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_types.h"
38 #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_btree.h"
48 #include "xfs_ialloc_btree.h"
49 #include "xfs_dir_sf.h"
50 #include "xfs_dir2_sf.h"
51 #include "xfs_attr_sf.h"
52 #include "xfs_dinode.h"
53 #include "xfs_inode.h"
54 #include "xfs_inode_item.h"
56 #include "xfs_alloc.h"
57 #include "xfs_ialloc.h"
58 #include "xfs_log_priv.h"
59 #include "xfs_buf_item.h"
60 #include "xfs_log_recover.h"
61 #include "xfs_extfree_item.h"
62 #include "xfs_trans_priv.h"
63 #include "xfs_quota.h"
66 STATIC
int xlog_find_zeroed(xlog_t
*, xfs_daddr_t
*);
67 STATIC
int xlog_clear_stale_blocks(xlog_t
*, xfs_lsn_t
);
68 STATIC
void xlog_recover_insert_item_backq(xlog_recover_item_t
**q
,
69 xlog_recover_item_t
*item
);
71 STATIC
void xlog_recover_check_summary(xlog_t
*);
72 STATIC
void xlog_recover_check_ail(xfs_mount_t
*, xfs_log_item_t
*, int);
74 #define xlog_recover_check_summary(log)
75 #define xlog_recover_check_ail(mp, lip, gen)
80 * Sector aligned buffer routines for buffer create/read/write/access
83 #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
84 ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
85 ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
86 #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
93 ASSERT(num_bblks
> 0);
95 if (log
->l_sectbb_log
) {
97 num_bblks
+= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, 1);
98 num_bblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, num_bblks
);
100 return xfs_buf_get_noaddr(BBTOB(num_bblks
), log
->l_mp
->m_logdev_targp
);
112 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
123 if (log
->l_sectbb_log
) {
124 blk_no
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, blk_no
);
125 nbblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, nbblks
);
129 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
132 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
135 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
136 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
138 xfsbdstrat(log
->l_mp
, bp
);
139 if ((error
= xfs_iowait(bp
)))
140 xfs_ioerror_alert("xlog_bread", log
->l_mp
,
141 bp
, XFS_BUF_ADDR(bp
));
146 * Write out the buffer at the given block for the given number of blocks.
147 * The buffer is kept locked across the write and is returned locked.
148 * This can only be used for synchronous log writes.
159 if (log
->l_sectbb_log
) {
160 blk_no
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, blk_no
);
161 nbblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, nbblks
);
165 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
167 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
168 XFS_BUF_ZEROFLAGS(bp
);
171 XFS_BUF_PSEMA(bp
, PRIBIO
);
172 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
173 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
175 if ((error
= xfs_bwrite(log
->l_mp
, bp
)))
176 xfs_ioerror_alert("xlog_bwrite", log
->l_mp
,
177 bp
, XFS_BUF_ADDR(bp
));
190 if (!log
->l_sectbb_log
)
191 return XFS_BUF_PTR(bp
);
193 ptr
= XFS_BUF_PTR(bp
) + BBTOB((int)blk_no
& log
->l_sectbb_mask
);
194 ASSERT(XFS_BUF_SIZE(bp
) >=
195 BBTOB(nbblks
+ (blk_no
& log
->l_sectbb_mask
)));
201 * dump debug superblock and log record information
204 xlog_header_check_dump(
206 xlog_rec_header_t
*head
)
210 printk("%s: SB : uuid = ", __FUNCTION__
);
211 for (b
= 0; b
< 16; b
++)
212 printk("%02x",((unsigned char *)&mp
->m_sb
.sb_uuid
)[b
]);
213 printk(", fmt = %d\n", XLOG_FMT
);
214 printk(" log : uuid = ");
215 for (b
= 0; b
< 16; b
++)
216 printk("%02x",((unsigned char *)&head
->h_fs_uuid
)[b
]);
217 printk(", fmt = %d\n", INT_GET(head
->h_fmt
, ARCH_CONVERT
));
220 #define xlog_header_check_dump(mp, head)
224 * check log record header for recovery
227 xlog_header_check_recover(
229 xlog_rec_header_t
*head
)
231 ASSERT(INT_GET(head
->h_magicno
, ARCH_CONVERT
) == XLOG_HEADER_MAGIC_NUM
);
234 * IRIX doesn't write the h_fmt field and leaves it zeroed
235 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
236 * a dirty log created in IRIX.
238 if (unlikely(INT_GET(head
->h_fmt
, ARCH_CONVERT
) != XLOG_FMT
)) {
240 "XFS: dirty log written in incompatible format - can't recover");
241 xlog_header_check_dump(mp
, head
);
242 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
243 XFS_ERRLEVEL_HIGH
, mp
);
244 return XFS_ERROR(EFSCORRUPTED
);
245 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
247 "XFS: dirty log entry has mismatched uuid - can't recover");
248 xlog_header_check_dump(mp
, head
);
249 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
250 XFS_ERRLEVEL_HIGH
, mp
);
251 return XFS_ERROR(EFSCORRUPTED
);
257 * read the head block of the log and check the header
260 xlog_header_check_mount(
262 xlog_rec_header_t
*head
)
264 ASSERT(INT_GET(head
->h_magicno
, ARCH_CONVERT
) == XLOG_HEADER_MAGIC_NUM
);
266 if (uuid_is_nil(&head
->h_fs_uuid
)) {
268 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
269 * h_fs_uuid is nil, we assume this log was last mounted
270 * by IRIX and continue.
272 xlog_warn("XFS: nil uuid in log - IRIX style log");
273 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
274 xlog_warn("XFS: log has mismatched uuid - can't recover");
275 xlog_header_check_dump(mp
, head
);
276 XFS_ERROR_REPORT("xlog_header_check_mount",
277 XFS_ERRLEVEL_HIGH
, mp
);
278 return XFS_ERROR(EFSCORRUPTED
);
289 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *));
291 if (XFS_BUF_GETERROR(bp
)) {
293 * We're not going to bother about retrying
294 * this during recovery. One strike!
296 mp
= XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*);
297 xfs_ioerror_alert("xlog_recover_iodone",
298 mp
, bp
, XFS_BUF_ADDR(bp
));
299 xfs_force_shutdown(mp
, XFS_METADATA_IO_ERROR
);
301 XFS_BUF_SET_FSPRIVATE(bp
, NULL
);
302 XFS_BUF_CLR_IODONE_FUNC(bp
);
307 * This routine finds (to an approximation) the first block in the physical
308 * log which contains the given cycle. It uses a binary search algorithm.
309 * Note that the algorithm can not be perfect because the disk will not
310 * necessarily be perfect.
