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_uncached(log
->l_mp
->m_logdev_targp
,
122 * Return the address of the start of the given block number's data
123 * in a log buffer. The buffer covers a log sector-aligned region.
132 xfs_daddr_t offset
= blk_no
& ((xfs_daddr_t
)log
->l_sectBBsize
- 1);
134 ASSERT(BBTOB(offset
+ nbblks
) <= XFS_BUF_SIZE(bp
));
135 return XFS_BUF_PTR(bp
) + BBTOB(offset
);
140 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
151 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
152 xlog_warn("XFS: Invalid block length (0x%x) given for buffer",
154 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
158 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
159 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
162 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
164 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
167 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
168 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
170 xfsbdstrat(log
->l_mp
, bp
);
171 error
= xfs_iowait(bp
);
173 xfs_ioerror_alert("xlog_bread", log
->l_mp
,
174 bp
, XFS_BUF_ADDR(bp
));
188 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
192 *offset
= xlog_align(log
, blk_no
, nbblks
, bp
);
197 * Write out the buffer at the given block for the given number of blocks.
198 * The buffer is kept locked across the write and is returned locked.
199 * This can only be used for synchronous log writes.
210 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
211 xlog_warn("XFS: Invalid block length (0x%x) given for buffer",
213 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
217 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
218 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
221 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
223 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
224 XFS_BUF_ZEROFLAGS(bp
);
227 XFS_BUF_PSEMA(bp
, PRIBIO
);
228 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
229 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
231 if ((error
= xfs_bwrite(log
->l_mp
, bp
)))
232 xfs_ioerror_alert("xlog_bwrite", log
->l_mp
,
233 bp
, XFS_BUF_ADDR(bp
));
239 * dump debug superblock and log record information
242 xlog_header_check_dump(
244 xlog_rec_header_t
*head
)
246 cmn_err(CE_DEBUG
, "%s: SB : uuid = %pU, fmt = %d\n",
247 __func__
, &mp
->m_sb
.sb_uuid
, XLOG_FMT
);
248 cmn_err(CE_DEBUG
, " log : uuid = %pU, fmt = %d\n",
249 &head
->h_fs_uuid
, be32_to_cpu(head
->h_fmt
));
252 #define xlog_header_check_dump(mp, head)
256 * check log record header for recovery
259 xlog_header_check_recover(
261 xlog_rec_header_t
*head
)
263 ASSERT(be32_to_cpu(head
->h_magicno
) == XLOG_HEADER_MAGIC_NUM
);
266 * IRIX doesn't write the h_fmt field and leaves it zeroed
267 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
268 * a dirty log created in IRIX.
270 if (unlikely(be32_to_cpu(head
->h_fmt
) != XLOG_FMT
)) {
272 "XFS: dirty log written in incompatible format - can't recover");
273 xlog_header_check_dump(mp
, head
);
274 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
275 XFS_ERRLEVEL_HIGH
, mp
);
276 return XFS_ERROR(EFSCORRUPTED
);
277 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
279 "XFS: dirty log entry has mismatched uuid - can't recover");
280 xlog_header_check_dump(mp
, head
);
281 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
282 XFS_ERRLEVEL_HIGH
, mp
);
283 return XFS_ERROR(EFSCORRUPTED
);
289 * read the head block of the log and check the header
292 xlog_header_check_mount(
294 xlog_rec_header_t
*head
)
296 ASSERT(be32_to_cpu(head
->h_magicno
) == XLOG_HEADER_MAGIC_NUM
);
298 if (uuid_is_nil(&head
->h_fs_uuid
)) {
300 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
301 * h_fs_uuid is nil, we assume this log was last mounted
302 * by IRIX and continue.
304 xlog_warn("XFS: nil uuid in log - IRIX style log");
305 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
306 xlog_warn("XFS: log has mismatched uuid - can't recover");
307 xlog_header_check_dump(mp
, head
);
308 XFS_ERROR_REPORT("xlog_header_check_mount",
309 XFS_ERRLEVEL_HIGH
, mp
);
310 return XFS_ERROR(EFSCORRUPTED
);
319 if (XFS_BUF_GETERROR(bp
)) {
321 * We're not going to bother about retrying
322 * this during recovery. One strike!
324 xfs_ioerror_alert("xlog_recover_iodone",
325 bp
->b_mount
, bp
, XFS_BUF_ADDR(bp
));
326 xfs_force_shutdown(bp
->b_mount
, SHUTDOWN_META_IO_ERROR
);
329 XFS_BUF_CLR_IODONE_FUNC(bp
);
334 * This routine finds (to an approximation) the first block in the physical
335 * log which contains the given cycle. It uses a binary search algorithm.
336 * Note that the algorithm can not be perfect because the disk will not
337 * necessarily be perfect.
340 xlog_find_cycle_start(
343 xfs_daddr_t first_blk
,
344 xfs_daddr_t
*last_blk
,
354 mid_blk
= BLK_AVG(first_blk
, end_blk
);
355 while (mid_blk
!= first_blk
&& mid_blk
!= end_blk
) {
356 error
= xlog_bread(log
, mid_blk
, 1, bp
, &offset
);
359 mid_cycle
= xlog_get_cycle(offset
);
360 if (mid_cycle
== cycle
)
361 end_blk
= mid_blk
; /* last_half_cycle == mid_cycle */
363 first_blk
= mid_blk
; /* first_half_cycle == mid_cycle */
364 mid_blk
= BLK_AVG(first_blk
, end_blk
);
366 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == end_blk
) ||
367 (mid_blk
== end_blk
&& mid_blk
-1 == first_blk
));
375 * Check that a range of blocks does not contain stop_on_cycle_no.
376 * Fill in *new_blk with the block offset where such a block is
377 * found, or with -1 (an invalid block number) if there is no such
378 * block in the range. The scan needs to occur from front to back
379 * and the pointer into the region must be updated since a later
380 * routine will need to perform another test.
383 xlog_find_verify_cycle(
385 xfs_daddr_t start_blk
,
387 uint stop_on_cycle_no
,
388 xfs_daddr_t
*new_blk
)
394 xfs_caddr_t buf
= NULL
;
398 * Greedily allocate a buffer big enough to handle the full
399 * range of basic blocks we'll be examining. If that fails,
400 * try a smaller size. We need to be able to read at least
401 * a log sector, or we're out of luck.
403 bufblks
= 1 << ffs(nbblks
);
404 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
406 if (bufblks
< log
->l_sectBBsize
)
410 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
413 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
415 error
= xlog_bread(log
, i
, bcount
, bp
, &buf
);
419 for (j
= 0; j
< bcount
; j
++) {
420 cycle
= xlog_get_cycle(buf
);
421 if (cycle
== stop_on_cycle_no
) {
438 * Potentially backup over partial log record write.
440 * In the typical case, last_blk is the number of the block directly after
441 * a good log record. Therefore, we subtract one to get the block number
442 * of the last block in the given buffer. extra_bblks contains the number
443 * of blocks we would have read on a previous read. This happens when the
444 * last log record is split over the end of the physical log.
446 * extra_bblks is the number of blocks potentially verified on a previous
447 * call to this routine.
450 xlog_find_verify_log_record(
452 xfs_daddr_t start_blk
,
453 xfs_daddr_t
*last_blk
,
458 xfs_caddr_t offset
= NULL
;
459 xlog_rec_header_t
*head
= NULL
;
462 int num_blks
= *last_blk
- start_blk
;
465 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
467 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
468 if (!(bp
= xlog_get_bp(log
, 1)))
472 error
= xlog_bread(log
, start_blk
, num_blks
, bp
, &offset
);
475 offset
+= ((num_blks
- 1) << BBSHIFT
);
478 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
480 /* valid log record not found */
482 "XFS: Log inconsistent (didn't find previous header)");
484 error
= XFS_ERROR(EIO
);
489 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
494 head
= (xlog_rec_header_t
*)offset
;
496 if (XLOG_HEADER_MAGIC_NUM
== be32_to_cpu(head
->h_magicno
))
504 * We hit the beginning of the physical log & still no header. Return
505 * to caller. If caller can handle a return of -1, then this routine
506 * will be called again for the end of the physical log.
514 * We have the final block of the good log (the first block
515 * of the log record _before_ the head. So we check the uuid.
517 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
521 * We may have found a log record header before we expected one.
522 * last_blk will be the 1st block # with a given cycle #. We may end
523 * up reading an entire log record. In this case, we don't want to
524 * reset last_blk. Only when last_blk points in the middle of a log
525 * record do we update last_blk.
527 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
528 uint h_size
= be32_to_cpu(head
->h_size
);
530 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
531 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
537 if (*last_blk
- i
+ extra_bblks
!=
538 BTOBB(be32_to_cpu(head
->h_len
)) + xhdrs
)
547 * Head is defined to be the point of the log where the next log write
548 * write could go. This means that incomplete LR writes at the end are
549 * eliminated when calculating the head. We aren't guaranteed that previous
550 * LR have complete transactions. We only know that a cycle number of
551 * current cycle number -1 won't be present in the log if we start writing
552 * from our current block number.
554 * last_blk contains the block number of the first block with a given
557 * Return: zero if normal, non-zero if error.
562 xfs_daddr_t
*return_head_blk
)
566 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
568 uint first_half_cycle
, last_half_cycle
;
570 int error
, log_bbnum
= log
->l_logBBsize
;
572 /* Is the end of the log device zeroed? */
573 if ((error
= xlog_find_zeroed(log
, &first_blk
)) == -1) {
574 *return_head_blk
= first_blk
;
576 /* Is the whole lot zeroed? */
578 /* Linux XFS shouldn't generate totally zeroed logs -
579 * mkfs etc write a dummy unmount record to a fresh
580 * log so we can store the uuid in there
582 xlog_warn("XFS: totally zeroed log");
587 xlog_warn("XFS: empty log check failed");
591 first_blk
= 0; /* get cycle # of 1st block */
592 bp
= xlog_get_bp(log
, 1);
596 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
600 first_half_cycle
= xlog_get_cycle(offset
);
602 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
603 error
= xlog_bread(log
, last_blk
, 1, bp
, &offset
);
607 last_half_cycle
= xlog_get_cycle(offset
);
608 ASSERT(last_half_cycle
!= 0);
611 * If the 1st half cycle number is equal to the last half cycle number,
612 * then the entire log is stamped with the same cycle number. In this
613 * case, head_blk can't be set to zero (which makes sense). The below
614 * math doesn't work out properly with head_blk equal to zero. Instead,
615 * we set it to log_bbnum which is an invalid block number, but this
616 * value makes the math correct. If head_blk doesn't changed through
617 * all the tests below, *head_blk is set to zero at the very end rather
618 * than log_bbnum. In a sense, log_bbnum and zero are the same block
619 * in a circular file.
621 if (first_half_cycle
== last_half_cycle
) {
623 * In this case we believe that the entire log should have
624 * cycle number last_half_cycle. We need to scan backwards
625 * from the end verifying that there are no holes still
626 * containing last_half_cycle - 1. If we find such a hole,
627 * then the start of that hole will be the new head. The
628 * simple case looks like
629 * x | x ... | x - 1 | x
630 * Another case that fits this picture would be
631 * x | x + 1 | x ... | x
632 * In this case the head really is somewhere at the end of the
633 * log, as one of the latest writes at the beginning was
636 * x | x + 1 | x ... | x - 1 | x
637 * This is really the combination of the above two cases, and
638 * the head has to end up at the start of the x-1 hole at the
641 * In the 256k log case, we will read from the beginning to the
642 * end of the log and search for cycle numbers equal to x-1.
