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"
27 #include "xfs_mount.h"
28 #include "xfs_error.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_alloc.h"
36 #include "xfs_ialloc.h"
37 #include "xfs_log_priv.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_log_recover.h"
40 #include "xfs_extfree_item.h"
41 #include "xfs_trans_priv.h"
42 #include "xfs_quota.h"
44 #include "xfs_utils.h"
45 #include "xfs_trace.h"
47 STATIC
int xlog_find_zeroed(xlog_t
*, xfs_daddr_t
*);
48 STATIC
int xlog_clear_stale_blocks(xlog_t
*, xfs_lsn_t
);
50 STATIC
void xlog_recover_check_summary(xlog_t
*);
52 #define xlog_recover_check_summary(log)
56 * Sector aligned buffer routines for buffer create/read/write/access
60 * Verify the given count of basic blocks is valid number of blocks
61 * to specify for an operation involving the given XFS log buffer.
62 * Returns nonzero if the count is valid, 0 otherwise.
66 xlog_buf_bbcount_valid(
70 return bbcount
> 0 && bbcount
<= log
->l_logBBsize
;
74 * Allocate a buffer to hold log data. The buffer needs to be able
75 * to map to a range of nbblks basic blocks at any valid (basic
76 * block) offset within the log.
83 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
84 xlog_warn("XFS: Invalid block length (0x%x) given for buffer",
86 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
91 * We do log I/O in units of log sectors (a power-of-2
92 * multiple of the basic block size), so we round up the
93 * requested size to acommodate the basic blocks required
94 * for complete log sectors.
96 * In addition, the buffer may be used for a non-sector-
97 * aligned block offset, in which case an I/O of the
98 * requested size could extend beyond the end of the
99 * buffer. If the requested size is only 1 basic block it
100 * will never straddle a sector boundary, so this won't be
101 * an issue. Nor will this be a problem if the log I/O is
102 * done in basic blocks (sector size 1). But otherwise we
103 * extend the buffer by one extra log sector to ensure
104 * there's space to accomodate this possiblility.
106 if (nbblks
> 1 && log
->l_sectBBsize
> 1)
107 nbblks
+= log
->l_sectBBsize
;
108 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
110 return xfs_buf_get_noaddr(BBTOB(nbblks
), log
->l_mp
->m_logdev_targp
);
121 * Return the address of the start of the given block number's data
122 * in a log buffer. The buffer covers a log sector-aligned region.
131 xfs_daddr_t offset
= blk_no
& ((xfs_daddr_t
)log
->l_sectBBsize
- 1);
133 ASSERT(BBTOB(offset
+ nbblks
) <= XFS_BUF_SIZE(bp
));
134 return XFS_BUF_PTR(bp
) + BBTOB(offset
);
139 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
150 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
151 xlog_warn("XFS: Invalid block length (0x%x) given for buffer",
153 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
157 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
158 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
161 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
163 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
166 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
167 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
169 xfsbdstrat(log
->l_mp
, bp
);
170 error
= xfs_iowait(bp
);
172 xfs_ioerror_alert("xlog_bread", log
->l_mp
,
173 bp
, XFS_BUF_ADDR(bp
));
187 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
191 *offset
= xlog_align(log
, blk_no
, nbblks
, bp
);
196 * Write out the buffer at the given block for the given number of blocks.
197 * The buffer is kept locked across the write and is returned locked.
198 * This can only be used for synchronous log writes.
209 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
210 xlog_warn("XFS: Invalid block length (0x%x) given for buffer",
212 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
216 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
217 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
220 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
222 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
223 XFS_BUF_ZEROFLAGS(bp
);
226 XFS_BUF_PSEMA(bp
, PRIBIO
);
227 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
228 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
230 if ((error
= xfs_bwrite(log
->l_mp
, bp
)))
231 xfs_ioerror_alert("xlog_bwrite", log
->l_mp
,
232 bp
, XFS_BUF_ADDR(bp
));
238 * dump debug superblock and log record information
241 xlog_header_check_dump(
243 xlog_rec_header_t
*head
)
245 cmn_err(CE_DEBUG
, "%s: SB : uuid = %pU, fmt = %d\n",
246 __func__
, &mp
->m_sb
.sb_uuid
, XLOG_FMT
);
247 cmn_err(CE_DEBUG
, " log : uuid = %pU, fmt = %d\n",
248 &head
->h_fs_uuid
, be32_to_cpu(head
->h_fmt
));
251 #define xlog_header_check_dump(mp, head)
255 * check log record header for recovery
258 xlog_header_check_recover(
260 xlog_rec_header_t
*head
)
262 ASSERT(be32_to_cpu(head
->h_magicno
) == XLOG_HEADER_MAGIC_NUM
);
265 * IRIX doesn't write the h_fmt field and leaves it zeroed
266 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
267 * a dirty log created in IRIX.
269 if (unlikely(be32_to_cpu(head
->h_fmt
) != XLOG_FMT
)) {
271 "XFS: dirty log written in incompatible format - can't recover");
272 xlog_header_check_dump(mp
, head
);
273 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
274 XFS_ERRLEVEL_HIGH
, mp
);
275 return XFS_ERROR(EFSCORRUPTED
);
276 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
278 "XFS: dirty log entry has mismatched uuid - can't recover");
279 xlog_header_check_dump(mp
, head
);
280 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
281 XFS_ERRLEVEL_HIGH
, mp
);
282 return XFS_ERROR(EFSCORRUPTED
);
288 * read the head block of the log and check the header
291 xlog_header_check_mount(
293 xlog_rec_header_t
*head
)
295 ASSERT(be32_to_cpu(head
->h_magicno
) == XLOG_HEADER_MAGIC_NUM
);
297 if (uuid_is_nil(&head
->h_fs_uuid
)) {
299 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
300 * h_fs_uuid is nil, we assume this log was last mounted
301 * by IRIX and continue.
303 xlog_warn("XFS: nil uuid in log - IRIX style log");
304 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
305 xlog_warn("XFS: log has mismatched uuid - can't recover");
306 xlog_header_check_dump(mp
, head
);
307 XFS_ERROR_REPORT("xlog_header_check_mount",
308 XFS_ERRLEVEL_HIGH
, mp
);
309 return XFS_ERROR(EFSCORRUPTED
);
318 if (XFS_BUF_GETERROR(bp
)) {
320 * We're not going to bother about retrying
321 * this during recovery. One strike!
323 xfs_ioerror_alert("xlog_recover_iodone",
324 bp
->b_mount
, bp
, XFS_BUF_ADDR(bp
));
325 xfs_force_shutdown(bp
->b_mount
, SHUTDOWN_META_IO_ERROR
);
328 XFS_BUF_CLR_IODONE_FUNC(bp
);
333 * This routine finds (to an approximation) the first block in the physical
334 * log which contains the given cycle. It uses a binary search algorithm.
335 * Note that the algorithm can not be perfect because the disk will not
336 * necessarily be perfect.
339 xlog_find_cycle_start(
342 xfs_daddr_t first_blk
,
343 xfs_daddr_t
*last_blk
,
353 mid_blk
= BLK_AVG(first_blk
, end_blk
);
354 while (mid_blk
!= first_blk
&& mid_blk
!= end_blk
) {
355 error
= xlog_bread(log
, mid_blk
, 1, bp
, &offset
);
358 mid_cycle
= xlog_get_cycle(offset
);
359 if (mid_cycle
== cycle
)
360 end_blk
= mid_blk
; /* last_half_cycle == mid_cycle */
362 first_blk
= mid_blk
; /* first_half_cycle == mid_cycle */
363 mid_blk
= BLK_AVG(first_blk
, end_blk
);
365 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == end_blk
) ||
366 (mid_blk
== end_blk
&& mid_blk
-1 == first_blk
));
374 * Check that a range of blocks does not contain stop_on_cycle_no.
375 * Fill in *new_blk with the block offset where such a block is
376 * found, or with -1 (an invalid block number) if there is no such
377 * block in the range. The scan needs to occur from front to back
378 * and the pointer into the region must be updated since a later
379 * routine will need to perform another test.
382 xlog_find_verify_cycle(
384 xfs_daddr_t start_blk
,
386 uint stop_on_cycle_no
,
387 xfs_daddr_t
*new_blk
)
393 xfs_caddr_t buf
= NULL
;
397 * Greedily allocate a buffer big enough to handle the full
398 * range of basic blocks we'll be examining. If that fails,
399 * try a smaller size. We need to be able to read at least
400 * a log sector, or we're out of luck.
402 bufblks
= 1 << ffs(nbblks
);
403 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
405 if (bufblks
< log
->l_sectBBsize
)
409 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
412 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
414 error
= xlog_bread(log
, i
, bcount
, bp
, &buf
);
418 for (j
= 0; j
< bcount
; j
++) {
419 cycle
= xlog_get_cycle(buf
);
420 if (cycle
== stop_on_cycle_no
) {
437 * Potentially backup over partial log record write.
439 * In the typical case, last_blk is the number of the block directly after
440 * a good log record. Therefore, we subtract one to get the block number
441 * of the last block in the given buffer. extra_bblks contains the number
442 * of blocks we would have read on a previous read. This happens when the
443 * last log record is split over the end of the physical log.
445 * extra_bblks is the number of blocks potentially verified on a previous
446 * call to this routine.
449 xlog_find_verify_log_record(
451 xfs_daddr_t start_blk
,
452 xfs_daddr_t
*last_blk
,
457 xfs_caddr_t offset
= NULL
;
458 xlog_rec_header_t
*head
= NULL
;
461 int num_blks
= *last_blk
- start_blk
;
464 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
466 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
467 if (!(bp
= xlog_get_bp(log
, 1)))
471 error
= xlog_bread(log
, start_blk
, num_blks
, bp
, &offset
);
474 offset
+= ((num_blks
- 1) << BBSHIFT
);
477 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
479 /* valid log record not found */
481 "XFS: Log inconsistent (didn't find previous header)");
483 error
= XFS_ERROR(EIO
);
488 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
493 head
= (xlog_rec_header_t
*)offset
;
495 if (XLOG_HEADER_MAGIC_NUM
== be32_to_cpu(head
->h_magicno
))
503 * We hit the beginning of the physical log & still no header. Return
504 * to caller. If caller can handle a return of -1, then this routine
505 * will be called again for the end of the physical log.
513 * We have the final block of the good log (the first block
514 * of the log record _before_ the head. So we check the uuid.
516 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
520 * We may have found a log record header before we expected one.
521 * last_blk will be the 1st block # with a given cycle #. We may end
522 * up reading an entire log record. In this case, we don't want to
523 * reset last_blk. Only when last_blk points in the middle of a log
524 * record do we update last_blk.
526 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
527 uint h_size
= be32_to_cpu(head
->h_size
);
529 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
530 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
536 if (*last_blk
- i
+ extra_bblks
!=
537 BTOBB(be32_to_cpu(head
->h_len
)) + xhdrs
)
546 * Head is defined to be the point of the log where the next log write
547 * write could go. This means that incomplete LR writes at the end are
548 * eliminated when calculating the head. We aren't guaranteed that previous
549 * LR have complete transactions. We only know that a cycle number of
550 * current cycle number -1 won't be present in the log if we start writing
551 * from our current block number.
553 * last_blk contains the block number of the first block with a given
556 * Return: zero if normal, non-zero if error.
561 xfs_daddr_t
*return_head_blk
)
565 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
567 uint first_half_cycle
, last_half_cycle
;
569 int error
, log_bbnum
= log
->l_logBBsize
;
571 /* Is the end of the log device zeroed? */
572 if ((error
= xlog_find_zeroed(log
, &first_blk
)) == -1) {
573 *return_head_blk
= first_blk
;
575 /* Is the whole lot zeroed? */
577 /* Linux XFS shouldn't generate totally zeroed logs -
578 * mkfs etc write a dummy unmount record to a fresh
579 * log so we can store the uuid in there
581 xlog_warn("XFS: totally zeroed log");
586 xlog_warn("XFS: empty log check failed");
590 first_blk
= 0; /* get cycle # of 1st block */
591 bp
= xlog_get_bp(log
, 1);
595 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
599 first_half_cycle
= xlog_get_cycle(offset
);
601 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
602 error
= xlog_bread(log
, last_blk
, 1, bp
, &offset
);
606 last_half_cycle
= xlog_get_cycle(offset
);
607 ASSERT(last_half_cycle
!= 0);
610 * If the 1st half cycle number is equal to the last half cycle number,
611 * then the entire log is stamped with the same cycle number. In this
612 * case, head_blk can't be set to zero (which makes sense). The below
613 * math doesn't work out properly with head_blk equal to zero. Instead,
614 * we set it to log_bbnum which is an invalid block number, but this
615 * value makes the math correct. If head_blk doesn't changed through
616 * all the tests below, *head_blk is set to zero at the very end rather
617 * than log_bbnum. In a sense, log_bbnum and zero are the same block
618 * in a circular file.