313 xlog_find_cycle_start(
316 xfs_daddr_t first_blk
,
317 xfs_daddr_t
*last_blk
,
325 mid_blk
= BLK_AVG(first_blk
, *last_blk
);
326 while (mid_blk
!= first_blk
&& mid_blk
!= *last_blk
) {
327 if ((error
= xlog_bread(log
, mid_blk
, 1, bp
)))
329 offset
= xlog_align(log
, mid_blk
, 1, bp
);
330 mid_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
331 if (mid_cycle
== cycle
) {
333 /* last_half_cycle == mid_cycle */
336 /* first_half_cycle == mid_cycle */
338 mid_blk
= BLK_AVG(first_blk
, *last_blk
);
340 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == *last_blk
) ||
341 (mid_blk
== *last_blk
&& mid_blk
-1 == first_blk
));
347 * Check that the range of blocks does not contain the cycle number
348 * given. The scan needs to occur from front to back and the ptr into the
349 * region must be updated since a later routine will need to perform another
350 * test. If the region is completely good, we end up returning the same
353 * Set blkno to -1 if we encounter no errors. This is an invalid block number
354 * since we don't ever expect logs to get this large.
357 xlog_find_verify_cycle(
359 xfs_daddr_t start_blk
,
361 uint stop_on_cycle_no
,
362 xfs_daddr_t
*new_blk
)
368 xfs_caddr_t buf
= NULL
;
371 bufblks
= 1 << ffs(nbblks
);
373 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
374 /* can't get enough memory to do everything in one big buffer */
376 if (bufblks
<= log
->l_sectbb_log
)
380 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
383 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
385 if ((error
= xlog_bread(log
, i
, bcount
, bp
)))
388 buf
= xlog_align(log
, i
, bcount
, bp
);
389 for (j
= 0; j
< bcount
; j
++) {
390 cycle
= GET_CYCLE(buf
, ARCH_CONVERT
);
391 if (cycle
== stop_on_cycle_no
) {
408 * Potentially backup over partial log record write.
410 * In the typical case, last_blk is the number of the block directly after
411 * a good log record. Therefore, we subtract one to get the block number
412 * of the last block in the given buffer. extra_bblks contains the number
413 * of blocks we would have read on a previous read. This happens when the
414 * last log record is split over the end of the physical log.
416 * extra_bblks is the number of blocks potentially verified on a previous
417 * call to this routine.
420 xlog_find_verify_log_record(
422 xfs_daddr_t start_blk
,
423 xfs_daddr_t
*last_blk
,
428 xfs_caddr_t offset
= NULL
;
429 xlog_rec_header_t
*head
= NULL
;
432 int num_blks
= *last_blk
- start_blk
;
435 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
437 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
438 if (!(bp
= xlog_get_bp(log
, 1)))
442 if ((error
= xlog_bread(log
, start_blk
, num_blks
, bp
)))
444 offset
= xlog_align(log
, start_blk
, num_blks
, bp
);
445 offset
+= ((num_blks
- 1) << BBSHIFT
);
448 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
450 /* valid log record not found */
452 "XFS: Log inconsistent (didn't find previous header)");
454 error
= XFS_ERROR(EIO
);
459 if ((error
= xlog_bread(log
, i
, 1, bp
)))
461 offset
= xlog_align(log
, i
, 1, bp
);
464 head
= (xlog_rec_header_t
*)offset
;
466 if (XLOG_HEADER_MAGIC_NUM
==
467 INT_GET(head
->h_magicno
, ARCH_CONVERT
))
475 * We hit the beginning of the physical log & still no header. Return
476 * to caller. If caller can handle a return of -1, then this routine
477 * will be called again for the end of the physical log.
485 * We have the final block of the good log (the first block
486 * of the log record _before_ the head. So we check the uuid.
488 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
492 * We may have found a log record header before we expected one.
493 * last_blk will be the 1st block # with a given cycle #. We may end
494 * up reading an entire log record. In this case, we don't want to
495 * reset last_blk. Only when last_blk points in the middle of a log
496 * record do we update last_blk.
498 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
499 uint h_size
= INT_GET(head
->h_size
, ARCH_CONVERT
);
501 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
502 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
508 if (*last_blk
- i
+ extra_bblks
509 != BTOBB(INT_GET(head
->h_len
, ARCH_CONVERT
)) + xhdrs
)
518 * Head is defined to be the point of the log where the next log write
519 * write could go. This means that incomplete LR writes at the end are
520 * eliminated when calculating the head. We aren't guaranteed that previous
521 * LR have complete transactions. We only know that a cycle number of
522 * current cycle number -1 won't be present in the log if we start writing
523 * from our current block number.
525 * last_blk contains the block number of the first block with a given
528 * Return: zero if normal, non-zero if error.
533 xfs_daddr_t
*return_head_blk
)
537 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
539 uint first_half_cycle
, last_half_cycle
;
541 int error
, log_bbnum
= log
->l_logBBsize
;
543 /* Is the end of the log device zeroed? */
544 if ((error
= xlog_find_zeroed(log
, &first_blk
)) == -1) {
545 *return_head_blk
= first_blk
;
547 /* Is the whole lot zeroed? */
549 /* Linux XFS shouldn't generate totally zeroed logs -
550 * mkfs etc write a dummy unmount record to a fresh
551 * log so we can store the uuid in there
553 xlog_warn("XFS: totally zeroed log");
558 xlog_warn("XFS: empty log check failed");
562 first_blk
= 0; /* get cycle # of 1st block */
563 bp
= xlog_get_bp(log
, 1);
566 if ((error
= xlog_bread(log
, 0, 1, bp
)))
568 offset
= xlog_align(log
, 0, 1, bp
);
569 first_half_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
571 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
572 if ((error
= xlog_bread(log
, last_blk
, 1, bp
)))
574 offset
= xlog_align(log
, last_blk
, 1, bp
);
575 last_half_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
576 ASSERT(last_half_cycle
!= 0);
579 * If the 1st half cycle number is equal to the last half cycle number,
580 * then the entire log is stamped with the same cycle number. In this
581 * case, head_blk can't be set to zero (which makes sense). The below
582 * math doesn't work out properly with head_blk equal to zero. Instead,
583 * we set it to log_bbnum which is an invalid block number, but this
584 * value makes the math correct. If head_blk doesn't changed through
585 * all the tests below, *head_blk is set to zero at the very end rather
586 * than log_bbnum. In a sense, log_bbnum and zero are the same block
587 * in a circular file.
