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_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_trans.h"
32 #include "xfs_log_priv.h"
33 #include "xfs_log_recover.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_extfree_item.h"
36 #include "xfs_trans_priv.h"
37 #include "xfs_alloc.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_quota.h"
40 #include "xfs_cksum.h"
41 #include "xfs_trace.h"
42 #include "xfs_icache.h"
43 #include "xfs_bmap_btree.h"
44 #include "xfs_error.h"
47 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
54 xlog_clear_stale_blocks(
59 xlog_recover_check_summary(
62 #define xlog_recover_check_summary(log)
65 xlog_do_recovery_pass(
66 struct xlog
*, xfs_daddr_t
, xfs_daddr_t
, int, xfs_daddr_t
*);
69 * This structure is used during recovery to record the buf log items which
70 * have been canceled and should not be replayed.
72 struct xfs_buf_cancel
{
76 struct list_head bc_list
;
80 * Sector aligned buffer routines for buffer create/read/write/access
84 * Verify the given count of basic blocks is valid number of blocks
85 * to specify for an operation involving the given XFS log buffer.
86 * Returns nonzero if the count is valid, 0 otherwise.
90 xlog_buf_bbcount_valid(
94 return bbcount
> 0 && bbcount
<= log
->l_logBBsize
;
98 * Allocate a buffer to hold log data. The buffer needs to be able
99 * to map to a range of nbblks basic blocks at any valid (basic
100 * block) offset within the log.
109 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
110 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
112 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
117 * We do log I/O in units of log sectors (a power-of-2
118 * multiple of the basic block size), so we round up the
119 * requested size to accommodate the basic blocks required
120 * for complete log sectors.
122 * In addition, the buffer may be used for a non-sector-
123 * aligned block offset, in which case an I/O of the
124 * requested size could extend beyond the end of the
125 * buffer. If the requested size is only 1 basic block it
126 * will never straddle a sector boundary, so this won't be
127 * an issue. Nor will this be a problem if the log I/O is
128 * done in basic blocks (sector size 1). But otherwise we
129 * extend the buffer by one extra log sector to ensure
130 * there's space to accommodate this possibility.
132 if (nbblks
> 1 && log
->l_sectBBsize
> 1)
133 nbblks
+= log
->l_sectBBsize
;
134 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
136 bp
= xfs_buf_get_uncached(log
->l_mp
->m_logdev_targp
, nbblks
, 0);
150 * Return the address of the start of the given block number's data
151 * in a log buffer. The buffer covers a log sector-aligned region.
160 xfs_daddr_t offset
= blk_no
& ((xfs_daddr_t
)log
->l_sectBBsize
- 1);
162 ASSERT(offset
+ nbblks
<= bp
->b_length
);
163 return bp
->b_addr
+ BBTOB(offset
);
168 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
179 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
180 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
182 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
183 return -EFSCORRUPTED
;
186 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
187 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
190 ASSERT(nbblks
<= bp
->b_length
);
192 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
193 bp
->b_flags
|= XBF_READ
;
194 bp
->b_io_length
= nbblks
;
197 error
= xfs_buf_submit_wait(bp
);
198 if (error
&& !XFS_FORCED_SHUTDOWN(log
->l_mp
))
199 xfs_buf_ioerror_alert(bp
, __func__
);
213 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
217 *offset
= xlog_align(log
, blk_no
, nbblks
, bp
);
222 * Read at an offset into the buffer. Returns with the buffer in it's original
223 * state regardless of the result of the read.
228 xfs_daddr_t blk_no
, /* block to read from */
229 int nbblks
, /* blocks to read */
233 char *orig_offset
= bp
->b_addr
;
234 int orig_len
= BBTOB(bp
->b_length
);
237 error
= xfs_buf_associate_memory(bp
, offset
, BBTOB(nbblks
));
241 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
243 /* must reset buffer pointer even on error */
244 error2
= xfs_buf_associate_memory(bp
, orig_offset
, orig_len
);
251 * Write out the buffer at the given block for the given number of blocks.
252 * The buffer is kept locked across the write and is returned locked.
253 * This can only be used for synchronous log writes.
264 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
265 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
267 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
268 return -EFSCORRUPTED
;
271 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
272 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
275 ASSERT(nbblks
<= bp
->b_length
);
277 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
280 bp
->b_io_length
= nbblks
;
283 error
= xfs_bwrite(bp
);
285 xfs_buf_ioerror_alert(bp
, __func__
);
292 * dump debug superblock and log record information
295 xlog_header_check_dump(
297 xlog_rec_header_t
*head
)
299 xfs_debug(mp
, "%s: SB : uuid = %pU, fmt = %d",
300 __func__
, &mp
->m_sb
.sb_uuid
, XLOG_FMT
);
301 xfs_debug(mp
, " log : uuid = %pU, fmt = %d",
302 &head
->h_fs_uuid
, be32_to_cpu(head
->h_fmt
));
305 #define xlog_header_check_dump(mp, head)
309 * check log record header for recovery
312 xlog_header_check_recover(
314 xlog_rec_header_t
*head
)
316 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
319 * IRIX doesn't write the h_fmt field and leaves it zeroed
320 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
321 * a dirty log created in IRIX.
323 if (unlikely(head
->h_fmt
!= cpu_to_be32(XLOG_FMT
))) {
325 "dirty log written in incompatible format - can't recover");
326 xlog_header_check_dump(mp
, head
);
327 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
328 XFS_ERRLEVEL_HIGH
, mp
);
329 return -EFSCORRUPTED
;
330 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
332 "dirty log entry has mismatched uuid - can't recover");
333 xlog_header_check_dump(mp
, head
);
334 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
335 XFS_ERRLEVEL_HIGH
, mp
);
336 return -EFSCORRUPTED
;
342 * read the head block of the log and check the header
345 xlog_header_check_mount(
347 xlog_rec_header_t
*head
)
349 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
351 if (uuid_is_nil(&head
->h_fs_uuid
)) {
353 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
354 * h_fs_uuid is nil, we assume this log was last mounted
355 * by IRIX and continue.
357 xfs_warn(mp
, "nil uuid in log - IRIX style log");
358 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
359 xfs_warn(mp
, "log has mismatched uuid - can't recover");
360 xlog_header_check_dump(mp
, head
);
361 XFS_ERROR_REPORT("xlog_header_check_mount",
362 XFS_ERRLEVEL_HIGH
, mp
);
363 return -EFSCORRUPTED
;
374 * We're not going to bother about retrying
375 * this during recovery. One strike!
377 if (!XFS_FORCED_SHUTDOWN(bp
->b_target
->bt_mount
)) {
378 xfs_buf_ioerror_alert(bp
, __func__
);
379 xfs_force_shutdown(bp
->b_target
->bt_mount
,
380 SHUTDOWN_META_IO_ERROR
);
388 * This routine finds (to an approximation) the first block in the physical
389 * log which contains the given cycle. It uses a binary search algorithm.
390 * Note that the algorithm can not be perfect because the disk will not
391 * necessarily be perfect.
394 xlog_find_cycle_start(
397 xfs_daddr_t first_blk
,
398 xfs_daddr_t
*last_blk
,
408 mid_blk
= BLK_AVG(first_blk
, end_blk
);
409 while (mid_blk
!= first_blk
&& mid_blk
!= end_blk
) {
410 error
= xlog_bread(log
, mid_blk
, 1, bp
, &offset
);
413 mid_cycle
= xlog_get_cycle(offset
);
414 if (mid_cycle
== cycle
)
415 end_blk
= mid_blk
; /* last_half_cycle == mid_cycle */
417 first_blk
= mid_blk
; /* first_half_cycle == mid_cycle */
418 mid_blk
= BLK_AVG(first_blk
, end_blk
);
420 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == end_blk
) ||
421 (mid_blk
== end_blk
&& mid_blk
-1 == first_blk
));
429 * Check that a range of blocks does not contain stop_on_cycle_no.
430 * Fill in *new_blk with the block offset where such a block is
431 * found, or with -1 (an invalid block number) if there is no such
432 * block in the range. The scan needs to occur from front to back
433 * and the pointer into the region must be updated since a later
434 * routine will need to perform another test.
437 xlog_find_verify_cycle(
439 xfs_daddr_t start_blk
,
441 uint stop_on_cycle_no
,
442 xfs_daddr_t
*new_blk
)
452 * Greedily allocate a buffer big enough to handle the full
453 * range of basic blocks we'll be examining. If that fails,
454 * try a smaller size. We need to be able to read at least
455 * a log sector, or we're out of luck.
457 bufblks
= 1 << ffs(nbblks
);
458 while (bufblks
> log
->l_logBBsize
)
460 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
462 if (bufblks
< log
->l_sectBBsize
)
466 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
469 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
471 error
= xlog_bread(log
, i
, bcount
, bp
, &buf
);
475 for (j
= 0; j
< bcount
; j
++) {
476 cycle
= xlog_get_cycle(buf
);
477 if (cycle
== stop_on_cycle_no
) {
494 * Potentially backup over partial log record write.
496 * In the typical case, last_blk is the number of the block directly after
497 * a good log record. Therefore, we subtract one to get the block number
498 * of the last block in the given buffer. extra_bblks contains the number
499 * of blocks we would have read on a previous read. This happens when the
500 * last log record is split over the end of the physical log.
502 * extra_bblks is the number of blocks potentially verified on a previous
503 * call to this routine.
506 xlog_find_verify_log_record(
508 xfs_daddr_t start_blk
,
509 xfs_daddr_t
*last_blk
,
515 xlog_rec_header_t
*head
= NULL
;
518 int num_blks
= *last_blk
- start_blk
;
521 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
523 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
524 if (!(bp
= xlog_get_bp(log
, 1)))
528 error
= xlog_bread(log
, start_blk
, num_blks
, bp
, &offset
);
531 offset
+= ((num_blks
- 1) << BBSHIFT
);
534 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
536 /* valid log record not found */
538 "Log inconsistent (didn't find previous header)");
545 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
550 head
= (xlog_rec_header_t
*)offset
;
552 if (head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))
560 * We hit the beginning of the physical log & still no header. Return
561 * to caller. If caller can handle a return of -1, then this routine
562 * will be called again for the end of the physical log.
570 * We have the final block of the good log (the first block
571 * of the log record _before_ the head. So we check the uuid.
573 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
577 * We may have found a log record header before we expected one.
578 * last_blk will be the 1st block # with a given cycle #. We may end
579 * up reading an entire log record. In this case, we don't want to
580 * reset last_blk. Only when last_blk points in the middle of a log
581 * record do we update last_blk.
583 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
584 uint h_size
= be32_to_cpu(head
->h_size
);
586 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
587 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
593 if (*last_blk
- i
+ extra_bblks
!=
594 BTOBB(be32_to_cpu(head
->h_len
)) + xhdrs
)
603 * Head is defined to be the point of the log where the next log write
604 * could go. This means that incomplete LR writes at the end are
605 * eliminated when calculating the head. We aren't guaranteed that previous
606 * LR have complete transactions. We only know that a cycle number of
607 * current cycle number -1 won't be present in the log if we start writing
608 * from our current block number.
610 * last_blk contains the block number of the first block with a given
613 * Return: zero if normal, non-zero if error.
618 xfs_daddr_t
*return_head_blk
)
622 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
624 uint first_half_cycle
, last_half_cycle
;
626 int error
, log_bbnum
= log
->l_logBBsize
;
628 /* Is the end of the log device zeroed? */
629 error
= xlog_find_zeroed(log
, &first_blk
);
631 xfs_warn(log
->l_mp
, "empty log check failed");
635 *return_head_blk
= first_blk
;
637 /* Is the whole lot zeroed? */
639 /* Linux XFS shouldn't generate totally zeroed logs -
640 * mkfs etc write a dummy unmount record to a fresh
641 * log so we can store the uuid in there
643 xfs_warn(log
->l_mp
, "totally zeroed log");
649 first_blk
= 0; /* get cycle # of 1st block */
650 bp
= xlog_get_bp(log
, 1);
654 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
658 first_half_cycle
= xlog_get_cycle(offset
);
660 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
661 error
= xlog_bread(log
, last_blk
, 1, bp
, &offset
);
665 last_half_cycle
= xlog_get_cycle(offset
);
666 ASSERT(last_half_cycle
!= 0);
669 * If the 1st half cycle number is equal to the last half cycle number,
670 * then the entire log is stamped with the same cycle number. In this
671 * case, head_blk can't be set to zero (which makes sense). The below
672 * math doesn't work out properly with head_blk equal to zero. Instead,
673 * we set it to log_bbnum which is an invalid block number, but this
674 * value makes the math correct. If head_blk doesn't changed through
675 * all the tests below, *head_blk is set to zero at the very end rather
676 * than log_bbnum. In a sense, log_bbnum and zero are the same block
677 * in a circular file.
679 if (first_half_cycle
== last_half_cycle
) {
681 * In this case we believe that the entire log should have
682 * cycle number last_half_cycle. We need to scan backwards
683 * from the end verifying that there are no holes still
684 * containing last_half_cycle - 1. If we find such a hole,
685 * then the start of that hole will be the new head. The
686 * simple case looks like
687 * x | x ... | x - 1 | x
688 * Another case that fits this picture would be
689 * x | x + 1 | x ... | x
690 * In this case the head really is somewhere at the end of the
691 * log, as one of the latest writes at the beginning was
694 * x | x + 1 | x ... | x - 1 | x
695 * This is really the combination of the above two cases, and
696 * the head has to end up at the start of the x-1 hole at the
699 * In the 256k log case, we will read from the beginning to the
700 * end of the log and search for cycle numbers equal to x-1.
701 * We don't worry about the x+1 blocks that we encounter,
702 * because we know that they cannot be the head since the log
705 head_blk
= log_bbnum
;
706 stop_on_cycle
= last_half_cycle
- 1;
709 * In this case we want to find the first block with cycle
710 * number matching last_half_cycle. We expect the log to be
712 * x + 1 ... | x ... | x
713 * The first block with cycle number x (last_half_cycle) will
714 * be where the new head belongs. First we do a binary search
715 * for the first occurrence of last_half_cycle. The binary
716 * search may not be totally accurate, so then we scan back
717 * from there looking for occurrences of last_half_cycle before
718 * us. If that backwards scan wraps around the beginning of
719 * the log, then we look for occurrences of last_half_cycle - 1
720 * at the end of the log. The cases we're looking for look
722 * v binary search stopped here
723 * x + 1 ... | x | x + 1 | x ... | x
724 * ^ but we want to locate this spot
726 * <---------> less than scan distance
727 * x + 1 ... | x ... | x - 1 | x
728 * ^ we want to locate this spot
730 stop_on_cycle
= last_half_cycle
;
731 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
732 &head_blk
, last_half_cycle
)))
737 * Now validate the answer. Scan back some number of maximum possible
738 * blocks and make sure each one has the expected cycle number. The
739 * maximum is determined by the total possible amount of buffering
740 * in the in-core log. The following number can be made tighter if
741 * we actually look at the block size of the filesystem.
743 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
744 if (head_blk
>= num_scan_bblks
) {
746 * We are guaranteed that the entire check can be performed
749 start_blk
= head_blk
- num_scan_bblks
;
750 if ((error
= xlog_find_verify_cycle(log
,
751 start_blk
, num_scan_bblks
,
752 stop_on_cycle
, &new_blk
)))
756 } else { /* need to read 2 parts of log */
758 * We are going to scan backwards in the log in two parts.
759 * First we scan the physical end of the log. In this part
760 * of the log, we are looking for blocks with cycle number
761 * last_half_cycle - 1.
762 * If we find one, then we know that the log starts there, as
763 * we've found a hole that didn't get written in going around
764 * the end of the physical log. The simple case for this is
765 * x + 1 ... | x ... | x - 1 | x
766 * <---------> less than scan distance
767 * If all of the blocks at the end of the log have cycle number
768 * last_half_cycle, then we check the blocks at the start of
769 * the log looking for occurrences of last_half_cycle. If we
770 * find one, then our current estimate for the location of the
771 * first occurrence of last_half_cycle is wrong and we move
772 * back to the hole we've found. This case looks like
773 * x + 1 ... | x | x + 1 | x ...
774 * ^ binary search stopped here
775 * Another case we need to handle that only occurs in 256k
777 * x + 1 ... | x ... | x+1 | x ...
778 * ^ binary search stops here
779 * In a 256k log, the scan at the end of the log will see the
780 * x + 1 blocks. We need to skip past those since that is
781 * certainly not the head of the log. By searching for
782 * last_half_cycle-1 we accomplish that.
784 ASSERT(head_blk
<= INT_MAX
&&
785 (xfs_daddr_t
) num_scan_bblks
>= head_blk
);
786 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
787 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
788 num_scan_bblks
- (int)head_blk
,
789 (stop_on_cycle
- 1), &new_blk
)))
797 * Scan beginning of log now. The last part of the physical
798 * log is good. This scan needs to verify that it doesn't find
799 * the last_half_cycle.
