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_buf_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_target
->bt_mount
, bp
,
327 xfs_force_shutdown(bp
->b_target
->bt_mount
,
328 SHUTDOWN_META_IO_ERROR
);
330 XFS_BUF_CLR_IODONE_FUNC(bp
);
331 xfs_buf_ioend(bp
, 0);
335 * This routine finds (to an approximation) the first block in the physical
336 * log which contains the given cycle. It uses a binary search algorithm.
337 * Note that the algorithm can not be perfect because the disk will not
338 * necessarily be perfect.
341 xlog_find_cycle_start(
344 xfs_daddr_t first_blk
,
345 xfs_daddr_t
*last_blk
,
355 mid_blk
= BLK_AVG(first_blk
, end_blk
);
356 while (mid_blk
!= first_blk
&& mid_blk
!= end_blk
) {
357 error
= xlog_bread(log
, mid_blk
, 1, bp
, &offset
);
360 mid_cycle
= xlog_get_cycle(offset
);
361 if (mid_cycle
== cycle
)
362 end_blk
= mid_blk
; /* last_half_cycle == mid_cycle */
364 first_blk
= mid_blk
; /* first_half_cycle == mid_cycle */
365 mid_blk
= BLK_AVG(first_blk
, end_blk
);
367 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == end_blk
) ||
368 (mid_blk
== end_blk
&& mid_blk
-1 == first_blk
));
376 * Check that a range of blocks does not contain stop_on_cycle_no.
377 * Fill in *new_blk with the block offset where such a block is
378 * found, or with -1 (an invalid block number) if there is no such
379 * block in the range. The scan needs to occur from front to back
380 * and the pointer into the region must be updated since a later
381 * routine will need to perform another test.
384 xlog_find_verify_cycle(
386 xfs_daddr_t start_blk
,
388 uint stop_on_cycle_no
,
389 xfs_daddr_t
*new_blk
)
395 xfs_caddr_t buf
= NULL
;
399 * Greedily allocate a buffer big enough to handle the full
400 * range of basic blocks we'll be examining. If that fails,
401 * try a smaller size. We need to be able to read at least
402 * a log sector, or we're out of luck.
404 bufblks
= 1 << ffs(nbblks
);
405 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
407 if (bufblks
< log
->l_sectBBsize
)
411 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
414 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
416 error
= xlog_bread(log
, i
, bcount
, bp
, &buf
);
420 for (j
= 0; j
< bcount
; j
++) {
421 cycle
= xlog_get_cycle(buf
);
422 if (cycle
== stop_on_cycle_no
) {
439 * Potentially backup over partial log record write.
441 * In the typical case, last_blk is the number of the block directly after
442 * a good log record. Therefore, we subtract one to get the block number
443 * of the last block in the given buffer. extra_bblks contains the number
444 * of blocks we would have read on a previous read. This happens when the
445 * last log record is split over the end of the physical log.
447 * extra_bblks is the number of blocks potentially verified on a previous
448 * call to this routine.
451 xlog_find_verify_log_record(
453 xfs_daddr_t start_blk
,
454 xfs_daddr_t
*last_blk
,
459 xfs_caddr_t offset
= NULL
;
460 xlog_rec_header_t
*head
= NULL
;
463 int num_blks
= *last_blk
- start_blk
;
466 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
468 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
469 if (!(bp
= xlog_get_bp(log
, 1)))
473 error
= xlog_bread(log
, start_blk
, num_blks
, bp
, &offset
);
476 offset
+= ((num_blks
- 1) << BBSHIFT
);
479 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
481 /* valid log record not found */
483 "XFS: Log inconsistent (didn't find previous header)");
485 error
= XFS_ERROR(EIO
);
490 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
495 head
= (xlog_rec_header_t
*)offset
;
497 if (XLOG_HEADER_MAGIC_NUM
== be32_to_cpu(head
->h_magicno
))
505 * We hit the beginning of the physical log & still no header. Return
506 * to caller. If caller can handle a return of -1, then this routine
507 * will be called again for the end of the physical log.
515 * We have the final block of the good log (the first block
516 * of the log record _before_ the head. So we check the uuid.
518 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
522 * We may have found a log record header before we expected one.
523 * last_blk will be the 1st block # with a given cycle #. We may end
524 * up reading an entire log record. In this case, we don't want to
525 * reset last_blk. Only when last_blk points in the middle of a log
526 * record do we update last_blk.
528 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
529 uint h_size
= be32_to_cpu(head
->h_size
);
531 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
532 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
538 if (*last_blk
- i
+ extra_bblks
!=
539 BTOBB(be32_to_cpu(head
->h_len
)) + xhdrs
)
548 * Head is defined to be the point of the log where the next log write
549 * write could go. This means that incomplete LR writes at the end are
550 * eliminated when calculating the head. We aren't guaranteed that previous
551 * LR have complete transactions. We only know that a cycle number of
552 * current cycle number -1 won't be present in the log if we start writing
553 * from our current block number.
555 * last_blk contains the block number of the first block with a given
558 * Return: zero if normal, non-zero if error.
563 xfs_daddr_t
*return_head_blk
)
567 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
569 uint first_half_cycle
, last_half_cycle
;
571 int error
, log_bbnum
= log
->l_logBBsize
;
573 /* Is the end of the log device zeroed? */
574 if ((error
= xlog_find_zeroed(log
, &first_blk
)) == -1) {
575 *return_head_blk
= first_blk
;
577 /* Is the whole lot zeroed? */
579 /* Linux XFS shouldn't generate totally zeroed logs -
580 * mkfs etc write a dummy unmount record to a fresh
581 * log so we can store the uuid in there
583 xlog_warn("XFS: totally zeroed log");
588 xlog_warn("XFS: empty log check failed");
592 first_blk
= 0; /* get cycle # of 1st block */
593 bp
= xlog_get_bp(log
, 1);
597 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
601 first_half_cycle
= xlog_get_cycle(offset
);
603 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
604 error
= xlog_bread(log
, last_blk
, 1, bp
, &offset
);
608 last_half_cycle
= xlog_get_cycle(offset
);
609 ASSERT(last_half_cycle
!= 0);
612 * If the 1st half cycle number is equal to the last half cycle number,
613 * then the entire log is stamped with the same cycle number. In this
614 * case, head_blk can't be set to zero (which makes sense). The below
615 * math doesn't work out properly with head_blk equal to zero. Instead,
616 * we set it to log_bbnum which is an invalid block number, but this
617 * value makes the math correct. If head_blk doesn't changed through
618 * all the tests below, *head_blk is set to zero at the very end rather
619 * than log_bbnum. In a sense, log_bbnum and zero are the same block
620 * in a circular file.
622 if (first_half_cycle
== last_half_cycle
) {
624 * In this case we believe that the entire log should have
625 * cycle number last_half_cycle. We need to scan backwards
626 * from the end verifying that there are no holes still
627 * containing last_half_cycle - 1. If we find such a hole,
628 * then the start of that hole will be the new head. The
629 * simple case looks like
630 * x | x ... | x - 1 | x
631 * Another case that fits this picture would be
632 * x | x + 1 | x ... | x
633 * In this case the head really is somewhere at the end of the
634 * log, as one of the latest writes at the beginning was
637 * x | x + 1 | x ... | x - 1 | x
638 * This is really the combination of the above two cases, and
639 * the head has to end up at the start of the x-1 hole at the
642 * In the 256k log case, we will read from the beginning to the
643 * end of the log and search for cycle numbers equal to x-1.
644 * We don't worry about the x+1 blocks that we encounter,
645 * because we know that they cannot be the head since the log
648 head_blk
= log_bbnum
;
649 stop_on_cycle
= last_half_cycle
- 1;
652 * In this case we want to find the first block with cycle
653 * number matching last_half_cycle. We expect the log to be
655 * x + 1 ... | x ... | x
656 * The first block with cycle number x (last_half_cycle) will
657 * be where the new head belongs. First we do a binary search
658 * for the first occurrence of last_half_cycle. The binary
659 * search may not be totally accurate, so then we scan back
660 * from there looking for occurrences of last_half_cycle before
661 * us. If that backwards scan wraps around the beginning of
662 * the log, then we look for occurrences of last_half_cycle - 1
663 * at the end of the log. The cases we're looking for look
665 * v binary search stopped here
666 * x + 1 ... | x | x + 1 | x ... | x
667 * ^ but we want to locate this spot
669 * <---------> less than scan distance
670 * x + 1 ... | x ... | x - 1 | x
671 * ^ we want to locate this spot
673 stop_on_cycle
= last_half_cycle
;
674 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
675 &head_blk
, last_half_cycle
)))
680 * Now validate the answer. Scan back some number of maximum possible
681 * blocks and make sure each one has the expected cycle number. The
682 * maximum is determined by the total possible amount of buffering
683 * in the in-core log. The following number can be made tighter if
684 * we actually look at the block size of the filesystem.
686 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
687 if (head_blk
>= num_scan_bblks
) {
689 * We are guaranteed that the entire check can be performed
692 start_blk
= head_blk
- num_scan_bblks
;
693 if ((error
= xlog_find_verify_cycle(log
,
694 start_blk
, num_scan_bblks
,
695 stop_on_cycle
, &new_blk
)))
699 } else { /* need to read 2 parts of log */
701 * We are going to scan backwards in the log in two parts.
702 * First we scan the physical end of the log. In this part
703 * of the log, we are looking for blocks with cycle number
704 * last_half_cycle - 1.
705 * If we find one, then we know that the log starts there, as
706 * we've found a hole that didn't get written in going around
707 * the end of the physical log. The simple case for this is
708 * x + 1 ... | x ... | x - 1 | x
709 * <---------> less than scan distance
710 * If all of the blocks at the end of the log have cycle number
711 * last_half_cycle, then we check the blocks at the start of
712 * the log looking for occurrences of last_half_cycle. If we
713 * find one, then our current estimate for the location of the
714 * first occurrence of last_half_cycle is wrong and we move
715 * back to the hole we've found. This case looks like
716 * x + 1 ... | x | x + 1 | x ...
717 * ^ binary search stopped here
718 * Another case we need to handle that only occurs in 256k
720 * x + 1 ... | x ... | x+1 | x ...
721 * ^ binary search stops here
722 * In a 256k log, the scan at the end of the log will see the
723 * x + 1 blocks. We need to skip past those since that is
724 * certainly not the head of the log. By searching for
725 * last_half_cycle-1 we accomplish that.
727 ASSERT(head_blk
<= INT_MAX
&&
728 (xfs_daddr_t
) num_scan_bblks
>= head_blk
);
729 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
730 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
731 num_scan_bblks
- (int)head_blk
,
732 (stop_on_cycle
- 1), &new_blk
)))
740 * Scan beginning of log now. The last part of the physical
741 * log is good. This scan needs to verify that it doesn't find
742 * the last_half_cycle.