589 if (first_half_cycle
== last_half_cycle
) {
591 * In this case we believe that the entire log should have
592 * cycle number last_half_cycle. We need to scan backwards
593 * from the end verifying that there are no holes still
594 * containing last_half_cycle - 1. If we find such a hole,
595 * then the start of that hole will be the new head. The
596 * simple case looks like
597 * x | x ... | x - 1 | x
598 * Another case that fits this picture would be
599 * x | x + 1 | x ... | x
600 * In this case the head really is somwhere at the end of the
601 * log, as one of the latest writes at the beginning was
604 * x | x + 1 | x ... | x - 1 | x
605 * This is really the combination of the above two cases, and
606 * the head has to end up at the start of the x-1 hole at the
609 * In the 256k log case, we will read from the beginning to the
610 * end of the log and search for cycle numbers equal to x-1.
611 * We don't worry about the x+1 blocks that we encounter,
612 * because we know that they cannot be the head since the log
615 head_blk
= log_bbnum
;
616 stop_on_cycle
= last_half_cycle
- 1;
619 * In this case we want to find the first block with cycle
620 * number matching last_half_cycle. We expect the log to be
623 * The first block with cycle number x (last_half_cycle) will
624 * be where the new head belongs. First we do a binary search
625 * for the first occurrence of last_half_cycle. The binary
626 * search may not be totally accurate, so then we scan back
627 * from there looking for occurrences of last_half_cycle before
628 * us. If that backwards scan wraps around the beginning of
629 * the log, then we look for occurrences of last_half_cycle - 1
630 * at the end of the log. The cases we're looking for look
632 * x + 1 ... | x | x + 1 | x ...
633 * ^ binary search stopped here
635 * x + 1 ... | x ... | x - 1 | x
636 * <---------> less than scan distance
638 stop_on_cycle
= last_half_cycle
;
639 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
640 &head_blk
, last_half_cycle
)))
645 * Now validate the answer. Scan back some number of maximum possible
646 * blocks and make sure each one has the expected cycle number. The
647 * maximum is determined by the total possible amount of buffering
648 * in the in-core log. The following number can be made tighter if
649 * we actually look at the block size of the filesystem.
651 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
652 if (head_blk
>= num_scan_bblks
) {
654 * We are guaranteed that the entire check can be performed
657 start_blk
= head_blk
- num_scan_bblks
;
658 if ((error
= xlog_find_verify_cycle(log
,
659 start_blk
, num_scan_bblks
,
660 stop_on_cycle
, &new_blk
)))
664 } else { /* need to read 2 parts of log */
666 * We are going to scan backwards in the log in two parts.
667 * First we scan the physical end of the log. In this part
668 * of the log, we are looking for blocks with cycle number
669 * last_half_cycle - 1.
670 * If we find one, then we know that the log starts there, as
671 * we've found a hole that didn't get written in going around
672 * the end of the physical log. The simple case for this is
673 * x + 1 ... | x ... | x - 1 | x
674 * <---------> less than scan distance
675 * If all of the blocks at the end of the log have cycle number
676 * last_half_cycle, then we check the blocks at the start of
677 * the log looking for occurrences of last_half_cycle. If we
678 * find one, then our current estimate for the location of the
679 * first occurrence of last_half_cycle is wrong and we move
680 * back to the hole we've found. This case looks like
681 * x + 1 ... | x | x + 1 | x ...
682 * ^ binary search stopped here
683 * Another case we need to handle that only occurs in 256k
685 * x + 1 ... | x ... | x+1 | x ...
686 * ^ binary search stops here
687 * In a 256k log, the scan at the end of the log will see the
688 * x + 1 blocks. We need to skip past those since that is
689 * certainly not the head of the log. By searching for
690 * last_half_cycle-1 we accomplish that.
692 start_blk
= log_bbnum
- num_scan_bblks
+ head_blk
;
693 ASSERT(head_blk
<= INT_MAX
&&
694 (xfs_daddr_t
) num_scan_bblks
- head_blk
>= 0);
695 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
696 num_scan_bblks
- (int)head_blk
,
697 (stop_on_cycle
- 1), &new_blk
)))
705 * Scan beginning of log now. The last part of the physical
706 * log is good. This scan needs to verify that it doesn't find
707 * the last_half_cycle.
710 ASSERT(head_blk
<= INT_MAX
);
711 if ((error
= xlog_find_verify_cycle(log
,
712 start_blk
, (int)head_blk
,
713 stop_on_cycle
, &new_blk
)))
721 * Now we need to make sure head_blk is not pointing to a block in
722 * the middle of a log record.
724 num_scan_bblks
= XLOG_REC_SHIFT(log
);
725 if (head_blk
>= num_scan_bblks
) {
726 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
728 /* start ptr at last block ptr before head_blk */
729 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
730 &head_blk
, 0)) == -1) {
731 error
= XFS_ERROR(EIO
);
737 ASSERT(head_blk
<= INT_MAX
);
738 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
739 &head_blk
, 0)) == -1) {
740 /* We hit the beginning of the log during our search */
741 start_blk
= log_bbnum
- num_scan_bblks
+ head_blk
;
743 ASSERT(start_blk
<= INT_MAX
&&
744 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
745 ASSERT(head_blk
<= INT_MAX
);
746 if ((error
= xlog_find_verify_log_record(log
,
748 (int)head_blk
)) == -1) {
749 error
= XFS_ERROR(EIO
);
753 if (new_blk
!= log_bbnum
)
760 if (head_blk
== log_bbnum
)
761 *return_head_blk
= 0;
763 *return_head_blk
= head_blk
;
765 * When returning here, we have a good block number. Bad block
766 * means that during a previous crash, we didn't have a clean break
767 * from cycle number N to cycle number N-1. In this case, we need
768 * to find the first block with cycle number N-1.
776 xlog_warn("XFS: failed to find log head");
781 * Find the sync block number or the tail of the log.
783 * This will be the block number of the last record to have its
784 * associated buffers synced to disk. Every log record header has
785 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
786 * to get a sync block number. The only concern is to figure out which
787 * log record header to believe.
789 * The following algorithm uses the log record header with the largest
790 * lsn. The entire log record does not need to be valid. We only care
791 * that the header is valid.