643 * We don't worry about the x+1 blocks that we encounter,
644 * because we know that they cannot be the head since the log
647 head_blk
= log_bbnum
;
648 stop_on_cycle
= last_half_cycle
- 1;
651 * In this case we want to find the first block with cycle
652 * number matching last_half_cycle. We expect the log to be
654 * x + 1 ... | x ... | x
655 * The first block with cycle number x (last_half_cycle) will
656 * be where the new head belongs. First we do a binary search
657 * for the first occurrence of last_half_cycle. The binary
658 * search may not be totally accurate, so then we scan back
659 * from there looking for occurrences of last_half_cycle before
660 * us. If that backwards scan wraps around the beginning of
661 * the log, then we look for occurrences of last_half_cycle - 1
662 * at the end of the log. The cases we're looking for look
664 * v binary search stopped here
665 * x + 1 ... | x | x + 1 | x ... | x
666 * ^ but we want to locate this spot
668 * <---------> less than scan distance
669 * x + 1 ... | x ... | x - 1 | x
670 * ^ we want to locate this spot
672 stop_on_cycle
= last_half_cycle
;
673 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
674 &head_blk
, last_half_cycle
)))
679 * Now validate the answer. Scan back some number of maximum possible
680 * blocks and make sure each one has the expected cycle number. The
681 * maximum is determined by the total possible amount of buffering
682 * in the in-core log. The following number can be made tighter if
683 * we actually look at the block size of the filesystem.
685 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
686 if (head_blk
>= num_scan_bblks
) {
688 * We are guaranteed that the entire check can be performed
691 start_blk
= head_blk
- num_scan_bblks
;
692 if ((error
= xlog_find_verify_cycle(log
,
693 start_blk
, num_scan_bblks
,
694 stop_on_cycle
, &new_blk
)))
698 } else { /* need to read 2 parts of log */
700 * We are going to scan backwards in the log in two parts.
701 * First we scan the physical end of the log. In this part
702 * of the log, we are looking for blocks with cycle number
703 * last_half_cycle - 1.
704 * If we find one, then we know that the log starts there, as
705 * we've found a hole that didn't get written in going around
706 * the end of the physical log. The simple case for this is
707 * x + 1 ... | x ... | x - 1 | x
708 * <---------> less than scan distance
709 * If all of the blocks at the end of the log have cycle number
710 * last_half_cycle, then we check the blocks at the start of
711 * the log looking for occurrences of last_half_cycle. If we
712 * find one, then our current estimate for the location of the
713 * first occurrence of last_half_cycle is wrong and we move
714 * back to the hole we've found. This case looks like
715 * x + 1 ... | x | x + 1 | x ...
716 * ^ binary search stopped here
717 * Another case we need to handle that only occurs in 256k
719 * x + 1 ... | x ... | x+1 | x ...
720 * ^ binary search stops here
721 * In a 256k log, the scan at the end of the log will see the
722 * x + 1 blocks. We need to skip past those since that is
723 * certainly not the head of the log. By searching for
724 * last_half_cycle-1 we accomplish that.
726 ASSERT(head_blk
<= INT_MAX
&&
727 (xfs_daddr_t
) num_scan_bblks
>= head_blk
);
728 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
729 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
730 num_scan_bblks
- (int)head_blk
,
731 (stop_on_cycle
- 1), &new_blk
)))
739 * Scan beginning of log now. The last part of the physical
740 * log is good. This scan needs to verify that it doesn't find
741 * the last_half_cycle.
744 ASSERT(head_blk
<= INT_MAX
);
745 if ((error
= xlog_find_verify_cycle(log
,
746 start_blk
, (int)head_blk
,
747 stop_on_cycle
, &new_blk
)))
755 * Now we need to make sure head_blk is not pointing to a block in
756 * the middle of a log record.
758 num_scan_bblks
= XLOG_REC_SHIFT(log
);
759 if (head_blk
>= num_scan_bblks
) {
760 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
762 /* start ptr at last block ptr before head_blk */
763 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
764 &head_blk
, 0)) == -1) {
765 error
= XFS_ERROR(EIO
);
771 ASSERT(head_blk
<= INT_MAX
);
772 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
773 &head_blk
, 0)) == -1) {
774 /* We hit the beginning of the log during our search */
775 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
777 ASSERT(start_blk
<= INT_MAX
&&
778 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
779 ASSERT(head_blk
<= INT_MAX
);
780 if ((error
= xlog_find_verify_log_record(log
,
782 (int)head_blk
)) == -1) {
783 error
= XFS_ERROR(EIO
);
787 if (new_blk
!= log_bbnum
)
794 if (head_blk
== log_bbnum
)
795 *return_head_blk
= 0;
797 *return_head_blk
= head_blk
;
799 * When returning here, we have a good block number. Bad block
800 * means that during a previous crash, we didn't have a clean break
801 * from cycle number N to cycle number N-1. In this case, we need
802 * to find the first block with cycle number N-1.
810 xlog_warn("XFS: failed to find log head");
815 * Find the sync block number or the tail of the log.
817 * This will be the block number of the last record to have its
818 * associated buffers synced to disk. Every log record header has
819 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
820 * to get a sync block number. The only concern is to figure out which
821 * log record header to believe.
823 * The following algorithm uses the log record header with the largest
824 * lsn. The entire log record does not need to be valid. We only care
825 * that the header is valid.
827 * We could speed up search by using current head_blk buffer, but it is not
833 xfs_daddr_t
*head_blk
,
834 xfs_daddr_t
*tail_blk
)
836 xlog_rec_header_t
*rhead
;
837 xlog_op_header_t
*op_head
;
838 xfs_caddr_t offset
= NULL
;
841 xfs_daddr_t umount_data_blk
;
842 xfs_daddr_t after_umount_blk
;
849 * Find previous log record
851 if ((error
= xlog_find_head(log
, head_blk
)))
854 bp
= xlog_get_bp(log
, 1);
857 if (*head_blk
== 0) { /* special case */
858 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
862 if (xlog_get_cycle(offset
) == 0) {
864 /* leave all other log inited values alone */
870 * Search backwards looking for log record header block
872 ASSERT(*head_blk
< INT_MAX
);
873 for (i
= (int)(*head_blk
) - 1; i
>= 0; i
--) {
874 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
878 if (XLOG_HEADER_MAGIC_NUM
== be32_to_cpu(*(__be32
*)offset
)) {
884 * If we haven't found the log record header block, start looking
885 * again from the end of the physical log. XXXmiken: There should be
886 * a check here to make sure we didn't search more than N blocks in
890 for (i
= log
->l_logBBsize
- 1; i
>= (int)(*head_blk
); i
--) {
891 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
895 if (XLOG_HEADER_MAGIC_NUM
==
896 be32_to_cpu(*(__be32
*)offset
)) {
903 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
905 return XFS_ERROR(EIO
);
908 /* find blk_no of tail of log */
909 rhead
= (xlog_rec_header_t
*)offset
;
910 *tail_blk
= BLOCK_LSN(be64_to_cpu(rhead
->h_tail_lsn
));
913 * Reset log values according to the state of the log when we
914 * crashed. In the case where head_blk == 0, we bump curr_cycle
915 * one because the next write starts a new cycle rather than
916 * continuing the cycle of the last good log record. At this
917 * point we have guaranteed that all partial log records have been
918 * accounted for. Therefore, we know that the last good log record
919 * written was complete and ended exactly on the end boundary
920 * of the physical log.
922 log
->l_prev_block
= i
;
923 log
->l_curr_block
= (int)*head_blk
;
924 log
->l_curr_cycle
= be32_to_cpu(rhead
->h_cycle
);
927 log
->l_tail_lsn
= be64_to_cpu(rhead
->h_tail_lsn
);
928 log
->l_last_sync_lsn
= be64_to_cpu(rhead
->h_lsn
);
929 log
->l_grant_reserve_cycle
= log
->l_curr_cycle
;
930 log
->l_grant_reserve_bytes
= BBTOB(log
->l_curr_block
);
931 log
->l_grant_write_cycle
= log
->l_curr_cycle
;
932 log
->l_grant_write_bytes
= BBTOB(log
->l_curr_block
);
935 * Look for unmount record. If we find it, then we know there
936 * was a clean unmount. Since 'i' could be the last block in
937 * the physical log, we convert to a log block before comparing
940 * Save the current tail lsn to use to pass to
941 * xlog_clear_stale_blocks() below. We won't want to clear the
942 * unmount record if there is one, so we pass the lsn of the
943 * unmount record rather than the block after it.
945 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
946 int h_size
= be32_to_cpu(rhead
->h_size
);
947 int h_version
= be32_to_cpu(rhead
->h_version
);
949 if ((h_version
& XLOG_VERSION_2
) &&
950 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
951 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
952 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
960 after_umount_blk
= (i
+ hblks
+ (int)
961 BTOBB(be32_to_cpu(rhead
->h_len
))) % log
->l_logBBsize
;
962 tail_lsn
= log
->l_tail_lsn
;
963 if (*head_blk
== after_umount_blk
&&
964 be32_to_cpu(rhead
->h_num_logops
) == 1) {
965 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
966 error
= xlog_bread(log
, umount_data_blk
, 1, bp
, &offset
);
970 op_head
= (xlog_op_header_t
*)offset
;
971 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
973 * Set tail and last sync so that newly written
974 * log records will point recovery to after the
975 * current unmount record.
978 xlog_assign_lsn(log
->l_curr_cycle
,
980 log
->l_last_sync_lsn
=
981 xlog_assign_lsn(log
->l_curr_cycle
,
983 *tail_blk
= after_umount_blk
;
986 * Note that the unmount was clean. If the unmount
987 * was not clean, we need to know this to rebuild the
988 * superblock counters from the perag headers if we
989 * have a filesystem using non-persistent counters.
991 log
->l_mp
->m_flags
|= XFS_MOUNT_WAS_CLEAN
;
996 * Make sure that there are no blocks in front of the head
997 * with the same cycle number as the head. This can happen
998 * because we allow multiple outstanding log writes concurrently,
999 * and the later writes might make it out before earlier ones.
1001 * We use the lsn from before modifying it so that we'll never
1002 * overwrite the unmount record after a clean unmount.
1004 * Do this only if we are going to recover the filesystem
1006 * NOTE: This used to say "if (!readonly)"
1007 * However on Linux, we can & do recover a read-only filesystem.
1008 * We only skip recovery if NORECOVERY is specified on mount,
1009 * in which case we would not be here.
1011 * But... if the -device- itself is readonly, just skip this.
1012 * We can't recover this device anyway, so it won't matter.
1014 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
))
1015 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
1021 xlog_warn("XFS: failed to locate log tail");
1026 * Is the log zeroed at all?
1028 * The last binary search should be changed to perform an X block read
1029 * once X becomes small enough. You can then search linearly through
1030 * the X blocks. This will cut down on the number of reads we need to do.
1032 * If the log is partially zeroed, this routine will pass back the blkno
1033 * of the first block with cycle number 0. It won't have a complete LR
1037 * 0 => the log is completely written to
1038 * -1 => use *blk_no as the first block of the log
1039 * >0 => error has occurred
1044 xfs_daddr_t
*blk_no
)
1048 uint first_cycle
, last_cycle
;
1049 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1050 xfs_daddr_t num_scan_bblks
;
1051 int error
, log_bbnum
= log
->l_logBBsize
;
1055 /* check totally zeroed log */
1056 bp
= xlog_get_bp(log
, 1);
1059 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1063 first_cycle
= xlog_get_cycle(offset
);
1064 if (first_cycle
== 0) { /* completely zeroed log */
1070 /* check partially zeroed log */
1071 error
= xlog_bread(log
, log_bbnum
-1, 1, bp
, &offset
);
1075 last_cycle
= xlog_get_cycle(offset
);
1076 if (last_cycle
!= 0) { /* log completely written to */
1079 } else if (first_cycle
!= 1) {
1081 * If the cycle of the last block is zero, the cycle of
1082 * the first block must be 1. If it's not, maybe we're
1083 * not looking at a log... Bail out.
1085 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1086 return XFS_ERROR(EINVAL
);
1089 /* we have a partially zeroed log */
1090 last_blk
= log_bbnum
-1;
1091 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1095 * Validate the answer. Because there is no way to guarantee that
1096 * the entire log is made up of log records which are the same size,
1097 * we scan over the defined maximum blocks. At this point, the maximum
1098 * is not chosen to mean anything special. XXXmiken
1100 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1101 ASSERT(num_scan_bblks
<= INT_MAX
);
1103 if (last_blk
< num_scan_bblks
)
1104 num_scan_bblks
= last_blk
;
1105 start_blk
= last_blk
- num_scan_bblks
;
1108 * We search for any instances of cycle number 0 that occur before
1109 * our current estimate of the head. What we're trying to detect is
1110 * 1 ... | 0 | 1 | 0...