802 ASSERT(head_blk
<= INT_MAX
);
803 if ((error
= xlog_find_verify_cycle(log
,
804 start_blk
, (int)head_blk
,
805 stop_on_cycle
, &new_blk
)))
813 * Now we need to make sure head_blk is not pointing to a block in
814 * the middle of a log record.
816 num_scan_bblks
= XLOG_REC_SHIFT(log
);
817 if (head_blk
>= num_scan_bblks
) {
818 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
820 /* start ptr at last block ptr before head_blk */
821 error
= xlog_find_verify_log_record(log
, start_blk
, &head_blk
, 0);
828 ASSERT(head_blk
<= INT_MAX
);
829 error
= xlog_find_verify_log_record(log
, start_blk
, &head_blk
, 0);
833 /* We hit the beginning of the log during our search */
834 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
836 ASSERT(start_blk
<= INT_MAX
&&
837 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
838 ASSERT(head_blk
<= INT_MAX
);
839 error
= xlog_find_verify_log_record(log
, start_blk
,
840 &new_blk
, (int)head_blk
);
845 if (new_blk
!= log_bbnum
)
852 if (head_blk
== log_bbnum
)
853 *return_head_blk
= 0;
855 *return_head_blk
= head_blk
;
857 * When returning here, we have a good block number. Bad block
858 * means that during a previous crash, we didn't have a clean break
859 * from cycle number N to cycle number N-1. In this case, we need
860 * to find the first block with cycle number N-1.
868 xfs_warn(log
->l_mp
, "failed to find log head");
873 * Seek backwards in the log for log record headers.
875 * Given a starting log block, walk backwards until we find the provided number
876 * of records or hit the provided tail block. The return value is the number of
877 * records encountered or a negative error code. The log block and buffer
878 * pointer of the last record seen are returned in rblk and rhead respectively.
881 xlog_rseek_logrec_hdr(
883 xfs_daddr_t head_blk
,
884 xfs_daddr_t tail_blk
,
888 struct xlog_rec_header
**rhead
,
900 * Walk backwards from the head block until we hit the tail or the first
903 end_blk
= head_blk
> tail_blk
? tail_blk
: 0;
904 for (i
= (int) head_blk
- 1; i
>= end_blk
; i
--) {
905 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
909 if (*(__be32
*) offset
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
911 *rhead
= (struct xlog_rec_header
*) offset
;
912 if (++found
== count
)
918 * If we haven't hit the tail block or the log record header count,
919 * start looking again from the end of the physical log. Note that
920 * callers can pass head == tail if the tail is not yet known.
922 if (tail_blk
>= head_blk
&& found
!= count
) {
923 for (i
= log
->l_logBBsize
- 1; i
>= (int) tail_blk
; i
--) {
924 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
928 if (*(__be32
*)offset
==
929 cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
932 *rhead
= (struct xlog_rec_header
*) offset
;
933 if (++found
== count
)
946 * Seek forward in the log for log record headers.
948 * Given head and tail blocks, walk forward from the tail block until we find
949 * the provided number of records or hit the head block. The return value is the
950 * number of records encountered or a negative error code. The log block and
951 * buffer pointer of the last record seen are returned in rblk and rhead
955 xlog_seek_logrec_hdr(
957 xfs_daddr_t head_blk
,
958 xfs_daddr_t tail_blk
,
962 struct xlog_rec_header
**rhead
,
974 * Walk forward from the tail block until we hit the head or the last
977 end_blk
= head_blk
> tail_blk
? head_blk
: log
->l_logBBsize
- 1;
978 for (i
= (int) tail_blk
; i
<= end_blk
; i
++) {
979 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
983 if (*(__be32
*) offset
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
985 *rhead
= (struct xlog_rec_header
*) offset
;
986 if (++found
== count
)
992 * If we haven't hit the head block or the log record header count,
993 * start looking again from the start of the physical log.
995 if (tail_blk
> head_blk
&& found
!= count
) {
996 for (i
= 0; i
< (int) head_blk
; i
++) {
997 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
1001 if (*(__be32
*)offset
==
1002 cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
1005 *rhead
= (struct xlog_rec_header
*) offset
;
1006 if (++found
== count
)
1019 * Check the log tail for torn writes. This is required when torn writes are
1020 * detected at the head and the head had to be walked back to a previous record.
1021 * The tail of the previous record must now be verified to ensure the torn
1022 * writes didn't corrupt the previous tail.
1024 * Return an error if CRC verification fails as recovery cannot proceed.
1029 xfs_daddr_t head_blk
,
1030 xfs_daddr_t tail_blk
)
1032 struct xlog_rec_header
*thead
;
1034 xfs_daddr_t first_bad
;
1038 xfs_daddr_t tmp_head
;
1040 bp
= xlog_get_bp(log
, 1);
1045 * Seek XLOG_MAX_ICLOGS + 1 records past the current tail record to get
1046 * a temporary head block that points after the last possible
1047 * concurrently written record of the tail.
1049 count
= xlog_seek_logrec_hdr(log
, head_blk
, tail_blk
,
1050 XLOG_MAX_ICLOGS
+ 1, bp
, &tmp_head
, &thead
,
1058 * If the call above didn't find XLOG_MAX_ICLOGS + 1 records, we ran
1059 * into the actual log head. tmp_head points to the start of the record
1060 * so update it to the actual head block.
1062 if (count
< XLOG_MAX_ICLOGS
+ 1)
1063 tmp_head
= head_blk
;
1066 * We now have a tail and temporary head block that covers at least
1067 * XLOG_MAX_ICLOGS records from the tail. We need to verify that these
1068 * records were completely written. Run a CRC verification pass from
1069 * tail to head and return the result.
1071 error
= xlog_do_recovery_pass(log
, tmp_head
, tail_blk
,
1072 XLOG_RECOVER_CRCPASS
, &first_bad
);
1080 * Detect and trim torn writes from the head of the log.
1082 * Storage without sector atomicity guarantees can result in torn writes in the
1083 * log in the event of a crash. Our only means to detect this scenario is via
1084 * CRC verification. While we can't always be certain that CRC verification
1085 * failure is due to a torn write vs. an unrelated corruption, we do know that
1086 * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
1087 * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
1088 * the log and treat failures in this range as torn writes as a matter of
1089 * policy. In the event of CRC failure, the head is walked back to the last good
1090 * record in the log and the tail is updated from that record and verified.
1095 xfs_daddr_t
*head_blk
, /* in/out: unverified head */
1096 xfs_daddr_t
*tail_blk
, /* out: tail block */
1098 xfs_daddr_t
*rhead_blk
, /* start blk of last record */
1099 struct xlog_rec_header
**rhead
, /* ptr to last record */
1100 bool *wrapped
) /* last rec. wraps phys. log */
1102 struct xlog_rec_header
*tmp_rhead
;
1103 struct xfs_buf
*tmp_bp
;
1104 xfs_daddr_t first_bad
;
1105 xfs_daddr_t tmp_rhead_blk
;
1111 * Search backwards through the log looking for the log record header
1112 * block. This wraps all the way back around to the head so something is
1113 * seriously wrong if we can't find it.
1115 found
= xlog_rseek_logrec_hdr(log
, *head_blk
, *head_blk
, 1, bp
, rhead_blk
,
1120 xfs_warn(log
->l_mp
, "%s: couldn't find sync record", __func__
);
1124 *tail_blk
= BLOCK_LSN(be64_to_cpu((*rhead
)->h_tail_lsn
));
1127 * Now that we have a tail block, check the head of the log for torn
1128 * writes. Search again until we hit the tail or the maximum number of
1129 * log record I/Os that could have been in flight at one time. Use a
1130 * temporary buffer so we don't trash the rhead/bp pointer from the
1133 tmp_bp
= xlog_get_bp(log
, 1);
1136 error
= xlog_rseek_logrec_hdr(log
, *head_blk
, *tail_blk
,
1137 XLOG_MAX_ICLOGS
, tmp_bp
, &tmp_rhead_blk
,
1138 &tmp_rhead
, &tmp_wrapped
);
1139 xlog_put_bp(tmp_bp
);
1144 * Now run a CRC verification pass over the records starting at the
1145 * block found above to the current head. If a CRC failure occurs, the
1146 * log block of the first bad record is saved in first_bad.
1148 error
= xlog_do_recovery_pass(log
, *head_blk
, tmp_rhead_blk
,
1149 XLOG_RECOVER_CRCPASS
, &first_bad
);
1150 if (error
== -EFSBADCRC
) {
1152 * We've hit a potential torn write. Reset the error and warn
1157 "Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.",
1158 first_bad
, *head_blk
);
1161 * Get the header block and buffer pointer for the last good
1162 * record before the bad record.
1164 * Note that xlog_find_tail() clears the blocks at the new head
1165 * (i.e., the records with invalid CRC) if the cycle number
1166 * matches the the current cycle.
1168 found
= xlog_rseek_logrec_hdr(log
, first_bad
, *tail_blk
, 1, bp
,
1169 rhead_blk
, rhead
, wrapped
);
1172 if (found
== 0) /* XXX: right thing to do here? */
1176 * Reset the head block to the starting block of the first bad
1177 * log record and set the tail block based on the last good
1180 * Bail out if the updated head/tail match as this indicates
1181 * possible corruption outside of the acceptable
1182 * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
1184 *head_blk
= first_bad
;
1185 *tail_blk
= BLOCK_LSN(be64_to_cpu((*rhead
)->h_tail_lsn
));
1186 if (*head_blk
== *tail_blk
) {
1192 * Now verify the tail based on the updated head. This is
1193 * required because the torn writes trimmed from the head could
1194 * have been written over the tail of a previous record. Return
1195 * any errors since recovery cannot proceed if the tail is
1198 * XXX: This leaves a gap in truly robust protection from torn
1199 * writes in the log. If the head is behind the tail, the tail
1200 * pushes forward to create some space and then a crash occurs
1201 * causing the writes into the previous record's tail region to
1202 * tear, log recovery isn't able to recover.
1204 * How likely is this to occur? If possible, can we do something
1205 * more intelligent here? Is it safe to push the tail forward if
1206 * we can determine that the tail is within the range of the
1207 * torn write (e.g., the kernel can only overwrite the tail if
1208 * it has actually been pushed forward)? Alternatively, could we
1209 * somehow prevent this condition at runtime?
1211 error
= xlog_verify_tail(log
, *head_blk
, *tail_blk
);
1218 * Find the sync block number or the tail of the log.
1220 * This will be the block number of the last record to have its
1221 * associated buffers synced to disk. Every log record header has
1222 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
1223 * to get a sync block number. The only concern is to figure out which
1224 * log record header to believe.
1226 * The following algorithm uses the log record header with the largest
1227 * lsn. The entire log record does not need to be valid. We only care
1228 * that the header is valid.
1230 * We could speed up search by using current head_blk buffer, but it is not
1236 xfs_daddr_t
*head_blk
,
1237 xfs_daddr_t
*tail_blk
)
1239 xlog_rec_header_t
*rhead
;
1240 xlog_op_header_t
*op_head
;
1241 char *offset
= NULL
;
1244 xfs_daddr_t umount_data_blk
;
1245 xfs_daddr_t after_umount_blk
;
1246 xfs_daddr_t rhead_blk
;
1249 bool wrapped
= false;
1252 * Find previous log record
1254 if ((error
= xlog_find_head(log
, head_blk
)))
1257 bp
= xlog_get_bp(log
, 1);
1260 if (*head_blk
== 0) { /* special case */
1261 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1265 if (xlog_get_cycle(offset
) == 0) {
1267 /* leave all other log inited values alone */
1273 * Trim the head block back to skip over torn records. We can have
1274 * multiple log I/Os in flight at any time, so we assume CRC failures
1275 * back through the previous several records are torn writes and skip
1278 ASSERT(*head_blk
< INT_MAX
);
1279 error
= xlog_verify_head(log
, head_blk
, tail_blk
, bp
, &rhead_blk
,
1285 * Reset log values according to the state of the log when we
1286 * crashed. In the case where head_blk == 0, we bump curr_cycle
1287 * one because the next write starts a new cycle rather than
1288 * continuing the cycle of the last good log record. At this
1289 * point we have guaranteed that all partial log records have been
1290 * accounted for. Therefore, we know that the last good log record
1291 * written was complete and ended exactly on the end boundary
1292 * of the physical log.
1294 log
->l_prev_block
= rhead_blk
;
1295 log
->l_curr_block
= (int)*head_blk
;
1296 log
->l_curr_cycle
= be32_to_cpu(rhead
->h_cycle
);
1298 log
->l_curr_cycle
++;
1299 atomic64_set(&log
->l_tail_lsn
, be64_to_cpu(rhead
->h_tail_lsn
));
1300 atomic64_set(&log
->l_last_sync_lsn
, be64_to_cpu(rhead
->h_lsn
));
1301 xlog_assign_grant_head(&log
->l_reserve_head
.grant
, log
->l_curr_cycle
,
1302 BBTOB(log
->l_curr_block
));
1303 xlog_assign_grant_head(&log
->l_write_head
.grant
, log
->l_curr_cycle
,
1304 BBTOB(log
->l_curr_block
));
1307 * Look for unmount record. If we find it, then we know there
1308 * was a clean unmount. Since 'i' could be the last block in
1309 * the physical log, we convert to a log block before comparing
1312 * Save the current tail lsn to use to pass to
1313 * xlog_clear_stale_blocks() below. We won't want to clear the
1314 * unmount record if there is one, so we pass the lsn of the
1315 * unmount record rather than the block after it.
1317 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
1318 int h_size
= be32_to_cpu(rhead
->h_size
);
1319 int h_version
= be32_to_cpu(rhead
->h_version
);
1321 if ((h_version
& XLOG_VERSION_2
) &&
1322 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
1323 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
1324 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
1332 after_umount_blk
= rhead_blk
+ hblks
+ BTOBB(be32_to_cpu(rhead
->h_len
));
1333 after_umount_blk
= do_mod(after_umount_blk
, log
->l_logBBsize
);
1334 tail_lsn
= atomic64_read(&log
->l_tail_lsn
);
1335 if (*head_blk
== after_umount_blk
&&
1336 be32_to_cpu(rhead
->h_num_logops
) == 1) {
1337 umount_data_blk
= rhead_blk
+ hblks
;
1338 umount_data_blk
= do_mod(umount_data_blk
, log
->l_logBBsize
);
1339 error
= xlog_bread(log
, umount_data_blk
, 1, bp
, &offset
);
1343 op_head
= (xlog_op_header_t
*)offset
;
1344 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
1346 * Set tail and last sync so that newly written
1347 * log records will point recovery to after the
1348 * current unmount record.
1350 xlog_assign_atomic_lsn(&log
->l_tail_lsn
,
1351 log
->l_curr_cycle
, after_umount_blk
);
1352 xlog_assign_atomic_lsn(&log
->l_last_sync_lsn
,
1353 log
->l_curr_cycle
, after_umount_blk
);
1354 *tail_blk
= after_umount_blk
;
1357 * Note that the unmount was clean. If the unmount
1358 * was not clean, we need to know this to rebuild the
1359 * superblock counters from the perag headers if we
1360 * have a filesystem using non-persistent counters.
1362 log
->l_mp
->m_flags
|= XFS_MOUNT_WAS_CLEAN
;
1367 * Make sure that there are no blocks in front of the head
1368 * with the same cycle number as the head. This can happen
1369 * because we allow multiple outstanding log writes concurrently,
1370 * and the later writes might make it out before earlier ones.
1372 * We use the lsn from before modifying it so that we'll never
1373 * overwrite the unmount record after a clean unmount.
1375 * Do this only if we are going to recover the filesystem
1377 * NOTE: This used to say "if (!readonly)"
1378 * However on Linux, we can & do recover a read-only filesystem.
1379 * We only skip recovery if NORECOVERY is specified on mount,
1380 * in which case we would not be here.
1382 * But... if the -device- itself is readonly, just skip this.
1383 * We can't recover this device anyway, so it won't matter.
1385 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
))
1386 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
1392 xfs_warn(log
->l_mp
, "failed to locate log tail");
1397 * Is the log zeroed at all?
1399 * The last binary search should be changed to perform an X block read
1400 * once X becomes small enough. You can then search linearly through
1401 * the X blocks. This will cut down on the number of reads we need to do.
1403 * If the log is partially zeroed, this routine will pass back the blkno
1404 * of the first block with cycle number 0. It won't have a complete LR
1408 * 0 => the log is completely written to
1409 * 1 => use *blk_no as the first block of the log
1410 * <0 => error has occurred
1415 xfs_daddr_t
*blk_no
)
1419 uint first_cycle
, last_cycle
;
1420 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1421 xfs_daddr_t num_scan_bblks
;
1422 int error
, log_bbnum
= log
->l_logBBsize
;
1426 /* check totally zeroed log */
1427 bp
= xlog_get_bp(log
, 1);
1430 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1434 first_cycle
= xlog_get_cycle(offset
);
1435 if (first_cycle
== 0) { /* completely zeroed log */
1441 /* check partially zeroed log */
1442 error
= xlog_bread(log
, log_bbnum
-1, 1, bp
, &offset
);
1446 last_cycle
= xlog_get_cycle(offset
);
1447 if (last_cycle
!= 0) { /* log completely written to */
1450 } else if (first_cycle
!= 1) {
1452 * If the cycle of the last block is zero, the cycle of
1453 * the first block must be 1. If it's not, maybe we're
1454 * not looking at a log... Bail out.