793 * We could speed up search by using current head_blk buffer, but it is not
799 xfs_daddr_t
*head_blk
,
800 xfs_daddr_t
*tail_blk
,
803 xlog_rec_header_t
*rhead
;
804 xlog_op_header_t
*op_head
;
805 xfs_caddr_t offset
= NULL
;
808 xfs_daddr_t umount_data_blk
;
809 xfs_daddr_t after_umount_blk
;
816 * Find previous log record
818 if ((error
= xlog_find_head(log
, head_blk
)))
821 bp
= xlog_get_bp(log
, 1);
824 if (*head_blk
== 0) { /* special case */
825 if ((error
= xlog_bread(log
, 0, 1, bp
)))
827 offset
= xlog_align(log
, 0, 1, bp
);
828 if (GET_CYCLE(offset
, ARCH_CONVERT
) == 0) {
830 /* leave all other log inited values alone */
836 * Search backwards looking for log record header block
838 ASSERT(*head_blk
< INT_MAX
);
839 for (i
= (int)(*head_blk
) - 1; i
>= 0; i
--) {
840 if ((error
= xlog_bread(log
, i
, 1, bp
)))
842 offset
= xlog_align(log
, i
, 1, bp
);
843 if (XLOG_HEADER_MAGIC_NUM
==
844 INT_GET(*(uint
*)offset
, ARCH_CONVERT
)) {
850 * If we haven't found the log record header block, start looking
851 * again from the end of the physical log. XXXmiken: There should be
852 * a check here to make sure we didn't search more than N blocks in
856 for (i
= log
->l_logBBsize
- 1; i
>= (int)(*head_blk
); i
--) {
857 if ((error
= xlog_bread(log
, i
, 1, bp
)))
859 offset
= xlog_align(log
, i
, 1, bp
);
860 if (XLOG_HEADER_MAGIC_NUM
==
861 INT_GET(*(uint
*)offset
, ARCH_CONVERT
)) {
868 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
870 return XFS_ERROR(EIO
);
873 /* find blk_no of tail of log */
874 rhead
= (xlog_rec_header_t
*)offset
;
875 *tail_blk
= BLOCK_LSN(INT_GET(rhead
->h_tail_lsn
, ARCH_CONVERT
));
878 * Reset log values according to the state of the log when we
879 * crashed. In the case where head_blk == 0, we bump curr_cycle
880 * one because the next write starts a new cycle rather than
881 * continuing the cycle of the last good log record. At this
882 * point we have guaranteed that all partial log records have been
883 * accounted for. Therefore, we know that the last good log record
884 * written was complete and ended exactly on the end boundary
885 * of the physical log.
887 log
->l_prev_block
= i
;
888 log
->l_curr_block
= (int)*head_blk
;
889 log
->l_curr_cycle
= INT_GET(rhead
->h_cycle
, ARCH_CONVERT
);
892 log
->l_tail_lsn
= INT_GET(rhead
->h_tail_lsn
, ARCH_CONVERT
);
893 log
->l_last_sync_lsn
= INT_GET(rhead
->h_lsn
, ARCH_CONVERT
);
894 log
->l_grant_reserve_cycle
= log
->l_curr_cycle
;
895 log
->l_grant_reserve_bytes
= BBTOB(log
->l_curr_block
);
896 log
->l_grant_write_cycle
= log
->l_curr_cycle
;
897 log
->l_grant_write_bytes
= BBTOB(log
->l_curr_block
);
900 * Look for unmount record. If we find it, then we know there
901 * was a clean unmount. Since 'i' could be the last block in
902 * the physical log, we convert to a log block before comparing
905 * Save the current tail lsn to use to pass to
906 * xlog_clear_stale_blocks() below. We won't want to clear the
907 * unmount record if there is one, so we pass the lsn of the
908 * unmount record rather than the block after it.
910 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
911 int h_size
= INT_GET(rhead
->h_size
, ARCH_CONVERT
);
912 int h_version
= INT_GET(rhead
->h_version
, ARCH_CONVERT
);
914 if ((h_version
& XLOG_VERSION_2
) &&
915 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
916 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
917 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
925 after_umount_blk
= (i
+ hblks
+ (int)
926 BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
))) % log
->l_logBBsize
;
927 tail_lsn
= log
->l_tail_lsn
;
928 if (*head_blk
== after_umount_blk
&&
929 INT_GET(rhead
->h_num_logops
, ARCH_CONVERT
) == 1) {
930 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
931 if ((error
= xlog_bread(log
, umount_data_blk
, 1, bp
))) {
934 offset
= xlog_align(log
, umount_data_blk
, 1, bp
);
935 op_head
= (xlog_op_header_t
*)offset
;
936 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
938 * Set tail and last sync so that newly written
939 * log records will point recovery to after the
940 * current unmount record.
942 ASSIGN_ANY_LSN_HOST(log
->l_tail_lsn
, log
->l_curr_cycle
,
944 ASSIGN_ANY_LSN_HOST(log
->l_last_sync_lsn
, log
->l_curr_cycle
,
946 *tail_blk
= after_umount_blk
;
951 * Make sure that there are no blocks in front of the head
952 * with the same cycle number as the head. This can happen
953 * because we allow multiple outstanding log writes concurrently,
954 * and the later writes might make it out before earlier ones.
956 * We use the lsn from before modifying it so that we'll never
957 * overwrite the unmount record after a clean unmount.
959 * Do this only if we are going to recover the filesystem
961 * NOTE: This used to say "if (!readonly)"
962 * However on Linux, we can & do recover a read-only filesystem.
963 * We only skip recovery if NORECOVERY is specified on mount,
964 * in which case we would not be here.
966 * But... if the -device- itself is readonly, just skip this.
967 * We can't recover this device anyway, so it won't matter.
969 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
)) {
970 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
978 xlog_warn("XFS: failed to locate log tail");
983 * Is the log zeroed at all?
985 * The last binary search should be changed to perform an X block read
986 * once X becomes small enough. You can then search linearly through
987 * the X blocks. This will cut down on the number of reads we need to do.
989 * If the log is partially zeroed, this routine will pass back the blkno
990 * of the first block with cycle number 0. It won't have a complete LR
994 * 0 => the log is completely written to
995 * -1 => use *blk_no as the first block of the log
996 * >0 => error has occurred
1001 xfs_daddr_t
*blk_no
)
1005 uint first_cycle
, last_cycle
;
1006 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1007 xfs_daddr_t num_scan_bblks
;
1008 int error
, log_bbnum
= log
->l_logBBsize
;
1010 /* check totally zeroed log */
1011 bp
= xlog_get_bp(log
, 1);
1014 if ((error
= xlog_bread(log
, 0, 1, bp
)))
1016 offset
= xlog_align(log
, 0, 1, bp
);
1017 first_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
1018 if (first_cycle
== 0) { /* completely zeroed log */
1024 /* check partially zeroed log */
1025 if ((error
= xlog_bread(log
, log_bbnum
-1, 1, bp
)))
1027 offset
= xlog_align(log
, log_bbnum
-1, 1, bp
);
1028 last_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
1029 if (last_cycle
!= 0) { /* log completely written to */
1032 } else if (first_cycle
!= 1) {
1034 * If the cycle of the last block is zero, the cycle of
1035 * the first block must be 1. If it's not, maybe we're
1036 * not looking at a log... Bail out.