1111 * ^ binary search ends here
1113 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1114 (int)num_scan_bblks
, 0, &new_blk
)))
1120 * Potentially backup over partial log record write. We don't need
1121 * to search the end of the log because we know it is zero.
1123 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
1124 &last_blk
, 0)) == -1) {
1125 error
= XFS_ERROR(EIO
);
1139 * These are simple subroutines used by xlog_clear_stale_blocks() below
1140 * to initialize a buffer full of empty log record headers and write
1141 * them into the log.
1152 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1154 memset(buf
, 0, BBSIZE
);
1155 recp
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1156 recp
->h_cycle
= cpu_to_be32(cycle
);
1157 recp
->h_version
= cpu_to_be32(
1158 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1159 recp
->h_lsn
= cpu_to_be64(xlog_assign_lsn(cycle
, block
));
1160 recp
->h_tail_lsn
= cpu_to_be64(xlog_assign_lsn(tail_cycle
, tail_block
));
1161 recp
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1162 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1166 xlog_write_log_records(
1177 int sectbb
= log
->l_sectBBsize
;
1178 int end_block
= start_block
+ blocks
;
1184 * Greedily allocate a buffer big enough to handle the full
1185 * range of basic blocks to be written. If that fails, try
1186 * a smaller size. We need to be able to write at least a
1187 * log sector, or we're out of luck.
1189 bufblks
= 1 << ffs(blocks
);
1190 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1192 if (bufblks
< sectbb
)
1196 /* We may need to do a read at the start to fill in part of
1197 * the buffer in the starting sector not covered by the first
1200 balign
= round_down(start_block
, sectbb
);
1201 if (balign
!= start_block
) {
1202 error
= xlog_bread_noalign(log
, start_block
, 1, bp
);
1206 j
= start_block
- balign
;
1209 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1210 int bcount
, endcount
;
1212 bcount
= min(bufblks
, end_block
- start_block
);
1213 endcount
= bcount
- j
;
1215 /* We may need to do a read at the end to fill in part of
1216 * the buffer in the final sector not covered by the write.
1217 * If this is the same sector as the above read, skip it.
1219 ealign
= round_down(end_block
, sectbb
);
1220 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1221 offset
= XFS_BUF_PTR(bp
);
1222 balign
= BBTOB(ealign
- start_block
);
1223 error
= XFS_BUF_SET_PTR(bp
, offset
+ balign
,
1228 error
= xlog_bread_noalign(log
, ealign
, sectbb
, bp
);
1232 error
= XFS_BUF_SET_PTR(bp
, offset
, bufblks
);
1237 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1238 for (; j
< endcount
; j
++) {
1239 xlog_add_record(log
, offset
, cycle
, i
+j
,
1240 tail_cycle
, tail_block
);
1243 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1246 start_block
+= endcount
;
1256 * This routine is called to blow away any incomplete log writes out
1257 * in front of the log head. We do this so that we won't become confused
1258 * if we come up, write only a little bit more, and then crash again.
1259 * If we leave the partial log records out there, this situation could
1260 * cause us to think those partial writes are valid blocks since they
1261 * have the current cycle number. We get rid of them by overwriting them
1262 * with empty log records with the old cycle number rather than the
1265 * The tail lsn is passed in rather than taken from
1266 * the log so that we will not write over the unmount record after a
1267 * clean unmount in a 512 block log. Doing so would leave the log without
1268 * any valid log records in it until a new one was written. If we crashed
1269 * during that time we would not be able to recover.
1272 xlog_clear_stale_blocks(
1276 int tail_cycle
, head_cycle
;
1277 int tail_block
, head_block
;
1278 int tail_distance
, max_distance
;
1282 tail_cycle
= CYCLE_LSN(tail_lsn
);
1283 tail_block
= BLOCK_LSN(tail_lsn
);
1284 head_cycle
= log
->l_curr_cycle
;
1285 head_block
= log
->l_curr_block
;
1288 * Figure out the distance between the new head of the log
1289 * and the tail. We want to write over any blocks beyond the
1290 * head that we may have written just before the crash, but
1291 * we don't want to overwrite the tail of the log.
1293 if (head_cycle
== tail_cycle
) {
1295 * The tail is behind the head in the physical log,
1296 * so the distance from the head to the tail is the
1297 * distance from the head to the end of the log plus
1298 * the distance from the beginning of the log to the
1301 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1302 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1303 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1304 return XFS_ERROR(EFSCORRUPTED
);
1306 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1309 * The head is behind the tail in the physical log,
1310 * so the distance from the head to the tail is just
1311 * the tail block minus the head block.
1313 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1314 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1315 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1316 return XFS_ERROR(EFSCORRUPTED
);
1318 tail_distance
= tail_block
- head_block
;
1322 * If the head is right up against the tail, we can't clear
1325 if (tail_distance
<= 0) {
1326 ASSERT(tail_distance
== 0);
1330 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1332 * Take the smaller of the maximum amount of outstanding I/O
1333 * we could have and the distance to the tail to clear out.
1334 * We take the smaller so that we don't overwrite the tail and
1335 * we don't waste all day writing from the head to the tail
1338 max_distance
= MIN(max_distance
, tail_distance
);
1340 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1342 * We can stomp all the blocks we need to without
1343 * wrapping around the end of the log. Just do it
1344 * in a single write. Use the cycle number of the
1345 * current cycle minus one so that the log will look like:
1348 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1349 head_block
, max_distance
, tail_cycle
,
1355 * We need to wrap around the end of the physical log in
1356 * order to clear all the blocks. Do it in two separate
1357 * I/Os. The first write should be from the head to the
1358 * end of the physical log, and it should use the current
1359 * cycle number minus one just like above.
1361 distance
= log
->l_logBBsize
- head_block
;
1362 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1363 head_block
, distance
, tail_cycle
,
1370 * Now write the blocks at the start of the physical log.
1371 * This writes the remainder of the blocks we want to clear.
1372 * It uses the current cycle number since we're now on the
1373 * same cycle as the head so that we get:
1374 * n ... n ... | n - 1 ...
1375 * ^^^^^ blocks we're writing
1377 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1378 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1379 tail_cycle
, tail_block
);
1387 /******************************************************************************
1389 * Log recover routines
1391 ******************************************************************************
1394 STATIC xlog_recover_t
*
1395 xlog_recover_find_tid(
1396 struct hlist_head
*head
,
1399 xlog_recover_t
*trans
;
1400 struct hlist_node
*n
;
1402 hlist_for_each_entry(trans
, n
, head
, r_list
) {
1403 if (trans
->r_log_tid
== tid
)
1410 xlog_recover_new_tid(
1411 struct hlist_head
*head
,
1415 xlog_recover_t
*trans
;
1417 trans
= kmem_zalloc(sizeof(xlog_recover_t
), KM_SLEEP
);
1418 trans
->r_log_tid
= tid
;
1420 INIT_LIST_HEAD(&trans
->r_itemq
);
1422 INIT_HLIST_NODE(&trans
->r_list
);
1423 hlist_add_head(&trans
->r_list
, head
);
1427 xlog_recover_add_item(
1428 struct list_head
*head
)
1430 xlog_recover_item_t
*item
;
1432 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
1433 INIT_LIST_HEAD(&item
->ri_list
);
1434 list_add_tail(&item
->ri_list
, head
);
1438 xlog_recover_add_to_cont_trans(
1440 xlog_recover_t
*trans
,
1444 xlog_recover_item_t
*item
;
1445 xfs_caddr_t ptr
, old_ptr
;
1448 if (list_empty(&trans
->r_itemq
)) {
1449 /* finish copying rest of trans header */
1450 xlog_recover_add_item(&trans
->r_itemq
);
1451 ptr
= (xfs_caddr_t
) &trans
->r_theader
+
1452 sizeof(xfs_trans_header_t
) - len
;
1453 memcpy(ptr
, dp
, len
); /* d, s, l */
1456 /* take the tail entry */
1457 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1459 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
1460 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
1462 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, 0u);
1463 memcpy(&ptr
[old_len
], dp
, len
); /* d, s, l */
1464 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
1465 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
1466 trace_xfs_log_recover_item_add_cont(log
, trans
, item
, 0);
1471 * The next region to add is the start of a new region. It could be
1472 * a whole region or it could be the first part of a new region. Because
1473 * of this, the assumption here is that the type and size fields of all
1474 * format structures fit into the first 32 bits of the structure.
1476 * This works because all regions must be 32 bit aligned. Therefore, we
1477 * either have both fields or we have neither field. In the case we have
1478 * neither field, the data part of the region is zero length. We only have
1479 * a log_op_header and can throw away the header since a new one will appear
1480 * later. If we have at least 4 bytes, then we can determine how many regions
1481 * will appear in the current log item.
1484 xlog_recover_add_to_trans(
1486 xlog_recover_t
*trans
,
1490 xfs_inode_log_format_t
*in_f
; /* any will do */
1491 xlog_recover_item_t
*item
;
1496 if (list_empty(&trans
->r_itemq
)) {
1497 /* we need to catch log corruptions here */
1498 if (*(uint
*)dp
!= XFS_TRANS_HEADER_MAGIC
) {
1499 xlog_warn("XFS: xlog_recover_add_to_trans: "
1500 "bad header magic number");
1502 return XFS_ERROR(EIO
);
1504 if (len
== sizeof(xfs_trans_header_t
))
1505 xlog_recover_add_item(&trans
->r_itemq
);
1506 memcpy(&trans
->r_theader
, dp
, len
); /* d, s, l */
1510 ptr
= kmem_alloc(len
, KM_SLEEP
);
1511 memcpy(ptr
, dp
, len
);
1512 in_f
= (xfs_inode_log_format_t
*)ptr
;
1514 /* take the tail entry */
1515 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1516 if (item
->ri_total
!= 0 &&
1517 item
->ri_total
== item
->ri_cnt
) {
1518 /* tail item is in use, get a new one */
1519 xlog_recover_add_item(&trans
->r_itemq
);
1520 item
= list_entry(trans
->r_itemq
.prev
,
1521 xlog_recover_item_t
, ri_list
);
1524 if (item
->ri_total
== 0) { /* first region to be added */
1525 if (in_f
->ilf_size
== 0 ||
1526 in_f
->ilf_size
> XLOG_MAX_REGIONS_IN_ITEM
) {
1528 "XFS: bad number of regions (%d) in inode log format",
1531 return XFS_ERROR(EIO
);
1534 item
->ri_total
= in_f
->ilf_size
;
1536 kmem_zalloc(item
->ri_total
* sizeof(xfs_log_iovec_t
),
1539 ASSERT(item
->ri_total
> item
->ri_cnt
);
1540 /* Description region is ri_buf[0] */
1541 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
1542 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
1544 trace_xfs_log_recover_item_add(log
, trans
, item
, 0);
1549 * Sort the log items in the transaction. Cancelled buffers need
1550 * to be put first so they are processed before any items that might
1551 * modify the buffers. If they are cancelled, then the modifications
1552 * don't need to be replayed.
1555 xlog_recover_reorder_trans(
1557 xlog_recover_t
*trans
,
1560 xlog_recover_item_t
*item
, *n
;
1561 LIST_HEAD(sort_list
);
1563 list_splice_init(&trans
->r_itemq
, &sort_list
);
1564 list_for_each_entry_safe(item
, n
, &sort_list
, ri_list
) {
1565 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1567 switch (ITEM_TYPE(item
)) {
1569 if (!(buf_f
->blf_flags
& XFS_BLF_CANCEL
)) {
1570 trace_xfs_log_recover_item_reorder_head(log
,
1572 list_move(&item
->ri_list
, &trans
->r_itemq
);
1577 case XFS_LI_QUOTAOFF
:
1580 trace_xfs_log_recover_item_reorder_tail(log
,
1582 list_move_tail(&item
->ri_list
, &trans
->r_itemq
);
1586 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1588 return XFS_ERROR(EIO
);
1591 ASSERT(list_empty(&sort_list
));
1596 * Build up the table of buf cancel records so that we don't replay
1597 * cancelled data in the second pass. For buffer records that are
1598 * not cancel records, there is nothing to do here so we just return.
