2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_error.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_inode_item.h"
40 #include "xfs_alloc.h"
41 #include "xfs_ialloc.h"
42 #include "xfs_log_priv.h"
43 #include "xfs_buf_item.h"
44 #include "xfs_log_recover.h"
45 #include "xfs_extfree_item.h"
46 #include "xfs_trans_priv.h"
47 #include "xfs_quota.h"
50 STATIC
int xlog_find_zeroed(xlog_t
*, xfs_daddr_t
*);
51 STATIC
int xlog_clear_stale_blocks(xlog_t
*, xfs_lsn_t
);
52 STATIC
void xlog_recover_insert_item_backq(xlog_recover_item_t
**q
,
53 xlog_recover_item_t
*item
);
55 STATIC
void xlog_recover_check_summary(xlog_t
*);
56 STATIC
void xlog_recover_check_ail(xfs_mount_t
*, xfs_log_item_t
*, int);
58 #define xlog_recover_check_summary(log)
59 #define xlog_recover_check_ail(mp, lip, gen)
64 * Sector aligned buffer routines for buffer create/read/write/access
67 #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
68 ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
69 ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
70 #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
77 ASSERT(num_bblks
> 0);
79 if (log
->l_sectbb_log
) {
81 num_bblks
+= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, 1);
82 num_bblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, num_bblks
);
84 return xfs_buf_get_noaddr(BBTOB(num_bblks
), log
->l_mp
->m_logdev_targp
);
96 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
107 if (log
->l_sectbb_log
) {
108 blk_no
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, blk_no
);
109 nbblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, nbblks
);
113 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
116 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
119 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
120 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
122 xfsbdstrat(log
->l_mp
, bp
);
123 if ((error
= xfs_iowait(bp
)))
124 xfs_ioerror_alert("xlog_bread", log
->l_mp
,
125 bp
, XFS_BUF_ADDR(bp
));
130 * Write out the buffer at the given block for the given number of blocks.
131 * The buffer is kept locked across the write and is returned locked.
132 * This can only be used for synchronous log writes.
143 if (log
->l_sectbb_log
) {
144 blk_no
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, blk_no
);
145 nbblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, nbblks
);
149 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
151 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
152 XFS_BUF_ZEROFLAGS(bp
);
155 XFS_BUF_PSEMA(bp
, PRIBIO
);
156 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
157 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
159 if ((error
= xfs_bwrite(log
->l_mp
, bp
)))
160 xfs_ioerror_alert("xlog_bwrite", log
->l_mp
,
161 bp
, XFS_BUF_ADDR(bp
));
174 if (!log
->l_sectbb_log
)
175 return XFS_BUF_PTR(bp
);
177 ptr
= XFS_BUF_PTR(bp
) + BBTOB((int)blk_no
& log
->l_sectbb_mask
);
178 ASSERT(XFS_BUF_SIZE(bp
) >=
179 BBTOB(nbblks
+ (blk_no
& log
->l_sectbb_mask
)));
185 * dump debug superblock and log record information
188 xlog_header_check_dump(
190 xlog_rec_header_t
*head
)
194 cmn_err(CE_DEBUG
, "%s: SB : uuid = ", __FUNCTION__
);
195 for (b
= 0; b
< 16; b
++)
196 cmn_err(CE_DEBUG
, "%02x", ((uchar_t
*)&mp
->m_sb
.sb_uuid
)[b
]);
197 cmn_err(CE_DEBUG
, ", fmt = %d\n", XLOG_FMT
);
198 cmn_err(CE_DEBUG
, " log : uuid = ");
199 for (b
= 0; b
< 16; b
++)
200 cmn_err(CE_DEBUG
, "%02x",((uchar_t
*)&head
->h_fs_uuid
)[b
]);
201 cmn_err(CE_DEBUG
, ", fmt = %d\n", INT_GET(head
->h_fmt
, ARCH_CONVERT
));
204 #define xlog_header_check_dump(mp, head)
208 * check log record header for recovery
211 xlog_header_check_recover(
213 xlog_rec_header_t
*head
)
215 ASSERT(INT_GET(head
->h_magicno
, ARCH_CONVERT
) == XLOG_HEADER_MAGIC_NUM
);
218 * IRIX doesn't write the h_fmt field and leaves it zeroed
219 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
220 * a dirty log created in IRIX.
222 if (unlikely(INT_GET(head
->h_fmt
, ARCH_CONVERT
) != XLOG_FMT
)) {
224 "XFS: dirty log written in incompatible format - can't recover");
225 xlog_header_check_dump(mp
, head
);
226 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
227 XFS_ERRLEVEL_HIGH
, mp
);
228 return XFS_ERROR(EFSCORRUPTED
);
229 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
231 "XFS: dirty log entry has mismatched uuid - can't recover");
232 xlog_header_check_dump(mp
, head
);
233 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
234 XFS_ERRLEVEL_HIGH
, mp
);
235 return XFS_ERROR(EFSCORRUPTED
);
241 * read the head block of the log and check the header
244 xlog_header_check_mount(
246 xlog_rec_header_t
*head
)
248 ASSERT(INT_GET(head
->h_magicno
, ARCH_CONVERT
) == XLOG_HEADER_MAGIC_NUM
);
250 if (uuid_is_nil(&head
->h_fs_uuid
)) {
252 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
253 * h_fs_uuid is nil, we assume this log was last mounted
254 * by IRIX and continue.
256 xlog_warn("XFS: nil uuid in log - IRIX style log");
257 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
258 xlog_warn("XFS: log has mismatched uuid - can't recover");
259 xlog_header_check_dump(mp
, head
);
260 XFS_ERROR_REPORT("xlog_header_check_mount",
261 XFS_ERRLEVEL_HIGH
, mp
);
262 return XFS_ERROR(EFSCORRUPTED
);
273 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *));
275 if (XFS_BUF_GETERROR(bp
)) {
277 * We're not going to bother about retrying
278 * this during recovery. One strike!
280 mp
= XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*);
281 xfs_ioerror_alert("xlog_recover_iodone",
282 mp
, bp
, XFS_BUF_ADDR(bp
));
283 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
285 XFS_BUF_SET_FSPRIVATE(bp
, NULL
);
286 XFS_BUF_CLR_IODONE_FUNC(bp
);
291 * This routine finds (to an approximation) the first block in the physical
292 * log which contains the given cycle. It uses a binary search algorithm.
293 * Note that the algorithm can not be perfect because the disk will not
294 * necessarily be perfect.
297 xlog_find_cycle_start(
300 xfs_daddr_t first_blk
,
301 xfs_daddr_t
*last_blk
,
309 mid_blk
= BLK_AVG(first_blk
, *last_blk
);
310 while (mid_blk
!= first_blk
&& mid_blk
!= *last_blk
) {
311 if ((error
= xlog_bread(log
, mid_blk
, 1, bp
)))
313 offset
= xlog_align(log
, mid_blk
, 1, bp
);
314 mid_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
315 if (mid_cycle
== cycle
) {
317 /* last_half_cycle == mid_cycle */
320 /* first_half_cycle == mid_cycle */
322 mid_blk
= BLK_AVG(first_blk
, *last_blk
);
324 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == *last_blk
) ||
325 (mid_blk
== *last_blk
&& mid_blk
-1 == first_blk
));
331 * Check that the range of blocks does not contain the cycle number
332 * given. The scan needs to occur from front to back and the ptr into the
333 * region must be updated since a later routine will need to perform another
334 * test. If the region is completely good, we end up returning the same
337 * Set blkno to -1 if we encounter no errors. This is an invalid block number
338 * since we don't ever expect logs to get this large.
341 xlog_find_verify_cycle(
343 xfs_daddr_t start_blk
,
345 uint stop_on_cycle_no
,
346 xfs_daddr_t
*new_blk
)
352 xfs_caddr_t buf
= NULL
;
355 bufblks
= 1 << ffs(nbblks
);
357 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
358 /* can't get enough memory to do everything in one big buffer */
360 if (bufblks
<= log
->l_sectbb_log
)
364 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
367 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
369 if ((error
= xlog_bread(log
, i
, bcount
, bp
)))
372 buf
= xlog_align(log
, i
, bcount
, bp
);
373 for (j
= 0; j
< bcount
; j
++) {
374 cycle
= GET_CYCLE(buf
, ARCH_CONVERT
);
375 if (cycle
== stop_on_cycle_no
) {
392 * Potentially backup over partial log record write.
394 * In the typical case, last_blk is the number of the block directly after
395 * a good log record. Therefore, we subtract one to get the block number
396 * of the last block in the given buffer. extra_bblks contains the number
397 * of blocks we would have read on a previous read. This happens when the
398 * last log record is split over the end of the physical log.
400 * extra_bblks is the number of blocks potentially verified on a previous
401 * call to this routine.
404 xlog_find_verify_log_record(
406 xfs_daddr_t start_blk
,
407 xfs_daddr_t
*last_blk
,
412 xfs_caddr_t offset
= NULL
;
413 xlog_rec_header_t
*head
= NULL
;
416 int num_blks
= *last_blk
- start_blk
;
419 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
421 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
422 if (!(bp
= xlog_get_bp(log
, 1)))
426 if ((error
= xlog_bread(log
, start_blk
, num_blks
, bp
)))
428 offset
= xlog_align(log
, start_blk
, num_blks
, bp
);
429 offset
+= ((num_blks
- 1) << BBSHIFT
);
432 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
434 /* valid log record not found */
436 "XFS: Log inconsistent (didn't find previous header)");
438 error
= XFS_ERROR(EIO
);
443 if ((error
= xlog_bread(log
, i
, 1, bp
)))
445 offset
= xlog_align(log
, i
, 1, bp
);
448 head
= (xlog_rec_header_t
*)offset
;
450 if (XLOG_HEADER_MAGIC_NUM
==
451 INT_GET(head
->h_magicno
, ARCH_CONVERT
))
459 * We hit the beginning of the physical log & still no header. Return
460 * to caller. If caller can handle a return of -1, then this routine
461 * will be called again for the end of the physical log.
469 * We have the final block of the good log (the first block
470 * of the log record _before_ the head. So we check the uuid.
472 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
476 * We may have found a log record header before we expected one.
477 * last_blk will be the 1st block # with a given cycle #. We may end
478 * up reading an entire log record. In this case, we don't want to
479 * reset last_blk. Only when last_blk points in the middle of a log
480 * record do we update last_blk.
482 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
483 uint h_size
= INT_GET(head
->h_size
, ARCH_CONVERT
);
485 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
486 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
492 if (*last_blk
- i
+ extra_bblks
493 != BTOBB(INT_GET(head
->h_len
, ARCH_CONVERT
)) + xhdrs
)
502 * Head is defined to be the point of the log where the next log write
503 * write could go. This means that incomplete LR writes at the end are
504 * eliminated when calculating the head. We aren't guaranteed that previous
505 * LR have complete transactions. We only know that a cycle number of
506 * current cycle number -1 won't be present in the log if we start writing
507 * from our current block number.
509 * last_blk contains the block number of the first block with a given
512 * Return: zero if normal, non-zero if error.
517 xfs_daddr_t
*return_head_blk
)
521 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
523 uint first_half_cycle
, last_half_cycle
;
525 int error
, log_bbnum
= log
->l_logBBsize
;
527 /* Is the end of the log device zeroed? */
528 if ((error
= xlog_find_zeroed(log
, &first_blk
)) == -1) {
529 *return_head_blk
= first_blk
;
531 /* Is the whole lot zeroed? */
533 /* Linux XFS shouldn't generate totally zeroed logs -
534 * mkfs etc write a dummy unmount record to a fresh
535 * log so we can store the uuid in there
537 xlog_warn("XFS: totally zeroed log");
542 xlog_warn("XFS: empty log check failed");
546 first_blk
= 0; /* get cycle # of 1st block */
547 bp
= xlog_get_bp(log
, 1);
550 if ((error
= xlog_bread(log
, 0, 1, bp
)))
552 offset
= xlog_align(log
, 0, 1, bp
);
553 first_half_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
555 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
556 if ((error
= xlog_bread(log
, last_blk
, 1, bp
)))
558 offset
= xlog_align(log
, last_blk
, 1, bp
);
559 last_half_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
560 ASSERT(last_half_cycle
!= 0);
563 * If the 1st half cycle number is equal to the last half cycle number,
564 * then the entire log is stamped with the same cycle number. In this
565 * case, head_blk can't be set to zero (which makes sense). The below
566 * math doesn't work out properly with head_blk equal to zero. Instead,
567 * we set it to log_bbnum which is an invalid block number, but this
568 * value makes the math correct. If head_blk doesn't changed through
569 * all the tests below, *head_blk is set to zero at the very end rather
570 * than log_bbnum. In a sense, log_bbnum and zero are the same block
571 * in a circular file.
573 if (first_half_cycle
== last_half_cycle
) {
575 * In this case we believe that the entire log should have
576 * cycle number last_half_cycle. We need to scan backwards
577 * from the end verifying that there are no holes still
578 * containing last_half_cycle - 1. If we find such a hole,
579 * then the start of that hole will be the new head. The
580 * simple case looks like
581 * x | x ... | x - 1 | x
582 * Another case that fits this picture would be
583 * x | x + 1 | x ... | x
584 * In this case the head really is somewhere at the end of the
585 * log, as one of the latest writes at the beginning was
588 * x | x + 1 | x ... | x - 1 | x
589 * This is really the combination of the above two cases, and
590 * the head has to end up at the start of the x-1 hole at the
593 * In the 256k log case, we will read from the beginning to the
594 * end of the log and search for cycle numbers equal to x-1.
