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
);
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
);
278 XFS_BUF_ZEROFLAGS(bp
);
281 bp
->b_io_length
= nbblks
;
284 error
= xfs_bwrite(bp
);
286 xfs_buf_ioerror_alert(bp
, __func__
);
293 * dump debug superblock and log record information
296 xlog_header_check_dump(
298 xlog_rec_header_t
*head
)
300 xfs_debug(mp
, "%s: SB : uuid = %pU, fmt = %d",
301 __func__
, &mp
->m_sb
.sb_uuid
, XLOG_FMT
);
302 xfs_debug(mp
, " log : uuid = %pU, fmt = %d",
303 &head
->h_fs_uuid
, be32_to_cpu(head
->h_fmt
));
306 #define xlog_header_check_dump(mp, head)
310 * check log record header for recovery
313 xlog_header_check_recover(
315 xlog_rec_header_t
*head
)
317 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
320 * IRIX doesn't write the h_fmt field and leaves it zeroed
321 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
322 * a dirty log created in IRIX.
324 if (unlikely(head
->h_fmt
!= cpu_to_be32(XLOG_FMT
))) {
326 "dirty log written in incompatible format - can't recover");
327 xlog_header_check_dump(mp
, head
);
328 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
329 XFS_ERRLEVEL_HIGH
, mp
);
330 return -EFSCORRUPTED
;
331 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
333 "dirty log entry has mismatched uuid - can't recover");
334 xlog_header_check_dump(mp
, head
);
335 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
336 XFS_ERRLEVEL_HIGH
, mp
);
337 return -EFSCORRUPTED
;
343 * read the head block of the log and check the header
346 xlog_header_check_mount(
348 xlog_rec_header_t
*head
)
350 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
352 if (uuid_is_nil(&head
->h_fs_uuid
)) {
354 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
355 * h_fs_uuid is nil, we assume this log was last mounted
356 * by IRIX and continue.
358 xfs_warn(mp
, "nil uuid in log - IRIX style log");
359 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
360 xfs_warn(mp
, "log has mismatched uuid - can't recover");
361 xlog_header_check_dump(mp
, head
);
362 XFS_ERROR_REPORT("xlog_header_check_mount",
363 XFS_ERRLEVEL_HIGH
, mp
);
364 return -EFSCORRUPTED
;
375 * We're not going to bother about retrying
376 * this during recovery. One strike!
378 if (!XFS_FORCED_SHUTDOWN(bp
->b_target
->bt_mount
)) {
379 xfs_buf_ioerror_alert(bp
, __func__
);
380 xfs_force_shutdown(bp
->b_target
->bt_mount
,
381 SHUTDOWN_META_IO_ERROR
);
389 * This routine finds (to an approximation) the first block in the physical
390 * log which contains the given cycle. It uses a binary search algorithm.
391 * Note that the algorithm can not be perfect because the disk will not
392 * necessarily be perfect.
395 xlog_find_cycle_start(
398 xfs_daddr_t first_blk
,
399 xfs_daddr_t
*last_blk
,
409 mid_blk
= BLK_AVG(first_blk
, end_blk
);
410 while (mid_blk
!= first_blk
&& mid_blk
!= end_blk
) {
411 error
= xlog_bread(log
, mid_blk
, 1, bp
, &offset
);
414 mid_cycle
= xlog_get_cycle(offset
);
415 if (mid_cycle
== cycle
)
416 end_blk
= mid_blk
; /* last_half_cycle == mid_cycle */
418 first_blk
= mid_blk
; /* first_half_cycle == mid_cycle */
419 mid_blk
= BLK_AVG(first_blk
, end_blk
);
421 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == end_blk
) ||
422 (mid_blk
== end_blk
&& mid_blk
-1 == first_blk
));
430 * Check that a range of blocks does not contain stop_on_cycle_no.
431 * Fill in *new_blk with the block offset where such a block is
432 * found, or with -1 (an invalid block number) if there is no such
433 * block in the range. The scan needs to occur from front to back
434 * and the pointer into the region must be updated since a later
435 * routine will need to perform another test.
438 xlog_find_verify_cycle(
440 xfs_daddr_t start_blk
,
442 uint stop_on_cycle_no
,
443 xfs_daddr_t
*new_blk
)
453 * Greedily allocate a buffer big enough to handle the full
454 * range of basic blocks we'll be examining. If that fails,
455 * try a smaller size. We need to be able to read at least
456 * a log sector, or we're out of luck.
458 bufblks
= 1 << ffs(nbblks
);
459 while (bufblks
> log
->l_logBBsize
)
461 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
463 if (bufblks
< log
->l_sectBBsize
)
467 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
470 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
472 error
= xlog_bread(log
, i
, bcount
, bp
, &buf
);
476 for (j
= 0; j
< bcount
; j
++) {
477 cycle
= xlog_get_cycle(buf
);
478 if (cycle
== stop_on_cycle_no
) {
495 * Potentially backup over partial log record write.
497 * In the typical case, last_blk is the number of the block directly after
498 * a good log record. Therefore, we subtract one to get the block number
499 * of the last block in the given buffer. extra_bblks contains the number
500 * of blocks we would have read on a previous read. This happens when the
501 * last log record is split over the end of the physical log.
503 * extra_bblks is the number of blocks potentially verified on a previous
504 * call to this routine.
507 xlog_find_verify_log_record(
509 xfs_daddr_t start_blk
,
510 xfs_daddr_t
*last_blk
,
516 xlog_rec_header_t
*head
= NULL
;
519 int num_blks
= *last_blk
- start_blk
;
522 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
524 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
525 if (!(bp
= xlog_get_bp(log
, 1)))
529 error
= xlog_bread(log
, start_blk
, num_blks
, bp
, &offset
);
532 offset
+= ((num_blks
- 1) << BBSHIFT
);
535 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
537 /* valid log record not found */
539 "Log inconsistent (didn't find previous header)");
546 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
551 head
= (xlog_rec_header_t
*)offset
;
553 if (head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))
561 * We hit the beginning of the physical log & still no header. Return
562 * to caller. If caller can handle a return of -1, then this routine
563 * will be called again for the end of the physical log.
571 * We have the final block of the good log (the first block
572 * of the log record _before_ the head. So we check the uuid.
574 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
578 * We may have found a log record header before we expected one.
579 * last_blk will be the 1st block # with a given cycle #. We may end
580 * up reading an entire log record. In this case, we don't want to
581 * reset last_blk. Only when last_blk points in the middle of a log
582 * record do we update last_blk.
584 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
585 uint h_size
= be32_to_cpu(head
->h_size
);
587 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
588 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
594 if (*last_blk
- i
+ extra_bblks
!=
595 BTOBB(be32_to_cpu(head
->h_len
)) + xhdrs
)
604 * Head is defined to be the point of the log where the next log write
605 * could go. This means that incomplete LR writes at the end are
606 * eliminated when calculating the head. We aren't guaranteed that previous
607 * LR have complete transactions. We only know that a cycle number of
608 * current cycle number -1 won't be present in the log if we start writing
609 * from our current block number.
611 * last_blk contains the block number of the first block with a given
614 * Return: zero if normal, non-zero if error.
619 xfs_daddr_t
*return_head_blk
)
623 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
625 uint first_half_cycle
, last_half_cycle
;
627 int error
, log_bbnum
= log
->l_logBBsize
;
629 /* Is the end of the log device zeroed? */
630 error
= xlog_find_zeroed(log
, &first_blk
);
632 xfs_warn(log
->l_mp
, "empty log check failed");
636 *return_head_blk
= first_blk
;
638 /* Is the whole lot zeroed? */
640 /* Linux XFS shouldn't generate totally zeroed logs -
641 * mkfs etc write a dummy unmount record to a fresh
642 * log so we can store the uuid in there
644 xfs_warn(log
->l_mp
, "totally zeroed log");
650 first_blk
= 0; /* get cycle # of 1st block */
651 bp
= xlog_get_bp(log
, 1);
655 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
659 first_half_cycle
= xlog_get_cycle(offset
);
661 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
662 error
= xlog_bread(log
, last_blk
, 1, bp
, &offset
);
666 last_half_cycle
= xlog_get_cycle(offset
);
667 ASSERT(last_half_cycle
!= 0);
670 * If the 1st half cycle number is equal to the last half cycle number,
671 * then the entire log is stamped with the same cycle number. In this
672 * case, head_blk can't be set to zero (which makes sense). The below
673 * math doesn't work out properly with head_blk equal to zero. Instead,
674 * we set it to log_bbnum which is an invalid block number, but this
675 * value makes the math correct. If head_blk doesn't changed through
676 * all the tests below, *head_blk is set to zero at the very end rather
677 * than log_bbnum. In a sense, log_bbnum and zero are the same block
678 * in a circular file.
680 if (first_half_cycle
== last_half_cycle
) {
682 * In this case we believe that the entire log should have
683 * cycle number last_half_cycle. We need to scan backwards
684 * from the end verifying that there are no holes still
685 * containing last_half_cycle - 1. If we find such a hole,
686 * then the start of that hole will be the new head. The
687 * simple case looks like
688 * x | x ... | x - 1 | x
689 * Another case that fits this picture would be
690 * x | x + 1 | x ... | x
691 * In this case the head really is somewhere at the end of the
692 * log, as one of the latest writes at the beginning was
695 * x | x + 1 | x ... | x - 1 | x
696 * This is really the combination of the above two cases, and
697 * the head has to end up at the start of the x-1 hole at the
700 * In the 256k log case, we will read from the beginning to the
701 * end of the log and search for cycle numbers equal to x-1.
702 * We don't worry about the x+1 blocks that we encounter,
703 * because we know that they cannot be the head since the log
706 head_blk
= log_bbnum
;
707 stop_on_cycle
= last_half_cycle
- 1;
710 * In this case we want to find the first block with cycle
711 * number matching last_half_cycle. We expect the log to be
713 * x + 1 ... | x ... | x
714 * The first block with cycle number x (last_half_cycle) will
715 * be where the new head belongs. First we do a binary search
716 * for the first occurrence of last_half_cycle. The binary
717 * search may not be totally accurate, so then we scan back
718 * from there looking for occurrences of last_half_cycle before
719 * us. If that backwards scan wraps around the beginning of
720 * the log, then we look for occurrences of last_half_cycle - 1
721 * at the end of the log. The cases we're looking for look
723 * v binary search stopped here
724 * x + 1 ... | x | x + 1 | x ... | x
725 * ^ but we want to locate this spot
727 * <---------> less than scan distance
728 * x + 1 ... | x ... | x - 1 | x
729 * ^ we want to locate this spot
731 stop_on_cycle
= last_half_cycle
;
732 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
733 &head_blk
, last_half_cycle
)))
738 * Now validate the answer. Scan back some number of maximum possible
739 * blocks and make sure each one has the expected cycle number. The
740 * maximum is determined by the total possible amount of buffering
741 * in the in-core log. The following number can be made tighter if
742 * we actually look at the block size of the filesystem.
744 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
745 if (head_blk
>= num_scan_bblks
) {
747 * We are guaranteed that the entire check can be performed
750 start_blk
= head_blk
- num_scan_bblks
;
751 if ((error
= xlog_find_verify_cycle(log
,
752 start_blk
, num_scan_bblks
,
753 stop_on_cycle
, &new_blk
)))
757 } else { /* need to read 2 parts of log */
759 * We are going to scan backwards in the log in two parts.
760 * First we scan the physical end of the log. In this part
761 * of the log, we are looking for blocks with cycle number
762 * last_half_cycle - 1.
763 * If we find one, then we know that the log starts there, as
764 * we've found a hole that didn't get written in going around
765 * the end of the physical log. The simple case for this is
766 * x + 1 ... | x ... | x - 1 | x
767 * <---------> less than scan distance
768 * If all of the blocks at the end of the log have cycle number
769 * last_half_cycle, then we check the blocks at the start of
770 * the log looking for occurrences of last_half_cycle. If we
771 * find one, then our current estimate for the location of the
772 * first occurrence of last_half_cycle is wrong and we move
773 * back to the hole we've found. This case looks like
774 * x + 1 ... | x | x + 1 | x ...
775 * ^ binary search stopped here
776 * Another case we need to handle that only occurs in 256k
778 * x + 1 ... | x ... | x+1 | x ...
779 * ^ binary search stops here
780 * In a 256k log, the scan at the end of the log will see the
781 * x + 1 blocks. We need to skip past those since that is
782 * certainly not the head of the log. By searching for
783 * last_half_cycle-1 we accomplish that.
785 ASSERT(head_blk
<= INT_MAX
&&
786 (xfs_daddr_t
) num_scan_bblks
>= head_blk
);
787 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
788 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
789 num_scan_bblks
- (int)head_blk
,
790 (stop_on_cycle
- 1), &new_blk
)))
798 * Scan beginning of log now. The last part of the physical
799 * log is good. This scan needs to verify that it doesn't find
800 * the last_half_cycle.
803 ASSERT(head_blk
<= INT_MAX
);
804 if ((error
= xlog_find_verify_cycle(log
,
805 start_blk
, (int)head_blk
,
806 stop_on_cycle
, &new_blk
)))
814 * Now we need to make sure head_blk is not pointing to a block in
815 * the middle of a log record.
817 num_scan_bblks
= XLOG_REC_SHIFT(log
);
818 if (head_blk
>= num_scan_bblks
) {
819 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
821 /* start ptr at last block ptr before head_blk */
822 error
= xlog_find_verify_log_record(log
, start_blk
, &head_blk
, 0);
829 ASSERT(head_blk
<= INT_MAX
);
830 error
= xlog_find_verify_log_record(log
, start_blk
, &head_blk
, 0);
834 /* We hit the beginning of the log during our search */
835 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
837 ASSERT(start_blk
<= INT_MAX
&&
838 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
839 ASSERT(head_blk
<= INT_MAX
);
840 error
= xlog_find_verify_log_record(log
, start_blk
,
841 &new_blk
, (int)head_blk
);
846 if (new_blk
!= log_bbnum
)
853 if (head_blk
== log_bbnum
)
854 *return_head_blk
= 0;
856 *return_head_blk
= head_blk
;
858 * When returning here, we have a good block number. Bad block
859 * means that during a previous crash, we didn't have a clean break
860 * from cycle number N to cycle number N-1. In this case, we need
861 * to find the first block with cycle number N-1.
869 xfs_warn(log
->l_mp
, "failed to find log head");
874 * Seek backwards in the log for log record headers.
876 * Given a starting log block, walk backwards until we find the provided number
877 * of records or hit the provided tail block. The return value is the number of
878 * records encountered or a negative error code. The log block and buffer
879 * pointer of the last record seen are returned in rblk and rhead respectively.
882 xlog_rseek_logrec_hdr(
884 xfs_daddr_t head_blk
,
885 xfs_daddr_t tail_blk
,
889 struct xlog_rec_header
**rhead
,
901 * Walk backwards from the head block until we hit the tail or the first
904 end_blk
= head_blk
> tail_blk
? tail_blk
: 0;
905 for (i
= (int) head_blk
- 1; i
>= end_blk
; i
--) {
906 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
910 if (*(__be32
*) offset
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
912 *rhead
= (struct xlog_rec_header
*) offset
;
913 if (++found
== count
)
919 * If we haven't hit the tail block or the log record header count,
920 * start looking again from the end of the physical log. Note that
921 * callers can pass head == tail if the tail is not yet known.
923 if (tail_blk
>= head_blk
&& found
!= count
) {
924 for (i
= log
->l_logBBsize
- 1; i
>= (int) tail_blk
; i
--) {
925 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
929 if (*(__be32
*)offset
==
930 cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
933 *rhead
= (struct xlog_rec_header
*) offset
;
934 if (++found
== count
)
947 * Seek forward in the log for log record headers.
949 * Given head and tail blocks, walk forward from the tail block until we find
950 * the provided number of records or hit the head block. The return value is the
951 * number of records encountered or a negative error code. The log block and
952 * buffer pointer of the last record seen are returned in rblk and rhead
956 xlog_seek_logrec_hdr(
958 xfs_daddr_t head_blk
,
959 xfs_daddr_t tail_blk
,
963 struct xlog_rec_header
**rhead
,
975 * Walk forward from the tail block until we hit the head or the last
978 end_blk
= head_blk
> tail_blk
? head_blk
: log
->l_logBBsize
- 1;
979 for (i
= (int) tail_blk
; i
<= end_blk
; i
++) {
980 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
984 if (*(__be32
*) offset
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
986 *rhead
= (struct xlog_rec_header
*) offset
;
987 if (++found
== count
)
993 * If we haven't hit the head block or the log record header count,
994 * start looking again from the start of the physical log.
