2 * Copyright (c) 2000-2003 Silicon Graphics, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
28 * For further information regarding this notice, see:
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
34 #include "xfs_macros.h"
35 #include "xfs_types.h"
40 #include "xfs_trans.h"
43 #include "xfs_dmapi.h"
44 #include "xfs_mount.h"
45 #include "xfs_error.h"
46 #include "xfs_bmap_btree.h"
47 #include "xfs_alloc.h"
48 #include "xfs_attr_sf.h"
49 #include "xfs_dir_sf.h"
50 #include "xfs_dir2_sf.h"
51 #include "xfs_dinode.h"
53 #include "xfs_inode_item.h"
54 #include "xfs_inode.h"
55 #include "xfs_ialloc_btree.h"
56 #include "xfs_ialloc.h"
57 #include "xfs_error.h"
58 #include "xfs_log_priv.h"
59 #include "xfs_buf_item.h"
60 #include "xfs_alloc_btree.h"
61 #include "xfs_log_recover.h"
62 #include "xfs_extfree_item.h"
63 #include "xfs_trans_priv.h"
65 #include "xfs_quota.h"
68 STATIC
int xlog_find_zeroed(xlog_t
*, xfs_daddr_t
*);
69 STATIC
int xlog_clear_stale_blocks(xlog_t
*, xfs_lsn_t
);
70 STATIC
void xlog_recover_insert_item_backq(xlog_recover_item_t
**q
,
71 xlog_recover_item_t
*item
);
73 STATIC
void xlog_recover_check_summary(xlog_t
*);
74 STATIC
void xlog_recover_check_ail(xfs_mount_t
*, xfs_log_item_t
*, int);
76 #define xlog_recover_check_summary(log)
77 #define xlog_recover_check_ail(mp, lip, gen)
82 * Sector aligned buffer routines for buffer create/read/write/access
85 #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
86 ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
87 ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
88 #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
95 ASSERT(num_bblks
> 0);
97 if (log
->l_sectbb_log
) {
99 num_bblks
+= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, 1);
100 num_bblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, num_bblks
);
102 return XFS_ngetrbuf(BBTOB(num_bblks
), log
->l_mp
);
114 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
125 if (log
->l_sectbb_log
) {
126 blk_no
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, blk_no
);
127 nbblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, nbblks
);
131 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
134 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
137 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
138 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
140 xfsbdstrat(log
->l_mp
, bp
);
141 if ((error
= xfs_iowait(bp
)))
142 xfs_ioerror_alert("xlog_bread", log
->l_mp
,
143 bp
, XFS_BUF_ADDR(bp
));
148 * Write out the buffer at the given block for the given number of blocks.
149 * The buffer is kept locked across the write and is returned locked.
150 * This can only be used for synchronous log writes.
161 if (log
->l_sectbb_log
) {
162 blk_no
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, blk_no
);
163 nbblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, nbblks
);
167 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
169 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
170 XFS_BUF_ZEROFLAGS(bp
);
173 XFS_BUF_PSEMA(bp
, PRIBIO
);
174 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
175 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
177 if ((error
= xfs_bwrite(log
->l_mp
, bp
)))
178 xfs_ioerror_alert("xlog_bwrite", log
->l_mp
,
179 bp
, XFS_BUF_ADDR(bp
));
192 if (!log
->l_sectbb_log
)
193 return XFS_BUF_PTR(bp
);
195 ptr
= XFS_BUF_PTR(bp
) + BBTOB((int)blk_no
& log
->l_sectbb_mask
);
196 ASSERT(XFS_BUF_SIZE(bp
) >=
197 BBTOB(nbblks
+ (blk_no
& log
->l_sectbb_mask
)));
203 * dump debug superblock and log record information
206 xlog_header_check_dump(
208 xlog_rec_header_t
*head
)
212 printk("%s: SB : uuid = ", __FUNCTION__
);
213 for (b
= 0; b
< 16; b
++)
214 printk("%02x",((unsigned char *)&mp
->m_sb
.sb_uuid
)[b
]);
215 printk(", fmt = %d\n", XLOG_FMT
);
216 printk(" log : uuid = ");
217 for (b
= 0; b
< 16; b
++)
218 printk("%02x",((unsigned char *)&head
->h_fs_uuid
)[b
]);
219 printk(", fmt = %d\n", INT_GET(head
->h_fmt
, ARCH_CONVERT
));
222 #define xlog_header_check_dump(mp, head)
226 * check log record header for recovery
229 xlog_header_check_recover(
231 xlog_rec_header_t
*head
)
233 ASSERT(INT_GET(head
->h_magicno
, ARCH_CONVERT
) == XLOG_HEADER_MAGIC_NUM
);
236 * IRIX doesn't write the h_fmt field and leaves it zeroed
237 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
238 * a dirty log created in IRIX.
240 if (unlikely(INT_GET(head
->h_fmt
, ARCH_CONVERT
) != XLOG_FMT
)) {
242 "XFS: dirty log written in incompatible format - can't recover");
243 xlog_header_check_dump(mp
, head
);
244 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
245 XFS_ERRLEVEL_HIGH
, mp
);
246 return XFS_ERROR(EFSCORRUPTED
);
247 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
249 "XFS: dirty log entry has mismatched uuid - can't recover");
250 xlog_header_check_dump(mp
, head
);
251 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
252 XFS_ERRLEVEL_HIGH
, mp
);
253 return XFS_ERROR(EFSCORRUPTED
);
259 * read the head block of the log and check the header
262 xlog_header_check_mount(
264 xlog_rec_header_t
*head
)
266 ASSERT(INT_GET(head
->h_magicno
, ARCH_CONVERT
) == XLOG_HEADER_MAGIC_NUM
);
268 if (uuid_is_nil(&head
->h_fs_uuid
)) {
270 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
271 * h_fs_uuid is nil, we assume this log was last mounted
272 * by IRIX and continue.
274 xlog_warn("XFS: nil uuid in log - IRIX style log");
275 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
276 xlog_warn("XFS: log has mismatched uuid - can't recover");
277 xlog_header_check_dump(mp
, head
);
278 XFS_ERROR_REPORT("xlog_header_check_mount",
279 XFS_ERRLEVEL_HIGH
, mp
);
280 return XFS_ERROR(EFSCORRUPTED
);
291 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *));
293 if (XFS_BUF_GETERROR(bp
)) {
295 * We're not going to bother about retrying
296 * this during recovery. One strike!
298 mp
= XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*);
299 xfs_ioerror_alert("xlog_recover_iodone",
300 mp
, bp
, XFS_BUF_ADDR(bp
));
301 xfs_force_shutdown(mp
, XFS_METADATA_IO_ERROR
);
303 XFS_BUF_SET_FSPRIVATE(bp
, NULL
);
304 XFS_BUF_CLR_IODONE_FUNC(bp
);
309 * This routine finds (to an approximation) the first block in the physical
310 * log which contains the given cycle. It uses a binary search algorithm.
311 * Note that the algorithm can not be perfect because the disk will not
312 * necessarily be perfect.
315 xlog_find_cycle_start(
318 xfs_daddr_t first_blk
,
319 xfs_daddr_t
*last_blk
,
327 mid_blk
= BLK_AVG(first_blk
, *last_blk
);
328 while (mid_blk
!= first_blk
&& mid_blk
!= *last_blk
) {
329 if ((error
= xlog_bread(log
, mid_blk
, 1, bp
)))
331 offset
= xlog_align(log
, mid_blk
, 1, bp
);
332 mid_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
333 if (mid_cycle
== cycle
) {
335 /* last_half_cycle == mid_cycle */
338 /* first_half_cycle == mid_cycle */
340 mid_blk
= BLK_AVG(first_blk
, *last_blk
);
342 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == *last_blk
) ||
343 (mid_blk
== *last_blk
&& mid_blk
-1 == first_blk
));
349 * Check that the range of blocks does not contain the cycle number
350 * given. The scan needs to occur from front to back and the ptr into the
351 * region must be updated since a later routine will need to perform another
352 * test. If the region is completely good, we end up returning the same
355 * Set blkno to -1 if we encounter no errors. This is an invalid block number
356 * since we don't ever expect logs to get this large.
359 xlog_find_verify_cycle(
361 xfs_daddr_t start_blk
,
363 uint stop_on_cycle_no
,
364 xfs_daddr_t
*new_blk
)
370 xfs_caddr_t buf
= NULL
;
373 bufblks
= 1 << ffs(nbblks
);
375 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
376 /* can't get enough memory to do everything in one big buffer */
378 if (bufblks
<= log
->l_sectbb_log
)
382 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
385 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
387 if ((error
= xlog_bread(log
, i
, bcount
, bp
)))
390 buf
= xlog_align(log
, i
, bcount
, bp
);
391 for (j
= 0; j
< bcount
; j
++) {
392 cycle
= GET_CYCLE(buf
, ARCH_CONVERT
);
393 if (cycle
== stop_on_cycle_no
) {
410 * Potentially backup over partial log record write.
412 * In the typical case, last_blk is the number of the block directly after
413 * a good log record. Therefore, we subtract one to get the block number
414 * of the last block in the given buffer. extra_bblks contains the number
415 * of blocks we would have read on a previous read. This happens when the
416 * last log record is split over the end of the physical log.
418 * extra_bblks is the number of blocks potentially verified on a previous
419 * call to this routine.
422 xlog_find_verify_log_record(
424 xfs_daddr_t start_blk
,
425 xfs_daddr_t
*last_blk
,
430 xfs_caddr_t offset
= NULL
;
431 xlog_rec_header_t
*head
= NULL
;
434 int num_blks
= *last_blk
- start_blk
;
437 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
439 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
440 if (!(bp
= xlog_get_bp(log
, 1)))
444 if ((error
= xlog_bread(log
, start_blk
, num_blks
, bp
)))
446 offset
= xlog_align(log
, start_blk
, num_blks
, bp
);
447 offset
+= ((num_blks
- 1) << BBSHIFT
);
450 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
452 /* legal log record not found */
454 "XFS: Log inconsistent (didn't find previous header)");
456 error
= XFS_ERROR(EIO
);
461 if ((error
= xlog_bread(log
, i
, 1, bp
)))
463 offset
= xlog_align(log
, i
, 1, bp
);
466 head
= (xlog_rec_header_t
*)offset
;
468 if (XLOG_HEADER_MAGIC_NUM
==
469 INT_GET(head
->h_magicno
, ARCH_CONVERT
))
477 * We hit the beginning of the physical log & still no header. Return
478 * to caller. If caller can handle a return of -1, then this routine
479 * will be called again for the end of the physical log.
487 * We have the final block of the good log (the first block
488 * of the log record _before_ the head. So we check the uuid.
490 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
494 * We may have found a log record header before we expected one.
495 * last_blk will be the 1st block # with a given cycle #. We may end
496 * up reading an entire log record. In this case, we don't want to
497 * reset last_blk. Only when last_blk points in the middle of a log
498 * record do we update last_blk.
500 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
501 uint h_size
= INT_GET(head
->h_size
, ARCH_CONVERT
);
503 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
504 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
510 if (*last_blk
- i
+ extra_bblks
511 != BTOBB(INT_GET(head
->h_len
, ARCH_CONVERT
)) + xhdrs
)
520 * Head is defined to be the point of the log where the next log write
521 * write could go. This means that incomplete LR writes at the end are
522 * eliminated when calculating the head. We aren't guaranteed that previous
523 * LR have complete transactions. We only know that a cycle number of
524 * current cycle number -1 won't be present in the log if we start writing
525 * from our current block number.
527 * last_blk contains the block number of the first block with a given
530 * Return: zero if normal, non-zero if error.
535 xfs_daddr_t
*return_head_blk
)
539 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
541 uint first_half_cycle
, last_half_cycle
;
543 int error
, log_bbnum
= log
->l_logBBsize
;
545 /* Is the end of the log device zeroed? */
546 if ((error
= xlog_find_zeroed(log
, &first_blk
)) == -1) {
547 *return_head_blk
= first_blk
;
549 /* Is the whole lot zeroed? */
551 /* Linux XFS shouldn't generate totally zeroed logs -
552 * mkfs etc write a dummy unmount record to a fresh
553 * log so we can store the uuid in there
555 xlog_warn("XFS: totally zeroed log");
560 xlog_warn("XFS: empty log check failed");
564 first_blk
= 0; /* get cycle # of 1st block */
565 bp
= xlog_get_bp(log
, 1);
568 if ((error
= xlog_bread(log
, 0, 1, bp
)))
570 offset
= xlog_align(log
, 0, 1, bp
);
571 first_half_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
573 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
574 if ((error
= xlog_bread(log
, last_blk
, 1, bp
)))
576 offset
= xlog_align(log
, last_blk
, 1, bp
);
577 last_half_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
578 ASSERT(last_half_cycle
!= 0);
581 * If the 1st half cycle number is equal to the last half cycle number,
582 * then the entire log is stamped with the same cycle number. In this
583 * case, head_blk can't be set to zero (which makes sense). The below
584 * math doesn't work out properly with head_blk equal to zero. Instead,
585 * we set it to log_bbnum which is an illegal block number, but this
586 * value makes the math correct. If head_blk doesn't changed through
587 * all the tests below, *head_blk is set to zero at the very end rather
588 * than log_bbnum. In a sense, log_bbnum and zero are the same block
589 * in a circular file.
591 if (first_half_cycle
== last_half_cycle
) {
593 * In this case we believe that the entire log should have
594 * cycle number last_half_cycle. We need to scan backwards
595 * from the end verifying that there are no holes still
596 * containing last_half_cycle - 1. If we find such a hole,
597 * then the start of that hole will be the new head. The
598 * simple case looks like
599 * x | x ... | x - 1 | x
600 * Another case that fits this picture would be
601 * x | x + 1 | x ... | x
602 * In this case the head really is somwhere at the end of the
603 * log, as one of the latest writes at the beginning was
606 * x | x + 1 | x ... | x - 1 | x
607 * This is really the combination of the above two cases, and
608 * the head has to end up at the start of the x-1 hole at the
611 * In the 256k log case, we will read from the beginning to the
612 * end of the log and search for cycle numbers equal to x-1.