595 * We don't worry about the x+1 blocks that we encounter,
596 * because we know that they cannot be the head since the log
599 head_blk
= log_bbnum
;
600 stop_on_cycle
= last_half_cycle
- 1;
603 * In this case we want to find the first block with cycle
604 * number matching last_half_cycle. We expect the log to be
607 * The first block with cycle number x (last_half_cycle) will
608 * be where the new head belongs. First we do a binary search
609 * for the first occurrence of last_half_cycle. The binary
610 * search may not be totally accurate, so then we scan back
611 * from there looking for occurrences of last_half_cycle before
612 * us. If that backwards scan wraps around the beginning of
613 * the log, then we look for occurrences of last_half_cycle - 1
614 * at the end of the log. The cases we're looking for look
616 * x + 1 ... | x | x + 1 | x ...
617 * ^ binary search stopped here
619 * x + 1 ... | x ... | x - 1 | x
620 * <---------> less than scan distance
622 stop_on_cycle
= last_half_cycle
;
623 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
624 &head_blk
, last_half_cycle
)))
629 * Now validate the answer. Scan back some number of maximum possible
630 * blocks and make sure each one has the expected cycle number. The
631 * maximum is determined by the total possible amount of buffering
632 * in the in-core log. The following number can be made tighter if
633 * we actually look at the block size of the filesystem.
635 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
636 if (head_blk
>= num_scan_bblks
) {
638 * We are guaranteed that the entire check can be performed
641 start_blk
= head_blk
- num_scan_bblks
;
642 if ((error
= xlog_find_verify_cycle(log
,
643 start_blk
, num_scan_bblks
,
644 stop_on_cycle
, &new_blk
)))
648 } else { /* need to read 2 parts of log */
650 * We are going to scan backwards in the log in two parts.
651 * First we scan the physical end of the log. In this part
652 * of the log, we are looking for blocks with cycle number
653 * last_half_cycle - 1.
654 * If we find one, then we know that the log starts there, as
655 * we've found a hole that didn't get written in going around
656 * the end of the physical log. The simple case for this is
657 * x + 1 ... | x ... | x - 1 | x
658 * <---------> less than scan distance
659 * If all of the blocks at the end of the log have cycle number
660 * last_half_cycle, then we check the blocks at the start of
661 * the log looking for occurrences of last_half_cycle. If we
662 * find one, then our current estimate for the location of the
663 * first occurrence of last_half_cycle is wrong and we move
664 * back to the hole we've found. This case looks like
665 * x + 1 ... | x | x + 1 | x ...
666 * ^ binary search stopped here
667 * Another case we need to handle that only occurs in 256k
669 * x + 1 ... | x ... | x+1 | x ...
670 * ^ binary search stops here
671 * In a 256k log, the scan at the end of the log will see the
672 * x + 1 blocks. We need to skip past those since that is
673 * certainly not the head of the log. By searching for
674 * last_half_cycle-1 we accomplish that.
676 start_blk
= log_bbnum
- num_scan_bblks
+ head_blk
;
677 ASSERT(head_blk
<= INT_MAX
&&
678 (xfs_daddr_t
) num_scan_bblks
- head_blk
>= 0);
679 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
680 num_scan_bblks
- (int)head_blk
,
681 (stop_on_cycle
- 1), &new_blk
)))
689 * Scan beginning of log now. The last part of the physical
690 * log is good. This scan needs to verify that it doesn't find
691 * the last_half_cycle.
694 ASSERT(head_blk
<= INT_MAX
);
695 if ((error
= xlog_find_verify_cycle(log
,
696 start_blk
, (int)head_blk
,
697 stop_on_cycle
, &new_blk
)))
705 * Now we need to make sure head_blk is not pointing to a block in
706 * the middle of a log record.
708 num_scan_bblks
= XLOG_REC_SHIFT(log
);
709 if (head_blk
>= num_scan_bblks
) {
710 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
712 /* start ptr at last block ptr before head_blk */
713 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
714 &head_blk
, 0)) == -1) {
715 error
= XFS_ERROR(EIO
);
721 ASSERT(head_blk
<= INT_MAX
);
722 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
723 &head_blk
, 0)) == -1) {
724 /* We hit the beginning of the log during our search */
725 start_blk
= log_bbnum
- num_scan_bblks
+ head_blk
;
727 ASSERT(start_blk
<= INT_MAX
&&
728 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
729 ASSERT(head_blk
<= INT_MAX
);
730 if ((error
= xlog_find_verify_log_record(log
,
732 (int)head_blk
)) == -1) {
733 error
= XFS_ERROR(EIO
);
737 if (new_blk
!= log_bbnum
)
744 if (head_blk
== log_bbnum
)
745 *return_head_blk
= 0;
747 *return_head_blk
= head_blk
;
749 * When returning here, we have a good block number. Bad block
750 * means that during a previous crash, we didn't have a clean break
751 * from cycle number N to cycle number N-1. In this case, we need
752 * to find the first block with cycle number N-1.
760 xlog_warn("XFS: failed to find log head");
765 * Find the sync block number or the tail of the log.
767 * This will be the block number of the last record to have its
768 * associated buffers synced to disk. Every log record header has
769 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
770 * to get a sync block number. The only concern is to figure out which
771 * log record header to believe.
773 * The following algorithm uses the log record header with the largest
774 * lsn. The entire log record does not need to be valid. We only care
775 * that the header is valid.
777 * We could speed up search by using current head_blk buffer, but it is not
783 xfs_daddr_t
*head_blk
,
784 xfs_daddr_t
*tail_blk
)
786 xlog_rec_header_t
*rhead
;
787 xlog_op_header_t
*op_head
;
788 xfs_caddr_t offset
= NULL
;
791 xfs_daddr_t umount_data_blk
;
792 xfs_daddr_t after_umount_blk
;
799 * Find previous log record
801 if ((error
= xlog_find_head(log
, head_blk
)))
804 bp
= xlog_get_bp(log
, 1);
807 if (*head_blk
== 0) { /* special case */
808 if ((error
= xlog_bread(log
, 0, 1, bp
)))
810 offset
= xlog_align(log
, 0, 1, bp
);
811 if (GET_CYCLE(offset
, ARCH_CONVERT
) == 0) {
813 /* leave all other log inited values alone */
819 * Search backwards looking for log record header block
821 ASSERT(*head_blk
< INT_MAX
);
822 for (i
= (int)(*head_blk
) - 1; i
>= 0; i
--) {
823 if ((error
= xlog_bread(log
, i
, 1, bp
)))
825 offset
= xlog_align(log
, i
, 1, bp
);
826 if (XLOG_HEADER_MAGIC_NUM
==
827 INT_GET(*(uint
*)offset
, ARCH_CONVERT
)) {
833 * If we haven't found the log record header block, start looking
834 * again from the end of the physical log. XXXmiken: There should be
835 * a check here to make sure we didn't search more than N blocks in
839 for (i
= log
->l_logBBsize
- 1; i
>= (int)(*head_blk
); i
--) {
840 if ((error
= xlog_bread(log
, i
, 1, bp
)))
842 offset
= xlog_align(log
, i
, 1, bp
);
843 if (XLOG_HEADER_MAGIC_NUM
==
844 INT_GET(*(uint
*)offset
, ARCH_CONVERT
)) {
851 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
853 return XFS_ERROR(EIO
);
856 /* find blk_no of tail of log */
857 rhead
= (xlog_rec_header_t
*)offset
;
858 *tail_blk
= BLOCK_LSN(INT_GET(rhead
->h_tail_lsn
, ARCH_CONVERT
));
861 * Reset log values according to the state of the log when we
862 * crashed. In the case where head_blk == 0, we bump curr_cycle
863 * one because the next write starts a new cycle rather than
864 * continuing the cycle of the last good log record. At this
865 * point we have guaranteed that all partial log records have been
866 * accounted for. Therefore, we know that the last good log record
867 * written was complete and ended exactly on the end boundary
868 * of the physical log.
870 log
->l_prev_block
= i
;
871 log
->l_curr_block
= (int)*head_blk
;
872 log
->l_curr_cycle
= INT_GET(rhead
->h_cycle
, ARCH_CONVERT
);
875 log
->l_tail_lsn
= INT_GET(rhead
->h_tail_lsn
, ARCH_CONVERT
);
876 log
->l_last_sync_lsn
= INT_GET(rhead
->h_lsn
, ARCH_CONVERT
);
877 log
->l_grant_reserve_cycle
= log
->l_curr_cycle
;
878 log
->l_grant_reserve_bytes
= BBTOB(log
->l_curr_block
);
879 log
->l_grant_write_cycle
= log
->l_curr_cycle
;
880 log
->l_grant_write_bytes
= BBTOB(log
->l_curr_block
);
883 * Look for unmount record. If we find it, then we know there
884 * was a clean unmount. Since 'i' could be the last block in
885 * the physical log, we convert to a log block before comparing
888 * Save the current tail lsn to use to pass to
889 * xlog_clear_stale_blocks() below. We won't want to clear the
890 * unmount record if there is one, so we pass the lsn of the
891 * unmount record rather than the block after it.
893 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
894 int h_size
= INT_GET(rhead
->h_size
, ARCH_CONVERT
);
895 int h_version
= INT_GET(rhead
->h_version
, ARCH_CONVERT
);
897 if ((h_version
& XLOG_VERSION_2
) &&
898 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
899 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
900 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
908 after_umount_blk
= (i
+ hblks
+ (int)
909 BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
))) % log
->l_logBBsize
;
910 tail_lsn
= log
->l_tail_lsn
;
911 if (*head_blk
== after_umount_blk
&&
912 INT_GET(rhead
->h_num_logops
, ARCH_CONVERT
) == 1) {
913 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
914 if ((error
= xlog_bread(log
, umount_data_blk
, 1, bp
))) {
917 offset
= xlog_align(log
, umount_data_blk
, 1, bp
);
918 op_head
= (xlog_op_header_t
*)offset
;
919 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
921 * Set tail and last sync so that newly written
922 * log records will point recovery to after the
923 * current unmount record.
925 ASSIGN_ANY_LSN_HOST(log
->l_tail_lsn
, log
->l_curr_cycle
,
927 ASSIGN_ANY_LSN_HOST(log
->l_last_sync_lsn
, log
->l_curr_cycle
,
929 *tail_blk
= after_umount_blk
;
934 * Make sure that there are no blocks in front of the head
935 * with the same cycle number as the head. This can happen
936 * because we allow multiple outstanding log writes concurrently,
937 * and the later writes might make it out before earlier ones.
939 * We use the lsn from before modifying it so that we'll never
940 * overwrite the unmount record after a clean unmount.
942 * Do this only if we are going to recover the filesystem
944 * NOTE: This used to say "if (!readonly)"
945 * However on Linux, we can & do recover a read-only filesystem.
946 * We only skip recovery if NORECOVERY is specified on mount,
947 * in which case we would not be here.
949 * But... if the -device- itself is readonly, just skip this.
950 * We can't recover this device anyway, so it won't matter.
952 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
)) {
953 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
961 xlog_warn("XFS: failed to locate log tail");
966 * Is the log zeroed at all?
968 * The last binary search should be changed to perform an X block read
969 * once X becomes small enough. You can then search linearly through
970 * the X blocks. This will cut down on the number of reads we need to do.
972 * If the log is partially zeroed, this routine will pass back the blkno
973 * of the first block with cycle number 0. It won't have a complete LR
977 * 0 => the log is completely written to
978 * -1 => use *blk_no as the first block of the log
979 * >0 => error has occurred
988 uint first_cycle
, last_cycle
;
989 xfs_daddr_t new_blk
, last_blk
, start_blk
;
990 xfs_daddr_t num_scan_bblks
;
991 int error
, log_bbnum
= log
->l_logBBsize
;
993 /* check totally zeroed log */
994 bp
= xlog_get_bp(log
, 1);
997 if ((error
= xlog_bread(log
, 0, 1, bp
)))
999 offset
= xlog_align(log
, 0, 1, bp
);
1000 first_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
1001 if (first_cycle
== 0) { /* completely zeroed log */
1007 /* check partially zeroed log */
1008 if ((error
= xlog_bread(log
, log_bbnum
-1, 1, bp
)))
1010 offset
= xlog_align(log
, log_bbnum
-1, 1, bp
);
1011 last_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
1012 if (last_cycle
!= 0) { /* log completely written to */
1015 } else if (first_cycle
!= 1) {
1017 * If the cycle of the last block is zero, the cycle of
1018 * the first block must be 1. If it's not, maybe we're
1019 * not looking at a log... Bail out.
1021 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1022 return XFS_ERROR(EINVAL
);
1025 /* we have a partially zeroed log */
1026 last_blk
= log_bbnum
-1;
1027 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1031 * Validate the answer. Because there is no way to guarantee that
1032 * the entire log is made up of log records which are the same size,
1033 * we scan over the defined maximum blocks. At this point, the maximum
1034 * is not chosen to mean anything special. XXXmiken
1036 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1037 ASSERT(num_scan_bblks
<= INT_MAX
);
1039 if (last_blk
< num_scan_bblks
)
1040 num_scan_bblks
= last_blk
;
1041 start_blk
= last_blk
- num_scan_bblks
;
1044 * We search for any instances of cycle number 0 that occur before
1045 * our current estimate of the head. What we're trying to detect is
1046 * 1 ... | 0 | 1 | 0...
1047 * ^ binary search ends here
1049 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1050 (int)num_scan_bblks
, 0, &new_blk
)))
1056 * Potentially backup over partial log record write. We don't need
1057 * to search the end of the log because we know it is zero.
1059 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
1060 &last_blk
, 0)) == -1) {
1061 error
= XFS_ERROR(EIO
);
1075 * These are simple subroutines used by xlog_clear_stale_blocks() below
1076 * to initialize a buffer full of empty log record headers and write
1077 * them into the log.