996 if (tail_blk
> head_blk
&& found
!= count
) {
997 for (i
= 0; i
< (int) head_blk
; i
++) {
998 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
1002 if (*(__be32
*)offset
==
1003 cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
1006 *rhead
= (struct xlog_rec_header
*) offset
;
1007 if (++found
== count
)
1020 * Check the log tail for torn writes. This is required when torn writes are
1021 * detected at the head and the head had to be walked back to a previous record.
1022 * The tail of the previous record must now be verified to ensure the torn
1023 * writes didn't corrupt the previous tail.
1025 * Return an error if CRC verification fails as recovery cannot proceed.
1030 xfs_daddr_t head_blk
,
1031 xfs_daddr_t tail_blk
)
1033 struct xlog_rec_header
*thead
;
1035 xfs_daddr_t first_bad
;
1039 xfs_daddr_t tmp_head
;
1041 bp
= xlog_get_bp(log
, 1);
1046 * Seek XLOG_MAX_ICLOGS + 1 records past the current tail record to get
1047 * a temporary head block that points after the last possible
1048 * concurrently written record of the tail.
1050 count
= xlog_seek_logrec_hdr(log
, head_blk
, tail_blk
,
1051 XLOG_MAX_ICLOGS
+ 1, bp
, &tmp_head
, &thead
,
1059 * If the call above didn't find XLOG_MAX_ICLOGS + 1 records, we ran
1060 * into the actual log head. tmp_head points to the start of the record
1061 * so update it to the actual head block.
1063 if (count
< XLOG_MAX_ICLOGS
+ 1)
1064 tmp_head
= head_blk
;
1067 * We now have a tail and temporary head block that covers at least
1068 * XLOG_MAX_ICLOGS records from the tail. We need to verify that these
1069 * records were completely written. Run a CRC verification pass from
1070 * tail to head and return the result.
1072 error
= xlog_do_recovery_pass(log
, tmp_head
, tail_blk
,
1073 XLOG_RECOVER_CRCPASS
, &first_bad
);
1081 * Detect and trim torn writes from the head of the log.
1083 * Storage without sector atomicity guarantees can result in torn writes in the
1084 * log in the event of a crash. Our only means to detect this scenario is via
1085 * CRC verification. While we can't always be certain that CRC verification
1086 * failure is due to a torn write vs. an unrelated corruption, we do know that
1087 * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
1088 * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
1089 * the log and treat failures in this range as torn writes as a matter of
1090 * policy. In the event of CRC failure, the head is walked back to the last good
1091 * record in the log and the tail is updated from that record and verified.
1096 xfs_daddr_t
*head_blk
, /* in/out: unverified head */
1097 xfs_daddr_t
*tail_blk
, /* out: tail block */
1099 xfs_daddr_t
*rhead_blk
, /* start blk of last record */
1100 struct xlog_rec_header
**rhead
, /* ptr to last record */
1101 bool *wrapped
) /* last rec. wraps phys. log */
1103 struct xlog_rec_header
*tmp_rhead
;
1104 struct xfs_buf
*tmp_bp
;
1105 xfs_daddr_t first_bad
;
1106 xfs_daddr_t tmp_rhead_blk
;
1112 * Check the head of the log for torn writes. Search backwards from the
1113 * head until we hit the tail or the maximum number of log record I/Os
1114 * that could have been in flight at one time. Use a temporary buffer so
1115 * we don't trash the rhead/bp pointers from the caller.
1117 tmp_bp
= xlog_get_bp(log
, 1);
1120 error
= xlog_rseek_logrec_hdr(log
, *head_blk
, *tail_blk
,
1121 XLOG_MAX_ICLOGS
, tmp_bp
, &tmp_rhead_blk
,
1122 &tmp_rhead
, &tmp_wrapped
);
1123 xlog_put_bp(tmp_bp
);
1128 * Now run a CRC verification pass over the records starting at the
1129 * block found above to the current head. If a CRC failure occurs, the
1130 * log block of the first bad record is saved in first_bad.
1132 error
= xlog_do_recovery_pass(log
, *head_blk
, tmp_rhead_blk
,
1133 XLOG_RECOVER_CRCPASS
, &first_bad
);
1134 if (error
== -EFSBADCRC
) {
1136 * We've hit a potential torn write. Reset the error and warn
1141 "Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.",
1142 first_bad
, *head_blk
);
1145 * Get the header block and buffer pointer for the last good
1146 * record before the bad record.
1148 * Note that xlog_find_tail() clears the blocks at the new head
1149 * (i.e., the records with invalid CRC) if the cycle number
1150 * matches the the current cycle.
1152 found
= xlog_rseek_logrec_hdr(log
, first_bad
, *tail_blk
, 1, bp
,
1153 rhead_blk
, rhead
, wrapped
);
1156 if (found
== 0) /* XXX: right thing to do here? */
1160 * Reset the head block to the starting block of the first bad
1161 * log record and set the tail block based on the last good
1164 * Bail out if the updated head/tail match as this indicates
1165 * possible corruption outside of the acceptable
1166 * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
1168 *head_blk
= first_bad
;
1169 *tail_blk
= BLOCK_LSN(be64_to_cpu((*rhead
)->h_tail_lsn
));
1170 if (*head_blk
== *tail_blk
) {
1176 * Now verify the tail based on the updated head. This is
1177 * required because the torn writes trimmed from the head could
1178 * have been written over the tail of a previous record. Return
1179 * any errors since recovery cannot proceed if the tail is
1182 * XXX: This leaves a gap in truly robust protection from torn
1183 * writes in the log. If the head is behind the tail, the tail
1184 * pushes forward to create some space and then a crash occurs
1185 * causing the writes into the previous record's tail region to
1186 * tear, log recovery isn't able to recover.
1188 * How likely is this to occur? If possible, can we do something
1189 * more intelligent here? Is it safe to push the tail forward if
1190 * we can determine that the tail is within the range of the
1191 * torn write (e.g., the kernel can only overwrite the tail if
1192 * it has actually been pushed forward)? Alternatively, could we
1193 * somehow prevent this condition at runtime?
1195 error
= xlog_verify_tail(log
, *head_blk
, *tail_blk
);
1202 * Check whether the head of the log points to an unmount record. In other
1203 * words, determine whether the log is clean. If so, update the in-core state
1207 xlog_check_unmount_rec(
1209 xfs_daddr_t
*head_blk
,
1210 xfs_daddr_t
*tail_blk
,
1211 struct xlog_rec_header
*rhead
,
1212 xfs_daddr_t rhead_blk
,
1216 struct xlog_op_header
*op_head
;
1217 xfs_daddr_t umount_data_blk
;
1218 xfs_daddr_t after_umount_blk
;
1226 * Look for unmount record. If we find it, then we know there was a
1227 * clean unmount. Since 'i' could be the last block in the physical
1228 * log, we convert to a log block before comparing to the head_blk.
1230 * Save the current tail lsn to use to pass to xlog_clear_stale_blocks()
1231 * below. We won't want to clear the unmount record if there is one, so
1232 * we pass the lsn of the unmount record rather than the block after it.
1234 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
1235 int h_size
= be32_to_cpu(rhead
->h_size
);
1236 int h_version
= be32_to_cpu(rhead
->h_version
);
1238 if ((h_version
& XLOG_VERSION_2
) &&
1239 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
1240 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
1241 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
1249 after_umount_blk
= rhead_blk
+ hblks
+ BTOBB(be32_to_cpu(rhead
->h_len
));
1250 after_umount_blk
= do_mod(after_umount_blk
, log
->l_logBBsize
);
1251 if (*head_blk
== after_umount_blk
&&
1252 be32_to_cpu(rhead
->h_num_logops
) == 1) {
1253 umount_data_blk
= rhead_blk
+ hblks
;
1254 umount_data_blk
= do_mod(umount_data_blk
, log
->l_logBBsize
);
1255 error
= xlog_bread(log
, umount_data_blk
, 1, bp
, &offset
);
1259 op_head
= (struct xlog_op_header
*)offset
;
1260 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
1262 * Set tail and last sync so that newly written log
1263 * records will point recovery to after the current
1266 xlog_assign_atomic_lsn(&log
->l_tail_lsn
,
1267 log
->l_curr_cycle
, after_umount_blk
);
1268 xlog_assign_atomic_lsn(&log
->l_last_sync_lsn
,
1269 log
->l_curr_cycle
, after_umount_blk
);
1270 *tail_blk
= after_umount_blk
;
1282 xfs_daddr_t head_blk
,
1283 struct xlog_rec_header
*rhead
,
1284 xfs_daddr_t rhead_blk
,
1288 * Reset log values according to the state of the log when we
1289 * crashed. In the case where head_blk == 0, we bump curr_cycle
1290 * one because the next write starts a new cycle rather than
1291 * continuing the cycle of the last good log record. At this
1292 * point we have guaranteed that all partial log records have been
1293 * accounted for. Therefore, we know that the last good log record
1294 * written was complete and ended exactly on the end boundary
1295 * of the physical log.
1297 log
->l_prev_block
= rhead_blk
;
1298 log
->l_curr_block
= (int)head_blk
;
1299 log
->l_curr_cycle
= be32_to_cpu(rhead
->h_cycle
);
1301 log
->l_curr_cycle
++;
1302 atomic64_set(&log
->l_tail_lsn
, be64_to_cpu(rhead
->h_tail_lsn
));
1303 atomic64_set(&log
->l_last_sync_lsn
, be64_to_cpu(rhead
->h_lsn
));
1304 xlog_assign_grant_head(&log
->l_reserve_head
.grant
, log
->l_curr_cycle
,
1305 BBTOB(log
->l_curr_block
));
1306 xlog_assign_grant_head(&log
->l_write_head
.grant
, log
->l_curr_cycle
,
1307 BBTOB(log
->l_curr_block
));
1311 * Find the sync block number or the tail of the log.
1313 * This will be the block number of the last record to have its
1314 * associated buffers synced to disk. Every log record header has
1315 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
1316 * to get a sync block number. The only concern is to figure out which
1317 * log record header to believe.
1319 * The following algorithm uses the log record header with the largest
1320 * lsn. The entire log record does not need to be valid. We only care
1321 * that the header is valid.
1323 * We could speed up search by using current head_blk buffer, but it is not
1329 xfs_daddr_t
*head_blk
,
1330 xfs_daddr_t
*tail_blk
)
1332 xlog_rec_header_t
*rhead
;
1333 char *offset
= NULL
;
1336 xfs_daddr_t rhead_blk
;
1338 bool wrapped
= false;
1342 * Find previous log record
1344 if ((error
= xlog_find_head(log
, head_blk
)))
1346 ASSERT(*head_blk
< INT_MAX
);
1348 bp
= xlog_get_bp(log
, 1);
1351 if (*head_blk
== 0) { /* special case */
1352 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1356 if (xlog_get_cycle(offset
) == 0) {
1358 /* leave all other log inited values alone */
1364 * Search backwards through the log looking for the log record header
1365 * block. This wraps all the way back around to the head so something is
1366 * seriously wrong if we can't find it.
1368 error
= xlog_rseek_logrec_hdr(log
, *head_blk
, *head_blk
, 1, bp
,
1369 &rhead_blk
, &rhead
, &wrapped
);
1373 xfs_warn(log
->l_mp
, "%s: couldn't find sync record", __func__
);
1376 *tail_blk
= BLOCK_LSN(be64_to_cpu(rhead
->h_tail_lsn
));
1379 * Set the log state based on the current head record.
1381 xlog_set_state(log
, *head_blk
, rhead
, rhead_blk
, wrapped
);
1382 tail_lsn
= atomic64_read(&log
->l_tail_lsn
);
1385 * Look for an unmount record at the head of the log. This sets the log
1386 * state to determine whether recovery is necessary.
1388 error
= xlog_check_unmount_rec(log
, head_blk
, tail_blk
, rhead
,
1389 rhead_blk
, bp
, &clean
);
1394 * Verify the log head if the log is not clean (e.g., we have anything
1395 * but an unmount record at the head). This uses CRC verification to
1396 * detect and trim torn writes. If discovered, CRC failures are
1397 * considered torn writes and the log head is trimmed accordingly.
1399 * Note that we can only run CRC verification when the log is dirty
1400 * because there's no guarantee that the log data behind an unmount
1401 * record is compatible with the current architecture.
1404 xfs_daddr_t orig_head
= *head_blk
;
1406 error
= xlog_verify_head(log
, head_blk
, tail_blk
, bp
,
1407 &rhead_blk
, &rhead
, &wrapped
);
1411 /* update in-core state again if the head changed */
1412 if (*head_blk
!= orig_head
) {
1413 xlog_set_state(log
, *head_blk
, rhead
, rhead_blk
,
1415 tail_lsn
= atomic64_read(&log
->l_tail_lsn
);
1416 error
= xlog_check_unmount_rec(log
, head_blk
, tail_blk
,
1417 rhead
, rhead_blk
, bp
,
1425 * Note that the unmount was clean. If the unmount was not clean, we
1426 * need to know this to rebuild the superblock counters from the perag
1427 * headers if we have a filesystem using non-persistent counters.
1430 log
->l_mp
->m_flags
|= XFS_MOUNT_WAS_CLEAN
;
1433 * Make sure that there are no blocks in front of the head
1434 * with the same cycle number as the head. This can happen
1435 * because we allow multiple outstanding log writes concurrently,
1436 * and the later writes might make it out before earlier ones.
1438 * We use the lsn from before modifying it so that we'll never
1439 * overwrite the unmount record after a clean unmount.
1441 * Do this only if we are going to recover the filesystem
1443 * NOTE: This used to say "if (!readonly)"
1444 * However on Linux, we can & do recover a read-only filesystem.
1445 * We only skip recovery if NORECOVERY is specified on mount,
1446 * in which case we would not be here.
1448 * But... if the -device- itself is readonly, just skip this.
1449 * We can't recover this device anyway, so it won't matter.
1451 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
))
1452 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
1458 xfs_warn(log
->l_mp
, "failed to locate log tail");
1463 * Is the log zeroed at all?
1465 * The last binary search should be changed to perform an X block read
1466 * once X becomes small enough. You can then search linearly through
1467 * the X blocks. This will cut down on the number of reads we need to do.
1469 * If the log is partially zeroed, this routine will pass back the blkno
1470 * of the first block with cycle number 0. It won't have a complete LR
1474 * 0 => the log is completely written to
1475 * 1 => use *blk_no as the first block of the log
1476 * <0 => error has occurred
1481 xfs_daddr_t
*blk_no
)
1485 uint first_cycle
, last_cycle
;
1486 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1487 xfs_daddr_t num_scan_bblks
;
1488 int error
, log_bbnum
= log
->l_logBBsize
;
1492 /* check totally zeroed log */
1493 bp
= xlog_get_bp(log
, 1);
1496 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1500 first_cycle
= xlog_get_cycle(offset
);
1501 if (first_cycle
== 0) { /* completely zeroed log */
1507 /* check partially zeroed log */
1508 error
= xlog_bread(log
, log_bbnum
-1, 1, bp
, &offset
);
1512 last_cycle
= xlog_get_cycle(offset
);
1513 if (last_cycle
!= 0) { /* log completely written to */
1516 } else if (first_cycle
!= 1) {
1518 * If the cycle of the last block is zero, the cycle of
1519 * the first block must be 1. If it's not, maybe we're
1520 * not looking at a log... Bail out.
1523 "Log inconsistent or not a log (last==0, first!=1)");
1528 /* we have a partially zeroed log */
1529 last_blk
= log_bbnum
-1;
1530 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1534 * Validate the answer. Because there is no way to guarantee that
1535 * the entire log is made up of log records which are the same size,
1536 * we scan over the defined maximum blocks. At this point, the maximum
1537 * is not chosen to mean anything special. XXXmiken
1539 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1540 ASSERT(num_scan_bblks
<= INT_MAX
);
1542 if (last_blk
< num_scan_bblks
)
1543 num_scan_bblks
= last_blk
;
1544 start_blk
= last_blk
- num_scan_bblks
;
1547 * We search for any instances of cycle number 0 that occur before
1548 * our current estimate of the head. What we're trying to detect is
1549 * 1 ... | 0 | 1 | 0...
1550 * ^ binary search ends here
1552 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1553 (int)num_scan_bblks
, 0, &new_blk
)))
1559 * Potentially backup over partial log record write. We don't need
1560 * to search the end of the log because we know it is zero.
1562 error
= xlog_find_verify_log_record(log
, start_blk
, &last_blk
, 0);
1577 * These are simple subroutines used by xlog_clear_stale_blocks() below
1578 * to initialize a buffer full of empty log record headers and write
1579 * them into the log.