1600 * If we get a cancel record which is already in the table, this indicates
1601 * that the buffer was cancelled multiple times. In order to ensure
1602 * that during pass 2 we keep the record in the table until we reach its
1603 * last occurrence in the log, we keep a reference count in the cancel
1604 * record in the table to tell us how many times we expect to see this
1605 * record during the second pass.
1608 xlog_recover_do_buffer_pass1(
1610 xfs_buf_log_format_t
*buf_f
)
1612 xfs_buf_cancel_t
*bcp
;
1613 xfs_buf_cancel_t
*nextp
;
1614 xfs_buf_cancel_t
*prevp
;
1615 xfs_buf_cancel_t
**bucket
;
1616 xfs_daddr_t blkno
= 0;
1620 switch (buf_f
->blf_type
) {
1622 blkno
= buf_f
->blf_blkno
;
1623 len
= buf_f
->blf_len
;
1624 flags
= buf_f
->blf_flags
;
1629 * If this isn't a cancel buffer item, then just return.
1631 if (!(flags
& XFS_BLF_CANCEL
)) {
1632 trace_xfs_log_recover_buf_not_cancel(log
, buf_f
);
1637 * Insert an xfs_buf_cancel record into the hash table of
1638 * them. If there is already an identical record, bump
1639 * its reference count.
1641 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1642 XLOG_BC_TABLE_SIZE
];
1644 * If the hash bucket is empty then just insert a new record into
1647 if (*bucket
== NULL
) {
1648 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1650 bcp
->bc_blkno
= blkno
;
1652 bcp
->bc_refcount
= 1;
1653 bcp
->bc_next
= NULL
;
1659 * The hash bucket is not empty, so search for duplicates of our
1660 * record. If we find one them just bump its refcount. If not
1661 * then add us at the end of the list.
1665 while (nextp
!= NULL
) {
1666 if (nextp
->bc_blkno
== blkno
&& nextp
->bc_len
== len
) {
1667 nextp
->bc_refcount
++;
1668 trace_xfs_log_recover_buf_cancel_ref_inc(log
, buf_f
);
1672 nextp
= nextp
->bc_next
;
1674 ASSERT(prevp
!= NULL
);
1675 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1677 bcp
->bc_blkno
= blkno
;
1679 bcp
->bc_refcount
= 1;
1680 bcp
->bc_next
= NULL
;
1681 prevp
->bc_next
= bcp
;
1682 trace_xfs_log_recover_buf_cancel_add(log
, buf_f
);
1686 * Check to see whether the buffer being recovered has a corresponding
1687 * entry in the buffer cancel record table. If it does then return 1
1688 * so that it will be cancelled, otherwise return 0. If the buffer is
1689 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
1690 * the refcount on the entry in the table and remove it from the table
1691 * if this is the last reference.
1693 * We remove the cancel record from the table when we encounter its
1694 * last occurrence in the log so that if the same buffer is re-used
1695 * again after its last cancellation we actually replay the changes
1696 * made at that point.
1699 xlog_check_buffer_cancelled(
1705 xfs_buf_cancel_t
*bcp
;
1706 xfs_buf_cancel_t
*prevp
;
1707 xfs_buf_cancel_t
**bucket
;
1709 if (log
->l_buf_cancel_table
== NULL
) {
1711 * There is nothing in the table built in pass one,
1712 * so this buffer must not be cancelled.
1714 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1718 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1719 XLOG_BC_TABLE_SIZE
];
1723 * There is no corresponding entry in the table built
1724 * in pass one, so this buffer has not been cancelled.
1726 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1731 * Search for an entry in the buffer cancel table that
1732 * matches our buffer.
1735 while (bcp
!= NULL
) {
1736 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
) {
1738 * We've go a match, so return 1 so that the
1739 * recovery of this buffer is cancelled.
1740 * If this buffer is actually a buffer cancel
1741 * log item, then decrement the refcount on the
1742 * one in the table and remove it if this is the
1745 if (flags
& XFS_BLF_CANCEL
) {
1747 if (bcp
->bc_refcount
== 0) {
1748 if (prevp
== NULL
) {
1749 *bucket
= bcp
->bc_next
;
1751 prevp
->bc_next
= bcp
->bc_next
;
1762 * We didn't find a corresponding entry in the table, so
1763 * return 0 so that the buffer is NOT cancelled.
1765 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1770 xlog_recover_do_buffer_pass2(
1772 xfs_buf_log_format_t
*buf_f
)
1774 xfs_daddr_t blkno
= 0;
1778 switch (buf_f
->blf_type
) {
1780 blkno
= buf_f
->blf_blkno
;
1781 flags
= buf_f
->blf_flags
;
1782 len
= buf_f
->blf_len
;
1786 return xlog_check_buffer_cancelled(log
, blkno
, len
, flags
);
1790 * Perform recovery for a buffer full of inodes. In these buffers,
1791 * the only data which should be recovered is that which corresponds
1792 * to the di_next_unlinked pointers in the on disk inode structures.
1793 * The rest of the data for the inodes is always logged through the
1794 * inodes themselves rather than the inode buffer and is recovered
1795 * in xlog_recover_do_inode_trans().
1797 * The only time when buffers full of inodes are fully recovered is
1798 * when the buffer is full of newly allocated inodes. In this case
1799 * the buffer will not be marked as an inode buffer and so will be
1800 * sent to xlog_recover_do_reg_buffer() below during recovery.
1803 xlog_recover_do_inode_buffer(
1805 xlog_recover_item_t
*item
,
1807 xfs_buf_log_format_t
*buf_f
)
1815 int next_unlinked_offset
;
1817 xfs_agino_t
*logged_nextp
;
1818 xfs_agino_t
*buffer_nextp
;
1819 unsigned int *data_map
= NULL
;
1820 unsigned int map_size
= 0;
1822 trace_xfs_log_recover_buf_inode_buf(mp
->m_log
, buf_f
);
1824 switch (buf_f
->blf_type
) {
1826 data_map
= buf_f
->blf_data_map
;
1827 map_size
= buf_f
->blf_map_size
;
1831 * Set the variables corresponding to the current region to
1832 * 0 so that we'll initialize them on the first pass through
1840 inodes_per_buf
= XFS_BUF_COUNT(bp
) >> mp
->m_sb
.sb_inodelog
;
1841 for (i
= 0; i
< inodes_per_buf
; i
++) {
1842 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1843 offsetof(xfs_dinode_t
, di_next_unlinked
);
1845 while (next_unlinked_offset
>=
1846 (reg_buf_offset
+ reg_buf_bytes
)) {
1848 * The next di_next_unlinked field is beyond
1849 * the current logged region. Find the next
1850 * logged region that contains or is beyond
1851 * the current di_next_unlinked field.
1854 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1857 * If there are no more logged regions in the
1858 * buffer, then we're done.
1864 nbits
= xfs_contig_bits(data_map
, map_size
,
1867 reg_buf_offset
= bit
<< XFS_BLF_SHIFT
;
1868 reg_buf_bytes
= nbits
<< XFS_BLF_SHIFT
;
1873 * If the current logged region starts after the current
1874 * di_next_unlinked field, then move on to the next
1875 * di_next_unlinked field.
1877 if (next_unlinked_offset
< reg_buf_offset
) {
1881 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1882 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLF_CHUNK
) == 0);
1883 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <= XFS_BUF_COUNT(bp
));
1886 * The current logged region contains a copy of the
1887 * current di_next_unlinked field. Extract its value
1888 * and copy it to the buffer copy.
1890 logged_nextp
= item
->ri_buf
[item_index
].i_addr
+
1891 next_unlinked_offset
- reg_buf_offset
;
1892 if (unlikely(*logged_nextp
== 0)) {
1893 xfs_fs_cmn_err(CE_ALERT
, mp
,
1894 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1896 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1897 XFS_ERRLEVEL_LOW
, mp
);
1898 return XFS_ERROR(EFSCORRUPTED
);
1901 buffer_nextp
= (xfs_agino_t
*)xfs_buf_offset(bp
,
1902 next_unlinked_offset
);
1903 *buffer_nextp
= *logged_nextp
;
1910 * Perform a 'normal' buffer recovery. Each logged region of the
1911 * buffer should be copied over the corresponding region in the
1912 * given buffer. The bitmap in the buf log format structure indicates
1913 * where to place the logged data.
1917 xlog_recover_do_reg_buffer(
1918 struct xfs_mount
*mp
,
1919 xlog_recover_item_t
*item
,
1921 xfs_buf_log_format_t
*buf_f
)
1926 unsigned int *data_map
= NULL
;
1927 unsigned int map_size
= 0;
1930 trace_xfs_log_recover_buf_reg_buf(mp
->m_log
, buf_f
);
1932 switch (buf_f
->blf_type
) {
1934 data_map
= buf_f
->blf_data_map
;
1935 map_size
= buf_f
->blf_map_size
;
1939 i
= 1; /* 0 is the buf format structure */
1941 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1944 nbits
= xfs_contig_bits(data_map
, map_size
, bit
);
1946 ASSERT(item
->ri_buf
[i
].i_addr
!= NULL
);
1947 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLF_CHUNK
== 0);
1948 ASSERT(XFS_BUF_COUNT(bp
) >=
1949 ((uint
)bit
<< XFS_BLF_SHIFT
)+(nbits
<<XFS_BLF_SHIFT
));
1952 * Do a sanity check if this is a dquot buffer. Just checking
1953 * the first dquot in the buffer should do. XXXThis is
1954 * probably a good thing to do for other buf types also.
1957 if (buf_f
->blf_flags
&
1958 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
1959 if (item
->ri_buf
[i
].i_addr
== NULL
) {
1961 "XFS: NULL dquot in %s.", __func__
);
1964 if (item
->ri_buf
[i
].i_len
< sizeof(xfs_disk_dquot_t
)) {
1966 "XFS: dquot too small (%d) in %s.",
1967 item
->ri_buf
[i
].i_len
, __func__
);
1970 error
= xfs_qm_dqcheck(item
->ri_buf
[i
].i_addr
,
1971 -1, 0, XFS_QMOPT_DOWARN
,
1972 "dquot_buf_recover");
1977 memcpy(xfs_buf_offset(bp
,
1978 (uint
)bit
<< XFS_BLF_SHIFT
), /* dest */
1979 item
->ri_buf
[i
].i_addr
, /* source */
1980 nbits
<<XFS_BLF_SHIFT
); /* length */
1986 /* Shouldn't be any more regions */
1987 ASSERT(i
== item
->ri_total
);
1991 * Do some primitive error checking on ondisk dquot data structures.
1995 xfs_disk_dquot_t
*ddq
,
1997 uint type
, /* used only when IO_dorepair is true */
2001 xfs_dqblk_t
*d
= (xfs_dqblk_t
*)ddq
;
2005 * We can encounter an uninitialized dquot buffer for 2 reasons:
2006 * 1. If we crash while deleting the quotainode(s), and those blks got
2007 * used for user data. This is because we take the path of regular
2008 * file deletion; however, the size field of quotainodes is never
2009 * updated, so all the tricks that we play in itruncate_finish
2010 * don't quite matter.
2012 * 2. We don't play the quota buffers when there's a quotaoff logitem.
2013 * But the allocation will be replayed so we'll end up with an
2014 * uninitialized quota block.
2016 * This is all fine; things are still consistent, and we haven't lost
2017 * any quota information. Just don't complain about bad dquot blks.