613 * We don't worry about the x+1 blocks that we encounter,
614 * because we know that they cannot be the head since the log
617 head_blk
= log_bbnum
;
618 stop_on_cycle
= last_half_cycle
- 1;
621 * In this case we want to find the first block with cycle
622 * number matching last_half_cycle. We expect the log to be
625 * The first block with cycle number x (last_half_cycle) will
626 * be where the new head belongs. First we do a binary search
627 * for the first occurrence of last_half_cycle. The binary
628 * search may not be totally accurate, so then we scan back
629 * from there looking for occurrences of last_half_cycle before
630 * us. If that backwards scan wraps around the beginning of
631 * the log, then we look for occurrences of last_half_cycle - 1
632 * at the end of the log. The cases we're looking for look
634 * x + 1 ... | x | x + 1 | x ...
635 * ^ binary search stopped here
637 * x + 1 ... | x ... | x - 1 | x
638 * <---------> less than scan distance
640 stop_on_cycle
= last_half_cycle
;
641 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
642 &head_blk
, last_half_cycle
)))
647 * Now validate the answer. Scan back some number of maximum possible
648 * blocks and make sure each one has the expected cycle number. The
649 * maximum is determined by the total possible amount of buffering
650 * in the in-core log. The following number can be made tighter if
651 * we actually look at the block size of the filesystem.
653 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
654 if (head_blk
>= num_scan_bblks
) {
656 * We are guaranteed that the entire check can be performed
659 start_blk
= head_blk
- num_scan_bblks
;
660 if ((error
= xlog_find_verify_cycle(log
,
661 start_blk
, num_scan_bblks
,
662 stop_on_cycle
, &new_blk
)))
666 } else { /* need to read 2 parts of log */
668 * We are going to scan backwards in the log in two parts.
669 * First we scan the physical end of the log. In this part
670 * of the log, we are looking for blocks with cycle number
671 * last_half_cycle - 1.
672 * If we find one, then we know that the log starts there, as
673 * we've found a hole that didn't get written in going around
674 * the end of the physical log. The simple case for this is
675 * x + 1 ... | x ... | x - 1 | x
676 * <---------> less than scan distance
677 * If all of the blocks at the end of the log have cycle number
678 * last_half_cycle, then we check the blocks at the start of
679 * the log looking for occurrences of last_half_cycle. If we
680 * find one, then our current estimate for the location of the
681 * first occurrence of last_half_cycle is wrong and we move
682 * back to the hole we've found. This case looks like
683 * x + 1 ... | x | x + 1 | x ...
684 * ^ binary search stopped here
685 * Another case we need to handle that only occurs in 256k
687 * x + 1 ... | x ... | x+1 | x ...
688 * ^ binary search stops here
689 * In a 256k log, the scan at the end of the log will see the
690 * x + 1 blocks. We need to skip past those since that is
691 * certainly not the head of the log. By searching for
692 * last_half_cycle-1 we accomplish that.
694 start_blk
= log_bbnum
- num_scan_bblks
+ head_blk
;
695 ASSERT(head_blk
<= INT_MAX
&&
696 (xfs_daddr_t
) num_scan_bblks
- head_blk
>= 0);
697 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
698 num_scan_bblks
- (int)head_blk
,
699 (stop_on_cycle
- 1), &new_blk
)))
707 * Scan beginning of log now. The last part of the physical
708 * log is good. This scan needs to verify that it doesn't find
709 * the last_half_cycle.
712 ASSERT(head_blk
<= INT_MAX
);
713 if ((error
= xlog_find_verify_cycle(log
,
714 start_blk
, (int)head_blk
,
715 stop_on_cycle
, &new_blk
)))
723 * Now we need to make sure head_blk is not pointing to a block in
724 * the middle of a log record.
726 num_scan_bblks
= XLOG_REC_SHIFT(log
);
727 if (head_blk
>= num_scan_bblks
) {
728 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
730 /* start ptr at last block ptr before head_blk */
731 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
732 &head_blk
, 0)) == -1) {
733 error
= XFS_ERROR(EIO
);
739 ASSERT(head_blk
<= INT_MAX
);
740 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
741 &head_blk
, 0)) == -1) {
742 /* We hit the beginning of the log during our search */
743 start_blk
= log_bbnum
- num_scan_bblks
+ head_blk
;
745 ASSERT(start_blk
<= INT_MAX
&&
746 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
747 ASSERT(head_blk
<= INT_MAX
);
748 if ((error
= xlog_find_verify_log_record(log
,
750 (int)head_blk
)) == -1) {
751 error
= XFS_ERROR(EIO
);
755 if (new_blk
!= log_bbnum
)
762 if (head_blk
== log_bbnum
)
763 *return_head_blk
= 0;
765 *return_head_blk
= head_blk
;
767 * When returning here, we have a good block number. Bad block
768 * means that during a previous crash, we didn't have a clean break
769 * from cycle number N to cycle number N-1. In this case, we need
770 * to find the first block with cycle number N-1.
778 xlog_warn("XFS: failed to find log head");
783 * Find the sync block number or the tail of the log.
785 * This will be the block number of the last record to have its
786 * associated buffers synced to disk. Every log record header has
787 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
788 * to get a sync block number. The only concern is to figure out which
789 * log record header to believe.
791 * The following algorithm uses the log record header with the largest
792 * lsn. The entire log record does not need to be valid. We only care
793 * that the header is valid.
795 * We could speed up search by using current head_blk buffer, but it is not
801 xfs_daddr_t
*head_blk
,
802 xfs_daddr_t
*tail_blk
,
805 xlog_rec_header_t
*rhead
;
806 xlog_op_header_t
*op_head
;
807 xfs_caddr_t offset
= NULL
;
810 xfs_daddr_t umount_data_blk
;
811 xfs_daddr_t after_umount_blk
;
818 * Find previous log record
820 if ((error
= xlog_find_head(log
, head_blk
)))
823 bp
= xlog_get_bp(log
, 1);
826 if (*head_blk
== 0) { /* special case */
827 if ((error
= xlog_bread(log
, 0, 1, bp
)))
829 offset
= xlog_align(log
, 0, 1, bp
);
830 if (GET_CYCLE(offset
, ARCH_CONVERT
) == 0) {
832 /* leave all other log inited values alone */
838 * Search backwards looking for log record header block
840 ASSERT(*head_blk
< INT_MAX
);
841 for (i
= (int)(*head_blk
) - 1; i
>= 0; i
--) {
842 if ((error
= xlog_bread(log
, i
, 1, bp
)))
844 offset
= xlog_align(log
, i
, 1, bp
);
845 if (XLOG_HEADER_MAGIC_NUM
==
846 INT_GET(*(uint
*)offset
, ARCH_CONVERT
)) {
852 * If we haven't found the log record header block, start looking
853 * again from the end of the physical log. XXXmiken: There should be
854 * a check here to make sure we didn't search more than N blocks in
858 for (i
= log
->l_logBBsize
- 1; i
>= (int)(*head_blk
); i
--) {
859 if ((error
= xlog_bread(log
, i
, 1, bp
)))
861 offset
= xlog_align(log
, i
, 1, bp
);
862 if (XLOG_HEADER_MAGIC_NUM
==
863 INT_GET(*(uint
*)offset
, ARCH_CONVERT
)) {
870 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
872 return XFS_ERROR(EIO
);
875 /* find blk_no of tail of log */
876 rhead
= (xlog_rec_header_t
*)offset
;
877 *tail_blk
= BLOCK_LSN(rhead
->h_tail_lsn
, ARCH_CONVERT
);
880 * Reset log values according to the state of the log when we
881 * crashed. In the case where head_blk == 0, we bump curr_cycle
882 * one because the next write starts a new cycle rather than
883 * continuing the cycle of the last good log record. At this
884 * point we have guaranteed that all partial log records have been
885 * accounted for. Therefore, we know that the last good log record
886 * written was complete and ended exactly on the end boundary
887 * of the physical log.
889 log
->l_prev_block
= i
;
890 log
->l_curr_block
= (int)*head_blk
;
891 log
->l_curr_cycle
= INT_GET(rhead
->h_cycle
, ARCH_CONVERT
);
894 log
->l_tail_lsn
= INT_GET(rhead
->h_tail_lsn
, ARCH_CONVERT
);
895 log
->l_last_sync_lsn
= INT_GET(rhead
->h_lsn
, ARCH_CONVERT
);
896 log
->l_grant_reserve_cycle
= log
->l_curr_cycle
;
897 log
->l_grant_reserve_bytes
= BBTOB(log
->l_curr_block
);
898 log
->l_grant_write_cycle
= log
->l_curr_cycle
;
899 log
->l_grant_write_bytes
= BBTOB(log
->l_curr_block
);
902 * Look for unmount record. If we find it, then we know there
903 * was a clean unmount. Since 'i' could be the last block in
904 * the physical log, we convert to a log block before comparing
907 * Save the current tail lsn to use to pass to
908 * xlog_clear_stale_blocks() below. We won't want to clear the
909 * unmount record if there is one, so we pass the lsn of the
910 * unmount record rather than the block after it.
912 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
913 int h_size
= INT_GET(rhead
->h_size
, ARCH_CONVERT
);
914 int h_version
= INT_GET(rhead
->h_version
, ARCH_CONVERT
);
916 if ((h_version
& XLOG_VERSION_2
) &&
917 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
918 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
919 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
927 after_umount_blk
= (i
+ hblks
+ (int)
928 BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
))) % log
->l_logBBsize
;
929 tail_lsn
= log
->l_tail_lsn
;
930 if (*head_blk
== after_umount_blk
&&
931 INT_GET(rhead
->h_num_logops
, ARCH_CONVERT
) == 1) {
932 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
933 if ((error
= xlog_bread(log
, umount_data_blk
, 1, bp
))) {
936 offset
= xlog_align(log
, umount_data_blk
, 1, bp
);
937 op_head
= (xlog_op_header_t
*)offset
;
938 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
940 * Set tail and last sync so that newly written
941 * log records will point recovery to after the
942 * current unmount record.
944 ASSIGN_ANY_LSN(log
->l_tail_lsn
, log
->l_curr_cycle
,
945 after_umount_blk
, ARCH_NOCONVERT
);
946 ASSIGN_ANY_LSN(log
->l_last_sync_lsn
, log
->l_curr_cycle
,
947 after_umount_blk
, ARCH_NOCONVERT
);
948 *tail_blk
= after_umount_blk
;
953 * Make sure that there are no blocks in front of the head
954 * with the same cycle number as the head. This can happen
955 * because we allow multiple outstanding log writes concurrently,
956 * and the later writes might make it out before earlier ones.
958 * We use the lsn from before modifying it so that we'll never
959 * overwrite the unmount record after a clean unmount.
961 * Do this only if we are going to recover the filesystem
963 * NOTE: This used to say "if (!readonly)"
964 * However on Linux, we can & do recover a read-only filesystem.
965 * We only skip recovery if NORECOVERY is specified on mount,
966 * in which case we would not be here.
968 * But... if the -device- itself is readonly, just skip this.
969 * We can't recover this device anyway, so it won't matter.
971 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
)) {
972 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
980 xlog_warn("XFS: failed to locate log tail");
985 * Is the log zeroed at all?
987 * The last binary search should be changed to perform an X block read
988 * once X becomes small enough. You can then search linearly through
989 * the X blocks. This will cut down on the number of reads we need to do.
991 * If the log is partially zeroed, this routine will pass back the blkno
992 * of the first block with cycle number 0. It won't have a complete LR
996 * 0 => the log is completely written to
997 * -1 => use *blk_no as the first block of the log
998 * >0 => error has occurred
1003 xfs_daddr_t
*blk_no
)
1007 uint first_cycle
, last_cycle
;
1008 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1009 xfs_daddr_t num_scan_bblks
;
1010 int error
, log_bbnum
= log
->l_logBBsize
;
1012 /* check totally zeroed log */
1013 bp
= xlog_get_bp(log
, 1);
1016 if ((error
= xlog_bread(log
, 0, 1, bp
)))
1018 offset
= xlog_align(log
, 0, 1, bp
);
1019 first_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
1020 if (first_cycle
== 0) { /* completely zeroed log */
1026 /* check partially zeroed log */
1027 if ((error
= xlog_bread(log
, log_bbnum
-1, 1, bp
)))
1029 offset
= xlog_align(log
, log_bbnum
-1, 1, bp
);
1030 last_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
1031 if (last_cycle
!= 0) { /* log completely written to */
1034 } else if (first_cycle
!= 1) {
1036 * If the cycle of the last block is zero, the cycle of
1037 * the first block must be 1. If it's not, maybe we're
1038 * not looking at a log... Bail out.
1040 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1041 return XFS_ERROR(EINVAL
);
1044 /* we have a partially zeroed log */
1045 last_blk
= log_bbnum
-1;
1046 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1050 * Validate the answer. Because there is no way to guarantee that
1051 * the entire log is made up of log records which are the same size,
1052 * we scan over the defined maximum blocks. At this point, the maximum
1053 * is not chosen to mean anything special. XXXmiken
1055 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1056 ASSERT(num_scan_bblks
<= INT_MAX
);
1058 if (last_blk
< num_scan_bblks
)
1059 num_scan_bblks
= last_blk
;
1060 start_blk
= last_blk
- num_scan_bblks
;
1063 * We search for any instances of cycle number 0 that occur before
1064 * our current estimate of the head. What we're trying to detect is
1065 * 1 ... | 0 | 1 | 0...
1066 * ^ binary search ends here
1068 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1069 (int)num_scan_bblks
, 0, &new_blk
)))
1075 * Potentially backup over partial log record write. We don't need
1076 * to search the end of the log because we know it is zero.
