;
3409 /* divide length by 4 to get # words */
3410 for (i
=0; i
< be32_to_cpu(rhead
->h_len
) >> 2; i
++) {
3411 chksum
^= be32_to_cpu(*up
);
3414 if (chksum
!= be32_to_cpu(rhead
->h_chksum
)) {
3415 if (rhead
->h_chksum
||
3416 ((log
->l_flags
& XLOG_CHKSUM_MISMATCH
) == 0)) {
3418 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
3419 be32_to_cpu(rhead
->h_chksum
), chksum
);
3421 "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3422 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3424 "XFS: LogR this is a LogV2 filesystem\n");
3426 log
->l_flags
|= XLOG_CHKSUM_MISMATCH
;
3431 #define xlog_unpack_data_checksum(rhead, dp, log)
3436 xlog_rec_header_t
*rhead
,
3442 for (i
= 0; i
< BTOBB(be32_to_cpu(rhead
->h_len
)) &&
3443 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3444 *(__be32
*)dp
= *(__be32
*)&rhead
->h_cycle_data
[i
];
3448 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3449 xlog_in_core_2_t
*xhdr
= (xlog_in_core_2_t
*)rhead
;
3450 for ( ; i
< BTOBB(be32_to_cpu(rhead
->h_len
)); i
++) {
3451 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3452 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3453 *(__be32
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
3458 xlog_unpack_data_checksum(rhead
, dp
, log
);
3462 xlog_valid_rec_header(
3464 xlog_rec_header_t
*rhead
,
3469 if (unlikely(be32_to_cpu(rhead
->h_magicno
) != XLOG_HEADER_MAGIC_NUM
)) {
3470 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3471 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3472 return XFS_ERROR(EFSCORRUPTED
);
3475 (!rhead
->h_version
||
3476 (be32_to_cpu(rhead
->h_version
) & (~XLOG_VERSION_OKBITS
))))) {
3477 xlog_warn("XFS: %s: unrecognised log version (%d).",
3478 __func__
, be32_to_cpu(rhead
->h_version
));
3479 return XFS_ERROR(EIO
);
3482 /* LR body must have data or it wouldn't have been written */
3483 hlen
= be32_to_cpu(rhead
->h_len
);
3484 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
3485 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3486 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3487 return XFS_ERROR(EFSCORRUPTED
);
3489 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
3490 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3491 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3492 return XFS_ERROR(EFSCORRUPTED
);
3498 * Read the log from tail to head and process the log records found.
3499 * Handle the two cases where the tail and head are in the same cycle
3500 * and where the active portion of the log wraps around the end of
3501 * the physical log separately. The pass parameter is passed through
3502 * to the routines called to process the data and is not looked at
3506 xlog_do_recovery_pass(
3508 xfs_daddr_t head_blk
,
3509 xfs_daddr_t tail_blk
,
3512 xlog_rec_header_t
*rhead
;
3514 xfs_caddr_t bufaddr
, offset
;
3515 xfs_buf_t
*hbp
, *dbp
;
3516 int error
= 0, h_size
;
3517 int bblks
, split_bblks
;
3518 int hblks
, split_hblks
, wrapped_hblks
;
3519 xlog_recover_t
*rhash
[XLOG_RHASH_SIZE
];
3521 ASSERT(head_blk
!= tail_blk
);
3524 * Read the header of the tail block and get the iclog buffer size from
3525 * h_size. Use this to tell how many sectors make up the log header.
3527 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3529 * When using variable length iclogs, read first sector of
3530 * iclog header and extract the header size from it. Get a
3531 * new hbp that is the correct size.