1590 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1592 memset(buf
, 0, BBSIZE
);
1593 recp
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1594 recp
->h_cycle
= cpu_to_be32(cycle
);
1595 recp
->h_version
= cpu_to_be32(
1596 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1597 recp
->h_lsn
= cpu_to_be64(xlog_assign_lsn(cycle
, block
));
1598 recp
->h_tail_lsn
= cpu_to_be64(xlog_assign_lsn(tail_cycle
, tail_block
));
1599 recp
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1600 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1604 xlog_write_log_records(
1615 int sectbb
= log
->l_sectBBsize
;
1616 int end_block
= start_block
+ blocks
;
1622 * Greedily allocate a buffer big enough to handle the full
1623 * range of basic blocks to be written. If that fails, try
1624 * a smaller size. We need to be able to write at least a
1625 * log sector, or we're out of luck.
1627 bufblks
= 1 << ffs(blocks
);
1628 while (bufblks
> log
->l_logBBsize
)
1630 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1632 if (bufblks
< sectbb
)
1636 /* We may need to do a read at the start to fill in part of
1637 * the buffer in the starting sector not covered by the first
1640 balign
= round_down(start_block
, sectbb
);
1641 if (balign
!= start_block
) {
1642 error
= xlog_bread_noalign(log
, start_block
, 1, bp
);
1646 j
= start_block
- balign
;
1649 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1650 int bcount
, endcount
;
1652 bcount
= min(bufblks
, end_block
- start_block
);
1653 endcount
= bcount
- j
;
1655 /* We may need to do a read at the end to fill in part of
1656 * the buffer in the final sector not covered by the write.
1657 * If this is the same sector as the above read, skip it.
1659 ealign
= round_down(end_block
, sectbb
);
1660 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1661 offset
= bp
->b_addr
+ BBTOB(ealign
- start_block
);
1662 error
= xlog_bread_offset(log
, ealign
, sectbb
,
1669 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1670 for (; j
< endcount
; j
++) {
1671 xlog_add_record(log
, offset
, cycle
, i
+j
,
1672 tail_cycle
, tail_block
);
1675 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1678 start_block
+= endcount
;
1688 * This routine is called to blow away any incomplete log writes out
1689 * in front of the log head. We do this so that we won't become confused
1690 * if we come up, write only a little bit more, and then crash again.
1691 * If we leave the partial log records out there, this situation could
1692 * cause us to think those partial writes are valid blocks since they
1693 * have the current cycle number. We get rid of them by overwriting them
1694 * with empty log records with the old cycle number rather than the
1697 * The tail lsn is passed in rather than taken from
1698 * the log so that we will not write over the unmount record after a
1699 * clean unmount in a 512 block log. Doing so would leave the log without
1700 * any valid log records in it until a new one was written. If we crashed
1701 * during that time we would not be able to recover.
1704 xlog_clear_stale_blocks(
1708 int tail_cycle
, head_cycle
;
1709 int tail_block
, head_block
;
1710 int tail_distance
, max_distance
;
1714 tail_cycle
= CYCLE_LSN(tail_lsn
);
1715 tail_block
= BLOCK_LSN(tail_lsn
);
1716 head_cycle
= log
->l_curr_cycle
;
1717 head_block
= log
->l_curr_block
;
1720 * Figure out the distance between the new head of the log
1721 * and the tail. We want to write over any blocks beyond the
1722 * head that we may have written just before the crash, but
1723 * we don't want to overwrite the tail of the log.
1725 if (head_cycle
== tail_cycle
) {
1727 * The tail is behind the head in the physical log,
1728 * so the distance from the head to the tail is the
1729 * distance from the head to the end of the log plus
1730 * the distance from the beginning of the log to the
1733 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1734 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1735 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1736 return -EFSCORRUPTED
;
1738 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1741 * The head is behind the tail in the physical log,
1742 * so the distance from the head to the tail is just
1743 * the tail block minus the head block.
1745 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1746 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1747 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1748 return -EFSCORRUPTED
;
1750 tail_distance
= tail_block
- head_block
;
1754 * If the head is right up against the tail, we can't clear
1757 if (tail_distance
<= 0) {
1758 ASSERT(tail_distance
== 0);
1762 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1764 * Take the smaller of the maximum amount of outstanding I/O
1765 * we could have and the distance to the tail to clear out.
1766 * We take the smaller so that we don't overwrite the tail and
1767 * we don't waste all day writing from the head to the tail
1770 max_distance
= MIN(max_distance
, tail_distance
);
1772 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1774 * We can stomp all the blocks we need to without
1775 * wrapping around the end of the log. Just do it
1776 * in a single write. Use the cycle number of the
1777 * current cycle minus one so that the log will look like:
1780 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1781 head_block
, max_distance
, tail_cycle
,
1787 * We need to wrap around the end of the physical log in
1788 * order to clear all the blocks. Do it in two separate
1789 * I/Os. The first write should be from the head to the
1790 * end of the physical log, and it should use the current
1791 * cycle number minus one just like above.
1793 distance
= log
->l_logBBsize
- head_block
;
1794 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1795 head_block
, distance
, tail_cycle
,
1802 * Now write the blocks at the start of the physical log.
1803 * This writes the remainder of the blocks we want to clear.
1804 * It uses the current cycle number since we're now on the
1805 * same cycle as the head so that we get:
1806 * n ... n ... | n - 1 ...
1807 * ^^^^^ blocks we're writing
1809 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1810 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1811 tail_cycle
, tail_block
);
1819 /******************************************************************************
1821 * Log recover routines
1823 ******************************************************************************
1827 * Sort the log items in the transaction.
1829 * The ordering constraints are defined by the inode allocation and unlink
1830 * behaviour. The rules are:
1832 * 1. Every item is only logged once in a given transaction. Hence it
1833 * represents the last logged state of the item. Hence ordering is
1834 * dependent on the order in which operations need to be performed so
1835 * required initial conditions are always met.
1837 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1838 * there's nothing to replay from them so we can simply cull them
1839 * from the transaction. However, we can't do that until after we've
1840 * replayed all the other items because they may be dependent on the
1841 * cancelled buffer and replaying the cancelled buffer can remove it
1842 * form the cancelled buffer table. Hence they have tobe done last.
1844 * 3. Inode allocation buffers must be replayed before inode items that
1845 * read the buffer and replay changes into it. For filesystems using the
1846 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1847 * treated the same as inode allocation buffers as they create and
1848 * initialise the buffers directly.
1850 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1851 * This ensures that inodes are completely flushed to the inode buffer
1852 * in a "free" state before we remove the unlinked inode list pointer.
1854 * Hence the ordering needs to be inode allocation buffers first, inode items
1855 * second, inode unlink buffers third and cancelled buffers last.
1857 * But there's a problem with that - we can't tell an inode allocation buffer
1858 * apart from a regular buffer, so we can't separate them. We can, however,
1859 * tell an inode unlink buffer from the others, and so we can separate them out
1860 * from all the other buffers and move them to last.
1862 * Hence, 4 lists, in order from head to tail:
1863 * - buffer_list for all buffers except cancelled/inode unlink buffers
1864 * - item_list for all non-buffer items
1865 * - inode_buffer_list for inode unlink buffers
1866 * - cancel_list for the cancelled buffers
1868 * Note that we add objects to the tail of the lists so that first-to-last
1869 * ordering is preserved within the lists. Adding objects to the head of the
1870 * list means when we traverse from the head we walk them in last-to-first
1871 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1872 * but for all other items there may be specific ordering that we need to
1876 xlog_recover_reorder_trans(
1878 struct xlog_recover
*trans
,
1881 xlog_recover_item_t
*item
, *n
;
1883 LIST_HEAD(sort_list
);
1884 LIST_HEAD(cancel_list
);
1885 LIST_HEAD(buffer_list
);
1886 LIST_HEAD(inode_buffer_list
);
1887 LIST_HEAD(inode_list
);
1889 list_splice_init(&trans
->r_itemq
, &sort_list
);
1890 list_for_each_entry_safe(item
, n
, &sort_list
, ri_list
) {
1891 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1893 switch (ITEM_TYPE(item
)) {
1894 case XFS_LI_ICREATE
:
1895 list_move_tail(&item
->ri_list
, &buffer_list
);
1898 if (buf_f
->blf_flags
& XFS_BLF_CANCEL
) {
1899 trace_xfs_log_recover_item_reorder_head(log
,
1901 list_move(&item
->ri_list
, &cancel_list
);
1904 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
1905 list_move(&item
->ri_list
, &inode_buffer_list
);
1908 list_move_tail(&item
->ri_list
, &buffer_list
);
1912 case XFS_LI_QUOTAOFF
:
1915 trace_xfs_log_recover_item_reorder_tail(log
,
1917 list_move_tail(&item
->ri_list
, &inode_list
);
1921 "%s: unrecognized type of log operation",
1925 * return the remaining items back to the transaction
1926 * item list so they can be freed in caller.
1928 if (!list_empty(&sort_list
))
1929 list_splice_init(&sort_list
, &trans
->r_itemq
);
1935 ASSERT(list_empty(&sort_list
));
1936 if (!list_empty(&buffer_list
))
1937 list_splice(&buffer_list
, &trans
->r_itemq
);
1938 if (!list_empty(&inode_list
))
1939 list_splice_tail(&inode_list
, &trans
->r_itemq
);
1940 if (!list_empty(&inode_buffer_list
))
1941 list_splice_tail(&inode_buffer_list
, &trans
->r_itemq
);
1942 if (!list_empty(&cancel_list
))
1943 list_splice_tail(&cancel_list
, &trans
->r_itemq
);
1948 * Build up the table of buf cancel records so that we don't replay
1949 * cancelled data in the second pass. For buffer records that are
1950 * not cancel records, there is nothing to do here so we just return.
1952 * If we get a cancel record which is already in the table, this indicates
1953 * that the buffer was cancelled multiple times. In order to ensure
1954 * that during pass 2 we keep the record in the table until we reach its
1955 * last occurrence in the log, we keep a reference count in the cancel
1956 * record in the table to tell us how many times we expect to see this
1957 * record during the second pass.
1960 xlog_recover_buffer_pass1(
1962 struct xlog_recover_item
*item
)
1964 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1965 struct list_head
*bucket
;
1966 struct xfs_buf_cancel
*bcp
;
1969 * If this isn't a cancel buffer item, then just return.
1971 if (!(buf_f
->blf_flags
& XFS_BLF_CANCEL
)) {
1972 trace_xfs_log_recover_buf_not_cancel(log
, buf_f
);
1977 * Insert an xfs_buf_cancel record into the hash table of them.
1978 * If there is already an identical record, bump its reference count.
1980 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, buf_f
->blf_blkno
);
1981 list_for_each_entry(bcp
, bucket
, bc_list
) {
1982 if (bcp
->bc_blkno
== buf_f
->blf_blkno
&&
1983 bcp
->bc_len
== buf_f
->blf_len
) {
1985 trace_xfs_log_recover_buf_cancel_ref_inc(log
, buf_f
);
1990 bcp
= kmem_alloc(sizeof(struct xfs_buf_cancel
), KM_SLEEP
);
1991 bcp
->bc_blkno
= buf_f
->blf_blkno
;
1992 bcp
->bc_len
= buf_f
->blf_len
;
1993 bcp
->bc_refcount
= 1;
1994 list_add_tail(&bcp
->bc_list
, bucket
);
1996 trace_xfs_log_recover_buf_cancel_add(log
, buf_f
);
2001 * Check to see whether the buffer being recovered has a corresponding
2002 * entry in the buffer cancel record table. If it is, return the cancel
2003 * buffer structure to the caller.
2005 STATIC
struct xfs_buf_cancel
*
2006 xlog_peek_buffer_cancelled(
2012 struct list_head
*bucket
;
2013 struct xfs_buf_cancel
*bcp
;
2015 if (!log
->l_buf_cancel_table
) {
2016 /* empty table means no cancelled buffers in the log */
2017 ASSERT(!(flags
& XFS_BLF_CANCEL
));
2021 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, blkno
);
2022 list_for_each_entry(bcp
, bucket
, bc_list
) {
2023 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
)
2028 * We didn't find a corresponding entry in the table, so return 0 so
2029 * that the buffer is NOT cancelled.
2031 ASSERT(!(flags
& XFS_BLF_CANCEL
));
2036 * If the buffer is being cancelled then return 1 so that it will be cancelled,
2037 * otherwise return 0. If the buffer is actually a buffer cancel item
2038 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
2039 * table and remove it from the table if this is the last reference.
2041 * We remove the cancel record from the table when we encounter its last
2042 * occurrence in the log so that if the same buffer is re-used again after its
2043 * last cancellation we actually replay the changes made at that point.
2046 xlog_check_buffer_cancelled(
2052 struct xfs_buf_cancel
*bcp
;
2054 bcp
= xlog_peek_buffer_cancelled(log
, blkno
, len
, flags
);
2059 * We've go a match, so return 1 so that the recovery of this buffer
2060 * is cancelled. If this buffer is actually a buffer cancel log
2061 * item, then decrement the refcount on the one in the table and
2062 * remove it if this is the last reference.
2064 if (flags
& XFS_BLF_CANCEL
) {
2065 if (--bcp
->bc_refcount
== 0) {
2066 list_del(&bcp
->bc_list
);
2074 * Perform recovery for a buffer full of inodes. In these buffers, the only
2075 * data which should be recovered is that which corresponds to the
2076 * di_next_unlinked pointers in the on disk inode structures. The rest of the
2077 * data for the inodes is always logged through the inodes themselves rather
2078 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
2080 * The only time when buffers full of inodes are fully recovered is when the
2081 * buffer is full of newly allocated inodes. In this case the buffer will
2082 * not be marked as an inode buffer and so will be sent to
2083 * xlog_recover_do_reg_buffer() below during recovery.
2086 xlog_recover_do_inode_buffer(
2087 struct xfs_mount
*mp
,
2088 xlog_recover_item_t
*item
,
2090 xfs_buf_log_format_t
*buf_f
)
2096 int reg_buf_offset
= 0;
2097 int reg_buf_bytes
= 0;
2098 int next_unlinked_offset
;
2100 xfs_agino_t
*logged_nextp
;
2101 xfs_agino_t
*buffer_nextp
;
2103 trace_xfs_log_recover_buf_inode_buf(mp
->m_log
, buf_f
);
2106 * Post recovery validation only works properly on CRC enabled
2109 if (xfs_sb_version_hascrc(&mp
->m_sb
))
2110 bp
->b_ops
= &xfs_inode_buf_ops
;
2112 inodes_per_buf
= BBTOB(bp
->b_io_length
) >> mp
->m_sb
.sb_inodelog
;
2113 for (i
= 0; i
< inodes_per_buf
; i
++) {
2114 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
2115 offsetof(xfs_dinode_t
, di_next_unlinked
);
2117 while (next_unlinked_offset
>=
2118 (reg_buf_offset
+ reg_buf_bytes
)) {
2120 * The next di_next_unlinked field is beyond
2121 * the current logged region. Find the next
2122 * logged region that contains or is beyond
2123 * the current di_next_unlinked field.
2126 bit
= xfs_next_bit(buf_f
->blf_data_map
,
2127 buf_f
->blf_map_size
, bit
);
2130 * If there are no more logged regions in the
2131 * buffer, then we're done.
2136 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
2137 buf_f
->blf_map_size
, bit
);
2139 reg_buf_offset
= bit
<< XFS_BLF_SHIFT
;
2140 reg_buf_bytes
= nbits
<< XFS_BLF_SHIFT
;
2145 * If the current logged region starts after the current
2146 * di_next_unlinked field, then move on to the next
2147 * di_next_unlinked field.
2149 if (next_unlinked_offset
< reg_buf_offset
)
2152 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
2153 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLF_CHUNK
) == 0);
2154 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <=
2155 BBTOB(bp
->b_io_length
));
2158 * The current logged region contains a copy of the
2159 * current di_next_unlinked field. Extract its value
2160 * and copy it to the buffer copy.
2162 logged_nextp
= item
->ri_buf
[item_index
].i_addr
+
2163 next_unlinked_offset
- reg_buf_offset
;
2164 if (unlikely(*logged_nextp
== 0)) {
2166 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
2167 "Trying to replay bad (0) inode di_next_unlinked field.",
2169 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
2170 XFS_ERRLEVEL_LOW
, mp
);
2171 return -EFSCORRUPTED
;
2174 buffer_nextp
= xfs_buf_offset(bp
, next_unlinked_offset
);
2175 *buffer_nextp
= *logged_nextp
;
2178 * If necessary, recalculate the CRC in the on-disk inode. We
2179 * have to leave the inode in a consistent state for whoever
2182 xfs_dinode_calc_crc(mp
,
2183 xfs_buf_offset(bp
, i
* mp
->m_sb
.sb_inodesize
));
2191 * V5 filesystems know the age of the buffer on disk being recovered. We can
2192 * have newer objects on disk than we are replaying, and so for these cases we
2193 * don't want to replay the current change as that will make the buffer contents
2194 * temporarily invalid on disk.