620 if (first_half_cycle
== last_half_cycle
) {
622 * In this case we believe that the entire log should have
623 * cycle number last_half_cycle. We need to scan backwards
624 * from the end verifying that there are no holes still
625 * containing last_half_cycle - 1. If we find such a hole,
626 * then the start of that hole will be the new head. The
627 * simple case looks like
628 * x | x ... | x - 1 | x
629 * Another case that fits this picture would be
630 * x | x + 1 | x ... | x
631 * In this case the head really is somewhere at the end of the
632 * log, as one of the latest writes at the beginning was
635 * x | x + 1 | x ... | x - 1 | x
636 * This is really the combination of the above two cases, and
637 * the head has to end up at the start of the x-1 hole at the
640 * In the 256k log case, we will read from the beginning to the
641 * end of the log and search for cycle numbers equal to x-1.
642 * We don't worry about the x+1 blocks that we encounter,
643 * because we know that they cannot be the head since the log
646 head_blk
= log_bbnum
;
647 stop_on_cycle
= last_half_cycle
- 1;
650 * In this case we want to find the first block with cycle
651 * number matching last_half_cycle. We expect the log to be
653 * x + 1 ... | x ... | x
654 * The first block with cycle number x (last_half_cycle) will
655 * be where the new head belongs. First we do a binary search
656 * for the first occurrence of last_half_cycle. The binary
657 * search may not be totally accurate, so then we scan back
658 * from there looking for occurrences of last_half_cycle before
659 * us. If that backwards scan wraps around the beginning of
660 * the log, then we look for occurrences of last_half_cycle - 1
661 * at the end of the log. The cases we're looking for look
663 * v binary search stopped here
664 * x + 1 ... | x | x + 1 | x ... | x
665 * ^ but we want to locate this spot
667 * <---------> less than scan distance
668 * x + 1 ... | x ... | x - 1 | x
669 * ^ we want to locate this spot
671 stop_on_cycle
= last_half_cycle
;
672 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
673 &head_blk
, last_half_cycle
)))
678 * Now validate the answer. Scan back some number of maximum possible
679 * blocks and make sure each one has the expected cycle number. The
680 * maximum is determined by the total possible amount of buffering
681 * in the in-core log. The following number can be made tighter if
682 * we actually look at the block size of the filesystem.
684 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
685 if (head_blk
>= num_scan_bblks
) {
687 * We are guaranteed that the entire check can be performed
690 start_blk
= head_blk
- num_scan_bblks
;
691 if ((error
= xlog_find_verify_cycle(log
,
692 start_blk
, num_scan_bblks
,
693 stop_on_cycle
, &new_blk
)))
697 } else { /* need to read 2 parts of log */
699 * We are going to scan backwards in the log in two parts.
700 * First we scan the physical end of the log. In this part
701 * of the log, we are looking for blocks with cycle number
702 * last_half_cycle - 1.
703 * If we find one, then we know that the log starts there, as
704 * we've found a hole that didn't get written in going around
705 * the end of the physical log. The simple case for this is
706 * x + 1 ... | x ... | x - 1 | x
707 * <---------> less than scan distance
708 * If all of the blocks at the end of the log have cycle number
709 * last_half_cycle, then we check the blocks at the start of
710 * the log looking for occurrences of last_half_cycle. If we
711 * find one, then our current estimate for the location of the
712 * first occurrence of last_half_cycle is wrong and we move
713 * back to the hole we've found. This case looks like
714 * x + 1 ... | x | x + 1 | x ...
715 * ^ binary search stopped here
716 * Another case we need to handle that only occurs in 256k
718 * x + 1 ... | x ... | x+1 | x ...
719 * ^ binary search stops here
720 * In a 256k log, the scan at the end of the log will see the
721 * x + 1 blocks. We need to skip past those since that is
722 * certainly not the head of the log. By searching for
723 * last_half_cycle-1 we accomplish that.
725 ASSERT(head_blk
<= INT_MAX
&&
726 (xfs_daddr_t
) num_scan_bblks
>= head_blk
);
727 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
728 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
729 num_scan_bblks
- (int)head_blk
,
730 (stop_on_cycle
- 1), &new_blk
)))
738 * Scan beginning of log now. The last part of the physical
739 * log is good. This scan needs to verify that it doesn't find
740 * the last_half_cycle.
743 ASSERT(head_blk
<= INT_MAX
);
744 if ((error
= xlog_find_verify_cycle(log
,
745 start_blk
, (int)head_blk
,
746 stop_on_cycle
, &new_blk
)))
754 * Now we need to make sure head_blk is not pointing to a block in
755 * the middle of a log record.
757 num_scan_bblks
= XLOG_REC_SHIFT(log
);
758 if (head_blk
>= num_scan_bblks
) {
759 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
761 /* start ptr at last block ptr before head_blk */
762 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
763 &head_blk
, 0)) == -1) {
764 error
= XFS_ERROR(EIO
);
770 ASSERT(head_blk
<= INT_MAX
);
771 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
772 &head_blk
, 0)) == -1) {
773 /* We hit the beginning of the log during our search */
774 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
776 ASSERT(start_blk
<= INT_MAX
&&
777 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
778 ASSERT(head_blk
<= INT_MAX
);
779 if ((error
= xlog_find_verify_log_record(log
,
781 (int)head_blk
)) == -1) {
782 error
= XFS_ERROR(EIO
);
786 if (new_blk
!= log_bbnum
)
793 if (head_blk
== log_bbnum
)
794 *return_head_blk
= 0;
796 *return_head_blk
= head_blk
;
798 * When returning here, we have a good block number. Bad block
799 * means that during a previous crash, we didn't have a clean break
800 * from cycle number N to cycle number N-1. In this case, we need
801 * to find the first block with cycle number N-1.
809 xlog_warn("XFS: failed to find log head");
814 * Find the sync block number or the tail of the log.
816 * This will be the block number of the last record to have its
817 * associated buffers synced to disk. Every log record header has
818 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
819 * to get a sync block number. The only concern is to figure out which
820 * log record header to believe.
822 * The following algorithm uses the log record header with the largest
823 * lsn. The entire log record does not need to be valid. We only care
824 * that the header is valid.
826 * We could speed up search by using current head_blk buffer, but it is not
832 xfs_daddr_t
*head_blk
,
833 xfs_daddr_t
*tail_blk
)
835 xlog_rec_header_t
*rhead
;
836 xlog_op_header_t
*op_head
;
837 xfs_caddr_t offset
= NULL
;
840 xfs_daddr_t umount_data_blk
;
841 xfs_daddr_t after_umount_blk
;
848 * Find previous log record
850 if ((error
= xlog_find_head(log
, head_blk
)))
853 bp
= xlog_get_bp(log
, 1);
856 if (*head_blk
== 0) { /* special case */
857 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
861 if (xlog_get_cycle(offset
) == 0) {
863 /* leave all other log inited values alone */
869 * Search backwards looking for log record header block
871 ASSERT(*head_blk
< INT_MAX
);
872 for (i
= (int)(*head_blk
) - 1; i
>= 0; i
--) {
873 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
877 if (XLOG_HEADER_MAGIC_NUM
== be32_to_cpu(*(__be32
*)offset
)) {
883 * If we haven't found the log record header block, start looking
884 * again from the end of the physical log. XXXmiken: There should be
885 * a check here to make sure we didn't search more than N blocks in
889 for (i
= log
->l_logBBsize
- 1; i
>= (int)(*head_blk
); i
--) {
890 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
894 if (XLOG_HEADER_MAGIC_NUM
==
895 be32_to_cpu(*(__be32
*)offset
)) {
902 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
904 return XFS_ERROR(EIO
);
907 /* find blk_no of tail of log */
908 rhead
= (xlog_rec_header_t
*)offset
;
909 *tail_blk
= BLOCK_LSN(be64_to_cpu(rhead
->h_tail_lsn
));
912 * Reset log values according to the state of the log when we
913 * crashed. In the case where head_blk == 0, we bump curr_cycle
914 * one because the next write starts a new cycle rather than
915 * continuing the cycle of the last good log record. At this
916 * point we have guaranteed that all partial log records have been
917 * accounted for. Therefore, we know that the last good log record
918 * written was complete and ended exactly on the end boundary
919 * of the physical log.
921 log
->l_prev_block
= i
;
922 log
->l_curr_block
= (int)*head_blk
;
923 log
->l_curr_cycle
= be32_to_cpu(rhead
->h_cycle
);
926 log
->l_tail_lsn
= be64_to_cpu(rhead
->h_tail_lsn
);
927 log
->l_last_sync_lsn
= be64_to_cpu(rhead
->h_lsn
);
928 log
->l_grant_reserve_cycle
= log
->l_curr_cycle
;
929 log
->l_grant_reserve_bytes
= BBTOB(log
->l_curr_block
);
930 log
->l_grant_write_cycle
= log
->l_curr_cycle
;
931 log
->l_grant_write_bytes
= BBTOB(log
->l_curr_block
);
934 * Look for unmount record. If we find it, then we know there
935 * was a clean unmount. Since 'i' could be the last block in
936 * the physical log, we convert to a log block before comparing
939 * Save the current tail lsn to use to pass to
940 * xlog_clear_stale_blocks() below. We won't want to clear the
941 * unmount record if there is one, so we pass the lsn of the
942 * unmount record rather than the block after it.
944 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
945 int h_size
= be32_to_cpu(rhead
->h_size
);
946 int h_version
= be32_to_cpu(rhead
->h_version
);
948 if ((h_version
& XLOG_VERSION_2
) &&
949 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
950 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
951 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
959 after_umount_blk
= (i
+ hblks
+ (int)
960 BTOBB(be32_to_cpu(rhead
->h_len
))) % log
->l_logBBsize
;
961 tail_lsn
= log
->l_tail_lsn
;
962 if (*head_blk
== after_umount_blk
&&
963 be32_to_cpu(rhead
->h_num_logops
) == 1) {
964 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
965 error
= xlog_bread(log
, umount_data_blk
, 1, bp
, &offset
);
969 op_head
= (xlog_op_header_t
*)offset
;
970 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
972 * Set tail and last sync so that newly written
973 * log records will point recovery to after the
974 * current unmount record.
977 xlog_assign_lsn(log
->l_curr_cycle
,
979 log
->l_last_sync_lsn
=
980 xlog_assign_lsn(log
->l_curr_cycle
,
982 *tail_blk
= after_umount_blk
;
985 * Note that the unmount was clean. If the unmount
986 * was not clean, we need to know this to rebuild the
987 * superblock counters from the perag headers if we
988 * have a filesystem using non-persistent counters.
990 log
->l_mp
->m_flags
|= XFS_MOUNT_WAS_CLEAN
;
995 * Make sure that there are no blocks in front of the head
996 * with the same cycle number as the head. This can happen
997 * because we allow multiple outstanding log writes concurrently,
998 * and the later writes might make it out before earlier ones.
1000 * We use the lsn from before modifying it so that we'll never
1001 * overwrite the unmount record after a clean unmount.
1003 * Do this only if we are going to recover the filesystem
1005 * NOTE: This used to say "if (!readonly)"
1006 * However on Linux, we can & do recover a read-only filesystem.
1007 * We only skip recovery if NORECOVERY is specified on mount,
1008 * in which case we would not be here.
1010 * But... if the -device- itself is readonly, just skip this.
1011 * We can't recover this device anyway, so it won't matter.
1013 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
))
1014 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
1020 xlog_warn("XFS: failed to locate log tail");
1025 * Is the log zeroed at all?
1027 * The last binary search should be changed to perform an X block read
1028 * once X becomes small enough. You can then search linearly through
1029 * the X blocks. This will cut down on the number of reads we need to do.