1078 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
1079 &last_blk
, 0)) == -1) {
1080 error
= XFS_ERROR(EIO
);
1094 * These are simple subroutines used by xlog_clear_stale_blocks() below
1095 * to initialize a buffer full of empty log record headers and write
1096 * them into the log.
1107 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1109 memset(buf
, 0, BBSIZE
);
1110 INT_SET(recp
->h_magicno
, ARCH_CONVERT
, XLOG_HEADER_MAGIC_NUM
);
1111 INT_SET(recp
->h_cycle
, ARCH_CONVERT
, cycle
);
1112 INT_SET(recp
->h_version
, ARCH_CONVERT
,
1113 XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
) ? 2 : 1);
1114 ASSIGN_ANY_LSN(recp
->h_lsn
, cycle
, block
, ARCH_CONVERT
);
1115 ASSIGN_ANY_LSN(recp
->h_tail_lsn
, tail_cycle
, tail_block
, ARCH_CONVERT
);
1116 INT_SET(recp
->h_fmt
, ARCH_CONVERT
, XLOG_FMT
);
1117 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1121 xlog_write_log_records(
1132 int sectbb
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, 1);
1133 int end_block
= start_block
+ blocks
;
1138 bufblks
= 1 << ffs(blocks
);
1139 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1141 if (bufblks
<= log
->l_sectbb_log
)
1145 /* We may need to do a read at the start to fill in part of
1146 * the buffer in the starting sector not covered by the first
1149 balign
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, start_block
);
1150 if (balign
!= start_block
) {
1151 if ((error
= xlog_bread(log
, start_block
, 1, bp
))) {
1155 j
= start_block
- balign
;
1158 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1159 int bcount
, endcount
;
1161 bcount
= min(bufblks
, end_block
- start_block
);
1162 endcount
= bcount
- j
;
1164 /* We may need to do a read at the end to fill in part of
1165 * the buffer in the final sector not covered by the write.
1166 * If this is the same sector as the above read, skip it.
1168 ealign
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, end_block
);
1169 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1170 offset
= XFS_BUF_PTR(bp
);
1171 balign
= BBTOB(ealign
- start_block
);
1172 XFS_BUF_SET_PTR(bp
, offset
+ balign
, BBTOB(sectbb
));
1173 if ((error
= xlog_bread(log
, ealign
, sectbb
, bp
)))
1175 XFS_BUF_SET_PTR(bp
, offset
, bufblks
);
1178 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1179 for (; j
< endcount
; j
++) {
1180 xlog_add_record(log
, offset
, cycle
, i
+j
,
1181 tail_cycle
, tail_block
);
1184 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1187 start_block
+= endcount
;
1195 * This routine is called to blow away any incomplete log writes out
1196 * in front of the log head. We do this so that we won't become confused
1197 * if we come up, write only a little bit more, and then crash again.
1198 * If we leave the partial log records out there, this situation could
1199 * cause us to think those partial writes are valid blocks since they
1200 * have the current cycle number. We get rid of them by overwriting them
1201 * with empty log records with the old cycle number rather than the
1204 * The tail lsn is passed in rather than taken from
1205 * the log so that we will not write over the unmount record after a
1206 * clean unmount in a 512 block log. Doing so would leave the log without
1207 * any valid log records in it until a new one was written. If we crashed
1208 * during that time we would not be able to recover.
1211 xlog_clear_stale_blocks(
1215 int tail_cycle
, head_cycle
;
1216 int tail_block
, head_block
;
1217 int tail_distance
, max_distance
;
1221 tail_cycle
= CYCLE_LSN(tail_lsn
, ARCH_NOCONVERT
);
1222 tail_block
= BLOCK_LSN(tail_lsn
, ARCH_NOCONVERT
);
1223 head_cycle
= log
->l_curr_cycle
;
1224 head_block
= log
->l_curr_block
;
1227 * Figure out the distance between the new head of the log
1228 * and the tail. We want to write over any blocks beyond the
1229 * head that we may have written just before the crash, but
1230 * we don't want to overwrite the tail of the log.
1232 if (head_cycle
== tail_cycle
) {
1234 * The tail is behind the head in the physical log,
1235 * so the distance from the head to the tail is the
1236 * distance from the head to the end of the log plus
1237 * the distance from the beginning of the log to the
1240 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1241 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1242 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1243 return XFS_ERROR(EFSCORRUPTED
);
1245 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1248 * The head is behind the tail in the physical log,
1249 * so the distance from the head to the tail is just
1250 * the tail block minus the head block.
1252 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1253 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1254 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1255 return XFS_ERROR(EFSCORRUPTED
);
1257 tail_distance
= tail_block
- head_block
;
1261 * If the head is right up against the tail, we can't clear
1264 if (tail_distance
<= 0) {
1265 ASSERT(tail_distance
== 0);
1269 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1271 * Take the smaller of the maximum amount of outstanding I/O
1272 * we could have and the distance to the tail to clear out.
1273 * We take the smaller so that we don't overwrite the tail and
1274 * we don't waste all day writing from the head to the tail
1277 max_distance
= MIN(max_distance
, tail_distance
);
1279 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1281 * We can stomp all the blocks we need to without
1282 * wrapping around the end of the log. Just do it
1283 * in a single write. Use the cycle number of the
1284 * current cycle minus one so that the log will look like:
1287 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1288 head_block
, max_distance
, tail_cycle
,
1294 * We need to wrap around the end of the physical log in
1295 * order to clear all the blocks. Do it in two separate
1296 * I/Os. The first write should be from the head to the
1297 * end of the physical log, and it should use the current
1298 * cycle number minus one just like above.
1300 distance
= log
->l_logBBsize
- head_block
;
1301 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1302 head_block
, distance
, tail_cycle
,
1309 * Now write the blocks at the start of the physical log.
1310 * This writes the remainder of the blocks we want to clear.
1311 * It uses the current cycle number since we're now on the
1312 * same cycle as the head so that we get:
1313 * n ... n ... | n - 1 ...
1314 * ^^^^^ blocks we're writing
1316 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1317 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1318 tail_cycle
, tail_block
);
1326 /******************************************************************************
1328 * Log recover routines
1330 ******************************************************************************
1333 STATIC xlog_recover_t
*
1334 xlog_recover_find_tid(
1338 xlog_recover_t
*p
= q
;
1341 if (p
->r_log_tid
== tid
)
1349 xlog_recover_put_hashq(
1351 xlog_recover_t
*trans
)
1358 xlog_recover_add_item(
1359 xlog_recover_item_t
**itemq
)
1361 xlog_recover_item_t
*item
;
1363 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), 0);
1364 xlog_recover_insert_item_backq(itemq
, item
);
1368 xlog_recover_add_to_cont_trans(
1369 xlog_recover_t
*trans
,
1373 xlog_recover_item_t
*item
;
1374 xfs_caddr_t ptr
, old_ptr
;
1377 item
= trans
->r_itemq
;
1379 /* finish copying rest of trans header */
1380 xlog_recover_add_item(&trans
->r_itemq
);
1381 ptr
= (xfs_caddr_t
) &trans
->r_theader
+
1382 sizeof(xfs_trans_header_t
) - len
;
1383 memcpy(ptr
, dp
, len
); /* d, s, l */
1386 item
= item
->ri_prev
;
1388 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
1389 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
1391 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, 0);
1392 memcpy(&ptr
[old_len
], dp
, len
); /* d, s, l */
1393 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
1394 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
1399 * The next region to add is the start of a new region. It could be
1400 * a whole region or it could be the first part of a new region. Because
1401 * of this, the assumption here is that the type and size fields of all
1402 * format structures fit into the first 32 bits of the structure.
1404 * This works because all regions must be 32 bit aligned. Therefore, we
1405 * either have both fields or we have neither field. In the case we have
1406 * neither field, the data part of the region is zero length. We only have
1407 * a log_op_header and can throw away the header since a new one will appear
1408 * later. If we have at least 4 bytes, then we can determine how many regions
1409 * will appear in the current log item.
1412 xlog_recover_add_to_trans(
1413 xlog_recover_t
*trans
,
1417 xfs_inode_log_format_t
*in_f
; /* any will do */
1418 xlog_recover_item_t
*item
;
1423 item
= trans
->r_itemq
;
1425 ASSERT(*(uint
*)dp
== XFS_TRANS_HEADER_MAGIC
);
1426 if (len
== sizeof(xfs_trans_header_t
))
1427 xlog_recover_add_item(&trans
->r_itemq
);
1428 memcpy(&trans
->r_theader
, dp
, len
); /* d, s, l */
1432 ptr
= kmem_alloc(len
, KM_SLEEP
);
1433 memcpy(ptr
, dp
, len
);
1434 in_f
= (xfs_inode_log_format_t
*)ptr
;
1436 if (item
->ri_prev
->ri_total
!= 0 &&
1437 item
->ri_prev
->ri_total
== item
->ri_prev
->ri_cnt
) {
1438 xlog_recover_add_item(&trans
->r_itemq
);
1440 item
= trans
->r_itemq
;
1441 item
= item
->ri_prev
;
1443 if (item
->ri_total
== 0) { /* first region to be added */
1444 item
->ri_total
= in_f
->ilf_size
;
1445 ASSERT(item
->ri_total
<= XLOG_MAX_REGIONS_IN_ITEM
);
1446 item
->ri_buf
= kmem_zalloc((item
->ri_total
*
1447 sizeof(xfs_log_iovec_t
)), 0);
1449 ASSERT(item
->ri_total
> item
->ri_cnt
);
1450 /* Description region is ri_buf[0] */
1451 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
1452 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
1458 xlog_recover_new_tid(
1463 xlog_recover_t
*trans
;
1465 trans
= kmem_zalloc(sizeof(xlog_recover_t
), KM_SLEEP
);
1466 trans
->r_log_tid
= tid
;
1468 xlog_recover_put_hashq(q
, trans
);
1472 xlog_recover_unlink_tid(
1474 xlog_recover_t
*trans
)
1485 if (tp
->r_next
== trans
) {
1493 "XFS: xlog_recover_unlink_tid: trans not found");
1495 return XFS_ERROR(EIO
);
1497 tp
->r_next
= tp
->r_next
->r_next
;
1503 xlog_recover_insert_item_backq(
1504 xlog_recover_item_t
**q
,
1505 xlog_recover_item_t
*item
)
1508 item
->ri_prev
= item
->ri_next
= item
;
1512 item
->ri_prev
= (*q
)->ri_prev
;
1513 (*q
)->ri_prev
= item
;
1514 item
->ri_prev
->ri_next
= item
;
1519 xlog_recover_insert_item_frontq(
1520 xlog_recover_item_t
**q
,
1521 xlog_recover_item_t
*item
)
1523 xlog_recover_insert_item_backq(q
, item
);
1528 xlog_recover_reorder_trans(
1530 xlog_recover_t
*trans
)
1532 xlog_recover_item_t
*first_item
, *itemq
, *itemq_next
;
1534 first_item
= itemq
= trans
->r_itemq
;
1535 trans
->r_itemq
= NULL
;
1537 itemq_next
= itemq
->ri_next
;
1538 switch (ITEM_TYPE(itemq
)) {
1540 case XFS_LI_6_1_BUF
:
1541 case XFS_LI_5_3_BUF
:
1542 xlog_recover_insert_item_frontq(&trans
->r_itemq
, itemq
);
1545 case XFS_LI_6_1_INODE
:
1546 case XFS_LI_5_3_INODE
:
1548 case XFS_LI_QUOTAOFF
:
1551 xlog_recover_insert_item_backq(&trans
->r_itemq
, itemq
);
1555 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1557 return XFS_ERROR(EIO
);
1560 } while (first_item
!= itemq
);
1565 * Build up the table of buf cancel records so that we don't replay
1566 * cancelled data in the second pass. For buffer records that are
1567 * not cancel records, there is nothing to do here so we just return.
1569 * If we get a cancel record which is already in the table, this indicates
1570 * that the buffer was cancelled multiple times. In order to ensure
1571 * that during pass 2 we keep the record in the table until we reach its
1572 * last occurrence in the log, we keep a reference count in the cancel
1573 * record in the table to tell us how many times we expect to see this
1574 * record during the second pass.
1577 xlog_recover_do_buffer_pass1(
1579 xfs_buf_log_format_t
*buf_f
)
1581 xfs_buf_cancel_t
*bcp
;
1582 xfs_buf_cancel_t
*nextp
;
1583 xfs_buf_cancel_t
*prevp
;
1584 xfs_buf_cancel_t
**bucket
;
1585 xfs_buf_log_format_v1_t
*obuf_f
;
1586 xfs_daddr_t blkno
= 0;
1590 switch (buf_f
->blf_type
) {
1592 blkno
= buf_f
->blf_blkno
;
1593 len
= buf_f
->blf_len
;
1594 flags
= buf_f
->blf_flags
;
1596 case XFS_LI_6_1_BUF
:
1597 case XFS_LI_5_3_BUF
:
1598 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1599 blkno
= (xfs_daddr_t
) obuf_f
->blf_blkno
;
1600 len
= obuf_f
->blf_len
;
1601 flags
= obuf_f
->blf_flags
;
1606 * If this isn't a cancel buffer item, then just return.
1608 if (!(flags
& XFS_BLI_CANCEL
))
1612 * Insert an xfs_buf_cancel record into the hash table of
1613 * them. If there is already an identical record, bump
1614 * its reference count.
1616 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1617 XLOG_BC_TABLE_SIZE
];
1619 * If the hash bucket is empty then just insert a new record into
1622 if (*bucket
== NULL
) {
1623 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1625 bcp
->bc_blkno
= blkno
;
1627 bcp
->bc_refcount
= 1;
1628 bcp
->bc_next
= NULL
;
1634 * The hash bucket is not empty, so search for duplicates of our
1635 * record. If we find one them just bump its refcount. If not
1636 * then add us at the end of the list.