1088 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1090 memset(buf
, 0, BBSIZE
);
1091 INT_SET(recp
->h_magicno
, ARCH_CONVERT
, XLOG_HEADER_MAGIC_NUM
);
1092 INT_SET(recp
->h_cycle
, ARCH_CONVERT
, cycle
);
1093 INT_SET(recp
->h_version
, ARCH_CONVERT
,
1094 XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
) ? 2 : 1);
1095 ASSIGN_ANY_LSN_DISK(recp
->h_lsn
, cycle
, block
);
1096 ASSIGN_ANY_LSN_DISK(recp
->h_tail_lsn
, tail_cycle
, tail_block
);
1097 INT_SET(recp
->h_fmt
, ARCH_CONVERT
, XLOG_FMT
);
1098 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1102 xlog_write_log_records(
1113 int sectbb
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, 1);
1114 int end_block
= start_block
+ blocks
;
1119 bufblks
= 1 << ffs(blocks
);
1120 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1122 if (bufblks
<= log
->l_sectbb_log
)
1126 /* We may need to do a read at the start to fill in part of
1127 * the buffer in the starting sector not covered by the first
1130 balign
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, start_block
);
1131 if (balign
!= start_block
) {
1132 if ((error
= xlog_bread(log
, start_block
, 1, bp
))) {
1136 j
= start_block
- balign
;
1139 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1140 int bcount
, endcount
;
1142 bcount
= min(bufblks
, end_block
- start_block
);
1143 endcount
= bcount
- j
;
1145 /* We may need to do a read at the end to fill in part of
1146 * the buffer in the final sector not covered by the write.
1147 * If this is the same sector as the above read, skip it.
1149 ealign
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, end_block
);
1150 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1151 offset
= XFS_BUF_PTR(bp
);
1152 balign
= BBTOB(ealign
- start_block
);
1153 XFS_BUF_SET_PTR(bp
, offset
+ balign
, BBTOB(sectbb
));
1154 if ((error
= xlog_bread(log
, ealign
, sectbb
, bp
)))
1156 XFS_BUF_SET_PTR(bp
, offset
, bufblks
);
1159 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1160 for (; j
< endcount
; j
++) {
1161 xlog_add_record(log
, offset
, cycle
, i
+j
,
1162 tail_cycle
, tail_block
);
1165 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1168 start_block
+= endcount
;
1176 * This routine is called to blow away any incomplete log writes out
1177 * in front of the log head. We do this so that we won't become confused
1178 * if we come up, write only a little bit more, and then crash again.
1179 * If we leave the partial log records out there, this situation could
1180 * cause us to think those partial writes are valid blocks since they
1181 * have the current cycle number. We get rid of them by overwriting them
1182 * with empty log records with the old cycle number rather than the
1185 * The tail lsn is passed in rather than taken from
1186 * the log so that we will not write over the unmount record after a
1187 * clean unmount in a 512 block log. Doing so would leave the log without
1188 * any valid log records in it until a new one was written. If we crashed
1189 * during that time we would not be able to recover.
1192 xlog_clear_stale_blocks(
1196 int tail_cycle
, head_cycle
;
1197 int tail_block
, head_block
;
1198 int tail_distance
, max_distance
;
1202 tail_cycle
= CYCLE_LSN(tail_lsn
);
1203 tail_block
= BLOCK_LSN(tail_lsn
);
1204 head_cycle
= log
->l_curr_cycle
;
1205 head_block
= log
->l_curr_block
;
1208 * Figure out the distance between the new head of the log
1209 * and the tail. We want to write over any blocks beyond the
1210 * head that we may have written just before the crash, but
1211 * we don't want to overwrite the tail of the log.
1213 if (head_cycle
== tail_cycle
) {
1215 * The tail is behind the head in the physical log,
1216 * so the distance from the head to the tail is the
1217 * distance from the head to the end of the log plus
1218 * the distance from the beginning of the log to the
1221 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1222 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1223 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1224 return XFS_ERROR(EFSCORRUPTED
);
1226 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1229 * The head is behind the tail in the physical log,
1230 * so the distance from the head to the tail is just
1231 * the tail block minus the head block.
1233 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1234 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1235 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1236 return XFS_ERROR(EFSCORRUPTED
);
1238 tail_distance
= tail_block
- head_block
;
1242 * If the head is right up against the tail, we can't clear
1245 if (tail_distance
<= 0) {
1246 ASSERT(tail_distance
== 0);
1250 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1252 * Take the smaller of the maximum amount of outstanding I/O
1253 * we could have and the distance to the tail to clear out.
1254 * We take the smaller so that we don't overwrite the tail and
1255 * we don't waste all day writing from the head to the tail
1258 max_distance
= MIN(max_distance
, tail_distance
);
1260 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1262 * We can stomp all the blocks we need to without
1263 * wrapping around the end of the log. Just do it
1264 * in a single write. Use the cycle number of the
1265 * current cycle minus one so that the log will look like:
1268 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1269 head_block
, max_distance
, tail_cycle
,
1275 * We need to wrap around the end of the physical log in
1276 * order to clear all the blocks. Do it in two separate
1277 * I/Os. The first write should be from the head to the
1278 * end of the physical log, and it should use the current
1279 * cycle number minus one just like above.
1281 distance
= log
->l_logBBsize
- head_block
;
1282 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1283 head_block
, distance
, tail_cycle
,
1290 * Now write the blocks at the start of the physical log.
1291 * This writes the remainder of the blocks we want to clear.
1292 * It uses the current cycle number since we're now on the
1293 * same cycle as the head so that we get:
1294 * n ... n ... | n - 1 ...
1295 * ^^^^^ blocks we're writing
1297 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1298 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1299 tail_cycle
, tail_block
);
1307 /******************************************************************************
1309 * Log recover routines
1311 ******************************************************************************
1314 STATIC xlog_recover_t
*
1315 xlog_recover_find_tid(
1319 xlog_recover_t
*p
= q
;
1322 if (p
->r_log_tid
== tid
)
1330 xlog_recover_put_hashq(
1332 xlog_recover_t
*trans
)
1339 xlog_recover_add_item(
1340 xlog_recover_item_t
**itemq
)
1342 xlog_recover_item_t
*item
;
1344 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
1345 xlog_recover_insert_item_backq(itemq
, item
);
1349 xlog_recover_add_to_cont_trans(
1350 xlog_recover_t
*trans
,
1354 xlog_recover_item_t
*item
;
1355 xfs_caddr_t ptr
, old_ptr
;
1358 item
= trans
->r_itemq
;
1360 /* finish copying rest of trans header */
1361 xlog_recover_add_item(&trans
->r_itemq
);
1362 ptr
= (xfs_caddr_t
) &trans
->r_theader
+
1363 sizeof(xfs_trans_header_t
) - len
;
1364 memcpy(ptr
, dp
, len
); /* d, s, l */
1367 item
= item
->ri_prev
;
1369 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
1370 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
1372 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, 0u);
1373 memcpy(&ptr
[old_len
], dp
, len
); /* d, s, l */
1374 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
1375 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
1380 * The next region to add is the start of a new region. It could be
1381 * a whole region or it could be the first part of a new region. Because
1382 * of this, the assumption here is that the type and size fields of all
1383 * format structures fit into the first 32 bits of the structure.
1385 * This works because all regions must be 32 bit aligned. Therefore, we
1386 * either have both fields or we have neither field. In the case we have
1387 * neither field, the data part of the region is zero length. We only have
1388 * a log_op_header and can throw away the header since a new one will appear
1389 * later. If we have at least 4 bytes, then we can determine how many regions
1390 * will appear in the current log item.
1393 xlog_recover_add_to_trans(
1394 xlog_recover_t
*trans
,
1398 xfs_inode_log_format_t
*in_f
; /* any will do */
1399 xlog_recover_item_t
*item
;
1404 item
= trans
->r_itemq
;
1406 ASSERT(*(uint
*)dp
== XFS_TRANS_HEADER_MAGIC
);
1407 if (len
== sizeof(xfs_trans_header_t
))
1408 xlog_recover_add_item(&trans
->r_itemq
);
1409 memcpy(&trans
->r_theader
, dp
, len
); /* d, s, l */
1413 ptr
= kmem_alloc(len
, KM_SLEEP
);
1414 memcpy(ptr
, dp
, len
);
1415 in_f
= (xfs_inode_log_format_t
*)ptr
;
1417 if (item
->ri_prev
->ri_total
!= 0 &&
1418 item
->ri_prev
->ri_total
== item
->ri_prev
->ri_cnt
) {
1419 xlog_recover_add_item(&trans
->r_itemq
);
1421 item
= trans
->r_itemq
;
1422 item
= item
->ri_prev
;
1424 if (item
->ri_total
== 0) { /* first region to be added */
1425 item
->ri_total
= in_f
->ilf_size
;
1426 ASSERT(item
->ri_total
<= XLOG_MAX_REGIONS_IN_ITEM
);
1427 item
->ri_buf
= kmem_zalloc((item
->ri_total
*
1428 sizeof(xfs_log_iovec_t
)), KM_SLEEP
);
1430 ASSERT(item
->ri_total
> item
->ri_cnt
);
1431 /* Description region is ri_buf[0] */
1432 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
1433 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
1439 xlog_recover_new_tid(
1444 xlog_recover_t
*trans
;
1446 trans
= kmem_zalloc(sizeof(xlog_recover_t
), KM_SLEEP
);
1447 trans
->r_log_tid
= tid
;
1449 xlog_recover_put_hashq(q
, trans
);
1453 xlog_recover_unlink_tid(
1455 xlog_recover_t
*trans
)
1466 if (tp
->r_next
== trans
) {
1474 "XFS: xlog_recover_unlink_tid: trans not found");
1476 return XFS_ERROR(EIO
);
1478 tp
->r_next
= tp
->r_next
->r_next
;
1484 xlog_recover_insert_item_backq(
1485 xlog_recover_item_t
**q
,
1486 xlog_recover_item_t
*item
)
1489 item
->ri_prev
= item
->ri_next
= item
;
1493 item
->ri_prev
= (*q
)->ri_prev
;
1494 (*q
)->ri_prev
= item
;
1495 item
->ri_prev
->ri_next
= item
;
1500 xlog_recover_insert_item_frontq(
1501 xlog_recover_item_t
**q
,
1502 xlog_recover_item_t
*item
)
1504 xlog_recover_insert_item_backq(q
, item
);
1509 xlog_recover_reorder_trans(
1511 xlog_recover_t
*trans
)
1513 xlog_recover_item_t
*first_item
, *itemq
, *itemq_next
;
1514 xfs_buf_log_format_t
*buf_f
;
1515 xfs_buf_log_format_v1_t
*obuf_f
;
1518 first_item
= itemq
= trans
->r_itemq
;
1519 trans
->r_itemq
= NULL
;
1521 itemq_next
= itemq
->ri_next
;
1522 buf_f
= (xfs_buf_log_format_t
*)itemq
->ri_buf
[0].i_addr
;
1523 switch (ITEM_TYPE(itemq
)) {
1525 flags
= buf_f
->blf_flags
;
1527 case XFS_LI_6_1_BUF
:
1528 case XFS_LI_5_3_BUF
:
1529 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1530 flags
= obuf_f
->blf_flags
;
1534 switch (ITEM_TYPE(itemq
)) {
1536 case XFS_LI_6_1_BUF
:
1537 case XFS_LI_5_3_BUF
:
1538 if (!(flags
& XFS_BLI_CANCEL
)) {
1539 xlog_recover_insert_item_frontq(&trans
->r_itemq
,
1544 case XFS_LI_6_1_INODE
:
1545 case XFS_LI_5_3_INODE
:
1547 case XFS_LI_QUOTAOFF
:
1550 xlog_recover_insert_item_backq(&trans
->r_itemq
, itemq
);
1554 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1556 return XFS_ERROR(EIO
);
1559 } while (first_item
!= itemq
);
1564 * Build up the table of buf cancel records so that we don't replay
1565 * cancelled data in the second pass. For buffer records that are
1566 * not cancel records, there is nothing to do here so we just return.
1568 * If we get a cancel record which is already in the table, this indicates
1569 * that the buffer was cancelled multiple times. In order to ensure
1570 * that during pass 2 we keep the record in the table until we reach its
1571 * last occurrence in the log, we keep a reference count in the cancel
1572 * record in the table to tell us how many times we expect to see this
1573 * record during the second pass.
1576 xlog_recover_do_buffer_pass1(
1578 xfs_buf_log_format_t
*buf_f
)
1580 xfs_buf_cancel_t
*bcp
;
1581 xfs_buf_cancel_t
*nextp
;
1582 xfs_buf_cancel_t
*prevp
;
1583 xfs_buf_cancel_t
**bucket
;
1584 xfs_buf_log_format_v1_t
*obuf_f
;
1585 xfs_daddr_t blkno
= 0;
1589 switch (buf_f
->blf_type
) {
1591 blkno
= buf_f
->blf_blkno
;
1592 len
= buf_f
->blf_len
;
1593 flags
= buf_f
->blf_flags
;
1595 case XFS_LI_6_1_BUF
:
1596 case XFS_LI_5_3_BUF
:
1597 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1598 blkno
= (xfs_daddr_t
) obuf_f
->blf_blkno
;
1599 len
= obuf_f
->blf_len
;
1600 flags
= obuf_f
->blf_flags
;
1605 * If this isn't a cancel buffer item, then just return.
1607 if (!(flags
& XFS_BLI_CANCEL
))
1611 * Insert an xfs_buf_cancel record into the hash table of
1612 * them. If there is already an identical record, bump
1613 * its reference count.