745 ASSERT(head_blk
<= INT_MAX
);
746 if ((error
= xlog_find_verify_cycle(log
,
747 start_blk
, (int)head_blk
,
748 stop_on_cycle
, &new_blk
)))
756 * Now we need to make sure head_blk is not pointing to a block in
757 * the middle of a log record.
759 num_scan_bblks
= XLOG_REC_SHIFT(log
);
760 if (head_blk
>= num_scan_bblks
) {
761 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
763 /* start ptr at last block ptr before head_blk */
764 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
765 &head_blk
, 0)) == -1) {
766 error
= XFS_ERROR(EIO
);
772 ASSERT(head_blk
<= INT_MAX
);
773 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
774 &head_blk
, 0)) == -1) {
775 /* We hit the beginning of the log during our search */
776 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
778 ASSERT(start_blk
<= INT_MAX
&&
779 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
780 ASSERT(head_blk
<= INT_MAX
);
781 if ((error
= xlog_find_verify_log_record(log
,
783 (int)head_blk
)) == -1) {
784 error
= XFS_ERROR(EIO
);
788 if (new_blk
!= log_bbnum
)
795 if (head_blk
== log_bbnum
)
796 *return_head_blk
= 0;
798 *return_head_blk
= head_blk
;
800 * When returning here, we have a good block number. Bad block
801 * means that during a previous crash, we didn't have a clean break
802 * from cycle number N to cycle number N-1. In this case, we need
803 * to find the first block with cycle number N-1.
811 xlog_warn("XFS: failed to find log head");
816 * Find the sync block number or the tail of the log.
818 * This will be the block number of the last record to have its
819 * associated buffers synced to disk. Every log record header has
820 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
821 * to get a sync block number. The only concern is to figure out which
822 * log record header to believe.
824 * The following algorithm uses the log record header with the largest
825 * lsn. The entire log record does not need to be valid. We only care
826 * that the header is valid.
828 * We could speed up search by using current head_blk buffer, but it is not
834 xfs_daddr_t
*head_blk
,
835 xfs_daddr_t
*tail_blk
)
837 xlog_rec_header_t
*rhead
;
838 xlog_op_header_t
*op_head
;
839 xfs_caddr_t offset
= NULL
;
842 xfs_daddr_t umount_data_blk
;
843 xfs_daddr_t after_umount_blk
;
850 * Find previous log record
852 if ((error
= xlog_find_head(log
, head_blk
)))
855 bp
= xlog_get_bp(log
, 1);
858 if (*head_blk
== 0) { /* special case */
859 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
863 if (xlog_get_cycle(offset
) == 0) {
865 /* leave all other log inited values alone */
871 * Search backwards looking for log record header block
873 ASSERT(*head_blk
< INT_MAX
);
874 for (i
= (int)(*head_blk
) - 1; i
>= 0; i
--) {
875 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
879 if (XLOG_HEADER_MAGIC_NUM
== be32_to_cpu(*(__be32
*)offset
)) {
885 * If we haven't found the log record header block, start looking
886 * again from the end of the physical log. XXXmiken: There should be
887 * a check here to make sure we didn't search more than N blocks in
891 for (i
= log
->l_logBBsize
- 1; i
>= (int)(*head_blk
); i
--) {
892 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
896 if (XLOG_HEADER_MAGIC_NUM
==
897 be32_to_cpu(*(__be32
*)offset
)) {
904 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
906 return XFS_ERROR(EIO
);
909 /* find blk_no of tail of log */
910 rhead
= (xlog_rec_header_t
*)offset
;
911 *tail_blk
= BLOCK_LSN(be64_to_cpu(rhead
->h_tail_lsn
));
914 * Reset log values according to the state of the log when we
915 * crashed. In the case where head_blk == 0, we bump curr_cycle
916 * one because the next write starts a new cycle rather than
917 * continuing the cycle of the last good log record. At this
918 * point we have guaranteed that all partial log records have been
919 * accounted for. Therefore, we know that the last good log record
920 * written was complete and ended exactly on the end boundary
921 * of the physical log.
923 log
->l_prev_block
= i
;
924 log
->l_curr_block
= (int)*head_blk
;
925 log
->l_curr_cycle
= be32_to_cpu(rhead
->h_cycle
);
928 log
->l_tail_lsn
= be64_to_cpu(rhead
->h_tail_lsn
);
929 log
->l_last_sync_lsn
= be64_to_cpu(rhead
->h_lsn
);
930 log
->l_grant_reserve_cycle
= log
->l_curr_cycle
;
931 log
->l_grant_reserve_bytes
= BBTOB(log
->l_curr_block
);
932 log
->l_grant_write_cycle
= log
->l_curr_cycle
;
933 log
->l_grant_write_bytes
= BBTOB(log
->l_curr_block
);
936 * Look for unmount record. If we find it, then we know there
937 * was a clean unmount. Since 'i' could be the last block in
938 * the physical log, we convert to a log block before comparing
941 * Save the current tail lsn to use to pass to
942 * xlog_clear_stale_blocks() below. We won't want to clear the
943 * unmount record if there is one, so we pass the lsn of the
944 * unmount record rather than the block after it.
946 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
947 int h_size
= be32_to_cpu(rhead
->h_size
);
948 int h_version
= be32_to_cpu(rhead
->h_version
);
950 if ((h_version
& XLOG_VERSION_2
) &&
951 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
952 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
953 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
961 after_umount_blk
= (i
+ hblks
+ (int)
962 BTOBB(be32_to_cpu(rhead
->h_len
))) % log
->l_logBBsize
;
963 tail_lsn
= log
->l_tail_lsn
;
964 if (*head_blk
== after_umount_blk
&&
965 be32_to_cpu(rhead
->h_num_logops
) == 1) {
966 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
967 error
= xlog_bread(log
, umount_data_blk
, 1, bp
, &offset
);
971 op_head
= (xlog_op_header_t
*)offset
;
972 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
974 * Set tail and last sync so that newly written
975 * log records will point recovery to after the
976 * current unmount record.
979 xlog_assign_lsn(log
->l_curr_cycle
,
981 log
->l_last_sync_lsn
=
982 xlog_assign_lsn(log
->l_curr_cycle
,
984 *tail_blk
= after_umount_blk
;
987 * Note that the unmount was clean. If the unmount
988 * was not clean, we need to know this to rebuild the
989 * superblock counters from the perag headers if we
990 * have a filesystem using non-persistent counters.
992 log
->l_mp
->m_flags
|= XFS_MOUNT_WAS_CLEAN
;
997 * Make sure that there are no blocks in front of the head
998 * with the same cycle number as the head. This can happen
999 * because we allow multiple outstanding log writes concurrently,
1000 * and the later writes might make it out before earlier ones.
1002 * We use the lsn from before modifying it so that we'll never
1003 * overwrite the unmount record after a clean unmount.
1005 * Do this only if we are going to recover the filesystem
1007 * NOTE: This used to say "if (!readonly)"
1008 * However on Linux, we can & do recover a read-only filesystem.
1009 * We only skip recovery if NORECOVERY is specified on mount,
1010 * in which case we would not be here.
1012 * But... if the -device- itself is readonly, just skip this.
1013 * We can't recover this device anyway, so it won't matter.
1015 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
))
1016 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
1022 xlog_warn("XFS: failed to locate log tail");
1027 * Is the log zeroed at all?
1029 * The last binary search should be changed to perform an X block read
1030 * once X becomes small enough. You can then search linearly through
1031 * the X blocks. This will cut down on the number of reads we need to do.
1033 * If the log is partially zeroed, this routine will pass back the blkno
1034 * of the first block with cycle number 0. It won't have a complete LR
1038 * 0 => the log is completely written to
1039 * -1 => use *blk_no as the first block of the log
1040 * >0 => error has occurred
1045 xfs_daddr_t
*blk_no
)
1049 uint first_cycle
, last_cycle
;
1050 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1051 xfs_daddr_t num_scan_bblks
;
1052 int error
, log_bbnum
= log
->l_logBBsize
;
1056 /* check totally zeroed log */
1057 bp
= xlog_get_bp(log
, 1);
1060 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1064 first_cycle
= xlog_get_cycle(offset
);
1065 if (first_cycle
== 0) { /* completely zeroed log */
1071 /* check partially zeroed log */
1072 error
= xlog_bread(log
, log_bbnum
-1, 1, bp
, &offset
);
1076 last_cycle
= xlog_get_cycle(offset
);
1077 if (last_cycle
!= 0) { /* log completely written to */
1080 } else if (first_cycle
!= 1) {
1082 * If the cycle of the last block is zero, the cycle of
1083 * the first block must be 1. If it's not, maybe we're
1084 * not looking at a log... Bail out.
1086 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1087 return XFS_ERROR(EINVAL
);
1090 /* we have a partially zeroed log */
1091 last_blk
= log_bbnum
-1;
1092 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1096 * Validate the answer. Because there is no way to guarantee that
1097 * the entire log is made up of log records which are the same size,
1098 * we scan over the defined maximum blocks. At this point, the maximum
1099 * is not chosen to mean anything special. XXXmiken
1101 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1102 ASSERT(num_scan_bblks
<= INT_MAX
);
1104 if (last_blk
< num_scan_bblks
)
1105 num_scan_bblks
= last_blk
;
1106 start_blk
= last_blk
- num_scan_bblks
;
1109 * We search for any instances of cycle number 0 that occur before
1110 * our current estimate of the head. What we're trying to detect is
1111 * 1 ... | 0 | 1 | 0...
1112 * ^ binary search ends here
1114 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1115 (int)num_scan_bblks
, 0, &new_blk
)))
1121 * Potentially backup over partial log record write. We don't need
1122 * to search the end of the log because we know it is zero.
1124 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
1125 &last_blk
, 0)) == -1) {
1126 error
= XFS_ERROR(EIO
);
1140 * These are simple subroutines used by xlog_clear_stale_blocks() below
1141 * to initialize a buffer full of empty log record headers and write
1142 * them into the log.
1153 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1155 memset(buf
, 0, BBSIZE
);
1156 recp
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1157 recp
->h_cycle
= cpu_to_be32(cycle
);
1158 recp
->h_version
= cpu_to_be32(
1159 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1160 recp
->h_lsn
= cpu_to_be64(xlog_assign_lsn(cycle
, block
));
1161 recp
->h_tail_lsn
= cpu_to_be64(xlog_assign_lsn(tail_cycle
, tail_block
));
1162 recp
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1163 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1167 xlog_write_log_records(
1178 int sectbb
= log
->l_sectBBsize
;
1179 int end_block
= start_block
+ blocks
;
1185 * Greedily allocate a buffer big enough to handle the full
1186 * range of basic blocks to be written. If that fails, try
1187 * a smaller size. We need to be able to write at least a
1188 * log sector, or we're out of luck.