1640 while (nextp
!= NULL
) {
1641 if (nextp
->bc_blkno
== blkno
&& nextp
->bc_len
== len
) {
1642 nextp
->bc_refcount
++;
1646 nextp
= nextp
->bc_next
;
1648 ASSERT(prevp
!= NULL
);
1649 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1651 bcp
->bc_blkno
= blkno
;
1653 bcp
->bc_refcount
= 1;
1654 bcp
->bc_next
= NULL
;
1655 prevp
->bc_next
= bcp
;
1659 * Check to see whether the buffer being recovered has a corresponding
1660 * entry in the buffer cancel record table. If it does then return 1
1661 * so that it will be cancelled, otherwise return 0. If the buffer is
1662 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1663 * the refcount on the entry in the table and remove it from the table
1664 * if this is the last reference.
1666 * We remove the cancel record from the table when we encounter its
1667 * last occurrence in the log so that if the same buffer is re-used
1668 * again after its last cancellation we actually replay the changes
1669 * made at that point.
1672 xlog_recover_do_buffer_pass2(
1674 xfs_buf_log_format_t
*buf_f
)
1676 xfs_buf_cancel_t
*bcp
;
1677 xfs_buf_cancel_t
*prevp
;
1678 xfs_buf_cancel_t
**bucket
;
1679 xfs_buf_log_format_v1_t
*obuf_f
;
1680 xfs_daddr_t blkno
= 0;
1684 switch (buf_f
->blf_type
) {
1686 blkno
= buf_f
->blf_blkno
;
1687 flags
= buf_f
->blf_flags
;
1688 len
= buf_f
->blf_len
;
1690 case XFS_LI_6_1_BUF
:
1691 case XFS_LI_5_3_BUF
:
1692 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1693 blkno
= (xfs_daddr_t
) obuf_f
->blf_blkno
;
1694 flags
= obuf_f
->blf_flags
;
1695 len
= (xfs_daddr_t
) obuf_f
->blf_len
;
1698 if (log
->l_buf_cancel_table
== NULL
) {
1700 * There is nothing in the table built in pass one,
1701 * so this buffer must not be cancelled.
1703 ASSERT(!(flags
& XFS_BLI_CANCEL
));
1707 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1708 XLOG_BC_TABLE_SIZE
];
1712 * There is no corresponding entry in the table built
1713 * in pass one, so this buffer has not been cancelled.
1715 ASSERT(!(flags
& XFS_BLI_CANCEL
));
1720 * Search for an entry in the buffer cancel table that
1721 * matches our buffer.
1724 while (bcp
!= NULL
) {
1725 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
) {
1727 * We've go a match, so return 1 so that the
1728 * recovery of this buffer is cancelled.
1729 * If this buffer is actually a buffer cancel
1730 * log item, then decrement the refcount on the
1731 * one in the table and remove it if this is the
1734 if (flags
& XFS_BLI_CANCEL
) {
1736 if (bcp
->bc_refcount
== 0) {
1737 if (prevp
== NULL
) {
1738 *bucket
= bcp
->bc_next
;
1740 prevp
->bc_next
= bcp
->bc_next
;
1743 sizeof(xfs_buf_cancel_t
));
1752 * We didn't find a corresponding entry in the table, so
1753 * return 0 so that the buffer is NOT cancelled.
1755 ASSERT(!(flags
& XFS_BLI_CANCEL
));
1760 * Perform recovery for a buffer full of inodes. In these buffers,
1761 * the only data which should be recovered is that which corresponds
1762 * to the di_next_unlinked pointers in the on disk inode structures.
1763 * The rest of the data for the inodes is always logged through the
1764 * inodes themselves rather than the inode buffer and is recovered
1765 * in xlog_recover_do_inode_trans().
1767 * The only time when buffers full of inodes are fully recovered is
1768 * when the buffer is full of newly allocated inodes. In this case
1769 * the buffer will not be marked as an inode buffer and so will be
1770 * sent to xlog_recover_do_reg_buffer() below during recovery.
1773 xlog_recover_do_inode_buffer(
1775 xlog_recover_item_t
*item
,
1777 xfs_buf_log_format_t
*buf_f
)
1785 int next_unlinked_offset
;
1787 xfs_agino_t
*logged_nextp
;
1788 xfs_agino_t
*buffer_nextp
;
1789 xfs_buf_log_format_v1_t
*obuf_f
;
1790 unsigned int *data_map
= NULL
;
1791 unsigned int map_size
= 0;
1793 switch (buf_f
->blf_type
) {
1795 data_map
= buf_f
->blf_data_map
;
1796 map_size
= buf_f
->blf_map_size
;
1798 case XFS_LI_6_1_BUF
:
1799 case XFS_LI_5_3_BUF
:
1800 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1801 data_map
= obuf_f
->blf_data_map
;
1802 map_size
= obuf_f
->blf_map_size
;
1806 * Set the variables corresponding to the current region to
1807 * 0 so that we'll initialize them on the first pass through
1815 inodes_per_buf
= XFS_BUF_COUNT(bp
) >> mp
->m_sb
.sb_inodelog
;
1816 for (i
= 0; i
< inodes_per_buf
; i
++) {
1817 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1818 offsetof(xfs_dinode_t
, di_next_unlinked
);
1820 while (next_unlinked_offset
>=
1821 (reg_buf_offset
+ reg_buf_bytes
)) {
1823 * The next di_next_unlinked field is beyond
1824 * the current logged region. Find the next
1825 * logged region that contains or is beyond
1826 * the current di_next_unlinked field.
1829 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1832 * If there are no more logged regions in the
1833 * buffer, then we're done.
1839 nbits
= xfs_contig_bits(data_map
, map_size
,
1841 reg_buf_offset
= bit
<< XFS_BLI_SHIFT
;
1842 reg_buf_bytes
= nbits
<< XFS_BLI_SHIFT
;
1847 * If the current logged region starts after the current
1848 * di_next_unlinked field, then move on to the next
1849 * di_next_unlinked field.
1851 if (next_unlinked_offset
< reg_buf_offset
) {
1855 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1856 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLI_CHUNK
) == 0);
1857 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <= XFS_BUF_COUNT(bp
));
1860 * The current logged region contains a copy of the
1861 * current di_next_unlinked field. Extract its value
1862 * and copy it to the buffer copy.
1864 logged_nextp
= (xfs_agino_t
*)
1865 ((char *)(item
->ri_buf
[item_index
].i_addr
) +
1866 (next_unlinked_offset
- reg_buf_offset
));
1867 if (unlikely(*logged_nextp
== 0)) {
1868 xfs_fs_cmn_err(CE_ALERT
, mp
,
1869 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1871 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1872 XFS_ERRLEVEL_LOW
, mp
);
1873 return XFS_ERROR(EFSCORRUPTED
);
1876 buffer_nextp
= (xfs_agino_t
*)xfs_buf_offset(bp
,
1877 next_unlinked_offset
);
1878 INT_SET(*buffer_nextp
, ARCH_CONVERT
, *logged_nextp
);
1885 * Perform a 'normal' buffer recovery. Each logged region of the
1886 * buffer should be copied over the corresponding region in the
1887 * given buffer. The bitmap in the buf log format structure indicates
1888 * where to place the logged data.
1892 xlog_recover_do_reg_buffer(
1894 xlog_recover_item_t
*item
,
1896 xfs_buf_log_format_t
*buf_f
)
1901 xfs_buf_log_format_v1_t
*obuf_f
;
1902 unsigned int *data_map
= NULL
;
1903 unsigned int map_size
= 0;
1906 switch (buf_f
->blf_type
) {
1908 data_map
= buf_f
->blf_data_map
;
1909 map_size
= buf_f
->blf_map_size
;
1911 case XFS_LI_6_1_BUF
:
1912 case XFS_LI_5_3_BUF
:
1913 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1914 data_map
= obuf_f
->blf_data_map
;
1915 map_size
= obuf_f
->blf_map_size
;
1919 i
= 1; /* 0 is the buf format structure */
1921 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1924 nbits
= xfs_contig_bits(data_map
, map_size
, bit
);
1925 ASSERT(item
->ri_buf
[i
].i_addr
!= 0);
1926 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLI_CHUNK
== 0);
1927 ASSERT(XFS_BUF_COUNT(bp
) >=
1928 ((uint
)bit
<< XFS_BLI_SHIFT
)+(nbits
<<XFS_BLI_SHIFT
));
1931 * Do a sanity check if this is a dquot buffer. Just checking
1932 * the first dquot in the buffer should do. XXXThis is
1933 * probably a good thing to do for other buf types also.
1936 if (buf_f
->blf_flags
& (XFS_BLI_UDQUOT_BUF
|XFS_BLI_GDQUOT_BUF
)) {
1937 error
= xfs_qm_dqcheck((xfs_disk_dquot_t
*)
1938 item
->ri_buf
[i
].i_addr
,
1939 -1, 0, XFS_QMOPT_DOWARN
,
1940 "dquot_buf_recover");
1943 memcpy(xfs_buf_offset(bp
,
1944 (uint
)bit
<< XFS_BLI_SHIFT
), /* dest */
1945 item
->ri_buf
[i
].i_addr
, /* source */
1946 nbits
<<XFS_BLI_SHIFT
); /* length */
1951 /* Shouldn't be any more regions */
1952 ASSERT(i
== item
->ri_total
);
1956 * Do some primitive error checking on ondisk dquot data structures.
1960 xfs_disk_dquot_t
*ddq
,
1962 uint type
, /* used only when IO_dorepair is true */
1966 xfs_dqblk_t
*d
= (xfs_dqblk_t
*)ddq
;
1970 * We can encounter an uninitialized dquot buffer for 2 reasons:
1971 * 1. If we crash while deleting the quotainode(s), and those blks got
1972 * used for user data. This is because we take the path of regular
1973 * file deletion; however, the size field of quotainodes is never
1974 * updated, so all the tricks that we play in itruncate_finish
1975 * don't quite matter.
1977 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1978 * But the allocation will be replayed so we'll end up with an
1979 * uninitialized quota block.
1981 * This is all fine; things are still consistent, and we haven't lost
1982 * any quota information. Just don't complain about bad dquot blks.
1984 if (INT_GET(ddq
->d_magic
, ARCH_CONVERT
) != XFS_DQUOT_MAGIC
) {
1985 if (flags
& XFS_QMOPT_DOWARN
)
1987 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
1989 INT_GET(ddq
->d_magic
, ARCH_CONVERT
), XFS_DQUOT_MAGIC
);
1992 if (INT_GET(ddq
->d_version
, ARCH_CONVERT
) != XFS_DQUOT_VERSION
) {
1993 if (flags
& XFS_QMOPT_DOWARN
)
1995 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
1997 INT_GET(ddq
->d_magic
, ARCH_CONVERT
), XFS_DQUOT_VERSION
);
2001 if (INT_GET(ddq
->d_flags
, ARCH_CONVERT
) != XFS_DQ_USER
&&
2002 INT_GET(ddq
->d_flags
, ARCH_CONVERT
) != XFS_DQ_GROUP
) {
2003 if (flags
& XFS_QMOPT_DOWARN
)
2005 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2006 str
, id
, INT_GET(ddq
->d_flags
, ARCH_CONVERT
));
2010 if (id
!= -1 && id
!= INT_GET(ddq
->d_id
, ARCH_CONVERT
)) {
2011 if (flags
& XFS_QMOPT_DOWARN
)
2013 "%s : ondisk-dquot 0x%x, ID mismatch: "
2014 "0x%x expected, found id 0x%x",
2015 str
, ddq
, id
, INT_GET(ddq
->d_id
, ARCH_CONVERT
));
2020 if (INT_GET(ddq
->d_blk_softlimit
, ARCH_CONVERT
) &&
2021 INT_GET(ddq
->d_bcount
, ARCH_CONVERT
) >=
2022 INT_GET(ddq
->d_blk_softlimit
, ARCH_CONVERT
)) {
2023 if (INT_ISZERO(ddq
->d_btimer
, ARCH_CONVERT
) &&
2024 !INT_ISZERO(ddq
->d_id
, ARCH_CONVERT
)) {
2025 if (flags
& XFS_QMOPT_DOWARN
)
2027 "%s : Dquot ID 0x%x (0x%x) "
2028 "BLK TIMER NOT STARTED",
2030 INT_GET(ddq
->d_id
, ARCH_CONVERT
), ddq
);
2034 if (INT_GET(ddq
->d_ino_softlimit
, ARCH_CONVERT
) &&
2035 INT_GET(ddq
->d_icount
, ARCH_CONVERT
) >=
2036 INT_GET(ddq
->d_ino_softlimit
, ARCH_CONVERT
)) {
2037 if (INT_ISZERO(ddq
->d_itimer
, ARCH_CONVERT
) &&
2038 !INT_ISZERO(ddq
->d_id
, ARCH_CONVERT
)) {
2039 if (flags
& XFS_QMOPT_DOWARN
)
2041 "%s : Dquot ID 0x%x (0x%x) "
2042 "INODE TIMER NOT STARTED",
2044 INT_GET(ddq
->d_id
, ARCH_CONVERT
), ddq
);
2050 if (!errs
|| !(flags
& XFS_QMOPT_DQREPAIR
))
2053 if (flags
& XFS_QMOPT_DOWARN
)
2054 cmn_err(CE_NOTE
, "Re-initializing dquot ID 0x%x", id
);
2057 * Typically, a repair is only requested by quotacheck.
2060 ASSERT(flags
& XFS_QMOPT_DQREPAIR
);
2061 memset(d
, 0, sizeof(xfs_dqblk_t
));
2062 INT_SET(d
->dd_diskdq
.d_magic
, ARCH_CONVERT
, XFS_DQUOT_MAGIC
);
2063 INT_SET(d
->dd_diskdq
.d_version
, ARCH_CONVERT
, XFS_DQUOT_VERSION
);
2064 INT_SET(d
->dd_diskdq
.d_id
, ARCH_CONVERT
, id
);
2065 INT_SET(d
->dd_diskdq
.d_flags
, ARCH_CONVERT
, type
);
2071 * Perform a dquot buffer recovery.