1615 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1616 XLOG_BC_TABLE_SIZE
];
1618 * If the hash bucket is empty then just insert a new record into
1621 if (*bucket
== NULL
) {
1622 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1624 bcp
->bc_blkno
= blkno
;
1626 bcp
->bc_refcount
= 1;
1627 bcp
->bc_next
= NULL
;
1633 * The hash bucket is not empty, so search for duplicates of our
1634 * record. If we find one them just bump its refcount. If not
1635 * then add us at the end of the list.
1639 while (nextp
!= NULL
) {
1640 if (nextp
->bc_blkno
== blkno
&& nextp
->bc_len
== len
) {
1641 nextp
->bc_refcount
++;
1645 nextp
= nextp
->bc_next
;
1647 ASSERT(prevp
!= NULL
);
1648 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1650 bcp
->bc_blkno
= blkno
;
1652 bcp
->bc_refcount
= 1;
1653 bcp
->bc_next
= NULL
;
1654 prevp
->bc_next
= bcp
;
1658 * Check to see whether the buffer being recovered has a corresponding
1659 * entry in the buffer cancel record table. If it does then return 1
1660 * so that it will be cancelled, otherwise return 0. If the buffer is
1661 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1662 * the refcount on the entry in the table and remove it from the table
1663 * if this is the last reference.
1665 * We remove the cancel record from the table when we encounter its
1666 * last occurrence in the log so that if the same buffer is re-used
1667 * again after its last cancellation we actually replay the changes
1668 * made at that point.
1671 xlog_check_buffer_cancelled(
1677 xfs_buf_cancel_t
*bcp
;
1678 xfs_buf_cancel_t
*prevp
;
1679 xfs_buf_cancel_t
**bucket
;
1681 if (log
->l_buf_cancel_table
== NULL
) {
1683 * There is nothing in the table built in pass one,
1684 * so this buffer must not be cancelled.
1686 ASSERT(!(flags
& XFS_BLI_CANCEL
));
1690 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1691 XLOG_BC_TABLE_SIZE
];
1695 * There is no corresponding entry in the table built
1696 * in pass one, so this buffer has not been cancelled.
1698 ASSERT(!(flags
& XFS_BLI_CANCEL
));
1703 * Search for an entry in the buffer cancel table that
1704 * matches our buffer.
1707 while (bcp
!= NULL
) {
1708 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
) {
1710 * We've go a match, so return 1 so that the
1711 * recovery of this buffer is cancelled.
1712 * If this buffer is actually a buffer cancel
1713 * log item, then decrement the refcount on the
1714 * one in the table and remove it if this is the
1717 if (flags
& XFS_BLI_CANCEL
) {
1719 if (bcp
->bc_refcount
== 0) {
1720 if (prevp
== NULL
) {
1721 *bucket
= bcp
->bc_next
;
1723 prevp
->bc_next
= bcp
->bc_next
;
1726 sizeof(xfs_buf_cancel_t
));
1735 * We didn't find a corresponding entry in the table, so
1736 * return 0 so that the buffer is NOT cancelled.
1738 ASSERT(!(flags
& XFS_BLI_CANCEL
));
1743 xlog_recover_do_buffer_pass2(
1745 xfs_buf_log_format_t
*buf_f
)
1747 xfs_buf_log_format_v1_t
*obuf_f
;
1748 xfs_daddr_t blkno
= 0;
1752 switch (buf_f
->blf_type
) {
1754 blkno
= buf_f
->blf_blkno
;
1755 flags
= buf_f
->blf_flags
;
1756 len
= buf_f
->blf_len
;
1758 case XFS_LI_6_1_BUF
:
1759 case XFS_LI_5_3_BUF
:
1760 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1761 blkno
= (xfs_daddr_t
) obuf_f
->blf_blkno
;
1762 flags
= obuf_f
->blf_flags
;
1763 len
= (xfs_daddr_t
) obuf_f
->blf_len
;
1767 return xlog_check_buffer_cancelled(log
, blkno
, len
, flags
);
1771 * Perform recovery for a buffer full of inodes. In these buffers,
1772 * the only data which should be recovered is that which corresponds
1773 * to the di_next_unlinked pointers in the on disk inode structures.
1774 * The rest of the data for the inodes is always logged through the
1775 * inodes themselves rather than the inode buffer and is recovered
1776 * in xlog_recover_do_inode_trans().
1778 * The only time when buffers full of inodes are fully recovered is
1779 * when the buffer is full of newly allocated inodes. In this case
1780 * the buffer will not be marked as an inode buffer and so will be
1781 * sent to xlog_recover_do_reg_buffer() below during recovery.
1784 xlog_recover_do_inode_buffer(
1786 xlog_recover_item_t
*item
,
1788 xfs_buf_log_format_t
*buf_f
)
1796 int next_unlinked_offset
;
1798 xfs_agino_t
*logged_nextp
;
1799 xfs_agino_t
*buffer_nextp
;
1800 xfs_buf_log_format_v1_t
*obuf_f
;
1801 unsigned int *data_map
= NULL
;
1802 unsigned int map_size
= 0;
1804 switch (buf_f
->blf_type
) {
1806 data_map
= buf_f
->blf_data_map
;
1807 map_size
= buf_f
->blf_map_size
;
1809 case XFS_LI_6_1_BUF
:
1810 case XFS_LI_5_3_BUF
:
1811 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1812 data_map
= obuf_f
->blf_data_map
;
1813 map_size
= obuf_f
->blf_map_size
;
1817 * Set the variables corresponding to the current region to
1818 * 0 so that we'll initialize them on the first pass through
1826 inodes_per_buf
= XFS_BUF_COUNT(bp
) >> mp
->m_sb
.sb_inodelog
;
1827 for (i
= 0; i
< inodes_per_buf
; i
++) {
1828 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1829 offsetof(xfs_dinode_t
, di_next_unlinked
);
1831 while (next_unlinked_offset
>=
1832 (reg_buf_offset
+ reg_buf_bytes
)) {
1834 * The next di_next_unlinked field is beyond
1835 * the current logged region. Find the next
1836 * logged region that contains or is beyond
1837 * the current di_next_unlinked field.
1840 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1843 * If there are no more logged regions in the
1844 * buffer, then we're done.
1850 nbits
= xfs_contig_bits(data_map
, map_size
,
1853 reg_buf_offset
= bit
<< XFS_BLI_SHIFT
;
1854 reg_buf_bytes
= nbits
<< XFS_BLI_SHIFT
;
1859 * If the current logged region starts after the current
1860 * di_next_unlinked field, then move on to the next
1861 * di_next_unlinked field.
1863 if (next_unlinked_offset
< reg_buf_offset
) {
1867 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1868 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLI_CHUNK
) == 0);
1869 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <= XFS_BUF_COUNT(bp
));
1872 * The current logged region contains a copy of the
1873 * current di_next_unlinked field. Extract its value
1874 * and copy it to the buffer copy.
1876 logged_nextp
= (xfs_agino_t
*)
1877 ((char *)(item
->ri_buf
[item_index
].i_addr
) +
1878 (next_unlinked_offset
- reg_buf_offset
));
1879 if (unlikely(*logged_nextp
== 0)) {
1880 xfs_fs_cmn_err(CE_ALERT
, mp
,
1881 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1883 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1884 XFS_ERRLEVEL_LOW
, mp
);
1885 return XFS_ERROR(EFSCORRUPTED
);
1888 buffer_nextp
= (xfs_agino_t
*)xfs_buf_offset(bp
,
1889 next_unlinked_offset
);
1890 *buffer_nextp
= *logged_nextp
;
1897 * Perform a 'normal' buffer recovery. Each logged region of the
1898 * buffer should be copied over the corresponding region in the
1899 * given buffer. The bitmap in the buf log format structure indicates
1900 * where to place the logged data.
1904 xlog_recover_do_reg_buffer(
1906 xlog_recover_item_t
*item
,
1908 xfs_buf_log_format_t
*buf_f
)
1913 xfs_buf_log_format_v1_t
*obuf_f
;
1914 unsigned int *data_map
= NULL
;
1915 unsigned int map_size
= 0;
1918 switch (buf_f
->blf_type
) {
1920 data_map
= buf_f
->blf_data_map
;
1921 map_size
= buf_f
->blf_map_size
;
1923 case XFS_LI_6_1_BUF
:
1924 case XFS_LI_5_3_BUF
:
1925 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1926 data_map
= obuf_f
->blf_data_map
;
1927 map_size
= obuf_f
->blf_map_size
;
1931 i
= 1; /* 0 is the buf format structure */
1933 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1936 nbits
= xfs_contig_bits(data_map
, map_size
, bit
);
1938 ASSERT(item
->ri_buf
[i
].i_addr
!= 0);
1939 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLI_CHUNK
== 0);
1940 ASSERT(XFS_BUF_COUNT(bp
) >=
1941 ((uint
)bit
<< XFS_BLI_SHIFT
)+(nbits
<<XFS_BLI_SHIFT
));
1944 * Do a sanity check if this is a dquot buffer. Just checking
1945 * the first dquot in the buffer should do. XXXThis is
1946 * probably a good thing to do for other buf types also.
1949 if (buf_f
->blf_flags
&
1950 (XFS_BLI_UDQUOT_BUF
|XFS_BLI_PDQUOT_BUF
|XFS_BLI_GDQUOT_BUF
)) {
1951 error
= xfs_qm_dqcheck((xfs_disk_dquot_t
*)
1952 item
->ri_buf
[i
].i_addr
,
1953 -1, 0, XFS_QMOPT_DOWARN
,
1954 "dquot_buf_recover");
1957 memcpy(xfs_buf_offset(bp
,
1958 (uint
)bit
<< XFS_BLI_SHIFT
), /* dest */
1959 item
->ri_buf
[i
].i_addr
, /* source */
1960 nbits
<<XFS_BLI_SHIFT
); /* length */
1965 /* Shouldn't be any more regions */
1966 ASSERT(i
== item
->ri_total
);
1970 * Do some primitive error checking on ondisk dquot data structures.
1974 xfs_disk_dquot_t
*ddq
,
1976 uint type
, /* used only when IO_dorepair is true */
1980 xfs_dqblk_t
*d
= (xfs_dqblk_t
*)ddq
;
1984 * We can encounter an uninitialized dquot buffer for 2 reasons:
1985 * 1. If we crash while deleting the quotainode(s), and those blks got
1986 * used for user data. This is because we take the path of regular
1987 * file deletion; however, the size field of quotainodes is never
1988 * updated, so all the tricks that we play in itruncate_finish
1989 * don't quite matter.
1991 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1992 * But the allocation will be replayed so we'll end up with an
1993 * uninitialized quota block.
1995 * This is all fine; things are still consistent, and we haven't lost
1996 * any quota information. Just don't complain about bad dquot blks.
1998 if (be16_to_cpu(ddq
->d_magic
) != XFS_DQUOT_MAGIC
) {
1999 if (flags
& XFS_QMOPT_DOWARN
)
2001 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
2002 str
, id
, be16_to_cpu(ddq
->d_magic
), XFS_DQUOT_MAGIC
);
2005 if (ddq
->d_version
!= XFS_DQUOT_VERSION
) {
2006 if (flags
& XFS_QMOPT_DOWARN
)
2008 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2009 str
, id
, ddq
->d_version
, XFS_DQUOT_VERSION
);
2013 if (ddq
->d_flags
!= XFS_DQ_USER
&&
2014 ddq
->d_flags
!= XFS_DQ_PROJ
&&
2015 ddq
->d_flags
!= XFS_DQ_GROUP
) {
2016 if (flags
& XFS_QMOPT_DOWARN
)
2018 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2019 str
, id
, ddq
->d_flags
);
2023 if (id
!= -1 && id
!= be32_to_cpu(ddq
->d_id
)) {
2024 if (flags
& XFS_QMOPT_DOWARN
)
2026 "%s : ondisk-dquot 0x%p, ID mismatch: "
2027 "0x%x expected, found id 0x%x",
2028 str
, ddq
, id
, be32_to_cpu(ddq
->d_id
));
2032 if (!errs
&& ddq
->d_id
) {
2033 if (ddq
->d_blk_softlimit
&&
2034 be64_to_cpu(ddq
->d_bcount
) >=
2035 be64_to_cpu(ddq
->d_blk_softlimit
)) {
2036 if (!ddq
->d_btimer
) {
2037 if (flags
& XFS_QMOPT_DOWARN
)
2039 "%s : Dquot ID 0x%x (0x%p) "
2040 "BLK TIMER NOT STARTED",
2041 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2045 if (ddq
->d_ino_softlimit
&&
2046 be64_to_cpu(ddq
->d_icount
) >=
2047 be64_to_cpu(ddq
->d_ino_softlimit
)) {
2048 if (!ddq
->d_itimer
) {
2049 if (flags
& XFS_QMOPT_DOWARN
)
2051 "%s : Dquot ID 0x%x (0x%p) "
2052 "INODE TIMER NOT STARTED",
2053 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2057 if (ddq
->d_rtb_softlimit
&&
2058 be64_to_cpu(ddq
->d_rtbcount
) >=
2059 be64_to_cpu(ddq
->d_rtb_softlimit
)) {
2060 if (!ddq
->d_rtbtimer
) {
2061 if (flags
& XFS_QMOPT_DOWARN
)
2063 "%s : Dquot ID 0x%x (0x%p) "
2064 "RTBLK TIMER NOT STARTED",
2065 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2071 if (!errs
|| !(flags
& XFS_QMOPT_DQREPAIR
))
2074 if (flags
& XFS_QMOPT_DOWARN
)
2075 cmn_err(CE_NOTE
, "Re-initializing dquot ID 0x%x", id
);
2078 * Typically, a repair is only requested by quotacheck.