1457 "Log inconsistent or not a log (last==0, first!=1)");
1462 /* we have a partially zeroed log */
1463 last_blk
= log_bbnum
-1;
1464 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1468 * Validate the answer. Because there is no way to guarantee that
1469 * the entire log is made up of log records which are the same size,
1470 * we scan over the defined maximum blocks. At this point, the maximum
1471 * is not chosen to mean anything special. XXXmiken
1473 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1474 ASSERT(num_scan_bblks
<= INT_MAX
);
1476 if (last_blk
< num_scan_bblks
)
1477 num_scan_bblks
= last_blk
;
1478 start_blk
= last_blk
- num_scan_bblks
;
1481 * We search for any instances of cycle number 0 that occur before
1482 * our current estimate of the head. What we're trying to detect is
1483 * 1 ... | 0 | 1 | 0...
1484 * ^ binary search ends here
1486 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1487 (int)num_scan_bblks
, 0, &new_blk
)))
1493 * Potentially backup over partial log record write. We don't need
1494 * to search the end of the log because we know it is zero.
1496 error
= xlog_find_verify_log_record(log
, start_blk
, &last_blk
, 0);
1511 * These are simple subroutines used by xlog_clear_stale_blocks() below
1512 * to initialize a buffer full of empty log record headers and write
1513 * them into the log.
1524 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1526 memset(buf
, 0, BBSIZE
);
1527 recp
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1528 recp
->h_cycle
= cpu_to_be32(cycle
);
1529 recp
->h_version
= cpu_to_be32(
1530 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1531 recp
->h_lsn
= cpu_to_be64(xlog_assign_lsn(cycle
, block
));
1532 recp
->h_tail_lsn
= cpu_to_be64(xlog_assign_lsn(tail_cycle
, tail_block
));
1533 recp
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1534 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1538 xlog_write_log_records(
1549 int sectbb
= log
->l_sectBBsize
;
1550 int end_block
= start_block
+ blocks
;
1556 * Greedily allocate a buffer big enough to handle the full
1557 * range of basic blocks to be written. If that fails, try
1558 * a smaller size. We need to be able to write at least a
1559 * log sector, or we're out of luck.
1561 bufblks
= 1 << ffs(blocks
);
1562 while (bufblks
> log
->l_logBBsize
)
1564 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1566 if (bufblks
< sectbb
)
1570 /* We may need to do a read at the start to fill in part of
1571 * the buffer in the starting sector not covered by the first
1574 balign
= round_down(start_block
, sectbb
);
1575 if (balign
!= start_block
) {
1576 error
= xlog_bread_noalign(log
, start_block
, 1, bp
);
1580 j
= start_block
- balign
;
1583 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1584 int bcount
, endcount
;
1586 bcount
= min(bufblks
, end_block
- start_block
);
1587 endcount
= bcount
- j
;
1589 /* We may need to do a read at the end to fill in part of
1590 * the buffer in the final sector not covered by the write.
1591 * If this is the same sector as the above read, skip it.
1593 ealign
= round_down(end_block
, sectbb
);
1594 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1595 offset
= bp
->b_addr
+ BBTOB(ealign
- start_block
);
1596 error
= xlog_bread_offset(log
, ealign
, sectbb
,
1603 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1604 for (; j
< endcount
; j
++) {
1605 xlog_add_record(log
, offset
, cycle
, i
+j
,
1606 tail_cycle
, tail_block
);
1609 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1612 start_block
+= endcount
;
1622 * This routine is called to blow away any incomplete log writes out
1623 * in front of the log head. We do this so that we won't become confused
1624 * if we come up, write only a little bit more, and then crash again.
1625 * If we leave the partial log records out there, this situation could
1626 * cause us to think those partial writes are valid blocks since they
1627 * have the current cycle number. We get rid of them by overwriting them
1628 * with empty log records with the old cycle number rather than the
1631 * The tail lsn is passed in rather than taken from
1632 * the log so that we will not write over the unmount record after a
1633 * clean unmount in a 512 block log. Doing so would leave the log without
1634 * any valid log records in it until a new one was written. If we crashed
1635 * during that time we would not be able to recover.
1638 xlog_clear_stale_blocks(
1642 int tail_cycle
, head_cycle
;
1643 int tail_block
, head_block
;
1644 int tail_distance
, max_distance
;
1648 tail_cycle
= CYCLE_LSN(tail_lsn
);
1649 tail_block
= BLOCK_LSN(tail_lsn
);
1650 head_cycle
= log
->l_curr_cycle
;
1651 head_block
= log
->l_curr_block
;
1654 * Figure out the distance between the new head of the log
1655 * and the tail. We want to write over any blocks beyond the
1656 * head that we may have written just before the crash, but
1657 * we don't want to overwrite the tail of the log.
1659 if (head_cycle
== tail_cycle
) {
1661 * The tail is behind the head in the physical log,
1662 * so the distance from the head to the tail is the
1663 * distance from the head to the end of the log plus
1664 * the distance from the beginning of the log to the
1667 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1668 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1669 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1670 return -EFSCORRUPTED
;
1672 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1675 * The head is behind the tail in the physical log,
1676 * so the distance from the head to the tail is just
1677 * the tail block minus the head block.
1679 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1680 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1681 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1682 return -EFSCORRUPTED
;
1684 tail_distance
= tail_block
- head_block
;
1688 * If the head is right up against the tail, we can't clear
1691 if (tail_distance
<= 0) {
1692 ASSERT(tail_distance
== 0);
1696 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1698 * Take the smaller of the maximum amount of outstanding I/O
1699 * we could have and the distance to the tail to clear out.
1700 * We take the smaller so that we don't overwrite the tail and
1701 * we don't waste all day writing from the head to the tail
1704 max_distance
= MIN(max_distance
, tail_distance
);
1706 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1708 * We can stomp all the blocks we need to without
1709 * wrapping around the end of the log. Just do it
1710 * in a single write. Use the cycle number of the
1711 * current cycle minus one so that the log will look like:
1714 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1715 head_block
, max_distance
, tail_cycle
,
1721 * We need to wrap around the end of the physical log in
1722 * order to clear all the blocks. Do it in two separate
1723 * I/Os. The first write should be from the head to the
1724 * end of the physical log, and it should use the current
1725 * cycle number minus one just like above.
1727 distance
= log
->l_logBBsize
- head_block
;
1728 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1729 head_block
, distance
, tail_cycle
,
1736 * Now write the blocks at the start of the physical log.
1737 * This writes the remainder of the blocks we want to clear.
1738 * It uses the current cycle number since we're now on the
1739 * same cycle as the head so that we get:
1740 * n ... n ... | n - 1 ...
1741 * ^^^^^ blocks we're writing
1743 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1744 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1745 tail_cycle
, tail_block
);
1753 /******************************************************************************
1755 * Log recover routines
1757 ******************************************************************************
1761 * Sort the log items in the transaction.
1763 * The ordering constraints are defined by the inode allocation and unlink
1764 * behaviour. The rules are:
1766 * 1. Every item is only logged once in a given transaction. Hence it
1767 * represents the last logged state of the item. Hence ordering is
1768 * dependent on the order in which operations need to be performed so
1769 * required initial conditions are always met.
1771 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1772 * there's nothing to replay from them so we can simply cull them
1773 * from the transaction. However, we can't do that until after we've
1774 * replayed all the other items because they may be dependent on the
1775 * cancelled buffer and replaying the cancelled buffer can remove it
1776 * form the cancelled buffer table. Hence they have tobe done last.
1778 * 3. Inode allocation buffers must be replayed before inode items that
1779 * read the buffer and replay changes into it. For filesystems using the
1780 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1781 * treated the same as inode allocation buffers as they create and
1782 * initialise the buffers directly.
1784 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1785 * This ensures that inodes are completely flushed to the inode buffer
1786 * in a "free" state before we remove the unlinked inode list pointer.
1788 * Hence the ordering needs to be inode allocation buffers first, inode items
1789 * second, inode unlink buffers third and cancelled buffers last.
1791 * But there's a problem with that - we can't tell an inode allocation buffer
1792 * apart from a regular buffer, so we can't separate them. We can, however,
1793 * tell an inode unlink buffer from the others, and so we can separate them out
1794 * from all the other buffers and move them to last.
1796 * Hence, 4 lists, in order from head to tail:
1797 * - buffer_list for all buffers except cancelled/inode unlink buffers
1798 * - item_list for all non-buffer items
1799 * - inode_buffer_list for inode unlink buffers
1800 * - cancel_list for the cancelled buffers
1802 * Note that we add objects to the tail of the lists so that first-to-last
1803 * ordering is preserved within the lists. Adding objects to the head of the
1804 * list means when we traverse from the head we walk them in last-to-first
1805 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1806 * but for all other items there may be specific ordering that we need to
1810 xlog_recover_reorder_trans(
1812 struct xlog_recover
*trans
,
1815 xlog_recover_item_t
*item
, *n
;
1817 LIST_HEAD(sort_list
);
1818 LIST_HEAD(cancel_list
);
1819 LIST_HEAD(buffer_list
);
1820 LIST_HEAD(inode_buffer_list
);
1821 LIST_HEAD(inode_list
);
1823 list_splice_init(&trans
->r_itemq
, &sort_list
);
1824 list_for_each_entry_safe(item
, n
, &sort_list
, ri_list
) {
1825 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1827 switch (ITEM_TYPE(item
)) {
1828 case XFS_LI_ICREATE
:
1829 list_move_tail(&item
->ri_list
, &buffer_list
);
1832 if (buf_f
->blf_flags
& XFS_BLF_CANCEL
) {
1833 trace_xfs_log_recover_item_reorder_head(log
,
1835 list_move(&item
->ri_list
, &cancel_list
);
1838 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
1839 list_move(&item
->ri_list
, &inode_buffer_list
);
1842 list_move_tail(&item
->ri_list
, &buffer_list
);
1846 case XFS_LI_QUOTAOFF
:
1849 trace_xfs_log_recover_item_reorder_tail(log
,
1851 list_move_tail(&item
->ri_list
, &inode_list
);
1855 "%s: unrecognized type of log operation",
1859 * return the remaining items back to the transaction
1860 * item list so they can be freed in caller.
1862 if (!list_empty(&sort_list
))
1863 list_splice_init(&sort_list
, &trans
->r_itemq
);
1869 ASSERT(list_empty(&sort_list
));
1870 if (!list_empty(&buffer_list
))
1871 list_splice(&buffer_list
, &trans
->r_itemq
);
1872 if (!list_empty(&inode_list
))
1873 list_splice_tail(&inode_list
, &trans
->r_itemq
);
1874 if (!list_empty(&inode_buffer_list
))
1875 list_splice_tail(&inode_buffer_list
, &trans
->r_itemq
);
1876 if (!list_empty(&cancel_list
))
1877 list_splice_tail(&cancel_list
, &trans
->r_itemq
);
1882 * Build up the table of buf cancel records so that we don't replay
1883 * cancelled data in the second pass. For buffer records that are
1884 * not cancel records, there is nothing to do here so we just return.
1886 * If we get a cancel record which is already in the table, this indicates
1887 * that the buffer was cancelled multiple times. In order to ensure
1888 * that during pass 2 we keep the record in the table until we reach its
1889 * last occurrence in the log, we keep a reference count in the cancel
1890 * record in the table to tell us how many times we expect to see this
1891 * record during the second pass.
1894 xlog_recover_buffer_pass1(
1896 struct xlog_recover_item
*item
)
1898 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1899 struct list_head
*bucket
;
1900 struct xfs_buf_cancel
*bcp
;
1903 * If this isn't a cancel buffer item, then just return.
1905 if (!(buf_f
->blf_flags
& XFS_BLF_CANCEL
)) {
1906 trace_xfs_log_recover_buf_not_cancel(log
, buf_f
);
1911 * Insert an xfs_buf_cancel record into the hash table of them.
1912 * If there is already an identical record, bump its reference count.
1914 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, buf_f
->blf_blkno
);
1915 list_for_each_entry(bcp
, bucket
, bc_list
) {
1916 if (bcp
->bc_blkno
== buf_f
->blf_blkno
&&
1917 bcp
->bc_len
== buf_f
->blf_len
) {
1919 trace_xfs_log_recover_buf_cancel_ref_inc(log
, buf_f
);
1924 bcp
= kmem_alloc(sizeof(struct xfs_buf_cancel
), KM_SLEEP
);
1925 bcp
->bc_blkno
= buf_f
->blf_blkno
;
1926 bcp
->bc_len
= buf_f
->blf_len
;
1927 bcp
->bc_refcount
= 1;
1928 list_add_tail(&bcp
->bc_list
, bucket
);
1930 trace_xfs_log_recover_buf_cancel_add(log
, buf_f
);
1935 * Check to see whether the buffer being recovered has a corresponding
1936 * entry in the buffer cancel record table. If it is, return the cancel
1937 * buffer structure to the caller.
1939 STATIC
struct xfs_buf_cancel
*
1940 xlog_peek_buffer_cancelled(
1946 struct list_head
*bucket
;
1947 struct xfs_buf_cancel
*bcp
;
1949 if (!log
->l_buf_cancel_table
) {
1950 /* empty table means no cancelled buffers in the log */
1951 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1955 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, blkno
);
1956 list_for_each_entry(bcp
, bucket
, bc_list
) {
1957 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
)
1962 * We didn't find a corresponding entry in the table, so return 0 so
1963 * that the buffer is NOT cancelled.
1965 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1970 * If the buffer is being cancelled then return 1 so that it will be cancelled,
1971 * otherwise return 0. If the buffer is actually a buffer cancel item
1972 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
1973 * table and remove it from the table if this is the last reference.
1975 * We remove the cancel record from the table when we encounter its last
1976 * occurrence in the log so that if the same buffer is re-used again after its
1977 * last cancellation we actually replay the changes made at that point.
1980 xlog_check_buffer_cancelled(
1986 struct xfs_buf_cancel
*bcp
;
1988 bcp
= xlog_peek_buffer_cancelled(log
, blkno
, len
, flags
);
1993 * We've go a match, so return 1 so that the recovery of this buffer
1994 * is cancelled. If this buffer is actually a buffer cancel log
1995 * item, then decrement the refcount on the one in the table and
1996 * remove it if this is the last reference.
1998 if (flags
& XFS_BLF_CANCEL
) {
1999 if (--bcp
->bc_refcount
== 0) {
2000 list_del(&bcp
->bc_list
);
2008 * Perform recovery for a buffer full of inodes. In these buffers, the only
2009 * data which should be recovered is that which corresponds to the
2010 * di_next_unlinked pointers in the on disk inode structures. The rest of the
2011 * data for the inodes is always logged through the inodes themselves rather
2012 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
2014 * The only time when buffers full of inodes are fully recovered is when the
2015 * buffer is full of newly allocated inodes. In this case the buffer will
2016 * not be marked as an inode buffer and so will be sent to
2017 * xlog_recover_do_reg_buffer() below during recovery.
2020 xlog_recover_do_inode_buffer(
2021 struct xfs_mount
*mp
,
2022 xlog_recover_item_t
*item
,
2024 xfs_buf_log_format_t
*buf_f
)
2030 int reg_buf_offset
= 0;
2031 int reg_buf_bytes
= 0;
2032 int next_unlinked_offset
;
2034 xfs_agino_t
*logged_nextp
;
2035 xfs_agino_t
*buffer_nextp
;
2037 trace_xfs_log_recover_buf_inode_buf(mp
->m_log
, buf_f
);
2040 * Post recovery validation only works properly on CRC enabled
2043 if (xfs_sb_version_hascrc(&mp
->m_sb
))
2044 bp
->b_ops
= &xfs_inode_buf_ops
;
2046 inodes_per_buf
= BBTOB(bp
->b_io_length
) >> mp
->m_sb
.sb_inodelog
;
2047 for (i
= 0; i
< inodes_per_buf
; i
++) {
2048 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
2049 offsetof(xfs_dinode_t
, di_next_unlinked
);
2051 while (next_unlinked_offset
>=
2052 (reg_buf_offset
+ reg_buf_bytes
)) {
2054 * The next di_next_unlinked field is beyond
2055 * the current logged region. Find the next
2056 * logged region that contains or is beyond
2057 * the current di_next_unlinked field.
2060 bit
= xfs_next_bit(buf_f
->blf_data_map
,
2061 buf_f
->blf_map_size
, bit
);
2064 * If there are no more logged regions in the
2065 * buffer, then we're done.
2070 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
2071 buf_f
->blf_map_size
, bit
);
2073 reg_buf_offset
= bit
<< XFS_BLF_SHIFT
;
2074 reg_buf_bytes
= nbits
<< XFS_BLF_SHIFT
;
2079 * If the current logged region starts after the current
2080 * di_next_unlinked field, then move on to the next
2081 * di_next_unlinked field.