1031 * If the log is partially zeroed, this routine will pass back the blkno
1032 * of the first block with cycle number 0. It won't have a complete LR
1036 * 0 => the log is completely written to
1037 * -1 => use *blk_no as the first block of the log
1038 * >0 => error has occurred
1043 xfs_daddr_t
*blk_no
)
1047 uint first_cycle
, last_cycle
;
1048 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1049 xfs_daddr_t num_scan_bblks
;
1050 int error
, log_bbnum
= log
->l_logBBsize
;
1054 /* check totally zeroed log */
1055 bp
= xlog_get_bp(log
, 1);
1058 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1062 first_cycle
= xlog_get_cycle(offset
);
1063 if (first_cycle
== 0) { /* completely zeroed log */
1069 /* check partially zeroed log */
1070 error
= xlog_bread(log
, log_bbnum
-1, 1, bp
, &offset
);
1074 last_cycle
= xlog_get_cycle(offset
);
1075 if (last_cycle
!= 0) { /* log completely written to */
1078 } else if (first_cycle
!= 1) {
1080 * If the cycle of the last block is zero, the cycle of
1081 * the first block must be 1. If it's not, maybe we're
1082 * not looking at a log... Bail out.
1084 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1085 return XFS_ERROR(EINVAL
);
1088 /* we have a partially zeroed log */
1089 last_blk
= log_bbnum
-1;
1090 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1094 * Validate the answer. Because there is no way to guarantee that
1095 * the entire log is made up of log records which are the same size,
1096 * we scan over the defined maximum blocks. At this point, the maximum
1097 * is not chosen to mean anything special. XXXmiken
1099 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1100 ASSERT(num_scan_bblks
<= INT_MAX
);
1102 if (last_blk
< num_scan_bblks
)
1103 num_scan_bblks
= last_blk
;
1104 start_blk
= last_blk
- num_scan_bblks
;
1107 * We search for any instances of cycle number 0 that occur before
1108 * our current estimate of the head. What we're trying to detect is
1109 * 1 ... | 0 | 1 | 0...
1110 * ^ binary search ends here
1112 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1113 (int)num_scan_bblks
, 0, &new_blk
)))
1119 * Potentially backup over partial log record write. We don't need
1120 * to search the end of the log because we know it is zero.
1122 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
1123 &last_blk
, 0)) == -1) {
1124 error
= XFS_ERROR(EIO
);
1138 * These are simple subroutines used by xlog_clear_stale_blocks() below
1139 * to initialize a buffer full of empty log record headers and write
1140 * them into the log.
1151 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1153 memset(buf
, 0, BBSIZE
);
1154 recp
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1155 recp
->h_cycle
= cpu_to_be32(cycle
);
1156 recp
->h_version
= cpu_to_be32(
1157 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1158 recp
->h_lsn
= cpu_to_be64(xlog_assign_lsn(cycle
, block
));
1159 recp
->h_tail_lsn
= cpu_to_be64(xlog_assign_lsn(tail_cycle
, tail_block
));
1160 recp
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1161 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1165 xlog_write_log_records(
1176 int sectbb
= log
->l_sectBBsize
;
1177 int end_block
= start_block
+ blocks
;
1183 * Greedily allocate a buffer big enough to handle the full
1184 * range of basic blocks to be written. If that fails, try
1185 * a smaller size. We need to be able to write at least a
1186 * log sector, or we're out of luck.
1188 bufblks
= 1 << ffs(blocks
);
1189 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1191 if (bufblks
< sectbb
)
1195 /* We may need to do a read at the start to fill in part of
1196 * the buffer in the starting sector not covered by the first
1199 balign
= round_down(start_block
, sectbb
);
1200 if (balign
!= start_block
) {
1201 error
= xlog_bread_noalign(log
, start_block
, 1, bp
);
1205 j
= start_block
- balign
;
1208 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1209 int bcount
, endcount
;
1211 bcount
= min(bufblks
, end_block
- start_block
);
1212 endcount
= bcount
- j
;
1214 /* We may need to do a read at the end to fill in part of
1215 * the buffer in the final sector not covered by the write.
1216 * If this is the same sector as the above read, skip it.
1218 ealign
= round_down(end_block
, sectbb
);
1219 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1220 offset
= XFS_BUF_PTR(bp
);
1221 balign
= BBTOB(ealign
- start_block
);
1222 error
= XFS_BUF_SET_PTR(bp
, offset
+ balign
,
1227 error
= xlog_bread_noalign(log
, ealign
, sectbb
, bp
);
1231 error
= XFS_BUF_SET_PTR(bp
, offset
, bufblks
);
1236 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1237 for (; j
< endcount
; j
++) {
1238 xlog_add_record(log
, offset
, cycle
, i
+j
,
1239 tail_cycle
, tail_block
);
1242 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1245 start_block
+= endcount
;
1255 * This routine is called to blow away any incomplete log writes out
1256 * in front of the log head. We do this so that we won't become confused
1257 * if we come up, write only a little bit more, and then crash again.
1258 * If we leave the partial log records out there, this situation could
1259 * cause us to think those partial writes are valid blocks since they
1260 * have the current cycle number. We get rid of them by overwriting them
1261 * with empty log records with the old cycle number rather than the
1264 * The tail lsn is passed in rather than taken from
1265 * the log so that we will not write over the unmount record after a
1266 * clean unmount in a 512 block log. Doing so would leave the log without
1267 * any valid log records in it until a new one was written. If we crashed
1268 * during that time we would not be able to recover.
1271 xlog_clear_stale_blocks(
1275 int tail_cycle
, head_cycle
;
1276 int tail_block
, head_block
;
1277 int tail_distance
, max_distance
;
1281 tail_cycle
= CYCLE_LSN(tail_lsn
);
1282 tail_block
= BLOCK_LSN(tail_lsn
);
1283 head_cycle
= log
->l_curr_cycle
;
1284 head_block
= log
->l_curr_block
;
1287 * Figure out the distance between the new head of the log
1288 * and the tail. We want to write over any blocks beyond the
1289 * head that we may have written just before the crash, but
1290 * we don't want to overwrite the tail of the log.
1292 if (head_cycle
== tail_cycle
) {
1294 * The tail is behind the head in the physical log,
1295 * so the distance from the head to the tail is the
1296 * distance from the head to the end of the log plus
1297 * the distance from the beginning of the log to the
1300 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1301 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1302 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1303 return XFS_ERROR(EFSCORRUPTED
);
1305 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1308 * The head is behind the tail in the physical log,
1309 * so the distance from the head to the tail is just
1310 * the tail block minus the head block.
1312 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1313 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1314 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1315 return XFS_ERROR(EFSCORRUPTED
);
1317 tail_distance
= tail_block
- head_block
;
1321 * If the head is right up against the tail, we can't clear
1324 if (tail_distance
<= 0) {
1325 ASSERT(tail_distance
== 0);
1329 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1331 * Take the smaller of the maximum amount of outstanding I/O
1332 * we could have and the distance to the tail to clear out.
1333 * We take the smaller so that we don't overwrite the tail and
1334 * we don't waste all day writing from the head to the tail
1337 max_distance
= MIN(max_distance
, tail_distance
);
1339 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1341 * We can stomp all the blocks we need to without
1342 * wrapping around the end of the log. Just do it
1343 * in a single write. Use the cycle number of the
1344 * current cycle minus one so that the log will look like:
1347 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1348 head_block
, max_distance
, tail_cycle
,
1354 * We need to wrap around the end of the physical log in
1355 * order to clear all the blocks. Do it in two separate
1356 * I/Os. The first write should be from the head to the
1357 * end of the physical log, and it should use the current
1358 * cycle number minus one just like above.
1360 distance
= log
->l_logBBsize
- head_block
;
1361 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1362 head_block
, distance
, tail_cycle
,
1369 * Now write the blocks at the start of the physical log.
1370 * This writes the remainder of the blocks we want to clear.
1371 * It uses the current cycle number since we're now on the
1372 * same cycle as the head so that we get:
1373 * n ... n ... | n - 1 ...
1374 * ^^^^^ blocks we're writing
1376 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1377 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1378 tail_cycle
, tail_block
);
1386 /******************************************************************************
1388 * Log recover routines
1390 ******************************************************************************
1393 STATIC xlog_recover_t
*
1394 xlog_recover_find_tid(
1395 struct hlist_head
*head
,
1398 xlog_recover_t
*trans
;
1399 struct hlist_node
*n
;
1401 hlist_for_each_entry(trans
, n
, head
, r_list
) {
1402 if (trans
->r_log_tid
== tid
)
1409 xlog_recover_new_tid(
1410 struct hlist_head
*head
,
1414 xlog_recover_t
*trans
;
1416 trans
= kmem_zalloc(sizeof(xlog_recover_t
), KM_SLEEP
);
1417 trans
->r_log_tid
= tid
;
1419 INIT_LIST_HEAD(&trans
->r_itemq
);
1421 INIT_HLIST_NODE(&trans
->r_list
);
1422 hlist_add_head(&trans
->r_list
, head
);
1426 xlog_recover_add_item(
1427 struct list_head
*head
)
1429 xlog_recover_item_t
*item
;
1431 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
1432 INIT_LIST_HEAD(&item
->ri_list
);
1433 list_add_tail(&item
->ri_list
, head
);
1437 xlog_recover_add_to_cont_trans(
1439 xlog_recover_t
*trans
,
1443 xlog_recover_item_t
*item
;
1444 xfs_caddr_t ptr
, old_ptr
;
1447 if (list_empty(&trans
->r_itemq
)) {
1448 /* finish copying rest of trans header */
1449 xlog_recover_add_item(&trans
->r_itemq
);
1450 ptr
= (xfs_caddr_t
) &trans
->r_theader
+
1451 sizeof(xfs_trans_header_t
) - len
;
1452 memcpy(ptr
, dp
, len
); /* d, s, l */
1455 /* take the tail entry */
1456 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1458 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
1459 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
1461 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, 0u);
1462 memcpy(&ptr
[old_len
], dp
, len
); /* d, s, l */
1463 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
1464 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
1465 trace_xfs_log_recover_item_add_cont(log
, trans
, item
, 0);
1470 * The next region to add is the start of a new region. It could be
1471 * a whole region or it could be the first part of a new region. Because
1472 * of this, the assumption here is that the type and size fields of all
1473 * format structures fit into the first 32 bits of the structure.
1475 * This works because all regions must be 32 bit aligned. Therefore, we
1476 * either have both fields or we have neither field. In the case we have
1477 * neither field, the data part of the region is zero length. We only have
1478 * a log_op_header and can throw away the header since a new one will appear
1479 * later. If we have at least 4 bytes, then we can determine how many regions
1480 * will appear in the current log item.
1483 xlog_recover_add_to_trans(
1485 xlog_recover_t
*trans
,
1489 xfs_inode_log_format_t
*in_f
; /* any will do */
1490 xlog_recover_item_t
*item
;
1495 if (list_empty(&trans
->r_itemq
)) {
1496 /* we need to catch log corruptions here */
1497 if (*(uint
*)dp
!= XFS_TRANS_HEADER_MAGIC
) {
1498 xlog_warn("XFS: xlog_recover_add_to_trans: "
1499 "bad header magic number");
1501 return XFS_ERROR(EIO
);
1503 if (len
== sizeof(xfs_trans_header_t
))
1504 xlog_recover_add_item(&trans
->r_itemq
);
1505 memcpy(&trans
->r_theader
, dp
, len
); /* d, s, l */
1509 ptr
= kmem_alloc(len
, KM_SLEEP
);
1510 memcpy(ptr
, dp
, len
);
1511 in_f
= (xfs_inode_log_format_t
*)ptr
;
1513 /* take the tail entry */
1514 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1515 if (item
->ri_total
!= 0 &&
1516 item
->ri_total
== item
->ri_cnt
) {
1517 /* tail item is in use, get a new one */
1518 xlog_recover_add_item(&trans
->r_itemq
);
1519 item
= list_entry(trans
->r_itemq
.prev
,
1520 xlog_recover_item_t
, ri_list
);
1523 if (item
->ri_total
== 0) { /* first region to be added */
1524 if (in_f
->ilf_size
== 0 ||
1525 in_f
->ilf_size
> XLOG_MAX_REGIONS_IN_ITEM
) {
1527 "XFS: bad number of regions (%d) in inode log format",
1530 return XFS_ERROR(EIO
);
1533 item
->ri_total
= in_f
->ilf_size
;
1535 kmem_zalloc(item
->ri_total
* sizeof(xfs_log_iovec_t
),
1538 ASSERT(item
->ri_total
> item
->ri_cnt
);
1539 /* Description region is ri_buf[0] */
1540 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
1541 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
1543 trace_xfs_log_recover_item_add(log
, trans
, item
, 0);
1548 * Sort the log items in the transaction. Cancelled buffers need
1549 * to be put first so they are processed before any items that might
1550 * modify the buffers. If they are cancelled, then the modifications
1551 * don't need to be replayed.