3533 hbp
= xlog_get_bp(log
, 1);
3537 error
= xlog_bread(log
, tail_blk
, 1, hbp
, &offset
);
3541 rhead
= (xlog_rec_header_t
*)offset
;
3542 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
3545 h_size
= be32_to_cpu(rhead
->h_size
);
3546 if ((be32_to_cpu(rhead
->h_version
) & XLOG_VERSION_2
) &&
3547 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
3548 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
3549 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
3552 hbp
= xlog_get_bp(log
, hblks
);
3557 ASSERT(log
->l_sectbb_log
== 0);
3559 hbp
= xlog_get_bp(log
, 1);
3560 h_size
= XLOG_BIG_RECORD_BSIZE
;
3565 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
3571 memset(rhash
, 0, sizeof(rhash
));
3572 if (tail_blk
<= head_blk
) {
3573 for (blk_no
= tail_blk
; blk_no
< head_blk
; ) {
3574 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3578 rhead
= (xlog_rec_header_t
*)offset
;
3579 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3583 /* blocks in data section */
3584 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3585 error
= xlog_bread(log
, blk_no
+ hblks
, bblks
, dbp
,
3590 xlog_unpack_data(rhead
, offset
, log
);
3591 if ((error
= xlog_recover_process_data(log
,
3592 rhash
, rhead
, offset
, pass
)))
3594 blk_no
+= bblks
+ hblks
;
3598 * Perform recovery around the end of the physical log.
3599 * When the head is not on the same cycle number as the tail,
3600 * we can't do a sequential recovery as above.
3603 while (blk_no
< log
->l_logBBsize
) {
3605 * Check for header wrapping around physical end-of-log
3610 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
3611 /* Read header in one read */
3612 error
= xlog_bread(log
, blk_no
, hblks
, hbp
,
3617 /* This LR is split across physical log end */
3618 if (blk_no
!= log
->l_logBBsize
) {
3619 /* some data before physical log end */
3620 ASSERT(blk_no
<= INT_MAX
);
3621 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
3622 ASSERT(split_hblks
> 0);
3623 error
= xlog_bread(log
, blk_no
,
3631 * Note: this black magic still works with
3632 * large sector sizes (non-512) only because:
3633 * - we increased the buffer size originally
3634 * by 1 sector giving us enough extra space
3635 * for the second read;
3636 * - the log start is guaranteed to be sector
3638 * - we read the log end (LR header start)
3639 * _first_, then the log start (LR header end)
3640 * - order is important.
3642 wrapped_hblks
= hblks
- split_hblks
;
3643 bufaddr
= XFS_BUF_PTR(hbp
);
3644 error
= XFS_BUF_SET_PTR(hbp
,
3645 bufaddr
+ BBTOB(split_hblks
),
3646 BBTOB(hblks
- split_hblks
));
3650 error
= xlog_bread_noalign(log
, 0,
3651 wrapped_hblks
, hbp
);
3655 error
= XFS_BUF_SET_PTR(hbp
, bufaddr
,
3661 offset
= xlog_align(log
, 0,
3662 wrapped_hblks
, hbp
);
3664 rhead
= (xlog_rec_header_t
*)offset
;
3665 error
= xlog_valid_rec_header(log
, rhead
,
3666 split_hblks
? blk_no
: 0);
3670 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3673 /* Read in data for log record */
3674 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
3675 error
= xlog_bread(log
, blk_no
, bblks
, dbp
,
3680 /* This log record is split across the
3681 * physical end of log */
3684 if (blk_no
!= log
->l_logBBsize
) {
3685 /* some data is before the physical
3687 ASSERT(!wrapped_hblks
);
3688 ASSERT(blk_no
<= INT_MAX
);
3690 log
->l_logBBsize
- (int)blk_no
;
3691 ASSERT(split_bblks
> 0);
3692 error
= xlog_bread(log
, blk_no
,
3700 * Note: this black magic still works with
3701 * large sector sizes (non-512) only because:
3702 * - we increased the buffer size originally
3703 * by 1 sector giving us enough extra space
3704 * for the second read;
3705 * - the log start is guaranteed to be sector
3707 * - we read the log end (LR header start)
3708 * _first_, then the log start (LR header end)
3709 * - order is important.