2083 if (next_unlinked_offset
< reg_buf_offset
)
2086 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
2087 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLF_CHUNK
) == 0);
2088 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <=
2089 BBTOB(bp
->b_io_length
));
2092 * The current logged region contains a copy of the
2093 * current di_next_unlinked field. Extract its value
2094 * and copy it to the buffer copy.
2096 logged_nextp
= item
->ri_buf
[item_index
].i_addr
+
2097 next_unlinked_offset
- reg_buf_offset
;
2098 if (unlikely(*logged_nextp
== 0)) {
2100 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
2101 "Trying to replay bad (0) inode di_next_unlinked field.",
2103 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
2104 XFS_ERRLEVEL_LOW
, mp
);
2105 return -EFSCORRUPTED
;
2108 buffer_nextp
= xfs_buf_offset(bp
, next_unlinked_offset
);
2109 *buffer_nextp
= *logged_nextp
;
2112 * If necessary, recalculate the CRC in the on-disk inode. We
2113 * have to leave the inode in a consistent state for whoever
2116 xfs_dinode_calc_crc(mp
,
2117 xfs_buf_offset(bp
, i
* mp
->m_sb
.sb_inodesize
));
2125 * V5 filesystems know the age of the buffer on disk being recovered. We can
2126 * have newer objects on disk than we are replaying, and so for these cases we
2127 * don't want to replay the current change as that will make the buffer contents
2128 * temporarily invalid on disk.
2130 * The magic number might not match the buffer type we are going to recover
2131 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
2132 * extract the LSN of the existing object in the buffer based on it's current
2133 * magic number. If we don't recognise the magic number in the buffer, then
2134 * return a LSN of -1 so that the caller knows it was an unrecognised block and
2135 * so can recover the buffer.
2137 * Note: we cannot rely solely on magic number matches to determine that the
2138 * buffer has a valid LSN - we also need to verify that it belongs to this
2139 * filesystem, so we need to extract the object's LSN and compare it to that
2140 * which we read from the superblock. If the UUIDs don't match, then we've got a
2141 * stale metadata block from an old filesystem instance that we need to recover
2145 xlog_recover_get_buf_lsn(
2146 struct xfs_mount
*mp
,
2152 void *blk
= bp
->b_addr
;
2156 /* v4 filesystems always recover immediately */
2157 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2158 goto recover_immediately
;
2160 magic32
= be32_to_cpu(*(__be32
*)blk
);
2162 case XFS_ABTB_CRC_MAGIC
:
2163 case XFS_ABTC_CRC_MAGIC
:
2164 case XFS_ABTB_MAGIC
:
2165 case XFS_ABTC_MAGIC
:
2166 case XFS_IBT_CRC_MAGIC
:
2167 case XFS_IBT_MAGIC
: {
2168 struct xfs_btree_block
*btb
= blk
;
2170 lsn
= be64_to_cpu(btb
->bb_u
.s
.bb_lsn
);
2171 uuid
= &btb
->bb_u
.s
.bb_uuid
;
2174 case XFS_BMAP_CRC_MAGIC
:
2175 case XFS_BMAP_MAGIC
: {
2176 struct xfs_btree_block
*btb
= blk
;
2178 lsn
= be64_to_cpu(btb
->bb_u
.l
.bb_lsn
);
2179 uuid
= &btb
->bb_u
.l
.bb_uuid
;
2183 lsn
= be64_to_cpu(((struct xfs_agf
*)blk
)->agf_lsn
);
2184 uuid
= &((struct xfs_agf
*)blk
)->agf_uuid
;
2186 case XFS_AGFL_MAGIC
:
2187 lsn
= be64_to_cpu(((struct xfs_agfl
*)blk
)->agfl_lsn
);
2188 uuid
= &((struct xfs_agfl
*)blk
)->agfl_uuid
;
2191 lsn
= be64_to_cpu(((struct xfs_agi
*)blk
)->agi_lsn
);
2192 uuid
= &((struct xfs_agi
*)blk
)->agi_uuid
;
2194 case XFS_SYMLINK_MAGIC
:
2195 lsn
= be64_to_cpu(((struct xfs_dsymlink_hdr
*)blk
)->sl_lsn
);
2196 uuid
= &((struct xfs_dsymlink_hdr
*)blk
)->sl_uuid
;
2198 case XFS_DIR3_BLOCK_MAGIC
:
2199 case XFS_DIR3_DATA_MAGIC
:
2200 case XFS_DIR3_FREE_MAGIC
:
2201 lsn
= be64_to_cpu(((struct xfs_dir3_blk_hdr
*)blk
)->lsn
);
2202 uuid
= &((struct xfs_dir3_blk_hdr
*)blk
)->uuid
;
2204 case XFS_ATTR3_RMT_MAGIC
:
2206 * Remote attr blocks are written synchronously, rather than
2207 * being logged. That means they do not contain a valid LSN
2208 * (i.e. transactionally ordered) in them, and hence any time we
2209 * see a buffer to replay over the top of a remote attribute
2210 * block we should simply do so.
2212 goto recover_immediately
;
2215 * superblock uuids are magic. We may or may not have a
2216 * sb_meta_uuid on disk, but it will be set in the in-core
2217 * superblock. We set the uuid pointer for verification
2218 * according to the superblock feature mask to ensure we check
2219 * the relevant UUID in the superblock.
2221 lsn
= be64_to_cpu(((struct xfs_dsb
*)blk
)->sb_lsn
);
2222 if (xfs_sb_version_hasmetauuid(&mp
->m_sb
))
2223 uuid
= &((struct xfs_dsb
*)blk
)->sb_meta_uuid
;
2225 uuid
= &((struct xfs_dsb
*)blk
)->sb_uuid
;
2231 if (lsn
!= (xfs_lsn_t
)-1) {
2232 if (!uuid_equal(&mp
->m_sb
.sb_meta_uuid
, uuid
))
2233 goto recover_immediately
;
2237 magicda
= be16_to_cpu(((struct xfs_da_blkinfo
*)blk
)->magic
);
2239 case XFS_DIR3_LEAF1_MAGIC
:
2240 case XFS_DIR3_LEAFN_MAGIC
:
2241 case XFS_DA3_NODE_MAGIC
:
2242 lsn
= be64_to_cpu(((struct xfs_da3_blkinfo
*)blk
)->lsn
);
2243 uuid
= &((struct xfs_da3_blkinfo
*)blk
)->uuid
;
2249 if (lsn
!= (xfs_lsn_t
)-1) {
2250 if (!uuid_equal(&mp
->m_sb
.sb_uuid
, uuid
))
2251 goto recover_immediately
;
2256 * We do individual object checks on dquot and inode buffers as they
2257 * have their own individual LSN records. Also, we could have a stale
2258 * buffer here, so we have to at least recognise these buffer types.
2260 * A notd complexity here is inode unlinked list processing - it logs
2261 * the inode directly in the buffer, but we don't know which inodes have
2262 * been modified, and there is no global buffer LSN. Hence we need to
2263 * recover all inode buffer types immediately. This problem will be
2264 * fixed by logical logging of the unlinked list modifications.
2266 magic16
= be16_to_cpu(*(__be16
*)blk
);
2268 case XFS_DQUOT_MAGIC
:
2269 case XFS_DINODE_MAGIC
:
2270 goto recover_immediately
;
2275 /* unknown buffer contents, recover immediately */
2277 recover_immediately
:
2278 return (xfs_lsn_t
)-1;
2283 * Validate the recovered buffer is of the correct type and attach the
2284 * appropriate buffer operations to them for writeback. Magic numbers are in a
2286 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2287 * the first 32 bits of the buffer (most blocks),
2288 * inside a struct xfs_da_blkinfo at the start of the buffer.
2291 xlog_recover_validate_buf_type(
2292 struct xfs_mount
*mp
,
2294 xfs_buf_log_format_t
*buf_f
)
2296 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
2302 * We can only do post recovery validation on items on CRC enabled
2303 * fielsystems as we need to know when the buffer was written to be able
2304 * to determine if we should have replayed the item. If we replay old
2305 * metadata over a newer buffer, then it will enter a temporarily
2306 * inconsistent state resulting in verification failures. Hence for now
2307 * just avoid the verification stage for non-crc filesystems
2309 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2312 magic32
= be32_to_cpu(*(__be32
*)bp
->b_addr
);
2313 magic16
= be16_to_cpu(*(__be16
*)bp
->b_addr
);
2314 magicda
= be16_to_cpu(info
->magic
);
2315 switch (xfs_blft_from_flags(buf_f
)) {
2316 case XFS_BLFT_BTREE_BUF
:
2318 case XFS_ABTB_CRC_MAGIC
:
2319 case XFS_ABTC_CRC_MAGIC
:
2320 case XFS_ABTB_MAGIC
:
2321 case XFS_ABTC_MAGIC
:
2322 bp
->b_ops
= &xfs_allocbt_buf_ops
;
2324 case XFS_IBT_CRC_MAGIC
:
2325 case XFS_FIBT_CRC_MAGIC
:
2327 case XFS_FIBT_MAGIC
:
2328 bp
->b_ops
= &xfs_inobt_buf_ops
;
2330 case XFS_BMAP_CRC_MAGIC
:
2331 case XFS_BMAP_MAGIC
:
2332 bp
->b_ops
= &xfs_bmbt_buf_ops
;
2335 xfs_warn(mp
, "Bad btree block magic!");
2340 case XFS_BLFT_AGF_BUF
:
2341 if (magic32
!= XFS_AGF_MAGIC
) {
2342 xfs_warn(mp
, "Bad AGF block magic!");
2346 bp
->b_ops
= &xfs_agf_buf_ops
;
2348 case XFS_BLFT_AGFL_BUF
:
2349 if (magic32
!= XFS_AGFL_MAGIC
) {
2350 xfs_warn(mp
, "Bad AGFL block magic!");
2354 bp
->b_ops
= &xfs_agfl_buf_ops
;
2356 case XFS_BLFT_AGI_BUF
:
2357 if (magic32
!= XFS_AGI_MAGIC
) {
2358 xfs_warn(mp
, "Bad AGI block magic!");
2362 bp
->b_ops
= &xfs_agi_buf_ops
;
2364 case XFS_BLFT_UDQUOT_BUF
:
2365 case XFS_BLFT_PDQUOT_BUF
:
2366 case XFS_BLFT_GDQUOT_BUF
:
2367 #ifdef CONFIG_XFS_QUOTA
2368 if (magic16
!= XFS_DQUOT_MAGIC
) {
2369 xfs_warn(mp
, "Bad DQUOT block magic!");
2373 bp
->b_ops
= &xfs_dquot_buf_ops
;
2376 "Trying to recover dquots without QUOTA support built in!");
2380 case XFS_BLFT_DINO_BUF
:
2381 if (magic16
!= XFS_DINODE_MAGIC
) {
2382 xfs_warn(mp
, "Bad INODE block magic!");
2386 bp
->b_ops
= &xfs_inode_buf_ops
;
2388 case XFS_BLFT_SYMLINK_BUF
:
2389 if (magic32
!= XFS_SYMLINK_MAGIC
) {
2390 xfs_warn(mp
, "Bad symlink block magic!");
2394 bp
->b_ops
= &xfs_symlink_buf_ops
;
2396 case XFS_BLFT_DIR_BLOCK_BUF
:
2397 if (magic32
!= XFS_DIR2_BLOCK_MAGIC
&&
2398 magic32
!= XFS_DIR3_BLOCK_MAGIC
) {
2399 xfs_warn(mp
, "Bad dir block magic!");
2403 bp
->b_ops
= &xfs_dir3_block_buf_ops
;
2405 case XFS_BLFT_DIR_DATA_BUF
:
2406 if (magic32
!= XFS_DIR2_DATA_MAGIC
&&
2407 magic32
!= XFS_DIR3_DATA_MAGIC
) {
2408 xfs_warn(mp
, "Bad dir data magic!");
2412 bp
->b_ops
= &xfs_dir3_data_buf_ops
;
2414 case XFS_BLFT_DIR_FREE_BUF
:
2415 if (magic32
!= XFS_DIR2_FREE_MAGIC
&&
2416 magic32
!= XFS_DIR3_FREE_MAGIC
) {
2417 xfs_warn(mp
, "Bad dir3 free magic!");
2421 bp
->b_ops
= &xfs_dir3_free_buf_ops
;
2423 case XFS_BLFT_DIR_LEAF1_BUF
:
2424 if (magicda
!= XFS_DIR2_LEAF1_MAGIC
&&
2425 magicda
!= XFS_DIR3_LEAF1_MAGIC
) {
2426 xfs_warn(mp
, "Bad dir leaf1 magic!");
2430 bp
->b_ops
= &xfs_dir3_leaf1_buf_ops
;
2432 case XFS_BLFT_DIR_LEAFN_BUF
:
2433 if (magicda
!= XFS_DIR2_LEAFN_MAGIC
&&
2434 magicda
!= XFS_DIR3_LEAFN_MAGIC
) {
2435 xfs_warn(mp
, "Bad dir leafn magic!");
2439 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
2441 case XFS_BLFT_DA_NODE_BUF
:
2442 if (magicda
!= XFS_DA_NODE_MAGIC
&&
2443 magicda
!= XFS_DA3_NODE_MAGIC
) {
2444 xfs_warn(mp
, "Bad da node magic!");
2448 bp
->b_ops
= &xfs_da3_node_buf_ops
;
2450 case XFS_BLFT_ATTR_LEAF_BUF
:
2451 if (magicda
!= XFS_ATTR_LEAF_MAGIC
&&
2452 magicda
!= XFS_ATTR3_LEAF_MAGIC
) {
2453 xfs_warn(mp
, "Bad attr leaf magic!");
2457 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
2459 case XFS_BLFT_ATTR_RMT_BUF
:
2460 if (magic32
!= XFS_ATTR3_RMT_MAGIC
) {
2461 xfs_warn(mp
, "Bad attr remote magic!");
2465 bp
->b_ops
= &xfs_attr3_rmt_buf_ops
;
2467 case XFS_BLFT_SB_BUF
:
2468 if (magic32
!= XFS_SB_MAGIC
) {
2469 xfs_warn(mp
, "Bad SB block magic!");
2473 bp
->b_ops
= &xfs_sb_buf_ops
;
2476 xfs_warn(mp
, "Unknown buffer type %d!",
2477 xfs_blft_from_flags(buf_f
));
2483 * Perform a 'normal' buffer recovery. Each logged region of the
2484 * buffer should be copied over the corresponding region in the
2485 * given buffer. The bitmap in the buf log format structure indicates
2486 * where to place the logged data.
2489 xlog_recover_do_reg_buffer(
2490 struct xfs_mount
*mp
,
2491 xlog_recover_item_t
*item
,
2493 xfs_buf_log_format_t
*buf_f
)
2500 trace_xfs_log_recover_buf_reg_buf(mp
->m_log
, buf_f
);
2503 i
= 1; /* 0 is the buf format structure */
2505 bit
= xfs_next_bit(buf_f
->blf_data_map
,
2506 buf_f
->blf_map_size
, bit
);
2509 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
2510 buf_f
->blf_map_size
, bit
);
2512 ASSERT(item
->ri_buf
[i
].i_addr
!= NULL
);
2513 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLF_CHUNK
== 0);
2514 ASSERT(BBTOB(bp
->b_io_length
) >=
2515 ((uint
)bit
<< XFS_BLF_SHIFT
) + (nbits
<< XFS_BLF_SHIFT
));
2518 * The dirty regions logged in the buffer, even though
2519 * contiguous, may span multiple chunks. This is because the
2520 * dirty region may span a physical page boundary in a buffer
2521 * and hence be split into two separate vectors for writing into
2522 * the log. Hence we need to trim nbits back to the length of
2523 * the current region being copied out of the log.
2525 if (item
->ri_buf
[i
].i_len
< (nbits
<< XFS_BLF_SHIFT
))
2526 nbits
= item
->ri_buf
[i
].i_len
>> XFS_BLF_SHIFT
;
2529 * Do a sanity check if this is a dquot buffer. Just checking
2530 * the first dquot in the buffer should do. XXXThis is
2531 * probably a good thing to do for other buf types also.
2534 if (buf_f
->blf_flags
&
2535 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2536 if (item
->ri_buf
[i
].i_addr
== NULL
) {
2538 "XFS: NULL dquot in %s.", __func__
);
2541 if (item
->ri_buf
[i
].i_len
< sizeof(xfs_disk_dquot_t
)) {
2543 "XFS: dquot too small (%d) in %s.",
2544 item
->ri_buf
[i
].i_len
, __func__
);
2547 error
= xfs_dqcheck(mp
, item
->ri_buf
[i
].i_addr
,
2548 -1, 0, XFS_QMOPT_DOWARN
,
2549 "dquot_buf_recover");
2554 memcpy(xfs_buf_offset(bp
,
2555 (uint
)bit
<< XFS_BLF_SHIFT
), /* dest */
2556 item
->ri_buf
[i
].i_addr
, /* source */
2557 nbits
<<XFS_BLF_SHIFT
); /* length */
2563 /* Shouldn't be any more regions */
2564 ASSERT(i
== item
->ri_total
);
2566 xlog_recover_validate_buf_type(mp
, bp
, buf_f
);
2570 * Perform a dquot buffer recovery.