1554 xlog_recover_reorder_trans(
1556 xlog_recover_t
*trans
,
1559 xlog_recover_item_t
*item
, *n
;
1560 LIST_HEAD(sort_list
);
1562 list_splice_init(&trans
->r_itemq
, &sort_list
);
1563 list_for_each_entry_safe(item
, n
, &sort_list
, ri_list
) {
1564 xfs_buf_log_format_t
*buf_f
;
1566 buf_f
= (xfs_buf_log_format_t
*)item
->ri_buf
[0].i_addr
;
1568 switch (ITEM_TYPE(item
)) {
1570 if (!(buf_f
->blf_flags
& XFS_BLF_CANCEL
)) {
1571 trace_xfs_log_recover_item_reorder_head(log
,
1573 list_move(&item
->ri_list
, &trans
->r_itemq
);
1578 case XFS_LI_QUOTAOFF
:
1581 trace_xfs_log_recover_item_reorder_tail(log
,
1583 list_move_tail(&item
->ri_list
, &trans
->r_itemq
);
1587 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1589 return XFS_ERROR(EIO
);
1592 ASSERT(list_empty(&sort_list
));
1597 * Build up the table of buf cancel records so that we don't replay
1598 * cancelled data in the second pass. For buffer records that are
1599 * not cancel records, there is nothing to do here so we just return.
1601 * If we get a cancel record which is already in the table, this indicates
1602 * that the buffer was cancelled multiple times. In order to ensure
1603 * that during pass 2 we keep the record in the table until we reach its
1604 * last occurrence in the log, we keep a reference count in the cancel
1605 * record in the table to tell us how many times we expect to see this
1606 * record during the second pass.
1609 xlog_recover_do_buffer_pass1(
1611 xfs_buf_log_format_t
*buf_f
)
1613 xfs_buf_cancel_t
*bcp
;
1614 xfs_buf_cancel_t
*nextp
;
1615 xfs_buf_cancel_t
*prevp
;
1616 xfs_buf_cancel_t
**bucket
;
1617 xfs_daddr_t blkno
= 0;
1621 switch (buf_f
->blf_type
) {
1623 blkno
= buf_f
->blf_blkno
;
1624 len
= buf_f
->blf_len
;
1625 flags
= buf_f
->blf_flags
;
1630 * If this isn't a cancel buffer item, then just return.
1632 if (!(flags
& XFS_BLF_CANCEL
)) {
1633 trace_xfs_log_recover_buf_not_cancel(log
, buf_f
);
1638 * Insert an xfs_buf_cancel record into the hash table of
1639 * them. If there is already an identical record, bump
1640 * its reference count.
1642 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1643 XLOG_BC_TABLE_SIZE
];
1645 * If the hash bucket is empty then just insert a new record into
1648 if (*bucket
== NULL
) {
1649 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1651 bcp
->bc_blkno
= blkno
;
1653 bcp
->bc_refcount
= 1;
1654 bcp
->bc_next
= NULL
;
1660 * The hash bucket is not empty, so search for duplicates of our
1661 * record. If we find one them just bump its refcount. If not
1662 * then add us at the end of the list.
1666 while (nextp
!= NULL
) {
1667 if (nextp
->bc_blkno
== blkno
&& nextp
->bc_len
== len
) {
1668 nextp
->bc_refcount
++;
1669 trace_xfs_log_recover_buf_cancel_ref_inc(log
, buf_f
);
1673 nextp
= nextp
->bc_next
;
1675 ASSERT(prevp
!= NULL
);
1676 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1678 bcp
->bc_blkno
= blkno
;
1680 bcp
->bc_refcount
= 1;
1681 bcp
->bc_next
= NULL
;
1682 prevp
->bc_next
= bcp
;
1683 trace_xfs_log_recover_buf_cancel_add(log
, buf_f
);
1687 * Check to see whether the buffer being recovered has a corresponding
1688 * entry in the buffer cancel record table. If it does then return 1
1689 * so that it will be cancelled, otherwise return 0. If the buffer is
1690 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
1691 * the refcount on the entry in the table and remove it from the table
1692 * if this is the last reference.
1694 * We remove the cancel record from the table when we encounter its
1695 * last occurrence in the log so that if the same buffer is re-used
1696 * again after its last cancellation we actually replay the changes
1697 * made at that point.
1700 xlog_check_buffer_cancelled(
1706 xfs_buf_cancel_t
*bcp
;
1707 xfs_buf_cancel_t
*prevp
;
1708 xfs_buf_cancel_t
**bucket
;
1710 if (log
->l_buf_cancel_table
== NULL
) {
1712 * There is nothing in the table built in pass one,
1713 * so this buffer must not be cancelled.
1715 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1719 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1720 XLOG_BC_TABLE_SIZE
];
1724 * There is no corresponding entry in the table built
1725 * in pass one, so this buffer has not been cancelled.
1727 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1732 * Search for an entry in the buffer cancel table that
1733 * matches our buffer.
1736 while (bcp
!= NULL
) {
1737 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
) {
1739 * We've go a match, so return 1 so that the
1740 * recovery of this buffer is cancelled.
1741 * If this buffer is actually a buffer cancel
1742 * log item, then decrement the refcount on the
1743 * one in the table and remove it if this is the
1746 if (flags
& XFS_BLF_CANCEL
) {
1748 if (bcp
->bc_refcount
== 0) {
1749 if (prevp
== NULL
) {
1750 *bucket
= bcp
->bc_next
;
1752 prevp
->bc_next
= bcp
->bc_next
;
1763 * We didn't find a corresponding entry in the table, so
1764 * return 0 so that the buffer is NOT cancelled.
1766 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1771 xlog_recover_do_buffer_pass2(
1773 xfs_buf_log_format_t
*buf_f
)
1775 xfs_daddr_t blkno
= 0;
1779 switch (buf_f
->blf_type
) {
1781 blkno
= buf_f
->blf_blkno
;
1782 flags
= buf_f
->blf_flags
;
1783 len
= buf_f
->blf_len
;
1787 return xlog_check_buffer_cancelled(log
, blkno
, len
, flags
);
1791 * Perform recovery for a buffer full of inodes. In these buffers,
1792 * the only data which should be recovered is that which corresponds
1793 * to the di_next_unlinked pointers in the on disk inode structures.
1794 * The rest of the data for the inodes is always logged through the
1795 * inodes themselves rather than the inode buffer and is recovered
1796 * in xlog_recover_do_inode_trans().
1798 * The only time when buffers full of inodes are fully recovered is
1799 * when the buffer is full of newly allocated inodes. In this case
1800 * the buffer will not be marked as an inode buffer and so will be
1801 * sent to xlog_recover_do_reg_buffer() below during recovery.
1804 xlog_recover_do_inode_buffer(
1806 xlog_recover_item_t
*item
,
1808 xfs_buf_log_format_t
*buf_f
)
1816 int next_unlinked_offset
;
1818 xfs_agino_t
*logged_nextp
;
1819 xfs_agino_t
*buffer_nextp
;
1820 unsigned int *data_map
= NULL
;
1821 unsigned int map_size
= 0;
1823 trace_xfs_log_recover_buf_inode_buf(mp
->m_log
, buf_f
);
1825 switch (buf_f
->blf_type
) {
1827 data_map
= buf_f
->blf_data_map
;
1828 map_size
= buf_f
->blf_map_size
;
1832 * Set the variables corresponding to the current region to
1833 * 0 so that we'll initialize them on the first pass through
1841 inodes_per_buf
= XFS_BUF_COUNT(bp
) >> mp
->m_sb
.sb_inodelog
;
1842 for (i
= 0; i
< inodes_per_buf
; i
++) {
1843 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1844 offsetof(xfs_dinode_t
, di_next_unlinked
);
1846 while (next_unlinked_offset
>=
1847 (reg_buf_offset
+ reg_buf_bytes
)) {
1849 * The next di_next_unlinked field is beyond
1850 * the current logged region. Find the next
1851 * logged region that contains or is beyond
1852 * the current di_next_unlinked field.
1855 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1858 * If there are no more logged regions in the
1859 * buffer, then we're done.
1865 nbits
= xfs_contig_bits(data_map
, map_size
,
1868 reg_buf_offset
= bit
<< XFS_BLF_SHIFT
;
1869 reg_buf_bytes
= nbits
<< XFS_BLF_SHIFT
;
1874 * If the current logged region starts after the current
1875 * di_next_unlinked field, then move on to the next
1876 * di_next_unlinked field.
1878 if (next_unlinked_offset
< reg_buf_offset
) {
1882 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1883 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLF_CHUNK
) == 0);
1884 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <= XFS_BUF_COUNT(bp
));
1887 * The current logged region contains a copy of the
1888 * current di_next_unlinked field. Extract its value
1889 * and copy it to the buffer copy.
1891 logged_nextp
= (xfs_agino_t
*)
1892 ((char *)(item
->ri_buf
[item_index
].i_addr
) +
1893 (next_unlinked_offset
- reg_buf_offset
));
1894 if (unlikely(*logged_nextp
== 0)) {
1895 xfs_fs_cmn_err(CE_ALERT
, mp
,
1896 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1898 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1899 XFS_ERRLEVEL_LOW
, mp
);
1900 return XFS_ERROR(EFSCORRUPTED
);
1903 buffer_nextp
= (xfs_agino_t
*)xfs_buf_offset(bp
,
1904 next_unlinked_offset
);
1905 *buffer_nextp
= *logged_nextp
;
1912 * Perform a 'normal' buffer recovery. Each logged region of the
1913 * buffer should be copied over the corresponding region in the
1914 * given buffer. The bitmap in the buf log format structure indicates
1915 * where to place the logged data.
1919 xlog_recover_do_reg_buffer(
1920 struct xfs_mount
*mp
,
1921 xlog_recover_item_t
*item
,
1923 xfs_buf_log_format_t
*buf_f
)
1928 unsigned int *data_map
= NULL
;
1929 unsigned int map_size
= 0;
1932 trace_xfs_log_recover_buf_reg_buf(mp
->m_log
, buf_f
);
1934 switch (buf_f
->blf_type
) {
1936 data_map
= buf_f
->blf_data_map
;
1937 map_size
= buf_f
->blf_map_size
;
1941 i
= 1; /* 0 is the buf format structure */
1943 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1946 nbits
= xfs_contig_bits(data_map
, map_size
, bit
);
1948 ASSERT(item
->ri_buf
[i
].i_addr
!= NULL
);
1949 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLF_CHUNK
== 0);
1950 ASSERT(XFS_BUF_COUNT(bp
) >=
1951 ((uint
)bit
<< XFS_BLF_SHIFT
)+(nbits
<<XFS_BLF_SHIFT
));
1954 * Do a sanity check if this is a dquot buffer. Just checking
1955 * the first dquot in the buffer should do. XXXThis is
1956 * probably a good thing to do for other buf types also.
1959 if (buf_f
->blf_flags
&
1960 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
1961 if (item
->ri_buf
[i
].i_addr
== NULL
) {
1963 "XFS: NULL dquot in %s.", __func__
);
1966 if (item
->ri_buf
[i
].i_len
< sizeof(xfs_disk_dquot_t
)) {
1968 "XFS: dquot too small (%d) in %s.",
1969 item
->ri_buf
[i
].i_len
, __func__
);
1972 error
= xfs_qm_dqcheck((xfs_disk_dquot_t
*)
1973 item
->ri_buf
[i
].i_addr
,
1974 -1, 0, XFS_QMOPT_DOWARN
,
1975 "dquot_buf_recover");
1980 memcpy(xfs_buf_offset(bp
,
1981 (uint
)bit
<< XFS_BLF_SHIFT
), /* dest */
1982 item
->ri_buf
[i
].i_addr
, /* source */
1983 nbits
<<XFS_BLF_SHIFT
); /* length */
1989 /* Shouldn't be any more regions */
1990 ASSERT(i
== item
->ri_total
);
1994 * Do some primitive error checking on ondisk dquot data structures.
1998 xfs_disk_dquot_t
*ddq
,
2000 uint type
, /* used only when IO_dorepair is true */
2004 xfs_dqblk_t
*d
= (xfs_dqblk_t
*)ddq
;
2008 * We can encounter an uninitialized dquot buffer for 2 reasons:
2009 * 1. If we crash while deleting the quotainode(s), and those blks got
2010 * used for user data. This is because we take the path of regular
2011 * file deletion; however, the size field of quotainodes is never
2012 * updated, so all the tricks that we play in itruncate_finish
2013 * don't quite matter.
2015 * 2. We don't play the quota buffers when there's a quotaoff logitem.