1190 bufblks
= 1 << ffs(blocks
);
1191 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1193 if (bufblks
< sectbb
)
1197 /* We may need to do a read at the start to fill in part of
1198 * the buffer in the starting sector not covered by the first
1201 balign
= round_down(start_block
, sectbb
);
1202 if (balign
!= start_block
) {
1203 error
= xlog_bread_noalign(log
, start_block
, 1, bp
);
1207 j
= start_block
- balign
;
1210 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1211 int bcount
, endcount
;
1213 bcount
= min(bufblks
, end_block
- start_block
);
1214 endcount
= bcount
- j
;
1216 /* We may need to do a read at the end to fill in part of
1217 * the buffer in the final sector not covered by the write.
1218 * If this is the same sector as the above read, skip it.
1220 ealign
= round_down(end_block
, sectbb
);
1221 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1222 offset
= XFS_BUF_PTR(bp
);
1223 balign
= BBTOB(ealign
- start_block
);
1224 error
= XFS_BUF_SET_PTR(bp
, offset
+ balign
,
1229 error
= xlog_bread_noalign(log
, ealign
, sectbb
, bp
);
1233 error
= XFS_BUF_SET_PTR(bp
, offset
, bufblks
);
1238 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1239 for (; j
< endcount
; j
++) {
1240 xlog_add_record(log
, offset
, cycle
, i
+j
,
1241 tail_cycle
, tail_block
);
1244 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1247 start_block
+= endcount
;
1257 * This routine is called to blow away any incomplete log writes out
1258 * in front of the log head. We do this so that we won't become confused
1259 * if we come up, write only a little bit more, and then crash again.
1260 * If we leave the partial log records out there, this situation could
1261 * cause us to think those partial writes are valid blocks since they
1262 * have the current cycle number. We get rid of them by overwriting them
1263 * with empty log records with the old cycle number rather than the
1266 * The tail lsn is passed in rather than taken from
1267 * the log so that we will not write over the unmount record after a
1268 * clean unmount in a 512 block log. Doing so would leave the log without
1269 * any valid log records in it until a new one was written. If we crashed
1270 * during that time we would not be able to recover.
1273 xlog_clear_stale_blocks(
1277 int tail_cycle
, head_cycle
;
1278 int tail_block
, head_block
;
1279 int tail_distance
, max_distance
;
1283 tail_cycle
= CYCLE_LSN(tail_lsn
);
1284 tail_block
= BLOCK_LSN(tail_lsn
);
1285 head_cycle
= log
->l_curr_cycle
;
1286 head_block
= log
->l_curr_block
;
1289 * Figure out the distance between the new head of the log
1290 * and the tail. We want to write over any blocks beyond the
1291 * head that we may have written just before the crash, but
1292 * we don't want to overwrite the tail of the log.
1294 if (head_cycle
== tail_cycle
) {
1296 * The tail is behind the head in the physical log,
1297 * so the distance from the head to the tail is the
1298 * distance from the head to the end of the log plus
1299 * the distance from the beginning of the log to the
1302 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1303 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1304 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1305 return XFS_ERROR(EFSCORRUPTED
);
1307 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1310 * The head is behind the tail in the physical log,
1311 * so the distance from the head to the tail is just
1312 * the tail block minus the head block.
1314 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1315 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1316 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1317 return XFS_ERROR(EFSCORRUPTED
);
1319 tail_distance
= tail_block
- head_block
;
1323 * If the head is right up against the tail, we can't clear
1326 if (tail_distance
<= 0) {
1327 ASSERT(tail_distance
== 0);
1331 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1333 * Take the smaller of the maximum amount of outstanding I/O
1334 * we could have and the distance to the tail to clear out.
1335 * We take the smaller so that we don't overwrite the tail and
1336 * we don't waste all day writing from the head to the tail
1339 max_distance
= MIN(max_distance
, tail_distance
);
1341 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1343 * We can stomp all the blocks we need to without
1344 * wrapping around the end of the log. Just do it
1345 * in a single write. Use the cycle number of the
1346 * current cycle minus one so that the log will look like:
1349 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1350 head_block
, max_distance
, tail_cycle
,
1356 * We need to wrap around the end of the physical log in
1357 * order to clear all the blocks. Do it in two separate
1358 * I/Os. The first write should be from the head to the
1359 * end of the physical log, and it should use the current
1360 * cycle number minus one just like above.
1362 distance
= log
->l_logBBsize
- head_block
;
1363 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1364 head_block
, distance
, tail_cycle
,
1371 * Now write the blocks at the start of the physical log.
1372 * This writes the remainder of the blocks we want to clear.
1373 * It uses the current cycle number since we're now on the
1374 * same cycle as the head so that we get:
1375 * n ... n ... | n - 1 ...
1376 * ^^^^^ blocks we're writing
1378 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1379 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1380 tail_cycle
, tail_block
);
1388 /******************************************************************************
1390 * Log recover routines
1392 ******************************************************************************
1395 STATIC xlog_recover_t
*
1396 xlog_recover_find_tid(
1397 struct hlist_head
*head
,
1400 xlog_recover_t
*trans
;
1401 struct hlist_node
*n
;
1403 hlist_for_each_entry(trans
, n
, head
, r_list
) {
1404 if (trans
->r_log_tid
== tid
)
1411 xlog_recover_new_tid(
1412 struct hlist_head
*head
,
1416 xlog_recover_t
*trans
;
1418 trans
= kmem_zalloc(sizeof(xlog_recover_t
), KM_SLEEP
);
1419 trans
->r_log_tid
= tid
;
1421 INIT_LIST_HEAD(&trans
->r_itemq
);
1423 INIT_HLIST_NODE(&trans
->r_list
);
1424 hlist_add_head(&trans
->r_list
, head
);
1428 xlog_recover_add_item(
1429 struct list_head
*head
)
1431 xlog_recover_item_t
*item
;
1433 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
1434 INIT_LIST_HEAD(&item
->ri_list
);
1435 list_add_tail(&item
->ri_list
, head
);
1439 xlog_recover_add_to_cont_trans(
1441 xlog_recover_t
*trans
,
1445 xlog_recover_item_t
*item
;
1446 xfs_caddr_t ptr
, old_ptr
;
1449 if (list_empty(&trans
->r_itemq
)) {
1450 /* finish copying rest of trans header */
1451 xlog_recover_add_item(&trans
->r_itemq
);
1452 ptr
= (xfs_caddr_t
) &trans
->r_theader
+
1453 sizeof(xfs_trans_header_t
) - len
;
1454 memcpy(ptr
, dp
, len
); /* d, s, l */
1457 /* take the tail entry */
1458 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1460 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
1461 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
1463 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, 0u);
1464 memcpy(&ptr
[old_len
], dp
, len
); /* d, s, l */
1465 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
1466 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
1467 trace_xfs_log_recover_item_add_cont(log
, trans
, item
, 0);
1472 * The next region to add is the start of a new region. It could be
1473 * a whole region or it could be the first part of a new region. Because
1474 * of this, the assumption here is that the type and size fields of all
1475 * format structures fit into the first 32 bits of the structure.
1477 * This works because all regions must be 32 bit aligned. Therefore, we
1478 * either have both fields or we have neither field. In the case we have
1479 * neither field, the data part of the region is zero length. We only have
1480 * a log_op_header and can throw away the header since a new one will appear
1481 * later. If we have at least 4 bytes, then we can determine how many regions
1482 * will appear in the current log item.
1485 xlog_recover_add_to_trans(
1487 xlog_recover_t
*trans
,
1491 xfs_inode_log_format_t
*in_f
; /* any will do */
1492 xlog_recover_item_t
*item
;
1497 if (list_empty(&trans
->r_itemq
)) {
1498 /* we need to catch log corruptions here */
1499 if (*(uint
*)dp
!= XFS_TRANS_HEADER_MAGIC
) {
1500 xlog_warn("XFS: xlog_recover_add_to_trans: "
1501 "bad header magic number");
1503 return XFS_ERROR(EIO
);
1505 if (len
== sizeof(xfs_trans_header_t
))
1506 xlog_recover_add_item(&trans
->r_itemq
);
1507 memcpy(&trans
->r_theader
, dp
, len
); /* d, s, l */
1511 ptr
= kmem_alloc(len
, KM_SLEEP
);
1512 memcpy(ptr
, dp
, len
);
1513 in_f
= (xfs_inode_log_format_t
*)ptr
;
1515 /* take the tail entry */
1516 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1517 if (item
->ri_total
!= 0 &&
1518 item
->ri_total
== item
->ri_cnt
) {
1519 /* tail item is in use, get a new one */
1520 xlog_recover_add_item(&trans
->r_itemq
);
1521 item
= list_entry(trans
->r_itemq
.prev
,
1522 xlog_recover_item_t
, ri_list
);
1525 if (item
->ri_total
== 0) { /* first region to be added */
1526 if (in_f
->ilf_size
== 0 ||
1527 in_f
->ilf_size
> XLOG_MAX_REGIONS_IN_ITEM
) {
1529 "XFS: bad number of regions (%d) in inode log format",
1532 return XFS_ERROR(EIO
);
1535 item
->ri_total
= in_f
->ilf_size
;
1537 kmem_zalloc(item
->ri_total
* sizeof(xfs_log_iovec_t
),
1540 ASSERT(item
->ri_total
> item
->ri_cnt
);
1541 /* Description region is ri_buf[0] */
1542 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
1543 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
1545 trace_xfs_log_recover_item_add(log
, trans
, item
, 0);
1550 * Sort the log items in the transaction. Cancelled buffers need
1551 * to be put first so they are processed before any items that might
1552 * modify the buffers. If they are cancelled, then the modifications
1553 * don't need to be replayed.
1556 xlog_recover_reorder_trans(
1558 xlog_recover_t
*trans
,
1561 xlog_recover_item_t
*item
, *n
;
1562 LIST_HEAD(sort_list
);
1564 list_splice_init(&trans
->r_itemq
, &sort_list
);
1565 list_for_each_entry_safe(item
, n
, &sort_list
, ri_list
) {
1566 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1568 switch (ITEM_TYPE(item
)) {
1570 if (!(buf_f
->blf_flags
& XFS_BLF_CANCEL
)) {
1571 trace_xfs_log_recover_item_reorder_head(log
,
1573 list_move(&item
->ri_list
, &trans
->r_itemq
);
1578 case XFS_LI_QUOTAOFF
:
1581 trace_xfs_log_recover_item_reorder_tail(log
,
1583 list_move_tail(&item
->ri_list
, &trans
->r_itemq
);
1587 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1589 return XFS_ERROR(EIO
);
1592 ASSERT(list_empty(&sort_list
));
1597 * Build up the table of buf cancel records so that we don't replay
1598 * cancelled data in the second pass. For buffer records that are
1599 * not cancel records, there is nothing to do here so we just return.