2081 ASSERT(flags
& XFS_QMOPT_DQREPAIR
);
2082 memset(d
, 0, sizeof(xfs_dqblk_t
));
2084 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
2085 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
2086 d
->dd_diskdq
.d_flags
= type
;
2087 d
->dd_diskdq
.d_id
= cpu_to_be32(id
);
2093 * Perform a dquot buffer recovery.
2094 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2095 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2096 * Else, treat it as a regular buffer and do recovery.
2099 xlog_recover_do_dquot_buffer(
2102 xlog_recover_item_t
*item
,
2104 xfs_buf_log_format_t
*buf_f
)
2109 * Filesystems are required to send in quota flags at mount time.
2111 if (mp
->m_qflags
== 0) {
2116 if (buf_f
->blf_flags
& XFS_BLI_UDQUOT_BUF
)
2117 type
|= XFS_DQ_USER
;
2118 if (buf_f
->blf_flags
& XFS_BLI_PDQUOT_BUF
)
2119 type
|= XFS_DQ_PROJ
;
2120 if (buf_f
->blf_flags
& XFS_BLI_GDQUOT_BUF
)
2121 type
|= XFS_DQ_GROUP
;
2123 * This type of quotas was turned off, so ignore this buffer
2125 if (log
->l_quotaoffs_flag
& type
)
2128 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2132 * This routine replays a modification made to a buffer at runtime.
2133 * There are actually two types of buffer, regular and inode, which
2134 * are handled differently. Inode buffers are handled differently
2135 * in that we only recover a specific set of data from them, namely
2136 * the inode di_next_unlinked fields. This is because all other inode
2137 * data is actually logged via inode records and any data we replay
2138 * here which overlaps that may be stale.
2140 * When meta-data buffers are freed at run time we log a buffer item
2141 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2142 * of the buffer in the log should not be replayed at recovery time.
2143 * This is so that if the blocks covered by the buffer are reused for
2144 * file data before we crash we don't end up replaying old, freed
2145 * meta-data into a user's file.
2147 * To handle the cancellation of buffer log items, we make two passes
2148 * over the log during recovery. During the first we build a table of
2149 * those buffers which have been cancelled, and during the second we
2150 * only replay those buffers which do not have corresponding cancel
2151 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2152 * for more details on the implementation of the table of cancel records.
2155 xlog_recover_do_buffer_trans(
2157 xlog_recover_item_t
*item
,
2160 xfs_buf_log_format_t
*buf_f
;
2161 xfs_buf_log_format_v1_t
*obuf_f
;
2170 buf_f
= (xfs_buf_log_format_t
*)item
->ri_buf
[0].i_addr
;
2172 if (pass
== XLOG_RECOVER_PASS1
) {
2174 * In this pass we're only looking for buf items
2175 * with the XFS_BLI_CANCEL bit set.
2177 xlog_recover_do_buffer_pass1(log
, buf_f
);
2181 * In this pass we want to recover all the buffers
2182 * which have not been cancelled and are not
2183 * cancellation buffers themselves. The routine
2184 * we call here will tell us whether or not to
2185 * continue with the replay of this buffer.
2187 cancel
= xlog_recover_do_buffer_pass2(log
, buf_f
);
2192 switch (buf_f
->blf_type
) {
2194 blkno
= buf_f
->blf_blkno
;
2195 len
= buf_f
->blf_len
;
2196 flags
= buf_f
->blf_flags
;
2198 case XFS_LI_6_1_BUF
:
2199 case XFS_LI_5_3_BUF
:
2200 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
2201 blkno
= obuf_f
->blf_blkno
;
2202 len
= obuf_f
->blf_len
;
2203 flags
= obuf_f
->blf_flags
;
2206 xfs_fs_cmn_err(CE_ALERT
, log
->l_mp
,
2207 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2208 buf_f
->blf_type
, log
->l_mp
->m_logname
?
2209 log
->l_mp
->m_logname
: "internal");
2210 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2211 XFS_ERRLEVEL_LOW
, log
->l_mp
);
2212 return XFS_ERROR(EFSCORRUPTED
);
2216 if (flags
& XFS_BLI_INODE_BUF
) {
2217 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, blkno
, len
,
2220 bp
= xfs_buf_read(mp
->m_ddev_targp
, blkno
, len
, 0);
2222 if (XFS_BUF_ISERROR(bp
)) {
2223 xfs_ioerror_alert("xlog_recover_do..(read#1)", log
->l_mp
,
2225 error
= XFS_BUF_GETERROR(bp
);
2231 if (flags
& XFS_BLI_INODE_BUF
) {
2232 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2234 (XFS_BLI_UDQUOT_BUF
|XFS_BLI_PDQUOT_BUF
|XFS_BLI_GDQUOT_BUF
)) {
2235 xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2237 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2240 return XFS_ERROR(error
);
2243 * Perform delayed write on the buffer. Asynchronous writes will be
2244 * slower when taking into account all the buffers to be flushed.
2246 * Also make sure that only inode buffers with good sizes stay in
2247 * the buffer cache. The kernel moves inodes in buffers of 1 block
2248 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2249 * buffers in the log can be a different size if the log was generated
2250 * by an older kernel using unclustered inode buffers or a newer kernel
2251 * running with a different inode cluster size. Regardless, if the
2252 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2253 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2254 * the buffer out of the buffer cache so that the buffer won't
2255 * overlap with future reads of those inodes.
2257 if (XFS_DINODE_MAGIC
==
2258 INT_GET(*((__uint16_t
*)(xfs_buf_offset(bp
, 0))), ARCH_CONVERT
) &&
2259 (XFS_BUF_COUNT(bp
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2260 (__uint32_t
)XFS_INODE_CLUSTER_SIZE(log
->l_mp
)))) {
2262 error
= xfs_bwrite(mp
, bp
);
2264 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
||
2265 XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*) == mp
);
2266 XFS_BUF_SET_FSPRIVATE(bp
, mp
);
2267 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2268 xfs_bdwrite(mp
, bp
);
2275 xlog_recover_do_inode_trans(
2277 xlog_recover_item_t
*item
,
2280 xfs_inode_log_format_t
*in_f
;
2292 xfs_dinode_core_t
*dicp
;
2295 if (pass
== XLOG_RECOVER_PASS1
) {
2299 if (item
->ri_buf
[0].i_len
== sizeof(xfs_inode_log_format_t
)) {
2300 in_f
= (xfs_inode_log_format_t
*)item
->ri_buf
[0].i_addr
;
2302 in_f
= (xfs_inode_log_format_t
*)kmem_alloc(
2303 sizeof(xfs_inode_log_format_t
), KM_SLEEP
);
2305 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
2309 ino
= in_f
->ilf_ino
;
2311 if (ITEM_TYPE(item
) == XFS_LI_INODE
) {
2312 imap
.im_blkno
= (xfs_daddr_t
)in_f
->ilf_blkno
;
2313 imap
.im_len
= in_f
->ilf_len
;
2314 imap
.im_boffset
= in_f
->ilf_boffset
;
2317 * It's an old inode format record. We don't know where
2318 * its cluster is located on disk, and we can't allow
2319 * xfs_imap() to figure it out because the inode btrees
2320 * are not ready to be used. Therefore do not pass the
2321 * XFS_IMAP_LOOKUP flag to xfs_imap(). This will give
2322 * us only the single block in which the inode lives
2323 * rather than its cluster, so we must make sure to
2324 * invalidate the buffer when we write it out below.
2327 xfs_imap(log
->l_mp
, NULL
, ino
, &imap
, 0);
2331 * Inode buffers can be freed, look out for it,
2332 * and do not replay the inode.
2334 if (xlog_check_buffer_cancelled(log
, imap
.im_blkno
, imap
.im_len
, 0)) {
2339 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, imap
.im_blkno
, imap
.im_len
,
2341 if (XFS_BUF_ISERROR(bp
)) {
2342 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp
,
2344 error
= XFS_BUF_GETERROR(bp
);
2349 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2350 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
2353 * Make sure the place we're flushing out to really looks
2356 if (unlikely(INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) != XFS_DINODE_MAGIC
)) {
2358 xfs_fs_cmn_err(CE_ALERT
, mp
,
2359 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2361 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2362 XFS_ERRLEVEL_LOW
, mp
);
2363 error
= EFSCORRUPTED
;
2366 dicp
= (xfs_dinode_core_t
*)(item
->ri_buf
[1].i_addr
);
2367 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2369 xfs_fs_cmn_err(CE_ALERT
, mp
,
2370 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2372 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2373 XFS_ERRLEVEL_LOW
, mp
);
2374 error
= EFSCORRUPTED
;
2378 /* Skip replay when the on disk inode is newer than the log one */
2379 if (dicp
->di_flushiter
<
2380 INT_GET(dip
->di_core
.di_flushiter
, ARCH_CONVERT
)) {
2382 * Deal with the wrap case, DI_MAX_FLUSH is less
2383 * than smaller numbers
2385 if ((INT_GET(dip
->di_core
.di_flushiter
, ARCH_CONVERT
)
2387 (dicp
->di_flushiter
< (DI_MAX_FLUSH
>>1))) {
2395 /* Take the opportunity to reset the flush iteration count */
2396 dicp
->di_flushiter
= 0;
2398 if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFREG
)) {
2399 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2400 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2401 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2402 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2404 xfs_fs_cmn_err(CE_ALERT
, mp
,
2405 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2406 item
, dip
, bp
, ino
);
2407 error
= EFSCORRUPTED
;
2410 } else if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFDIR
)) {
2411 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2412 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2413 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2414 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2415 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2417 xfs_fs_cmn_err(CE_ALERT
, mp
,
2418 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2419 item
, dip
, bp
, ino
);
2420 error
= EFSCORRUPTED
;
2424 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2425 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2426 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2428 xfs_fs_cmn_err(CE_ALERT
, mp
,
2429 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2431 dicp
->di_nextents
+ dicp
->di_anextents
,
2433 error
= EFSCORRUPTED
;
2436 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2437 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2438 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2440 xfs_fs_cmn_err(CE_ALERT
, mp
,
2441 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2442 item
, dip
, bp
, ino
, dicp
->di_forkoff
);
2443 error
= EFSCORRUPTED
;
2446 if (unlikely(item
->ri_buf
[1].i_len
> sizeof(xfs_dinode_core_t
))) {
2447 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2448 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2450 xfs_fs_cmn_err(CE_ALERT
, mp
,
2451 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2452 item
->ri_buf
[1].i_len
, item
);
2453 error
= EFSCORRUPTED
;
2457 /* The core is in in-core format */
2458 xfs_xlate_dinode_core((xfs_caddr_t
)&dip
->di_core
,
2459 (xfs_dinode_core_t
*)item
->ri_buf
[1].i_addr
, -1);
2461 /* the rest is in on-disk format */
2462 if (item
->ri_buf
[1].i_len
> sizeof(xfs_dinode_core_t
)) {
2463 memcpy((xfs_caddr_t
) dip
+ sizeof(xfs_dinode_core_t
),
2464 item
->ri_buf
[1].i_addr
+ sizeof(xfs_dinode_core_t
),
2465 item
->ri_buf
[1].i_len
- sizeof(xfs_dinode_core_t
));
2468 fields
= in_f
->ilf_fields
;
2469 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
2471 INT_SET(dip
->di_u
.di_dev
, ARCH_CONVERT
, in_f
->ilf_u
.ilfu_rdev
);
2475 dip
->di_u
.di_muuid
= in_f
->ilf_u
.ilfu_uuid
;
2479 if (in_f
->ilf_size
== 2)
2480 goto write_inode_buffer
;
2481 len
= item
->ri_buf
[2].i_len
;
2482 src
= item
->ri_buf
[2].i_addr
;
2483 ASSERT(in_f
->ilf_size
<= 4);
2484 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
2485 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
2486 (len
== in_f
->ilf_dsize
));
2488 switch (fields
& XFS_ILOG_DFORK
) {
2489 case XFS_ILOG_DDATA
:
2491 memcpy(&dip
->di_u
, src
, len
);
2494 case XFS_ILOG_DBROOT
:
2495 xfs_bmbt_to_bmdr((xfs_bmbt_block_t
*)src
, len
,
2496 &(dip
->di_u
.di_bmbt
),
2497 XFS_DFORK_DSIZE(dip
, mp
));
2502 * There are no data fork flags set.
2504 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
2509 * If we logged any attribute data, recover it. There may or
2510 * may not have been any other non-core data logged in this
2513 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2514 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
2519 len
= item
->ri_buf
[attr_index
].i_len
;
2520 src
= item
->ri_buf
[attr_index
].i_addr
;
2521 ASSERT(len
== in_f
->ilf_asize
);
2523 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2524 case XFS_ILOG_ADATA
:
2526 dest
= XFS_DFORK_APTR(dip
);
2527 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
2528 memcpy(dest
, src
, len
);
2531 case XFS_ILOG_ABROOT
:
2532 dest
= XFS_DFORK_APTR(dip
);
2533 xfs_bmbt_to_bmdr((xfs_bmbt_block_t
*)src
, len
,
2534 (xfs_bmdr_block_t
*)dest
,
2535 XFS_DFORK_ASIZE(dip
, mp
));
2539 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2548 if (ITEM_TYPE(item
) == XFS_LI_INODE
) {
2549 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
||
2550 XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*) == mp
);
2551 XFS_BUF_SET_FSPRIVATE(bp
, mp
);
2552 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2553 xfs_bdwrite(mp
, bp
);
2556 error
= xfs_bwrite(mp
, bp
);
2561 kmem_free(in_f
, sizeof(*in_f
));
2562 return XFS_ERROR(error
);
2566 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2567 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2571 xlog_recover_do_quotaoff_trans(
2573 xlog_recover_item_t
*item
,
2576 xfs_qoff_logformat_t
*qoff_f
;
2578 if (pass
== XLOG_RECOVER_PASS2
) {
2582 qoff_f
= (xfs_qoff_logformat_t
*)item
->ri_buf
[0].i_addr
;
2586 * The logitem format's flag tells us if this was user quotaoff,
2587 * group/project quotaoff or both.