2016 * But the allocation will be replayed so we'll end up with an
2017 * uninitialized quota block.
2019 * This is all fine; things are still consistent, and we haven't lost
2020 * any quota information. Just don't complain about bad dquot blks.
2022 if (be16_to_cpu(ddq
->d_magic
) != XFS_DQUOT_MAGIC
) {
2023 if (flags
& XFS_QMOPT_DOWARN
)
2025 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
2026 str
, id
, be16_to_cpu(ddq
->d_magic
), XFS_DQUOT_MAGIC
);
2029 if (ddq
->d_version
!= XFS_DQUOT_VERSION
) {
2030 if (flags
& XFS_QMOPT_DOWARN
)
2032 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2033 str
, id
, ddq
->d_version
, XFS_DQUOT_VERSION
);
2037 if (ddq
->d_flags
!= XFS_DQ_USER
&&
2038 ddq
->d_flags
!= XFS_DQ_PROJ
&&
2039 ddq
->d_flags
!= XFS_DQ_GROUP
) {
2040 if (flags
& XFS_QMOPT_DOWARN
)
2042 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2043 str
, id
, ddq
->d_flags
);
2047 if (id
!= -1 && id
!= be32_to_cpu(ddq
->d_id
)) {
2048 if (flags
& XFS_QMOPT_DOWARN
)
2050 "%s : ondisk-dquot 0x%p, ID mismatch: "
2051 "0x%x expected, found id 0x%x",
2052 str
, ddq
, id
, be32_to_cpu(ddq
->d_id
));
2056 if (!errs
&& ddq
->d_id
) {
2057 if (ddq
->d_blk_softlimit
&&
2058 be64_to_cpu(ddq
->d_bcount
) >=
2059 be64_to_cpu(ddq
->d_blk_softlimit
)) {
2060 if (!ddq
->d_btimer
) {
2061 if (flags
& XFS_QMOPT_DOWARN
)
2063 "%s : Dquot ID 0x%x (0x%p) "
2064 "BLK TIMER NOT STARTED",
2065 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2069 if (ddq
->d_ino_softlimit
&&
2070 be64_to_cpu(ddq
->d_icount
) >=
2071 be64_to_cpu(ddq
->d_ino_softlimit
)) {
2072 if (!ddq
->d_itimer
) {
2073 if (flags
& XFS_QMOPT_DOWARN
)
2075 "%s : Dquot ID 0x%x (0x%p) "
2076 "INODE TIMER NOT STARTED",
2077 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2081 if (ddq
->d_rtb_softlimit
&&
2082 be64_to_cpu(ddq
->d_rtbcount
) >=
2083 be64_to_cpu(ddq
->d_rtb_softlimit
)) {
2084 if (!ddq
->d_rtbtimer
) {
2085 if (flags
& XFS_QMOPT_DOWARN
)
2087 "%s : Dquot ID 0x%x (0x%p) "
2088 "RTBLK TIMER NOT STARTED",
2089 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2095 if (!errs
|| !(flags
& XFS_QMOPT_DQREPAIR
))
2098 if (flags
& XFS_QMOPT_DOWARN
)
2099 cmn_err(CE_NOTE
, "Re-initializing dquot ID 0x%x", id
);
2102 * Typically, a repair is only requested by quotacheck.
2105 ASSERT(flags
& XFS_QMOPT_DQREPAIR
);
2106 memset(d
, 0, sizeof(xfs_dqblk_t
));
2108 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
2109 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
2110 d
->dd_diskdq
.d_flags
= type
;
2111 d
->dd_diskdq
.d_id
= cpu_to_be32(id
);
2117 * Perform a dquot buffer recovery.
2118 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2119 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2120 * Else, treat it as a regular buffer and do recovery.
2123 xlog_recover_do_dquot_buffer(
2126 xlog_recover_item_t
*item
,
2128 xfs_buf_log_format_t
*buf_f
)
2132 trace_xfs_log_recover_buf_dquot_buf(log
, buf_f
);
2135 * Filesystems are required to send in quota flags at mount time.
2137 if (mp
->m_qflags
== 0) {
2142 if (buf_f
->blf_flags
& XFS_BLF_UDQUOT_BUF
)
2143 type
|= XFS_DQ_USER
;
2144 if (buf_f
->blf_flags
& XFS_BLF_PDQUOT_BUF
)
2145 type
|= XFS_DQ_PROJ
;
2146 if (buf_f
->blf_flags
& XFS_BLF_GDQUOT_BUF
)
2147 type
|= XFS_DQ_GROUP
;
2149 * This type of quotas was turned off, so ignore this buffer
2151 if (log
->l_quotaoffs_flag
& type
)
2154 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2158 * This routine replays a modification made to a buffer at runtime.
2159 * There are actually two types of buffer, regular and inode, which
2160 * are handled differently. Inode buffers are handled differently
2161 * in that we only recover a specific set of data from them, namely
2162 * the inode di_next_unlinked fields. This is because all other inode
2163 * data is actually logged via inode records and any data we replay
2164 * here which overlaps that may be stale.
2166 * When meta-data buffers are freed at run time we log a buffer item
2167 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2168 * of the buffer in the log should not be replayed at recovery time.
2169 * This is so that if the blocks covered by the buffer are reused for
2170 * file data before we crash we don't end up replaying old, freed
2171 * meta-data into a user's file.
2173 * To handle the cancellation of buffer log items, we make two passes
2174 * over the log during recovery. During the first we build a table of
2175 * those buffers which have been cancelled, and during the second we
2176 * only replay those buffers which do not have corresponding cancel
2177 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2178 * for more details on the implementation of the table of cancel records.
2181 xlog_recover_do_buffer_trans(
2183 xlog_recover_item_t
*item
,
2186 xfs_buf_log_format_t
*buf_f
;
2196 buf_f
= (xfs_buf_log_format_t
*)item
->ri_buf
[0].i_addr
;
2198 if (pass
== XLOG_RECOVER_PASS1
) {
2200 * In this pass we're only looking for buf items
2201 * with the XFS_BLF_CANCEL bit set.
2203 xlog_recover_do_buffer_pass1(log
, buf_f
);
2207 * In this pass we want to recover all the buffers
2208 * which have not been cancelled and are not
2209 * cancellation buffers themselves. The routine
2210 * we call here will tell us whether or not to
2211 * continue with the replay of this buffer.
2213 cancel
= xlog_recover_do_buffer_pass2(log
, buf_f
);
2215 trace_xfs_log_recover_buf_cancel(log
, buf_f
);
2219 trace_xfs_log_recover_buf_recover(log
, buf_f
);
2220 switch (buf_f
->blf_type
) {
2222 blkno
= buf_f
->blf_blkno
;
2223 len
= buf_f
->blf_len
;
2224 flags
= buf_f
->blf_flags
;
2227 xfs_fs_cmn_err(CE_ALERT
, log
->l_mp
,
2228 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2229 buf_f
->blf_type
, log
->l_mp
->m_logname
?
2230 log
->l_mp
->m_logname
: "internal");
2231 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2232 XFS_ERRLEVEL_LOW
, log
->l_mp
);
2233 return XFS_ERROR(EFSCORRUPTED
);
2237 buf_flags
= XBF_LOCK
;
2238 if (!(flags
& XFS_BLF_INODE_BUF
))
2239 buf_flags
|= XBF_MAPPED
;
2241 bp
= xfs_buf_read(mp
->m_ddev_targp
, blkno
, len
, buf_flags
);
2242 if (XFS_BUF_ISERROR(bp
)) {
2243 xfs_ioerror_alert("xlog_recover_do..(read#1)", log
->l_mp
,
2245 error
= XFS_BUF_GETERROR(bp
);
2251 if (flags
& XFS_BLF_INODE_BUF
) {
2252 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2254 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2255 xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2257 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2260 return XFS_ERROR(error
);
2263 * Perform delayed write on the buffer. Asynchronous writes will be
2264 * slower when taking into account all the buffers to be flushed.
2266 * Also make sure that only inode buffers with good sizes stay in
2267 * the buffer cache. The kernel moves inodes in buffers of 1 block
2268 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2269 * buffers in the log can be a different size if the log was generated
2270 * by an older kernel using unclustered inode buffers or a newer kernel
2271 * running with a different inode cluster size. Regardless, if the
2272 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2273 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2274 * the buffer out of the buffer cache so that the buffer won't
2275 * overlap with future reads of those inodes.
2277 if (XFS_DINODE_MAGIC
==
2278 be16_to_cpu(*((__be16
*)xfs_buf_offset(bp
, 0))) &&
2279 (XFS_BUF_COUNT(bp
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2280 (__uint32_t
)XFS_INODE_CLUSTER_SIZE(log
->l_mp
)))) {
2282 error
= xfs_bwrite(mp
, bp
);
2284 ASSERT(bp
->b_mount
== NULL
|| bp
->b_mount
== mp
);
2286 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2287 xfs_bdwrite(mp
, bp
);
2294 xlog_recover_do_inode_trans(
2296 xlog_recover_item_t
*item
,
2299 xfs_inode_log_format_t
*in_f
;
2310 xfs_icdinode_t
*dicp
;
2313 if (pass
== XLOG_RECOVER_PASS1
) {
2317 if (item
->ri_buf
[0].i_len
== sizeof(xfs_inode_log_format_t
)) {
2318 in_f
= (xfs_inode_log_format_t
*)item
->ri_buf
[0].i_addr
;
2320 in_f
= (xfs_inode_log_format_t
*)kmem_alloc(
2321 sizeof(xfs_inode_log_format_t
), KM_SLEEP
);
2323 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
2327 ino
= in_f
->ilf_ino
;
2331 * Inode buffers can be freed, look out for it,
2332 * and do not replay the inode.
2334 if (xlog_check_buffer_cancelled(log
, in_f
->ilf_blkno
,
2335 in_f
->ilf_len
, 0)) {
2337 trace_xfs_log_recover_inode_cancel(log
, in_f
);
2340 trace_xfs_log_recover_inode_recover(log
, in_f
);
2342 bp
= xfs_buf_read(mp
->m_ddev_targp
, in_f
->ilf_blkno
, in_f
->ilf_len
,
2344 if (XFS_BUF_ISERROR(bp
)) {
2345 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp
,
2346 bp
, in_f
->ilf_blkno
);
2347 error
= XFS_BUF_GETERROR(bp
);
2352 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2353 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, in_f
->ilf_boffset
);
2356 * Make sure the place we're flushing out to really looks
2359 if (unlikely(be16_to_cpu(dip
->di_magic
) != XFS_DINODE_MAGIC
)) {
2361 xfs_fs_cmn_err(CE_ALERT
, mp
,
2362 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2364 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2365 XFS_ERRLEVEL_LOW
, mp
);
2366 error
= EFSCORRUPTED
;
2369 dicp
= (xfs_icdinode_t
*)(item
->ri_buf
[1].i_addr
);
2370 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2372 xfs_fs_cmn_err(CE_ALERT
, mp
,
2373 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2375 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2376 XFS_ERRLEVEL_LOW
, mp
);
2377 error
= EFSCORRUPTED
;
2381 /* Skip replay when the on disk inode is newer than the log one */
2382 if (dicp
->di_flushiter
< be16_to_cpu(dip
->di_flushiter
)) {
2384 * Deal with the wrap case, DI_MAX_FLUSH is less
2385 * than smaller numbers
2387 if (be16_to_cpu(dip
->di_flushiter
) == DI_MAX_FLUSH
&&
2388 dicp
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
2392 trace_xfs_log_recover_inode_skip(log
, in_f
);
2397 /* Take the opportunity to reset the flush iteration count */
2398 dicp
->di_flushiter
= 0;
2400 if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFREG
)) {
2401 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2402 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2403 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2404 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2406 xfs_fs_cmn_err(CE_ALERT
, mp
,
2407 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2408 item
, dip
, bp
, ino
);
2409 error
= EFSCORRUPTED
;
2412 } else if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFDIR
)) {
2413 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2414 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2415 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2416 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2417 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2419 xfs_fs_cmn_err(CE_ALERT
, mp
,
2420 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2421 item
, dip
, bp
, ino
);
2422 error
= EFSCORRUPTED
;
2426 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2427 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2428 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2430 xfs_fs_cmn_err(CE_ALERT
, mp
,
2431 "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",
2433 dicp
->di_nextents
+ dicp
->di_anextents
,
2435 error
= EFSCORRUPTED
;
2438 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2439 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2440 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2442 xfs_fs_cmn_err(CE_ALERT
, mp
,
2443 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2444 item
, dip
, bp
, ino
, dicp
->di_forkoff
);
2445 error
= EFSCORRUPTED
;
2448 if (unlikely(item
->ri_buf
[1].i_len
> sizeof(struct xfs_icdinode
))) {
2449 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2450 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2452 xfs_fs_cmn_err(CE_ALERT
, mp
,
2453 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2454 item
->ri_buf
[1].i_len
, item
);
2455 error
= EFSCORRUPTED
;
2459 /* The core is in in-core format */
2460 xfs_dinode_to_disk(dip
, (xfs_icdinode_t
*)item
->ri_buf
[1].i_addr
);
2462 /* the rest is in on-disk format */
2463 if (item
->ri_buf
[1].i_len
> sizeof(struct xfs_icdinode
)) {
2464 memcpy((xfs_caddr_t
) dip
+ sizeof(struct xfs_icdinode
),
2465 item
->ri_buf
[1].i_addr
+ sizeof(struct xfs_icdinode
),
2466 item
->ri_buf
[1].i_len
- sizeof(struct xfs_icdinode
));
2469 fields
= in_f
->ilf_fields
;
2470 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
2472 xfs_dinode_put_rdev(dip
, in_f
->ilf_u
.ilfu_rdev
);
2475 memcpy(XFS_DFORK_DPTR(dip
),
2476 &in_f
->ilf_u
.ilfu_uuid
,
2481 if (in_f
->ilf_size
== 2)
2482 goto write_inode_buffer
;
2483 len
= item
->ri_buf
[2].i_len
;
2484 src
= item
->ri_buf
[2].i_addr
;
2485 ASSERT(in_f
->ilf_size
<= 4);
2486 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
2487 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
2488 (len
== in_f
->ilf_dsize
));
2490 switch (fields
& XFS_ILOG_DFORK
) {
2491 case XFS_ILOG_DDATA
:
2493 memcpy(XFS_DFORK_DPTR(dip
), src
, len
);
2496 case XFS_ILOG_DBROOT
:
2497 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
, len
,
2498 (xfs_bmdr_block_t
*)XFS_DFORK_DPTR(dip
),
2499 XFS_DFORK_DSIZE(dip
, mp
));
2504 * There are no data fork flags set.