2072 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2073 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2074 * Else, treat it as a regular buffer and do recovery.
2077 xlog_recover_do_dquot_buffer(
2080 xlog_recover_item_t
*item
,
2082 xfs_buf_log_format_t
*buf_f
)
2087 * Filesystems are required to send in quota flags at mount time.
2089 if (mp
->m_qflags
== 0) {
2094 if (buf_f
->blf_flags
& XFS_BLI_UDQUOT_BUF
)
2095 type
|= XFS_DQ_USER
;
2096 if (buf_f
->blf_flags
& XFS_BLI_GDQUOT_BUF
)
2097 type
|= XFS_DQ_GROUP
;
2099 * This type of quotas was turned off, so ignore this buffer
2101 if (log
->l_quotaoffs_flag
& type
)
2104 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2108 * This routine replays a modification made to a buffer at runtime.
2109 * There are actually two types of buffer, regular and inode, which
2110 * are handled differently. Inode buffers are handled differently
2111 * in that we only recover a specific set of data from them, namely
2112 * the inode di_next_unlinked fields. This is because all other inode
2113 * data is actually logged via inode records and any data we replay
2114 * here which overlaps that may be stale.
2116 * When meta-data buffers are freed at run time we log a buffer item
2117 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2118 * of the buffer in the log should not be replayed at recovery time.
2119 * This is so that if the blocks covered by the buffer are reused for
2120 * file data before we crash we don't end up replaying old, freed
2121 * meta-data into a user's file.
2123 * To handle the cancellation of buffer log items, we make two passes
2124 * over the log during recovery. During the first we build a table of
2125 * those buffers which have been cancelled, and during the second we
2126 * only replay those buffers which do not have corresponding cancel
2127 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2128 * for more details on the implementation of the table of cancel records.
2131 xlog_recover_do_buffer_trans(
2133 xlog_recover_item_t
*item
,
2136 xfs_buf_log_format_t
*buf_f
;
2137 xfs_buf_log_format_v1_t
*obuf_f
;
2146 buf_f
= (xfs_buf_log_format_t
*)item
->ri_buf
[0].i_addr
;
2148 if (pass
== XLOG_RECOVER_PASS1
) {
2150 * In this pass we're only looking for buf items
2151 * with the XFS_BLI_CANCEL bit set.
2153 xlog_recover_do_buffer_pass1(log
, buf_f
);
2157 * In this pass we want to recover all the buffers
2158 * which have not been cancelled and are not
2159 * cancellation buffers themselves. The routine
2160 * we call here will tell us whether or not to
2161 * continue with the replay of this buffer.
2163 cancel
= xlog_recover_do_buffer_pass2(log
, buf_f
);
2168 switch (buf_f
->blf_type
) {
2170 blkno
= buf_f
->blf_blkno
;
2171 len
= buf_f
->blf_len
;
2172 flags
= buf_f
->blf_flags
;
2174 case XFS_LI_6_1_BUF
:
2175 case XFS_LI_5_3_BUF
:
2176 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
2177 blkno
= obuf_f
->blf_blkno
;
2178 len
= obuf_f
->blf_len
;
2179 flags
= obuf_f
->blf_flags
;
2182 xfs_fs_cmn_err(CE_ALERT
, log
->l_mp
,
2183 "xfs_log_recover: unknown buffer type 0x%x, dev 0x%x",
2184 buf_f
->blf_type
, log
->l_dev
);
2185 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2186 XFS_ERRLEVEL_LOW
, log
->l_mp
);
2187 return XFS_ERROR(EFSCORRUPTED
);
2191 if (flags
& XFS_BLI_INODE_BUF
) {
2192 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, blkno
, len
,
2195 bp
= xfs_buf_read(mp
->m_ddev_targp
, blkno
, len
, 0);
2197 if (XFS_BUF_ISERROR(bp
)) {
2198 xfs_ioerror_alert("xlog_recover_do..(read#1)", log
->l_mp
,
2200 error
= XFS_BUF_GETERROR(bp
);
2206 if (flags
& XFS_BLI_INODE_BUF
) {
2207 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2208 } else if (flags
& (XFS_BLI_UDQUOT_BUF
| XFS_BLI_GDQUOT_BUF
)) {
2209 xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2211 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2214 return XFS_ERROR(error
);
2217 * Perform delayed write on the buffer. Asynchronous writes will be
2218 * slower when taking into account all the buffers to be flushed.
2220 * Also make sure that only inode buffers with good sizes stay in
2221 * the buffer cache. The kernel moves inodes in buffers of 1 block
2222 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2223 * buffers in the log can be a different size if the log was generated
2224 * by an older kernel using unclustered inode buffers or a newer kernel
2225 * running with a different inode cluster size. Regardless, if the
2226 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2227 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2228 * the buffer out of the buffer cache so that the buffer won't
2229 * overlap with future reads of those inodes.
2231 if (XFS_DINODE_MAGIC
==
2232 INT_GET(*((__uint16_t
*)(xfs_buf_offset(bp
, 0))), ARCH_CONVERT
) &&
2233 (XFS_BUF_COUNT(bp
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2234 (__uint32_t
)XFS_INODE_CLUSTER_SIZE(log
->l_mp
)))) {
2236 error
= xfs_bwrite(mp
, bp
);
2238 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
||
2239 XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*) == mp
);
2240 XFS_BUF_SET_FSPRIVATE(bp
, mp
);
2241 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2242 xfs_bdwrite(mp
, bp
);
2249 xlog_recover_do_inode_trans(
2251 xlog_recover_item_t
*item
,
2254 xfs_inode_log_format_t
*in_f
;
2266 xfs_dinode_core_t
*dicp
;
2268 if (pass
== XLOG_RECOVER_PASS1
) {
2272 in_f
= (xfs_inode_log_format_t
*)item
->ri_buf
[0].i_addr
;
2273 ino
= in_f
->ilf_ino
;
2275 if (ITEM_TYPE(item
) == XFS_LI_INODE
) {
2276 imap
.im_blkno
= (xfs_daddr_t
)in_f
->ilf_blkno
;
2277 imap
.im_len
= in_f
->ilf_len
;
2278 imap
.im_boffset
= in_f
->ilf_boffset
;
2281 * It's an old inode format record. We don't know where
2282 * its cluster is located on disk, and we can't allow
2283 * xfs_imap() to figure it out because the inode btrees
2284 * are not ready to be used. Therefore do not pass the
2285 * XFS_IMAP_LOOKUP flag to xfs_imap(). This will give
2286 * us only the single block in which the inode lives
2287 * rather than its cluster, so we must make sure to
2288 * invalidate the buffer when we write it out below.
2291 xfs_imap(log
->l_mp
, 0, ino
, &imap
, 0);
2293 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, imap
.im_blkno
, imap
.im_len
,
2295 if (XFS_BUF_ISERROR(bp
)) {
2296 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp
,
2298 error
= XFS_BUF_GETERROR(bp
);
2303 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2304 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
2307 * Make sure the place we're flushing out to really looks
2310 if (unlikely(INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) != XFS_DINODE_MAGIC
)) {
2312 xfs_fs_cmn_err(CE_ALERT
, mp
,
2313 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2315 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2316 XFS_ERRLEVEL_LOW
, mp
);
2317 return XFS_ERROR(EFSCORRUPTED
);
2319 dicp
= (xfs_dinode_core_t
*)(item
->ri_buf
[1].i_addr
);
2320 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2322 xfs_fs_cmn_err(CE_ALERT
, mp
,
2323 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2325 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2326 XFS_ERRLEVEL_LOW
, mp
);
2327 return XFS_ERROR(EFSCORRUPTED
);
2329 if (unlikely((dicp
->di_mode
& IFMT
) == IFREG
)) {
2330 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2331 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2332 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2333 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2335 xfs_fs_cmn_err(CE_ALERT
, mp
,
2336 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2337 item
, dip
, bp
, ino
);
2338 return XFS_ERROR(EFSCORRUPTED
);
2340 } else if (unlikely((dicp
->di_mode
& IFMT
) == IFDIR
)) {
2341 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2342 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2343 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2344 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2345 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2347 xfs_fs_cmn_err(CE_ALERT
, mp
,
2348 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2349 item
, dip
, bp
, ino
);
2350 return XFS_ERROR(EFSCORRUPTED
);
2353 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2354 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2355 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2357 xfs_fs_cmn_err(CE_ALERT
, mp
,
2358 "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",
2360 dicp
->di_nextents
+ dicp
->di_anextents
,
2362 return XFS_ERROR(EFSCORRUPTED
);
2364 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2365 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2366 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2368 xfs_fs_cmn_err(CE_ALERT
, mp
,
2369 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2370 item
, dip
, bp
, ino
, dicp
->di_forkoff
);
2371 return XFS_ERROR(EFSCORRUPTED
);
2373 if (unlikely(item
->ri_buf
[1].i_len
> sizeof(xfs_dinode_core_t
))) {
2374 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2375 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2377 xfs_fs_cmn_err(CE_ALERT
, mp
,
2378 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2379 item
->ri_buf
[1].i_len
, item
);
2380 return XFS_ERROR(EFSCORRUPTED
);
2383 /* The core is in in-core format */
2384 xfs_xlate_dinode_core((xfs_caddr_t
)&dip
->di_core
,
2385 (xfs_dinode_core_t
*)item
->ri_buf
[1].i_addr
,
2387 /* the rest is in on-disk format */
2388 if (item
->ri_buf
[1].i_len
> sizeof(xfs_dinode_core_t
)) {
2389 memcpy((xfs_caddr_t
) dip
+ sizeof(xfs_dinode_core_t
),
2390 item
->ri_buf
[1].i_addr
+ sizeof(xfs_dinode_core_t
),
2391 item
->ri_buf
[1].i_len
- sizeof(xfs_dinode_core_t
));
2394 fields
= in_f
->ilf_fields
;
2395 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
2397 INT_SET(dip
->di_u
.di_dev
, ARCH_CONVERT
, in_f
->ilf_u
.ilfu_rdev
);
2401 dip
->di_u
.di_muuid
= in_f
->ilf_u
.ilfu_uuid
;
2405 if (in_f
->ilf_size
== 2)
2406 goto write_inode_buffer
;
2407 len
= item
->ri_buf
[2].i_len
;
2408 src
= item
->ri_buf
[2].i_addr
;
2409 ASSERT(in_f
->ilf_size
<= 4);
2410 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
2411 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
2412 (len
== in_f
->ilf_dsize
));
2414 switch (fields
& XFS_ILOG_DFORK
) {
2415 case XFS_ILOG_DDATA
:
2417 memcpy(&dip
->di_u
, src
, len
);
2420 case XFS_ILOG_DBROOT
:
2421 xfs_bmbt_to_bmdr((xfs_bmbt_block_t
*)src
, len
,
2422 &(dip
->di_u
.di_bmbt
),
2423 XFS_DFORK_DSIZE(dip
, mp
));
2428 * There are no data fork flags set.
2430 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
2435 * If we logged any attribute data, recover it. There may or
2436 * may not have been any other non-core data logged in this
2439 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2440 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
2445 len
= item
->ri_buf
[attr_index
].i_len
;
2446 src
= item
->ri_buf
[attr_index
].i_addr
;
2447 ASSERT(len
== in_f
->ilf_asize
);
2449 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2450 case XFS_ILOG_ADATA
:
2452 dest
= XFS_DFORK_APTR(dip
);
2453 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
2454 memcpy(dest
, src
, len
);
2457 case XFS_ILOG_ABROOT
:
2458 dest
= XFS_DFORK_APTR(dip
);
2459 xfs_bmbt_to_bmdr((xfs_bmbt_block_t
*)src
, len
,
2460 (xfs_bmdr_block_t
*)dest
,
2461 XFS_DFORK_ASIZE(dip
, mp
));
2465 xlog_warn("XFS: xlog_recover_do_inode_trans: Illegal flag");
2468 return XFS_ERROR(EIO
);
2473 if (ITEM_TYPE(item
) == XFS_LI_INODE
) {
2474 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
||
2475 XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*) == mp
);
2476 XFS_BUF_SET_FSPRIVATE(bp
, mp
);
2477 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2478 xfs_bdwrite(mp
, bp
);
2481 error
= xfs_bwrite(mp
, bp
);
2488 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2489 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2493 xlog_recover_do_quotaoff_trans(
2495 xlog_recover_item_t
*item
,
2498 xfs_qoff_logformat_t
*qoff_f
;
2500 if (pass
== XLOG_RECOVER_PASS2
) {
2504 qoff_f
= (xfs_qoff_logformat_t
*)item
->ri_buf
[0].i_addr
;
2508 * The logitem format's flag tells us if this was user quotaoff,
2509 * group quotaoff or both.
2511 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2512 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2513 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2514 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2520 * Recover a dquot record
2523 xlog_recover_do_dquot_trans(
2525 xlog_recover_item_t
*item
,
2530 struct xfs_disk_dquot
*ddq
, *recddq
;
2532 xfs_dq_logformat_t
*dq_f
;
2535 if (pass
== XLOG_RECOVER_PASS1
) {
2541 * Filesystems are required to send in quota flags at mount time.
2543 if (mp
->m_qflags
== 0)
2546 recddq
= (xfs_disk_dquot_t
*)item
->ri_buf
[1].i_addr
;
2549 * This type of quotas was turned off, so ignore this record.
2551 type
= INT_GET(recddq
->d_flags
, ARCH_CONVERT
) &
2552 (XFS_DQ_USER
| XFS_DQ_GROUP
);
2554 if (log
->l_quotaoffs_flag
& type
)
2558 * At this point we know that quota was _not_ turned off.
2559 * Since the mount flags are not indicating to us otherwise, this
2560 * must mean that quota is on, and the dquot needs to be replayed.