2589 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2590 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2591 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
2592 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
2593 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2594 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2600 * Recover a dquot record
2603 xlog_recover_do_dquot_trans(
2605 xlog_recover_item_t
*item
,
2610 struct xfs_disk_dquot
*ddq
, *recddq
;
2612 xfs_dq_logformat_t
*dq_f
;
2615 if (pass
== XLOG_RECOVER_PASS1
) {
2621 * Filesystems are required to send in quota flags at mount time.
2623 if (mp
->m_qflags
== 0)
2626 recddq
= (xfs_disk_dquot_t
*)item
->ri_buf
[1].i_addr
;
2629 * This type of quotas was turned off, so ignore this record.
2631 type
= INT_GET(recddq
->d_flags
, ARCH_CONVERT
) &
2632 (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
2634 if (log
->l_quotaoffs_flag
& type
)
2638 * At this point we know that quota was _not_ turned off.
2639 * Since the mount flags are not indicating to us otherwise, this
2640 * must mean that quota is on, and the dquot needs to be replayed.
2641 * Remember that we may not have fully recovered the superblock yet,
2642 * so we can't do the usual trick of looking at the SB quota bits.
2644 * The other possibility, of course, is that the quota subsystem was
2645 * removed since the last mount - ENOSYS.
2647 dq_f
= (xfs_dq_logformat_t
*)item
->ri_buf
[0].i_addr
;
2649 if ((error
= xfs_qm_dqcheck(recddq
,
2651 0, XFS_QMOPT_DOWARN
,
2652 "xlog_recover_do_dquot_trans (log copy)"))) {
2653 return XFS_ERROR(EIO
);
2655 ASSERT(dq_f
->qlf_len
== 1);
2657 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
2659 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
),
2662 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp
,
2663 bp
, dq_f
->qlf_blkno
);
2667 ddq
= (xfs_disk_dquot_t
*)xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
2670 * At least the magic num portion should be on disk because this
2671 * was among a chunk of dquots created earlier, and we did some
2672 * minimal initialization then.
2674 if (xfs_qm_dqcheck(ddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2675 "xlog_recover_do_dquot_trans")) {
2677 return XFS_ERROR(EIO
);
2680 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
2682 ASSERT(dq_f
->qlf_size
== 2);
2683 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
||
2684 XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*) == mp
);
2685 XFS_BUF_SET_FSPRIVATE(bp
, mp
);
2686 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2687 xfs_bdwrite(mp
, bp
);
2693 * This routine is called to create an in-core extent free intent
2694 * item from the efi format structure which was logged on disk.
2695 * It allocates an in-core efi, copies the extents from the format
2696 * structure into it, and adds the efi to the AIL with the given
2700 xlog_recover_do_efi_trans(
2702 xlog_recover_item_t
*item
,
2708 xfs_efi_log_item_t
*efip
;
2709 xfs_efi_log_format_t
*efi_formatp
;
2712 if (pass
== XLOG_RECOVER_PASS1
) {
2716 efi_formatp
= (xfs_efi_log_format_t
*)item
->ri_buf
[0].i_addr
;
2719 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
2720 if ((error
= xfs_efi_copy_format(&(item
->ri_buf
[0]),
2721 &(efip
->efi_format
)))) {
2722 xfs_efi_item_free(efip
);
2725 efip
->efi_next_extent
= efi_formatp
->efi_nextents
;
2726 efip
->efi_flags
|= XFS_EFI_COMMITTED
;
2730 * xfs_trans_update_ail() drops the AIL lock.
2732 xfs_trans_update_ail(mp
, (xfs_log_item_t
*)efip
, lsn
, s
);
2738 * This routine is called when an efd format structure is found in
2739 * a committed transaction in the log. It's purpose is to cancel
2740 * the corresponding efi if it was still in the log. To do this
2741 * it searches the AIL for the efi with an id equal to that in the
2742 * efd format structure. If we find it, we remove the efi from the
2746 xlog_recover_do_efd_trans(
2748 xlog_recover_item_t
*item
,
2752 xfs_efd_log_format_t
*efd_formatp
;
2753 xfs_efi_log_item_t
*efip
= NULL
;
2754 xfs_log_item_t
*lip
;
2759 if (pass
== XLOG_RECOVER_PASS1
) {
2763 efd_formatp
= (xfs_efd_log_format_t
*)item
->ri_buf
[0].i_addr
;
2764 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
2765 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
2766 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
2767 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
2768 efi_id
= efd_formatp
->efd_efi_id
;
2771 * Search for the efi with the id in the efd format structure
2776 lip
= xfs_trans_first_ail(mp
, &gen
);
2777 while (lip
!= NULL
) {
2778 if (lip
->li_type
== XFS_LI_EFI
) {
2779 efip
= (xfs_efi_log_item_t
*)lip
;
2780 if (efip
->efi_format
.efi_id
== efi_id
) {
2782 * xfs_trans_delete_ail() drops the
2785 xfs_trans_delete_ail(mp
, lip
, s
);
2789 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
2793 * If we found it, then free it up. If it wasn't there, it
2794 * must have been overwritten in the log. Oh well.
2797 xfs_efi_item_free(efip
);
2804 * Perform the transaction
2806 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2807 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2810 xlog_recover_do_trans(
2812 xlog_recover_t
*trans
,
2816 xlog_recover_item_t
*item
, *first_item
;
2818 if ((error
= xlog_recover_reorder_trans(log
, trans
)))
2820 first_item
= item
= trans
->r_itemq
;
2823 * we don't need to worry about the block number being
2824 * truncated in > 1 TB buffers because in user-land,
2825 * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so
2826 * the blknos will get through the user-mode buffer
2827 * cache properly. The only bad case is o32 kernels
2828 * where xfs_daddr_t is 32-bits but mount will warn us
2829 * off a > 1 TB filesystem before we get here.
2831 if ((ITEM_TYPE(item
) == XFS_LI_BUF
) ||
2832 (ITEM_TYPE(item
) == XFS_LI_6_1_BUF
) ||
2833 (ITEM_TYPE(item
) == XFS_LI_5_3_BUF
)) {
2834 if ((error
= xlog_recover_do_buffer_trans(log
, item
,
2837 } else if ((ITEM_TYPE(item
) == XFS_LI_INODE
)) {
2838 if ((error
= xlog_recover_do_inode_trans(log
, item
,
2841 } else if (ITEM_TYPE(item
) == XFS_LI_EFI
) {
2842 if ((error
= xlog_recover_do_efi_trans(log
, item
, trans
->r_lsn
,
2845 } else if (ITEM_TYPE(item
) == XFS_LI_EFD
) {
2846 xlog_recover_do_efd_trans(log
, item
, pass
);
2847 } else if (ITEM_TYPE(item
) == XFS_LI_DQUOT
) {
2848 if ((error
= xlog_recover_do_dquot_trans(log
, item
,
2851 } else if ((ITEM_TYPE(item
) == XFS_LI_QUOTAOFF
)) {
2852 if ((error
= xlog_recover_do_quotaoff_trans(log
, item
,
2856 xlog_warn("XFS: xlog_recover_do_trans");
2858 error
= XFS_ERROR(EIO
);
2861 item
= item
->ri_next
;
2862 } while (first_item
!= item
);
2868 * Free up any resources allocated by the transaction
2870 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2873 xlog_recover_free_trans(
2874 xlog_recover_t
*trans
)
2876 xlog_recover_item_t
*first_item
, *item
, *free_item
;
2879 item
= first_item
= trans
->r_itemq
;
2882 item
= item
->ri_next
;
2883 /* Free the regions in the item. */
2884 for (i
= 0; i
< free_item
->ri_cnt
; i
++) {
2885 kmem_free(free_item
->ri_buf
[i
].i_addr
,
2886 free_item
->ri_buf
[i
].i_len
);
2888 /* Free the item itself */
2889 kmem_free(free_item
->ri_buf
,
2890 (free_item
->ri_total
* sizeof(xfs_log_iovec_t
)));
2891 kmem_free(free_item
, sizeof(xlog_recover_item_t
));
2892 } while (first_item
!= item
);
2893 /* Free the transaction recover structure */
2894 kmem_free(trans
, sizeof(xlog_recover_t
));
2898 xlog_recover_commit_trans(
2901 xlog_recover_t
*trans
,
2906 if ((error
= xlog_recover_unlink_tid(q
, trans
)))
2908 if ((error
= xlog_recover_do_trans(log
, trans
, pass
)))
2910 xlog_recover_free_trans(trans
); /* no error */
2915 xlog_recover_unmount_trans(
2916 xlog_recover_t
*trans
)
2918 /* Do nothing now */
2919 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2924 * There are two valid states of the r_state field. 0 indicates that the
2925 * transaction structure is in a normal state. We have either seen the
2926 * start of the transaction or the last operation we added was not a partial
2927 * operation. If the last operation we added to the transaction was a
2928 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2930 * NOTE: skip LRs with 0 data length.
2933 xlog_recover_process_data(
2935 xlog_recover_t
*rhash
[],
2936 xlog_rec_header_t
*rhead
,
2942 xlog_op_header_t
*ohead
;
2943 xlog_recover_t
*trans
;
2949 lp
= dp
+ INT_GET(rhead
->h_len
, ARCH_CONVERT
);
2950 num_logops
= INT_GET(rhead
->h_num_logops
, ARCH_CONVERT
);
2952 /* check the log format matches our own - else we can't recover */
2953 if (xlog_header_check_recover(log
->l_mp
, rhead
))
2954 return (XFS_ERROR(EIO
));
2956 while ((dp
< lp
) && num_logops
) {
2957 ASSERT(dp
+ sizeof(xlog_op_header_t
) <= lp
);
2958 ohead
= (xlog_op_header_t
*)dp
;
2959 dp
+= sizeof(xlog_op_header_t
);
2960 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
2961 ohead
->oh_clientid
!= XFS_LOG
) {
2963 "XFS: xlog_recover_process_data: bad clientid");
2965 return (XFS_ERROR(EIO
));
2967 tid
= INT_GET(ohead
->oh_tid
, ARCH_CONVERT
);
2968 hash
= XLOG_RHASH(tid
);
2969 trans
= xlog_recover_find_tid(rhash
[hash
], tid
);
2970 if (trans
== NULL
) { /* not found; add new tid */
2971 if (ohead
->oh_flags
& XLOG_START_TRANS
)
2972 xlog_recover_new_tid(&rhash
[hash
], tid
,
2973 INT_GET(rhead
->h_lsn
, ARCH_CONVERT
));
2975 ASSERT(dp
+INT_GET(ohead
->oh_len
, ARCH_CONVERT
) <= lp
);
2976 flags
= ohead
->oh_flags
& ~XLOG_END_TRANS
;
2977 if (flags
& XLOG_WAS_CONT_TRANS
)
2978 flags
&= ~XLOG_CONTINUE_TRANS
;
2980 case XLOG_COMMIT_TRANS
:
2981 error
= xlog_recover_commit_trans(log
,
2982 &rhash
[hash
], trans
, pass
);
2984 case XLOG_UNMOUNT_TRANS
:
2985 error
= xlog_recover_unmount_trans(trans
);
2987 case XLOG_WAS_CONT_TRANS
:
2988 error
= xlog_recover_add_to_cont_trans(trans
,
2989 dp
, INT_GET(ohead
->oh_len
,
2992 case XLOG_START_TRANS
:
2994 "XFS: xlog_recover_process_data: bad transaction");
2996 error
= XFS_ERROR(EIO
);
2999 case XLOG_CONTINUE_TRANS
:
3000 error
= xlog_recover_add_to_trans(trans
,
3001 dp
, INT_GET(ohead
->oh_len
,
3006 "XFS: xlog_recover_process_data: bad flag");
3008 error
= XFS_ERROR(EIO
);
3014 dp
+= INT_GET(ohead
->oh_len
, ARCH_CONVERT
);
3021 * Process an extent free intent item that was recovered from
3022 * the log. We need to free the extents that it describes.
3025 xlog_recover_process_efi(
3027 xfs_efi_log_item_t
*efip
)
3029 xfs_efd_log_item_t
*efdp
;
3033 xfs_fsblock_t startblock_fsb
;
3035 ASSERT(!(efip
->efi_flags
& XFS_EFI_RECOVERED
));
3038 * First check the validity of the extents described by the
3039 * EFI. If any are bad, then assume that all are bad and
3040 * just toss the EFI.
3042 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3043 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3044 startblock_fsb
= XFS_BB_TO_FSB(mp
,
3045 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
3046 if ((startblock_fsb
== 0) ||
3047 (extp
->ext_len
== 0) ||
3048 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
3049 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
3051 * This will pull the EFI from the AIL and
3052 * free the memory associated with it.
3054 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
3059 tp
= xfs_trans_alloc(mp
, 0);
3060 xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0, 0, 0);
3061 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
3063 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3064 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3065 xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
3066 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
3070 efip
->efi_flags
|= XFS_EFI_RECOVERED
;
3071 xfs_trans_commit(tp
, 0, NULL
);
3075 * Verify that once we've encountered something other than an EFI
3076 * in the AIL that there are no more EFIs in the AIL.
3080 xlog_recover_check_ail(
3082 xfs_log_item_t
*lip
,
3088 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
3089 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
3091 * The check will be bogus if we restart from the
3092 * beginning of the AIL, so ASSERT that we don't.