2571 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2572 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2573 * Else, treat it as a regular buffer and do recovery.
2575 * Return false if the buffer was tossed and true if we recovered the buffer to
2576 * indicate to the caller if the buffer needs writing.
2579 xlog_recover_do_dquot_buffer(
2580 struct xfs_mount
*mp
,
2582 struct xlog_recover_item
*item
,
2584 struct xfs_buf_log_format
*buf_f
)
2588 trace_xfs_log_recover_buf_dquot_buf(log
, buf_f
);
2591 * Filesystems are required to send in quota flags at mount time.
2597 if (buf_f
->blf_flags
& XFS_BLF_UDQUOT_BUF
)
2598 type
|= XFS_DQ_USER
;
2599 if (buf_f
->blf_flags
& XFS_BLF_PDQUOT_BUF
)
2600 type
|= XFS_DQ_PROJ
;
2601 if (buf_f
->blf_flags
& XFS_BLF_GDQUOT_BUF
)
2602 type
|= XFS_DQ_GROUP
;
2604 * This type of quotas was turned off, so ignore this buffer
2606 if (log
->l_quotaoffs_flag
& type
)
2609 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2614 * This routine replays a modification made to a buffer at runtime.
2615 * There are actually two types of buffer, regular and inode, which
2616 * are handled differently. Inode buffers are handled differently
2617 * in that we only recover a specific set of data from them, namely
2618 * the inode di_next_unlinked fields. This is because all other inode
2619 * data is actually logged via inode records and any data we replay
2620 * here which overlaps that may be stale.
2622 * When meta-data buffers are freed at run time we log a buffer item
2623 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2624 * of the buffer in the log should not be replayed at recovery time.
2625 * This is so that if the blocks covered by the buffer are reused for
2626 * file data before we crash we don't end up replaying old, freed
2627 * meta-data into a user's file.
2629 * To handle the cancellation of buffer log items, we make two passes
2630 * over the log during recovery. During the first we build a table of
2631 * those buffers which have been cancelled, and during the second we
2632 * only replay those buffers which do not have corresponding cancel
2633 * records in the table. See xlog_recover_buffer_pass[1,2] above
2634 * for more details on the implementation of the table of cancel records.
2637 xlog_recover_buffer_pass2(
2639 struct list_head
*buffer_list
,
2640 struct xlog_recover_item
*item
,
2641 xfs_lsn_t current_lsn
)
2643 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
2644 xfs_mount_t
*mp
= log
->l_mp
;
2651 * In this pass we only want to recover all the buffers which have
2652 * not been cancelled and are not cancellation buffers themselves.
2654 if (xlog_check_buffer_cancelled(log
, buf_f
->blf_blkno
,
2655 buf_f
->blf_len
, buf_f
->blf_flags
)) {
2656 trace_xfs_log_recover_buf_cancel(log
, buf_f
);
2660 trace_xfs_log_recover_buf_recover(log
, buf_f
);
2663 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
)
2664 buf_flags
|= XBF_UNMAPPED
;
2666 bp
= xfs_buf_read(mp
->m_ddev_targp
, buf_f
->blf_blkno
, buf_f
->blf_len
,
2670 error
= bp
->b_error
;
2672 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#1)");
2677 * Recover the buffer only if we get an LSN from it and it's less than
2678 * the lsn of the transaction we are replaying.
2680 * Note that we have to be extremely careful of readahead here.
2681 * Readahead does not attach verfiers to the buffers so if we don't
2682 * actually do any replay after readahead because of the LSN we found
2683 * in the buffer if more recent than that current transaction then we
2684 * need to attach the verifier directly. Failure to do so can lead to
2685 * future recovery actions (e.g. EFI and unlinked list recovery) can
2686 * operate on the buffers and they won't get the verifier attached. This
2687 * can lead to blocks on disk having the correct content but a stale
2690 * It is safe to assume these clean buffers are currently up to date.
2691 * If the buffer is dirtied by a later transaction being replayed, then
2692 * the verifier will be reset to match whatever recover turns that
2695 lsn
= xlog_recover_get_buf_lsn(mp
, bp
);
2696 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
2697 xlog_recover_validate_buf_type(mp
, bp
, buf_f
);
2701 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
2702 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2705 } else if (buf_f
->blf_flags
&
2706 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2709 dirty
= xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2713 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2717 * Perform delayed write on the buffer. Asynchronous writes will be
2718 * slower when taking into account all the buffers to be flushed.
2720 * Also make sure that only inode buffers with good sizes stay in
2721 * the buffer cache. The kernel moves inodes in buffers of 1 block
2722 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
2723 * buffers in the log can be a different size if the log was generated
2724 * by an older kernel using unclustered inode buffers or a newer kernel
2725 * running with a different inode cluster size. Regardless, if the
2726 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2727 * for *our* value of mp->m_inode_cluster_size, then we need to keep
2728 * the buffer out of the buffer cache so that the buffer won't
2729 * overlap with future reads of those inodes.
2731 if (XFS_DINODE_MAGIC
==
2732 be16_to_cpu(*((__be16
*)xfs_buf_offset(bp
, 0))) &&
2733 (BBTOB(bp
->b_io_length
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2734 (__uint32_t
)log
->l_mp
->m_inode_cluster_size
))) {
2736 error
= xfs_bwrite(bp
);
2738 ASSERT(bp
->b_target
->bt_mount
== mp
);
2739 bp
->b_iodone
= xlog_recover_iodone
;
2740 xfs_buf_delwri_queue(bp
, buffer_list
);
2749 * Inode fork owner changes
2751 * If we have been told that we have to reparent the inode fork, it's because an
2752 * extent swap operation on a CRC enabled filesystem has been done and we are
2753 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2756 * The complexity here is that we don't have an inode context to work with, so
2757 * after we've replayed the inode we need to instantiate one. This is where the
2760 * We are in the middle of log recovery, so we can't run transactions. That
2761 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2762 * that will result in the corresponding iput() running the inode through
2763 * xfs_inactive(). If we've just replayed an inode core that changes the link
2764 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2765 * transactions (bad!).
2767 * So, to avoid this, we instantiate an inode directly from the inode core we've
2768 * just recovered. We have the buffer still locked, and all we really need to
2769 * instantiate is the inode core and the forks being modified. We can do this
2770 * manually, then run the inode btree owner change, and then tear down the
2771 * xfs_inode without having to run any transactions at all.
2773 * Also, because we don't have a transaction context available here but need to
2774 * gather all the buffers we modify for writeback so we pass the buffer_list
2775 * instead for the operation to use.
2779 xfs_recover_inode_owner_change(
2780 struct xfs_mount
*mp
,
2781 struct xfs_dinode
*dip
,
2782 struct xfs_inode_log_format
*in_f
,
2783 struct list_head
*buffer_list
)
2785 struct xfs_inode
*ip
;
2788 ASSERT(in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
));
2790 ip
= xfs_inode_alloc(mp
, in_f
->ilf_ino
);
2794 /* instantiate the inode */
2795 xfs_dinode_from_disk(&ip
->i_d
, dip
);
2796 ASSERT(ip
->i_d
.di_version
>= 3);
2798 error
= xfs_iformat_fork(ip
, dip
);
2803 if (in_f
->ilf_fields
& XFS_ILOG_DOWNER
) {
2804 ASSERT(in_f
->ilf_fields
& XFS_ILOG_DBROOT
);
2805 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_DATA_FORK
,
2806 ip
->i_ino
, buffer_list
);
2811 if (in_f
->ilf_fields
& XFS_ILOG_AOWNER
) {
2812 ASSERT(in_f
->ilf_fields
& XFS_ILOG_ABROOT
);
2813 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_ATTR_FORK
,
2814 ip
->i_ino
, buffer_list
);
2825 xlog_recover_inode_pass2(
2827 struct list_head
*buffer_list
,
2828 struct xlog_recover_item
*item
,
2829 xfs_lsn_t current_lsn
)
2831 xfs_inode_log_format_t
*in_f
;
2832 xfs_mount_t
*mp
= log
->l_mp
;
2841 xfs_icdinode_t
*dicp
;
2845 if (item
->ri_buf
[0].i_len
== sizeof(xfs_inode_log_format_t
)) {
2846 in_f
= item
->ri_buf
[0].i_addr
;
2848 in_f
= kmem_alloc(sizeof(xfs_inode_log_format_t
), KM_SLEEP
);
2850 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
2856 * Inode buffers can be freed, look out for it,
2857 * and do not replay the inode.
2859 if (xlog_check_buffer_cancelled(log
, in_f
->ilf_blkno
,
2860 in_f
->ilf_len
, 0)) {
2862 trace_xfs_log_recover_inode_cancel(log
, in_f
);
2865 trace_xfs_log_recover_inode_recover(log
, in_f
);
2867 bp
= xfs_buf_read(mp
->m_ddev_targp
, in_f
->ilf_blkno
, in_f
->ilf_len
, 0,
2868 &xfs_inode_buf_ops
);
2873 error
= bp
->b_error
;
2875 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#2)");
2878 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2879 dip
= xfs_buf_offset(bp
, in_f
->ilf_boffset
);
2882 * Make sure the place we're flushing out to really looks
2885 if (unlikely(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
))) {
2887 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2888 __func__
, dip
, bp
, in_f
->ilf_ino
);
2889 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2890 XFS_ERRLEVEL_LOW
, mp
);
2891 error
= -EFSCORRUPTED
;
2894 dicp
= item
->ri_buf
[1].i_addr
;
2895 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2897 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2898 __func__
, item
, in_f
->ilf_ino
);
2899 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2900 XFS_ERRLEVEL_LOW
, mp
);
2901 error
= -EFSCORRUPTED
;
2906 * If the inode has an LSN in it, recover the inode only if it's less
2907 * than the lsn of the transaction we are replaying. Note: we still
2908 * need to replay an owner change even though the inode is more recent
2909 * than the transaction as there is no guarantee that all the btree
2910 * blocks are more recent than this transaction, too.
2912 if (dip
->di_version
>= 3) {
2913 xfs_lsn_t lsn
= be64_to_cpu(dip
->di_lsn
);
2915 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
2916 trace_xfs_log_recover_inode_skip(log
, in_f
);
2918 goto out_owner_change
;
2923 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2924 * are transactional and if ordering is necessary we can determine that
2925 * more accurately by the LSN field in the V3 inode core. Don't trust
2926 * the inode versions we might be changing them here - use the
2927 * superblock flag to determine whether we need to look at di_flushiter
2928 * to skip replay when the on disk inode is newer than the log one
2930 if (!xfs_sb_version_hascrc(&mp
->m_sb
) &&
2931 dicp
->di_flushiter
< be16_to_cpu(dip
->di_flushiter
)) {
2933 * Deal with the wrap case, DI_MAX_FLUSH is less
2934 * than smaller numbers
2936 if (be16_to_cpu(dip
->di_flushiter
) == DI_MAX_FLUSH
&&
2937 dicp
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
2940 trace_xfs_log_recover_inode_skip(log
, in_f
);
2946 /* Take the opportunity to reset the flush iteration count */
2947 dicp
->di_flushiter
= 0;
2949 if (unlikely(S_ISREG(dicp
->di_mode
))) {
2950 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2951 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2952 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2953 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2955 "%s: Bad regular inode log record, rec ptr 0x%p, "
2956 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2957 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2958 error
= -EFSCORRUPTED
;
2961 } else if (unlikely(S_ISDIR(dicp
->di_mode
))) {
2962 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2963 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2964 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2965 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2966 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2968 "%s: Bad dir inode log record, rec ptr 0x%p, "
2969 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2970 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2971 error
= -EFSCORRUPTED
;
2975 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2976 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2977 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2979 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2980 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2981 __func__
, item
, dip
, bp
, in_f
->ilf_ino
,
2982 dicp
->di_nextents
+ dicp
->di_anextents
,
2984 error
= -EFSCORRUPTED
;
2987 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2988 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2989 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2991 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2992 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__
,
2993 item
, dip
, bp
, in_f
->ilf_ino
, dicp
->di_forkoff
);
2994 error
= -EFSCORRUPTED
;
2997 isize
= xfs_icdinode_size(dicp
->di_version
);
2998 if (unlikely(item
->ri_buf
[1].i_len
> isize
)) {
2999 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
3000 XFS_ERRLEVEL_LOW
, mp
, dicp
);
3002 "%s: Bad inode log record length %d, rec ptr 0x%p",
3003 __func__
, item
->ri_buf
[1].i_len
, item
);
3004 error
= -EFSCORRUPTED
;
3008 /* The core is in in-core format */
3009 xfs_dinode_to_disk(dip
, dicp
);
3011 /* the rest is in on-disk format */
3012 if (item
->ri_buf
[1].i_len
> isize
) {
3013 memcpy((char *)dip
+ isize
,
3014 item
->ri_buf
[1].i_addr
+ isize
,
3015 item
->ri_buf
[1].i_len
- isize
);
3018 fields
= in_f
->ilf_fields
;
3019 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
3021 xfs_dinode_put_rdev(dip
, in_f
->ilf_u
.ilfu_rdev
);
3024 memcpy(XFS_DFORK_DPTR(dip
),
3025 &in_f
->ilf_u
.ilfu_uuid
,
3030 if (in_f
->ilf_size
== 2)
3031 goto out_owner_change
;
3032 len
= item
->ri_buf
[2].i_len
;
3033 src
= item
->ri_buf
[2].i_addr
;
3034 ASSERT(in_f
->ilf_size
<= 4);
3035 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
3036 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
3037 (len
== in_f
->ilf_dsize
));
3039 switch (fields
& XFS_ILOG_DFORK
) {
3040 case XFS_ILOG_DDATA
:
3042 memcpy(XFS_DFORK_DPTR(dip
), src
, len
);
3045 case XFS_ILOG_DBROOT
:
3046 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
, len
,
3047 (xfs_bmdr_block_t
*)XFS_DFORK_DPTR(dip
),
3048 XFS_DFORK_DSIZE(dip
, mp
));
3053 * There are no data fork flags set.
3055 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
3060 * If we logged any attribute data, recover it. There may or
3061 * may not have been any other non-core data logged in this
3064 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
3065 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
3070 len
= item
->ri_buf
[attr_index
].i_len
;
3071 src
= item
->ri_buf
[attr_index
].i_addr
;
3072 ASSERT(len
== in_f
->ilf_asize
);
3074 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
3075 case XFS_ILOG_ADATA
:
3077 dest
= XFS_DFORK_APTR(dip
);
3078 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
3079 memcpy(dest
, src
, len
);
3082 case XFS_ILOG_ABROOT
:
3083 dest
= XFS_DFORK_APTR(dip
);
3084 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
,
3085 len
, (xfs_bmdr_block_t
*)dest
,
3086 XFS_DFORK_ASIZE(dip
, mp
));
3090 xfs_warn(log
->l_mp
, "%s: Invalid flag", __func__
);
3098 if (in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
))
3099 error
= xfs_recover_inode_owner_change(mp
, dip
, in_f
,
3101 /* re-generate the checksum. */
3102 xfs_dinode_calc_crc(log
->l_mp
, dip
);
3104 ASSERT(bp
->b_target
->bt_mount
== mp
);
3105 bp
->b_iodone
= xlog_recover_iodone
;
3106 xfs_buf_delwri_queue(bp
, buffer_list
);
3117 * Recover QUOTAOFF records. We simply make a note of it in the xlog
3118 * structure, so that we know not to do any dquot item or dquot buffer recovery,
3122 xlog_recover_quotaoff_pass1(
3124 struct xlog_recover_item
*item
)
3126 xfs_qoff_logformat_t
*qoff_f
= item
->ri_buf
[0].i_addr
;
3130 * The logitem format's flag tells us if this was user quotaoff,
3131 * group/project quotaoff or both.
3133 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
3134 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
3135 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
3136 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
3137 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
3138 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
3144 * Recover a dquot record
3147 xlog_recover_dquot_pass2(
3149 struct list_head
*buffer_list
,
3150 struct xlog_recover_item
*item
,
3151 xfs_lsn_t current_lsn
)
3153 xfs_mount_t
*mp
= log
->l_mp
;
3155 struct xfs_disk_dquot
*ddq
, *recddq
;
3157 xfs_dq_logformat_t
*dq_f
;
3162 * Filesystems are required to send in quota flags at mount time.
3164 if (mp
->m_qflags
== 0)
3167 recddq
= item
->ri_buf
[1].i_addr
;
3168 if (recddq
== NULL
) {
3169 xfs_alert(log
->l_mp
, "NULL dquot in %s.", __func__
);
3172 if (item
->ri_buf
[1].i_len
< sizeof(xfs_disk_dquot_t
)) {
3173 xfs_alert(log
->l_mp
, "dquot too small (%d) in %s.",
3174 item
->ri_buf
[1].i_len
, __func__
);
3179 * This type of quotas was turned off, so ignore this record.