2196 * The magic number might not match the buffer type we are going to recover
2197 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
2198 * extract the LSN of the existing object in the buffer based on it's current
2199 * magic number. If we don't recognise the magic number in the buffer, then
2200 * return a LSN of -1 so that the caller knows it was an unrecognised block and
2201 * so can recover the buffer.
2203 * Note: we cannot rely solely on magic number matches to determine that the
2204 * buffer has a valid LSN - we also need to verify that it belongs to this
2205 * filesystem, so we need to extract the object's LSN and compare it to that
2206 * which we read from the superblock. If the UUIDs don't match, then we've got a
2207 * stale metadata block from an old filesystem instance that we need to recover
2211 xlog_recover_get_buf_lsn(
2212 struct xfs_mount
*mp
,
2218 void *blk
= bp
->b_addr
;
2222 /* v4 filesystems always recover immediately */
2223 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2224 goto recover_immediately
;
2226 magic32
= be32_to_cpu(*(__be32
*)blk
);
2228 case XFS_ABTB_CRC_MAGIC
:
2229 case XFS_ABTC_CRC_MAGIC
:
2230 case XFS_ABTB_MAGIC
:
2231 case XFS_ABTC_MAGIC
:
2232 case XFS_IBT_CRC_MAGIC
:
2233 case XFS_IBT_MAGIC
: {
2234 struct xfs_btree_block
*btb
= blk
;
2236 lsn
= be64_to_cpu(btb
->bb_u
.s
.bb_lsn
);
2237 uuid
= &btb
->bb_u
.s
.bb_uuid
;
2240 case XFS_BMAP_CRC_MAGIC
:
2241 case XFS_BMAP_MAGIC
: {
2242 struct xfs_btree_block
*btb
= blk
;
2244 lsn
= be64_to_cpu(btb
->bb_u
.l
.bb_lsn
);
2245 uuid
= &btb
->bb_u
.l
.bb_uuid
;
2249 lsn
= be64_to_cpu(((struct xfs_agf
*)blk
)->agf_lsn
);
2250 uuid
= &((struct xfs_agf
*)blk
)->agf_uuid
;
2252 case XFS_AGFL_MAGIC
:
2253 lsn
= be64_to_cpu(((struct xfs_agfl
*)blk
)->agfl_lsn
);
2254 uuid
= &((struct xfs_agfl
*)blk
)->agfl_uuid
;
2257 lsn
= be64_to_cpu(((struct xfs_agi
*)blk
)->agi_lsn
);
2258 uuid
= &((struct xfs_agi
*)blk
)->agi_uuid
;
2260 case XFS_SYMLINK_MAGIC
:
2261 lsn
= be64_to_cpu(((struct xfs_dsymlink_hdr
*)blk
)->sl_lsn
);
2262 uuid
= &((struct xfs_dsymlink_hdr
*)blk
)->sl_uuid
;
2264 case XFS_DIR3_BLOCK_MAGIC
:
2265 case XFS_DIR3_DATA_MAGIC
:
2266 case XFS_DIR3_FREE_MAGIC
:
2267 lsn
= be64_to_cpu(((struct xfs_dir3_blk_hdr
*)blk
)->lsn
);
2268 uuid
= &((struct xfs_dir3_blk_hdr
*)blk
)->uuid
;
2270 case XFS_ATTR3_RMT_MAGIC
:
2272 * Remote attr blocks are written synchronously, rather than
2273 * being logged. That means they do not contain a valid LSN
2274 * (i.e. transactionally ordered) in them, and hence any time we
2275 * see a buffer to replay over the top of a remote attribute
2276 * block we should simply do so.
2278 goto recover_immediately
;
2281 * superblock uuids are magic. We may or may not have a
2282 * sb_meta_uuid on disk, but it will be set in the in-core
2283 * superblock. We set the uuid pointer for verification
2284 * according to the superblock feature mask to ensure we check
2285 * the relevant UUID in the superblock.
2287 lsn
= be64_to_cpu(((struct xfs_dsb
*)blk
)->sb_lsn
);
2288 if (xfs_sb_version_hasmetauuid(&mp
->m_sb
))
2289 uuid
= &((struct xfs_dsb
*)blk
)->sb_meta_uuid
;
2291 uuid
= &((struct xfs_dsb
*)blk
)->sb_uuid
;
2297 if (lsn
!= (xfs_lsn_t
)-1) {
2298 if (!uuid_equal(&mp
->m_sb
.sb_meta_uuid
, uuid
))
2299 goto recover_immediately
;
2303 magicda
= be16_to_cpu(((struct xfs_da_blkinfo
*)blk
)->magic
);
2305 case XFS_DIR3_LEAF1_MAGIC
:
2306 case XFS_DIR3_LEAFN_MAGIC
:
2307 case XFS_DA3_NODE_MAGIC
:
2308 lsn
= be64_to_cpu(((struct xfs_da3_blkinfo
*)blk
)->lsn
);
2309 uuid
= &((struct xfs_da3_blkinfo
*)blk
)->uuid
;
2315 if (lsn
!= (xfs_lsn_t
)-1) {
2316 if (!uuid_equal(&mp
->m_sb
.sb_uuid
, uuid
))
2317 goto recover_immediately
;
2322 * We do individual object checks on dquot and inode buffers as they
2323 * have their own individual LSN records. Also, we could have a stale
2324 * buffer here, so we have to at least recognise these buffer types.
2326 * A notd complexity here is inode unlinked list processing - it logs
2327 * the inode directly in the buffer, but we don't know which inodes have
2328 * been modified, and there is no global buffer LSN. Hence we need to
2329 * recover all inode buffer types immediately. This problem will be
2330 * fixed by logical logging of the unlinked list modifications.
2332 magic16
= be16_to_cpu(*(__be16
*)blk
);
2334 case XFS_DQUOT_MAGIC
:
2335 case XFS_DINODE_MAGIC
:
2336 goto recover_immediately
;
2341 /* unknown buffer contents, recover immediately */
2343 recover_immediately
:
2344 return (xfs_lsn_t
)-1;
2349 * Validate the recovered buffer is of the correct type and attach the
2350 * appropriate buffer operations to them for writeback. Magic numbers are in a
2352 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2353 * the first 32 bits of the buffer (most blocks),
2354 * inside a struct xfs_da_blkinfo at the start of the buffer.
2357 xlog_recover_validate_buf_type(
2358 struct xfs_mount
*mp
,
2360 xfs_buf_log_format_t
*buf_f
)
2362 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
2368 * We can only do post recovery validation on items on CRC enabled
2369 * fielsystems as we need to know when the buffer was written to be able
2370 * to determine if we should have replayed the item. If we replay old
2371 * metadata over a newer buffer, then it will enter a temporarily
2372 * inconsistent state resulting in verification failures. Hence for now
2373 * just avoid the verification stage for non-crc filesystems
2375 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2378 magic32
= be32_to_cpu(*(__be32
*)bp
->b_addr
);
2379 magic16
= be16_to_cpu(*(__be16
*)bp
->b_addr
);
2380 magicda
= be16_to_cpu(info
->magic
);
2381 switch (xfs_blft_from_flags(buf_f
)) {
2382 case XFS_BLFT_BTREE_BUF
:
2384 case XFS_ABTB_CRC_MAGIC
:
2385 case XFS_ABTC_CRC_MAGIC
:
2386 case XFS_ABTB_MAGIC
:
2387 case XFS_ABTC_MAGIC
:
2388 bp
->b_ops
= &xfs_allocbt_buf_ops
;
2390 case XFS_IBT_CRC_MAGIC
:
2391 case XFS_FIBT_CRC_MAGIC
:
2393 case XFS_FIBT_MAGIC
:
2394 bp
->b_ops
= &xfs_inobt_buf_ops
;
2396 case XFS_BMAP_CRC_MAGIC
:
2397 case XFS_BMAP_MAGIC
:
2398 bp
->b_ops
= &xfs_bmbt_buf_ops
;
2401 xfs_warn(mp
, "Bad btree block magic!");
2406 case XFS_BLFT_AGF_BUF
:
2407 if (magic32
!= XFS_AGF_MAGIC
) {
2408 xfs_warn(mp
, "Bad AGF block magic!");
2412 bp
->b_ops
= &xfs_agf_buf_ops
;
2414 case XFS_BLFT_AGFL_BUF
:
2415 if (magic32
!= XFS_AGFL_MAGIC
) {
2416 xfs_warn(mp
, "Bad AGFL block magic!");
2420 bp
->b_ops
= &xfs_agfl_buf_ops
;
2422 case XFS_BLFT_AGI_BUF
:
2423 if (magic32
!= XFS_AGI_MAGIC
) {
2424 xfs_warn(mp
, "Bad AGI block magic!");
2428 bp
->b_ops
= &xfs_agi_buf_ops
;
2430 case XFS_BLFT_UDQUOT_BUF
:
2431 case XFS_BLFT_PDQUOT_BUF
:
2432 case XFS_BLFT_GDQUOT_BUF
:
2433 #ifdef CONFIG_XFS_QUOTA
2434 if (magic16
!= XFS_DQUOT_MAGIC
) {
2435 xfs_warn(mp
, "Bad DQUOT block magic!");
2439 bp
->b_ops
= &xfs_dquot_buf_ops
;
2442 "Trying to recover dquots without QUOTA support built in!");
2446 case XFS_BLFT_DINO_BUF
:
2447 if (magic16
!= XFS_DINODE_MAGIC
) {
2448 xfs_warn(mp
, "Bad INODE block magic!");
2452 bp
->b_ops
= &xfs_inode_buf_ops
;
2454 case XFS_BLFT_SYMLINK_BUF
:
2455 if (magic32
!= XFS_SYMLINK_MAGIC
) {
2456 xfs_warn(mp
, "Bad symlink block magic!");
2460 bp
->b_ops
= &xfs_symlink_buf_ops
;
2462 case XFS_BLFT_DIR_BLOCK_BUF
:
2463 if (magic32
!= XFS_DIR2_BLOCK_MAGIC
&&
2464 magic32
!= XFS_DIR3_BLOCK_MAGIC
) {
2465 xfs_warn(mp
, "Bad dir block magic!");
2469 bp
->b_ops
= &xfs_dir3_block_buf_ops
;
2471 case XFS_BLFT_DIR_DATA_BUF
:
2472 if (magic32
!= XFS_DIR2_DATA_MAGIC
&&
2473 magic32
!= XFS_DIR3_DATA_MAGIC
) {
2474 xfs_warn(mp
, "Bad dir data magic!");
2478 bp
->b_ops
= &xfs_dir3_data_buf_ops
;
2480 case XFS_BLFT_DIR_FREE_BUF
:
2481 if (magic32
!= XFS_DIR2_FREE_MAGIC
&&
2482 magic32
!= XFS_DIR3_FREE_MAGIC
) {
2483 xfs_warn(mp
, "Bad dir3 free magic!");
2487 bp
->b_ops
= &xfs_dir3_free_buf_ops
;
2489 case XFS_BLFT_DIR_LEAF1_BUF
:
2490 if (magicda
!= XFS_DIR2_LEAF1_MAGIC
&&
2491 magicda
!= XFS_DIR3_LEAF1_MAGIC
) {
2492 xfs_warn(mp
, "Bad dir leaf1 magic!");
2496 bp
->b_ops
= &xfs_dir3_leaf1_buf_ops
;
2498 case XFS_BLFT_DIR_LEAFN_BUF
:
2499 if (magicda
!= XFS_DIR2_LEAFN_MAGIC
&&
2500 magicda
!= XFS_DIR3_LEAFN_MAGIC
) {
2501 xfs_warn(mp
, "Bad dir leafn magic!");
2505 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
2507 case XFS_BLFT_DA_NODE_BUF
:
2508 if (magicda
!= XFS_DA_NODE_MAGIC
&&
2509 magicda
!= XFS_DA3_NODE_MAGIC
) {
2510 xfs_warn(mp
, "Bad da node magic!");
2514 bp
->b_ops
= &xfs_da3_node_buf_ops
;
2516 case XFS_BLFT_ATTR_LEAF_BUF
:
2517 if (magicda
!= XFS_ATTR_LEAF_MAGIC
&&
2518 magicda
!= XFS_ATTR3_LEAF_MAGIC
) {
2519 xfs_warn(mp
, "Bad attr leaf magic!");
2523 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
2525 case XFS_BLFT_ATTR_RMT_BUF
:
2526 if (magic32
!= XFS_ATTR3_RMT_MAGIC
) {
2527 xfs_warn(mp
, "Bad attr remote magic!");
2531 bp
->b_ops
= &xfs_attr3_rmt_buf_ops
;
2533 case XFS_BLFT_SB_BUF
:
2534 if (magic32
!= XFS_SB_MAGIC
) {
2535 xfs_warn(mp
, "Bad SB block magic!");
2539 bp
->b_ops
= &xfs_sb_buf_ops
;
2542 xfs_warn(mp
, "Unknown buffer type %d!",
2543 xfs_blft_from_flags(buf_f
));
2549 * Perform a 'normal' buffer recovery. Each logged region of the
2550 * buffer should be copied over the corresponding region in the
2551 * given buffer. The bitmap in the buf log format structure indicates
2552 * where to place the logged data.
2555 xlog_recover_do_reg_buffer(
2556 struct xfs_mount
*mp
,
2557 xlog_recover_item_t
*item
,
2559 xfs_buf_log_format_t
*buf_f
)
2566 trace_xfs_log_recover_buf_reg_buf(mp
->m_log
, buf_f
);
2569 i
= 1; /* 0 is the buf format structure */
2571 bit
= xfs_next_bit(buf_f
->blf_data_map
,
2572 buf_f
->blf_map_size
, bit
);
2575 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
2576 buf_f
->blf_map_size
, bit
);
2578 ASSERT(item
->ri_buf
[i
].i_addr
!= NULL
);
2579 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLF_CHUNK
== 0);
2580 ASSERT(BBTOB(bp
->b_io_length
) >=
2581 ((uint
)bit
<< XFS_BLF_SHIFT
) + (nbits
<< XFS_BLF_SHIFT
));
2584 * The dirty regions logged in the buffer, even though
2585 * contiguous, may span multiple chunks. This is because the
2586 * dirty region may span a physical page boundary in a buffer
2587 * and hence be split into two separate vectors for writing into
2588 * the log. Hence we need to trim nbits back to the length of
2589 * the current region being copied out of the log.
2591 if (item
->ri_buf
[i
].i_len
< (nbits
<< XFS_BLF_SHIFT
))
2592 nbits
= item
->ri_buf
[i
].i_len
>> XFS_BLF_SHIFT
;
2595 * Do a sanity check if this is a dquot buffer. Just checking
2596 * the first dquot in the buffer should do. XXXThis is
2597 * probably a good thing to do for other buf types also.
2600 if (buf_f
->blf_flags
&
2601 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2602 if (item
->ri_buf
[i
].i_addr
== NULL
) {
2604 "XFS: NULL dquot in %s.", __func__
);
2607 if (item
->ri_buf
[i
].i_len
< sizeof(xfs_disk_dquot_t
)) {
2609 "XFS: dquot too small (%d) in %s.",
2610 item
->ri_buf
[i
].i_len
, __func__
);
2613 error
= xfs_dqcheck(mp
, item
->ri_buf
[i
].i_addr
,
2614 -1, 0, XFS_QMOPT_DOWARN
,
2615 "dquot_buf_recover");
2620 memcpy(xfs_buf_offset(bp
,
2621 (uint
)bit
<< XFS_BLF_SHIFT
), /* dest */
2622 item
->ri_buf
[i
].i_addr
, /* source */
2623 nbits
<<XFS_BLF_SHIFT
); /* length */
2629 /* Shouldn't be any more regions */
2630 ASSERT(i
== item
->ri_total
);
2632 xlog_recover_validate_buf_type(mp
, bp
, buf_f
);
2636 * Perform a dquot buffer recovery.
2637 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2638 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2639 * Else, treat it as a regular buffer and do recovery.
2641 * Return false if the buffer was tossed and true if we recovered the buffer to
2642 * indicate to the caller if the buffer needs writing.
2645 xlog_recover_do_dquot_buffer(
2646 struct xfs_mount
*mp
,
2648 struct xlog_recover_item
*item
,
2650 struct xfs_buf_log_format
*buf_f
)
2654 trace_xfs_log_recover_buf_dquot_buf(log
, buf_f
);
2657 * Filesystems are required to send in quota flags at mount time.
2663 if (buf_f
->blf_flags
& XFS_BLF_UDQUOT_BUF
)
2664 type
|= XFS_DQ_USER
;
2665 if (buf_f
->blf_flags
& XFS_BLF_PDQUOT_BUF
)
2666 type
|= XFS_DQ_PROJ
;
2667 if (buf_f
->blf_flags
& XFS_BLF_GDQUOT_BUF
)
2668 type
|= XFS_DQ_GROUP
;
2670 * This type of quotas was turned off, so ignore this buffer
2672 if (log
->l_quotaoffs_flag
& type
)
2675 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2680 * This routine replays a modification made to a buffer at runtime.