2019 if (be16_to_cpu(ddq
->d_magic
) != XFS_DQUOT_MAGIC
) {
2020 if (flags
& XFS_QMOPT_DOWARN
)
2022 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
2023 str
, id
, be16_to_cpu(ddq
->d_magic
), XFS_DQUOT_MAGIC
);
2026 if (ddq
->d_version
!= XFS_DQUOT_VERSION
) {
2027 if (flags
& XFS_QMOPT_DOWARN
)
2029 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2030 str
, id
, ddq
->d_version
, XFS_DQUOT_VERSION
);
2034 if (ddq
->d_flags
!= XFS_DQ_USER
&&
2035 ddq
->d_flags
!= XFS_DQ_PROJ
&&
2036 ddq
->d_flags
!= XFS_DQ_GROUP
) {
2037 if (flags
& XFS_QMOPT_DOWARN
)
2039 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2040 str
, id
, ddq
->d_flags
);
2044 if (id
!= -1 && id
!= be32_to_cpu(ddq
->d_id
)) {
2045 if (flags
& XFS_QMOPT_DOWARN
)
2047 "%s : ondisk-dquot 0x%p, ID mismatch: "
2048 "0x%x expected, found id 0x%x",
2049 str
, ddq
, id
, be32_to_cpu(ddq
->d_id
));
2053 if (!errs
&& ddq
->d_id
) {
2054 if (ddq
->d_blk_softlimit
&&
2055 be64_to_cpu(ddq
->d_bcount
) >=
2056 be64_to_cpu(ddq
->d_blk_softlimit
)) {
2057 if (!ddq
->d_btimer
) {
2058 if (flags
& XFS_QMOPT_DOWARN
)
2060 "%s : Dquot ID 0x%x (0x%p) "
2061 "BLK TIMER NOT STARTED",
2062 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2066 if (ddq
->d_ino_softlimit
&&
2067 be64_to_cpu(ddq
->d_icount
) >=
2068 be64_to_cpu(ddq
->d_ino_softlimit
)) {
2069 if (!ddq
->d_itimer
) {
2070 if (flags
& XFS_QMOPT_DOWARN
)
2072 "%s : Dquot ID 0x%x (0x%p) "
2073 "INODE TIMER NOT STARTED",
2074 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2078 if (ddq
->d_rtb_softlimit
&&
2079 be64_to_cpu(ddq
->d_rtbcount
) >=
2080 be64_to_cpu(ddq
->d_rtb_softlimit
)) {
2081 if (!ddq
->d_rtbtimer
) {
2082 if (flags
& XFS_QMOPT_DOWARN
)
2084 "%s : Dquot ID 0x%x (0x%p) "
2085 "RTBLK TIMER NOT STARTED",
2086 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2092 if (!errs
|| !(flags
& XFS_QMOPT_DQREPAIR
))
2095 if (flags
& XFS_QMOPT_DOWARN
)
2096 cmn_err(CE_NOTE
, "Re-initializing dquot ID 0x%x", id
);
2099 * Typically, a repair is only requested by quotacheck.
2102 ASSERT(flags
& XFS_QMOPT_DQREPAIR
);
2103 memset(d
, 0, sizeof(xfs_dqblk_t
));
2105 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
2106 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
2107 d
->dd_diskdq
.d_flags
= type
;
2108 d
->dd_diskdq
.d_id
= cpu_to_be32(id
);
2114 * Perform a dquot buffer recovery.
2115 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2116 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2117 * Else, treat it as a regular buffer and do recovery.
2120 xlog_recover_do_dquot_buffer(
2123 xlog_recover_item_t
*item
,
2125 xfs_buf_log_format_t
*buf_f
)
2129 trace_xfs_log_recover_buf_dquot_buf(log
, buf_f
);
2132 * Filesystems are required to send in quota flags at mount time.
2134 if (mp
->m_qflags
== 0) {
2139 if (buf_f
->blf_flags
& XFS_BLF_UDQUOT_BUF
)
2140 type
|= XFS_DQ_USER
;
2141 if (buf_f
->blf_flags
& XFS_BLF_PDQUOT_BUF
)
2142 type
|= XFS_DQ_PROJ
;
2143 if (buf_f
->blf_flags
& XFS_BLF_GDQUOT_BUF
)
2144 type
|= XFS_DQ_GROUP
;
2146 * This type of quotas was turned off, so ignore this buffer
2148 if (log
->l_quotaoffs_flag
& type
)
2151 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2155 * This routine replays a modification made to a buffer at runtime.
2156 * There are actually two types of buffer, regular and inode, which
2157 * are handled differently. Inode buffers are handled differently
2158 * in that we only recover a specific set of data from them, namely
2159 * the inode di_next_unlinked fields. This is because all other inode
2160 * data is actually logged via inode records and any data we replay
2161 * here which overlaps that may be stale.
2163 * When meta-data buffers are freed at run time we log a buffer item
2164 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2165 * of the buffer in the log should not be replayed at recovery time.
2166 * This is so that if the blocks covered by the buffer are reused for
2167 * file data before we crash we don't end up replaying old, freed
2168 * meta-data into a user's file.
2170 * To handle the cancellation of buffer log items, we make two passes
2171 * over the log during recovery. During the first we build a table of
2172 * those buffers which have been cancelled, and during the second we
2173 * only replay those buffers which do not have corresponding cancel
2174 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2175 * for more details on the implementation of the table of cancel records.
2178 xlog_recover_do_buffer_trans(
2180 xlog_recover_item_t
*item
,
2183 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
2193 if (pass
== XLOG_RECOVER_PASS1
) {
2195 * In this pass we're only looking for buf items
2196 * with the XFS_BLF_CANCEL bit set.
2198 xlog_recover_do_buffer_pass1(log
, buf_f
);
2202 * In this pass we want to recover all the buffers
2203 * which have not been cancelled and are not
2204 * cancellation buffers themselves. The routine
2205 * we call here will tell us whether or not to
2206 * continue with the replay of this buffer.
2208 cancel
= xlog_recover_do_buffer_pass2(log
, buf_f
);
2210 trace_xfs_log_recover_buf_cancel(log
, buf_f
);
2214 trace_xfs_log_recover_buf_recover(log
, buf_f
);
2215 switch (buf_f
->blf_type
) {
2217 blkno
= buf_f
->blf_blkno
;
2218 len
= buf_f
->blf_len
;
2219 flags
= buf_f
->blf_flags
;
2222 xfs_fs_cmn_err(CE_ALERT
, log
->l_mp
,
2223 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2224 buf_f
->blf_type
, log
->l_mp
->m_logname
?
2225 log
->l_mp
->m_logname
: "internal");
2226 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2227 XFS_ERRLEVEL_LOW
, log
->l_mp
);
2228 return XFS_ERROR(EFSCORRUPTED
);
2232 buf_flags
= XBF_LOCK
;
2233 if (!(flags
& XFS_BLF_INODE_BUF
))
2234 buf_flags
|= XBF_MAPPED
;
2236 bp
= xfs_buf_read(mp
->m_ddev_targp
, blkno
, len
, buf_flags
);
2237 if (XFS_BUF_ISERROR(bp
)) {
2238 xfs_ioerror_alert("xlog_recover_do..(read#1)", log
->l_mp
,
2240 error
= XFS_BUF_GETERROR(bp
);
2246 if (flags
& XFS_BLF_INODE_BUF
) {
2247 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2249 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2250 xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2252 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2255 return XFS_ERROR(error
);
2258 * Perform delayed write on the buffer. Asynchronous writes will be
2259 * slower when taking into account all the buffers to be flushed.
2261 * Also make sure that only inode buffers with good sizes stay in
2262 * the buffer cache. The kernel moves inodes in buffers of 1 block
2263 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2264 * buffers in the log can be a different size if the log was generated
2265 * by an older kernel using unclustered inode buffers or a newer kernel
2266 * running with a different inode cluster size. Regardless, if the
2267 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2268 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2269 * the buffer out of the buffer cache so that the buffer won't
2270 * overlap with future reads of those inodes.
2272 if (XFS_DINODE_MAGIC
==
2273 be16_to_cpu(*((__be16
*)xfs_buf_offset(bp
, 0))) &&
2274 (XFS_BUF_COUNT(bp
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2275 (__uint32_t
)XFS_INODE_CLUSTER_SIZE(log
->l_mp
)))) {
2277 error
= xfs_bwrite(mp
, bp
);
2279 ASSERT(bp
->b_mount
== NULL
|| bp
->b_mount
== mp
);
2281 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2282 xfs_bdwrite(mp
, bp
);
2289 xlog_recover_do_inode_trans(
2291 xlog_recover_item_t
*item
,
2294 xfs_inode_log_format_t
*in_f
;
2305 xfs_icdinode_t
*dicp
;
2308 if (pass
== XLOG_RECOVER_PASS1
) {
2312 if (item
->ri_buf
[0].i_len
== sizeof(xfs_inode_log_format_t
)) {
2313 in_f
= item
->ri_buf
[0].i_addr
;
2315 in_f
= kmem_alloc(sizeof(xfs_inode_log_format_t
), KM_SLEEP
);
2317 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
2321 ino
= in_f
->ilf_ino
;
2325 * Inode buffers can be freed, look out for it,
2326 * and do not replay the inode.
2328 if (xlog_check_buffer_cancelled(log
, in_f
->ilf_blkno
,
2329 in_f
->ilf_len
, 0)) {
2331 trace_xfs_log_recover_inode_cancel(log
, in_f
);
2334 trace_xfs_log_recover_inode_recover(log
, in_f
);
2336 bp
= xfs_buf_read(mp
->m_ddev_targp
, in_f
->ilf_blkno
, in_f
->ilf_len
,
2338 if (XFS_BUF_ISERROR(bp
)) {
2339 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp
,
2340 bp
, in_f
->ilf_blkno
);
2341 error
= XFS_BUF_GETERROR(bp
);
2346 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2347 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, in_f
->ilf_boffset
);
2350 * Make sure the place we're flushing out to really looks
2353 if (unlikely(be16_to_cpu(dip
->di_magic
) != XFS_DINODE_MAGIC
)) {
2355 xfs_fs_cmn_err(CE_ALERT
, mp
,
2356 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2358 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2359 XFS_ERRLEVEL_LOW
, mp
);
2360 error
= EFSCORRUPTED
;
2363 dicp
= item
->ri_buf
[1].i_addr
;
2364 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2366 xfs_fs_cmn_err(CE_ALERT
, mp
,
2367 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2369 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2370 XFS_ERRLEVEL_LOW
, mp
);
2371 error
= EFSCORRUPTED
;
2375 /* Skip replay when the on disk inode is newer than the log one */
2376 if (dicp
->di_flushiter
< be16_to_cpu(dip
->di_flushiter
)) {
2378 * Deal with the wrap case, DI_MAX_FLUSH is less
2379 * than smaller numbers
2381 if (be16_to_cpu(dip
->di_flushiter
) == DI_MAX_FLUSH
&&
2382 dicp
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
2386 trace_xfs_log_recover_inode_skip(log
, in_f
);
2391 /* Take the opportunity to reset the flush iteration count */
2392 dicp
->di_flushiter
= 0;
2394 if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFREG
)) {
2395 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2396 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2397 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2398 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2400 xfs_fs_cmn_err(CE_ALERT
, mp
,
2401 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2402 item
, dip
, bp
, ino
);
2403 error
= EFSCORRUPTED
;
2406 } else if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFDIR
)) {
2407 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2408 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2409 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2410 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2411 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2413 xfs_fs_cmn_err(CE_ALERT
, mp
,
2414 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2415 item
, dip
, bp
, ino
);
2416 error
= EFSCORRUPTED
;
2420 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2421 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2422 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2424 xfs_fs_cmn_err(CE_ALERT
, mp
,
2425 "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",
2427 dicp
->di_nextents
+ dicp
->di_anextents
,
2429 error
= EFSCORRUPTED
;
2432 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2433 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2434 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2436 xfs_fs_cmn_err(CE_ALERT
, mp
,
2437 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2438 item
, dip
, bp
, ino
, dicp
->di_forkoff
);
2439 error
= EFSCORRUPTED
;
2442 if (unlikely(item
->ri_buf
[1].i_len
> sizeof(struct xfs_icdinode
))) {
2443 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2444 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2446 xfs_fs_cmn_err(CE_ALERT
, mp
,
2447 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2448 item
->ri_buf
[1].i_len
, item
);
2449 error
= EFSCORRUPTED
;
2453 /* The core is in in-core format */
2454 xfs_dinode_to_disk(dip
, item
->ri_buf
[1].i_addr
);
2456 /* the rest is in on-disk format */
2457 if (item
->ri_buf
[1].i_len
> sizeof(struct xfs_icdinode
)) {
2458 memcpy((xfs_caddr_t
) dip
+ sizeof(struct xfs_icdinode
),
2459 item
->ri_buf
[1].i_addr
+ sizeof(struct xfs_icdinode
),
2460 item
->ri_buf
[1].i_len
- sizeof(struct xfs_icdinode
));
2463 fields
= in_f
->ilf_fields
;
2464 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
2466 xfs_dinode_put_rdev(dip
, in_f
->ilf_u
.ilfu_rdev
);
2469 memcpy(XFS_DFORK_DPTR(dip
),
2470 &in_f
->ilf_u
.ilfu_uuid
,
2475 if (in_f
->ilf_size
== 2)
2476 goto write_inode_buffer
;
2477 len
= item
->ri_buf
[2].i_len
;
2478 src
= item
->ri_buf
[2].i_addr
;
2479 ASSERT(in_f
->ilf_size
<= 4);
2480 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
2481 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
2482 (len
== in_f
->ilf_dsize
));
2484 switch (fields
& XFS_ILOG_DFORK
) {
2485 case XFS_ILOG_DDATA
:
2487 memcpy(XFS_DFORK_DPTR(dip
), src
, len
);
2490 case XFS_ILOG_DBROOT
:
2491 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
, len
,
2492 (xfs_bmdr_block_t
*)XFS_DFORK_DPTR(dip
),
2493 XFS_DFORK_DSIZE(dip
, mp
));
2498 * There are no data fork flags set.