3181 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
3183 if (log
->l_quotaoffs_flag
& type
)
3187 * At this point we know that quota was _not_ turned off.
3188 * Since the mount flags are not indicating to us otherwise, this
3189 * must mean that quota is on, and the dquot needs to be replayed.
3190 * Remember that we may not have fully recovered the superblock yet,
3191 * so we can't do the usual trick of looking at the SB quota bits.
3193 * The other possibility, of course, is that the quota subsystem was
3194 * removed since the last mount - ENOSYS.
3196 dq_f
= item
->ri_buf
[0].i_addr
;
3198 error
= xfs_dqcheck(mp
, recddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
3199 "xlog_recover_dquot_pass2 (log copy)");
3202 ASSERT(dq_f
->qlf_len
== 1);
3205 * At this point we are assuming that the dquots have been allocated
3206 * and hence the buffer has valid dquots stamped in it. It should,
3207 * therefore, pass verifier validation. If the dquot is bad, then the
3208 * we'll return an error here, so we don't need to specifically check
3209 * the dquot in the buffer after the verifier has run.
3211 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
, dq_f
->qlf_blkno
,
3212 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
), 0, &bp
,
3213 &xfs_dquot_buf_ops
);
3218 ddq
= xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
3221 * If the dquot has an LSN in it, recover the dquot only if it's less
3222 * than the lsn of the transaction we are replaying.
3224 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
3225 struct xfs_dqblk
*dqb
= (struct xfs_dqblk
*)ddq
;
3226 xfs_lsn_t lsn
= be64_to_cpu(dqb
->dd_lsn
);
3228 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
3233 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
3234 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
3235 xfs_update_cksum((char *)ddq
, sizeof(struct xfs_dqblk
),
3239 ASSERT(dq_f
->qlf_size
== 2);
3240 ASSERT(bp
->b_target
->bt_mount
== mp
);
3241 bp
->b_iodone
= xlog_recover_iodone
;
3242 xfs_buf_delwri_queue(bp
, buffer_list
);
3250 * This routine is called to create an in-core extent free intent
3251 * item from the efi format structure which was logged on disk.
3252 * It allocates an in-core efi, copies the extents from the format
3253 * structure into it, and adds the efi to the AIL with the given
3257 xlog_recover_efi_pass2(
3259 struct xlog_recover_item
*item
,
3263 struct xfs_mount
*mp
= log
->l_mp
;
3264 struct xfs_efi_log_item
*efip
;
3265 struct xfs_efi_log_format
*efi_formatp
;
3267 efi_formatp
= item
->ri_buf
[0].i_addr
;
3269 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
3270 error
= xfs_efi_copy_format(&item
->ri_buf
[0], &efip
->efi_format
);
3272 xfs_efi_item_free(efip
);
3275 atomic_set(&efip
->efi_next_extent
, efi_formatp
->efi_nextents
);
3277 spin_lock(&log
->l_ailp
->xa_lock
);
3279 * The EFI has two references. One for the EFD and one for EFI to ensure
3280 * it makes it into the AIL. Insert the EFI into the AIL directly and
3281 * drop the EFI reference. Note that xfs_trans_ail_update() drops the
3284 xfs_trans_ail_update(log
->l_ailp
, &efip
->efi_item
, lsn
);
3285 xfs_efi_release(efip
);
3291 * This routine is called when an EFD format structure is found in a committed
3292 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
3293 * was still in the log. To do this it searches the AIL for the EFI with an id
3294 * equal to that in the EFD format structure. If we find it we drop the EFD
3295 * reference, which removes the EFI from the AIL and frees it.
3298 xlog_recover_efd_pass2(
3300 struct xlog_recover_item
*item
)
3302 xfs_efd_log_format_t
*efd_formatp
;
3303 xfs_efi_log_item_t
*efip
= NULL
;
3304 xfs_log_item_t
*lip
;
3306 struct xfs_ail_cursor cur
;
3307 struct xfs_ail
*ailp
= log
->l_ailp
;
3309 efd_formatp
= item
->ri_buf
[0].i_addr
;
3310 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
3311 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
3312 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
3313 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
3314 efi_id
= efd_formatp
->efd_efi_id
;
3317 * Search for the EFI with the id in the EFD format structure in the
3320 spin_lock(&ailp
->xa_lock
);
3321 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3322 while (lip
!= NULL
) {
3323 if (lip
->li_type
== XFS_LI_EFI
) {
3324 efip
= (xfs_efi_log_item_t
*)lip
;
3325 if (efip
->efi_format
.efi_id
== efi_id
) {
3327 * Drop the EFD reference to the EFI. This
3328 * removes the EFI from the AIL and frees it.
3330 spin_unlock(&ailp
->xa_lock
);
3331 xfs_efi_release(efip
);
3332 spin_lock(&ailp
->xa_lock
);
3336 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3339 xfs_trans_ail_cursor_done(&cur
);
3340 spin_unlock(&ailp
->xa_lock
);
3346 * This routine is called when an inode create format structure is found in a
3347 * committed transaction in the log. It's purpose is to initialise the inodes
3348 * being allocated on disk. This requires us to get inode cluster buffers that
3349 * match the range to be intialised, stamped with inode templates and written
3350 * by delayed write so that subsequent modifications will hit the cached buffer
3351 * and only need writing out at the end of recovery.
3354 xlog_recover_do_icreate_pass2(
3356 struct list_head
*buffer_list
,
3357 xlog_recover_item_t
*item
)
3359 struct xfs_mount
*mp
= log
->l_mp
;
3360 struct xfs_icreate_log
*icl
;
3361 xfs_agnumber_t agno
;
3362 xfs_agblock_t agbno
;
3365 xfs_agblock_t length
;
3366 int blks_per_cluster
;
3372 icl
= (struct xfs_icreate_log
*)item
->ri_buf
[0].i_addr
;
3373 if (icl
->icl_type
!= XFS_LI_ICREATE
) {
3374 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad type");
3378 if (icl
->icl_size
!= 1) {
3379 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad icl size");
3383 agno
= be32_to_cpu(icl
->icl_ag
);
3384 if (agno
>= mp
->m_sb
.sb_agcount
) {
3385 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad agno");
3388 agbno
= be32_to_cpu(icl
->icl_agbno
);
3389 if (!agbno
|| agbno
== NULLAGBLOCK
|| agbno
>= mp
->m_sb
.sb_agblocks
) {
3390 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad agbno");
3393 isize
= be32_to_cpu(icl
->icl_isize
);
3394 if (isize
!= mp
->m_sb
.sb_inodesize
) {
3395 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad isize");
3398 count
= be32_to_cpu(icl
->icl_count
);
3400 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad count");
3403 length
= be32_to_cpu(icl
->icl_length
);
3404 if (!length
|| length
>= mp
->m_sb
.sb_agblocks
) {
3405 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad length");
3410 * The inode chunk is either full or sparse and we only support
3411 * m_ialloc_min_blks sized sparse allocations at this time.
3413 if (length
!= mp
->m_ialloc_blks
&&
3414 length
!= mp
->m_ialloc_min_blks
) {
3416 "%s: unsupported chunk length", __FUNCTION__
);
3420 /* verify inode count is consistent with extent length */
3421 if ((count
>> mp
->m_sb
.sb_inopblog
) != length
) {
3423 "%s: inconsistent inode count and chunk length",
3429 * The icreate transaction can cover multiple cluster buffers and these
3430 * buffers could have been freed and reused. Check the individual
3431 * buffers for cancellation so we don't overwrite anything written after
3434 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
3435 bb_per_cluster
= XFS_FSB_TO_BB(mp
, blks_per_cluster
);
3436 nbufs
= length
/ blks_per_cluster
;
3437 for (i
= 0, cancel_count
= 0; i
< nbufs
; i
++) {
3440 daddr
= XFS_AGB_TO_DADDR(mp
, agno
,
3441 agbno
+ i
* blks_per_cluster
);
3442 if (xlog_check_buffer_cancelled(log
, daddr
, bb_per_cluster
, 0))
3447 * We currently only use icreate for a single allocation at a time. This
3448 * means we should expect either all or none of the buffers to be
3449 * cancelled. Be conservative and skip replay if at least one buffer is
3450 * cancelled, but warn the user that something is awry if the buffers
3451 * are not consistent.
3453 * XXX: This must be refined to only skip cancelled clusters once we use
3454 * icreate for multiple chunk allocations.
3456 ASSERT(!cancel_count
|| cancel_count
== nbufs
);
3458 if (cancel_count
!= nbufs
)
3460 "WARNING: partial inode chunk cancellation, skipped icreate.");
3461 trace_xfs_log_recover_icreate_cancel(log
, icl
);
3465 trace_xfs_log_recover_icreate_recover(log
, icl
);
3466 return xfs_ialloc_inode_init(mp
, NULL
, buffer_list
, count
, agno
, agbno
,
3467 length
, be32_to_cpu(icl
->icl_gen
));
3471 xlog_recover_buffer_ra_pass2(
3473 struct xlog_recover_item
*item
)
3475 struct xfs_buf_log_format
*buf_f
= item
->ri_buf
[0].i_addr
;
3476 struct xfs_mount
*mp
= log
->l_mp
;
3478 if (xlog_peek_buffer_cancelled(log
, buf_f
->blf_blkno
,
3479 buf_f
->blf_len
, buf_f
->blf_flags
)) {
3483 xfs_buf_readahead(mp
->m_ddev_targp
, buf_f
->blf_blkno
,
3484 buf_f
->blf_len
, NULL
);
3488 xlog_recover_inode_ra_pass2(
3490 struct xlog_recover_item
*item
)
3492 struct xfs_inode_log_format ilf_buf
;
3493 struct xfs_inode_log_format
*ilfp
;
3494 struct xfs_mount
*mp
= log
->l_mp
;
3497 if (item
->ri_buf
[0].i_len
== sizeof(struct xfs_inode_log_format
)) {
3498 ilfp
= item
->ri_buf
[0].i_addr
;
3501 memset(ilfp
, 0, sizeof(*ilfp
));
3502 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], ilfp
);
3507 if (xlog_peek_buffer_cancelled(log
, ilfp
->ilf_blkno
, ilfp
->ilf_len
, 0))
3510 xfs_buf_readahead(mp
->m_ddev_targp
, ilfp
->ilf_blkno
,
3511 ilfp
->ilf_len
, &xfs_inode_buf_ra_ops
);
3515 xlog_recover_dquot_ra_pass2(
3517 struct xlog_recover_item
*item
)
3519 struct xfs_mount
*mp
= log
->l_mp
;
3520 struct xfs_disk_dquot
*recddq
;
3521 struct xfs_dq_logformat
*dq_f
;
3526 if (mp
->m_qflags
== 0)
3529 recddq
= item
->ri_buf
[1].i_addr
;
3532 if (item
->ri_buf
[1].i_len
< sizeof(struct xfs_disk_dquot
))
3535 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
3537 if (log
->l_quotaoffs_flag
& type
)
3540 dq_f
= item
->ri_buf
[0].i_addr
;
3542 ASSERT(dq_f
->qlf_len
== 1);
3544 len
= XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
);
3545 if (xlog_peek_buffer_cancelled(log
, dq_f
->qlf_blkno
, len
, 0))
3548 xfs_buf_readahead(mp
->m_ddev_targp
, dq_f
->qlf_blkno
, len
,
3549 &xfs_dquot_buf_ra_ops
);
3553 xlog_recover_ra_pass2(
3555 struct xlog_recover_item
*item
)
3557 switch (ITEM_TYPE(item
)) {
3559 xlog_recover_buffer_ra_pass2(log
, item
);
3562 xlog_recover_inode_ra_pass2(log
, item
);
3565 xlog_recover_dquot_ra_pass2(log
, item
);
3569 case XFS_LI_QUOTAOFF
:
3576 xlog_recover_commit_pass1(
3578 struct xlog_recover
*trans
,
3579 struct xlog_recover_item
*item
)
3581 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS1
);
3583 switch (ITEM_TYPE(item
)) {
3585 return xlog_recover_buffer_pass1(log
, item
);
3586 case XFS_LI_QUOTAOFF
:
3587 return xlog_recover_quotaoff_pass1(log
, item
);
3592 case XFS_LI_ICREATE
:
3593 /* nothing to do in pass 1 */
3596 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
3597 __func__
, ITEM_TYPE(item
));
3604 xlog_recover_commit_pass2(
3606 struct xlog_recover
*trans
,
3607 struct list_head
*buffer_list
,
3608 struct xlog_recover_item
*item
)
3610 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS2
);
3612 switch (ITEM_TYPE(item
)) {
3614 return xlog_recover_buffer_pass2(log
, buffer_list
, item
,
3617 return xlog_recover_inode_pass2(log
, buffer_list
, item
,
3620 return xlog_recover_efi_pass2(log
, item
, trans
->r_lsn
);
3622 return xlog_recover_efd_pass2(log
, item
);
3624 return xlog_recover_dquot_pass2(log
, buffer_list
, item
,
3626 case XFS_LI_ICREATE
:
3627 return xlog_recover_do_icreate_pass2(log
, buffer_list
, item
);
3628 case XFS_LI_QUOTAOFF
:
3629 /* nothing to do in pass2 */
3632 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
3633 __func__
, ITEM_TYPE(item
));
3640 xlog_recover_items_pass2(
3642 struct xlog_recover
*trans
,
3643 struct list_head
*buffer_list
,
3644 struct list_head
*item_list
)
3646 struct xlog_recover_item
*item
;
3649 list_for_each_entry(item
, item_list
, ri_list
) {
3650 error
= xlog_recover_commit_pass2(log
, trans
,
3660 * Perform the transaction.
3662 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3663 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3666 xlog_recover_commit_trans(
3668 struct xlog_recover
*trans
,
3673 int items_queued
= 0;
3674 struct xlog_recover_item
*item
;
3675 struct xlog_recover_item
*next
;
3676 LIST_HEAD (buffer_list
);
3677 LIST_HEAD (ra_list
);
3678 LIST_HEAD (done_list
);
3680 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
3682 hlist_del(&trans
->r_list
);
3684 error
= xlog_recover_reorder_trans(log
, trans
, pass
);
3688 list_for_each_entry_safe(item
, next
, &trans
->r_itemq
, ri_list
) {
3690 case XLOG_RECOVER_PASS1
:
3691 error
= xlog_recover_commit_pass1(log
, trans
, item
);
3693 case XLOG_RECOVER_PASS2
:
3694 xlog_recover_ra_pass2(log
, item
);
3695 list_move_tail(&item
->ri_list
, &ra_list
);
3697 if (items_queued
>= XLOG_RECOVER_COMMIT_QUEUE_MAX
) {
3698 error
= xlog_recover_items_pass2(log
, trans
,
3699 &buffer_list
, &ra_list
);
3700 list_splice_tail_init(&ra_list
, &done_list
);
3714 if (!list_empty(&ra_list
)) {
3716 error
= xlog_recover_items_pass2(log
, trans
,
3717 &buffer_list
, &ra_list
);
3718 list_splice_tail_init(&ra_list
, &done_list
);
3721 if (!list_empty(&done_list
))
3722 list_splice_init(&done_list
, &trans
->r_itemq
);
3724 error2
= xfs_buf_delwri_submit(&buffer_list
);
3725 return error
? error
: error2
;
3729 xlog_recover_add_item(
3730 struct list_head
*head
)
3732 xlog_recover_item_t
*item
;
3734 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
3735 INIT_LIST_HEAD(&item
->ri_list
);
3736 list_add_tail(&item
->ri_list
, head
);
3740 xlog_recover_add_to_cont_trans(
3742 struct xlog_recover
*trans
,
3746 xlog_recover_item_t
*item
;
3747 char *ptr
, *old_ptr
;
3751 * If the transaction is empty, the header was split across this and the
3752 * previous record. Copy the rest of the header.
3754 if (list_empty(&trans
->r_itemq
)) {
3755 ASSERT(len
<= sizeof(struct xfs_trans_header
));
3756 if (len
> sizeof(struct xfs_trans_header
)) {
3757 xfs_warn(log
->l_mp
, "%s: bad header length", __func__
);
3761 xlog_recover_add_item(&trans
->r_itemq
);
3762 ptr
= (char *)&trans
->r_theader
+
3763 sizeof(struct xfs_trans_header
) - len
;
3764 memcpy(ptr
, dp
, len
);
3768 /* take the tail entry */
3769 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
3771 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
3772 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
3774 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, KM_SLEEP
);
3775 memcpy(&ptr
[old_len
], dp
, len
);
3776 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
3777 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
3778 trace_xfs_log_recover_item_add_cont(log
, trans
, item
, 0);
3783 * The next region to add is the start of a new region. It could be
3784 * a whole region or it could be the first part of a new region. Because
3785 * of this, the assumption here is that the type and size fields of all
3786 * format structures fit into the first 32 bits of the structure.
3788 * This works because all regions must be 32 bit aligned. Therefore, we
3789 * either have both fields or we have neither field. In the case we have
3790 * neither field, the data part of the region is zero length. We only have
3791 * a log_op_header and can throw away the header since a new one will appear
3792 * later. If we have at least 4 bytes, then we can determine how many regions
3793 * will appear in the current log item.