1038 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1039 return XFS_ERROR(EINVAL
);
1042 /* we have a partially zeroed log */
1043 last_blk
= log_bbnum
-1;
1044 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1048 * Validate the answer. Because there is no way to guarantee that
1049 * the entire log is made up of log records which are the same size,
1050 * we scan over the defined maximum blocks. At this point, the maximum
1051 * is not chosen to mean anything special. XXXmiken
1053 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1054 ASSERT(num_scan_bblks
<= INT_MAX
);
1056 if (last_blk
< num_scan_bblks
)
1057 num_scan_bblks
= last_blk
;
1058 start_blk
= last_blk
- num_scan_bblks
;
1061 * We search for any instances of cycle number 0 that occur before
1062 * our current estimate of the head. What we're trying to detect is
1063 * 1 ... | 0 | 1 | 0...
1064 * ^ binary search ends here
1066 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1067 (int)num_scan_bblks
, 0, &new_blk
)))
1073 * Potentially backup over partial log record write. We don't need
1074 * to search the end of the log because we know it is zero.
1076 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
1077 &last_blk
, 0)) == -1) {
1078 error
= XFS_ERROR(EIO
);
1092 * These are simple subroutines used by xlog_clear_stale_blocks() below
1093 * to initialize a buffer full of empty log record headers and write
1094 * them into the log.
1105 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1107 memset(buf
, 0, BBSIZE
);
1108 INT_SET(recp
->h_magicno
, ARCH_CONVERT
, XLOG_HEADER_MAGIC_NUM
);
1109 INT_SET(recp
->h_cycle
, ARCH_CONVERT
, cycle
);
1110 INT_SET(recp
->h_version
, ARCH_CONVERT
,
1111 XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
) ? 2 : 1);
1112 ASSIGN_ANY_LSN_DISK(recp
->h_lsn
, cycle
, block
);
1113 ASSIGN_ANY_LSN_DISK(recp
->h_tail_lsn
, tail_cycle
, tail_block
);
1114 INT_SET(recp
->h_fmt
, ARCH_CONVERT
, XLOG_FMT
);
1115 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1119 xlog_write_log_records(
1130 int sectbb
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, 1);
1131 int end_block
= start_block
+ blocks
;
1136 bufblks
= 1 << ffs(blocks
);
1137 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1139 if (bufblks
<= log
->l_sectbb_log
)
1143 /* We may need to do a read at the start to fill in part of
1144 * the buffer in the starting sector not covered by the first
1147 balign
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, start_block
);
1148 if (balign
!= start_block
) {
1149 if ((error
= xlog_bread(log
, start_block
, 1, bp
))) {
1153 j
= start_block
- balign
;
1156 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1157 int bcount
, endcount
;
1159 bcount
= min(bufblks
, end_block
- start_block
);
1160 endcount
= bcount
- j
;
1162 /* We may need to do a read at the end to fill in part of
1163 * the buffer in the final sector not covered by the write.
1164 * If this is the same sector as the above read, skip it.
1166 ealign
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, end_block
);
1167 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1168 offset
= XFS_BUF_PTR(bp
);
1169 balign
= BBTOB(ealign
- start_block
);
1170 XFS_BUF_SET_PTR(bp
, offset
+ balign
, BBTOB(sectbb
));
1171 if ((error
= xlog_bread(log
, ealign
, sectbb
, bp
)))
1173 XFS_BUF_SET_PTR(bp
, offset
, bufblks
);
1176 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1177 for (; j
< endcount
; j
++) {
1178 xlog_add_record(log
, offset
, cycle
, i
+j
,
1179 tail_cycle
, tail_block
);
1182 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1185 start_block
+= endcount
;
1193 * This routine is called to blow away any incomplete log writes out
1194 * in front of the log head. We do this so that we won't become confused
1195 * if we come up, write only a little bit more, and then crash again.
1196 * If we leave the partial log records out there, this situation could
1197 * cause us to think those partial writes are valid blocks since they
1198 * have the current cycle number. We get rid of them by overwriting them
1199 * with empty log records with the old cycle number rather than the
1202 * The tail lsn is passed in rather than taken from
1203 * the log so that we will not write over the unmount record after a
1204 * clean unmount in a 512 block log. Doing so would leave the log without
1205 * any valid log records in it until a new one was written. If we crashed
1206 * during that time we would not be able to recover.
1209 xlog_clear_stale_blocks(
1213 int tail_cycle
, head_cycle
;
1214 int tail_block
, head_block
;
1215 int tail_distance
, max_distance
;
1219 tail_cycle
= CYCLE_LSN(tail_lsn
);
1220 tail_block
= BLOCK_LSN(tail_lsn
);
1221 head_cycle
= log
->l_curr_cycle
;
1222 head_block
= log
->l_curr_block
;
1225 * Figure out the distance between the new head of the log
1226 * and the tail. We want to write over any blocks beyond the
1227 * head that we may have written just before the crash, but
1228 * we don't want to overwrite the tail of the log.
1230 if (head_cycle
== tail_cycle
) {
1232 * The tail is behind the head in the physical log,
1233 * so the distance from the head to the tail is the
1234 * distance from the head to the end of the log plus
1235 * the distance from the beginning of the log to the
1238 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1239 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1240 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1241 return XFS_ERROR(EFSCORRUPTED
);
1243 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1246 * The head is behind the tail in the physical log,
1247 * so the distance from the head to the tail is just
1248 * the tail block minus the head block.
1250 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1251 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1252 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1253 return XFS_ERROR(EFSCORRUPTED
);
1255 tail_distance
= tail_block
- head_block
;
1259 * If the head is right up against the tail, we can't clear
1262 if (tail_distance
<= 0) {
1263 ASSERT(tail_distance
== 0);
1267 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1269 * Take the smaller of the maximum amount of outstanding I/O
1270 * we could have and the distance to the tail to clear out.
1271 * We take the smaller so that we don't overwrite the tail and
1272 * we don't waste all day writing from the head to the tail
1275 max_distance
= MIN(max_distance
, tail_distance
);
1277 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1279 * We can stomp all the blocks we need to without
1280 * wrapping around the end of the log. Just do it
1281 * in a single write. Use the cycle number of the
1282 * current cycle minus one so that the log will look like:
1285 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1286 head_block
, max_distance
, tail_cycle
,
1292 * We need to wrap around the end of the physical log in
1293 * order to clear all the blocks. Do it in two separate
1294 * I/Os. The first write should be from the head to the
1295 * end of the physical log, and it should use the current
1296 * cycle number minus one just like above.
1298 distance
= log
->l_logBBsize
- head_block
;
1299 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1300 head_block
, distance
, tail_cycle
,
1307 * Now write the blocks at the start of the physical log.
1308 * This writes the remainder of the blocks we want to clear.
1309 * It uses the current cycle number since we're now on the
1310 * same cycle as the head so that we get:
1311 * n ... n ... | n - 1 ...