2681 * There are actually two types of buffer, regular and inode, which
2682 * are handled differently. Inode buffers are handled differently
2683 * in that we only recover a specific set of data from them, namely
2684 * the inode di_next_unlinked fields. This is because all other inode
2685 * data is actually logged via inode records and any data we replay
2686 * here which overlaps that may be stale.
2688 * When meta-data buffers are freed at run time we log a buffer item
2689 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2690 * of the buffer in the log should not be replayed at recovery time.
2691 * This is so that if the blocks covered by the buffer are reused for
2692 * file data before we crash we don't end up replaying old, freed
2693 * meta-data into a user's file.
2695 * To handle the cancellation of buffer log items, we make two passes
2696 * over the log during recovery. During the first we build a table of
2697 * those buffers which have been cancelled, and during the second we
2698 * only replay those buffers which do not have corresponding cancel
2699 * records in the table. See xlog_recover_buffer_pass[1,2] above
2700 * for more details on the implementation of the table of cancel records.
2703 xlog_recover_buffer_pass2(
2705 struct list_head
*buffer_list
,
2706 struct xlog_recover_item
*item
,
2707 xfs_lsn_t current_lsn
)
2709 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
2710 xfs_mount_t
*mp
= log
->l_mp
;
2717 * In this pass we only want to recover all the buffers which have
2718 * not been cancelled and are not cancellation buffers themselves.
2720 if (xlog_check_buffer_cancelled(log
, buf_f
->blf_blkno
,
2721 buf_f
->blf_len
, buf_f
->blf_flags
)) {
2722 trace_xfs_log_recover_buf_cancel(log
, buf_f
);
2726 trace_xfs_log_recover_buf_recover(log
, buf_f
);
2729 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
)
2730 buf_flags
|= XBF_UNMAPPED
;
2732 bp
= xfs_buf_read(mp
->m_ddev_targp
, buf_f
->blf_blkno
, buf_f
->blf_len
,
2736 error
= bp
->b_error
;
2738 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#1)");
2743 * Recover the buffer only if we get an LSN from it and it's less than
2744 * the lsn of the transaction we are replaying.
2746 * Note that we have to be extremely careful of readahead here.
2747 * Readahead does not attach verfiers to the buffers so if we don't
2748 * actually do any replay after readahead because of the LSN we found
2749 * in the buffer if more recent than that current transaction then we
2750 * need to attach the verifier directly. Failure to do so can lead to
2751 * future recovery actions (e.g. EFI and unlinked list recovery) can
2752 * operate on the buffers and they won't get the verifier attached. This
2753 * can lead to blocks on disk having the correct content but a stale
2756 * It is safe to assume these clean buffers are currently up to date.
2757 * If the buffer is dirtied by a later transaction being replayed, then
2758 * the verifier will be reset to match whatever recover turns that
2761 lsn
= xlog_recover_get_buf_lsn(mp
, bp
);
2762 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
2763 xlog_recover_validate_buf_type(mp
, bp
, buf_f
);
2767 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
2768 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2771 } else if (buf_f
->blf_flags
&
2772 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2775 dirty
= xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2779 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2783 * Perform delayed write on the buffer. Asynchronous writes will be
2784 * slower when taking into account all the buffers to be flushed.
2786 * Also make sure that only inode buffers with good sizes stay in
2787 * the buffer cache. The kernel moves inodes in buffers of 1 block
2788 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
2789 * buffers in the log can be a different size if the log was generated
2790 * by an older kernel using unclustered inode buffers or a newer kernel
2791 * running with a different inode cluster size. Regardless, if the
2792 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2793 * for *our* value of mp->m_inode_cluster_size, then we need to keep
2794 * the buffer out of the buffer cache so that the buffer won't
2795 * overlap with future reads of those inodes.
2797 if (XFS_DINODE_MAGIC
==
2798 be16_to_cpu(*((__be16
*)xfs_buf_offset(bp
, 0))) &&
2799 (BBTOB(bp
->b_io_length
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2800 (__uint32_t
)log
->l_mp
->m_inode_cluster_size
))) {
2802 error
= xfs_bwrite(bp
);
2804 ASSERT(bp
->b_target
->bt_mount
== mp
);
2805 bp
->b_iodone
= xlog_recover_iodone
;
2806 xfs_buf_delwri_queue(bp
, buffer_list
);
2815 * Inode fork owner changes
2817 * If we have been told that we have to reparent the inode fork, it's because an
2818 * extent swap operation on a CRC enabled filesystem has been done and we are
2819 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2822 * The complexity here is that we don't have an inode context to work with, so
2823 * after we've replayed the inode we need to instantiate one. This is where the
2826 * We are in the middle of log recovery, so we can't run transactions. That
2827 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2828 * that will result in the corresponding iput() running the inode through
2829 * xfs_inactive(). If we've just replayed an inode core that changes the link
2830 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2831 * transactions (bad!).
2833 * So, to avoid this, we instantiate an inode directly from the inode core we've
2834 * just recovered. We have the buffer still locked, and all we really need to
2835 * instantiate is the inode core and the forks being modified. We can do this
2836 * manually, then run the inode btree owner change, and then tear down the
2837 * xfs_inode without having to run any transactions at all.
2839 * Also, because we don't have a transaction context available here but need to
2840 * gather all the buffers we modify for writeback so we pass the buffer_list
2841 * instead for the operation to use.
2845 xfs_recover_inode_owner_change(
2846 struct xfs_mount
*mp
,
2847 struct xfs_dinode
*dip
,
2848 struct xfs_inode_log_format
*in_f
,
2849 struct list_head
*buffer_list
)
2851 struct xfs_inode
*ip
;
2854 ASSERT(in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
));
2856 ip
= xfs_inode_alloc(mp
, in_f
->ilf_ino
);
2860 /* instantiate the inode */
2861 xfs_dinode_from_disk(&ip
->i_d
, dip
);
2862 ASSERT(ip
->i_d
.di_version
>= 3);
2864 error
= xfs_iformat_fork(ip
, dip
);
2869 if (in_f
->ilf_fields
& XFS_ILOG_DOWNER
) {
2870 ASSERT(in_f
->ilf_fields
& XFS_ILOG_DBROOT
);
2871 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_DATA_FORK
,
2872 ip
->i_ino
, buffer_list
);
2877 if (in_f
->ilf_fields
& XFS_ILOG_AOWNER
) {
2878 ASSERT(in_f
->ilf_fields
& XFS_ILOG_ABROOT
);
2879 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_ATTR_FORK
,
2880 ip
->i_ino
, buffer_list
);
2891 xlog_recover_inode_pass2(
2893 struct list_head
*buffer_list
,
2894 struct xlog_recover_item
*item
,
2895 xfs_lsn_t current_lsn
)
2897 xfs_inode_log_format_t
*in_f
;
2898 xfs_mount_t
*mp
= log
->l_mp
;
2907 xfs_icdinode_t
*dicp
;
2911 if (item
->ri_buf
[0].i_len
== sizeof(xfs_inode_log_format_t
)) {
2912 in_f
= item
->ri_buf
[0].i_addr
;
2914 in_f
= kmem_alloc(sizeof(xfs_inode_log_format_t
), KM_SLEEP
);
2916 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
2922 * Inode buffers can be freed, look out for it,
2923 * and do not replay the inode.
2925 if (xlog_check_buffer_cancelled(log
, in_f
->ilf_blkno
,
2926 in_f
->ilf_len
, 0)) {
2928 trace_xfs_log_recover_inode_cancel(log
, in_f
);
2931 trace_xfs_log_recover_inode_recover(log
, in_f
);
2933 bp
= xfs_buf_read(mp
->m_ddev_targp
, in_f
->ilf_blkno
, in_f
->ilf_len
, 0,
2934 &xfs_inode_buf_ops
);
2939 error
= bp
->b_error
;
2941 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#2)");
2944 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2945 dip
= xfs_buf_offset(bp
, in_f
->ilf_boffset
);
2948 * Make sure the place we're flushing out to really looks
2951 if (unlikely(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
))) {
2953 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2954 __func__
, dip
, bp
, in_f
->ilf_ino
);
2955 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2956 XFS_ERRLEVEL_LOW
, mp
);
2957 error
= -EFSCORRUPTED
;
2960 dicp
= item
->ri_buf
[1].i_addr
;
2961 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2963 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2964 __func__
, item
, in_f
->ilf_ino
);
2965 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2966 XFS_ERRLEVEL_LOW
, mp
);
2967 error
= -EFSCORRUPTED
;
2972 * If the inode has an LSN in it, recover the inode only if it's less
2973 * than the lsn of the transaction we are replaying. Note: we still
2974 * need to replay an owner change even though the inode is more recent
2975 * than the transaction as there is no guarantee that all the btree
2976 * blocks are more recent than this transaction, too.
2978 if (dip
->di_version
>= 3) {
2979 xfs_lsn_t lsn
= be64_to_cpu(dip
->di_lsn
);
2981 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
2982 trace_xfs_log_recover_inode_skip(log
, in_f
);
2984 goto out_owner_change
;
2989 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2990 * are transactional and if ordering is necessary we can determine that
2991 * more accurately by the LSN field in the V3 inode core. Don't trust
2992 * the inode versions we might be changing them here - use the
2993 * superblock flag to determine whether we need to look at di_flushiter
2994 * to skip replay when the on disk inode is newer than the log one
2996 if (!xfs_sb_version_hascrc(&mp
->m_sb
) &&
2997 dicp
->di_flushiter
< be16_to_cpu(dip
->di_flushiter
)) {
2999 * Deal with the wrap case, DI_MAX_FLUSH is less
3000 * than smaller numbers
3002 if (be16_to_cpu(dip
->di_flushiter
) == DI_MAX_FLUSH
&&
3003 dicp
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
3006 trace_xfs_log_recover_inode_skip(log
, in_f
);
3012 /* Take the opportunity to reset the flush iteration count */
3013 dicp
->di_flushiter
= 0;
3015 if (unlikely(S_ISREG(dicp
->di_mode
))) {
3016 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3017 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
3018 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
3019 XFS_ERRLEVEL_LOW
, mp
, dicp
);
3021 "%s: Bad regular inode log record, rec ptr 0x%p, "
3022 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
3023 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
3024 error
= -EFSCORRUPTED
;
3027 } else if (unlikely(S_ISDIR(dicp
->di_mode
))) {
3028 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3029 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
3030 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
3031 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
3032 XFS_ERRLEVEL_LOW
, mp
, dicp
);
3034 "%s: Bad dir inode log record, rec ptr 0x%p, "
3035 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
3036 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
3037 error
= -EFSCORRUPTED
;
3041 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
3042 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
3043 XFS_ERRLEVEL_LOW
, mp
, dicp
);
3045 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
3046 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
3047 __func__
, item
, dip
, bp
, in_f
->ilf_ino
,
3048 dicp
->di_nextents
+ dicp
->di_anextents
,
3050 error
= -EFSCORRUPTED
;
3053 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
3054 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
3055 XFS_ERRLEVEL_LOW
, mp
, dicp
);
3057 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
3058 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__
,
3059 item
, dip
, bp
, in_f
->ilf_ino
, dicp
->di_forkoff
);
3060 error
= -EFSCORRUPTED
;
3063 isize
= xfs_icdinode_size(dicp
->di_version
);
3064 if (unlikely(item
->ri_buf
[1].i_len
> isize
)) {
3065 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
3066 XFS_ERRLEVEL_LOW
, mp
, dicp
);
3068 "%s: Bad inode log record length %d, rec ptr 0x%p",
3069 __func__
, item
->ri_buf
[1].i_len
, item
);
3070 error
= -EFSCORRUPTED
;
3074 /* The core is in in-core format */
3075 xfs_dinode_to_disk(dip
, dicp
);
3077 /* the rest is in on-disk format */
3078 if (item
->ri_buf
[1].i_len
> isize
) {
3079 memcpy((char *)dip
+ isize
,
3080 item
->ri_buf
[1].i_addr
+ isize
,
3081 item
->ri_buf
[1].i_len
- isize
);
3084 fields
= in_f
->ilf_fields
;
3085 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
3087 xfs_dinode_put_rdev(dip
, in_f
->ilf_u
.ilfu_rdev
);
3090 memcpy(XFS_DFORK_DPTR(dip
),
3091 &in_f
->ilf_u
.ilfu_uuid
,
3096 if (in_f
->ilf_size
== 2)
3097 goto out_owner_change
;
3098 len
= item
->ri_buf
[2].i_len
;
3099 src
= item
->ri_buf
[2].i_addr
;
3100 ASSERT(in_f
->ilf_size
<= 4);
3101 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
3102 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
3103 (len
== in_f
->ilf_dsize
));
3105 switch (fields
& XFS_ILOG_DFORK
) {
3106 case XFS_ILOG_DDATA
:
3108 memcpy(XFS_DFORK_DPTR(dip
), src
, len
);
3111 case XFS_ILOG_DBROOT
:
3112 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
, len
,
3113 (xfs_bmdr_block_t
*)XFS_DFORK_DPTR(dip
),
3114 XFS_DFORK_DSIZE(dip
, mp
));
3119 * There are no data fork flags set.
3121 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
3126 * If we logged any attribute data, recover it. There may or
3127 * may not have been any other non-core data logged in this
3130 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
3131 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
3136 len
= item
->ri_buf
[attr_index
].i_len
;
3137 src
= item
->ri_buf
[attr_index
].i_addr
;
3138 ASSERT(len
== in_f
->ilf_asize
);
3140 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
3141 case XFS_ILOG_ADATA
:
3143 dest
= XFS_DFORK_APTR(dip
);
3144 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
3145 memcpy(dest
, src
, len
);
3148 case XFS_ILOG_ABROOT
:
3149 dest
= XFS_DFORK_APTR(dip
);
3150 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
,
3151 len
, (xfs_bmdr_block_t
*)dest
,
3152 XFS_DFORK_ASIZE(dip
, mp
));
3156 xfs_warn(log
->l_mp
, "%s: Invalid flag", __func__
);
3164 if (in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
))
3165 error
= xfs_recover_inode_owner_change(mp
, dip
, in_f
,
3167 /* re-generate the checksum. */
3168 xfs_dinode_calc_crc(log
->l_mp
, dip
);
3170 ASSERT(bp
->b_target
->bt_mount
== mp
);
3171 bp
->b_iodone
= xlog_recover_iodone
;
3172 xfs_buf_delwri_queue(bp
, buffer_list
);
3183 * Recover QUOTAOFF records. We simply make a note of it in the xlog
3184 * structure, so that we know not to do any dquot item or dquot buffer recovery,
3188 xlog_recover_quotaoff_pass1(
3190 struct xlog_recover_item
*item
)
3192 xfs_qoff_logformat_t
*qoff_f
= item
->ri_buf
[0].i_addr
;
3196 * The logitem format's flag tells us if this was user quotaoff,
3197 * group/project quotaoff or both.
3199 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
3200 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
3201 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
3202 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
3203 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
3204 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
3210 * Recover a dquot record
3213 xlog_recover_dquot_pass2(
3215 struct list_head
*buffer_list
,
3216 struct xlog_recover_item
*item
,
3217 xfs_lsn_t current_lsn
)
3219 xfs_mount_t
*mp
= log
->l_mp
;
3221 struct xfs_disk_dquot
*ddq
, *recddq
;
3223 xfs_dq_logformat_t
*dq_f
;
3228 * Filesystems are required to send in quota flags at mount time.
3230 if (mp
->m_qflags
== 0)
3233 recddq
= item
->ri_buf
[1].i_addr
;
3234 if (recddq
== NULL
) {
3235 xfs_alert(log
->l_mp
, "NULL dquot in %s.", __func__
);
3238 if (item
->ri_buf
[1].i_len
< sizeof(xfs_disk_dquot_t
)) {
3239 xfs_alert(log
->l_mp
, "dquot too small (%d) in %s.",
3240 item
->ri_buf
[1].i_len
, __func__
);
3245 * This type of quotas was turned off, so ignore this record.
3247 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
3249 if (log
->l_quotaoffs_flag
& type
)
3253 * At this point we know that quota was _not_ turned off.
3254 * Since the mount flags are not indicating to us otherwise, this
3255 * must mean that quota is on, and the dquot needs to be replayed.
3256 * Remember that we may not have fully recovered the superblock yet,
3257 * so we can't do the usual trick of looking at the SB quota bits.
3259 * The other possibility, of course, is that the quota subsystem was
3260 * removed since the last mount - ENOSYS.