3796 xlog_recover_add_to_trans(
3798 struct xlog_recover
*trans
,
3802 xfs_inode_log_format_t
*in_f
; /* any will do */
3803 xlog_recover_item_t
*item
;
3808 if (list_empty(&trans
->r_itemq
)) {
3809 /* we need to catch log corruptions here */
3810 if (*(uint
*)dp
!= XFS_TRANS_HEADER_MAGIC
) {
3811 xfs_warn(log
->l_mp
, "%s: bad header magic number",
3817 if (len
> sizeof(struct xfs_trans_header
)) {
3818 xfs_warn(log
->l_mp
, "%s: bad header length", __func__
);
3824 * The transaction header can be arbitrarily split across op
3825 * records. If we don't have the whole thing here, copy what we
3826 * do have and handle the rest in the next record.
3828 if (len
== sizeof(struct xfs_trans_header
))
3829 xlog_recover_add_item(&trans
->r_itemq
);
3830 memcpy(&trans
->r_theader
, dp
, len
);
3834 ptr
= kmem_alloc(len
, KM_SLEEP
);
3835 memcpy(ptr
, dp
, len
);
3836 in_f
= (xfs_inode_log_format_t
*)ptr
;
3838 /* take the tail entry */
3839 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
3840 if (item
->ri_total
!= 0 &&
3841 item
->ri_total
== item
->ri_cnt
) {
3842 /* tail item is in use, get a new one */
3843 xlog_recover_add_item(&trans
->r_itemq
);
3844 item
= list_entry(trans
->r_itemq
.prev
,
3845 xlog_recover_item_t
, ri_list
);
3848 if (item
->ri_total
== 0) { /* first region to be added */
3849 if (in_f
->ilf_size
== 0 ||
3850 in_f
->ilf_size
> XLOG_MAX_REGIONS_IN_ITEM
) {
3852 "bad number of regions (%d) in inode log format",
3859 item
->ri_total
= in_f
->ilf_size
;
3861 kmem_zalloc(item
->ri_total
* sizeof(xfs_log_iovec_t
),
3864 ASSERT(item
->ri_total
> item
->ri_cnt
);
3865 /* Description region is ri_buf[0] */
3866 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
3867 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
3869 trace_xfs_log_recover_item_add(log
, trans
, item
, 0);
3874 * Free up any resources allocated by the transaction
3876 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3879 xlog_recover_free_trans(
3880 struct xlog_recover
*trans
)
3882 xlog_recover_item_t
*item
, *n
;
3885 list_for_each_entry_safe(item
, n
, &trans
->r_itemq
, ri_list
) {
3886 /* Free the regions in the item. */
3887 list_del(&item
->ri_list
);
3888 for (i
= 0; i
< item
->ri_cnt
; i
++)
3889 kmem_free(item
->ri_buf
[i
].i_addr
);
3890 /* Free the item itself */
3891 kmem_free(item
->ri_buf
);
3894 /* Free the transaction recover structure */
3899 * On error or completion, trans is freed.
3902 xlog_recovery_process_trans(
3904 struct xlog_recover
*trans
,
3911 bool freeit
= false;
3913 /* mask off ophdr transaction container flags */
3914 flags
&= ~XLOG_END_TRANS
;
3915 if (flags
& XLOG_WAS_CONT_TRANS
)
3916 flags
&= ~XLOG_CONTINUE_TRANS
;
3919 * Callees must not free the trans structure. We'll decide if we need to
3920 * free it or not based on the operation being done and it's result.
3923 /* expected flag values */
3925 case XLOG_CONTINUE_TRANS
:
3926 error
= xlog_recover_add_to_trans(log
, trans
, dp
, len
);
3928 case XLOG_WAS_CONT_TRANS
:
3929 error
= xlog_recover_add_to_cont_trans(log
, trans
, dp
, len
);
3931 case XLOG_COMMIT_TRANS
:
3932 error
= xlog_recover_commit_trans(log
, trans
, pass
);
3933 /* success or fail, we are now done with this transaction. */
3937 /* unexpected flag values */
3938 case XLOG_UNMOUNT_TRANS
:
3939 /* just skip trans */
3940 xfs_warn(log
->l_mp
, "%s: Unmount LR", __func__
);
3943 case XLOG_START_TRANS
:
3945 xfs_warn(log
->l_mp
, "%s: bad flag 0x%x", __func__
, flags
);
3950 if (error
|| freeit
)
3951 xlog_recover_free_trans(trans
);
3956 * Lookup the transaction recovery structure associated with the ID in the
3957 * current ophdr. If the transaction doesn't exist and the start flag is set in
3958 * the ophdr, then allocate a new transaction for future ID matches to find.
3959 * Either way, return what we found during the lookup - an existing transaction
3962 STATIC
struct xlog_recover
*
3963 xlog_recover_ophdr_to_trans(
3964 struct hlist_head rhash
[],
3965 struct xlog_rec_header
*rhead
,
3966 struct xlog_op_header
*ohead
)
3968 struct xlog_recover
*trans
;
3970 struct hlist_head
*rhp
;
3972 tid
= be32_to_cpu(ohead
->oh_tid
);
3973 rhp
= &rhash
[XLOG_RHASH(tid
)];
3974 hlist_for_each_entry(trans
, rhp
, r_list
) {
3975 if (trans
->r_log_tid
== tid
)
3980 * skip over non-start transaction headers - we could be
3981 * processing slack space before the next transaction starts
3983 if (!(ohead
->oh_flags
& XLOG_START_TRANS
))
3986 ASSERT(be32_to_cpu(ohead
->oh_len
) == 0);
3989 * This is a new transaction so allocate a new recovery container to
3990 * hold the recovery ops that will follow.
3992 trans
= kmem_zalloc(sizeof(struct xlog_recover
), KM_SLEEP
);
3993 trans
->r_log_tid
= tid
;
3994 trans
->r_lsn
= be64_to_cpu(rhead
->h_lsn
);
3995 INIT_LIST_HEAD(&trans
->r_itemq
);
3996 INIT_HLIST_NODE(&trans
->r_list
);
3997 hlist_add_head(&trans
->r_list
, rhp
);
4000 * Nothing more to do for this ophdr. Items to be added to this new
4001 * transaction will be in subsequent ophdr containers.
4007 xlog_recover_process_ophdr(
4009 struct hlist_head rhash
[],
4010 struct xlog_rec_header
*rhead
,
4011 struct xlog_op_header
*ohead
,
4016 struct xlog_recover
*trans
;
4019 /* Do we understand who wrote this op? */
4020 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
4021 ohead
->oh_clientid
!= XFS_LOG
) {
4022 xfs_warn(log
->l_mp
, "%s: bad clientid 0x%x",
4023 __func__
, ohead
->oh_clientid
);
4029 * Check the ophdr contains all the data it is supposed to contain.
4031 len
= be32_to_cpu(ohead
->oh_len
);
4032 if (dp
+ len
> end
) {
4033 xfs_warn(log
->l_mp
, "%s: bad length 0x%x", __func__
, len
);
4038 trans
= xlog_recover_ophdr_to_trans(rhash
, rhead
, ohead
);
4040 /* nothing to do, so skip over this ophdr */
4044 return xlog_recovery_process_trans(log
, trans
, dp
, len
,
4045 ohead
->oh_flags
, pass
);
4049 * There are two valid states of the r_state field. 0 indicates that the
4050 * transaction structure is in a normal state. We have either seen the
4051 * start of the transaction or the last operation we added was not a partial
4052 * operation. If the last operation we added to the transaction was a
4053 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
4055 * NOTE: skip LRs with 0 data length.
4058 xlog_recover_process_data(
4060 struct hlist_head rhash
[],
4061 struct xlog_rec_header
*rhead
,
4065 struct xlog_op_header
*ohead
;
4070 end
= dp
+ be32_to_cpu(rhead
->h_len
);
4071 num_logops
= be32_to_cpu(rhead
->h_num_logops
);
4073 /* check the log format matches our own - else we can't recover */
4074 if (xlog_header_check_recover(log
->l_mp
, rhead
))
4077 while ((dp
< end
) && num_logops
) {
4079 ohead
= (struct xlog_op_header
*)dp
;
4080 dp
+= sizeof(*ohead
);
4083 /* errors will abort recovery */
4084 error
= xlog_recover_process_ophdr(log
, rhash
, rhead
, ohead
,
4089 dp
+= be32_to_cpu(ohead
->oh_len
);
4096 * Process an extent free intent item that was recovered from
4097 * the log. We need to free the extents that it describes.
4100 xlog_recover_process_efi(
4102 xfs_efi_log_item_t
*efip
)
4104 xfs_efd_log_item_t
*efdp
;
4109 xfs_fsblock_t startblock_fsb
;
4111 ASSERT(!test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
));
4114 * First check the validity of the extents described by the
4115 * EFI. If any are bad, then assume that all are bad and
4116 * just toss the EFI.
4118 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
4119 extp
= &(efip
->efi_format
.efi_extents
[i
]);
4120 startblock_fsb
= XFS_BB_TO_FSB(mp
,
4121 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
4122 if ((startblock_fsb
== 0) ||
4123 (extp
->ext_len
== 0) ||
4124 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
4125 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
4127 * This will pull the EFI from the AIL and
4128 * free the memory associated with it.
4130 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
4131 xfs_efi_release(efip
);
4136 tp
= xfs_trans_alloc(mp
, 0);
4137 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
4140 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
4142 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
4143 extp
= &(efip
->efi_format
.efi_extents
[i
]);
4144 error
= xfs_trans_free_extent(tp
, efdp
, extp
->ext_start
,
4151 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
4152 error
= xfs_trans_commit(tp
);
4156 xfs_trans_cancel(tp
);
4161 * When this is called, all of the EFIs which did not have
4162 * corresponding EFDs should be in the AIL. What we do now
4163 * is free the extents associated with each one.
4165 * Since we process the EFIs in normal transactions, they
4166 * will be removed at some point after the commit. This prevents
4167 * us from just walking down the list processing each one.
4168 * We'll use a flag in the EFI to skip those that we've already
4169 * processed and use the AIL iteration mechanism's generation
4170 * count to try to speed this up at least a bit.
4172 * When we start, we know that the EFIs are the only things in
4173 * the AIL. As we process them, however, other items are added
4174 * to the AIL. Since everything added to the AIL must come after
4175 * everything already in the AIL, we stop processing as soon as
4176 * we see something other than an EFI in the AIL.
4179 xlog_recover_process_efis(
4182 struct xfs_log_item
*lip
;
4183 struct xfs_efi_log_item
*efip
;
4185 struct xfs_ail_cursor cur
;
4186 struct xfs_ail
*ailp
;
4189 spin_lock(&ailp
->xa_lock
);
4190 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
4191 while (lip
!= NULL
) {
4193 * We're done when we see something other than an EFI.
4194 * There should be no EFIs left in the AIL now.
4196 if (lip
->li_type
!= XFS_LI_EFI
) {
4198 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
4199 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
4205 * Skip EFIs that we've already processed.
4207 efip
= container_of(lip
, struct xfs_efi_log_item
, efi_item
);
4208 if (test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
)) {
4209 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
4213 spin_unlock(&ailp
->xa_lock
);
4214 error
= xlog_recover_process_efi(log
->l_mp
, efip
);
4215 spin_lock(&ailp
->xa_lock
);
4218 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
4221 xfs_trans_ail_cursor_done(&cur
);
4222 spin_unlock(&ailp
->xa_lock
);
4227 * A cancel occurs when the mount has failed and we're bailing out. Release all
4228 * pending EFIs so they don't pin the AIL.
4231 xlog_recover_cancel_efis(
4234 struct xfs_log_item
*lip
;
4235 struct xfs_efi_log_item
*efip
;
4237 struct xfs_ail_cursor cur
;
4238 struct xfs_ail
*ailp
;
4241 spin_lock(&ailp
->xa_lock
);
4242 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
4243 while (lip
!= NULL
) {
4245 * We're done when we see something other than an EFI.
4246 * There should be no EFIs left in the AIL now.
4248 if (lip
->li_type
!= XFS_LI_EFI
) {
4250 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
4251 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
4256 efip
= container_of(lip
, struct xfs_efi_log_item
, efi_item
);
4258 spin_unlock(&ailp
->xa_lock
);
4259 xfs_efi_release(efip
);
4260 spin_lock(&ailp
->xa_lock
);
4262 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
4265 xfs_trans_ail_cursor_done(&cur
);
4266 spin_unlock(&ailp
->xa_lock
);
4271 * This routine performs a transaction to null out a bad inode pointer
4272 * in an agi unlinked inode hash bucket.
4275 xlog_recover_clear_agi_bucket(
4277 xfs_agnumber_t agno
,
4286 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
4287 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_clearagi
, 0, 0);
4291 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
4295 agi
= XFS_BUF_TO_AGI(agibp
);
4296 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
4297 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
4298 (sizeof(xfs_agino_t
) * bucket
);
4299 xfs_trans_log_buf(tp
, agibp
, offset
,
4300 (offset
+ sizeof(xfs_agino_t
) - 1));
4302 error
= xfs_trans_commit(tp
);
4308 xfs_trans_cancel(tp
);
4310 xfs_warn(mp
, "%s: failed to clear agi %d. Continuing.", __func__
, agno
);
4315 xlog_recover_process_one_iunlink(
4316 struct xfs_mount
*mp
,
4317 xfs_agnumber_t agno
,
4321 struct xfs_buf
*ibp
;
4322 struct xfs_dinode
*dip
;
4323 struct xfs_inode
*ip
;
4327 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
4328 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
);
4333 * Get the on disk inode to find the next inode in the bucket.
4335 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &ibp
, 0, 0);
4339 ASSERT(ip
->i_d
.di_nlink
== 0);
4340 ASSERT(ip
->i_d
.di_mode
!= 0);
4342 /* setup for the next pass */
4343 agino
= be32_to_cpu(dip
->di_next_unlinked
);
4347 * Prevent any DMAPI event from being sent when the reference on
4348 * the inode is dropped.
4350 ip
->i_d
.di_dmevmask
= 0;
4359 * We can't read in the inode this bucket points to, or this inode
4360 * is messed up. Just ditch this bucket of inodes. We will lose
4361 * some inodes and space, but at least we won't hang.
4363 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
4364 * clear the inode pointer in the bucket.
4366 xlog_recover_clear_agi_bucket(mp
, agno
, bucket
);
4371 * xlog_iunlink_recover
4373 * This is called during recovery to process any inodes which
4374 * we unlinked but not freed when the system crashed. These
4375 * inodes will be on the lists in the AGI blocks. What we do
4376 * here is scan all the AGIs and fully truncate and free any
4377 * inodes found on the lists. Each inode is removed from the
4378 * lists when it has been fully truncated and is freed. The
4379 * freeing of the inode and its removal from the list must be
4383 xlog_recover_process_iunlinks(
4387 xfs_agnumber_t agno
;
4398 * Prevent any DMAPI event from being sent while in this function.
4400 mp_dmevmask
= mp
->m_dmevmask
;
4403 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
4405 * Find the agi for this ag.
4407 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
4410 * AGI is b0rked. Don't process it.
4412 * We should probably mark the filesystem as corrupt
4413 * after we've recovered all the ag's we can....
4418 * Unlock the buffer so that it can be acquired in the normal
4419 * course of the transaction to truncate and free each inode.
4420 * Because we are not racing with anyone else here for the AGI
4421 * buffer, we don't even need to hold it locked to read the
4422 * initial unlinked bucket entries out of the buffer. We keep
4423 * buffer reference though, so that it stays pinned in memory
4424 * while we need the buffer.
4426 agi
= XFS_BUF_TO_AGI(agibp
);
4427 xfs_buf_unlock(agibp
);
4429 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
4430 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
4431 while (agino
!= NULLAGINO
) {
4432 agino
= xlog_recover_process_one_iunlink(mp
,
4433 agno
, agino
, bucket
);
4436 xfs_buf_rele(agibp
);
4439 mp
->m_dmevmask
= mp_dmevmask
;
4444 struct xlog_rec_header
*rhead
,
4450 for (i
= 0; i
< BTOBB(be32_to_cpu(rhead
->h_len
)) &&
4451 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
4452 *(__be32
*)dp
= *(__be32
*)&rhead
->h_cycle_data
[i
];
4456 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
4457 xlog_in_core_2_t
*xhdr
= (xlog_in_core_2_t
*)rhead
;
4458 for ( ; i
< BTOBB(be32_to_cpu(rhead
->h_len
)); i
++) {
4459 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
4460 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
4461 *(__be32
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
4470 * CRC check, unpack and process a log record.
4473 xlog_recover_process(
4475 struct hlist_head rhash
[],
4476 struct xlog_rec_header
*rhead
,
4483 crc
= xlog_cksum(log
, rhead
, dp
, be32_to_cpu(rhead
->h_len
));
4486 * Nothing else to do if this is a CRC verification pass. Just return
4487 * if this a record with a non-zero crc. Unfortunately, mkfs always
4488 * sets h_crc to 0 so we must consider this valid even on v5 supers.