1312 * ^^^^^ blocks we're writing
1314 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1315 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1316 tail_cycle
, tail_block
);
1324 /******************************************************************************
1326 * Log recover routines
1328 ******************************************************************************
1331 STATIC xlog_recover_t
*
1332 xlog_recover_find_tid(
1336 xlog_recover_t
*p
= q
;
1339 if (p
->r_log_tid
== tid
)
1347 xlog_recover_put_hashq(
1349 xlog_recover_t
*trans
)
1356 xlog_recover_add_item(
1357 xlog_recover_item_t
**itemq
)
1359 xlog_recover_item_t
*item
;
1361 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
1362 xlog_recover_insert_item_backq(itemq
, item
);
1366 xlog_recover_add_to_cont_trans(
1367 xlog_recover_t
*trans
,
1371 xlog_recover_item_t
*item
;
1372 xfs_caddr_t ptr
, old_ptr
;
1375 item
= trans
->r_itemq
;
1377 /* finish copying rest of trans header */
1378 xlog_recover_add_item(&trans
->r_itemq
);
1379 ptr
= (xfs_caddr_t
) &trans
->r_theader
+
1380 sizeof(xfs_trans_header_t
) - len
;
1381 memcpy(ptr
, dp
, len
); /* d, s, l */
1384 item
= item
->ri_prev
;
1386 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
1387 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
1389 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, 0u);
1390 memcpy(&ptr
[old_len
], dp
, len
); /* d, s, l */
1391 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
1392 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
1397 * The next region to add is the start of a new region. It could be
1398 * a whole region or it could be the first part of a new region. Because
1399 * of this, the assumption here is that the type and size fields of all
1400 * format structures fit into the first 32 bits of the structure.
1402 * This works because all regions must be 32 bit aligned. Therefore, we
1403 * either have both fields or we have neither field. In the case we have
1404 * neither field, the data part of the region is zero length. We only have
1405 * a log_op_header and can throw away the header since a new one will appear
1406 * later. If we have at least 4 bytes, then we can determine how many regions
1407 * will appear in the current log item.
1410 xlog_recover_add_to_trans(
1411 xlog_recover_t
*trans
,
1415 xfs_inode_log_format_t
*in_f
; /* any will do */
1416 xlog_recover_item_t
*item
;
1421 item
= trans
->r_itemq
;
1423 ASSERT(*(uint
*)dp
== XFS_TRANS_HEADER_MAGIC
);
1424 if (len
== sizeof(xfs_trans_header_t
))
1425 xlog_recover_add_item(&trans
->r_itemq
);
1426 memcpy(&trans
->r_theader
, dp
, len
); /* d, s, l */
1430 ptr
= kmem_alloc(len
, KM_SLEEP
);
1431 memcpy(ptr
, dp
, len
);
1432 in_f
= (xfs_inode_log_format_t
*)ptr
;
1434 if (item
->ri_prev
->ri_total
!= 0 &&
1435 item
->ri_prev
->ri_total
== item
->ri_prev
->ri_cnt
) {
1436 xlog_recover_add_item(&trans
->r_itemq
);
1438 item
= trans
->r_itemq
;
1439 item
= item
->ri_prev
;
1441 if (item
->ri_total
== 0) { /* first region to be added */
1442 item
->ri_total
= in_f
->ilf_size
;
1443 ASSERT(item
->ri_total
<= XLOG_MAX_REGIONS_IN_ITEM
);
1444 item
->ri_buf
= kmem_zalloc((item
->ri_total
*
1445 sizeof(xfs_log_iovec_t
)), KM_SLEEP
);
1447 ASSERT(item
->ri_total
> item
->ri_cnt
);
1448 /* Description region is ri_buf[0] */
1449 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
1450 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
1456 xlog_recover_new_tid(
1461 xlog_recover_t
*trans
;
1463 trans
= kmem_zalloc(sizeof(xlog_recover_t
), KM_SLEEP
);
1464 trans
->r_log_tid
= tid
;
1466 xlog_recover_put_hashq(q
, trans
);
1470 xlog_recover_unlink_tid(
1472 xlog_recover_t
*trans
)
1483 if (tp
->r_next
== trans
) {
1491 "XFS: xlog_recover_unlink_tid: trans not found");
1493 return XFS_ERROR(EIO
);
1495 tp
->r_next
= tp
->r_next
->r_next
;
1501 xlog_recover_insert_item_backq(
1502 xlog_recover_item_t
**q
,
1503 xlog_recover_item_t
*item
)
1506 item
->ri_prev
= item
->ri_next
= item
;
1510 item
->ri_prev
= (*q
)->ri_prev
;
1511 (*q
)->ri_prev
= item
;
1512 item
->ri_prev
->ri_next
= item
;
1517 xlog_recover_insert_item_frontq(
1518 xlog_recover_item_t
**q
,
1519 xlog_recover_item_t
*item
)
1521 xlog_recover_insert_item_backq(q
, item
);
1526 xlog_recover_reorder_trans(
1528 xlog_recover_t
*trans
)
1530 xlog_recover_item_t
*first_item
, *itemq
, *itemq_next
;
1531 xfs_buf_log_format_t
*buf_f
;
1532 xfs_buf_log_format_v1_t
*obuf_f
;
1535 first_item
= itemq
= trans
->r_itemq
;
1536 trans
->r_itemq
= NULL
;
1538 itemq_next
= itemq
->ri_next
;
1539 buf_f
= (xfs_buf_log_format_t
*)itemq
->ri_buf
[0].i_addr
;
1540 switch (ITEM_TYPE(itemq
)) {
1542 flags
= buf_f
->blf_flags
;
1544 case XFS_LI_6_1_BUF
:
1545 case XFS_LI_5_3_BUF
:
1546 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1547 flags
= obuf_f
->blf_flags
;
1551 switch (ITEM_TYPE(itemq
)) {
1553 case XFS_LI_6_1_BUF
:
1554 case XFS_LI_5_3_BUF
:
1555 if (!(flags
& XFS_BLI_CANCEL
)) {
1556 xlog_recover_insert_item_frontq(&trans
->r_itemq
,
1561 case XFS_LI_6_1_INODE
:
1562 case XFS_LI_5_3_INODE
:
1564 case XFS_LI_QUOTAOFF
:
1567 xlog_recover_insert_item_backq(&trans
->r_itemq
, itemq
);
1571 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1573 return XFS_ERROR(EIO
);
1576 } while (first_item
!= itemq
);
1581 * Build up the table of buf cancel records so that we don't replay
1582 * cancelled data in the second pass. For buffer records that are
1583 * not cancel records, there is nothing to do here so we just return.