2500 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
2505 * If we logged any attribute data, recover it. There may or
2506 * may not have been any other non-core data logged in this
2509 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2510 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
2515 len
= item
->ri_buf
[attr_index
].i_len
;
2516 src
= item
->ri_buf
[attr_index
].i_addr
;
2517 ASSERT(len
== in_f
->ilf_asize
);
2519 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2520 case XFS_ILOG_ADATA
:
2522 dest
= XFS_DFORK_APTR(dip
);
2523 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
2524 memcpy(dest
, src
, len
);
2527 case XFS_ILOG_ABROOT
:
2528 dest
= XFS_DFORK_APTR(dip
);
2529 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
,
2530 len
, (xfs_bmdr_block_t
*)dest
,
2531 XFS_DFORK_ASIZE(dip
, mp
));
2535 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2544 ASSERT(bp
->b_mount
== NULL
|| bp
->b_mount
== mp
);
2546 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2547 xfs_bdwrite(mp
, bp
);
2551 return XFS_ERROR(error
);
2555 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2556 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2560 xlog_recover_do_quotaoff_trans(
2562 xlog_recover_item_t
*item
,
2565 xfs_qoff_logformat_t
*qoff_f
;
2567 if (pass
== XLOG_RECOVER_PASS2
) {
2571 qoff_f
= item
->ri_buf
[0].i_addr
;
2575 * The logitem format's flag tells us if this was user quotaoff,
2576 * group/project quotaoff or both.
2578 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2579 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2580 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
2581 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
2582 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2583 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2589 * Recover a dquot record
2592 xlog_recover_do_dquot_trans(
2594 xlog_recover_item_t
*item
,
2599 struct xfs_disk_dquot
*ddq
, *recddq
;
2601 xfs_dq_logformat_t
*dq_f
;
2604 if (pass
== XLOG_RECOVER_PASS1
) {
2610 * Filesystems are required to send in quota flags at mount time.
2612 if (mp
->m_qflags
== 0)
2615 recddq
= item
->ri_buf
[1].i_addr
;
2616 if (recddq
== NULL
) {
2618 "XFS: NULL dquot in %s.", __func__
);
2619 return XFS_ERROR(EIO
);
2621 if (item
->ri_buf
[1].i_len
< sizeof(xfs_disk_dquot_t
)) {
2623 "XFS: dquot too small (%d) in %s.",
2624 item
->ri_buf
[1].i_len
, __func__
);
2625 return XFS_ERROR(EIO
);
2629 * This type of quotas was turned off, so ignore this record.
2631 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
2633 if (log
->l_quotaoffs_flag
& type
)
2637 * At this point we know that quota was _not_ turned off.
2638 * Since the mount flags are not indicating to us otherwise, this
2639 * must mean that quota is on, and the dquot needs to be replayed.
2640 * Remember that we may not have fully recovered the superblock yet,
2641 * so we can't do the usual trick of looking at the SB quota bits.
2643 * The other possibility, of course, is that the quota subsystem was
2644 * removed since the last mount - ENOSYS.
2646 dq_f
= item
->ri_buf
[0].i_addr
;
2648 if ((error
= xfs_qm_dqcheck(recddq
,
2650 0, XFS_QMOPT_DOWARN
,
2651 "xlog_recover_do_dquot_trans (log copy)"))) {
2652 return XFS_ERROR(EIO
);
2654 ASSERT(dq_f
->qlf_len
== 1);
2656 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
2658 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
),
2661 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp
,
2662 bp
, dq_f
->qlf_blkno
);
2666 ddq
= (xfs_disk_dquot_t
*)xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
2669 * At least the magic num portion should be on disk because this
2670 * was among a chunk of dquots created earlier, and we did some
2671 * minimal initialization then.
2673 if (xfs_qm_dqcheck(ddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2674 "xlog_recover_do_dquot_trans")) {
2676 return XFS_ERROR(EIO
);
2679 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
2681 ASSERT(dq_f
->qlf_size
== 2);
2682 ASSERT(bp
->b_mount
== NULL
|| bp
->b_mount
== mp
);
2684 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2685 xfs_bdwrite(mp
, bp
);
2691 * This routine is called to create an in-core extent free intent
2692 * item from the efi format structure which was logged on disk.
2693 * It allocates an in-core efi, copies the extents from the format
2694 * structure into it, and adds the efi to the AIL with the given
2698 xlog_recover_do_efi_trans(
2700 xlog_recover_item_t
*item
,
2706 xfs_efi_log_item_t
*efip
;
2707 xfs_efi_log_format_t
*efi_formatp
;
2709 if (pass
== XLOG_RECOVER_PASS1
) {
2713 efi_formatp
= item
->ri_buf
[0].i_addr
;
2716 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
2717 if ((error
= xfs_efi_copy_format(&(item
->ri_buf
[0]),
2718 &(efip
->efi_format
)))) {
2719 xfs_efi_item_free(efip
);
2722 efip
->efi_next_extent
= efi_formatp
->efi_nextents
;
2723 efip
->efi_flags
|= XFS_EFI_COMMITTED
;
2725 spin_lock(&log
->l_ailp
->xa_lock
);
2727 * xfs_trans_ail_update() drops the AIL lock.
2729 xfs_trans_ail_update(log
->l_ailp
, (xfs_log_item_t
*)efip
, lsn
);
2735 * This routine is called when an efd format structure is found in
2736 * a committed transaction in the log. It's purpose is to cancel
2737 * the corresponding efi if it was still in the log. To do this
2738 * it searches the AIL for the efi with an id equal to that in the
2739 * efd format structure. If we find it, we remove the efi from the
2743 xlog_recover_do_efd_trans(
2745 xlog_recover_item_t
*item
,
2748 xfs_efd_log_format_t
*efd_formatp
;
2749 xfs_efi_log_item_t
*efip
= NULL
;
2750 xfs_log_item_t
*lip
;
2752 struct xfs_ail_cursor cur
;
2753 struct xfs_ail
*ailp
= log
->l_ailp
;
2755 if (pass
== XLOG_RECOVER_PASS1
) {
2759 efd_formatp
= item
->ri_buf
[0].i_addr
;
2760 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
2761 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
2762 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
2763 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
2764 efi_id
= efd_formatp
->efd_efi_id
;
2767 * Search for the efi with the id in the efd format structure
2770 spin_lock(&ailp
->xa_lock
);
2771 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
2772 while (lip
!= NULL
) {
2773 if (lip
->li_type
== XFS_LI_EFI
) {
2774 efip
= (xfs_efi_log_item_t
*)lip
;
2775 if (efip
->efi_format
.efi_id
== efi_id
) {
2777 * xfs_trans_ail_delete() drops the
2780 xfs_trans_ail_delete(ailp
, lip
);
2781 xfs_efi_item_free(efip
);
2782 spin_lock(&ailp
->xa_lock
);
2786 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
2788 xfs_trans_ail_cursor_done(ailp
, &cur
);
2789 spin_unlock(&ailp
->xa_lock
);
2793 * Perform the transaction
2795 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2796 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2799 xlog_recover_do_trans(
2801 xlog_recover_t
*trans
,
2805 xlog_recover_item_t
*item
;
2807 error
= xlog_recover_reorder_trans(log
, trans
, pass
);
2811 list_for_each_entry(item
, &trans
->r_itemq
, ri_list
) {
2812 trace_xfs_log_recover_item_recover(log
, trans
, item
, pass
);
2813 switch (ITEM_TYPE(item
)) {
2815 error
= xlog_recover_do_buffer_trans(log
, item
, pass
);
2818 error
= xlog_recover_do_inode_trans(log
, item
, pass
);
2821 error
= xlog_recover_do_efi_trans(log
, item
,
2822 trans
->r_lsn
, pass
);
2825 xlog_recover_do_efd_trans(log
, item
, pass
);
2829 error
= xlog_recover_do_dquot_trans(log
, item
, pass
);
2831 case XFS_LI_QUOTAOFF
:
2832 error
= xlog_recover_do_quotaoff_trans(log
, item
,
2837 "XFS: invalid item type (%d) xlog_recover_do_trans", ITEM_TYPE(item
));
2839 error
= XFS_ERROR(EIO
);
2851 * Free up any resources allocated by the transaction
2853 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2856 xlog_recover_free_trans(
2857 xlog_recover_t
*trans
)
2859 xlog_recover_item_t
*item
, *n
;
2862 list_for_each_entry_safe(item
, n
, &trans
->r_itemq
, ri_list
) {
2863 /* Free the regions in the item. */
2864 list_del(&item
->ri_list
);
2865 for (i
= 0; i
< item
->ri_cnt
; i
++)
2866 kmem_free(item
->ri_buf
[i
].i_addr
);
2867 /* Free the item itself */
2868 kmem_free(item
->ri_buf
);
2871 /* Free the transaction recover structure */
2876 xlog_recover_commit_trans(
2878 xlog_recover_t
*trans
,
2883 hlist_del(&trans
->r_list
);
2884 if ((error
= xlog_recover_do_trans(log
, trans
, pass
)))
2886 xlog_recover_free_trans(trans
); /* no error */
2891 xlog_recover_unmount_trans(
2892 xlog_recover_t
*trans
)
2894 /* Do nothing now */
2895 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2900 * There are two valid states of the r_state field. 0 indicates that the
2901 * transaction structure is in a normal state. We have either seen the
2902 * start of the transaction or the last operation we added was not a partial
2903 * operation. If the last operation we added to the transaction was a
2904 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2906 * NOTE: skip LRs with 0 data length.