4489 * Otherwise, return EFSBADCRC on failure so the callers up the stack
4490 * know precisely what failed.
4492 if (pass
== XLOG_RECOVER_CRCPASS
) {
4493 if (rhead
->h_crc
&& crc
!= le32_to_cpu(rhead
->h_crc
))
4499 * We're in the normal recovery path. Issue a warning if and only if the
4500 * CRC in the header is non-zero. This is an advisory warning and the
4501 * zero CRC check prevents warnings from being emitted when upgrading
4502 * the kernel from one that does not add CRCs by default.
4504 if (crc
!= le32_to_cpu(rhead
->h_crc
)) {
4505 if (rhead
->h_crc
|| xfs_sb_version_hascrc(&log
->l_mp
->m_sb
)) {
4506 xfs_alert(log
->l_mp
,
4507 "log record CRC mismatch: found 0x%x, expected 0x%x.",
4508 le32_to_cpu(rhead
->h_crc
),
4510 xfs_hex_dump(dp
, 32);
4514 * If the filesystem is CRC enabled, this mismatch becomes a
4515 * fatal log corruption failure.
4517 if (xfs_sb_version_hascrc(&log
->l_mp
->m_sb
))
4518 return -EFSCORRUPTED
;
4521 error
= xlog_unpack_data(rhead
, dp
, log
);
4525 return xlog_recover_process_data(log
, rhash
, rhead
, dp
, pass
);
4529 xlog_valid_rec_header(
4531 struct xlog_rec_header
*rhead
,
4536 if (unlikely(rhead
->h_magicno
!= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))) {
4537 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4538 XFS_ERRLEVEL_LOW
, log
->l_mp
);
4539 return -EFSCORRUPTED
;
4542 (!rhead
->h_version
||
4543 (be32_to_cpu(rhead
->h_version
) & (~XLOG_VERSION_OKBITS
))))) {
4544 xfs_warn(log
->l_mp
, "%s: unrecognised log version (%d).",
4545 __func__
, be32_to_cpu(rhead
->h_version
));
4549 /* LR body must have data or it wouldn't have been written */
4550 hlen
= be32_to_cpu(rhead
->h_len
);
4551 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
4552 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4553 XFS_ERRLEVEL_LOW
, log
->l_mp
);
4554 return -EFSCORRUPTED
;
4556 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
4557 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4558 XFS_ERRLEVEL_LOW
, log
->l_mp
);
4559 return -EFSCORRUPTED
;
4565 * Read the log from tail to head and process the log records found.
4566 * Handle the two cases where the tail and head are in the same cycle
4567 * and where the active portion of the log wraps around the end of
4568 * the physical log separately. The pass parameter is passed through
4569 * to the routines called to process the data and is not looked at
4573 xlog_do_recovery_pass(
4575 xfs_daddr_t head_blk
,
4576 xfs_daddr_t tail_blk
,
4578 xfs_daddr_t
*first_bad
) /* out: first bad log rec */
4580 xlog_rec_header_t
*rhead
;
4582 xfs_daddr_t rhead_blk
;
4584 xfs_buf_t
*hbp
, *dbp
;
4585 int error
= 0, h_size
, h_len
;
4586 int bblks
, split_bblks
;
4587 int hblks
, split_hblks
, wrapped_hblks
;
4588 struct hlist_head rhash
[XLOG_RHASH_SIZE
];
4590 ASSERT(head_blk
!= tail_blk
);
4594 * Read the header of the tail block and get the iclog buffer size from
4595 * h_size. Use this to tell how many sectors make up the log header.
4597 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
4599 * When using variable length iclogs, read first sector of
4600 * iclog header and extract the header size from it. Get a
4601 * new hbp that is the correct size.
4603 hbp
= xlog_get_bp(log
, 1);
4607 error
= xlog_bread(log
, tail_blk
, 1, hbp
, &offset
);
4611 rhead
= (xlog_rec_header_t
*)offset
;
4612 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
4617 * xfsprogs has a bug where record length is based on lsunit but
4618 * h_size (iclog size) is hardcoded to 32k. Now that we
4619 * unconditionally CRC verify the unmount record, this means the
4620 * log buffer can be too small for the record and cause an
4623 * Detect this condition here. Use lsunit for the buffer size as
4624 * long as this looks like the mkfs case. Otherwise, return an
4625 * error to avoid a buffer overrun.
4627 h_size
= be32_to_cpu(rhead
->h_size
);
4628 h_len
= be32_to_cpu(rhead
->h_len
);
4629 if (h_len
> h_size
) {
4630 if (h_len
<= log
->l_mp
->m_logbsize
&&
4631 be32_to_cpu(rhead
->h_num_logops
) == 1) {
4633 "invalid iclog size (%d bytes), using lsunit (%d bytes)",
4634 h_size
, log
->l_mp
->m_logbsize
);
4635 h_size
= log
->l_mp
->m_logbsize
;
4637 return -EFSCORRUPTED
;
4640 if ((be32_to_cpu(rhead
->h_version
) & XLOG_VERSION_2
) &&
4641 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
4642 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
4643 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
4646 hbp
= xlog_get_bp(log
, hblks
);
4651 ASSERT(log
->l_sectBBsize
== 1);
4653 hbp
= xlog_get_bp(log
, 1);
4654 h_size
= XLOG_BIG_RECORD_BSIZE
;
4659 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
4665 memset(rhash
, 0, sizeof(rhash
));
4666 blk_no
= rhead_blk
= tail_blk
;
4667 if (tail_blk
> head_blk
) {
4669 * Perform recovery around the end of the physical log.
4670 * When the head is not on the same cycle number as the tail,
4671 * we can't do a sequential recovery.
4673 while (blk_no
< log
->l_logBBsize
) {
4675 * Check for header wrapping around physical end-of-log
4677 offset
= hbp
->b_addr
;
4680 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
4681 /* Read header in one read */
4682 error
= xlog_bread(log
, blk_no
, hblks
, hbp
,
4687 /* This LR is split across physical log end */
4688 if (blk_no
!= log
->l_logBBsize
) {
4689 /* some data before physical log end */
4690 ASSERT(blk_no
<= INT_MAX
);
4691 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
4692 ASSERT(split_hblks
> 0);
4693 error
= xlog_bread(log
, blk_no
,
4701 * Note: this black magic still works with
4702 * large sector sizes (non-512) only because:
4703 * - we increased the buffer size originally
4704 * by 1 sector giving us enough extra space
4705 * for the second read;
4706 * - the log start is guaranteed to be sector
4708 * - we read the log end (LR header start)
4709 * _first_, then the log start (LR header end)
4710 * - order is important.
4712 wrapped_hblks
= hblks
- split_hblks
;
4713 error
= xlog_bread_offset(log
, 0,
4715 offset
+ BBTOB(split_hblks
));
4719 rhead
= (xlog_rec_header_t
*)offset
;
4720 error
= xlog_valid_rec_header(log
, rhead
,
4721 split_hblks
? blk_no
: 0);
4725 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
4728 /* Read in data for log record */
4729 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
4730 error
= xlog_bread(log
, blk_no
, bblks
, dbp
,
4735 /* This log record is split across the
4736 * physical end of log */
4737 offset
= dbp
->b_addr
;
4739 if (blk_no
!= log
->l_logBBsize
) {
4740 /* some data is before the physical
4742 ASSERT(!wrapped_hblks
);
4743 ASSERT(blk_no
<= INT_MAX
);
4745 log
->l_logBBsize
- (int)blk_no
;
4746 ASSERT(split_bblks
> 0);
4747 error
= xlog_bread(log
, blk_no
,
4755 * Note: this black magic still works with
4756 * large sector sizes (non-512) only because:
4757 * - we increased the buffer size originally
4758 * by 1 sector giving us enough extra space
4759 * for the second read;
4760 * - the log start is guaranteed to be sector
4762 * - we read the log end (LR header start)
4763 * _first_, then the log start (LR header end)
4764 * - order is important.
4766 error
= xlog_bread_offset(log
, 0,
4767 bblks
- split_bblks
, dbp
,
4768 offset
+ BBTOB(split_bblks
));
4773 error
= xlog_recover_process(log
, rhash
, rhead
, offset
,
4782 ASSERT(blk_no
>= log
->l_logBBsize
);
4783 blk_no
-= log
->l_logBBsize
;
4787 /* read first part of physical log */
4788 while (blk_no
< head_blk
) {
4789 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
4793 rhead
= (xlog_rec_header_t
*)offset
;
4794 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
4798 /* blocks in data section */
4799 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
4800 error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
,
4805 error
= xlog_recover_process(log
, rhash
, rhead
, offset
, pass
);
4809 blk_no
+= bblks
+ hblks
;
4818 if (error
&& first_bad
)
4819 *first_bad
= rhead_blk
;
4825 * Do the recovery of the log. We actually do this in two phases.
4826 * The two passes are necessary in order to implement the function
4827 * of cancelling a record written into the log. The first pass
4828 * determines those things which have been cancelled, and the
4829 * second pass replays log items normally except for those which
4830 * have been cancelled. The handling of the replay and cancellations
4831 * takes place in the log item type specific routines.
4833 * The table of items which have cancel records in the log is allocated
4834 * and freed at this level, since only here do we know when all of
4835 * the log recovery has been completed.
4838 xlog_do_log_recovery(
4840 xfs_daddr_t head_blk
,
4841 xfs_daddr_t tail_blk
)
4845 ASSERT(head_blk
!= tail_blk
);
4848 * First do a pass to find all of the cancelled buf log items.
4849 * Store them in the buf_cancel_table for use in the second pass.
4851 log
->l_buf_cancel_table
= kmem_zalloc(XLOG_BC_TABLE_SIZE
*
4852 sizeof(struct list_head
),
4854 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
4855 INIT_LIST_HEAD(&log
->l_buf_cancel_table
[i
]);
4857 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
4858 XLOG_RECOVER_PASS1
, NULL
);
4860 kmem_free(log
->l_buf_cancel_table
);
4861 log
->l_buf_cancel_table
= NULL
;
4865 * Then do a second pass to actually recover the items in the log.
4866 * When it is complete free the table of buf cancel items.
4868 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
4869 XLOG_RECOVER_PASS2
, NULL
);
4874 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
4875 ASSERT(list_empty(&log
->l_buf_cancel_table
[i
]));
4879 kmem_free(log
->l_buf_cancel_table
);
4880 log
->l_buf_cancel_table
= NULL
;
4886 * Do the actual recovery
4891 xfs_daddr_t head_blk
,
4892 xfs_daddr_t tail_blk
)
4899 * First replay the images in the log.
4901 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
4906 * If IO errors happened during recovery, bail out.
4908 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
4913 * We now update the tail_lsn since much of the recovery has completed
4914 * and there may be space available to use. If there were no extent
4915 * or iunlinks, we can free up the entire log and set the tail_lsn to
4916 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4917 * lsn of the last known good LR on disk. If there are extent frees
4918 * or iunlinks they will have some entries in the AIL; so we look at
4919 * the AIL to determine how to set the tail_lsn.
4921 xlog_assign_tail_lsn(log
->l_mp
);
4924 * Now that we've finished replaying all buffer and inode
4925 * updates, re-read in the superblock and reverify it.
4927 bp
= xfs_getsb(log
->l_mp
, 0);
4928 bp
->b_flags
&= ~(XBF_DONE
| XBF_ASYNC
);
4929 ASSERT(!(bp
->b_flags
& XBF_WRITE
));
4930 bp
->b_flags
|= XBF_READ
;
4931 bp
->b_ops
= &xfs_sb_buf_ops
;
4933 error
= xfs_buf_submit_wait(bp
);
4935 if (!XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
4936 xfs_buf_ioerror_alert(bp
, __func__
);
4943 /* Convert superblock from on-disk format */
4944 sbp
= &log
->l_mp
->m_sb
;
4945 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
4946 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
4947 ASSERT(xfs_sb_good_version(sbp
));
4948 xfs_reinit_percpu_counters(log
->l_mp
);
4953 xlog_recover_check_summary(log
);
4955 /* Normal transactions can now occur */
4956 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
4961 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4963 * Return error or zero.
4969 xfs_daddr_t head_blk
, tail_blk
;
4972 /* find the tail of the log */
4973 error
= xlog_find_tail(log
, &head_blk
, &tail_blk
);
4978 * The superblock was read before the log was available and thus the LSN
4979 * could not be verified. Check the superblock LSN against the current
4980 * LSN now that it's known.
4982 if (xfs_sb_version_hascrc(&log
->l_mp
->m_sb
) &&
4983 !xfs_log_check_lsn(log
->l_mp
, log
->l_mp
->m_sb
.sb_lsn
))
4986 if (tail_blk
!= head_blk
) {
4987 /* There used to be a comment here:
4989 * disallow recovery on read-only mounts. note -- mount
4990 * checks for ENOSPC and turns it into an intelligent
4992 * ...but this is no longer true. Now, unless you specify
4993 * NORECOVERY (in which case this function would never be
4994 * called), we just go ahead and recover. We do this all
4995 * under the vfs layer, so we can get away with it unless
4996 * the device itself is read-only, in which case we fail.
4998 if ((error
= xfs_dev_is_read_only(log
->l_mp
, "recovery"))) {
5003 * Version 5 superblock log feature mask validation. We know the
5004 * log is dirty so check if there are any unknown log features
5005 * in what we need to recover. If there are unknown features
5006 * (e.g. unsupported transactions, then simply reject the
5007 * attempt at recovery before touching anything.
5009 if (XFS_SB_VERSION_NUM(&log
->l_mp
->m_sb
) == XFS_SB_VERSION_5
&&
5010 xfs_sb_has_incompat_log_feature(&log
->l_mp
->m_sb
,
5011 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
)) {
5013 "Superblock has unknown incompatible log features (0x%x) enabled.",
5014 (log
->l_mp
->m_sb
.sb_features_log_incompat
&
5015 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
));
5017 "The log can not be fully and/or safely recovered by this kernel.");
5019 "Please recover the log on a kernel that supports the unknown features.");
5024 * Delay log recovery if the debug hook is set. This is debug
5025 * instrumention to coordinate simulation of I/O failures with
5028 if (xfs_globals
.log_recovery_delay
) {
5029 xfs_notice(log
->l_mp
,
5030 "Delaying log recovery for %d seconds.",
5031 xfs_globals
.log_recovery_delay
);
5032 msleep(xfs_globals
.log_recovery_delay
* 1000);
5035 xfs_notice(log
->l_mp
, "Starting recovery (logdev: %s)",
5036 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
5039 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
5040 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
5046 * In the first part of recovery we replay inodes and buffers and build
5047 * up the list of extent free items which need to be processed. Here
5048 * we process the extent free items and clean up the on disk unlinked
5049 * inode lists. This is separated from the first part of recovery so
5050 * that the root and real-time bitmap inodes can be read in from disk in
5051 * between the two stages. This is necessary so that we can free space
5052 * in the real-time portion of the file system.
5055 xlog_recover_finish(
5059 * Now we're ready to do the transactions needed for the
5060 * rest of recovery. Start with completing all the extent
5061 * free intent records and then process the unlinked inode
5062 * lists. At this point, we essentially run in normal mode
5063 * except that we're still performing recovery actions
5064 * rather than accepting new requests.
5066 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
5068 error
= xlog_recover_process_efis(log
);
5070 xfs_alert(log
->l_mp
, "Failed to recover EFIs");
5074 * Sync the log to get all the EFIs out of the AIL.
5075 * This isn't absolutely necessary, but it helps in
5076 * case the unlink transactions would have problems
5077 * pushing the EFIs out of the way.
5079 xfs_log_force(log
->l_mp
, XFS_LOG_SYNC
);
5081 xlog_recover_process_iunlinks(log
);
5083 xlog_recover_check_summary(log
);
5085 xfs_notice(log
->l_mp
, "Ending recovery (logdev: %s)",
5086 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
5088 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
5090 xfs_info(log
->l_mp
, "Ending clean mount");
5096 xlog_recover_cancel(
5101 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
)
5102 error
= xlog_recover_cancel_efis(log
);
5109 * Read all of the agf and agi counters and check that they
5110 * are consistent with the superblock counters.
5113 xlog_recover_check_summary(
5120 xfs_agnumber_t agno
;
5121 __uint64_t freeblks
;
5131 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
5132 error
= xfs_read_agf(mp
, NULL
, agno
, 0, &agfbp
);
5134 xfs_alert(mp
, "%s agf read failed agno %d error %d",
5135 __func__
, agno
, error
);
5137 agfp
= XFS_BUF_TO_AGF(agfbp
);
5138 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
5139 be32_to_cpu(agfp
->agf_flcount
);
5140 xfs_buf_relse(agfbp
);
5143 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
5145 xfs_alert(mp
, "%s agi read failed agno %d error %d",
5146 __func__
, agno
, error
);
5148 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agibp
);
5150 itotal
+= be32_to_cpu(agi
->agi_count
);
5151 ifree
+= be32_to_cpu(agi
->agi_freecount
);
5152 xfs_buf_relse(agibp
);