3262 dq_f
= item
->ri_buf
[0].i_addr
;
3264 error
= xfs_dqcheck(mp
, recddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
3265 "xlog_recover_dquot_pass2 (log copy)");
3268 ASSERT(dq_f
->qlf_len
== 1);
3271 * At this point we are assuming that the dquots have been allocated
3272 * and hence the buffer has valid dquots stamped in it. It should,
3273 * therefore, pass verifier validation. If the dquot is bad, then the
3274 * we'll return an error here, so we don't need to specifically check
3275 * the dquot in the buffer after the verifier has run.
3277 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
, dq_f
->qlf_blkno
,
3278 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
), 0, &bp
,
3279 &xfs_dquot_buf_ops
);
3284 ddq
= xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
3287 * If the dquot has an LSN in it, recover the dquot only if it's less
3288 * than the lsn of the transaction we are replaying.
3290 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
3291 struct xfs_dqblk
*dqb
= (struct xfs_dqblk
*)ddq
;
3292 xfs_lsn_t lsn
= be64_to_cpu(dqb
->dd_lsn
);
3294 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
3299 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
3300 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
3301 xfs_update_cksum((char *)ddq
, sizeof(struct xfs_dqblk
),
3305 ASSERT(dq_f
->qlf_size
== 2);
3306 ASSERT(bp
->b_target
->bt_mount
== mp
);
3307 bp
->b_iodone
= xlog_recover_iodone
;
3308 xfs_buf_delwri_queue(bp
, buffer_list
);
3316 * This routine is called to create an in-core extent free intent
3317 * item from the efi format structure which was logged on disk.
3318 * It allocates an in-core efi, copies the extents from the format
3319 * structure into it, and adds the efi to the AIL with the given
3323 xlog_recover_efi_pass2(
3325 struct xlog_recover_item
*item
,
3329 struct xfs_mount
*mp
= log
->l_mp
;
3330 struct xfs_efi_log_item
*efip
;
3331 struct xfs_efi_log_format
*efi_formatp
;
3333 efi_formatp
= item
->ri_buf
[0].i_addr
;
3335 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
3336 error
= xfs_efi_copy_format(&item
->ri_buf
[0], &efip
->efi_format
);
3338 xfs_efi_item_free(efip
);
3341 atomic_set(&efip
->efi_next_extent
, efi_formatp
->efi_nextents
);
3343 spin_lock(&log
->l_ailp
->xa_lock
);
3345 * The EFI has two references. One for the EFD and one for EFI to ensure
3346 * it makes it into the AIL. Insert the EFI into the AIL directly and
3347 * drop the EFI reference. Note that xfs_trans_ail_update() drops the
3350 xfs_trans_ail_update(log
->l_ailp
, &efip
->efi_item
, lsn
);
3351 xfs_efi_release(efip
);
3357 * This routine is called when an EFD format structure is found in a committed
3358 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
3359 * was still in the log. To do this it searches the AIL for the EFI with an id
3360 * equal to that in the EFD format structure. If we find it we drop the EFD
3361 * reference, which removes the EFI from the AIL and frees it.
3364 xlog_recover_efd_pass2(
3366 struct xlog_recover_item
*item
)
3368 xfs_efd_log_format_t
*efd_formatp
;
3369 xfs_efi_log_item_t
*efip
= NULL
;
3370 xfs_log_item_t
*lip
;
3372 struct xfs_ail_cursor cur
;
3373 struct xfs_ail
*ailp
= log
->l_ailp
;
3375 efd_formatp
= item
->ri_buf
[0].i_addr
;
3376 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
3377 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
3378 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
3379 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
3380 efi_id
= efd_formatp
->efd_efi_id
;
3383 * Search for the EFI with the id in the EFD format structure in the
3386 spin_lock(&ailp
->xa_lock
);
3387 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3388 while (lip
!= NULL
) {
3389 if (lip
->li_type
== XFS_LI_EFI
) {
3390 efip
= (xfs_efi_log_item_t
*)lip
;
3391 if (efip
->efi_format
.efi_id
== efi_id
) {
3393 * Drop the EFD reference to the EFI. This
3394 * removes the EFI from the AIL and frees it.
3396 spin_unlock(&ailp
->xa_lock
);
3397 xfs_efi_release(efip
);
3398 spin_lock(&ailp
->xa_lock
);
3402 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3405 xfs_trans_ail_cursor_done(&cur
);
3406 spin_unlock(&ailp
->xa_lock
);
3412 * This routine is called when an inode create format structure is found in a
3413 * committed transaction in the log. It's purpose is to initialise the inodes
3414 * being allocated on disk. This requires us to get inode cluster buffers that
3415 * match the range to be intialised, stamped with inode templates and written
3416 * by delayed write so that subsequent modifications will hit the cached buffer
3417 * and only need writing out at the end of recovery.
3420 xlog_recover_do_icreate_pass2(
3422 struct list_head
*buffer_list
,
3423 xlog_recover_item_t
*item
)
3425 struct xfs_mount
*mp
= log
->l_mp
;
3426 struct xfs_icreate_log
*icl
;
3427 xfs_agnumber_t agno
;
3428 xfs_agblock_t agbno
;
3431 xfs_agblock_t length
;
3432 int blks_per_cluster
;
3438 icl
= (struct xfs_icreate_log
*)item
->ri_buf
[0].i_addr
;
3439 if (icl
->icl_type
!= XFS_LI_ICREATE
) {
3440 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad type");
3444 if (icl
->icl_size
!= 1) {
3445 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad icl size");
3449 agno
= be32_to_cpu(icl
->icl_ag
);
3450 if (agno
>= mp
->m_sb
.sb_agcount
) {
3451 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad agno");
3454 agbno
= be32_to_cpu(icl
->icl_agbno
);
3455 if (!agbno
|| agbno
== NULLAGBLOCK
|| agbno
>= mp
->m_sb
.sb_agblocks
) {
3456 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad agbno");
3459 isize
= be32_to_cpu(icl
->icl_isize
);
3460 if (isize
!= mp
->m_sb
.sb_inodesize
) {
3461 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad isize");
3464 count
= be32_to_cpu(icl
->icl_count
);
3466 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad count");
3469 length
= be32_to_cpu(icl
->icl_length
);
3470 if (!length
|| length
>= mp
->m_sb
.sb_agblocks
) {
3471 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad length");
3476 * The inode chunk is either full or sparse and we only support
3477 * m_ialloc_min_blks sized sparse allocations at this time.
3479 if (length
!= mp
->m_ialloc_blks
&&
3480 length
!= mp
->m_ialloc_min_blks
) {
3482 "%s: unsupported chunk length", __FUNCTION__
);
3486 /* verify inode count is consistent with extent length */
3487 if ((count
>> mp
->m_sb
.sb_inopblog
) != length
) {
3489 "%s: inconsistent inode count and chunk length",
3495 * The icreate transaction can cover multiple cluster buffers and these
3496 * buffers could have been freed and reused. Check the individual
3497 * buffers for cancellation so we don't overwrite anything written after
3500 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
3501 bb_per_cluster
= XFS_FSB_TO_BB(mp
, blks_per_cluster
);
3502 nbufs
= length
/ blks_per_cluster
;
3503 for (i
= 0, cancel_count
= 0; i
< nbufs
; i
++) {
3506 daddr
= XFS_AGB_TO_DADDR(mp
, agno
,
3507 agbno
+ i
* blks_per_cluster
);
3508 if (xlog_check_buffer_cancelled(log
, daddr
, bb_per_cluster
, 0))
3513 * We currently only use icreate for a single allocation at a time. This
3514 * means we should expect either all or none of the buffers to be
3515 * cancelled. Be conservative and skip replay if at least one buffer is
3516 * cancelled, but warn the user that something is awry if the buffers
3517 * are not consistent.
3519 * XXX: This must be refined to only skip cancelled clusters once we use
3520 * icreate for multiple chunk allocations.
3522 ASSERT(!cancel_count
|| cancel_count
== nbufs
);
3524 if (cancel_count
!= nbufs
)
3526 "WARNING: partial inode chunk cancellation, skipped icreate.");
3527 trace_xfs_log_recover_icreate_cancel(log
, icl
);
3531 trace_xfs_log_recover_icreate_recover(log
, icl
);
3532 return xfs_ialloc_inode_init(mp
, NULL
, buffer_list
, count
, agno
, agbno
,
3533 length
, be32_to_cpu(icl
->icl_gen
));
3537 xlog_recover_buffer_ra_pass2(
3539 struct xlog_recover_item
*item
)
3541 struct xfs_buf_log_format
*buf_f
= item
->ri_buf
[0].i_addr
;
3542 struct xfs_mount
*mp
= log
->l_mp
;
3544 if (xlog_peek_buffer_cancelled(log
, buf_f
->blf_blkno
,
3545 buf_f
->blf_len
, buf_f
->blf_flags
)) {
3549 xfs_buf_readahead(mp
->m_ddev_targp
, buf_f
->blf_blkno
,
3550 buf_f
->blf_len
, NULL
);
3554 xlog_recover_inode_ra_pass2(
3556 struct xlog_recover_item
*item
)
3558 struct xfs_inode_log_format ilf_buf
;
3559 struct xfs_inode_log_format
*ilfp
;
3560 struct xfs_mount
*mp
= log
->l_mp
;
3563 if (item
->ri_buf
[0].i_len
== sizeof(struct xfs_inode_log_format
)) {
3564 ilfp
= item
->ri_buf
[0].i_addr
;
3567 memset(ilfp
, 0, sizeof(*ilfp
));
3568 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], ilfp
);
3573 if (xlog_peek_buffer_cancelled(log
, ilfp
->ilf_blkno
, ilfp
->ilf_len
, 0))
3576 xfs_buf_readahead(mp
->m_ddev_targp
, ilfp
->ilf_blkno
,
3577 ilfp
->ilf_len
, &xfs_inode_buf_ra_ops
);
3581 xlog_recover_dquot_ra_pass2(
3583 struct xlog_recover_item
*item
)
3585 struct xfs_mount
*mp
= log
->l_mp
;
3586 struct xfs_disk_dquot
*recddq
;
3587 struct xfs_dq_logformat
*dq_f
;
3592 if (mp
->m_qflags
== 0)
3595 recddq
= item
->ri_buf
[1].i_addr
;
3598 if (item
->ri_buf
[1].i_len
< sizeof(struct xfs_disk_dquot
))
3601 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
3603 if (log
->l_quotaoffs_flag
& type
)
3606 dq_f
= item
->ri_buf
[0].i_addr
;
3608 ASSERT(dq_f
->qlf_len
== 1);
3610 len
= XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
);
3611 if (xlog_peek_buffer_cancelled(log
, dq_f
->qlf_blkno
, len
, 0))
3614 xfs_buf_readahead(mp
->m_ddev_targp
, dq_f
->qlf_blkno
, len
,
3615 &xfs_dquot_buf_ra_ops
);
3619 xlog_recover_ra_pass2(
3621 struct xlog_recover_item
*item
)
3623 switch (ITEM_TYPE(item
)) {
3625 xlog_recover_buffer_ra_pass2(log
, item
);
3628 xlog_recover_inode_ra_pass2(log
, item
);
3631 xlog_recover_dquot_ra_pass2(log
, item
);
3635 case XFS_LI_QUOTAOFF
:
3642 xlog_recover_commit_pass1(
3644 struct xlog_recover
*trans
,
3645 struct xlog_recover_item
*item
)
3647 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS1
);
3649 switch (ITEM_TYPE(item
)) {
3651 return xlog_recover_buffer_pass1(log
, item
);
3652 case XFS_LI_QUOTAOFF
:
3653 return xlog_recover_quotaoff_pass1(log
, item
);
3658 case XFS_LI_ICREATE
:
3659 /* nothing to do in pass 1 */
3662 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
3663 __func__
, ITEM_TYPE(item
));
3670 xlog_recover_commit_pass2(
3672 struct xlog_recover
*trans
,
3673 struct list_head
*buffer_list
,
3674 struct xlog_recover_item
*item
)
3676 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS2
);
3678 switch (ITEM_TYPE(item
)) {
3680 return xlog_recover_buffer_pass2(log
, buffer_list
, item
,
3683 return xlog_recover_inode_pass2(log
, buffer_list
, item
,
3686 return xlog_recover_efi_pass2(log
, item
, trans
->r_lsn
);
3688 return xlog_recover_efd_pass2(log
, item
);
3690 return xlog_recover_dquot_pass2(log
, buffer_list
, item
,
3692 case XFS_LI_ICREATE
:
3693 return xlog_recover_do_icreate_pass2(log
, buffer_list
, item
);
3694 case XFS_LI_QUOTAOFF
:
3695 /* nothing to do in pass2 */
3698 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
3699 __func__
, ITEM_TYPE(item
));
3706 xlog_recover_items_pass2(
3708 struct xlog_recover
*trans
,
3709 struct list_head
*buffer_list
,
3710 struct list_head
*item_list
)
3712 struct xlog_recover_item
*item
;
3715 list_for_each_entry(item
, item_list
, ri_list
) {
3716 error
= xlog_recover_commit_pass2(log
, trans
,
3726 * Perform the transaction.
3728 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3729 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3732 xlog_recover_commit_trans(
3734 struct xlog_recover
*trans
,
3739 int items_queued
= 0;
3740 struct xlog_recover_item
*item
;
3741 struct xlog_recover_item
*next
;
3742 LIST_HEAD (buffer_list
);
3743 LIST_HEAD (ra_list
);
3744 LIST_HEAD (done_list
);
3746 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
3748 hlist_del(&trans
->r_list
);
3750 error
= xlog_recover_reorder_trans(log
, trans
, pass
);
3754 list_for_each_entry_safe(item
, next
, &trans
->r_itemq
, ri_list
) {
3756 case XLOG_RECOVER_PASS1
:
3757 error
= xlog_recover_commit_pass1(log
, trans
, item
);
3759 case XLOG_RECOVER_PASS2
:
3760 xlog_recover_ra_pass2(log
, item
);
3761 list_move_tail(&item
->ri_list
, &ra_list
);
3763 if (items_queued
>= XLOG_RECOVER_COMMIT_QUEUE_MAX
) {
3764 error
= xlog_recover_items_pass2(log
, trans
,
3765 &buffer_list
, &ra_list
);
3766 list_splice_tail_init(&ra_list
, &done_list
);
3780 if (!list_empty(&ra_list
)) {
3782 error
= xlog_recover_items_pass2(log
, trans
,
3783 &buffer_list
, &ra_list
);
3784 list_splice_tail_init(&ra_list
, &done_list
);
3787 if (!list_empty(&done_list
))
3788 list_splice_init(&done_list
, &trans
->r_itemq
);
3790 error2
= xfs_buf_delwri_submit(&buffer_list
);
3791 return error
? error
: error2
;
3795 xlog_recover_add_item(
3796 struct list_head
*head
)
3798 xlog_recover_item_t
*item
;
3800 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
3801 INIT_LIST_HEAD(&item
->ri_list
);
3802 list_add_tail(&item
->ri_list
, head
);
3806 xlog_recover_add_to_cont_trans(
3808 struct xlog_recover
*trans
,
3812 xlog_recover_item_t
*item
;
3813 char *ptr
, *old_ptr
;
3817 * If the transaction is empty, the header was split across this and the
3818 * previous record. Copy the rest of the header.
3820 if (list_empty(&trans
->r_itemq
)) {
3821 ASSERT(len
<= sizeof(struct xfs_trans_header
));
3822 if (len
> sizeof(struct xfs_trans_header
)) {
3823 xfs_warn(log
->l_mp
, "%s: bad header length", __func__
);
3827 xlog_recover_add_item(&trans
->r_itemq
);
3828 ptr
= (char *)&trans
->r_theader
+
3829 sizeof(struct xfs_trans_header
) - len
;
3830 memcpy(ptr
, dp
, len
);
3834 /* take the tail entry */
3835 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
3837 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
3838 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
3840 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, KM_SLEEP
);
3841 memcpy(&ptr
[old_len
], dp
, len
);
3842 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
3843 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
3844 trace_xfs_log_recover_item_add_cont(log
, trans
, item
, 0);
3849 * The next region to add is the start of a new region. It could be
3850 * a whole region or it could be the first part of a new region. Because
3851 * of this, the assumption here is that the type and size fields of all
3852 * format structures fit into the first 32 bits of the structure.
3854 * This works because all regions must be 32 bit aligned. Therefore, we
3855 * either have both fields or we have neither field. In the case we have
3856 * neither field, the data part of the region is zero length. We only have
3857 * a log_op_header and can throw away the header since a new one will appear
3858 * later. If we have at least 4 bytes, then we can determine how many regions
3859 * will appear in the current log item.