2909 xlog_recover_process_data(
2911 struct hlist_head rhash
[],
2912 xlog_rec_header_t
*rhead
,
2918 xlog_op_header_t
*ohead
;
2919 xlog_recover_t
*trans
;
2925 lp
= dp
+ be32_to_cpu(rhead
->h_len
);
2926 num_logops
= be32_to_cpu(rhead
->h_num_logops
);
2928 /* check the log format matches our own - else we can't recover */
2929 if (xlog_header_check_recover(log
->l_mp
, rhead
))
2930 return (XFS_ERROR(EIO
));
2932 while ((dp
< lp
) && num_logops
) {
2933 ASSERT(dp
+ sizeof(xlog_op_header_t
) <= lp
);
2934 ohead
= (xlog_op_header_t
*)dp
;
2935 dp
+= sizeof(xlog_op_header_t
);
2936 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
2937 ohead
->oh_clientid
!= XFS_LOG
) {
2939 "XFS: xlog_recover_process_data: bad clientid");
2941 return (XFS_ERROR(EIO
));
2943 tid
= be32_to_cpu(ohead
->oh_tid
);
2944 hash
= XLOG_RHASH(tid
);
2945 trans
= xlog_recover_find_tid(&rhash
[hash
], tid
);
2946 if (trans
== NULL
) { /* not found; add new tid */
2947 if (ohead
->oh_flags
& XLOG_START_TRANS
)
2948 xlog_recover_new_tid(&rhash
[hash
], tid
,
2949 be64_to_cpu(rhead
->h_lsn
));
2951 if (dp
+ be32_to_cpu(ohead
->oh_len
) > lp
) {
2953 "XFS: xlog_recover_process_data: bad length");
2955 return (XFS_ERROR(EIO
));
2957 flags
= ohead
->oh_flags
& ~XLOG_END_TRANS
;
2958 if (flags
& XLOG_WAS_CONT_TRANS
)
2959 flags
&= ~XLOG_CONTINUE_TRANS
;
2961 case XLOG_COMMIT_TRANS
:
2962 error
= xlog_recover_commit_trans(log
,
2965 case XLOG_UNMOUNT_TRANS
:
2966 error
= xlog_recover_unmount_trans(trans
);
2968 case XLOG_WAS_CONT_TRANS
:
2969 error
= xlog_recover_add_to_cont_trans(log
,
2971 be32_to_cpu(ohead
->oh_len
));
2973 case XLOG_START_TRANS
:
2975 "XFS: xlog_recover_process_data: bad transaction");
2977 error
= XFS_ERROR(EIO
);
2980 case XLOG_CONTINUE_TRANS
:
2981 error
= xlog_recover_add_to_trans(log
, trans
,
2982 dp
, be32_to_cpu(ohead
->oh_len
));
2986 "XFS: xlog_recover_process_data: bad flag");
2988 error
= XFS_ERROR(EIO
);
2994 dp
+= be32_to_cpu(ohead
->oh_len
);
3001 * Process an extent free intent item that was recovered from
3002 * the log. We need to free the extents that it describes.
3005 xlog_recover_process_efi(
3007 xfs_efi_log_item_t
*efip
)
3009 xfs_efd_log_item_t
*efdp
;
3014 xfs_fsblock_t startblock_fsb
;
3016 ASSERT(!(efip
->efi_flags
& XFS_EFI_RECOVERED
));
3019 * First check the validity of the extents described by the
3020 * EFI. If any are bad, then assume that all are bad and
3021 * just toss the EFI.
3023 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3024 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3025 startblock_fsb
= XFS_BB_TO_FSB(mp
,
3026 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
3027 if ((startblock_fsb
== 0) ||
3028 (extp
->ext_len
== 0) ||
3029 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
3030 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
3032 * This will pull the EFI from the AIL and
3033 * free the memory associated with it.
3035 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
3036 return XFS_ERROR(EIO
);
3040 tp
= xfs_trans_alloc(mp
, 0);
3041 error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0, 0, 0);
3044 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
3046 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3047 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3048 error
= xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
3051 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
3055 efip
->efi_flags
|= XFS_EFI_RECOVERED
;
3056 error
= xfs_trans_commit(tp
, 0);
3060 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3065 * When this is called, all of the EFIs which did not have
3066 * corresponding EFDs should be in the AIL. What we do now
3067 * is free the extents associated with each one.
3069 * Since we process the EFIs in normal transactions, they
3070 * will be removed at some point after the commit. This prevents
3071 * us from just walking down the list processing each one.
3072 * We'll use a flag in the EFI to skip those that we've already
3073 * processed and use the AIL iteration mechanism's generation
3074 * count to try to speed this up at least a bit.
3076 * When we start, we know that the EFIs are the only things in
3077 * the AIL. As we process them, however, other items are added
3078 * to the AIL. Since everything added to the AIL must come after
3079 * everything already in the AIL, we stop processing as soon as
3080 * we see something other than an EFI in the AIL.
3083 xlog_recover_process_efis(
3086 xfs_log_item_t
*lip
;
3087 xfs_efi_log_item_t
*efip
;
3089 struct xfs_ail_cursor cur
;
3090 struct xfs_ail
*ailp
;
3093 spin_lock(&ailp
->xa_lock
);
3094 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3095 while (lip
!= NULL
) {
3097 * We're done when we see something other than an EFI.
3098 * There should be no EFIs left in the AIL now.
3100 if (lip
->li_type
!= XFS_LI_EFI
) {
3102 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
3103 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
3109 * Skip EFIs that we've already processed.
3111 efip
= (xfs_efi_log_item_t
*)lip
;
3112 if (efip
->efi_flags
& XFS_EFI_RECOVERED
) {
3113 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3117 spin_unlock(&ailp
->xa_lock
);
3118 error
= xlog_recover_process_efi(log
->l_mp
, efip
);
3119 spin_lock(&ailp
->xa_lock
);
3122 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3125 xfs_trans_ail_cursor_done(ailp
, &cur
);
3126 spin_unlock(&ailp
->xa_lock
);
3131 * This routine performs a transaction to null out a bad inode pointer
3132 * in an agi unlinked inode hash bucket.
3135 xlog_recover_clear_agi_bucket(
3137 xfs_agnumber_t agno
,
3146 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3147 error
= xfs_trans_reserve(tp
, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp
),
3152 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
3156 agi
= XFS_BUF_TO_AGI(agibp
);
3157 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
3158 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3159 (sizeof(xfs_agino_t
) * bucket
);
3160 xfs_trans_log_buf(tp
, agibp
, offset
,
3161 (offset
+ sizeof(xfs_agino_t
) - 1));
3163 error
= xfs_trans_commit(tp
, 0);
3169 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3171 xfs_fs_cmn_err(CE_WARN
, mp
, "xlog_recover_clear_agi_bucket: "
3172 "failed to clear agi %d. Continuing.", agno
);
3177 xlog_recover_process_one_iunlink(
3178 struct xfs_mount
*mp
,
3179 xfs_agnumber_t agno
,
3183 struct xfs_buf
*ibp
;
3184 struct xfs_dinode
*dip
;
3185 struct xfs_inode
*ip
;
3189 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3190 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
);
3195 * Get the on disk inode to find the next inode in the bucket.
3197 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &ibp
, XBF_LOCK
);
3201 ASSERT(ip
->i_d
.di_nlink
== 0);
3202 ASSERT(ip
->i_d
.di_mode
!= 0);
3204 /* setup for the next pass */
3205 agino
= be32_to_cpu(dip
->di_next_unlinked
);
3209 * Prevent any DMAPI event from being sent when the reference on
3210 * the inode is dropped.
3212 ip
->i_d
.di_dmevmask
= 0;
3221 * We can't read in the inode this bucket points to, or this inode
3222 * is messed up. Just ditch this bucket of inodes. We will lose
3223 * some inodes and space, but at least we won't hang.
3225 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3226 * clear the inode pointer in the bucket.
3228 xlog_recover_clear_agi_bucket(mp
, agno
, bucket
);
3233 * xlog_iunlink_recover
3235 * This is called during recovery to process any inodes which
3236 * we unlinked but not freed when the system crashed. These
3237 * inodes will be on the lists in the AGI blocks. What we do
3238 * here is scan all the AGIs and fully truncate and free any
3239 * inodes found on the lists. Each inode is removed from the
3240 * lists when it has been fully truncated and is freed. The
3241 * freeing of the inode and its removal from the list must be
3245 xlog_recover_process_iunlinks(
3249 xfs_agnumber_t agno
;
3260 * Prevent any DMAPI event from being sent while in this function.
3262 mp_dmevmask
= mp
->m_dmevmask
;
3265 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3267 * Find the agi for this ag.
3269 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3272 * AGI is b0rked. Don't process it.
3274 * We should probably mark the filesystem as corrupt
3275 * after we've recovered all the ag's we can....
3279 agi
= XFS_BUF_TO_AGI(agibp
);
3281 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3282 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
3283 while (agino
!= NULLAGINO
) {
3285 * Release the agi buffer so that it can
3286 * be acquired in the normal course of the
3287 * transaction to truncate and free the inode.
3289 xfs_buf_relse(agibp
);
3291 agino
= xlog_recover_process_one_iunlink(mp
,
3292 agno
, agino
, bucket
);
3295 * Reacquire the agibuffer and continue around
3296 * the loop. This should never fail as we know
3297 * the buffer was good earlier on.
3299 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3301 agi
= XFS_BUF_TO_AGI(agibp
);
3306 * Release the buffer for the current agi so we can
3307 * go on to the next one.
3309 xfs_buf_relse(agibp
);
3312 mp
->m_dmevmask
= mp_dmevmask
;
3318 xlog_pack_data_checksum(
3320 xlog_in_core_t
*iclog
,
3327 up
= (__be32
*)iclog
->ic_datap
;
3328 /* divide length by 4 to get # words */
3329 for (i
= 0; i
< (size
>> 2); i
++) {
3330 chksum
^= be32_to_cpu(*up
);
3333 iclog
->ic_header
.h_chksum
= cpu_to_be32(chksum
);
3336 #define xlog_pack_data_checksum(log, iclog, size)
3340 * Stamp cycle number in every block
3345 xlog_in_core_t
*iclog
,
3349 int size
= iclog
->ic_offset
+ roundoff
;
3353 xlog_pack_data_checksum(log
, iclog
, size
);
3355 cycle_lsn
= CYCLE_LSN_DISK(iclog
->ic_header
.h_lsn
);
3357 dp
= iclog
->ic_datap
;
3358 for (i
= 0; i
< BTOBB(size
) &&
3359 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3360 iclog
->ic_header
.h_cycle_data
[i
] = *(__be32
*)dp
;
3361 *(__be32
*)dp
= cycle_lsn
;
3365 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3366 xlog_in_core_2_t
*xhdr
= iclog
->ic_data
;
3368 for ( ; i
< BTOBB(size
); i
++) {
3369 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3370 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3371 xhdr
[j
].hic_xheader
.xh_cycle_data
[k
] = *(__be32
*)dp
;
3372 *(__be32
*)dp
= cycle_lsn
;
3376 for (i
= 1; i
< log
->l_iclog_heads
; i
++) {
3377 xhdr
[i
].hic_xheader
.xh_cycle
= cycle_lsn
;
3384 xlog_rec_header_t
*rhead
,
3390 for (i
= 0; i
< BTOBB(be32_to_cpu(rhead
->h_len
)) &&
3391 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3392 *(__be32
*)dp
= *(__be32
*)&rhead
->h_cycle_data
[i
];
3396 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3397 xlog_in_core_2_t
*xhdr
= (xlog_in_core_2_t
*)rhead
;
3398 for ( ; i
< BTOBB(be32_to_cpu(rhead
->h_len
)); i
++) {
3399 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3400 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3401 *(__be32
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
3408 xlog_valid_rec_header(
3410 xlog_rec_header_t
*rhead
,
3415 if (unlikely(be32_to_cpu(rhead
->h_magicno
) != XLOG_HEADER_MAGIC_NUM
)) {
3416 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3417 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3418 return XFS_ERROR(EFSCORRUPTED
);
3421 (!rhead
->h_version
||
3422 (be32_to_cpu(rhead
->h_version
) & (~XLOG_VERSION_OKBITS
))))) {
3423 xlog_warn("XFS: %s: unrecognised log version (%d).",
3424 __func__
, be32_to_cpu(rhead
->h_version
));
3425 return XFS_ERROR(EIO
);
3428 /* LR body must have data or it wouldn't have been written */
3429 hlen
= be32_to_cpu(rhead
->h_len
);
3430 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
3431 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3432 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3433 return XFS_ERROR(EFSCORRUPTED
);
3435 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
3436 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3437 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3438 return XFS_ERROR(EFSCORRUPTED
);
3444 * Read the log from tail to head and process the log records found.
3445 * Handle the two cases where the tail and head are in the same cycle
3446 * and where the active portion of the log wraps around the end of
3447 * the physical log separately. The pass parameter is passed through
3448 * to the routines called to process the data and is not looked at
3452 xlog_do_recovery_pass(
3454 xfs_daddr_t head_blk
,
3455 xfs_daddr_t tail_blk
,
3458 xlog_rec_header_t
*rhead
;
3461 xfs_buf_t
*hbp
, *dbp
;
3462 int error
= 0, h_size
;
3463 int bblks
, split_bblks
;
3464 int hblks
, split_hblks
, wrapped_hblks
;
3465 struct hlist_head rhash
[XLOG_RHASH_SIZE
];
3467 ASSERT(head_blk
!= tail_blk
);
3470 * Read the header of the tail block and get the iclog buffer size from
3471 * h_size. Use this to tell how many sectors make up the log header.