2506 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
2511 * If we logged any attribute data, recover it. There may or
2512 * may not have been any other non-core data logged in this
2515 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2516 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
2521 len
= item
->ri_buf
[attr_index
].i_len
;
2522 src
= item
->ri_buf
[attr_index
].i_addr
;
2523 ASSERT(len
== in_f
->ilf_asize
);
2525 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2526 case XFS_ILOG_ADATA
:
2528 dest
= XFS_DFORK_APTR(dip
);
2529 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
2530 memcpy(dest
, src
, len
);
2533 case XFS_ILOG_ABROOT
:
2534 dest
= XFS_DFORK_APTR(dip
);
2535 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
,
2536 len
, (xfs_bmdr_block_t
*)dest
,
2537 XFS_DFORK_ASIZE(dip
, mp
));
2541 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2550 ASSERT(bp
->b_mount
== NULL
|| bp
->b_mount
== mp
);
2552 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2553 xfs_bdwrite(mp
, bp
);
2557 return XFS_ERROR(error
);
2561 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2562 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2566 xlog_recover_do_quotaoff_trans(
2568 xlog_recover_item_t
*item
,
2571 xfs_qoff_logformat_t
*qoff_f
;
2573 if (pass
== XLOG_RECOVER_PASS2
) {
2577 qoff_f
= (xfs_qoff_logformat_t
*)item
->ri_buf
[0].i_addr
;
2581 * The logitem format's flag tells us if this was user quotaoff,
2582 * group/project quotaoff or both.
2584 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2585 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2586 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
2587 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
2588 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2589 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2595 * Recover a dquot record
2598 xlog_recover_do_dquot_trans(
2600 xlog_recover_item_t
*item
,
2605 struct xfs_disk_dquot
*ddq
, *recddq
;
2607 xfs_dq_logformat_t
*dq_f
;
2610 if (pass
== XLOG_RECOVER_PASS1
) {
2616 * Filesystems are required to send in quota flags at mount time.
2618 if (mp
->m_qflags
== 0)
2621 recddq
= (xfs_disk_dquot_t
*)item
->ri_buf
[1].i_addr
;
2623 if (item
->ri_buf
[1].i_addr
== NULL
) {
2625 "XFS: NULL dquot in %s.", __func__
);
2626 return XFS_ERROR(EIO
);
2628 if (item
->ri_buf
[1].i_len
< sizeof(xfs_disk_dquot_t
)) {
2630 "XFS: dquot too small (%d) in %s.",
2631 item
->ri_buf
[1].i_len
, __func__
);
2632 return XFS_ERROR(EIO
);
2636 * This type of quotas was turned off, so ignore this record.
2638 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
2640 if (log
->l_quotaoffs_flag
& type
)
2644 * At this point we know that quota was _not_ turned off.
2645 * Since the mount flags are not indicating to us otherwise, this
2646 * must mean that quota is on, and the dquot needs to be replayed.
2647 * Remember that we may not have fully recovered the superblock yet,
2648 * so we can't do the usual trick of looking at the SB quota bits.
2650 * The other possibility, of course, is that the quota subsystem was
2651 * removed since the last mount - ENOSYS.
2653 dq_f
= (xfs_dq_logformat_t
*)item
->ri_buf
[0].i_addr
;
2655 if ((error
= xfs_qm_dqcheck(recddq
,
2657 0, XFS_QMOPT_DOWARN
,
2658 "xlog_recover_do_dquot_trans (log copy)"))) {
2659 return XFS_ERROR(EIO
);
2661 ASSERT(dq_f
->qlf_len
== 1);
2663 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
2665 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
),
2668 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp
,
2669 bp
, dq_f
->qlf_blkno
);
2673 ddq
= (xfs_disk_dquot_t
*)xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
2676 * At least the magic num portion should be on disk because this
2677 * was among a chunk of dquots created earlier, and we did some
2678 * minimal initialization then.
2680 if (xfs_qm_dqcheck(ddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2681 "xlog_recover_do_dquot_trans")) {
2683 return XFS_ERROR(EIO
);
2686 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
2688 ASSERT(dq_f
->qlf_size
== 2);
2689 ASSERT(bp
->b_mount
== NULL
|| bp
->b_mount
== mp
);
2691 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2692 xfs_bdwrite(mp
, bp
);
2698 * This routine is called to create an in-core extent free intent
2699 * item from the efi format structure which was logged on disk.
2700 * It allocates an in-core efi, copies the extents from the format
2701 * structure into it, and adds the efi to the AIL with the given
2705 xlog_recover_do_efi_trans(
2707 xlog_recover_item_t
*item
,
2713 xfs_efi_log_item_t
*efip
;
2714 xfs_efi_log_format_t
*efi_formatp
;
2716 if (pass
== XLOG_RECOVER_PASS1
) {
2720 efi_formatp
= (xfs_efi_log_format_t
*)item
->ri_buf
[0].i_addr
;
2723 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
2724 if ((error
= xfs_efi_copy_format(&(item
->ri_buf
[0]),
2725 &(efip
->efi_format
)))) {
2726 xfs_efi_item_free(efip
);
2729 efip
->efi_next_extent
= efi_formatp
->efi_nextents
;
2730 efip
->efi_flags
|= XFS_EFI_COMMITTED
;
2732 spin_lock(&log
->l_ailp
->xa_lock
);
2734 * xfs_trans_ail_update() drops the AIL lock.
2736 xfs_trans_ail_update(log
->l_ailp
, (xfs_log_item_t
*)efip
, lsn
);
2742 * This routine is called when an efd format structure is found in
2743 * a committed transaction in the log. It's purpose is to cancel
2744 * the corresponding efi if it was still in the log. To do this
2745 * it searches the AIL for the efi with an id equal to that in the
2746 * efd format structure. If we find it, we remove the efi from the
2750 xlog_recover_do_efd_trans(
2752 xlog_recover_item_t
*item
,
2755 xfs_efd_log_format_t
*efd_formatp
;
2756 xfs_efi_log_item_t
*efip
= NULL
;
2757 xfs_log_item_t
*lip
;
2759 struct xfs_ail_cursor cur
;
2760 struct xfs_ail
*ailp
= log
->l_ailp
;
2762 if (pass
== XLOG_RECOVER_PASS1
) {
2766 efd_formatp
= (xfs_efd_log_format_t
*)item
->ri_buf
[0].i_addr
;
2767 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
2768 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
2769 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
2770 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
2771 efi_id
= efd_formatp
->efd_efi_id
;
2774 * Search for the efi with the id in the efd format structure
2777 spin_lock(&ailp
->xa_lock
);
2778 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
2779 while (lip
!= NULL
) {
2780 if (lip
->li_type
== XFS_LI_EFI
) {
2781 efip
= (xfs_efi_log_item_t
*)lip
;
2782 if (efip
->efi_format
.efi_id
== efi_id
) {
2784 * xfs_trans_ail_delete() drops the
2787 xfs_trans_ail_delete(ailp
, lip
);
2788 xfs_efi_item_free(efip
);
2789 spin_lock(&ailp
->xa_lock
);
2793 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
2795 xfs_trans_ail_cursor_done(ailp
, &cur
);
2796 spin_unlock(&ailp
->xa_lock
);
2800 * Perform the transaction
2802 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2803 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2806 xlog_recover_do_trans(
2808 xlog_recover_t
*trans
,
2812 xlog_recover_item_t
*item
;
2814 error
= xlog_recover_reorder_trans(log
, trans
, pass
);
2818 list_for_each_entry(item
, &trans
->r_itemq
, ri_list
) {
2819 trace_xfs_log_recover_item_recover(log
, trans
, item
, pass
);
2820 switch (ITEM_TYPE(item
)) {
2822 error
= xlog_recover_do_buffer_trans(log
, item
, pass
);
2825 error
= xlog_recover_do_inode_trans(log
, item
, pass
);
2828 error
= xlog_recover_do_efi_trans(log
, item
,
2829 trans
->r_lsn
, pass
);
2832 xlog_recover_do_efd_trans(log
, item
, pass
);
2836 error
= xlog_recover_do_dquot_trans(log
, item
, pass
);
2838 case XFS_LI_QUOTAOFF
:
2839 error
= xlog_recover_do_quotaoff_trans(log
, item
,
2844 "XFS: invalid item type (%d) xlog_recover_do_trans", ITEM_TYPE(item
));
2846 error
= XFS_ERROR(EIO
);
2858 * Free up any resources allocated by the transaction
2860 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2863 xlog_recover_free_trans(
2864 xlog_recover_t
*trans
)
2866 xlog_recover_item_t
*item
, *n
;
2869 list_for_each_entry_safe(item
, n
, &trans
->r_itemq
, ri_list
) {
2870 /* Free the regions in the item. */
2871 list_del(&item
->ri_list
);
2872 for (i
= 0; i
< item
->ri_cnt
; i
++)
2873 kmem_free(item
->ri_buf
[i
].i_addr
);
2874 /* Free the item itself */
2875 kmem_free(item
->ri_buf
);
2878 /* Free the transaction recover structure */
2883 xlog_recover_commit_trans(
2885 xlog_recover_t
*trans
,
2890 hlist_del(&trans
->r_list
);
2891 if ((error
= xlog_recover_do_trans(log
, trans
, pass
)))
2893 xlog_recover_free_trans(trans
); /* no error */
2898 xlog_recover_unmount_trans(
2899 xlog_recover_t
*trans
)
2901 /* Do nothing now */
2902 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2907 * There are two valid states of the r_state field. 0 indicates that the
2908 * transaction structure is in a normal state. We have either seen the
2909 * start of the transaction or the last operation we added was not a partial
2910 * operation. If the last operation we added to the transaction was a
2911 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2913 * NOTE: skip LRs with 0 data length.