3862 xlog_recover_add_to_trans(
3864 struct xlog_recover
*trans
,
3868 xfs_inode_log_format_t
*in_f
; /* any will do */
3869 xlog_recover_item_t
*item
;
3874 if (list_empty(&trans
->r_itemq
)) {
3875 /* we need to catch log corruptions here */
3876 if (*(uint
*)dp
!= XFS_TRANS_HEADER_MAGIC
) {
3877 xfs_warn(log
->l_mp
, "%s: bad header magic number",
3883 if (len
> sizeof(struct xfs_trans_header
)) {
3884 xfs_warn(log
->l_mp
, "%s: bad header length", __func__
);
3890 * The transaction header can be arbitrarily split across op
3891 * records. If we don't have the whole thing here, copy what we
3892 * do have and handle the rest in the next record.
3894 if (len
== sizeof(struct xfs_trans_header
))
3895 xlog_recover_add_item(&trans
->r_itemq
);
3896 memcpy(&trans
->r_theader
, dp
, len
);
3900 ptr
= kmem_alloc(len
, KM_SLEEP
);
3901 memcpy(ptr
, dp
, len
);
3902 in_f
= (xfs_inode_log_format_t
*)ptr
;
3904 /* take the tail entry */
3905 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
3906 if (item
->ri_total
!= 0 &&
3907 item
->ri_total
== item
->ri_cnt
) {
3908 /* tail item is in use, get a new one */
3909 xlog_recover_add_item(&trans
->r_itemq
);
3910 item
= list_entry(trans
->r_itemq
.prev
,
3911 xlog_recover_item_t
, ri_list
);
3914 if (item
->ri_total
== 0) { /* first region to be added */
3915 if (in_f
->ilf_size
== 0 ||
3916 in_f
->ilf_size
> XLOG_MAX_REGIONS_IN_ITEM
) {
3918 "bad number of regions (%d) in inode log format",
3925 item
->ri_total
= in_f
->ilf_size
;
3927 kmem_zalloc(item
->ri_total
* sizeof(xfs_log_iovec_t
),
3930 ASSERT(item
->ri_total
> item
->ri_cnt
);
3931 /* Description region is ri_buf[0] */
3932 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
3933 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
3935 trace_xfs_log_recover_item_add(log
, trans
, item
, 0);
3940 * Free up any resources allocated by the transaction
3942 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3945 xlog_recover_free_trans(
3946 struct xlog_recover
*trans
)
3948 xlog_recover_item_t
*item
, *n
;
3951 list_for_each_entry_safe(item
, n
, &trans
->r_itemq
, ri_list
) {
3952 /* Free the regions in the item. */
3953 list_del(&item
->ri_list
);
3954 for (i
= 0; i
< item
->ri_cnt
; i
++)
3955 kmem_free(item
->ri_buf
[i
].i_addr
);
3956 /* Free the item itself */
3957 kmem_free(item
->ri_buf
);
3960 /* Free the transaction recover structure */
3965 * On error or completion, trans is freed.
3968 xlog_recovery_process_trans(
3970 struct xlog_recover
*trans
,
3977 bool freeit
= false;
3979 /* mask off ophdr transaction container flags */
3980 flags
&= ~XLOG_END_TRANS
;
3981 if (flags
& XLOG_WAS_CONT_TRANS
)
3982 flags
&= ~XLOG_CONTINUE_TRANS
;
3985 * Callees must not free the trans structure. We'll decide if we need to
3986 * free it or not based on the operation being done and it's result.
3989 /* expected flag values */
3991 case XLOG_CONTINUE_TRANS
:
3992 error
= xlog_recover_add_to_trans(log
, trans
, dp
, len
);
3994 case XLOG_WAS_CONT_TRANS
:
3995 error
= xlog_recover_add_to_cont_trans(log
, trans
, dp
, len
);
3997 case XLOG_COMMIT_TRANS
:
3998 error
= xlog_recover_commit_trans(log
, trans
, pass
);
3999 /* success or fail, we are now done with this transaction. */
4003 /* unexpected flag values */
4004 case XLOG_UNMOUNT_TRANS
:
4005 /* just skip trans */
4006 xfs_warn(log
->l_mp
, "%s: Unmount LR", __func__
);
4009 case XLOG_START_TRANS
:
4011 xfs_warn(log
->l_mp
, "%s: bad flag 0x%x", __func__
, flags
);
4016 if (error
|| freeit
)
4017 xlog_recover_free_trans(trans
);
4022 * Lookup the transaction recovery structure associated with the ID in the
4023 * current ophdr. If the transaction doesn't exist and the start flag is set in
4024 * the ophdr, then allocate a new transaction for future ID matches to find.
4025 * Either way, return what we found during the lookup - an existing transaction
4028 STATIC
struct xlog_recover
*
4029 xlog_recover_ophdr_to_trans(
4030 struct hlist_head rhash
[],
4031 struct xlog_rec_header
*rhead
,
4032 struct xlog_op_header
*ohead
)
4034 struct xlog_recover
*trans
;
4036 struct hlist_head
*rhp
;
4038 tid
= be32_to_cpu(ohead
->oh_tid
);
4039 rhp
= &rhash
[XLOG_RHASH(tid
)];
4040 hlist_for_each_entry(trans
, rhp
, r_list
) {
4041 if (trans
->r_log_tid
== tid
)
4046 * skip over non-start transaction headers - we could be
4047 * processing slack space before the next transaction starts
4049 if (!(ohead
->oh_flags
& XLOG_START_TRANS
))
4052 ASSERT(be32_to_cpu(ohead
->oh_len
) == 0);
4055 * This is a new transaction so allocate a new recovery container to
4056 * hold the recovery ops that will follow.
4058 trans
= kmem_zalloc(sizeof(struct xlog_recover
), KM_SLEEP
);
4059 trans
->r_log_tid
= tid
;
4060 trans
->r_lsn
= be64_to_cpu(rhead
->h_lsn
);
4061 INIT_LIST_HEAD(&trans
->r_itemq
);
4062 INIT_HLIST_NODE(&trans
->r_list
);
4063 hlist_add_head(&trans
->r_list
, rhp
);
4066 * Nothing more to do for this ophdr. Items to be added to this new
4067 * transaction will be in subsequent ophdr containers.
4073 xlog_recover_process_ophdr(
4075 struct hlist_head rhash
[],
4076 struct xlog_rec_header
*rhead
,
4077 struct xlog_op_header
*ohead
,
4082 struct xlog_recover
*trans
;
4085 /* Do we understand who wrote this op? */
4086 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
4087 ohead
->oh_clientid
!= XFS_LOG
) {
4088 xfs_warn(log
->l_mp
, "%s: bad clientid 0x%x",
4089 __func__
, ohead
->oh_clientid
);
4095 * Check the ophdr contains all the data it is supposed to contain.
4097 len
= be32_to_cpu(ohead
->oh_len
);
4098 if (dp
+ len
> end
) {
4099 xfs_warn(log
->l_mp
, "%s: bad length 0x%x", __func__
, len
);
4104 trans
= xlog_recover_ophdr_to_trans(rhash
, rhead
, ohead
);
4106 /* nothing to do, so skip over this ophdr */
4110 return xlog_recovery_process_trans(log
, trans
, dp
, len
,
4111 ohead
->oh_flags
, pass
);
4115 * There are two valid states of the r_state field. 0 indicates that the
4116 * transaction structure is in a normal state. We have either seen the
4117 * start of the transaction or the last operation we added was not a partial
4118 * operation. If the last operation we added to the transaction was a
4119 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
4121 * NOTE: skip LRs with 0 data length.
4124 xlog_recover_process_data(
4126 struct hlist_head rhash
[],
4127 struct xlog_rec_header
*rhead
,
4131 struct xlog_op_header
*ohead
;
4136 end
= dp
+ be32_to_cpu(rhead
->h_len
);
4137 num_logops
= be32_to_cpu(rhead
->h_num_logops
);
4139 /* check the log format matches our own - else we can't recover */
4140 if (xlog_header_check_recover(log
->l_mp
, rhead
))
4143 while ((dp
< end
) && num_logops
) {
4145 ohead
= (struct xlog_op_header
*)dp
;
4146 dp
+= sizeof(*ohead
);
4149 /* errors will abort recovery */
4150 error
= xlog_recover_process_ophdr(log
, rhash
, rhead
, ohead
,
4155 dp
+= be32_to_cpu(ohead
->oh_len
);
4162 * Process an extent free intent item that was recovered from
4163 * the log. We need to free the extents that it describes.
4166 xlog_recover_process_efi(
4168 xfs_efi_log_item_t
*efip
)
4170 xfs_efd_log_item_t
*efdp
;
4175 xfs_fsblock_t startblock_fsb
;
4177 ASSERT(!test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
));
4180 * First check the validity of the extents described by the
4181 * EFI. If any are bad, then assume that all are bad and
4182 * just toss the EFI.
4184 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
4185 extp
= &(efip
->efi_format
.efi_extents
[i
]);
4186 startblock_fsb
= XFS_BB_TO_FSB(mp
,
4187 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
4188 if ((startblock_fsb
== 0) ||
4189 (extp
->ext_len
== 0) ||
4190 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
4191 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
4193 * This will pull the EFI from the AIL and
4194 * free the memory associated with it.
4196 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
4197 xfs_efi_release(efip
);
4202 tp
= xfs_trans_alloc(mp
, 0);
4203 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
4206 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
4208 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
4209 extp
= &(efip
->efi_format
.efi_extents
[i
]);
4210 error
= xfs_trans_free_extent(tp
, efdp
, extp
->ext_start
,
4217 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
4218 error
= xfs_trans_commit(tp
);
4222 xfs_trans_cancel(tp
);
4227 * When this is called, all of the EFIs which did not have
4228 * corresponding EFDs should be in the AIL. What we do now
4229 * is free the extents associated with each one.
4231 * Since we process the EFIs in normal transactions, they
4232 * will be removed at some point after the commit. This prevents
4233 * us from just walking down the list processing each one.
4234 * We'll use a flag in the EFI to skip those that we've already
4235 * processed and use the AIL iteration mechanism's generation
4236 * count to try to speed this up at least a bit.
4238 * When we start, we know that the EFIs are the only things in
4239 * the AIL. As we process them, however, other items are added
4240 * to the AIL. Since everything added to the AIL must come after
4241 * everything already in the AIL, we stop processing as soon as
4242 * we see something other than an EFI in the AIL.
4245 xlog_recover_process_efis(
4248 struct xfs_log_item
*lip
;
4249 struct xfs_efi_log_item
*efip
;
4251 struct xfs_ail_cursor cur
;
4252 struct xfs_ail
*ailp
;
4255 spin_lock(&ailp
->xa_lock
);
4256 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
4257 while (lip
!= NULL
) {
4259 * We're done when we see something other than an EFI.
4260 * There should be no EFIs left in the AIL now.
4262 if (lip
->li_type
!= XFS_LI_EFI
) {
4264 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
4265 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
4271 * Skip EFIs that we've already processed.
4273 efip
= container_of(lip
, struct xfs_efi_log_item
, efi_item
);
4274 if (test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
)) {
4275 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
4279 spin_unlock(&ailp
->xa_lock
);
4280 error
= xlog_recover_process_efi(log
->l_mp
, efip
);
4281 spin_lock(&ailp
->xa_lock
);
4284 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
4287 xfs_trans_ail_cursor_done(&cur
);
4288 spin_unlock(&ailp
->xa_lock
);
4293 * A cancel occurs when the mount has failed and we're bailing out. Release all
4294 * pending EFIs so they don't pin the AIL.
4297 xlog_recover_cancel_efis(
4300 struct xfs_log_item
*lip
;
4301 struct xfs_efi_log_item
*efip
;
4303 struct xfs_ail_cursor cur
;
4304 struct xfs_ail
*ailp
;
4307 spin_lock(&ailp
->xa_lock
);
4308 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
4309 while (lip
!= NULL
) {
4311 * We're done when we see something other than an EFI.
4312 * There should be no EFIs left in the AIL now.
4314 if (lip
->li_type
!= XFS_LI_EFI
) {
4316 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
4317 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
4322 efip
= container_of(lip
, struct xfs_efi_log_item
, efi_item
);
4324 spin_unlock(&ailp
->xa_lock
);
4325 xfs_efi_release(efip
);
4326 spin_lock(&ailp
->xa_lock
);
4328 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
4331 xfs_trans_ail_cursor_done(&cur
);
4332 spin_unlock(&ailp
->xa_lock
);
4337 * This routine performs a transaction to null out a bad inode pointer
4338 * in an agi unlinked inode hash bucket.
4341 xlog_recover_clear_agi_bucket(
4343 xfs_agnumber_t agno
,
4352 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
4353 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_clearagi
, 0, 0);
4357 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
4361 agi
= XFS_BUF_TO_AGI(agibp
);
4362 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
4363 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
4364 (sizeof(xfs_agino_t
) * bucket
);
4365 xfs_trans_log_buf(tp
, agibp
, offset
,
4366 (offset
+ sizeof(xfs_agino_t
) - 1));
4368 error
= xfs_trans_commit(tp
);
4374 xfs_trans_cancel(tp
);
4376 xfs_warn(mp
, "%s: failed to clear agi %d. Continuing.", __func__
, agno
);
4381 xlog_recover_process_one_iunlink(
4382 struct xfs_mount
*mp
,
4383 xfs_agnumber_t agno
,
4387 struct xfs_buf
*ibp
;
4388 struct xfs_dinode
*dip
;
4389 struct xfs_inode
*ip
;
4393 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
4394 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
);
4399 * Get the on disk inode to find the next inode in the bucket.
4401 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &ibp
, 0, 0);
4405 ASSERT(ip
->i_d
.di_nlink
== 0);
4406 ASSERT(ip
->i_d
.di_mode
!= 0);
4408 /* setup for the next pass */
4409 agino
= be32_to_cpu(dip
->di_next_unlinked
);
4413 * Prevent any DMAPI event from being sent when the reference on
4414 * the inode is dropped.
4416 ip
->i_d
.di_dmevmask
= 0;
4425 * We can't read in the inode this bucket points to, or this inode
4426 * is messed up. Just ditch this bucket of inodes. We will lose
4427 * some inodes and space, but at least we won't hang.
4429 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
4430 * clear the inode pointer in the bucket.
4432 xlog_recover_clear_agi_bucket(mp
, agno
, bucket
);
4437 * xlog_iunlink_recover
4439 * This is called during recovery to process any inodes which
4440 * we unlinked but not freed when the system crashed. These
4441 * inodes will be on the lists in the AGI blocks. What we do
4442 * here is scan all the AGIs and fully truncate and free any
4443 * inodes found on the lists. Each inode is removed from the
4444 * lists when it has been fully truncated and is freed. The
4445 * freeing of the inode and its removal from the list must be
4449 xlog_recover_process_iunlinks(
4453 xfs_agnumber_t agno
;
4464 * Prevent any DMAPI event from being sent while in this function.
4466 mp_dmevmask
= mp
->m_dmevmask
;
4469 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
4471 * Find the agi for this ag.
4473 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
4476 * AGI is b0rked. Don't process it.
4478 * We should probably mark the filesystem as corrupt
4479 * after we've recovered all the ag's we can....
4484 * Unlock the buffer so that it can be acquired in the normal
4485 * course of the transaction to truncate and free each inode.
4486 * Because we are not racing with anyone else here for the AGI
4487 * buffer, we don't even need to hold it locked to read the
4488 * initial unlinked bucket entries out of the buffer. We keep
4489 * buffer reference though, so that it stays pinned in memory
4490 * while we need the buffer.
4492 agi
= XFS_BUF_TO_AGI(agibp
);
4493 xfs_buf_unlock(agibp
);
4495 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
4496 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
4497 while (agino
!= NULLAGINO
) {
4498 agino
= xlog_recover_process_one_iunlink(mp
,
4499 agno
, agino
, bucket
);
4502 xfs_buf_rele(agibp
);
4505 mp
->m_dmevmask
= mp_dmevmask
;
4510 struct xlog_rec_header
*rhead
,
4516 for (i
= 0; i
< BTOBB(be32_to_cpu(rhead
->h_len
)) &&
4517 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
4518 *(__be32
*)dp
= *(__be32
*)&rhead
->h_cycle_data
[i
];
4522 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
4523 xlog_in_core_2_t
*xhdr
= (xlog_in_core_2_t
*)rhead
;
4524 for ( ; i
< BTOBB(be32_to_cpu(rhead
->h_len
)); i
++) {
4525 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
4526 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
4527 *(__be32
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
4536 * CRC check, unpack and process a log record.
4539 xlog_recover_process(
4541 struct hlist_head rhash
[],
4542 struct xlog_rec_header
*rhead
,
4549 crc
= xlog_cksum(log
, rhead
, dp
, be32_to_cpu(rhead
->h_len
));
4552 * Nothing else to do if this is a CRC verification pass. Just return
4553 * if this a record with a non-zero crc. Unfortunately, mkfs always
4554 * sets h_crc to 0 so we must consider this valid even on v5 supers.
4555 * Otherwise, return EFSBADCRC on failure so the callers up the stack
4556 * know precisely what failed.