1585 * If we get a cancel record which is already in the table, this indicates
1586 * that the buffer was cancelled multiple times. In order to ensure
1587 * that during pass 2 we keep the record in the table until we reach its
1588 * last occurrence in the log, we keep a reference count in the cancel
1589 * record in the table to tell us how many times we expect to see this
1590 * record during the second pass.
1593 xlog_recover_do_buffer_pass1(
1595 xfs_buf_log_format_t
*buf_f
)
1597 xfs_buf_cancel_t
*bcp
;
1598 xfs_buf_cancel_t
*nextp
;
1599 xfs_buf_cancel_t
*prevp
;
1600 xfs_buf_cancel_t
**bucket
;
1601 xfs_buf_log_format_v1_t
*obuf_f
;
1602 xfs_daddr_t blkno
= 0;
1606 switch (buf_f
->blf_type
) {
1608 blkno
= buf_f
->blf_blkno
;
1609 len
= buf_f
->blf_len
;
1610 flags
= buf_f
->blf_flags
;
1612 case XFS_LI_6_1_BUF
:
1613 case XFS_LI_5_3_BUF
:
1614 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1615 blkno
= (xfs_daddr_t
) obuf_f
->blf_blkno
;
1616 len
= obuf_f
->blf_len
;
1617 flags
= obuf_f
->blf_flags
;
1622 * If this isn't a cancel buffer item, then just return.
1624 if (!(flags
& XFS_BLI_CANCEL
))
1628 * Insert an xfs_buf_cancel record into the hash table of
1629 * them. If there is already an identical record, bump
1630 * its reference count.
1632 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1633 XLOG_BC_TABLE_SIZE
];
1635 * If the hash bucket is empty then just insert a new record into
1638 if (*bucket
== NULL
) {
1639 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1641 bcp
->bc_blkno
= blkno
;
1643 bcp
->bc_refcount
= 1;
1644 bcp
->bc_next
= NULL
;
1650 * The hash bucket is not empty, so search for duplicates of our
1651 * record. If we find one them just bump its refcount. If not
1652 * then add us at the end of the list.
1656 while (nextp
!= NULL
) {
1657 if (nextp
->bc_blkno
== blkno
&& nextp
->bc_len
== len
) {
1658 nextp
->bc_refcount
++;
1662 nextp
= nextp
->bc_next
;
1664 ASSERT(prevp
!= NULL
);
1665 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1667 bcp
->bc_blkno
= blkno
;
1669 bcp
->bc_refcount
= 1;
1670 bcp
->bc_next
= NULL
;
1671 prevp
->bc_next
= bcp
;
1675 * Check to see whether the buffer being recovered has a corresponding
1676 * entry in the buffer cancel record table. If it does then return 1
1677 * so that it will be cancelled, otherwise return 0. If the buffer is
1678 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1679 * the refcount on the entry in the table and remove it from the table
1680 * if this is the last reference.
1682 * We remove the cancel record from the table when we encounter its
1683 * last occurrence in the log so that if the same buffer is re-used
1684 * again after its last cancellation we actually replay the changes
1685 * made at that point.
1688 xlog_check_buffer_cancelled(
1694 xfs_buf_cancel_t
*bcp
;
1695 xfs_buf_cancel_t
*prevp
;
1696 xfs_buf_cancel_t
**bucket
;
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 xlog_recover_do_buffer_pass2(
1762 xfs_buf_log_format_t
*buf_f
)
1764 xfs_buf_log_format_v1_t
*obuf_f
;
1765 xfs_daddr_t blkno
= 0;
1769 switch (buf_f
->blf_type
) {
1771 blkno
= buf_f
->blf_blkno
;
1772 flags
= buf_f
->blf_flags
;
1773 len
= buf_f
->blf_len
;
1775 case XFS_LI_6_1_BUF
:
1776 case XFS_LI_5_3_BUF
:
1777 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1778 blkno
= (xfs_daddr_t
) obuf_f
->blf_blkno
;
1779 flags
= obuf_f
->blf_flags
;
1780 len
= (xfs_daddr_t
) obuf_f
->blf_len
;
1784 return xlog_check_buffer_cancelled(log
, blkno
, len
, flags
);
1788 * Perform recovery for a buffer full of inodes. In these buffers,
1789 * the only data which should be recovered is that which corresponds
1790 * to the di_next_unlinked pointers in the on disk inode structures.
1791 * The rest of the data for the inodes is always logged through the
1792 * inodes themselves rather than the inode buffer and is recovered
1793 * in xlog_recover_do_inode_trans().
1795 * The only time when buffers full of inodes are fully recovered is
1796 * when the buffer is full of newly allocated inodes. In this case
1797 * the buffer will not be marked as an inode buffer and so will be
1798 * sent to xlog_recover_do_reg_buffer() below during recovery.
1801 xlog_recover_do_inode_buffer(
1803 xlog_recover_item_t
*item
,
1805 xfs_buf_log_format_t
*buf_f
)
1813 int next_unlinked_offset
;
1815 xfs_agino_t
*logged_nextp
;
1816 xfs_agino_t
*buffer_nextp
;
1817 xfs_buf_log_format_v1_t
*obuf_f
;
1818 unsigned int *data_map
= NULL
;
1819 unsigned int map_size
= 0;
1821 switch (buf_f
->blf_type
) {
1823 data_map
= buf_f
->blf_data_map
;
1824 map_size
= buf_f
->blf_map_size
;
1826 case XFS_LI_6_1_BUF
:
1827 case XFS_LI_5_3_BUF
:
1828 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1829 data_map
= obuf_f
->blf_data_map
;
1830 map_size
= obuf_f
->blf_map_size
;
1834 * Set the variables corresponding to the current region to
1835 * 0 so that we'll initialize them on the first pass through
1843 inodes_per_buf
= XFS_BUF_COUNT(bp
) >> mp
->m_sb
.sb_inodelog
;
1844 for (i
= 0; i
< inodes_per_buf
; i
++) {
1845 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1846 offsetof(xfs_dinode_t
, di_next_unlinked
);
1848 while (next_unlinked_offset
>=
1849 (reg_buf_offset
+ reg_buf_bytes
)) {
1851 * The next di_next_unlinked field is beyond
1852 * the current logged region. Find the next
1853 * logged region that contains or is beyond
1854 * the current di_next_unlinked field.
1857 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1860 * If there are no more logged regions in the
1861 * buffer, then we're done.
1867 nbits
= xfs_contig_bits(data_map
, map_size
,
1870 reg_buf_offset
= bit
<< XFS_BLI_SHIFT
;
1871 reg_buf_bytes
= nbits
<< XFS_BLI_SHIFT
;
1876 * If the current logged region starts after the current
1877 * di_next_unlinked field, then move on to the next
1878 * di_next_unlinked field.