2561 * Remember that we may not have fully recovered the superblock yet,
2562 * so we can't do the usual trick of looking at the SB quota bits.
2564 * The other possibility, of course, is that the quota subsystem was
2565 * removed since the last mount - ENOSYS.
2567 dq_f
= (xfs_dq_logformat_t
*)item
->ri_buf
[0].i_addr
;
2569 if ((error
= xfs_qm_dqcheck(recddq
,
2571 0, XFS_QMOPT_DOWARN
,
2572 "xlog_recover_do_dquot_trans (log copy)"))) {
2573 return XFS_ERROR(EIO
);
2575 ASSERT(dq_f
->qlf_len
== 1);
2577 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
2579 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
),
2582 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp
,
2583 bp
, dq_f
->qlf_blkno
);
2587 ddq
= (xfs_disk_dquot_t
*)xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
2590 * At least the magic num portion should be on disk because this
2591 * was among a chunk of dquots created earlier, and we did some
2592 * minimal initialization then.
2594 if (xfs_qm_dqcheck(ddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2595 "xlog_recover_do_dquot_trans")) {
2597 return XFS_ERROR(EIO
);
2600 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
2602 ASSERT(dq_f
->qlf_size
== 2);
2603 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
||
2604 XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*) == mp
);
2605 XFS_BUF_SET_FSPRIVATE(bp
, mp
);
2606 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2607 xfs_bdwrite(mp
, bp
);
2613 * This routine is called to create an in-core extent free intent
2614 * item from the efi format structure which was logged on disk.
2615 * It allocates an in-core efi, copies the extents from the format
2616 * structure into it, and adds the efi to the AIL with the given
2620 xlog_recover_do_efi_trans(
2622 xlog_recover_item_t
*item
,
2627 xfs_efi_log_item_t
*efip
;
2628 xfs_efi_log_format_t
*efi_formatp
;
2631 if (pass
== XLOG_RECOVER_PASS1
) {
2635 efi_formatp
= (xfs_efi_log_format_t
*)item
->ri_buf
[0].i_addr
;
2636 ASSERT(item
->ri_buf
[0].i_len
==
2637 (sizeof(xfs_efi_log_format_t
) +
2638 ((efi_formatp
->efi_nextents
- 1) * sizeof(xfs_extent_t
))));
2641 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
2642 memcpy((char *)&(efip
->efi_format
), (char *)efi_formatp
,
2643 sizeof(xfs_efi_log_format_t
) +
2644 ((efi_formatp
->efi_nextents
- 1) * sizeof(xfs_extent_t
)));
2645 efip
->efi_next_extent
= efi_formatp
->efi_nextents
;
2646 efip
->efi_flags
|= XFS_EFI_COMMITTED
;
2650 * xfs_trans_update_ail() drops the AIL lock.
2652 xfs_trans_update_ail(mp
, (xfs_log_item_t
*)efip
, lsn
, s
);
2657 * This routine is called when an efd format structure is found in
2658 * a committed transaction in the log. It's purpose is to cancel
2659 * the corresponding efi if it was still in the log. To do this
2660 * it searches the AIL for the efi with an id equal to that in the
2661 * efd format structure. If we find it, we remove the efi from the
2665 xlog_recover_do_efd_trans(
2667 xlog_recover_item_t
*item
,
2671 xfs_efd_log_format_t
*efd_formatp
;
2672 xfs_efi_log_item_t
*efip
= NULL
;
2673 xfs_log_item_t
*lip
;
2679 if (pass
== XLOG_RECOVER_PASS1
) {
2683 efd_formatp
= (xfs_efd_log_format_t
*)item
->ri_buf
[0].i_addr
;
2684 ASSERT(item
->ri_buf
[0].i_len
==
2685 (sizeof(xfs_efd_log_format_t
) +
2686 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_t
))));
2687 efi_id
= efd_formatp
->efd_efi_id
;
2690 * Search for the efi with the id in the efd format structure
2695 lip
= xfs_trans_first_ail(mp
, &gen
);
2696 while (lip
!= NULL
) {
2697 if (lip
->li_type
== XFS_LI_EFI
) {
2698 efip
= (xfs_efi_log_item_t
*)lip
;
2699 if (efip
->efi_format
.efi_id
== efi_id
) {
2701 * xfs_trans_delete_ail() drops the
2704 xfs_trans_delete_ail(mp
, lip
, s
);
2708 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
2715 * If we found it, then free it up. If it wasn't there, it
2716 * must have been overwritten in the log. Oh well.
2719 nexts
= efip
->efi_format
.efi_nextents
;
2720 if (nexts
> XFS_EFI_MAX_FAST_EXTENTS
) {
2721 kmem_free(lip
, sizeof(xfs_efi_log_item_t
) +
2722 ((nexts
- 1) * sizeof(xfs_extent_t
)));
2724 kmem_zone_free(xfs_efi_zone
, efip
);
2730 * Perform the transaction
2732 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2733 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2736 xlog_recover_do_trans(
2738 xlog_recover_t
*trans
,
2742 xlog_recover_item_t
*item
, *first_item
;
2744 if ((error
= xlog_recover_reorder_trans(log
, trans
)))
2746 first_item
= item
= trans
->r_itemq
;
2749 * we don't need to worry about the block number being
2750 * truncated in > 1 TB buffers because in user-land,
2751 * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so
2752 * the blkno's will get through the user-mode buffer
2753 * cache properly. The only bad case is o32 kernels
2754 * where xfs_daddr_t is 32-bits but mount will warn us
2755 * off a > 1 TB filesystem before we get here.
2757 if ((ITEM_TYPE(item
) == XFS_LI_BUF
) ||
2758 (ITEM_TYPE(item
) == XFS_LI_6_1_BUF
) ||
2759 (ITEM_TYPE(item
) == XFS_LI_5_3_BUF
)) {
2760 if ((error
= xlog_recover_do_buffer_trans(log
, item
,
2763 } else if ((ITEM_TYPE(item
) == XFS_LI_INODE
) ||
2764 (ITEM_TYPE(item
) == XFS_LI_6_1_INODE
) ||
2765 (ITEM_TYPE(item
) == XFS_LI_5_3_INODE
)) {
2766 if ((error
= xlog_recover_do_inode_trans(log
, item
,
2769 } else if (ITEM_TYPE(item
) == XFS_LI_EFI
) {
2770 xlog_recover_do_efi_trans(log
, item
, trans
->r_lsn
,
2772 } else if (ITEM_TYPE(item
) == XFS_LI_EFD
) {
2773 xlog_recover_do_efd_trans(log
, item
, pass
);
2774 } else if (ITEM_TYPE(item
) == XFS_LI_DQUOT
) {
2775 if ((error
= xlog_recover_do_dquot_trans(log
, item
,
2778 } else if ((ITEM_TYPE(item
) == XFS_LI_QUOTAOFF
)) {
2779 if ((error
= xlog_recover_do_quotaoff_trans(log
, item
,
2783 xlog_warn("XFS: xlog_recover_do_trans");
2785 error
= XFS_ERROR(EIO
);
2788 item
= item
->ri_next
;
2789 } while (first_item
!= item
);
2795 * Free up any resources allocated by the transaction
2797 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2800 xlog_recover_free_trans(
2801 xlog_recover_t
*trans
)
2803 xlog_recover_item_t
*first_item
, *item
, *free_item
;
2806 item
= first_item
= trans
->r_itemq
;
2809 item
= item
->ri_next
;
2810 /* Free the regions in the item. */
2811 for (i
= 0; i
< free_item
->ri_cnt
; i
++) {
2812 kmem_free(free_item
->ri_buf
[i
].i_addr
,
2813 free_item
->ri_buf
[i
].i_len
);
2815 /* Free the item itself */
2816 kmem_free(free_item
->ri_buf
,
2817 (free_item
->ri_total
* sizeof(xfs_log_iovec_t
)));
2818 kmem_free(free_item
, sizeof(xlog_recover_item_t
));
2819 } while (first_item
!= item
);
2820 /* Free the transaction recover structure */
2821 kmem_free(trans
, sizeof(xlog_recover_t
));
2825 xlog_recover_commit_trans(
2828 xlog_recover_t
*trans
,
2833 if ((error
= xlog_recover_unlink_tid(q
, trans
)))
2835 if ((error
= xlog_recover_do_trans(log
, trans
, pass
)))
2837 xlog_recover_free_trans(trans
); /* no error */
2842 xlog_recover_unmount_trans(
2843 xlog_recover_t
*trans
)
2845 /* Do nothing now */
2846 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2851 * There are two valid states of the r_state field. 0 indicates that the
2852 * transaction structure is in a normal state. We have either seen the
2853 * start of the transaction or the last operation we added was not a partial
2854 * operation. If the last operation we added to the transaction was a
2855 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2857 * NOTE: skip LRs with 0 data length.
2860 xlog_recover_process_data(
2862 xlog_recover_t
*rhash
[],
2863 xlog_rec_header_t
*rhead
,
2869 xlog_op_header_t
*ohead
;
2870 xlog_recover_t
*trans
;
2876 lp
= dp
+ INT_GET(rhead
->h_len
, ARCH_CONVERT
);
2877 num_logops
= INT_GET(rhead
->h_num_logops
, ARCH_CONVERT
);
2879 /* check the log format matches our own - else we can't recover */
2880 if (xlog_header_check_recover(log
->l_mp
, rhead
))
2881 return (XFS_ERROR(EIO
));
2883 while ((dp
< lp
) && num_logops
) {
2884 ASSERT(dp
+ sizeof(xlog_op_header_t
) <= lp
);
2885 ohead
= (xlog_op_header_t
*)dp
;
2886 dp
+= sizeof(xlog_op_header_t
);
2887 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
2888 ohead
->oh_clientid
!= XFS_LOG
) {
2890 "XFS: xlog_recover_process_data: bad clientid");
2892 return (XFS_ERROR(EIO
));
2894 tid
= INT_GET(ohead
->oh_tid
, ARCH_CONVERT
);
2895 hash
= XLOG_RHASH(tid
);
2896 trans
= xlog_recover_find_tid(rhash
[hash
], tid
);
2897 if (trans
== NULL
) { /* not found; add new tid */
2898 if (ohead
->oh_flags
& XLOG_START_TRANS
)
2899 xlog_recover_new_tid(&rhash
[hash
], tid
,
2900 INT_GET(rhead
->h_lsn
, ARCH_CONVERT
));
2902 ASSERT(dp
+INT_GET(ohead
->oh_len
, ARCH_CONVERT
) <= lp
);
2903 flags
= ohead
->oh_flags
& ~XLOG_END_TRANS
;
2904 if (flags
& XLOG_WAS_CONT_TRANS
)
2905 flags
&= ~XLOG_CONTINUE_TRANS
;
2907 case XLOG_COMMIT_TRANS
:
2908 error
= xlog_recover_commit_trans(log
,
2909 &rhash
[hash
], trans
, pass
);
2911 case XLOG_UNMOUNT_TRANS
:
2912 error
= xlog_recover_unmount_trans(trans
);
2914 case XLOG_WAS_CONT_TRANS
:
2915 error
= xlog_recover_add_to_cont_trans(trans
,
2916 dp
, INT_GET(ohead
->oh_len
,
2919 case XLOG_START_TRANS
:
2921 "XFS: xlog_recover_process_data: bad transaction");
2923 error
= XFS_ERROR(EIO
);
2926 case XLOG_CONTINUE_TRANS
:
2927 error
= xlog_recover_add_to_trans(trans
,
2928 dp
, INT_GET(ohead
->oh_len
,
2933 "XFS: xlog_recover_process_data: bad flag");
2935 error
= XFS_ERROR(EIO
);
2941 dp
+= INT_GET(ohead
->oh_len
, ARCH_CONVERT
);
2948 * Process an extent free intent item that was recovered from
2949 * the log. We need to free the extents that it describes.
2952 xlog_recover_process_efi(
2954 xfs_efi_log_item_t
*efip
)
2956 xfs_efd_log_item_t
*efdp
;
2960 xfs_fsblock_t startblock_fsb
;
2962 ASSERT(!(efip
->efi_flags
& XFS_EFI_RECOVERED
));
2965 * First check the validity of the extents described by the
2966 * EFI. If any are bad, then assume that all are bad and
2967 * just toss the EFI.
2969 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
2970 extp
= &(efip
->efi_format
.efi_extents
[i
]);
2971 startblock_fsb
= XFS_BB_TO_FSB(mp
,
2972 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
2973 if ((startblock_fsb
== 0) ||
2974 (extp
->ext_len
== 0) ||
2975 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
2976 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
2978 * This will pull the EFI from the AIL and
2979 * free the memory associated with it.
2981 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
2986 tp
= xfs_trans_alloc(mp
, 0);
2987 xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0, 0, 0);
2988 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
2990 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
2991 extp
= &(efip
->efi_format
.efi_extents
[i
]);
2992 xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
2993 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
2997 efip
->efi_flags
|= XFS_EFI_RECOVERED
;
2998 xfs_trans_commit(tp
, 0, NULL
);
3002 * Verify that once we've encountered something other than an EFI
3003 * in the AIL that there are no more EFIs in the AIL.
3007 xlog_recover_check_ail(
3009 xfs_log_item_t
*lip
,
3015 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
3016 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
3018 * The check will be bogus if we restart from the
3019 * beginning of the AIL, so ASSERT that we don't.
3020 * We never should since we're holding the AIL lock
3023 ASSERT(gen
== orig_gen
);
3024 } while (lip
!= NULL
);
3029 * When this is called, all of the EFIs which did not have
3030 * corresponding EFDs should be in the AIL. What we do now
3031 * is free the extents associated with each one.
3033 * Since we process the EFIs in normal transactions, they
3034 * will be removed at some point after the commit. This prevents
3035 * us from just walking down the list processing each one.
3036 * We'll use a flag in the EFI to skip those that we've already
3037 * processed and use the AIL iteration mechanism's generation
3038 * count to try to speed this up at least a bit.