4558 if (pass
== XLOG_RECOVER_CRCPASS
) {
4559 if (rhead
->h_crc
&& crc
!= rhead
->h_crc
)
4565 * We're in the normal recovery path. Issue a warning if and only if the
4566 * CRC in the header is non-zero. This is an advisory warning and the
4567 * zero CRC check prevents warnings from being emitted when upgrading
4568 * the kernel from one that does not add CRCs by default.
4570 if (crc
!= rhead
->h_crc
) {
4571 if (rhead
->h_crc
|| xfs_sb_version_hascrc(&log
->l_mp
->m_sb
)) {
4572 xfs_alert(log
->l_mp
,
4573 "log record CRC mismatch: found 0x%x, expected 0x%x.",
4574 le32_to_cpu(rhead
->h_crc
),
4576 xfs_hex_dump(dp
, 32);
4580 * If the filesystem is CRC enabled, this mismatch becomes a
4581 * fatal log corruption failure.
4583 if (xfs_sb_version_hascrc(&log
->l_mp
->m_sb
))
4584 return -EFSCORRUPTED
;
4587 error
= xlog_unpack_data(rhead
, dp
, log
);
4591 return xlog_recover_process_data(log
, rhash
, rhead
, dp
, pass
);
4595 xlog_valid_rec_header(
4597 struct xlog_rec_header
*rhead
,
4602 if (unlikely(rhead
->h_magicno
!= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))) {
4603 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4604 XFS_ERRLEVEL_LOW
, log
->l_mp
);
4605 return -EFSCORRUPTED
;
4608 (!rhead
->h_version
||
4609 (be32_to_cpu(rhead
->h_version
) & (~XLOG_VERSION_OKBITS
))))) {
4610 xfs_warn(log
->l_mp
, "%s: unrecognised log version (%d).",
4611 __func__
, be32_to_cpu(rhead
->h_version
));
4615 /* LR body must have data or it wouldn't have been written */
4616 hlen
= be32_to_cpu(rhead
->h_len
);
4617 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
4618 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4619 XFS_ERRLEVEL_LOW
, log
->l_mp
);
4620 return -EFSCORRUPTED
;
4622 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
4623 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4624 XFS_ERRLEVEL_LOW
, log
->l_mp
);
4625 return -EFSCORRUPTED
;
4631 * Read the log from tail to head and process the log records found.
4632 * Handle the two cases where the tail and head are in the same cycle
4633 * and where the active portion of the log wraps around the end of
4634 * the physical log separately. The pass parameter is passed through
4635 * to the routines called to process the data and is not looked at
4639 xlog_do_recovery_pass(
4641 xfs_daddr_t head_blk
,
4642 xfs_daddr_t tail_blk
,
4644 xfs_daddr_t
*first_bad
) /* out: first bad log rec */
4646 xlog_rec_header_t
*rhead
;
4648 xfs_daddr_t rhead_blk
;
4650 xfs_buf_t
*hbp
, *dbp
;
4651 int error
= 0, h_size
, h_len
;
4652 int bblks
, split_bblks
;
4653 int hblks
, split_hblks
, wrapped_hblks
;
4654 struct hlist_head rhash
[XLOG_RHASH_SIZE
];
4656 ASSERT(head_blk
!= tail_blk
);
4660 * Read the header of the tail block and get the iclog buffer size from
4661 * h_size. Use this to tell how many sectors make up the log header.
4663 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
4665 * When using variable length iclogs, read first sector of
4666 * iclog header and extract the header size from it. Get a
4667 * new hbp that is the correct size.
4669 hbp
= xlog_get_bp(log
, 1);
4673 error
= xlog_bread(log
, tail_blk
, 1, hbp
, &offset
);
4677 rhead
= (xlog_rec_header_t
*)offset
;
4678 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
4683 * xfsprogs has a bug where record length is based on lsunit but
4684 * h_size (iclog size) is hardcoded to 32k. Now that we
4685 * unconditionally CRC verify the unmount record, this means the
4686 * log buffer can be too small for the record and cause an
4689 * Detect this condition here. Use lsunit for the buffer size as
4690 * long as this looks like the mkfs case. Otherwise, return an
4691 * error to avoid a buffer overrun.
4693 h_size
= be32_to_cpu(rhead
->h_size
);
4694 h_len
= be32_to_cpu(rhead
->h_len
);
4695 if (h_len
> h_size
) {
4696 if (h_len
<= log
->l_mp
->m_logbsize
&&
4697 be32_to_cpu(rhead
->h_num_logops
) == 1) {
4699 "invalid iclog size (%d bytes), using lsunit (%d bytes)",
4700 h_size
, log
->l_mp
->m_logbsize
);
4701 h_size
= log
->l_mp
->m_logbsize
;
4703 return -EFSCORRUPTED
;
4706 if ((be32_to_cpu(rhead
->h_version
) & XLOG_VERSION_2
) &&
4707 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
4708 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
4709 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
4712 hbp
= xlog_get_bp(log
, hblks
);
4717 ASSERT(log
->l_sectBBsize
== 1);
4719 hbp
= xlog_get_bp(log
, 1);
4720 h_size
= XLOG_BIG_RECORD_BSIZE
;
4725 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
4731 memset(rhash
, 0, sizeof(rhash
));
4732 blk_no
= rhead_blk
= tail_blk
;
4733 if (tail_blk
> head_blk
) {
4735 * Perform recovery around the end of the physical log.
4736 * When the head is not on the same cycle number as the tail,
4737 * we can't do a sequential recovery.
4739 while (blk_no
< log
->l_logBBsize
) {
4741 * Check for header wrapping around physical end-of-log
4743 offset
= hbp
->b_addr
;
4746 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
4747 /* Read header in one read */
4748 error
= xlog_bread(log
, blk_no
, hblks
, hbp
,
4753 /* This LR is split across physical log end */
4754 if (blk_no
!= log
->l_logBBsize
) {
4755 /* some data before physical log end */
4756 ASSERT(blk_no
<= INT_MAX
);
4757 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
4758 ASSERT(split_hblks
> 0);
4759 error
= xlog_bread(log
, blk_no
,
4767 * Note: this black magic still works with
4768 * large sector sizes (non-512) only because:
4769 * - we increased the buffer size originally
4770 * by 1 sector giving us enough extra space
4771 * for the second read;
4772 * - the log start is guaranteed to be sector
4774 * - we read the log end (LR header start)
4775 * _first_, then the log start (LR header end)
4776 * - order is important.
4778 wrapped_hblks
= hblks
- split_hblks
;
4779 error
= xlog_bread_offset(log
, 0,
4781 offset
+ BBTOB(split_hblks
));
4785 rhead
= (xlog_rec_header_t
*)offset
;
4786 error
= xlog_valid_rec_header(log
, rhead
,
4787 split_hblks
? blk_no
: 0);
4791 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
4794 /* Read in data for log record */
4795 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
4796 error
= xlog_bread(log
, blk_no
, bblks
, dbp
,
4801 /* This log record is split across the
4802 * physical end of log */
4803 offset
= dbp
->b_addr
;
4805 if (blk_no
!= log
->l_logBBsize
) {
4806 /* some data is before the physical
4808 ASSERT(!wrapped_hblks
);
4809 ASSERT(blk_no
<= INT_MAX
);
4811 log
->l_logBBsize
- (int)blk_no
;
4812 ASSERT(split_bblks
> 0);
4813 error
= xlog_bread(log
, blk_no
,
4821 * Note: this black magic still works with
4822 * large sector sizes (non-512) only because:
4823 * - we increased the buffer size originally
4824 * by 1 sector giving us enough extra space
4825 * for the second read;
4826 * - the log start is guaranteed to be sector
4828 * - we read the log end (LR header start)
4829 * _first_, then the log start (LR header end)
4830 * - order is important.
4832 error
= xlog_bread_offset(log
, 0,
4833 bblks
- split_bblks
, dbp
,
4834 offset
+ BBTOB(split_bblks
));
4839 error
= xlog_recover_process(log
, rhash
, rhead
, offset
,
4848 ASSERT(blk_no
>= log
->l_logBBsize
);
4849 blk_no
-= log
->l_logBBsize
;
4853 /* read first part of physical log */
4854 while (blk_no
< head_blk
) {
4855 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
4859 rhead
= (xlog_rec_header_t
*)offset
;
4860 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
4864 /* blocks in data section */
4865 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
4866 error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
,
4871 error
= xlog_recover_process(log
, rhash
, rhead
, offset
, pass
);
4875 blk_no
+= bblks
+ hblks
;
4884 if (error
&& first_bad
)
4885 *first_bad
= rhead_blk
;
4891 * Do the recovery of the log. We actually do this in two phases.
4892 * The two passes are necessary in order to implement the function
4893 * of cancelling a record written into the log. The first pass
4894 * determines those things which have been cancelled, and the
4895 * second pass replays log items normally except for those which
4896 * have been cancelled. The handling of the replay and cancellations
4897 * takes place in the log item type specific routines.
4899 * The table of items which have cancel records in the log is allocated
4900 * and freed at this level, since only here do we know when all of
4901 * the log recovery has been completed.
4904 xlog_do_log_recovery(
4906 xfs_daddr_t head_blk
,
4907 xfs_daddr_t tail_blk
)
4911 ASSERT(head_blk
!= tail_blk
);
4914 * First do a pass to find all of the cancelled buf log items.
4915 * Store them in the buf_cancel_table for use in the second pass.
4917 log
->l_buf_cancel_table
= kmem_zalloc(XLOG_BC_TABLE_SIZE
*
4918 sizeof(struct list_head
),
4920 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
4921 INIT_LIST_HEAD(&log
->l_buf_cancel_table
[i
]);
4923 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
4924 XLOG_RECOVER_PASS1
, NULL
);
4926 kmem_free(log
->l_buf_cancel_table
);
4927 log
->l_buf_cancel_table
= NULL
;
4931 * Then do a second pass to actually recover the items in the log.
4932 * When it is complete free the table of buf cancel items.
4934 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
4935 XLOG_RECOVER_PASS2
, NULL
);
4940 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
4941 ASSERT(list_empty(&log
->l_buf_cancel_table
[i
]));
4945 kmem_free(log
->l_buf_cancel_table
);
4946 log
->l_buf_cancel_table
= NULL
;
4952 * Do the actual recovery
4957 xfs_daddr_t head_blk
,
4958 xfs_daddr_t tail_blk
)
4965 * First replay the images in the log.
4967 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
4972 * If IO errors happened during recovery, bail out.
4974 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
4979 * We now update the tail_lsn since much of the recovery has completed
4980 * and there may be space available to use. If there were no extent
4981 * or iunlinks, we can free up the entire log and set the tail_lsn to
4982 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4983 * lsn of the last known good LR on disk. If there are extent frees
4984 * or iunlinks they will have some entries in the AIL; so we look at
4985 * the AIL to determine how to set the tail_lsn.
4987 xlog_assign_tail_lsn(log
->l_mp
);
4990 * Now that we've finished replaying all buffer and inode
4991 * updates, re-read in the superblock and reverify it.
4993 bp
= xfs_getsb(log
->l_mp
, 0);
4995 ASSERT(!(XFS_BUF_ISWRITE(bp
)));
4997 XFS_BUF_UNASYNC(bp
);
4998 bp
->b_ops
= &xfs_sb_buf_ops
;
5000 error
= xfs_buf_submit_wait(bp
);
5002 if (!XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
5003 xfs_buf_ioerror_alert(bp
, __func__
);
5010 /* Convert superblock from on-disk format */
5011 sbp
= &log
->l_mp
->m_sb
;
5012 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
5013 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
5014 ASSERT(xfs_sb_good_version(sbp
));
5015 xfs_reinit_percpu_counters(log
->l_mp
);
5020 xlog_recover_check_summary(log
);
5022 /* Normal transactions can now occur */
5023 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
5028 * Perform recovery and re-initialize some log variables in xlog_find_tail.
5030 * Return error or zero.
5036 xfs_daddr_t head_blk
, tail_blk
;
5039 /* find the tail of the log */
5040 error
= xlog_find_tail(log
, &head_blk
, &tail_blk
);
5045 * The superblock was read before the log was available and thus the LSN
5046 * could not be verified. Check the superblock LSN against the current
5047 * LSN now that it's known.
5049 if (xfs_sb_version_hascrc(&log
->l_mp
->m_sb
) &&
5050 !xfs_log_check_lsn(log
->l_mp
, log
->l_mp
->m_sb
.sb_lsn
))
5053 if (tail_blk
!= head_blk
) {
5054 /* There used to be a comment here:
5056 * disallow recovery on read-only mounts. note -- mount
5057 * checks for ENOSPC and turns it into an intelligent
5059 * ...but this is no longer true. Now, unless you specify
5060 * NORECOVERY (in which case this function would never be
5061 * called), we just go ahead and recover. We do this all
5062 * under the vfs layer, so we can get away with it unless
5063 * the device itself is read-only, in which case we fail.
5065 if ((error
= xfs_dev_is_read_only(log
->l_mp
, "recovery"))) {
5070 * Version 5 superblock log feature mask validation. We know the
5071 * log is dirty so check if there are any unknown log features
5072 * in what we need to recover. If there are unknown features
5073 * (e.g. unsupported transactions, then simply reject the
5074 * attempt at recovery before touching anything.
5076 if (XFS_SB_VERSION_NUM(&log
->l_mp
->m_sb
) == XFS_SB_VERSION_5
&&
5077 xfs_sb_has_incompat_log_feature(&log
->l_mp
->m_sb
,
5078 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
)) {
5080 "Superblock has unknown incompatible log features (0x%x) enabled.",
5081 (log
->l_mp
->m_sb
.sb_features_log_incompat
&
5082 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
));
5084 "The log can not be fully and/or safely recovered by this kernel.");
5086 "Please recover the log on a kernel that supports the unknown features.");
5091 * Delay log recovery if the debug hook is set. This is debug
5092 * instrumention to coordinate simulation of I/O failures with
5095 if (xfs_globals
.log_recovery_delay
) {
5096 xfs_notice(log
->l_mp
,
5097 "Delaying log recovery for %d seconds.",
5098 xfs_globals
.log_recovery_delay
);
5099 msleep(xfs_globals
.log_recovery_delay
* 1000);
5102 xfs_notice(log
->l_mp
, "Starting recovery (logdev: %s)",
5103 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
5106 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
5107 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
5113 * In the first part of recovery we replay inodes and buffers and build
5114 * up the list of extent free items which need to be processed. Here
5115 * we process the extent free items and clean up the on disk unlinked
5116 * inode lists. This is separated from the first part of recovery so
5117 * that the root and real-time bitmap inodes can be read in from disk in
5118 * between the two stages. This is necessary so that we can free space
5119 * in the real-time portion of the file system.
5122 xlog_recover_finish(
5126 * Now we're ready to do the transactions needed for the
5127 * rest of recovery. Start with completing all the extent
5128 * free intent records and then process the unlinked inode
5129 * lists. At this point, we essentially run in normal mode
5130 * except that we're still performing recovery actions
5131 * rather than accepting new requests.
5133 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
5135 error
= xlog_recover_process_efis(log
);
5137 xfs_alert(log
->l_mp
, "Failed to recover EFIs");
5141 * Sync the log to get all the EFIs out of the AIL.
5142 * This isn't absolutely necessary, but it helps in
5143 * case the unlink transactions would have problems
5144 * pushing the EFIs out of the way.
5146 xfs_log_force(log
->l_mp
, XFS_LOG_SYNC
);
5148 xlog_recover_process_iunlinks(log
);
5150 xlog_recover_check_summary(log
);
5152 xfs_notice(log
->l_mp
, "Ending recovery (logdev: %s)",
5153 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
5155 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
5157 xfs_info(log
->l_mp
, "Ending clean mount");
5163 xlog_recover_cancel(
5168 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
)
5169 error
= xlog_recover_cancel_efis(log
);
5176 * Read all of the agf and agi counters and check that they
5177 * are consistent with the superblock counters.
5180 xlog_recover_check_summary(
5187 xfs_agnumber_t agno
;
5188 __uint64_t freeblks
;
5198 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
5199 error
= xfs_read_agf(mp
, NULL
, agno
, 0, &agfbp
);
5201 xfs_alert(mp
, "%s agf read failed agno %d error %d",
5202 __func__
, agno
, error
);
5204 agfp
= XFS_BUF_TO_AGF(agfbp
);
5205 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
5206 be32_to_cpu(agfp
->agf_flcount
);
5207 xfs_buf_relse(agfbp
);
5210 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
5212 xfs_alert(mp
, "%s agi read failed agno %d error %d",
5213 __func__
, agno
, error
);
5215 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agibp
);
5217 itotal
+= be32_to_cpu(agi
->agi_count
);
5218 ifree
+= be32_to_cpu(agi
->agi_freecount
);
5219 xfs_buf_relse(agibp
);