1601 * If we get a cancel record which is already in the table, this indicates
1602 * that the buffer was cancelled multiple times. In order to ensure
1603 * that during pass 2 we keep the record in the table until we reach its
1604 * last occurrence in the log, we keep a reference count in the cancel
1605 * record in the table to tell us how many times we expect to see this
1606 * record during the second pass.
1609 xlog_recover_do_buffer_pass1(
1611 xfs_buf_log_format_t
*buf_f
)
1613 xfs_buf_cancel_t
*bcp
;
1614 xfs_buf_cancel_t
*nextp
;
1615 xfs_buf_cancel_t
*prevp
;
1616 xfs_buf_cancel_t
**bucket
;
1617 xfs_daddr_t blkno
= 0;
1621 switch (buf_f
->blf_type
) {
1623 blkno
= buf_f
->blf_blkno
;
1624 len
= buf_f
->blf_len
;
1625 flags
= buf_f
->blf_flags
;
1630 * If this isn't a cancel buffer item, then just return.
1632 if (!(flags
& XFS_BLF_CANCEL
)) {
1633 trace_xfs_log_recover_buf_not_cancel(log
, buf_f
);
1638 * Insert an xfs_buf_cancel record into the hash table of
1639 * them. If there is already an identical record, bump
1640 * its reference count.
1642 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1643 XLOG_BC_TABLE_SIZE
];
1645 * If the hash bucket is empty then just insert a new record into
1648 if (*bucket
== NULL
) {
1649 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1651 bcp
->bc_blkno
= blkno
;
1653 bcp
->bc_refcount
= 1;
1654 bcp
->bc_next
= NULL
;
1660 * The hash bucket is not empty, so search for duplicates of our
1661 * record. If we find one them just bump its refcount. If not
1662 * then add us at the end of the list.
1666 while (nextp
!= NULL
) {
1667 if (nextp
->bc_blkno
== blkno
&& nextp
->bc_len
== len
) {
1668 nextp
->bc_refcount
++;
1669 trace_xfs_log_recover_buf_cancel_ref_inc(log
, buf_f
);
1673 nextp
= nextp
->bc_next
;
1675 ASSERT(prevp
!= NULL
);
1676 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1678 bcp
->bc_blkno
= blkno
;
1680 bcp
->bc_refcount
= 1;
1681 bcp
->bc_next
= NULL
;
1682 prevp
->bc_next
= bcp
;
1683 trace_xfs_log_recover_buf_cancel_add(log
, buf_f
);
1687 * Check to see whether the buffer being recovered has a corresponding
1688 * entry in the buffer cancel record table. If it does then return 1
1689 * so that it will be cancelled, otherwise return 0. If the buffer is
1690 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
1691 * the refcount on the entry in the table and remove it from the table
1692 * if this is the last reference.
1694 * We remove the cancel record from the table when we encounter its
1695 * last occurrence in the log so that if the same buffer is re-used
1696 * again after its last cancellation we actually replay the changes
1697 * made at that point.
1700 xlog_check_buffer_cancelled(
1706 xfs_buf_cancel_t
*bcp
;
1707 xfs_buf_cancel_t
*prevp
;
1708 xfs_buf_cancel_t
**bucket
;
1710 if (log
->l_buf_cancel_table
== NULL
) {
1712 * There is nothing in the table built in pass one,
1713 * so this buffer must not be cancelled.
1715 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1719 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1720 XLOG_BC_TABLE_SIZE
];
1724 * There is no corresponding entry in the table built
1725 * in pass one, so this buffer has not been cancelled.
1727 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1732 * Search for an entry in the buffer cancel table that
1733 * matches our buffer.
1736 while (bcp
!= NULL
) {
1737 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
) {
1739 * We've go a match, so return 1 so that the
1740 * recovery of this buffer is cancelled.
1741 * If this buffer is actually a buffer cancel
1742 * log item, then decrement the refcount on the
1743 * one in the table and remove it if this is the
1746 if (flags
& XFS_BLF_CANCEL
) {
1748 if (bcp
->bc_refcount
== 0) {
1749 if (prevp
== NULL
) {
1750 *bucket
= bcp
->bc_next
;
1752 prevp
->bc_next
= bcp
->bc_next
;
1763 * We didn't find a corresponding entry in the table, so
1764 * return 0 so that the buffer is NOT cancelled.
1766 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1771 xlog_recover_do_buffer_pass2(
1773 xfs_buf_log_format_t
*buf_f
)
1775 xfs_daddr_t blkno
= 0;
1779 switch (buf_f
->blf_type
) {
1781 blkno
= buf_f
->blf_blkno
;
1782 flags
= buf_f
->blf_flags
;
1783 len
= buf_f
->blf_len
;
1787 return xlog_check_buffer_cancelled(log
, blkno
, len
, flags
);
1791 * Perform recovery for a buffer full of inodes. In these buffers,
1792 * the only data which should be recovered is that which corresponds
1793 * to the di_next_unlinked pointers in the on disk inode structures.
1794 * The rest of the data for the inodes is always logged through the
1795 * inodes themselves rather than the inode buffer and is recovered
1796 * in xlog_recover_do_inode_trans().
1798 * The only time when buffers full of inodes are fully recovered is
1799 * when the buffer is full of newly allocated inodes. In this case
1800 * the buffer will not be marked as an inode buffer and so will be
1801 * sent to xlog_recover_do_reg_buffer() below during recovery.
1804 xlog_recover_do_inode_buffer(
1806 xlog_recover_item_t
*item
,
1808 xfs_buf_log_format_t
*buf_f
)
1816 int next_unlinked_offset
;
1818 xfs_agino_t
*logged_nextp
;
1819 xfs_agino_t
*buffer_nextp
;
1820 unsigned int *data_map
= NULL
;
1821 unsigned int map_size
= 0;
1823 trace_xfs_log_recover_buf_inode_buf(mp
->m_log
, buf_f
);
1825 switch (buf_f
->blf_type
) {
1827 data_map
= buf_f
->blf_data_map
;
1828 map_size
= buf_f
->blf_map_size
;
1832 * Set the variables corresponding to the current region to
1833 * 0 so that we'll initialize them on the first pass through
1841 inodes_per_buf
= XFS_BUF_COUNT(bp
) >> mp
->m_sb
.sb_inodelog
;
1842 for (i
= 0; i
< inodes_per_buf
; i
++) {
1843 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1844 offsetof(xfs_dinode_t
, di_next_unlinked
);
1846 while (next_unlinked_offset
>=
1847 (reg_buf_offset
+ reg_buf_bytes
)) {
1849 * The next di_next_unlinked field is beyond
1850 * the current logged region. Find the next
1851 * logged region that contains or is beyond
1852 * the current di_next_unlinked field.
1855 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1858 * If there are no more logged regions in the
1859 * buffer, then we're done.
1865 nbits
= xfs_contig_bits(data_map
, map_size
,
1868 reg_buf_offset
= bit
<< XFS_BLF_SHIFT
;
1869 reg_buf_bytes
= nbits
<< XFS_BLF_SHIFT
;
1874 * If the current logged region starts after the current
1875 * di_next_unlinked field, then move on to the next
1876 * di_next_unlinked field.
1878 if (next_unlinked_offset
< reg_buf_offset
) {
1882 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1883 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLF_CHUNK
) == 0);
1884 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <= XFS_BUF_COUNT(bp
));
1887 * The current logged region contains a copy of the
1888 * current di_next_unlinked field. Extract its value
1889 * and copy it to the buffer copy.
1891 logged_nextp
= item
->ri_buf
[item_index
].i_addr
+
1892 next_unlinked_offset
- reg_buf_offset
;
1893 if (unlikely(*logged_nextp
== 0)) {
1894 xfs_fs_cmn_err(CE_ALERT
, mp
,
1895 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1897 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1898 XFS_ERRLEVEL_LOW
, mp
);
1899 return XFS_ERROR(EFSCORRUPTED
);
1902 buffer_nextp
= (xfs_agino_t
*)xfs_buf_offset(bp
,
1903 next_unlinked_offset
);
1904 *buffer_nextp
= *logged_nextp
;
1911 * Perform a 'normal' buffer recovery. Each logged region of the
1912 * buffer should be copied over the corresponding region in the
1913 * given buffer. The bitmap in the buf log format structure indicates
1914 * where to place the logged data.
1918 xlog_recover_do_reg_buffer(
1919 struct xfs_mount
*mp
,
1920 xlog_recover_item_t
*item
,
1922 xfs_buf_log_format_t
*buf_f
)
1927 unsigned int *data_map
= NULL
;
1928 unsigned int map_size
= 0;
1931 trace_xfs_log_recover_buf_reg_buf(mp
->m_log
, buf_f
);
1933 switch (buf_f
->blf_type
) {
1935 data_map
= buf_f
->blf_data_map
;
1936 map_size
= buf_f
->blf_map_size
;
1940 i
= 1; /* 0 is the buf format structure */
1942 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1945 nbits
= xfs_contig_bits(data_map
, map_size
, bit
);
1947 ASSERT(item
->ri_buf
[i
].i_addr
!= NULL
);
1948 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLF_CHUNK
== 0);
1949 ASSERT(XFS_BUF_COUNT(bp
) >=
1950 ((uint
)bit
<< XFS_BLF_SHIFT
)+(nbits
<<XFS_BLF_SHIFT
));
1953 * Do a sanity check if this is a dquot buffer. Just checking
1954 * the first dquot in the buffer should do. XXXThis is
1955 * probably a good thing to do for other buf types also.
1958 if (buf_f
->blf_flags
&
1959 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
1960 if (item
->ri_buf
[i
].i_addr
== NULL
) {
1962 "XFS: NULL dquot in %s.", __func__
);
1965 if (item
->ri_buf
[i
].i_len
< sizeof(xfs_disk_dquot_t
)) {
1967 "XFS: dquot too small (%d) in %s.",
1968 item
->ri_buf
[i
].i_len
, __func__
);
1971 error
= xfs_qm_dqcheck(item
->ri_buf
[i
].i_addr
,
1972 -1, 0, XFS_QMOPT_DOWARN
,
1973 "dquot_buf_recover");
1978 memcpy(xfs_buf_offset(bp
,
1979 (uint
)bit
<< XFS_BLF_SHIFT
), /* dest */
1980 item
->ri_buf
[i
].i_addr
, /* source */
1981 nbits
<<XFS_BLF_SHIFT
); /* length */
1987 /* Shouldn't be any more regions */
1988 ASSERT(i
== item
->ri_total
);
1992 * Do some primitive error checking on ondisk dquot data structures.
1996 xfs_disk_dquot_t
*ddq
,
1998 uint type
, /* used only when IO_dorepair is true */
2002 xfs_dqblk_t
*d
= (xfs_dqblk_t
*)ddq
;
2006 * We can encounter an uninitialized dquot buffer for 2 reasons:
2007 * 1. If we crash while deleting the quotainode(s), and those blks got
2008 * used for user data. This is because we take the path of regular
2009 * file deletion; however, the size field of quotainodes is never
2010 * updated, so all the tricks that we play in itruncate_finish
2011 * don't quite matter.