3040 * When we start, we know that the EFIs are the only things in
3041 * the AIL. As we process them, however, other items are added
3042 * to the AIL. Since everything added to the AIL must come after
3043 * everything already in the AIL, we stop processing as soon as
3044 * we see something other than an EFI in the AIL.
3047 xlog_recover_process_efis(
3050 xfs_log_item_t
*lip
;
3051 xfs_efi_log_item_t
*efip
;
3059 lip
= xfs_trans_first_ail(mp
, &gen
);
3060 while (lip
!= NULL
) {
3062 * We're done when we see something other than an EFI.
3064 if (lip
->li_type
!= XFS_LI_EFI
) {
3065 xlog_recover_check_ail(mp
, lip
, gen
);
3070 * Skip EFIs that we've already processed.
3072 efip
= (xfs_efi_log_item_t
*)lip
;
3073 if (efip
->efi_flags
& XFS_EFI_RECOVERED
) {
3074 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
3079 xlog_recover_process_efi(mp
, efip
);
3081 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
3087 * This routine performs a transaction to null out a bad inode pointer
3088 * in an agi unlinked inode hash bucket.
3091 xlog_recover_clear_agi_bucket(
3093 xfs_agnumber_t agno
,
3102 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3103 xfs_trans_reserve(tp
, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp
), 0, 0, 0);
3105 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
3106 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
3107 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
3109 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3113 agi
= XFS_BUF_TO_AGI(agibp
);
3114 if (INT_GET(agi
->agi_magicnum
, ARCH_CONVERT
) != XFS_AGI_MAGIC
) {
3115 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3118 ASSERT(INT_GET(agi
->agi_magicnum
, ARCH_CONVERT
) == XFS_AGI_MAGIC
);
3120 INT_SET(agi
->agi_unlinked
[bucket
], ARCH_CONVERT
, NULLAGINO
);
3121 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3122 (sizeof(xfs_agino_t
) * bucket
);
3123 xfs_trans_log_buf(tp
, agibp
, offset
,
3124 (offset
+ sizeof(xfs_agino_t
) - 1));
3126 (void) xfs_trans_commit(tp
, 0, NULL
);
3130 * xlog_iunlink_recover
3132 * This is called during recovery to process any inodes which
3133 * we unlinked but not freed when the system crashed. These
3134 * inodes will be on the lists in the AGI blocks. What we do
3135 * here is scan all the AGIs and fully truncate and free any
3136 * inodes found on the lists. Each inode is removed from the
3137 * lists when it has been fully truncated and is freed. The
3138 * freeing of the inode and its removal from the list must be
3142 xlog_recover_process_iunlinks(
3146 xfs_agnumber_t agno
;
3161 * Prevent any DMAPI event from being sent while in this function.
3163 mp_dmevmask
= mp
->m_dmevmask
;
3166 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3168 * Find the agi for this ag.
3170 agibp
= xfs_buf_read(mp
->m_ddev_targp
,
3171 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
3172 XFS_FSS_TO_BB(mp
, 1), 0);
3173 if (XFS_BUF_ISERROR(agibp
)) {
3174 xfs_ioerror_alert("xlog_recover_process_iunlinks(#1)",
3176 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)));
3178 agi
= XFS_BUF_TO_AGI(agibp
);
3179 ASSERT(XFS_AGI_MAGIC
==
3180 INT_GET(agi
->agi_magicnum
, ARCH_CONVERT
));
3182 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3184 agino
= INT_GET(agi
->agi_unlinked
[bucket
], ARCH_CONVERT
);
3185 while (agino
!= NULLAGINO
) {
3188 * Release the agi buffer so that it can
3189 * be acquired in the normal course of the
3190 * transaction to truncate and free the inode.
3192 xfs_buf_relse(agibp
);
3194 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3195 error
= xfs_iget(mp
, NULL
, ino
, 0, &ip
, 0);
3196 ASSERT(error
|| (ip
!= NULL
));
3200 * Get the on disk inode to find the
3201 * next inode in the bucket.
3203 error
= xfs_itobp(mp
, NULL
, ip
, &dip
,
3205 ASSERT(error
|| (dip
!= NULL
));
3209 ASSERT(ip
->i_d
.di_nlink
== 0);
3211 /* setup for the next pass */
3212 agino
= INT_GET(dip
->di_next_unlinked
,
3216 * Prevent any DMAPI event from
3217 * being sent when the
3218 * reference on the inode is
3221 ip
->i_d
.di_dmevmask
= 0;
3224 * If this is a new inode, handle
3225 * it specially. Otherwise,
3226 * just drop our reference to the
3227 * inode. If there are no
3228 * other references, this will
3230 * xfs_inactive() which will
3231 * truncate the file and free
3234 if (ip
->i_d
.di_mode
== 0)
3235 xfs_iput_new(ip
, 0);
3237 VN_RELE(XFS_ITOV(ip
));
3240 * We can't read in the inode
3241 * this bucket points to, or
3242 * this inode is messed up. Just
3243 * ditch this bucket of inodes. We
3244 * will lose some inodes and space,
3245 * but at least we won't hang. Call
3246 * xlog_recover_clear_agi_bucket()
3247 * to perform a transaction to clear
3248 * the inode pointer in the bucket.
3250 xlog_recover_clear_agi_bucket(mp
, agno
,
3257 * Reacquire the agibuffer and continue around
3260 agibp
= xfs_buf_read(mp
->m_ddev_targp
,
3261 XFS_AG_DADDR(mp
, agno
,
3263 XFS_FSS_TO_BB(mp
, 1), 0);
3264 if (XFS_BUF_ISERROR(agibp
)) {
3266 "xlog_recover_process_iunlinks(#2)",
3268 XFS_AG_DADDR(mp
, agno
,
3269 XFS_AGI_DADDR(mp
)));
3271 agi
= XFS_BUF_TO_AGI(agibp
);
3272 ASSERT(XFS_AGI_MAGIC
== INT_GET(
3273 agi
->agi_magicnum
, ARCH_CONVERT
));
3278 * Release the buffer for the current agi so we can
3279 * go on to the next one.
3281 xfs_buf_relse(agibp
);
3284 mp
->m_dmevmask
= mp_dmevmask
;
3290 xlog_pack_data_checksum(
3292 xlog_in_core_t
*iclog
,
3299 up
= (uint
*)iclog
->ic_datap
;
3300 /* divide length by 4 to get # words */
3301 for (i
= 0; i
< (size
>> 2); i
++) {
3302 chksum
^= INT_GET(*up
, ARCH_CONVERT
);
3305 INT_SET(iclog
->ic_header
.h_chksum
, ARCH_CONVERT
, chksum
);
3308 #define xlog_pack_data_checksum(log, iclog, size)
3312 * Stamp cycle number in every block
3317 xlog_in_core_t
*iclog
)
3320 int size
= iclog
->ic_offset
+ iclog
->ic_roundoff
;
3323 xlog_in_core_2_t
*xhdr
;
3325 xlog_pack_data_checksum(log
, iclog
, size
);
3327 cycle_lsn
= CYCLE_LSN_NOCONV(iclog
->ic_header
.h_lsn
, ARCH_CONVERT
);
3329 dp
= iclog
->ic_datap
;
3330 for (i
= 0; i
< BTOBB(size
) &&
3331 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3332 iclog
->ic_header
.h_cycle_data
[i
] = *(uint
*)dp
;
3333 *(uint
*)dp
= cycle_lsn
;
3337 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3338 xhdr
= (xlog_in_core_2_t
*)&iclog
->ic_header
;
3339 for ( ; i
< BTOBB(size
); i
++) {
3340 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3341 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3342 xhdr
[j
].hic_xheader
.xh_cycle_data
[k
] = *(uint
*)dp
;
3343 *(uint
*)dp
= cycle_lsn
;
3347 for (i
= 1; i
< log
->l_iclog_heads
; i
++) {
3348 xhdr
[i
].hic_xheader
.xh_cycle
= cycle_lsn
;
3353 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3355 xlog_unpack_data_checksum(
3356 xlog_rec_header_t
*rhead
,
3360 uint
*up
= (uint
*)dp
;
3363 /* divide length by 4 to get # words */
3364 for (i
=0; i
< INT_GET(rhead
->h_len
, ARCH_CONVERT
) >> 2; i
++) {
3365 chksum
^= INT_GET(*up
, ARCH_CONVERT
);
3368 if (chksum
!= INT_GET(rhead
->h_chksum
, ARCH_CONVERT
)) {
3369 if (!INT_ISZERO(rhead
->h_chksum
, ARCH_CONVERT
) ||
3370 ((log
->l_flags
& XLOG_CHKSUM_MISMATCH
) == 0)) {
3372 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)",
3373 INT_GET(rhead
->h_chksum
, ARCH_CONVERT
), chksum
);
3375 "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3376 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3378 "XFS: LogR this is a LogV2 filesystem");
3380 log
->l_flags
|= XLOG_CHKSUM_MISMATCH
;
3385 #define xlog_unpack_data_checksum(rhead, dp, log)
3390 xlog_rec_header_t
*rhead
,
3395 xlog_in_core_2_t
*xhdr
;
3397 for (i
= 0; i
< BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
)) &&
3398 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3399 *(uint
*)dp
= *(uint
*)&rhead
->h_cycle_data
[i
];
3403 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3404 xhdr
= (xlog_in_core_2_t
*)rhead
;
3405 for ( ; i
< BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
)); i
++) {
3406 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3407 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3408 *(uint
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
3413 xlog_unpack_data_checksum(rhead
, dp
, log
);
3417 xlog_valid_rec_header(
3419 xlog_rec_header_t
*rhead
,
3425 (INT_GET(rhead
->h_magicno
, ARCH_CONVERT
) !=
3426 XLOG_HEADER_MAGIC_NUM
))) {
3427 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3428 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3429 return XFS_ERROR(EFSCORRUPTED
);
3432 (INT_ISZERO(rhead
->h_version
, ARCH_CONVERT
) ||
3433 (INT_GET(rhead
->h_version
, ARCH_CONVERT
) &
3434 (~XLOG_VERSION_OKBITS
)) != 0))) {
3435 xlog_warn("XFS: %s: unrecognised log version (%d).",
3436 __FUNCTION__
, INT_GET(rhead
->h_version
, ARCH_CONVERT
));
3437 return XFS_ERROR(EIO
);
3440 /* LR body must have data or it wouldn't have been written */
3441 bblks
= INT_GET(rhead
->h_len
, ARCH_CONVERT
);
3442 if (unlikely( bblks
<= 0 || bblks
> INT_MAX
)) {
3443 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3444 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3445 return XFS_ERROR(EFSCORRUPTED
);
3447 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
3448 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3449 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3450 return XFS_ERROR(EFSCORRUPTED
);
3456 * Read the log from tail to head and process the log records found.
3457 * Handle the two cases where the tail and head are in the same cycle
3458 * and where the active portion of the log wraps around the end of
3459 * the physical log separately. The pass parameter is passed through
3460 * to the routines called to process the data and is not looked at
3464 xlog_do_recovery_pass(
3466 xfs_daddr_t head_blk
,
3467 xfs_daddr_t tail_blk
,
3470 xlog_rec_header_t
*rhead
;
3472 xfs_caddr_t bufaddr
, offset
;
3473 xfs_buf_t
*hbp
, *dbp
;
3474 int error
= 0, h_size
;
3475 int bblks
, split_bblks
;
3476 int hblks
, split_hblks
, wrapped_hblks
;
3477 xlog_recover_t
*rhash
[XLOG_RHASH_SIZE
];
3479 ASSERT(head_blk
!= tail_blk
);
3482 * Read the header of the tail block and get the iclog buffer size from
3483 * h_size. Use this to tell how many sectors make up the log header.
3485 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3487 * When using variable length iclogs, read first sector of
3488 * iclog header and extract the header size from it. Get a
3489 * new hbp that is the correct size.
3491 hbp
= xlog_get_bp(log
, 1);
3494 if ((error
= xlog_bread(log
, tail_blk
, 1, hbp
)))
3496 offset
= xlog_align(log
, tail_blk
, 1, hbp
);
3497 rhead
= (xlog_rec_header_t
*)offset
;
3498 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
3501 h_size
= INT_GET(rhead
->h_size
, ARCH_CONVERT
);
3502 if ((INT_GET(rhead
->h_version
, ARCH_CONVERT
)
3503 & XLOG_VERSION_2
) &&
3504 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
3505 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
3506 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
3509 hbp
= xlog_get_bp(log
, hblks
);
3514 ASSERT(log
->l_sectbb_log
== 0);
3516 hbp
= xlog_get_bp(log
, 1);
3517 h_size
= XLOG_BIG_RECORD_BSIZE
;
3522 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
3528 memset(rhash
, 0, sizeof(rhash
));
3529 if (tail_blk
<= head_blk
) {
3530 for (blk_no
= tail_blk
; blk_no
< head_blk
; ) {
3531 if ((error
= xlog_bread(log
, blk_no
, hblks
, hbp
)))
3533 offset
= xlog_align(log
, blk_no
, hblks
, hbp
);
3534 rhead
= (xlog_rec_header_t
*)offset
;
3535 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3539 /* blocks in data section */
3540 bblks
= (int)BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
));
3541 error
= xlog_bread(log
, blk_no
+ hblks
, bblks
, dbp
);
3544 offset
= xlog_align(log
, blk_no
+ hblks
, bblks
, dbp
);
3545 xlog_unpack_data(rhead
, offset
, log
);
3546 if ((error
= xlog_recover_process_data(log
,
3547 rhash
, rhead
, offset
, pass
)))
3549 blk_no
+= bblks
+ hblks
;
3553 * Perform recovery around the end of the physical log.
3554 * When the head is not on the same cycle number as the tail,
3555 * we can't do a sequential recovery as above.