3093 * We never should since we're holding the AIL lock
3096 ASSERT(gen
== orig_gen
);
3097 } while (lip
!= NULL
);
3102 * When this is called, all of the EFIs which did not have
3103 * corresponding EFDs should be in the AIL. What we do now
3104 * is free the extents associated with each one.
3106 * Since we process the EFIs in normal transactions, they
3107 * will be removed at some point after the commit. This prevents
3108 * us from just walking down the list processing each one.
3109 * We'll use a flag in the EFI to skip those that we've already
3110 * processed and use the AIL iteration mechanism's generation
3111 * count to try to speed this up at least a bit.
3113 * When we start, we know that the EFIs are the only things in
3114 * the AIL. As we process them, however, other items are added
3115 * to the AIL. Since everything added to the AIL must come after
3116 * everything already in the AIL, we stop processing as soon as
3117 * we see something other than an EFI in the AIL.
3120 xlog_recover_process_efis(
3123 xfs_log_item_t
*lip
;
3124 xfs_efi_log_item_t
*efip
;
3132 lip
= xfs_trans_first_ail(mp
, &gen
);
3133 while (lip
!= NULL
) {
3135 * We're done when we see something other than an EFI.
3137 if (lip
->li_type
!= XFS_LI_EFI
) {
3138 xlog_recover_check_ail(mp
, lip
, gen
);
3143 * Skip EFIs that we've already processed.
3145 efip
= (xfs_efi_log_item_t
*)lip
;
3146 if (efip
->efi_flags
& XFS_EFI_RECOVERED
) {
3147 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
3152 xlog_recover_process_efi(mp
, efip
);
3154 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
3160 * This routine performs a transaction to null out a bad inode pointer
3161 * in an agi unlinked inode hash bucket.
3164 xlog_recover_clear_agi_bucket(
3166 xfs_agnumber_t agno
,
3175 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3176 xfs_trans_reserve(tp
, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp
), 0, 0, 0);
3178 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
3179 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
3180 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
3182 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3186 agi
= XFS_BUF_TO_AGI(agibp
);
3187 if (be32_to_cpu(agi
->agi_magicnum
) != XFS_AGI_MAGIC
) {
3188 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3192 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
3193 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3194 (sizeof(xfs_agino_t
) * bucket
);
3195 xfs_trans_log_buf(tp
, agibp
, offset
,
3196 (offset
+ sizeof(xfs_agino_t
) - 1));
3198 (void) xfs_trans_commit(tp
, 0, NULL
);
3202 * xlog_iunlink_recover
3204 * This is called during recovery to process any inodes which
3205 * we unlinked but not freed when the system crashed. These
3206 * inodes will be on the lists in the AGI blocks. What we do
3207 * here is scan all the AGIs and fully truncate and free any
3208 * inodes found on the lists. Each inode is removed from the
3209 * lists when it has been fully truncated and is freed. The
3210 * freeing of the inode and its removal from the list must be
3214 xlog_recover_process_iunlinks(
3218 xfs_agnumber_t agno
;
3233 * Prevent any DMAPI event from being sent while in this function.
3235 mp_dmevmask
= mp
->m_dmevmask
;
3238 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3240 * Find the agi for this ag.
3242 agibp
= xfs_buf_read(mp
->m_ddev_targp
,
3243 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
3244 XFS_FSS_TO_BB(mp
, 1), 0);
3245 if (XFS_BUF_ISERROR(agibp
)) {
3246 xfs_ioerror_alert("xlog_recover_process_iunlinks(#1)",
3248 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)));
3250 agi
= XFS_BUF_TO_AGI(agibp
);
3251 ASSERT(XFS_AGI_MAGIC
== be32_to_cpu(agi
->agi_magicnum
));
3253 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3255 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
3256 while (agino
!= NULLAGINO
) {
3259 * Release the agi buffer so that it can
3260 * be acquired in the normal course of the
3261 * transaction to truncate and free the inode.
3263 xfs_buf_relse(agibp
);
3265 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3266 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
, 0);
3267 ASSERT(error
|| (ip
!= NULL
));
3271 * Get the on disk inode to find the
3272 * next inode in the bucket.
3274 error
= xfs_itobp(mp
, NULL
, ip
, &dip
,
3276 ASSERT(error
|| (dip
!= NULL
));
3280 ASSERT(ip
->i_d
.di_nlink
== 0);
3282 /* setup for the next pass */
3283 agino
= INT_GET(dip
->di_next_unlinked
,
3287 * Prevent any DMAPI event from
3288 * being sent when the
3289 * reference on the inode is
3292 ip
->i_d
.di_dmevmask
= 0;
3295 * If this is a new inode, handle
3296 * it specially. Otherwise,
3297 * just drop our reference to the
3298 * inode. If there are no
3299 * other references, this will
3301 * xfs_inactive() which will
3302 * truncate the file and free
3305 if (ip
->i_d
.di_mode
== 0)
3306 xfs_iput_new(ip
, 0);
3308 VN_RELE(XFS_ITOV(ip
));
3311 * We can't read in the inode
3312 * this bucket points to, or
3313 * this inode is messed up. Just
3314 * ditch this bucket of inodes. We
3315 * will lose some inodes and space,
3316 * but at least we won't hang. Call
3317 * xlog_recover_clear_agi_bucket()
3318 * to perform a transaction to clear
3319 * the inode pointer in the bucket.
3321 xlog_recover_clear_agi_bucket(mp
, agno
,
3328 * Reacquire the agibuffer and continue around
3331 agibp
= xfs_buf_read(mp
->m_ddev_targp
,
3332 XFS_AG_DADDR(mp
, agno
,
3334 XFS_FSS_TO_BB(mp
, 1), 0);
3335 if (XFS_BUF_ISERROR(agibp
)) {
3337 "xlog_recover_process_iunlinks(#2)",
3339 XFS_AG_DADDR(mp
, agno
,
3340 XFS_AGI_DADDR(mp
)));
3342 agi
= XFS_BUF_TO_AGI(agibp
);
3343 ASSERT(XFS_AGI_MAGIC
== be32_to_cpu(
3344 agi
->agi_magicnum
));
3349 * Release the buffer for the current agi so we can
3350 * go on to the next one.
3352 xfs_buf_relse(agibp
);
3355 mp
->m_dmevmask
= mp_dmevmask
;
3361 xlog_pack_data_checksum(
3363 xlog_in_core_t
*iclog
,
3370 up
= (uint
*)iclog
->ic_datap
;
3371 /* divide length by 4 to get # words */
3372 for (i
= 0; i
< (size
>> 2); i
++) {
3373 chksum
^= INT_GET(*up
, ARCH_CONVERT
);
3376 INT_SET(iclog
->ic_header
.h_chksum
, ARCH_CONVERT
, chksum
);
3379 #define xlog_pack_data_checksum(log, iclog, size)
3383 * Stamp cycle number in every block
3388 xlog_in_core_t
*iclog
,
3392 int size
= iclog
->ic_offset
+ roundoff
;
3395 xlog_in_core_2_t
*xhdr
;
3397 xlog_pack_data_checksum(log
, iclog
, size
);
3399 cycle_lsn
= CYCLE_LSN_DISK(iclog
->ic_header
.h_lsn
);
3401 dp
= iclog
->ic_datap
;
3402 for (i
= 0; i
< BTOBB(size
) &&
3403 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3404 iclog
->ic_header
.h_cycle_data
[i
] = *(uint
*)dp
;
3405 *(uint
*)dp
= cycle_lsn
;
3409 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3410 xhdr
= (xlog_in_core_2_t
*)&iclog
->ic_header
;
3411 for ( ; i
< BTOBB(size
); i
++) {
3412 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3413 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3414 xhdr
[j
].hic_xheader
.xh_cycle_data
[k
] = *(uint
*)dp
;
3415 *(uint
*)dp
= cycle_lsn
;
3419 for (i
= 1; i
< log
->l_iclog_heads
; i
++) {
3420 xhdr
[i
].hic_xheader
.xh_cycle
= cycle_lsn
;
3425 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3427 xlog_unpack_data_checksum(
3428 xlog_rec_header_t
*rhead
,
3432 uint
*up
= (uint
*)dp
;
3436 /* divide length by 4 to get # words */
3437 for (i
=0; i
< INT_GET(rhead
->h_len
, ARCH_CONVERT
) >> 2; i
++) {
3438 chksum
^= INT_GET(*up
, ARCH_CONVERT
);
3441 if (chksum
!= INT_GET(rhead
->h_chksum
, ARCH_CONVERT
)) {
3442 if (rhead
->h_chksum
||
3443 ((log
->l_flags
& XLOG_CHKSUM_MISMATCH
) == 0)) {
3445 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
3446 INT_GET(rhead
->h_chksum
, ARCH_CONVERT
), chksum
);
3448 "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3449 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3451 "XFS: LogR this is a LogV2 filesystem\n");
3453 log
->l_flags
|= XLOG_CHKSUM_MISMATCH
;
3458 #define xlog_unpack_data_checksum(rhead, dp, log)
3463 xlog_rec_header_t
*rhead
,
3468 xlog_in_core_2_t
*xhdr
;
3470 for (i
= 0; i
< BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
)) &&
3471 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3472 *(uint
*)dp
= *(uint
*)&rhead
->h_cycle_data
[i
];
3476 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3477 xhdr
= (xlog_in_core_2_t
*)rhead
;
3478 for ( ; i
< BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
)); i
++) {
3479 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3480 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3481 *(uint
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
3486 xlog_unpack_data_checksum(rhead
, dp
, log
);
3490 xlog_valid_rec_header(
3492 xlog_rec_header_t
*rhead
,
3498 (INT_GET(rhead
->h_magicno
, ARCH_CONVERT
) !=
3499 XLOG_HEADER_MAGIC_NUM
))) {
3500 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3501 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3502 return XFS_ERROR(EFSCORRUPTED
);
3505 (!rhead
->h_version
||
3506 (INT_GET(rhead
->h_version
, ARCH_CONVERT
) &
3507 (~XLOG_VERSION_OKBITS
)) != 0))) {
3508 xlog_warn("XFS: %s: unrecognised log version (%d).",
3509 __FUNCTION__
, INT_GET(rhead
->h_version
, ARCH_CONVERT
));
3510 return XFS_ERROR(EIO
);
3513 /* LR body must have data or it wouldn't have been written */
3514 hlen
= INT_GET(rhead
->h_len
, ARCH_CONVERT
);
3515 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
3516 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3517 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3518 return XFS_ERROR(EFSCORRUPTED
);
3520 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
3521 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3522 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3523 return XFS_ERROR(EFSCORRUPTED
);
3529 * Read the log from tail to head and process the log records found.
3530 * Handle the two cases where the tail and head are in the same cycle
3531 * and where the active portion of the log wraps around the end of
3532 * the physical log separately. The pass parameter is passed through
3533 * to the routines called to process the data and is not looked at
3537 xlog_do_recovery_pass(
3539 xfs_daddr_t head_blk
,
3540 xfs_daddr_t tail_blk
,
3543 xlog_rec_header_t
*rhead
;
3545 xfs_caddr_t bufaddr
, offset
;
3546 xfs_buf_t
*hbp
, *dbp
;
3547 int error
= 0, h_size
;
3548 int bblks
, split_bblks
;
3549 int hblks
, split_hblks
, wrapped_hblks
;
3550 xlog_recover_t
*rhash
[XLOG_RHASH_SIZE
];
3552 ASSERT(head_blk
!= tail_blk
);
3555 * Read the header of the tail block and get the iclog buffer size from
3556 * h_size. Use this to tell how many sectors make up the log header.
3558 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3560 * When using variable length iclogs, read first sector of
3561 * iclog header and extract the header size from it. Get a
3562 * new hbp that is the correct size.
3564 hbp
= xlog_get_bp(log
, 1);
3567 if ((error
= xlog_bread(log
, tail_blk
, 1, hbp
)))
3569 offset
= xlog_align(log
, tail_blk
, 1, hbp
);
3570 rhead
= (xlog_rec_header_t
*)offset
;
3571 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
3574 h_size
= INT_GET(rhead
->h_size
, ARCH_CONVERT
);
3575 if ((INT_GET(rhead
->h_version
, ARCH_CONVERT
)
3576 & XLOG_VERSION_2
) &&
3577 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
3578 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
3579 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
3582 hbp
= xlog_get_bp(log
, hblks
);
3587 ASSERT(log
->l_sectbb_log
== 0);
3589 hbp
= xlog_get_bp(log
, 1);
3590 h_size
= XLOG_BIG_RECORD_BSIZE
;
3595 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
3601 memset(rhash
, 0, sizeof(rhash
));
3602 if (tail_blk
<= head_blk
) {
3603 for (blk_no
= tail_blk
; blk_no
< head_blk
; ) {
3604 if ((error
= xlog_bread(log
, blk_no
, hblks
, hbp
)))
3606 offset
= xlog_align(log
, blk_no
, hblks
, hbp
);
3607 rhead
= (xlog_rec_header_t
*)offset
;
3608 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3612 /* blocks in data section */
3613 bblks
= (int)BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
));
3614 error
= xlog_bread(log
, blk_no
+ hblks
, bblks
, dbp
);
3617 offset
= xlog_align(log
, blk_no
+ hblks
, bblks
, dbp
);
3618 xlog_unpack_data(rhead
, offset
, log
);
3619 if ((error
= xlog_recover_process_data(log
,
3620 rhash
, rhead
, offset
, pass
)))
3622 blk_no
+= bblks
+ hblks
;
3626 * Perform recovery around the end of the physical log.
3627 * When the head is not on the same cycle number as the tail,
3628 * we can't do a sequential recovery as above.