2013 * 2. We don't play the quota buffers when there's a quotaoff logitem.
2014 * But the allocation will be replayed so we'll end up with an
2015 * uninitialized quota block.
2017 * This is all fine; things are still consistent, and we haven't lost
2018 * any quota information. Just don't complain about bad dquot blks.
2020 if (be16_to_cpu(ddq
->d_magic
) != XFS_DQUOT_MAGIC
) {
2021 if (flags
& XFS_QMOPT_DOWARN
)
2023 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
2024 str
, id
, be16_to_cpu(ddq
->d_magic
), XFS_DQUOT_MAGIC
);
2027 if (ddq
->d_version
!= XFS_DQUOT_VERSION
) {
2028 if (flags
& XFS_QMOPT_DOWARN
)
2030 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2031 str
, id
, ddq
->d_version
, XFS_DQUOT_VERSION
);
2035 if (ddq
->d_flags
!= XFS_DQ_USER
&&
2036 ddq
->d_flags
!= XFS_DQ_PROJ
&&
2037 ddq
->d_flags
!= XFS_DQ_GROUP
) {
2038 if (flags
& XFS_QMOPT_DOWARN
)
2040 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2041 str
, id
, ddq
->d_flags
);
2045 if (id
!= -1 && id
!= be32_to_cpu(ddq
->d_id
)) {
2046 if (flags
& XFS_QMOPT_DOWARN
)
2048 "%s : ondisk-dquot 0x%p, ID mismatch: "
2049 "0x%x expected, found id 0x%x",
2050 str
, ddq
, id
, be32_to_cpu(ddq
->d_id
));
2054 if (!errs
&& ddq
->d_id
) {
2055 if (ddq
->d_blk_softlimit
&&
2056 be64_to_cpu(ddq
->d_bcount
) >=
2057 be64_to_cpu(ddq
->d_blk_softlimit
)) {
2058 if (!ddq
->d_btimer
) {
2059 if (flags
& XFS_QMOPT_DOWARN
)
2061 "%s : Dquot ID 0x%x (0x%p) "
2062 "BLK TIMER NOT STARTED",
2063 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2067 if (ddq
->d_ino_softlimit
&&
2068 be64_to_cpu(ddq
->d_icount
) >=
2069 be64_to_cpu(ddq
->d_ino_softlimit
)) {
2070 if (!ddq
->d_itimer
) {
2071 if (flags
& XFS_QMOPT_DOWARN
)
2073 "%s : Dquot ID 0x%x (0x%p) "
2074 "INODE TIMER NOT STARTED",
2075 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2079 if (ddq
->d_rtb_softlimit
&&
2080 be64_to_cpu(ddq
->d_rtbcount
) >=
2081 be64_to_cpu(ddq
->d_rtb_softlimit
)) {
2082 if (!ddq
->d_rtbtimer
) {
2083 if (flags
& XFS_QMOPT_DOWARN
)
2085 "%s : Dquot ID 0x%x (0x%p) "
2086 "RTBLK TIMER NOT STARTED",
2087 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2093 if (!errs
|| !(flags
& XFS_QMOPT_DQREPAIR
))
2096 if (flags
& XFS_QMOPT_DOWARN
)
2097 cmn_err(CE_NOTE
, "Re-initializing dquot ID 0x%x", id
);
2100 * Typically, a repair is only requested by quotacheck.
2103 ASSERT(flags
& XFS_QMOPT_DQREPAIR
);
2104 memset(d
, 0, sizeof(xfs_dqblk_t
));
2106 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
2107 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
2108 d
->dd_diskdq
.d_flags
= type
;
2109 d
->dd_diskdq
.d_id
= cpu_to_be32(id
);
2115 * Perform a dquot buffer recovery.
2116 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2117 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2118 * Else, treat it as a regular buffer and do recovery.
2121 xlog_recover_do_dquot_buffer(
2124 xlog_recover_item_t
*item
,
2126 xfs_buf_log_format_t
*buf_f
)
2130 trace_xfs_log_recover_buf_dquot_buf(log
, buf_f
);
2133 * Filesystems are required to send in quota flags at mount time.
2135 if (mp
->m_qflags
== 0) {
2140 if (buf_f
->blf_flags
& XFS_BLF_UDQUOT_BUF
)
2141 type
|= XFS_DQ_USER
;
2142 if (buf_f
->blf_flags
& XFS_BLF_PDQUOT_BUF
)
2143 type
|= XFS_DQ_PROJ
;
2144 if (buf_f
->blf_flags
& XFS_BLF_GDQUOT_BUF
)
2145 type
|= XFS_DQ_GROUP
;
2147 * This type of quotas was turned off, so ignore this buffer
2149 if (log
->l_quotaoffs_flag
& type
)
2152 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2156 * This routine replays a modification made to a buffer at runtime.
2157 * There are actually two types of buffer, regular and inode, which
2158 * are handled differently. Inode buffers are handled differently
2159 * in that we only recover a specific set of data from them, namely
2160 * the inode di_next_unlinked fields. This is because all other inode
2161 * data is actually logged via inode records and any data we replay
2162 * here which overlaps that may be stale.
2164 * When meta-data buffers are freed at run time we log a buffer item
2165 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2166 * of the buffer in the log should not be replayed at recovery time.
2167 * This is so that if the blocks covered by the buffer are reused for
2168 * file data before we crash we don't end up replaying old, freed
2169 * meta-data into a user's file.
2171 * To handle the cancellation of buffer log items, we make two passes
2172 * over the log during recovery. During the first we build a table of
2173 * those buffers which have been cancelled, and during the second we
2174 * only replay those buffers which do not have corresponding cancel
2175 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2176 * for more details on the implementation of the table of cancel records.
2179 xlog_recover_do_buffer_trans(
2181 xlog_recover_item_t
*item
,
2184 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
2194 if (pass
== XLOG_RECOVER_PASS1
) {
2196 * In this pass we're only looking for buf items
2197 * with the XFS_BLF_CANCEL bit set.
2199 xlog_recover_do_buffer_pass1(log
, buf_f
);
2203 * In this pass we want to recover all the buffers
2204 * which have not been cancelled and are not
2205 * cancellation buffers themselves. The routine
2206 * we call here will tell us whether or not to
2207 * continue with the replay of this buffer.
2209 cancel
= xlog_recover_do_buffer_pass2(log
, buf_f
);
2211 trace_xfs_log_recover_buf_cancel(log
, buf_f
);
2215 trace_xfs_log_recover_buf_recover(log
, buf_f
);
2216 switch (buf_f
->blf_type
) {
2218 blkno
= buf_f
->blf_blkno
;
2219 len
= buf_f
->blf_len
;
2220 flags
= buf_f
->blf_flags
;
2223 xfs_fs_cmn_err(CE_ALERT
, log
->l_mp
,
2224 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2225 buf_f
->blf_type
, log
->l_mp
->m_logname
?
2226 log
->l_mp
->m_logname
: "internal");
2227 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2228 XFS_ERRLEVEL_LOW
, log
->l_mp
);
2229 return XFS_ERROR(EFSCORRUPTED
);
2233 buf_flags
= XBF_LOCK
;
2234 if (!(flags
& XFS_BLF_INODE_BUF
))
2235 buf_flags
|= XBF_MAPPED
;
2237 bp
= xfs_buf_read(mp
->m_ddev_targp
, blkno
, len
, buf_flags
);
2238 if (XFS_BUF_ISERROR(bp
)) {
2239 xfs_ioerror_alert("xlog_recover_do..(read#1)", log
->l_mp
,
2241 error
= XFS_BUF_GETERROR(bp
);
2247 if (flags
& XFS_BLF_INODE_BUF
) {
2248 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2250 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2251 xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2253 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2256 return XFS_ERROR(error
);
2259 * Perform delayed write on the buffer. Asynchronous writes will be
2260 * slower when taking into account all the buffers to be flushed.
2262 * Also make sure that only inode buffers with good sizes stay in
2263 * the buffer cache. The kernel moves inodes in buffers of 1 block
2264 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2265 * buffers in the log can be a different size if the log was generated
2266 * by an older kernel using unclustered inode buffers or a newer kernel
2267 * running with a different inode cluster size. Regardless, if the
2268 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2269 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2270 * the buffer out of the buffer cache so that the buffer won't
2271 * overlap with future reads of those inodes.
2273 if (XFS_DINODE_MAGIC
==
2274 be16_to_cpu(*((__be16
*)xfs_buf_offset(bp
, 0))) &&
2275 (XFS_BUF_COUNT(bp
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2276 (__uint32_t
)XFS_INODE_CLUSTER_SIZE(log
->l_mp
)))) {
2278 error
= xfs_bwrite(mp
, bp
);
2280 ASSERT(bp
->b_target
->bt_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_target
->bt_mount
== mp
);
2545 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2546 xfs_bdwrite(mp
, bp
);
2550 return XFS_ERROR(error
);
2554 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2555 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2559 xlog_recover_do_quotaoff_trans(
2561 xlog_recover_item_t
*item
,
2564 xfs_qoff_logformat_t
*qoff_f
;
2566 if (pass
== XLOG_RECOVER_PASS2
) {
2570 qoff_f
= item
->ri_buf
[0].i_addr
;
2574 * The logitem format's flag tells us if this was user quotaoff,
2575 * group/project quotaoff or both.
2577 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2578 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2579 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
2580 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
2581 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2582 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2588 * Recover a dquot record
2591 xlog_recover_do_dquot_trans(
2593 xlog_recover_item_t
*item
,
2598 struct xfs_disk_dquot
*ddq
, *recddq
;
2600 xfs_dq_logformat_t
*dq_f
;
2603 if (pass
== XLOG_RECOVER_PASS1
) {
2609 * Filesystems are required to send in quota flags at mount time.
2611 if (mp
->m_qflags
== 0)
2614 recddq
= item
->ri_buf
[1].i_addr
;
2615 if (recddq
== NULL
) {
2617 "XFS: NULL dquot in %s.", __func__
);
2618 return XFS_ERROR(EIO
);
2620 if (item
->ri_buf
[1].i_len
< sizeof(xfs_disk_dquot_t
)) {
2622 "XFS: dquot too small (%d) in %s.",
2623 item
->ri_buf
[1].i_len
, __func__
);
2624 return XFS_ERROR(EIO
);
2628 * This type of quotas was turned off, so ignore this record.
2630 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
2632 if (log
->l_quotaoffs_flag
& type
)
2636 * At this point we know that quota was _not_ turned off.
2637 * Since the mount flags are not indicating to us otherwise, this
2638 * must mean that quota is on, and the dquot needs to be replayed.
2639 * Remember that we may not have fully recovered the superblock yet,
2640 * so we can't do the usual trick of looking at the SB quota bits.