3473 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3475 * When using variable length iclogs, read first sector of
3476 * iclog header and extract the header size from it. Get a
3477 * new hbp that is the correct size.
3479 hbp
= xlog_get_bp(log
, 1);
3483 error
= xlog_bread(log
, tail_blk
, 1, hbp
, &offset
);
3487 rhead
= (xlog_rec_header_t
*)offset
;
3488 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
3491 h_size
= be32_to_cpu(rhead
->h_size
);
3492 if ((be32_to_cpu(rhead
->h_version
) & XLOG_VERSION_2
) &&
3493 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
3494 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
3495 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
3498 hbp
= xlog_get_bp(log
, hblks
);
3503 ASSERT(log
->l_sectBBsize
== 1);
3505 hbp
= xlog_get_bp(log
, 1);
3506 h_size
= XLOG_BIG_RECORD_BSIZE
;
3511 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
3517 memset(rhash
, 0, sizeof(rhash
));
3518 if (tail_blk
<= head_blk
) {
3519 for (blk_no
= tail_blk
; blk_no
< head_blk
; ) {
3520 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3524 rhead
= (xlog_rec_header_t
*)offset
;
3525 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3529 /* blocks in data section */
3530 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3531 error
= xlog_bread(log
, blk_no
+ hblks
, bblks
, dbp
,
3536 xlog_unpack_data(rhead
, offset
, log
);
3537 if ((error
= xlog_recover_process_data(log
,
3538 rhash
, rhead
, offset
, pass
)))
3540 blk_no
+= bblks
+ hblks
;
3544 * Perform recovery around the end of the physical log.
3545 * When the head is not on the same cycle number as the tail,
3546 * we can't do a sequential recovery as above.
3549 while (blk_no
< log
->l_logBBsize
) {
3551 * Check for header wrapping around physical end-of-log
3553 offset
= XFS_BUF_PTR(hbp
);
3556 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
3557 /* Read header in one read */
3558 error
= xlog_bread(log
, blk_no
, hblks
, hbp
,
3563 /* This LR is split across physical log end */
3564 if (blk_no
!= log
->l_logBBsize
) {
3565 /* some data before physical log end */
3566 ASSERT(blk_no
<= INT_MAX
);
3567 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
3568 ASSERT(split_hblks
> 0);
3569 error
= xlog_bread(log
, blk_no
,
3577 * Note: this black magic still works with
3578 * large sector sizes (non-512) only because:
3579 * - we increased the buffer size originally
3580 * by 1 sector giving us enough extra space
3581 * for the second read;
3582 * - the log start is guaranteed to be sector
3584 * - we read the log end (LR header start)
3585 * _first_, then the log start (LR header end)
3586 * - order is important.
3588 wrapped_hblks
= hblks
- split_hblks
;
3589 error
= XFS_BUF_SET_PTR(hbp
,
3590 offset
+ BBTOB(split_hblks
),
3591 BBTOB(hblks
- split_hblks
));
3595 error
= xlog_bread_noalign(log
, 0,
3596 wrapped_hblks
, hbp
);
3600 error
= XFS_BUF_SET_PTR(hbp
, offset
,
3605 rhead
= (xlog_rec_header_t
*)offset
;
3606 error
= xlog_valid_rec_header(log
, rhead
,
3607 split_hblks
? blk_no
: 0);
3611 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3614 /* Read in data for log record */
3615 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
3616 error
= xlog_bread(log
, blk_no
, bblks
, dbp
,
3621 /* This log record is split across the
3622 * physical end of log */
3623 offset
= XFS_BUF_PTR(dbp
);
3625 if (blk_no
!= log
->l_logBBsize
) {
3626 /* some data is before the physical
3628 ASSERT(!wrapped_hblks
);
3629 ASSERT(blk_no
<= INT_MAX
);
3631 log
->l_logBBsize
- (int)blk_no
;
3632 ASSERT(split_bblks
> 0);
3633 error
= xlog_bread(log
, blk_no
,
3641 * Note: this black magic still works with
3642 * large sector sizes (non-512) only because:
3643 * - we increased the buffer size originally
3644 * by 1 sector giving us enough extra space
3645 * for the second read;
3646 * - the log start is guaranteed to be sector
3648 * - we read the log end (LR header start)
3649 * _first_, then the log start (LR header end)
3650 * - order is important.
3652 error
= XFS_BUF_SET_PTR(dbp
,
3653 offset
+ BBTOB(split_bblks
),
3654 BBTOB(bblks
- split_bblks
));
3658 error
= xlog_bread_noalign(log
, wrapped_hblks
,
3659 bblks
- split_bblks
,
3664 error
= XFS_BUF_SET_PTR(dbp
, offset
, h_size
);
3668 xlog_unpack_data(rhead
, offset
, log
);
3669 if ((error
= xlog_recover_process_data(log
, rhash
,
3670 rhead
, offset
, pass
)))
3675 ASSERT(blk_no
>= log
->l_logBBsize
);
3676 blk_no
-= log
->l_logBBsize
;
3678 /* read first part of physical log */
3679 while (blk_no
< head_blk
) {
3680 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3684 rhead
= (xlog_rec_header_t
*)offset
;
3685 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3689 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3690 error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
,
3695 xlog_unpack_data(rhead
, offset
, log
);
3696 if ((error
= xlog_recover_process_data(log
, rhash
,
3697 rhead
, offset
, pass
)))
3699 blk_no
+= bblks
+ hblks
;
3711 * Do the recovery of the log. We actually do this in two phases.
3712 * The two passes are necessary in order to implement the function
3713 * of cancelling a record written into the log. The first pass
3714 * determines those things which have been cancelled, and the
3715 * second pass replays log items normally except for those which
3716 * have been cancelled. The handling of the replay and cancellations
3717 * takes place in the log item type specific routines.
3719 * The table of items which have cancel records in the log is allocated
3720 * and freed at this level, since only here do we know when all of
3721 * the log recovery has been completed.
3724 xlog_do_log_recovery(
3726 xfs_daddr_t head_blk
,
3727 xfs_daddr_t tail_blk
)
3731 ASSERT(head_blk
!= tail_blk
);
3734 * First do a pass to find all of the cancelled buf log items.
3735 * Store them in the buf_cancel_table for use in the second pass.
3737 log
->l_buf_cancel_table
=
3738 (xfs_buf_cancel_t
**)kmem_zalloc(XLOG_BC_TABLE_SIZE
*
3739 sizeof(xfs_buf_cancel_t
*),
3741 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3742 XLOG_RECOVER_PASS1
);
3744 kmem_free(log
->l_buf_cancel_table
);
3745 log
->l_buf_cancel_table
= NULL
;
3749 * Then do a second pass to actually recover the items in the log.
3750 * When it is complete free the table of buf cancel items.
3752 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3753 XLOG_RECOVER_PASS2
);
3758 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
3759 ASSERT(log
->l_buf_cancel_table
[i
] == NULL
);
3763 kmem_free(log
->l_buf_cancel_table
);
3764 log
->l_buf_cancel_table
= NULL
;
3770 * Do the actual recovery
3775 xfs_daddr_t head_blk
,
3776 xfs_daddr_t tail_blk
)
3783 * First replay the images in the log.
3785 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
3790 XFS_bflush(log
->l_mp
->m_ddev_targp
);
3793 * If IO errors happened during recovery, bail out.
3795 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
3800 * We now update the tail_lsn since much of the recovery has completed
3801 * and there may be space available to use. If there were no extent
3802 * or iunlinks, we can free up the entire log and set the tail_lsn to
3803 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3804 * lsn of the last known good LR on disk. If there are extent frees
3805 * or iunlinks they will have some entries in the AIL; so we look at
3806 * the AIL to determine how to set the tail_lsn.
3808 xlog_assign_tail_lsn(log
->l_mp
);
3811 * Now that we've finished replaying all buffer and inode
3812 * updates, re-read in the superblock.
3814 bp
= xfs_getsb(log
->l_mp
, 0);
3816 ASSERT(!(XFS_BUF_ISWRITE(bp
)));
3817 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp
)));
3819 XFS_BUF_UNASYNC(bp
);
3820 xfsbdstrat(log
->l_mp
, bp
);
3821 error
= xfs_iowait(bp
);
3823 xfs_ioerror_alert("xlog_do_recover",
3824 log
->l_mp
, bp
, XFS_BUF_ADDR(bp
));
3830 /* Convert superblock from on-disk format */
3831 sbp
= &log
->l_mp
->m_sb
;
3832 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
3833 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
3834 ASSERT(xfs_sb_good_version(sbp
));
3837 /* We've re-read the superblock so re-initialize per-cpu counters */
3838 xfs_icsb_reinit_counters(log
->l_mp
);
3840 xlog_recover_check_summary(log
);
3842 /* Normal transactions can now occur */
3843 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
3848 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3850 * Return error or zero.
3856 xfs_daddr_t head_blk
, tail_blk
;
3859 /* find the tail of the log */
3860 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
)))
3863 if (tail_blk
!= head_blk
) {
3864 /* There used to be a comment here:
3866 * disallow recovery on read-only mounts. note -- mount
3867 * checks for ENOSPC and turns it into an intelligent
3869 * ...but this is no longer true. Now, unless you specify
3870 * NORECOVERY (in which case this function would never be
3871 * called), we just go ahead and recover. We do this all
3872 * under the vfs layer, so we can get away with it unless
3873 * the device itself is read-only, in which case we fail.
3875 if ((error
= xfs_dev_is_read_only(log
->l_mp
, "recovery"))) {
3880 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3881 log
->l_mp
->m_fsname
, log
->l_mp
->m_logname
?
3882 log
->l_mp
->m_logname
: "internal");
3884 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
3885 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
3891 * In the first part of recovery we replay inodes and buffers and build
3892 * up the list of extent free items which need to be processed. Here
3893 * we process the extent free items and clean up the on disk unlinked
3894 * inode lists. This is separated from the first part of recovery so
3895 * that the root and real-time bitmap inodes can be read in from disk in
3896 * between the two stages. This is necessary so that we can free space
3897 * in the real-time portion of the file system.
3900 xlog_recover_finish(
3904 * Now we're ready to do the transactions needed for the
3905 * rest of recovery. Start with completing all the extent
3906 * free intent records and then process the unlinked inode
3907 * lists. At this point, we essentially run in normal mode
3908 * except that we're still performing recovery actions
3909 * rather than accepting new requests.
3911 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
3913 error
= xlog_recover_process_efis(log
);
3916 "Failed to recover EFIs on filesystem: %s",
3917 log
->l_mp
->m_fsname
);
3921 * Sync the log to get all the EFIs out of the AIL.
3922 * This isn't absolutely necessary, but it helps in
3923 * case the unlink transactions would have problems
3924 * pushing the EFIs out of the way.
3926 xfs_log_force(log
->l_mp
, XFS_LOG_SYNC
);
3928 xlog_recover_process_iunlinks(log
);
3930 xlog_recover_check_summary(log
);
3933 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3934 log
->l_mp
->m_fsname
, log
->l_mp
->m_logname
?
3935 log
->l_mp
->m_logname
: "internal");
3936 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
3939 "!Ending clean XFS mount for filesystem: %s\n",
3940 log
->l_mp
->m_fsname
);
3948 * Read all of the agf and agi counters and check that they
3949 * are consistent with the superblock counters.
3952 xlog_recover_check_summary(
3959 xfs_agnumber_t agno
;
3960 __uint64_t freeblks
;
3970 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3971 error
= xfs_read_agf(mp
, NULL
, agno
, 0, &agfbp
);
3973 xfs_fs_cmn_err(CE_ALERT
, mp
,
3974 "xlog_recover_check_summary(agf)"
3975 "agf read failed agno %d error %d",
3978 agfp
= XFS_BUF_TO_AGF(agfbp
);
3979 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
3980 be32_to_cpu(agfp
->agf_flcount
);
3981 xfs_buf_relse(agfbp
);
3984 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3986 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agibp
);
3988 itotal
+= be32_to_cpu(agi
->agi_count
);
3989 ifree
+= be32_to_cpu(agi
->agi_freecount
);
3990 xfs_buf_relse(agibp
);