3711 bufaddr
= XFS_BUF_PTR(dbp
);
3712 error
= XFS_BUF_SET_PTR(dbp
,
3713 bufaddr
+ BBTOB(split_bblks
),
3714 BBTOB(bblks
- split_bblks
));
3718 error
= xlog_bread_noalign(log
, wrapped_hblks
,
3719 bblks
- split_bblks
,
3724 error
= XFS_BUF_SET_PTR(dbp
, bufaddr
, h_size
);
3729 offset
= xlog_align(log
, wrapped_hblks
,
3730 bblks
- split_bblks
, dbp
);
3732 xlog_unpack_data(rhead
, offset
, log
);
3733 if ((error
= xlog_recover_process_data(log
, rhash
,
3734 rhead
, offset
, pass
)))
3739 ASSERT(blk_no
>= log
->l_logBBsize
);
3740 blk_no
-= log
->l_logBBsize
;
3742 /* read first part of physical log */
3743 while (blk_no
< head_blk
) {
3744 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3748 rhead
= (xlog_rec_header_t
*)offset
;
3749 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3753 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3754 error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
,
3759 xlog_unpack_data(rhead
, offset
, log
);
3760 if ((error
= xlog_recover_process_data(log
, rhash
,
3761 rhead
, offset
, pass
)))
3763 blk_no
+= bblks
+ hblks
;
3775 * Do the recovery of the log. We actually do this in two phases.
3776 * The two passes are necessary in order to implement the function
3777 * of cancelling a record written into the log. The first pass
3778 * determines those things which have been cancelled, and the
3779 * second pass replays log items normally except for those which
3780 * have been cancelled. The handling of the replay and cancellations
3781 * takes place in the log item type specific routines.
3783 * The table of items which have cancel records in the log is allocated
3784 * and freed at this level, since only here do we know when all of
3785 * the log recovery has been completed.
3788 xlog_do_log_recovery(
3790 xfs_daddr_t head_blk
,
3791 xfs_daddr_t tail_blk
)
3795 ASSERT(head_blk
!= tail_blk
);
3798 * First do a pass to find all of the cancelled buf log items.
3799 * Store them in the buf_cancel_table for use in the second pass.
3801 log
->l_buf_cancel_table
=
3802 (xfs_buf_cancel_t
**)kmem_zalloc(XLOG_BC_TABLE_SIZE
*
3803 sizeof(xfs_buf_cancel_t
*),
3805 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3806 XLOG_RECOVER_PASS1
);
3808 kmem_free(log
->l_buf_cancel_table
);
3809 log
->l_buf_cancel_table
= NULL
;
3813 * Then do a second pass to actually recover the items in the log.
3814 * When it is complete free the table of buf cancel items.
3816 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3817 XLOG_RECOVER_PASS2
);
3822 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
3823 ASSERT(log
->l_buf_cancel_table
[i
] == NULL
);
3827 kmem_free(log
->l_buf_cancel_table
);
3828 log
->l_buf_cancel_table
= NULL
;
3834 * Do the actual recovery
3839 xfs_daddr_t head_blk
,
3840 xfs_daddr_t tail_blk
)
3847 * First replay the images in the log.
3849 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
3854 XFS_bflush(log
->l_mp
->m_ddev_targp
);
3857 * If IO errors happened during recovery, bail out.
3859 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
3864 * We now update the tail_lsn since much of the recovery has completed
3865 * and there may be space available to use. If there were no extent
3866 * or iunlinks, we can free up the entire log and set the tail_lsn to
3867 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3868 * lsn of the last known good LR on disk. If there are extent frees
3869 * or iunlinks they will have some entries in the AIL; so we look at
3870 * the AIL to determine how to set the tail_lsn.
3872 xlog_assign_tail_lsn(log
->l_mp
);
3875 * Now that we've finished replaying all buffer and inode
3876 * updates, re-read in the superblock.
3878 bp
= xfs_getsb(log
->l_mp
, 0);
3880 ASSERT(!(XFS_BUF_ISWRITE(bp
)));
3881 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp
)));
3883 XFS_BUF_UNASYNC(bp
);
3884 xfsbdstrat(log
->l_mp
, bp
);
3885 error
= xfs_iowait(bp
);
3887 xfs_ioerror_alert("xlog_do_recover",
3888 log
->l_mp
, bp
, XFS_BUF_ADDR(bp
));
3894 /* Convert superblock from on-disk format */
3895 sbp
= &log
->l_mp
->m_sb
;
3896 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
3897 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
3898 ASSERT(xfs_sb_good_version(sbp
));
3901 /* We've re-read the superblock so re-initialize per-cpu counters */
3902 xfs_icsb_reinit_counters(log
->l_mp
);
3904 xlog_recover_check_summary(log
);
3906 /* Normal transactions can now occur */
3907 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
3912 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3914 * Return error or zero.