2642 * The other possibility, of course, is that the quota subsystem was
2643 * removed since the last mount - ENOSYS.
2645 dq_f
= item
->ri_buf
[0].i_addr
;
2647 if ((error
= xfs_qm_dqcheck(recddq
,
2649 0, XFS_QMOPT_DOWARN
,
2650 "xlog_recover_do_dquot_trans (log copy)"))) {
2651 return XFS_ERROR(EIO
);
2653 ASSERT(dq_f
->qlf_len
== 1);
2655 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
2657 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
),
2660 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp
,
2661 bp
, dq_f
->qlf_blkno
);
2665 ddq
= (xfs_disk_dquot_t
*)xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
2668 * At least the magic num portion should be on disk because this
2669 * was among a chunk of dquots created earlier, and we did some
2670 * minimal initialization then.
2672 if (xfs_qm_dqcheck(ddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2673 "xlog_recover_do_dquot_trans")) {
2675 return XFS_ERROR(EIO
);
2678 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
2680 ASSERT(dq_f
->qlf_size
== 2);
2681 ASSERT(bp
->b_target
->bt_mount
== mp
);
2682 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2683 xfs_bdwrite(mp
, bp
);
2689 * This routine is called to create an in-core extent free intent
2690 * item from the efi format structure which was logged on disk.
2691 * It allocates an in-core efi, copies the extents from the format
2692 * structure into it, and adds the efi to the AIL with the given
2696 xlog_recover_do_efi_trans(
2698 xlog_recover_item_t
*item
,
2704 xfs_efi_log_item_t
*efip
;
2705 xfs_efi_log_format_t
*efi_formatp
;
2707 if (pass
== XLOG_RECOVER_PASS1
) {
2711 efi_formatp
= item
->ri_buf
[0].i_addr
;
2714 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
2715 if ((error
= xfs_efi_copy_format(&(item
->ri_buf
[0]),
2716 &(efip
->efi_format
)))) {
2717 xfs_efi_item_free(efip
);
2720 efip
->efi_next_extent
= efi_formatp
->efi_nextents
;
2721 efip
->efi_flags
|= XFS_EFI_COMMITTED
;
2723 spin_lock(&log
->l_ailp
->xa_lock
);
2725 * xfs_trans_ail_update() drops the AIL lock.
2727 xfs_trans_ail_update(log
->l_ailp
, (xfs_log_item_t
*)efip
, lsn
);
2733 * This routine is called when an efd format structure is found in
2734 * a committed transaction in the log. It's purpose is to cancel
2735 * the corresponding efi if it was still in the log. To do this
2736 * it searches the AIL for the efi with an id equal to that in the
2737 * efd format structure. If we find it, we remove the efi from the
2741 xlog_recover_do_efd_trans(
2743 xlog_recover_item_t
*item
,
2746 xfs_efd_log_format_t
*efd_formatp
;
2747 xfs_efi_log_item_t
*efip
= NULL
;
2748 xfs_log_item_t
*lip
;
2750 struct xfs_ail_cursor cur
;
2751 struct xfs_ail
*ailp
= log
->l_ailp
;
2753 if (pass
== XLOG_RECOVER_PASS1
) {
2757 efd_formatp
= item
->ri_buf
[0].i_addr
;
2758 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
2759 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
2760 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
2761 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
2762 efi_id
= efd_formatp
->efd_efi_id
;
2765 * Search for the efi with the id in the efd format structure
2768 spin_lock(&ailp
->xa_lock
);
2769 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
2770 while (lip
!= NULL
) {
2771 if (lip
->li_type
== XFS_LI_EFI
) {
2772 efip
= (xfs_efi_log_item_t
*)lip
;
2773 if (efip
->efi_format
.efi_id
== efi_id
) {
2775 * xfs_trans_ail_delete() drops the
2778 xfs_trans_ail_delete(ailp
, lip
);
2779 xfs_efi_item_free(efip
);
2780 spin_lock(&ailp
->xa_lock
);
2784 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
2786 xfs_trans_ail_cursor_done(ailp
, &cur
);
2787 spin_unlock(&ailp
->xa_lock
);
2791 * Perform the transaction
2793 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2794 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2797 xlog_recover_do_trans(
2799 xlog_recover_t
*trans
,
2803 xlog_recover_item_t
*item
;
2805 error
= xlog_recover_reorder_trans(log
, trans
, pass
);
2809 list_for_each_entry(item
, &trans
->r_itemq
, ri_list
) {
2810 trace_xfs_log_recover_item_recover(log
, trans
, item
, pass
);
2811 switch (ITEM_TYPE(item
)) {
2813 error
= xlog_recover_do_buffer_trans(log
, item
, pass
);
2816 error
= xlog_recover_do_inode_trans(log
, item
, pass
);
2819 error
= xlog_recover_do_efi_trans(log
, item
,
2820 trans
->r_lsn
, pass
);
2823 xlog_recover_do_efd_trans(log
, item
, pass
);
2827 error
= xlog_recover_do_dquot_trans(log
, item
, pass
);
2829 case XFS_LI_QUOTAOFF
:
2830 error
= xlog_recover_do_quotaoff_trans(log
, item
,
2835 "XFS: invalid item type (%d) xlog_recover_do_trans", ITEM_TYPE(item
));
2837 error
= XFS_ERROR(EIO
);
2849 * Free up any resources allocated by the transaction
2851 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2854 xlog_recover_free_trans(
2855 xlog_recover_t
*trans
)
2857 xlog_recover_item_t
*item
, *n
;
2860 list_for_each_entry_safe(item
, n
, &trans
->r_itemq
, ri_list
) {
2861 /* Free the regions in the item. */
2862 list_del(&item
->ri_list
);
2863 for (i
= 0; i
< item
->ri_cnt
; i
++)
2864 kmem_free(item
->ri_buf
[i
].i_addr
);
2865 /* Free the item itself */
2866 kmem_free(item
->ri_buf
);
2869 /* Free the transaction recover structure */
2874 xlog_recover_commit_trans(
2876 xlog_recover_t
*trans
,
2881 hlist_del(&trans
->r_list
);
2882 if ((error
= xlog_recover_do_trans(log
, trans
, pass
)))
2884 xlog_recover_free_trans(trans
); /* no error */
2889 xlog_recover_unmount_trans(
2890 xlog_recover_t
*trans
)
2892 /* Do nothing now */
2893 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2898 * There are two valid states of the r_state field. 0 indicates that the
2899 * transaction structure is in a normal state. We have either seen the
2900 * start of the transaction or the last operation we added was not a partial
2901 * operation. If the last operation we added to the transaction was a
2902 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2904 * NOTE: skip LRs with 0 data length.
2907 xlog_recover_process_data(
2909 struct hlist_head rhash
[],
2910 xlog_rec_header_t
*rhead
,
2916 xlog_op_header_t
*ohead
;
2917 xlog_recover_t
*trans
;
2923 lp
= dp
+ be32_to_cpu(rhead
->h_len
);
2924 num_logops
= be32_to_cpu(rhead
->h_num_logops
);
2926 /* check the log format matches our own - else we can't recover */
2927 if (xlog_header_check_recover(log
->l_mp
, rhead
))
2928 return (XFS_ERROR(EIO
));
2930 while ((dp
< lp
) && num_logops
) {
2931 ASSERT(dp
+ sizeof(xlog_op_header_t
) <= lp
);
2932 ohead
= (xlog_op_header_t
*)dp
;
2933 dp
+= sizeof(xlog_op_header_t
);
2934 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
2935 ohead
->oh_clientid
!= XFS_LOG
) {
2937 "XFS: xlog_recover_process_data: bad clientid");
2939 return (XFS_ERROR(EIO
));
2941 tid
= be32_to_cpu(ohead
->oh_tid
);
2942 hash
= XLOG_RHASH(tid
);
2943 trans
= xlog_recover_find_tid(&rhash
[hash
], tid
);
2944 if (trans
== NULL
) { /* not found; add new tid */
2945 if (ohead
->oh_flags
& XLOG_START_TRANS
)
2946 xlog_recover_new_tid(&rhash
[hash
], tid
,
2947 be64_to_cpu(rhead
->h_lsn
));
2949 if (dp
+ be32_to_cpu(ohead
->oh_len
) > lp
) {
2951 "XFS: xlog_recover_process_data: bad length");
2953 return (XFS_ERROR(EIO
));
2955 flags
= ohead
->oh_flags
& ~XLOG_END_TRANS
;
2956 if (flags
& XLOG_WAS_CONT_TRANS
)
2957 flags
&= ~XLOG_CONTINUE_TRANS
;
2959 case XLOG_COMMIT_TRANS
:
2960 error
= xlog_recover_commit_trans(log
,
2963 case XLOG_UNMOUNT_TRANS
:
2964 error
= xlog_recover_unmount_trans(trans
);
2966 case XLOG_WAS_CONT_TRANS
:
2967 error
= xlog_recover_add_to_cont_trans(log
,
2969 be32_to_cpu(ohead
->oh_len
));
2971 case XLOG_START_TRANS
:
2973 "XFS: xlog_recover_process_data: bad transaction");
2975 error
= XFS_ERROR(EIO
);
2978 case XLOG_CONTINUE_TRANS
:
2979 error
= xlog_recover_add_to_trans(log
, trans
,
2980 dp
, be32_to_cpu(ohead
->oh_len
));
2984 "XFS: xlog_recover_process_data: bad flag");
2986 error
= XFS_ERROR(EIO
);
2992 dp
+= be32_to_cpu(ohead
->oh_len
);
2999 * Process an extent free intent item that was recovered from
3000 * the log. We need to free the extents that it describes.
3003 xlog_recover_process_efi(
3005 xfs_efi_log_item_t
*efip
)
3007 xfs_efd_log_item_t
*efdp
;
3012 xfs_fsblock_t startblock_fsb
;
3014 ASSERT(!(efip
->efi_flags
& XFS_EFI_RECOVERED
));
3017 * First check the validity of the extents described by the
3018 * EFI. If any are bad, then assume that all are bad and
3019 * just toss the EFI.
3021 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3022 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3023 startblock_fsb
= XFS_BB_TO_FSB(mp
,
3024 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
3025 if ((startblock_fsb
== 0) ||
3026 (extp
->ext_len
== 0) ||
3027 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
3028 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
3030 * This will pull the EFI from the AIL and
3031 * free the memory associated with it.
3033 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
3034 return XFS_ERROR(EIO
);
3038 tp
= xfs_trans_alloc(mp
, 0);
3039 error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0, 0, 0);
3042 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
3044 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3045 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3046 error
= xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
3049 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
3053 efip
->efi_flags
|= XFS_EFI_RECOVERED
;
3054 error
= xfs_trans_commit(tp
, 0);
3058 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3063 * When this is called, all of the EFIs which did not have
3064 * corresponding EFDs should be in the AIL. What we do now
3065 * is free the extents associated with each one.