2916 xlog_recover_process_data(
2918 struct hlist_head rhash
[],
2919 xlog_rec_header_t
*rhead
,
2925 xlog_op_header_t
*ohead
;
2926 xlog_recover_t
*trans
;
2932 lp
= dp
+ be32_to_cpu(rhead
->h_len
);
2933 num_logops
= be32_to_cpu(rhead
->h_num_logops
);
2935 /* check the log format matches our own - else we can't recover */
2936 if (xlog_header_check_recover(log
->l_mp
, rhead
))
2937 return (XFS_ERROR(EIO
));
2939 while ((dp
< lp
) && num_logops
) {
2940 ASSERT(dp
+ sizeof(xlog_op_header_t
) <= lp
);
2941 ohead
= (xlog_op_header_t
*)dp
;
2942 dp
+= sizeof(xlog_op_header_t
);
2943 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
2944 ohead
->oh_clientid
!= XFS_LOG
) {
2946 "XFS: xlog_recover_process_data: bad clientid");
2948 return (XFS_ERROR(EIO
));
2950 tid
= be32_to_cpu(ohead
->oh_tid
);
2951 hash
= XLOG_RHASH(tid
);
2952 trans
= xlog_recover_find_tid(&rhash
[hash
], tid
);
2953 if (trans
== NULL
) { /* not found; add new tid */
2954 if (ohead
->oh_flags
& XLOG_START_TRANS
)
2955 xlog_recover_new_tid(&rhash
[hash
], tid
,
2956 be64_to_cpu(rhead
->h_lsn
));
2958 if (dp
+ be32_to_cpu(ohead
->oh_len
) > lp
) {
2960 "XFS: xlog_recover_process_data: bad length");
2962 return (XFS_ERROR(EIO
));
2964 flags
= ohead
->oh_flags
& ~XLOG_END_TRANS
;
2965 if (flags
& XLOG_WAS_CONT_TRANS
)
2966 flags
&= ~XLOG_CONTINUE_TRANS
;
2968 case XLOG_COMMIT_TRANS
:
2969 error
= xlog_recover_commit_trans(log
,
2972 case XLOG_UNMOUNT_TRANS
:
2973 error
= xlog_recover_unmount_trans(trans
);
2975 case XLOG_WAS_CONT_TRANS
:
2976 error
= xlog_recover_add_to_cont_trans(log
,
2978 be32_to_cpu(ohead
->oh_len
));
2980 case XLOG_START_TRANS
:
2982 "XFS: xlog_recover_process_data: bad transaction");
2984 error
= XFS_ERROR(EIO
);
2987 case XLOG_CONTINUE_TRANS
:
2988 error
= xlog_recover_add_to_trans(log
, trans
,
2989 dp
, be32_to_cpu(ohead
->oh_len
));
2993 "XFS: xlog_recover_process_data: bad flag");
2995 error
= XFS_ERROR(EIO
);
3001 dp
+= be32_to_cpu(ohead
->oh_len
);
3008 * Process an extent free intent item that was recovered from
3009 * the log. We need to free the extents that it describes.
3012 xlog_recover_process_efi(
3014 xfs_efi_log_item_t
*efip
)
3016 xfs_efd_log_item_t
*efdp
;
3021 xfs_fsblock_t startblock_fsb
;
3023 ASSERT(!(efip
->efi_flags
& XFS_EFI_RECOVERED
));
3026 * First check the validity of the extents described by the
3027 * EFI. If any are bad, then assume that all are bad and
3028 * just toss the EFI.
3030 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3031 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3032 startblock_fsb
= XFS_BB_TO_FSB(mp
,
3033 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
3034 if ((startblock_fsb
== 0) ||
3035 (extp
->ext_len
== 0) ||
3036 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
3037 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
3039 * This will pull the EFI from the AIL and
3040 * free the memory associated with it.
3042 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
3043 return XFS_ERROR(EIO
);
3047 tp
= xfs_trans_alloc(mp
, 0);
3048 error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0, 0, 0);
3051 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
3053 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3054 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3055 error
= xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
3058 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
3062 efip
->efi_flags
|= XFS_EFI_RECOVERED
;
3063 error
= xfs_trans_commit(tp
, 0);
3067 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3072 * When this is called, all of the EFIs which did not have
3073 * corresponding EFDs should be in the AIL. What we do now
3074 * is free the extents associated with each one.
3076 * Since we process the EFIs in normal transactions, they
3077 * will be removed at some point after the commit. This prevents
3078 * us from just walking down the list processing each one.
3079 * We'll use a flag in the EFI to skip those that we've already
3080 * processed and use the AIL iteration mechanism's generation
3081 * count to try to speed this up at least a bit.
3083 * When we start, we know that the EFIs are the only things in
3084 * the AIL. As we process them, however, other items are added
3085 * to the AIL. Since everything added to the AIL must come after
3086 * everything already in the AIL, we stop processing as soon as
3087 * we see something other than an EFI in the AIL.
3090 xlog_recover_process_efis(
3093 xfs_log_item_t
*lip
;
3094 xfs_efi_log_item_t
*efip
;
3096 struct xfs_ail_cursor cur
;
3097 struct xfs_ail
*ailp
;
3100 spin_lock(&ailp
->xa_lock
);
3101 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3102 while (lip
!= NULL
) {
3104 * We're done when we see something other than an EFI.
3105 * There should be no EFIs left in the AIL now.
3107 if (lip
->li_type
!= XFS_LI_EFI
) {
3109 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
3110 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
3116 * Skip EFIs that we've already processed.
3118 efip
= (xfs_efi_log_item_t
*)lip
;
3119 if (efip
->efi_flags
& XFS_EFI_RECOVERED
) {
3120 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3124 spin_unlock(&ailp
->xa_lock
);
3125 error
= xlog_recover_process_efi(log
->l_mp
, efip
);
3126 spin_lock(&ailp
->xa_lock
);
3129 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3132 xfs_trans_ail_cursor_done(ailp
, &cur
);
3133 spin_unlock(&ailp
->xa_lock
);
3138 * This routine performs a transaction to null out a bad inode pointer
3139 * in an agi unlinked inode hash bucket.
3142 xlog_recover_clear_agi_bucket(
3144 xfs_agnumber_t agno
,
3153 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3154 error
= xfs_trans_reserve(tp
, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp
),
3159 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
3163 agi
= XFS_BUF_TO_AGI(agibp
);
3164 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
3165 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3166 (sizeof(xfs_agino_t
) * bucket
);
3167 xfs_trans_log_buf(tp
, agibp
, offset
,
3168 (offset
+ sizeof(xfs_agino_t
) - 1));
3170 error
= xfs_trans_commit(tp
, 0);
3176 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3178 xfs_fs_cmn_err(CE_WARN
, mp
, "xlog_recover_clear_agi_bucket: "
3179 "failed to clear agi %d. Continuing.", agno
);
3184 xlog_recover_process_one_iunlink(
3185 struct xfs_mount
*mp
,
3186 xfs_agnumber_t agno
,
3190 struct xfs_buf
*ibp
;
3191 struct xfs_dinode
*dip
;
3192 struct xfs_inode
*ip
;
3196 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3197 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
);
3202 * Get the on disk inode to find the next inode in the bucket.
3204 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &ibp
, XBF_LOCK
);
3208 ASSERT(ip
->i_d
.di_nlink
== 0);
3209 ASSERT(ip
->i_d
.di_mode
!= 0);
3211 /* setup for the next pass */
3212 agino
= be32_to_cpu(dip
->di_next_unlinked
);
3216 * Prevent any DMAPI event from being sent when the reference on
3217 * the inode is dropped.
3219 ip
->i_d
.di_dmevmask
= 0;
3228 * We can't read in the inode this bucket points to, or this inode
3229 * is messed up. Just ditch this bucket of inodes. We will lose
3230 * some inodes and space, but at least we won't hang.
3232 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3233 * clear the inode pointer in the bucket.
3235 xlog_recover_clear_agi_bucket(mp
, agno
, bucket
);
3240 * xlog_iunlink_recover
3242 * This is called during recovery to process any inodes which
3243 * we unlinked but not freed when the system crashed. These
3244 * inodes will be on the lists in the AGI blocks. What we do
3245 * here is scan all the AGIs and fully truncate and free any
3246 * inodes found on the lists. Each inode is removed from the
3247 * lists when it has been fully truncated and is freed. The
3248 * freeing of the inode and its removal from the list must be
3252 xlog_recover_process_iunlinks(
3256 xfs_agnumber_t agno
;
3267 * Prevent any DMAPI event from being sent while in this function.
3269 mp_dmevmask
= mp
->m_dmevmask
;
3272 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3274 * Find the agi for this ag.
3276 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3279 * AGI is b0rked. Don't process it.
3281 * We should probably mark the filesystem as corrupt
3282 * after we've recovered all the ag's we can....
3286 agi
= XFS_BUF_TO_AGI(agibp
);
3288 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3289 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
3290 while (agino
!= NULLAGINO
) {
3292 * Release the agi buffer so that it can
3293 * be acquired in the normal course of the
3294 * transaction to truncate and free the inode.
3296 xfs_buf_relse(agibp
);
3298 agino
= xlog_recover_process_one_iunlink(mp
,
3299 agno
, agino
, bucket
);
3302 * Reacquire the agibuffer and continue around
3303 * the loop. This should never fail as we know
3304 * the buffer was good earlier on.
3306 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3308 agi
= XFS_BUF_TO_AGI(agibp
);
3313 * Release the buffer for the current agi so we can
3314 * go on to the next one.
3316 xfs_buf_relse(agibp
);
3319 mp
->m_dmevmask
= mp_dmevmask
;
3325 xlog_pack_data_checksum(
3327 xlog_in_core_t
*iclog
,
3334 up
= (__be32
*)iclog
->ic_datap
;
3335 /* divide length by 4 to get # words */
3336 for (i
= 0; i
< (size
>> 2); i
++) {
3337 chksum
^= be32_to_cpu(*up
);
3340 iclog
->ic_header
.h_chksum
= cpu_to_be32(chksum
);
3343 #define xlog_pack_data_checksum(log, iclog, size)
3347 * Stamp cycle number in every block
3352 xlog_in_core_t
*iclog
,
3356 int size
= iclog
->ic_offset
+ roundoff
;
3360 xlog_pack_data_checksum(log
, iclog
, size
);
3362 cycle_lsn
= CYCLE_LSN_DISK(iclog
->ic_header
.h_lsn
);
3364 dp
= iclog
->ic_datap
;
3365 for (i
= 0; i
< BTOBB(size
) &&
3366 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3367 iclog
->ic_header
.h_cycle_data
[i
] = *(__be32
*)dp
;
3368 *(__be32
*)dp
= cycle_lsn
;
3372 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3373 xlog_in_core_2_t
*xhdr
= iclog
->ic_data
;
3375 for ( ; i
< BTOBB(size
); i
++) {
3376 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3377 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3378 xhdr
[j
].hic_xheader
.xh_cycle_data
[k
] = *(__be32
*)dp
;
3379 *(__be32
*)dp
= cycle_lsn
;
3383 for (i
= 1; i
< log
->l_iclog_heads
; i
++) {
3384 xhdr
[i
].hic_xheader
.xh_cycle
= cycle_lsn
;
3391 xlog_rec_header_t
*rhead
,
3397 for (i
= 0; i
< BTOBB(be32_to_cpu(rhead
->h_len
)) &&
3398 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3399 *(__be32
*)dp
= *(__be32
*)&rhead
->h_cycle_data
[i
];
3403 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3404 xlog_in_core_2_t
*xhdr
= (xlog_in_core_2_t
*)rhead
;
3405 for ( ; i
< BTOBB(be32_to_cpu(rhead
->h_len
)); i
++) {
3406 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3407 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3408 *(__be32
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
3415 xlog_valid_rec_header(
3417 xlog_rec_header_t
*rhead
,
3422 if (unlikely(be32_to_cpu(rhead
->h_magicno
) != XLOG_HEADER_MAGIC_NUM
)) {
3423 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3424 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3425 return XFS_ERROR(EFSCORRUPTED
);
3428 (!rhead
->h_version
||
3429 (be32_to_cpu(rhead
->h_version
) & (~XLOG_VERSION_OKBITS
))))) {
3430 xlog_warn("XFS: %s: unrecognised log version (%d).",
3431 __func__
, be32_to_cpu(rhead
->h_version
));
3432 return XFS_ERROR(EIO
);
3435 /* LR body must have data or it wouldn't have been written */
3436 hlen
= be32_to_cpu(rhead
->h_len
);
3437 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
3438 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3439 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3440 return XFS_ERROR(EFSCORRUPTED
);
3442 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
3443 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3444 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3445 return XFS_ERROR(EFSCORRUPTED
);
3451 * Read the log from tail to head and process the log records found.
3452 * Handle the two cases where the tail and head are in the same cycle
3453 * and where the active portion of the log wraps around the end of
3454 * the physical log separately. The pass parameter is passed through
3455 * to the routines called to process the data and is not looked at
3459 xlog_do_recovery_pass(
3461 xfs_daddr_t head_blk
,
3462 xfs_daddr_t tail_blk
,
3465 xlog_rec_header_t
*rhead
;
3468 xfs_buf_t
*hbp
, *dbp
;
3469 int error
= 0, h_size
;
3470 int bblks
, split_bblks
;
3471 int hblks
, split_hblks
, wrapped_hblks
;
3472 struct hlist_head rhash
[XLOG_RHASH_SIZE
];
3474 ASSERT(head_blk
!= tail_blk
);
3477 * Read the header of the tail block and get the iclog buffer size from
3478 * h_size. Use this to tell how many sectors make up the log header.