3558 while (blk_no
< log
->l_logBBsize
) {
3560 * Check for header wrapping around physical end-of-log
3565 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
3566 /* Read header in one read */
3567 error
= xlog_bread(log
, blk_no
, hblks
, hbp
);
3570 offset
= xlog_align(log
, blk_no
, hblks
, hbp
);
3572 /* This LR is split across physical log end */
3573 if (blk_no
!= log
->l_logBBsize
) {
3574 /* some data before physical log end */
3575 ASSERT(blk_no
<= INT_MAX
);
3576 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
3577 ASSERT(split_hblks
> 0);
3578 if ((error
= xlog_bread(log
, blk_no
,
3581 offset
= xlog_align(log
, blk_no
,
3585 * Note: this black magic still works with
3586 * large sector sizes (non-512) only because:
3587 * - we increased the buffer size originally
3588 * by 1 sector giving us enough extra space
3589 * for the second read;
3590 * - the log start is guaranteed to be sector
3592 * - we read the log end (LR header start)
3593 * _first_, then the log start (LR header end)
3594 * - order is important.
3596 bufaddr
= XFS_BUF_PTR(hbp
);
3597 XFS_BUF_SET_PTR(hbp
,
3598 bufaddr
+ BBTOB(split_hblks
),
3599 BBTOB(hblks
- split_hblks
));
3600 wrapped_hblks
= hblks
- split_hblks
;
3601 error
= xlog_bread(log
, 0, wrapped_hblks
, hbp
);
3604 XFS_BUF_SET_PTR(hbp
, bufaddr
, hblks
);
3606 offset
= xlog_align(log
, 0,
3607 wrapped_hblks
, hbp
);
3609 rhead
= (xlog_rec_header_t
*)offset
;
3610 error
= xlog_valid_rec_header(log
, rhead
,
3611 split_hblks
? blk_no
: 0);
3615 bblks
= (int)BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
));
3618 /* Read in data for log record */
3619 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
3620 error
= xlog_bread(log
, blk_no
, bblks
, dbp
);
3623 offset
= xlog_align(log
, blk_no
, bblks
, dbp
);
3625 /* This log record is split across the
3626 * physical end of log */
3629 if (blk_no
!= log
->l_logBBsize
) {
3630 /* some data is before the physical
3632 ASSERT(!wrapped_hblks
);
3633 ASSERT(blk_no
<= INT_MAX
);
3635 log
->l_logBBsize
- (int)blk_no
;
3636 ASSERT(split_bblks
> 0);
3637 if ((error
= xlog_bread(log
, blk_no
,
3640 offset
= xlog_align(log
, blk_no
,
3644 * Note: this black magic still works with
3645 * large sector sizes (non-512) only because:
3646 * - we increased the buffer size originally
3647 * by 1 sector giving us enough extra space
3648 * for the second read;
3649 * - the log start is guaranteed to be sector
3651 * - we read the log end (LR header start)
3652 * _first_, then the log start (LR header end)
3653 * - order is important.
3655 bufaddr
= XFS_BUF_PTR(dbp
);
3656 XFS_BUF_SET_PTR(dbp
,
3657 bufaddr
+ BBTOB(split_bblks
),
3658 BBTOB(bblks
- split_bblks
));
3659 if ((error
= xlog_bread(log
, wrapped_hblks
,
3660 bblks
- split_bblks
, dbp
)))
3662 XFS_BUF_SET_PTR(dbp
, bufaddr
,
3663 XLOG_BIG_RECORD_BSIZE
);
3665 offset
= xlog_align(log
, wrapped_hblks
,
3666 bblks
- split_bblks
, dbp
);
3668 xlog_unpack_data(rhead
, offset
, log
);
3669 if ((error
= xlog_recover_process_data(log
, rhash
,
3670 rhead
, offset
, pass
)))
3675 ASSERT(blk_no
>= log
->l_logBBsize
);
3676 blk_no
-= log
->l_logBBsize
;
3678 /* read first part of physical log */
3679 while (blk_no
< head_blk
) {
3680 if ((error
= xlog_bread(log
, blk_no
, hblks
, hbp
)))
3682 offset
= xlog_align(log
, blk_no
, hblks
, hbp
);
3683 rhead
= (xlog_rec_header_t
*)offset
;
3684 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3687 bblks
= (int)BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
));
3688 if ((error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
)))
3690 offset
= xlog_align(log
, blk_no
+hblks
, bblks
, dbp
);
3691 xlog_unpack_data(rhead
, offset
, log
);
3692 if ((error
= xlog_recover_process_data(log
, rhash
,
3693 rhead
, offset
, pass
)))
3695 blk_no
+= bblks
+ hblks
;
3707 * Do the recovery of the log. We actually do this in two phases.
3708 * The two passes are necessary in order to implement the function
3709 * of cancelling a record written into the log. The first pass
3710 * determines those things which have been cancelled, and the
3711 * second pass replays log items normally except for those which
3712 * have been cancelled. The handling of the replay and cancellations
3713 * takes place in the log item type specific routines.
3715 * The table of items which have cancel records in the log is allocated
3716 * and freed at this level, since only here do we know when all of
3717 * the log recovery has been completed.
3720 xlog_do_log_recovery(
3722 xfs_daddr_t head_blk
,
3723 xfs_daddr_t tail_blk
)
3727 ASSERT(head_blk
!= tail_blk
);
3730 * First do a pass to find all of the cancelled buf log items.
3731 * Store them in the buf_cancel_table for use in the second pass.
3733 log
->l_buf_cancel_table
=
3734 (xfs_buf_cancel_t
**)kmem_zalloc(XLOG_BC_TABLE_SIZE
*
3735 sizeof(xfs_buf_cancel_t
*),
3737 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3738 XLOG_RECOVER_PASS1
);
3740 kmem_free(log
->l_buf_cancel_table
,
3741 XLOG_BC_TABLE_SIZE
* sizeof(xfs_buf_cancel_t
*));
3742 log
->l_buf_cancel_table
= NULL
;
3746 * Then do a second pass to actually recover the items in the log.
3747 * When it is complete free the table of buf cancel items.
3749 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3750 XLOG_RECOVER_PASS2
);
3755 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
3756 ASSERT(log
->l_buf_cancel_table
[i
] == NULL
);
3760 kmem_free(log
->l_buf_cancel_table
,
3761 XLOG_BC_TABLE_SIZE
* sizeof(xfs_buf_cancel_t
*));
3762 log
->l_buf_cancel_table
= NULL
;
3768 * Do the actual recovery
3773 xfs_daddr_t head_blk
,
3774 xfs_daddr_t tail_blk
)
3781 * First replay the images in the log.
3783 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
3788 XFS_bflush(log
->l_mp
->m_ddev_targp
);
3791 * If IO errors happened during recovery, bail out.
3793 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
3798 * We now update the tail_lsn since much of the recovery has completed
3799 * and there may be space available to use. If there were no extent
3800 * or iunlinks, we can free up the entire log and set the tail_lsn to
3801 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3802 * lsn of the last known good LR on disk. If there are extent frees
3803 * or iunlinks they will have some entries in the AIL; so we look at
3804 * the AIL to determine how to set the tail_lsn.
3806 xlog_assign_tail_lsn(log
->l_mp
);
3809 * Now that we've finished replaying all buffer and inode
3810 * updates, re-read in the superblock.
3812 bp
= xfs_getsb(log
->l_mp
, 0);
3815 xfsbdstrat(log
->l_mp
, bp
);
3816 if ((error
= xfs_iowait(bp
))) {
3817 xfs_ioerror_alert("xlog_do_recover",
3818 log
->l_mp
, bp
, XFS_BUF_ADDR(bp
));
3824 /* Convert superblock from on-disk format */
3825 sbp
= &log
->l_mp
->m_sb
;
3826 xfs_xlatesb(XFS_BUF_TO_SBP(bp
), sbp
, 1, ARCH_CONVERT
, XFS_SB_ALL_BITS
);
3827 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
3828 ASSERT(XFS_SB_GOOD_VERSION(sbp
));
3831 xlog_recover_check_summary(log
);
3833 /* Normal transactions can now occur */
3834 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
3839 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3841 * Return error or zero.
3848 xfs_daddr_t head_blk
, tail_blk
;
3851 /* find the tail of the log */
3852 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
, readonly
)))
3855 if (tail_blk
!= head_blk
) {
3856 /* There used to be a comment here:
3858 * disallow recovery on read-only mounts. note -- mount
3859 * checks for ENOSPC and turns it into an intelligent
3861 * ...but this is no longer true. Now, unless you specify
3862 * NORECOVERY (in which case this function would never be
3863 * called), we just go ahead and recover. We do this all
3864 * under the vfs layer, so we can get away with it unless
3865 * the device itself is read-only, in which case we fail.
3867 if ((error
= xfs_dev_is_read_only(log
->l_mp
,
3868 "recovery required"))) {
3873 "Starting XFS recovery on filesystem: %s (dev: %d/%d)",
3874 log
->l_mp
->m_fsname
, MAJOR(log
->l_dev
),
3877 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
3878 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
3884 * In the first part of recovery we replay inodes and buffers and build
3885 * up the list of extent free items which need to be processed. Here
3886 * we process the extent free items and clean up the on disk unlinked
3887 * inode lists. This is separated from the first part of recovery so
3888 * that the root and real-time bitmap inodes can be read in from disk in
3889 * between the two stages. This is necessary so that we can free space
3890 * in the real-time portion of the file system.
3893 xlog_recover_finish(
3898 * Now we're ready to do the transactions needed for the
3899 * rest of recovery. Start with completing all the extent
3900 * free intent records and then process the unlinked inode
3901 * lists. At this point, we essentially run in normal mode
3902 * except that we're still performing recovery actions
3903 * rather than accepting new requests.
3905 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
3906 xlog_recover_process_efis(log
);
3908 * Sync the log to get all the EFIs out of the AIL.
3909 * This isn't absolutely necessary, but it helps in
3910 * case the unlink transactions would have problems
3911 * pushing the EFIs out of the way.
3913 xfs_log_force(log
->l_mp
, (xfs_lsn_t
)0,
3914 (XFS_LOG_FORCE
| XFS_LOG_SYNC
));
3916 if ( (mfsi_flags
& XFS_MFSI_NOUNLINK
) == 0 ) {
3917 xlog_recover_process_iunlinks(log
);
3920 xlog_recover_check_summary(log
);
3923 "Ending XFS recovery on filesystem: %s (dev: %d/%d)",
3924 log
->l_mp
->m_fsname
, MAJOR(log
->l_dev
),
3927 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
3930 "!Ending clean XFS mount for filesystem: %s",
3931 log
->l_mp
->m_fsname
);
3939 * Read all of the agf and agi counters and check that they
3940 * are consistent with the superblock counters.
3943 xlog_recover_check_summary(
3951 xfs_daddr_t agfdaddr
;
3952 xfs_daddr_t agidaddr
;
3954 #ifdef XFS_LOUD_RECOVERY
3957 xfs_agnumber_t agno
;
3958 __uint64_t freeblks
;
3967 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3968 agfdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGF_DADDR(mp
));
3969 agfbp
= xfs_buf_read(mp
->m_ddev_targp
, agfdaddr
,
3970 XFS_FSS_TO_BB(mp
, 1), 0);
3971 if (XFS_BUF_ISERROR(agfbp
)) {
3972 xfs_ioerror_alert("xlog_recover_check_summary(agf)",
3973 mp
, agfbp
, agfdaddr
);
3975 agfp
= XFS_BUF_TO_AGF(agfbp
);
3976 ASSERT(XFS_AGF_MAGIC
==
3977 INT_GET(agfp
->agf_magicnum
, ARCH_CONVERT
));
3978 ASSERT(XFS_AGF_GOOD_VERSION(
3979 INT_GET(agfp
->agf_versionnum
, ARCH_CONVERT
)));
3980 ASSERT(INT_GET(agfp
->agf_seqno
, ARCH_CONVERT
) == agno
);
3982 freeblks
+= INT_GET(agfp
->agf_freeblks
, ARCH_CONVERT
) +
3983 INT_GET(agfp
->agf_flcount
, ARCH_CONVERT
);
3984 xfs_buf_relse(agfbp
);
3986 agidaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
3987 agibp
= xfs_buf_read(mp
->m_ddev_targp
, agidaddr
,
3988 XFS_FSS_TO_BB(mp
, 1), 0);
3989 if (XFS_BUF_ISERROR(agibp
)) {
3990 xfs_ioerror_alert("xlog_recover_check_summary(agi)",
3991 log
->l_mp
, agibp
, agidaddr
);
3993 agip
= XFS_BUF_TO_AGI(agibp
);
3994 ASSERT(XFS_AGI_MAGIC
==
3995 INT_GET(agip
->agi_magicnum
, ARCH_CONVERT
));
3996 ASSERT(XFS_AGI_GOOD_VERSION(
3997 INT_GET(agip
->agi_versionnum
, ARCH_CONVERT
)));
3998 ASSERT(INT_GET(agip
->agi_seqno
, ARCH_CONVERT
) == agno
);
4000 itotal
+= INT_GET(agip
->agi_count
, ARCH_CONVERT
);
4001 ifree
+= INT_GET(agip
->agi_freecount
, ARCH_CONVERT
);
4002 xfs_buf_relse(agibp
);
4005 sbbp
= xfs_getsb(mp
, 0);
4006 #ifdef XFS_LOUD_RECOVERY
4007 sbp
= XFS_BUF_TO_SBP(sbbp
);
4009 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4010 sbp
->sb_icount
, itotal
);
4012 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4013 sbp
->sb_ifree
, ifree
);
4015 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4016 sbp
->sb_fdblocks
, freeblks
);
4019 * This is turned off until I account for the allocation
4020 * btree blocks which live in free space.
4022 ASSERT(sbp
->sb_icount
== itotal
);
4023 ASSERT(sbp
->sb_ifree
== ifree
);
4024 ASSERT(sbp
->sb_fdblocks
== freeblks
);
4027 xfs_buf_relse(sbbp
);