3920 xfs_daddr_t head_blk
, tail_blk
;
3923 /* find the tail of the log */
3924 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
)))
3927 if (tail_blk
!= head_blk
) {
3928 /* There used to be a comment here:
3930 * disallow recovery on read-only mounts. note -- mount
3931 * checks for ENOSPC and turns it into an intelligent
3933 * ...but this is no longer true. Now, unless you specify
3934 * NORECOVERY (in which case this function would never be
3935 * called), we just go ahead and recover. We do this all
3936 * under the vfs layer, so we can get away with it unless
3937 * the device itself is read-only, in which case we fail.
3939 if ((error
= xfs_dev_is_read_only(log
->l_mp
, "recovery"))) {
3944 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3945 log
->l_mp
->m_fsname
, log
->l_mp
->m_logname
?
3946 log
->l_mp
->m_logname
: "internal");
3948 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
3949 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
3955 * In the first part of recovery we replay inodes and buffers and build
3956 * up the list of extent free items which need to be processed. Here
3957 * we process the extent free items and clean up the on disk unlinked
3958 * inode lists. This is separated from the first part of recovery so
3959 * that the root and real-time bitmap inodes can be read in from disk in
3960 * between the two stages. This is necessary so that we can free space
3961 * in the real-time portion of the file system.
3964 xlog_recover_finish(
3968 * Now we're ready to do the transactions needed for the
3969 * rest of recovery. Start with completing all the extent
3970 * free intent records and then process the unlinked inode
3971 * lists. At this point, we essentially run in normal mode
3972 * except that we're still performing recovery actions
3973 * rather than accepting new requests.
3975 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
3977 error
= xlog_recover_process_efis(log
);
3980 "Failed to recover EFIs on filesystem: %s",
3981 log
->l_mp
->m_fsname
);
3985 * Sync the log to get all the EFIs out of the AIL.
3986 * This isn't absolutely necessary, but it helps in
3987 * case the unlink transactions would have problems
3988 * pushing the EFIs out of the way.
3990 xfs_log_force(log
->l_mp
, (xfs_lsn_t
)0,
3991 (XFS_LOG_FORCE
| XFS_LOG_SYNC
));
3993 xlog_recover_process_iunlinks(log
);
3995 xlog_recover_check_summary(log
);
3998 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3999 log
->l_mp
->m_fsname
, log
->l_mp
->m_logname
?
4000 log
->l_mp
->m_logname
: "internal");
4001 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
4004 "!Ending clean XFS mount for filesystem: %s\n",
4005 log
->l_mp
->m_fsname
);
4013 * Read all of the agf and agi counters and check that they
4014 * are consistent with the superblock counters.
4017 xlog_recover_check_summary(
4025 #ifdef XFS_LOUD_RECOVERY
4028 xfs_agnumber_t agno
;
4029 __uint64_t freeblks
;
4039 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
4040 error
= xfs_read_agf(mp
, NULL
, agno
, 0, &agfbp
);
4042 xfs_fs_cmn_err(CE_ALERT
, mp
,
4043 "xlog_recover_check_summary(agf)"
4044 "agf read failed agno %d error %d",
4047 agfp
= XFS_BUF_TO_AGF(agfbp
);
4048 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
4049 be32_to_cpu(agfp
->agf_flcount
);
4050 xfs_buf_relse(agfbp
);
4053 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
4055 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agibp
);
4057 itotal
+= be32_to_cpu(agi
->agi_count
);
4058 ifree
+= be32_to_cpu(agi
->agi_freecount
);
4059 xfs_buf_relse(agibp
);
4063 sbbp
= xfs_getsb(mp
, 0);
4064 #ifdef XFS_LOUD_RECOVERY
4066 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(sbbp
));
4068 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4069 sbp
->sb_icount
, itotal
);
4071 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4072 sbp
->sb_ifree
, ifree
);
4074 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4075 sbp
->sb_fdblocks
, freeblks
);
4078 * This is turned off until I account for the allocation
4079 * btree blocks which live in free space.
4081 ASSERT(sbp
->sb_icount
== itotal
);
4082 ASSERT(sbp
->sb_ifree
== ifree
);
4083 ASSERT(sbp
->sb_fdblocks
== freeblks
);
4086 xfs_buf_relse(sbbp
);