3631 while (blk_no
< log
->l_logBBsize
) {
3633 * Check for header wrapping around physical end-of-log
3638 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
3639 /* Read header in one read */
3640 error
= xlog_bread(log
, blk_no
, hblks
, hbp
);
3643 offset
= xlog_align(log
, blk_no
, hblks
, hbp
);
3645 /* This LR is split across physical log end */
3646 if (blk_no
!= log
->l_logBBsize
) {
3647 /* some data before physical log end */
3648 ASSERT(blk_no
<= INT_MAX
);
3649 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
3650 ASSERT(split_hblks
> 0);
3651 if ((error
= xlog_bread(log
, blk_no
,
3654 offset
= xlog_align(log
, blk_no
,
3658 * Note: this black magic still works with
3659 * large sector sizes (non-512) only because:
3660 * - we increased the buffer size originally
3661 * by 1 sector giving us enough extra space
3662 * for the second read;
3663 * - the log start is guaranteed to be sector
3665 * - we read the log end (LR header start)
3666 * _first_, then the log start (LR header end)
3667 * - order is important.
3669 bufaddr
= XFS_BUF_PTR(hbp
);
3670 XFS_BUF_SET_PTR(hbp
,
3671 bufaddr
+ BBTOB(split_hblks
),
3672 BBTOB(hblks
- split_hblks
));
3673 wrapped_hblks
= hblks
- split_hblks
;
3674 error
= xlog_bread(log
, 0, wrapped_hblks
, hbp
);
3677 XFS_BUF_SET_PTR(hbp
, bufaddr
, BBTOB(hblks
));
3679 offset
= xlog_align(log
, 0,
3680 wrapped_hblks
, hbp
);
3682 rhead
= (xlog_rec_header_t
*)offset
;
3683 error
= xlog_valid_rec_header(log
, rhead
,
3684 split_hblks
? blk_no
: 0);
3688 bblks
= (int)BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
));
3691 /* Read in data for log record */
3692 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
3693 error
= xlog_bread(log
, blk_no
, bblks
, dbp
);
3696 offset
= xlog_align(log
, blk_no
, bblks
, dbp
);
3698 /* This log record is split across the
3699 * physical end of log */
3702 if (blk_no
!= log
->l_logBBsize
) {
3703 /* some data is before the physical
3705 ASSERT(!wrapped_hblks
);
3706 ASSERT(blk_no
<= INT_MAX
);
3708 log
->l_logBBsize
- (int)blk_no
;
3709 ASSERT(split_bblks
> 0);
3710 if ((error
= xlog_bread(log
, blk_no
,
3713 offset
= xlog_align(log
, blk_no
,
3717 * Note: this black magic still works with
3718 * large sector sizes (non-512) only because:
3719 * - we increased the buffer size originally
3720 * by 1 sector giving us enough extra space
3721 * for the second read;
3722 * - the log start is guaranteed to be sector
3724 * - we read the log end (LR header start)
3725 * _first_, then the log start (LR header end)
3726 * - order is important.
3728 bufaddr
= XFS_BUF_PTR(dbp
);
3729 XFS_BUF_SET_PTR(dbp
,
3730 bufaddr
+ BBTOB(split_bblks
),
3731 BBTOB(bblks
- split_bblks
));
3732 if ((error
= xlog_bread(log
, wrapped_hblks
,
3733 bblks
- split_bblks
, dbp
)))
3735 XFS_BUF_SET_PTR(dbp
, bufaddr
, h_size
);
3737 offset
= xlog_align(log
, wrapped_hblks
,
3738 bblks
- split_bblks
, dbp
);
3740 xlog_unpack_data(rhead
, offset
, log
);
3741 if ((error
= xlog_recover_process_data(log
, rhash
,
3742 rhead
, offset
, pass
)))
3747 ASSERT(blk_no
>= log
->l_logBBsize
);
3748 blk_no
-= log
->l_logBBsize
;
3750 /* read first part of physical log */
3751 while (blk_no
< head_blk
) {
3752 if ((error
= xlog_bread(log
, blk_no
, hblks
, hbp
)))
3754 offset
= xlog_align(log
, blk_no
, hblks
, hbp
);
3755 rhead
= (xlog_rec_header_t
*)offset
;
3756 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3759 bblks
= (int)BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
));
3760 if ((error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
)))
3762 offset
= xlog_align(log
, blk_no
+hblks
, bblks
, dbp
);
3763 xlog_unpack_data(rhead
, offset
, log
);
3764 if ((error
= xlog_recover_process_data(log
, rhash
,
3765 rhead
, offset
, pass
)))
3767 blk_no
+= bblks
+ hblks
;
3779 * Do the recovery of the log. We actually do this in two phases.
3780 * The two passes are necessary in order to implement the function
3781 * of cancelling a record written into the log. The first pass
3782 * determines those things which have been cancelled, and the
3783 * second pass replays log items normally except for those which
3784 * have been cancelled. The handling of the replay and cancellations
3785 * takes place in the log item type specific routines.
3787 * The table of items which have cancel records in the log is allocated
3788 * and freed at this level, since only here do we know when all of
3789 * the log recovery has been completed.
3792 xlog_do_log_recovery(
3794 xfs_daddr_t head_blk
,
3795 xfs_daddr_t tail_blk
)
3799 ASSERT(head_blk
!= tail_blk
);
3802 * First do a pass to find all of the cancelled buf log items.
3803 * Store them in the buf_cancel_table for use in the second pass.
3805 log
->l_buf_cancel_table
=
3806 (xfs_buf_cancel_t
**)kmem_zalloc(XLOG_BC_TABLE_SIZE
*
3807 sizeof(xfs_buf_cancel_t
*),
3809 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3810 XLOG_RECOVER_PASS1
);
3812 kmem_free(log
->l_buf_cancel_table
,
3813 XLOG_BC_TABLE_SIZE
* sizeof(xfs_buf_cancel_t
*));
3814 log
->l_buf_cancel_table
= NULL
;
3818 * Then do a second pass to actually recover the items in the log.
3819 * When it is complete free the table of buf cancel items.
3821 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3822 XLOG_RECOVER_PASS2
);
3827 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
3828 ASSERT(log
->l_buf_cancel_table
[i
] == NULL
);
3832 kmem_free(log
->l_buf_cancel_table
,
3833 XLOG_BC_TABLE_SIZE
* sizeof(xfs_buf_cancel_t
*));
3834 log
->l_buf_cancel_table
= NULL
;
3840 * Do the actual recovery
3845 xfs_daddr_t head_blk
,
3846 xfs_daddr_t tail_blk
)
3853 * First replay the images in the log.
3855 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
3860 XFS_bflush(log
->l_mp
->m_ddev_targp
);
3863 * If IO errors happened during recovery, bail out.
3865 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
3870 * We now update the tail_lsn since much of the recovery has completed
3871 * and there may be space available to use. If there were no extent
3872 * or iunlinks, we can free up the entire log and set the tail_lsn to
3873 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3874 * lsn of the last known good LR on disk. If there are extent frees
3875 * or iunlinks they will have some entries in the AIL; so we look at
3876 * the AIL to determine how to set the tail_lsn.
3878 xlog_assign_tail_lsn(log
->l_mp
);
3881 * Now that we've finished replaying all buffer and inode
3882 * updates, re-read in the superblock.
3884 bp
= xfs_getsb(log
->l_mp
, 0);
3887 xfsbdstrat(log
->l_mp
, bp
);
3888 if ((error
= xfs_iowait(bp
))) {
3889 xfs_ioerror_alert("xlog_do_recover",
3890 log
->l_mp
, bp
, XFS_BUF_ADDR(bp
));
3896 /* Convert superblock from on-disk format */
3897 sbp
= &log
->l_mp
->m_sb
;
3898 xfs_xlatesb(XFS_BUF_TO_SBP(bp
), sbp
, 1, XFS_SB_ALL_BITS
);
3899 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
3900 ASSERT(XFS_SB_GOOD_VERSION(sbp
));
3903 xlog_recover_check_summary(log
);
3905 /* Normal transactions can now occur */
3906 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
3911 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3913 * Return error or zero.
3919 xfs_daddr_t head_blk
, tail_blk
;
3922 /* find the tail of the log */
3923 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
)))
3926 if (tail_blk
!= head_blk
) {
3927 /* There used to be a comment here:
3929 * disallow recovery on read-only mounts. note -- mount
3930 * checks for ENOSPC and turns it into an intelligent
3932 * ...but this is no longer true. Now, unless you specify
3933 * NORECOVERY (in which case this function would never be
3934 * called), we just go ahead and recover. We do this all
3935 * under the vfs layer, so we can get away with it unless
3936 * the device itself is read-only, in which case we fail.
3938 if ((error
= xfs_dev_is_read_only(log
->l_mp
,
3939 "recovery required"))) {
3944 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3945 log
->l_mp
->m_fsname
, log
->l_mp
->m_logname
?
3946 log
->l_mp
->m_logname
: "internal");
3948 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
3949 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
3955 * In the first part of recovery we replay inodes and buffers and build
3956 * up the list of extent free items which need to be processed. Here
3957 * we process the extent free items and clean up the on disk unlinked
3958 * inode lists. This is separated from the first part of recovery so
3959 * that the root and real-time bitmap inodes can be read in from disk in
3960 * between the two stages. This is necessary so that we can free space
3961 * in the real-time portion of the file system.
3964 xlog_recover_finish(
3969 * Now we're ready to do the transactions needed for the
3970 * rest of recovery. Start with completing all the extent
3971 * free intent records and then process the unlinked inode
3972 * lists. At this point, we essentially run in normal mode
3973 * except that we're still performing recovery actions
3974 * rather than accepting new requests.
3976 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
3977 xlog_recover_process_efis(log
);
3979 * Sync the log to get all the EFIs out of the AIL.
3980 * This isn't absolutely necessary, but it helps in
3981 * case the unlink transactions would have problems
3982 * pushing the EFIs out of the way.
3984 xfs_log_force(log
->l_mp
, (xfs_lsn_t
)0,
3985 (XFS_LOG_FORCE
| XFS_LOG_SYNC
));
3987 if ( (mfsi_flags
& XFS_MFSI_NOUNLINK
) == 0 ) {
3988 xlog_recover_process_iunlinks(log
);
3991 xlog_recover_check_summary(log
);
3994 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3995 log
->l_mp
->m_fsname
, log
->l_mp
->m_logname
?
3996 log
->l_mp
->m_logname
: "internal");
3997 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
4000 "!Ending clean XFS mount for filesystem: %s\n",
4001 log
->l_mp
->m_fsname
);
4009 * Read all of the agf and agi counters and check that they
4010 * are consistent with the superblock counters.
4013 xlog_recover_check_summary(
4021 xfs_daddr_t agfdaddr
;
4022 xfs_daddr_t agidaddr
;
4024 #ifdef XFS_LOUD_RECOVERY
4027 xfs_agnumber_t agno
;
4028 __uint64_t freeblks
;
4037 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
4038 agfdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGF_DADDR(mp
));
4039 agfbp
= xfs_buf_read(mp
->m_ddev_targp
, agfdaddr
,
4040 XFS_FSS_TO_BB(mp
, 1), 0);
4041 if (XFS_BUF_ISERROR(agfbp
)) {
4042 xfs_ioerror_alert("xlog_recover_check_summary(agf)",
4043 mp
, agfbp
, agfdaddr
);
4045 agfp
= XFS_BUF_TO_AGF(agfbp
);
4046 ASSERT(XFS_AGF_MAGIC
== be32_to_cpu(agfp
->agf_magicnum
));
4047 ASSERT(XFS_AGF_GOOD_VERSION(be32_to_cpu(agfp
->agf_versionnum
)));
4048 ASSERT(be32_to_cpu(agfp
->agf_seqno
) == agno
);
4050 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
4051 be32_to_cpu(agfp
->agf_flcount
);
4052 xfs_buf_relse(agfbp
);
4054 agidaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
4055 agibp
= xfs_buf_read(mp
->m_ddev_targp
, agidaddr
,
4056 XFS_FSS_TO_BB(mp
, 1), 0);
4057 if (XFS_BUF_ISERROR(agibp
)) {
4058 xfs_ioerror_alert("xlog_recover_check_summary(agi)",
4059 mp
, agibp
, agidaddr
);
4061 agip
= XFS_BUF_TO_AGI(agibp
);
4062 ASSERT(XFS_AGI_MAGIC
== be32_to_cpu(agip
->agi_magicnum
));
4063 ASSERT(XFS_AGI_GOOD_VERSION(be32_to_cpu(agip
->agi_versionnum
)));
4064 ASSERT(be32_to_cpu(agip
->agi_seqno
) == agno
);
4066 itotal
+= be32_to_cpu(agip
->agi_count
);
4067 ifree
+= be32_to_cpu(agip
->agi_freecount
);
4068 xfs_buf_relse(agibp
);
4071 sbbp
= xfs_getsb(mp
, 0);
4072 #ifdef XFS_LOUD_RECOVERY
4074 xfs_xlatesb(XFS_BUF_TO_SBP(sbbp
), sbp
, 1, XFS_SB_ALL_BITS
);
4076 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4077 sbp
->sb_icount
, itotal
);
4079 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4080 sbp
->sb_ifree
, ifree
);
4082 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4083 sbp
->sb_fdblocks
, freeblks
);
4086 * This is turned off until I account for the allocation
4087 * btree blocks which live in free space.
4089 ASSERT(sbp
->sb_icount
== itotal
);
4090 ASSERT(sbp
->sb_ifree
== ifree
);
4091 ASSERT(sbp
->sb_fdblocks
== freeblks
);
4094 xfs_buf_relse(sbbp
);