3480 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3482 * When using variable length iclogs, read first sector of
3483 * iclog header and extract the header size from it. Get a
3484 * new hbp that is the correct size.
3486 hbp
= xlog_get_bp(log
, 1);
3490 error
= xlog_bread(log
, tail_blk
, 1, hbp
, &offset
);
3494 rhead
= (xlog_rec_header_t
*)offset
;
3495 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
3498 h_size
= be32_to_cpu(rhead
->h_size
);
3499 if ((be32_to_cpu(rhead
->h_version
) & XLOG_VERSION_2
) &&
3500 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
3501 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
3502 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
3505 hbp
= xlog_get_bp(log
, hblks
);
3510 ASSERT(log
->l_sectBBsize
== 1);
3512 hbp
= xlog_get_bp(log
, 1);
3513 h_size
= XLOG_BIG_RECORD_BSIZE
;
3518 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
3524 memset(rhash
, 0, sizeof(rhash
));
3525 if (tail_blk
<= head_blk
) {
3526 for (blk_no
= tail_blk
; blk_no
< head_blk
; ) {
3527 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3531 rhead
= (xlog_rec_header_t
*)offset
;
3532 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3536 /* blocks in data section */
3537 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3538 error
= xlog_bread(log
, blk_no
+ hblks
, bblks
, dbp
,
3543 xlog_unpack_data(rhead
, offset
, log
);
3544 if ((error
= xlog_recover_process_data(log
,
3545 rhash
, rhead
, offset
, pass
)))
3547 blk_no
+= bblks
+ hblks
;
3551 * Perform recovery around the end of the physical log.
3552 * When the head is not on the same cycle number as the tail,
3553 * we can't do a sequential recovery as above.
3556 while (blk_no
< log
->l_logBBsize
) {
3558 * Check for header wrapping around physical end-of-log
3560 offset
= XFS_BUF_PTR(hbp
);
3563 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
3564 /* Read header in one read */
3565 error
= xlog_bread(log
, blk_no
, hblks
, hbp
,
3570 /* This LR is split across physical log end */
3571 if (blk_no
!= log
->l_logBBsize
) {
3572 /* some data before physical log end */
3573 ASSERT(blk_no
<= INT_MAX
);
3574 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
3575 ASSERT(split_hblks
> 0);
3576 error
= xlog_bread(log
, blk_no
,
3584 * Note: this black magic still works with
3585 * large sector sizes (non-512) only because:
3586 * - we increased the buffer size originally
3587 * by 1 sector giving us enough extra space
3588 * for the second read;
3589 * - the log start is guaranteed to be sector
3591 * - we read the log end (LR header start)
3592 * _first_, then the log start (LR header end)
3593 * - order is important.
3595 wrapped_hblks
= hblks
- split_hblks
;
3596 error
= XFS_BUF_SET_PTR(hbp
,
3597 offset
+ BBTOB(split_hblks
),
3598 BBTOB(hblks
- split_hblks
));
3602 error
= xlog_bread_noalign(log
, 0,
3603 wrapped_hblks
, hbp
);
3607 error
= XFS_BUF_SET_PTR(hbp
, offset
,
3612 rhead
= (xlog_rec_header_t
*)offset
;
3613 error
= xlog_valid_rec_header(log
, rhead
,
3614 split_hblks
? blk_no
: 0);
3618 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3621 /* Read in data for log record */
3622 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
3623 error
= xlog_bread(log
, blk_no
, bblks
, dbp
,
3628 /* This log record is split across the
3629 * physical end of log */
3630 offset
= XFS_BUF_PTR(dbp
);
3632 if (blk_no
!= log
->l_logBBsize
) {
3633 /* some data is before the physical
3635 ASSERT(!wrapped_hblks
);
3636 ASSERT(blk_no
<= INT_MAX
);
3638 log
->l_logBBsize
- (int)blk_no
;
3639 ASSERT(split_bblks
> 0);
3640 error
= xlog_bread(log
, blk_no
,
3648 * Note: this black magic still works with
3649 * large sector sizes (non-512) only because:
3650 * - we increased the buffer size originally
3651 * by 1 sector giving us enough extra space
3652 * for the second read;
3653 * - the log start is guaranteed to be sector
3655 * - we read the log end (LR header start)
3656 * _first_, then the log start (LR header end)
3657 * - order is important.
3659 error
= XFS_BUF_SET_PTR(dbp
,
3660 offset
+ BBTOB(split_bblks
),
3661 BBTOB(bblks
- split_bblks
));
3665 error
= xlog_bread_noalign(log
, wrapped_hblks
,
3666 bblks
- split_bblks
,
3671 error
= XFS_BUF_SET_PTR(dbp
, offset
, h_size
);
3675 xlog_unpack_data(rhead
, offset
, log
);
3676 if ((error
= xlog_recover_process_data(log
, rhash
,
3677 rhead
, offset
, pass
)))
3682 ASSERT(blk_no
>= log
->l_logBBsize
);
3683 blk_no
-= log
->l_logBBsize
;
3685 /* read first part of physical log */
3686 while (blk_no
< head_blk
) {
3687 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3691 rhead
= (xlog_rec_header_t
*)offset
;
3692 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3696 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3697 error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
,
3702 xlog_unpack_data(rhead
, offset
, log
);
3703 if ((error
= xlog_recover_process_data(log
, rhash
,
3704 rhead
, offset
, pass
)))
3706 blk_no
+= bblks
+ hblks
;
3718 * Do the recovery of the log. We actually do this in two phases.
3719 * The two passes are necessary in order to implement the function
3720 * of cancelling a record written into the log. The first pass
3721 * determines those things which have been cancelled, and the
3722 * second pass replays log items normally except for those which
3723 * have been cancelled. The handling of the replay and cancellations
3724 * takes place in the log item type specific routines.
3726 * The table of items which have cancel records in the log is allocated
3727 * and freed at this level, since only here do we know when all of
3728 * the log recovery has been completed.
3731 xlog_do_log_recovery(
3733 xfs_daddr_t head_blk
,
3734 xfs_daddr_t tail_blk
)
3738 ASSERT(head_blk
!= tail_blk
);
3741 * First do a pass to find all of the cancelled buf log items.
3742 * Store them in the buf_cancel_table for use in the second pass.
3744 log
->l_buf_cancel_table
=
3745 (xfs_buf_cancel_t
**)kmem_zalloc(XLOG_BC_TABLE_SIZE
*
3746 sizeof(xfs_buf_cancel_t
*),
3748 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3749 XLOG_RECOVER_PASS1
);
3751 kmem_free(log
->l_buf_cancel_table
);
3752 log
->l_buf_cancel_table
= NULL
;
3756 * Then do a second pass to actually recover the items in the log.
3757 * When it is complete free the table of buf cancel items.
3759 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3760 XLOG_RECOVER_PASS2
);
3765 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
3766 ASSERT(log
->l_buf_cancel_table
[i
] == NULL
);
3770 kmem_free(log
->l_buf_cancel_table
);
3771 log
->l_buf_cancel_table
= NULL
;
3777 * Do the actual recovery
3782 xfs_daddr_t head_blk
,
3783 xfs_daddr_t tail_blk
)
3790 * First replay the images in the log.
3792 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
3797 XFS_bflush(log
->l_mp
->m_ddev_targp
);
3800 * If IO errors happened during recovery, bail out.
3802 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
3807 * We now update the tail_lsn since much of the recovery has completed
3808 * and there may be space available to use. If there were no extent
3809 * or iunlinks, we can free up the entire log and set the tail_lsn to
3810 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3811 * lsn of the last known good LR on disk. If there are extent frees
3812 * or iunlinks they will have some entries in the AIL; so we look at
3813 * the AIL to determine how to set the tail_lsn.
3815 xlog_assign_tail_lsn(log
->l_mp
);
3818 * Now that we've finished replaying all buffer and inode
3819 * updates, re-read in the superblock.
3821 bp
= xfs_getsb(log
->l_mp
, 0);
3823 ASSERT(!(XFS_BUF_ISWRITE(bp
)));
3824 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp
)));
3826 XFS_BUF_UNASYNC(bp
);
3827 xfsbdstrat(log
->l_mp
, bp
);
3828 error
= xfs_iowait(bp
);
3830 xfs_ioerror_alert("xlog_do_recover",
3831 log
->l_mp
, bp
, XFS_BUF_ADDR(bp
));
3837 /* Convert superblock from on-disk format */
3838 sbp
= &log
->l_mp
->m_sb
;
3839 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
3840 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
3841 ASSERT(xfs_sb_good_version(sbp
));
3844 /* We've re-read the superblock so re-initialize per-cpu counters */
3845 xfs_icsb_reinit_counters(log
->l_mp
);
3847 xlog_recover_check_summary(log
);
3849 /* Normal transactions can now occur */
3850 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
3855 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3857 * Return error or zero.
3863 xfs_daddr_t head_blk
, tail_blk
;
3866 /* find the tail of the log */
3867 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
)))
3870 if (tail_blk
!= head_blk
) {
3871 /* There used to be a comment here:
3873 * disallow recovery on read-only mounts. note -- mount
3874 * checks for ENOSPC and turns it into an intelligent
3876 * ...but this is no longer true. Now, unless you specify
3877 * NORECOVERY (in which case this function would never be
3878 * called), we just go ahead and recover. We do this all
3879 * under the vfs layer, so we can get away with it unless
3880 * the device itself is read-only, in which case we fail.
3882 if ((error
= xfs_dev_is_read_only(log
->l_mp
, "recovery"))) {
3887 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3888 log
->l_mp
->m_fsname
, log
->l_mp
->m_logname
?
3889 log
->l_mp
->m_logname
: "internal");
3891 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
3892 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
3898 * In the first part of recovery we replay inodes and buffers and build
3899 * up the list of extent free items which need to be processed. Here
3900 * we process the extent free items and clean up the on disk unlinked
3901 * inode lists. This is separated from the first part of recovery so
3902 * that the root and real-time bitmap inodes can be read in from disk in
3903 * between the two stages. This is necessary so that we can free space
3904 * in the real-time portion of the file system.
3907 xlog_recover_finish(
3911 * Now we're ready to do the transactions needed for the
3912 * rest of recovery. Start with completing all the extent
3913 * free intent records and then process the unlinked inode
3914 * lists. At this point, we essentially run in normal mode
3915 * except that we're still performing recovery actions
3916 * rather than accepting new requests.
3918 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
3920 error
= xlog_recover_process_efis(log
);
3923 "Failed to recover EFIs on filesystem: %s",
3924 log
->l_mp
->m_fsname
);
3928 * Sync the log to get all the EFIs out of the AIL.
3929 * This isn't absolutely necessary, but it helps in
3930 * case the unlink transactions would have problems
3931 * pushing the EFIs out of the way.
3933 xfs_log_force(log
->l_mp
, XFS_LOG_SYNC
);
3935 xlog_recover_process_iunlinks(log
);
3937 xlog_recover_check_summary(log
);
3940 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3941 log
->l_mp
->m_fsname
, log
->l_mp
->m_logname
?
3942 log
->l_mp
->m_logname
: "internal");
3943 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
3946 "!Ending clean XFS mount for filesystem: %s\n",
3947 log
->l_mp
->m_fsname
);
3955 * Read all of the agf and agi counters and check that they
3956 * are consistent with the superblock counters.
3959 xlog_recover_check_summary(
3966 xfs_agnumber_t agno
;
3967 __uint64_t freeblks
;
3977 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3978 error
= xfs_read_agf(mp
, NULL
, agno
, 0, &agfbp
);
3980 xfs_fs_cmn_err(CE_ALERT
, mp
,
3981 "xlog_recover_check_summary(agf)"
3982 "agf read failed agno %d error %d",
3985 agfp
= XFS_BUF_TO_AGF(agfbp
);
3986 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
3987 be32_to_cpu(agfp
->agf_flcount
);
3988 xfs_buf_relse(agfbp
);
3991 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3993 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agibp
);
3995 itotal
+= be32_to_cpu(agi
->agi_count
);
3996 ifree
+= be32_to_cpu(agi
->agi_freecount
);
3997 xfs_buf_relse(agibp
);