3067 * Since we process the EFIs in normal transactions, they
3068 * will be removed at some point after the commit. This prevents
3069 * us from just walking down the list processing each one.
3070 * We'll use a flag in the EFI to skip those that we've already
3071 * processed and use the AIL iteration mechanism's generation
3072 * count to try to speed this up at least a bit.
3074 * When we start, we know that the EFIs are the only things in
3075 * the AIL. As we process them, however, other items are added
3076 * to the AIL. Since everything added to the AIL must come after
3077 * everything already in the AIL, we stop processing as soon as
3078 * we see something other than an EFI in the AIL.
3081 xlog_recover_process_efis(
3084 xfs_log_item_t
*lip
;
3085 xfs_efi_log_item_t
*efip
;
3087 struct xfs_ail_cursor cur
;
3088 struct xfs_ail
*ailp
;
3091 spin_lock(&ailp
->xa_lock
);
3092 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3093 while (lip
!= NULL
) {
3095 * We're done when we see something other than an EFI.
3096 * There should be no EFIs left in the AIL now.
3098 if (lip
->li_type
!= XFS_LI_EFI
) {
3100 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
3101 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
3107 * Skip EFIs that we've already processed.
3109 efip
= (xfs_efi_log_item_t
*)lip
;
3110 if (efip
->efi_flags
& XFS_EFI_RECOVERED
) {
3111 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3115 spin_unlock(&ailp
->xa_lock
);
3116 error
= xlog_recover_process_efi(log
->l_mp
, efip
);
3117 spin_lock(&ailp
->xa_lock
);
3120 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3123 xfs_trans_ail_cursor_done(ailp
, &cur
);
3124 spin_unlock(&ailp
->xa_lock
);
3129 * This routine performs a transaction to null out a bad inode pointer
3130 * in an agi unlinked inode hash bucket.
3133 xlog_recover_clear_agi_bucket(
3135 xfs_agnumber_t agno
,
3144 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3145 error
= xfs_trans_reserve(tp
, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp
),
3150 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
3154 agi
= XFS_BUF_TO_AGI(agibp
);
3155 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
3156 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3157 (sizeof(xfs_agino_t
) * bucket
);
3158 xfs_trans_log_buf(tp
, agibp
, offset
,
3159 (offset
+ sizeof(xfs_agino_t
) - 1));
3161 error
= xfs_trans_commit(tp
, 0);
3167 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3169 xfs_fs_cmn_err(CE_WARN
, mp
, "xlog_recover_clear_agi_bucket: "
3170 "failed to clear agi %d. Continuing.", agno
);
3175 xlog_recover_process_one_iunlink(
3176 struct xfs_mount
*mp
,
3177 xfs_agnumber_t agno
,
3181 struct xfs_buf
*ibp
;
3182 struct xfs_dinode
*dip
;
3183 struct xfs_inode
*ip
;
3187 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3188 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
);
3193 * Get the on disk inode to find the next inode in the bucket.
3195 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &ibp
, XBF_LOCK
);
3199 ASSERT(ip
->i_d
.di_nlink
== 0);
3200 ASSERT(ip
->i_d
.di_mode
!= 0);
3202 /* setup for the next pass */
3203 agino
= be32_to_cpu(dip
->di_next_unlinked
);
3207 * Prevent any DMAPI event from being sent when the reference on
3208 * the inode is dropped.
3210 ip
->i_d
.di_dmevmask
= 0;
3219 * We can't read in the inode this bucket points to, or this inode
3220 * is messed up. Just ditch this bucket of inodes. We will lose
3221 * some inodes and space, but at least we won't hang.
3223 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3224 * clear the inode pointer in the bucket.
3226 xlog_recover_clear_agi_bucket(mp
, agno
, bucket
);
3231 * xlog_iunlink_recover
3233 * This is called during recovery to process any inodes which
3234 * we unlinked but not freed when the system crashed. These
3235 * inodes will be on the lists in the AGI blocks. What we do
3236 * here is scan all the AGIs and fully truncate and free any
3237 * inodes found on the lists. Each inode is removed from the
3238 * lists when it has been fully truncated and is freed. The
3239 * freeing of the inode and its removal from the list must be
3243 xlog_recover_process_iunlinks(
3247 xfs_agnumber_t agno
;
3258 * Prevent any DMAPI event from being sent while in this function.
3260 mp_dmevmask
= mp
->m_dmevmask
;
3263 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3265 * Find the agi for this ag.
3267 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3270 * AGI is b0rked. Don't process it.
3272 * We should probably mark the filesystem as corrupt
3273 * after we've recovered all the ag's we can....
3277 agi
= XFS_BUF_TO_AGI(agibp
);
3279 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3280 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
3281 while (agino
!= NULLAGINO
) {
3283 * Release the agi buffer so that it can
3284 * be acquired in the normal course of the
3285 * transaction to truncate and free the inode.
3287 xfs_buf_relse(agibp
);
3289 agino
= xlog_recover_process_one_iunlink(mp
,
3290 agno
, agino
, bucket
);
3293 * Reacquire the agibuffer and continue around
3294 * the loop. This should never fail as we know
3295 * the buffer was good earlier on.
3297 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3299 agi
= XFS_BUF_TO_AGI(agibp
);
3304 * Release the buffer for the current agi so we can
3305 * go on to the next one.
3307 xfs_buf_relse(agibp
);
3310 mp
->m_dmevmask
= mp_dmevmask
;
3316 xlog_pack_data_checksum(
3318 xlog_in_core_t
*iclog
,
3325 up
= (__be32
*)iclog
->ic_datap
;
3326 /* divide length by 4 to get # words */
3327 for (i
= 0; i
< (size
>> 2); i
++) {
3328 chksum
^= be32_to_cpu(*up
);
3331 iclog
->ic_header
.h_chksum
= cpu_to_be32(chksum
);
3334 #define xlog_pack_data_checksum(log, iclog, size)
3338 * Stamp cycle number in every block
3343 xlog_in_core_t
*iclog
,
3347 int size
= iclog
->ic_offset
+ roundoff
;
3351 xlog_pack_data_checksum(log
, iclog
, size
);
3353 cycle_lsn
= CYCLE_LSN_DISK(iclog
->ic_header
.h_lsn
);
3355 dp
= iclog
->ic_datap
;
3356 for (i
= 0; i
< BTOBB(size
) &&
3357 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3358 iclog
->ic_header
.h_cycle_data
[i
] = *(__be32
*)dp
;
3359 *(__be32
*)dp
= cycle_lsn
;
3363 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3364 xlog_in_core_2_t
*xhdr
= iclog
->ic_data
;
3366 for ( ; i
< BTOBB(size
); i
++) {
3367 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3368 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3369 xhdr
[j
].hic_xheader
.xh_cycle_data
[k
] = *(__be32
*)dp
;
3370 *(__be32
*)dp
= cycle_lsn
;
3374 for (i
= 1; i
< log
->l_iclog_heads
; i
++) {
3375 xhdr
[i
].hic_xheader
.xh_cycle
= cycle_lsn
;
3382 xlog_rec_header_t
*rhead
,
3388 for (i
= 0; i
< BTOBB(be32_to_cpu(rhead
->h_len
)) &&
3389 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3390 *(__be32
*)dp
= *(__be32
*)&rhead
->h_cycle_data
[i
];
3394 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3395 xlog_in_core_2_t
*xhdr
= (xlog_in_core_2_t
*)rhead
;
3396 for ( ; i
< BTOBB(be32_to_cpu(rhead
->h_len
)); i
++) {
3397 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3398 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3399 *(__be32
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
3406 xlog_valid_rec_header(
3408 xlog_rec_header_t
*rhead
,
3413 if (unlikely(be32_to_cpu(rhead
->h_magicno
) != XLOG_HEADER_MAGIC_NUM
)) {
3414 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3415 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3416 return XFS_ERROR(EFSCORRUPTED
);
3419 (!rhead
->h_version
||
3420 (be32_to_cpu(rhead
->h_version
) & (~XLOG_VERSION_OKBITS
))))) {
3421 xlog_warn("XFS: %s: unrecognised log version (%d).",
3422 __func__
, be32_to_cpu(rhead
->h_version
));
3423 return XFS_ERROR(EIO
);
3426 /* LR body must have data or it wouldn't have been written */
3427 hlen
= be32_to_cpu(rhead
->h_len
);
3428 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
3429 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3430 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3431 return XFS_ERROR(EFSCORRUPTED
);
3433 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
3434 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3435 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3436 return XFS_ERROR(EFSCORRUPTED
);
3442 * Read the log from tail to head and process the log records found.
3443 * Handle the two cases where the tail and head are in the same cycle
3444 * and where the active portion of the log wraps around the end of
3445 * the physical log separately. The pass parameter is passed through
3446 * to the routines called to process the data and is not looked at
3450 xlog_do_recovery_pass(
3452 xfs_daddr_t head_blk
,
3453 xfs_daddr_t tail_blk
,
3456 xlog_rec_header_t
*rhead
;
3459 xfs_buf_t
*hbp
, *dbp
;
3460 int error
= 0, h_size
;
3461 int bblks
, split_bblks
;
3462 int hblks
, split_hblks
, wrapped_hblks
;
3463 struct hlist_head rhash
[XLOG_RHASH_SIZE
];
3465 ASSERT(head_blk
!= tail_blk
);
3468 * Read the header of the tail block and get the iclog buffer size from
3469 * h_size. Use this to tell how many sectors make up the log header.
3471 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3473 * When using variable length iclogs, read first sector of
3474 * iclog header and extract the header size from it. Get a
3475 * new hbp that is the correct size.
3477 hbp
= xlog_get_bp(log
, 1);
3481 error
= xlog_bread(log
, tail_blk
, 1, hbp
, &offset
);
3485 rhead
= (xlog_rec_header_t
*)offset
;
3486 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
3489 h_size
= be32_to_cpu(rhead
->h_size
);
3490 if ((be32_to_cpu(rhead
->h_version
) & XLOG_VERSION_2
) &&
3491 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
3492 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
3493 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
3496 hbp
= xlog_get_bp(log
, hblks
);
3501 ASSERT(log
->l_sectBBsize
== 1);
3503 hbp
= xlog_get_bp(log
, 1);
3504 h_size
= XLOG_BIG_RECORD_BSIZE
;
3509 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
3515 memset(rhash
, 0, sizeof(rhash
));
3516 if (tail_blk
<= head_blk
) {
3517 for (blk_no
= tail_blk
; blk_no
< head_blk
; ) {
3518 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3522 rhead
= (xlog_rec_header_t
*)offset
;
3523 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3527 /* blocks in data section */
3528 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3529 error
= xlog_bread(log
, blk_no
+ hblks
, bblks
, dbp
,
3534 xlog_unpack_data(rhead
, offset
, log
);
3535 if ((error
= xlog_recover_process_data(log
,
3536 rhash
, rhead
, offset
, pass
)))
3538 blk_no
+= bblks
+ hblks
;
3542 * Perform recovery around the end of the physical log.
3543 * When the head is not on the same cycle number as the tail,
3544 * we can't do a sequential recovery as above.
3547 while (blk_no
< log
->l_logBBsize
) {
3549 * Check for header wrapping around physical end-of-log
3551 offset
= XFS_BUF_PTR(hbp
);
3554 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
3555 /* Read header in one read */
3556 error
= xlog_bread(log
, blk_no
, hblks
, hbp
,
3561 /* This LR is split across physical log end */
3562 if (blk_no
!= log
->l_logBBsize
) {
3563 /* some data before physical log end */
3564 ASSERT(blk_no
<= INT_MAX
);
3565 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
3566 ASSERT(split_hblks
> 0);
3567 error
= xlog_bread(log
, blk_no
,
3575 * Note: this black magic still works with
3576 * large sector sizes (non-512) only because:
3577 * - we increased the buffer size originally
3578 * by 1 sector giving us enough extra space
3579 * for the second read;
3580 * - the log start is guaranteed to be sector
3582 * - we read the log end (LR header start)
3583 * _first_, then the log start (LR header end)
3584 * - order is important.
3586 wrapped_hblks
= hblks
- split_hblks
;
3587 error
= XFS_BUF_SET_PTR(hbp
,
3588 offset
+ BBTOB(split_hblks
),
3589 BBTOB(hblks
- split_hblks
));
3593 error
= xlog_bread_noalign(log
, 0,
3594 wrapped_hblks
, hbp
);
3598 error
= XFS_BUF_SET_PTR(hbp
, offset
,
3603 rhead
= (xlog_rec_header_t
*)offset
;
3604 error
= xlog_valid_rec_header(log
, rhead
,
3605 split_hblks
? blk_no
: 0);
3609 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3612 /* Read in data for log record */
3613 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
3614 error
= xlog_bread(log
, blk_no
, bblks
, dbp
,
3619 /* This log record is split across the
3620 * physical end of log */
3621 offset
= XFS_BUF_PTR(dbp
);
3623 if (blk_no
!= log
->l_logBBsize
) {
3624 /* some data is before the physical
3626 ASSERT(!wrapped_hblks
);
3627 ASSERT(blk_no
<= INT_MAX
);
3629 log
->l_logBBsize
- (int)blk_no
;
3630 ASSERT(split_bblks
> 0);
3631 error
= xlog_bread(log
, blk_no
,
3639 * Note: this black magic still works with
3640 * large sector sizes (non-512) only because:
3641 * - we increased the buffer size originally
3642 * by 1 sector giving us enough extra space
3643 * for the second read;
3644 * - the log start is guaranteed to be sector
3646 * - we read the log end (LR header start)
3647 * _first_, then the log start (LR header end)
3648 * - order is important.
3650 error
= XFS_BUF_SET_PTR(dbp
,
3651 offset
+ BBTOB(split_bblks
),
3652 BBTOB(bblks
- split_bblks
));
3656 error
= xlog_bread_noalign(log
, wrapped_hblks
,
3657 bblks
- split_bblks
,
3662 error
= XFS_BUF_SET_PTR(dbp
, offset
, h_size
);
3666 xlog_unpack_data(rhead
, offset
, log
);
3667 if ((error
= xlog_recover_process_data(log
, rhash
,
3668 rhead
, offset
, pass
)))
3673 ASSERT(blk_no
>= log
->l_logBBsize
);
3674 blk_no
-= log
->l_logBBsize
;
3676 /* read first part of physical log */
3677 while (blk_no
< head_blk
) {
3678 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3682 rhead
= (xlog_rec_header_t
*)offset
;
3683 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3687 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3688 error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
,
3693 xlog_unpack_data(rhead
, offset
, log
);
3694 if ((error
= xlog_recover_process_data(log
, rhash
,
3695 rhead
, offset
, pass
)))
3697 blk_no
+= bblks
+ hblks
;
3709 * Do the recovery of the log. We actually do this in two phases.
3710 * The two passes are necessary in order to implement the function
3711 * of cancelling a record written into the log. The first pass
3712 * determines those things which have been cancelled, and the
3713 * second pass replays log items normally except for those which
3714 * have been cancelled. The handling of the replay and cancellations
3715 * takes place in the log item type specific routines.
3717 * The table of items which have cancel records in the log is allocated
3718 * and freed at this level, since only here do we know when all of
3719 * the log recovery has been completed.
3722 xlog_do_log_recovery(
3724 xfs_daddr_t head_blk
,
3725 xfs_daddr_t tail_blk
)
3729 ASSERT(head_blk
!= tail_blk
);
3732 * First do a pass to find all of the cancelled buf log items.
3733 * Store them in the buf_cancel_table for use in the second pass.
3735 log
->l_buf_cancel_table
=
3736 (xfs_buf_cancel_t
**)kmem_zalloc(XLOG_BC_TABLE_SIZE
*
3737 sizeof(xfs_buf_cancel_t
*),
3739 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3740 XLOG_RECOVER_PASS1
);
3742 kmem_free(log
->l_buf_cancel_table
);
3743 log
->l_buf_cancel_table
= NULL
;
3747 * Then do a second pass to actually recover the items in the log.
3748 * When it is complete free the table of buf cancel items.
3750 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3751 XLOG_RECOVER_PASS2
);
3756 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
3757 ASSERT(log
->l_buf_cancel_table
[i
] == NULL
);
3761 kmem_free(log
->l_buf_cancel_table
);
3762 log
->l_buf_cancel_table
= NULL
;
3768 * Do the actual recovery
3773 xfs_daddr_t head_blk
,
3774 xfs_daddr_t tail_blk
)
3781 * First replay the images in the log.
3783 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
3788 XFS_bflush(log
->l_mp
->m_ddev_targp
);
3791 * If IO errors happened during recovery, bail out.
3793 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
3798 * We now update the tail_lsn since much of the recovery has completed
3799 * and there may be space available to use. If there were no extent
3800 * or iunlinks, we can free up the entire log and set the tail_lsn to
3801 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3802 * lsn of the last known good LR on disk. If there are extent frees
3803 * or iunlinks they will have some entries in the AIL; so we look at
3804 * the AIL to determine how to set the tail_lsn.
3806 xlog_assign_tail_lsn(log
->l_mp
);
3809 * Now that we've finished replaying all buffer and inode
3810 * updates, re-read in the superblock.
3812 bp
= xfs_getsb(log
->l_mp
, 0);
3814 ASSERT(!(XFS_BUF_ISWRITE(bp
)));
3815 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp
)));
3817 XFS_BUF_UNASYNC(bp
);
3818 xfsbdstrat(log
->l_mp
, bp
);
3819 error
= xfs_buf_iowait(bp
);
3821 xfs_ioerror_alert("xlog_do_recover",
3822 log
->l_mp
, bp
, XFS_BUF_ADDR(bp
));
3828 /* Convert superblock from on-disk format */
3829 sbp
= &log
->l_mp
->m_sb
;
3830 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
3831 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
3832 ASSERT(xfs_sb_good_version(sbp
));
3835 /* We've re-read the superblock so re-initialize per-cpu counters */
3836 xfs_icsb_reinit_counters(log
->l_mp
);
3838 xlog_recover_check_summary(log
);
3840 /* Normal transactions can now occur */
3841 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
3846 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3848 * Return error or zero.
3854 xfs_daddr_t head_blk
, tail_blk
;
3857 /* find the tail of the log */
3858 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
)))
3861 if (tail_blk
!= head_blk
) {
3862 /* There used to be a comment here:
3864 * disallow recovery on read-only mounts. note -- mount
3865 * checks for ENOSPC and turns it into an intelligent
3867 * ...but this is no longer true. Now, unless you specify
3868 * NORECOVERY (in which case this function would never be
3869 * called), we just go ahead and recover. We do this all
3870 * under the vfs layer, so we can get away with it unless
3871 * the device itself is read-only, in which case we fail.
3873 if ((error
= xfs_dev_is_read_only(log
->l_mp
, "recovery"))) {
3878 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3879 log
->l_mp
->m_fsname
, log
->l_mp
->m_logname
?
3880 log
->l_mp
->m_logname
: "internal");
3882 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
3883 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
3889 * In the first part of recovery we replay inodes and buffers and build
3890 * up the list of extent free items which need to be processed. Here
3891 * we process the extent free items and clean up the on disk unlinked
3892 * inode lists. This is separated from the first part of recovery so
3893 * that the root and real-time bitmap inodes can be read in from disk in
3894 * between the two stages. This is necessary so that we can free space
3895 * in the real-time portion of the file system.
3898 xlog_recover_finish(
3902 * Now we're ready to do the transactions needed for the
3903 * rest of recovery. Start with completing all the extent
3904 * free intent records and then process the unlinked inode
3905 * lists. At this point, we essentially run in normal mode
3906 * except that we're still performing recovery actions
3907 * rather than accepting new requests.
3909 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
3911 error
= xlog_recover_process_efis(log
);
3914 "Failed to recover EFIs on filesystem: %s",
3915 log
->l_mp
->m_fsname
);
3919 * Sync the log to get all the EFIs out of the AIL.
3920 * This isn't absolutely necessary, but it helps in
3921 * case the unlink transactions would have problems
3922 * pushing the EFIs out of the way.
3924 xfs_log_force(log
->l_mp
, XFS_LOG_SYNC
);
3926 xlog_recover_process_iunlinks(log
);
3928 xlog_recover_check_summary(log
);
3931 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3932 log
->l_mp
->m_fsname
, log
->l_mp
->m_logname
?
3933 log
->l_mp
->m_logname
: "internal");
3934 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
3937 "!Ending clean XFS mount for filesystem: %s\n",
3938 log
->l_mp
->m_fsname
);
3946 * Read all of the agf and agi counters and check that they
3947 * are consistent with the superblock counters.
3950 xlog_recover_check_summary(
3957 xfs_agnumber_t agno
;
3958 __uint64_t freeblks
;
3968 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3969 error
= xfs_read_agf(mp
, NULL
, agno
, 0, &agfbp
);
3971 xfs_fs_cmn_err(CE_ALERT
, mp
,
3972 "xlog_recover_check_summary(agf)"
3973 "agf read failed agno %d error %d",
3976 agfp
= XFS_BUF_TO_AGF(agfbp
);
3977 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
3978 be32_to_cpu(agfp
->agf_flcount
);
3979 xfs_buf_relse(agfbp
);
3982 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3984 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agibp
);
3986 itotal
+= be32_to_cpu(agi
->agi_count
);
3987 ifree
+= be32_to_cpu(agi
->agi_freecount
);
3988 xfs_buf_relse(agibp
);