1880 if (next_unlinked_offset
< reg_buf_offset
) {
1884 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1885 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLI_CHUNK
) == 0);
1886 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <= XFS_BUF_COUNT(bp
));
1889 * The current logged region contains a copy of the
1890 * current di_next_unlinked field. Extract its value
1891 * and copy it to the buffer copy.
1893 logged_nextp
= (xfs_agino_t
*)
1894 ((char *)(item
->ri_buf
[item_index
].i_addr
) +
1895 (next_unlinked_offset
- reg_buf_offset
));
1896 if (unlikely(*logged_nextp
== 0)) {
1897 xfs_fs_cmn_err(CE_ALERT
, mp
,
1898 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1900 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1901 XFS_ERRLEVEL_LOW
, mp
);
1902 return XFS_ERROR(EFSCORRUPTED
);
1905 buffer_nextp
= (xfs_agino_t
*)xfs_buf_offset(bp
,
1906 next_unlinked_offset
);
1907 INT_SET(*buffer_nextp
, ARCH_CONVERT
, *logged_nextp
);
1914 * Perform a 'normal' buffer recovery. Each logged region of the
1915 * buffer should be copied over the corresponding region in the
1916 * given buffer. The bitmap in the buf log format structure indicates
1917 * where to place the logged data.
1921 xlog_recover_do_reg_buffer(
1923 xlog_recover_item_t
*item
,
1925 xfs_buf_log_format_t
*buf_f
)
1930 xfs_buf_log_format_v1_t
*obuf_f
;
1931 unsigned int *data_map
= NULL
;
1932 unsigned int map_size
= 0;
1935 switch (buf_f
->blf_type
) {
1937 data_map
= buf_f
->blf_data_map
;
1938 map_size
= buf_f
->blf_map_size
;
1940 case XFS_LI_6_1_BUF
:
1941 case XFS_LI_5_3_BUF
:
1942 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1943 data_map
= obuf_f
->blf_data_map
;
1944 map_size
= obuf_f
->blf_map_size
;
1948 i
= 1; /* 0 is the buf format structure */
1950 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1953 nbits
= xfs_contig_bits(data_map
, map_size
, bit
);
1955 ASSERT(item
->ri_buf
[i
].i_addr
!= 0);
1956 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLI_CHUNK
== 0);
1957 ASSERT(XFS_BUF_COUNT(bp
) >=
1958 ((uint
)bit
<< XFS_BLI_SHIFT
)+(nbits
<<XFS_BLI_SHIFT
));
1961 * Do a sanity check if this is a dquot buffer. Just checking
1962 * the first dquot in the buffer should do. XXXThis is
1963 * probably a good thing to do for other buf types also.
1966 if (buf_f
->blf_flags
&
1967 (XFS_BLI_UDQUOT_BUF
|XFS_BLI_PDQUOT_BUF
|XFS_BLI_GDQUOT_BUF
)) {
1968 error
= xfs_qm_dqcheck((xfs_disk_dquot_t
*)
1969 item
->ri_buf
[i
].i_addr
,
1970 -1, 0, XFS_QMOPT_DOWARN
,
1971 "dquot_buf_recover");
1974 memcpy(xfs_buf_offset(bp
,
1975 (uint
)bit
<< XFS_BLI_SHIFT
), /* dest */
1976 item
->ri_buf
[i
].i_addr
, /* source */
1977 nbits
<<XFS_BLI_SHIFT
); /* length */
1982 /* Shouldn't be any more regions */
1983 ASSERT(i
== item
->ri_total
);
1987 * Do some primitive error checking on ondisk dquot data structures.
1991 xfs_disk_dquot_t
*ddq
,
1993 uint type
, /* used only when IO_dorepair is true */
1997 xfs_dqblk_t
*d
= (xfs_dqblk_t
*)ddq
;
2001 * We can encounter an uninitialized dquot buffer for 2 reasons:
2002 * 1. If we crash while deleting the quotainode(s), and those blks got
2003 * used for user data. This is because we take the path of regular
2004 * file deletion; however, the size field of quotainodes is never
2005 * updated, so all the tricks that we play in itruncate_finish
2006 * don't quite matter.
2008 * 2. We don't play the quota buffers when there's a quotaoff logitem.
2009 * But the allocation will be replayed so we'll end up with an
2010 * uninitialized quota block.
2012 * This is all fine; things are still consistent, and we haven't lost
2013 * any quota information. Just don't complain about bad dquot blks.
2015 if (INT_GET(ddq
->d_magic
, ARCH_CONVERT
) != XFS_DQUOT_MAGIC
) {
2016 if (flags
& XFS_QMOPT_DOWARN
)
2018 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
2020 INT_GET(ddq
->d_magic
, ARCH_CONVERT
), XFS_DQUOT_MAGIC
);
2023 if (INT_GET(ddq
->d_version
, ARCH_CONVERT
) != XFS_DQUOT_VERSION
) {
2024 if (flags
& XFS_QMOPT_DOWARN
)
2026 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2028 INT_GET(ddq
->d_magic
, ARCH_CONVERT
), XFS_DQUOT_VERSION
);
2032 if (INT_GET(ddq
->d_flags
, ARCH_CONVERT
) != XFS_DQ_USER
&&
2033 INT_GET(ddq
->d_flags
, ARCH_CONVERT
) != XFS_DQ_PROJ
&&
2034 INT_GET(ddq
->d_flags
, ARCH_CONVERT
) != XFS_DQ_GROUP
) {
2035 if (flags
& XFS_QMOPT_DOWARN
)
2037 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2038 str
, id
, INT_GET(ddq
->d_flags
, ARCH_CONVERT
));
2042 if (id
!= -1 && id
!= INT_GET(ddq
->d_id
, ARCH_CONVERT
)) {
2043 if (flags
& XFS_QMOPT_DOWARN
)
2045 "%s : ondisk-dquot 0x%p, ID mismatch: "
2046 "0x%x expected, found id 0x%x",
2047 str
, ddq
, id
, INT_GET(ddq
->d_id
, ARCH_CONVERT
));
2051 if (!errs
&& ddq
->d_id
) {
2052 if (INT_GET(ddq
->d_blk_softlimit
, ARCH_CONVERT
) &&
2053 INT_GET(ddq
->d_bcount
, ARCH_CONVERT
) >=
2054 INT_GET(ddq
->d_blk_softlimit
, ARCH_CONVERT
)) {
2055 if (!ddq
->d_btimer
) {
2056 if (flags
& XFS_QMOPT_DOWARN
)
2058 "%s : Dquot ID 0x%x (0x%p) "
2059 "BLK TIMER NOT STARTED",
2061 INT_GET(ddq
->d_id
, ARCH_CONVERT
), ddq
);
2065 if (INT_GET(ddq
->d_ino_softlimit
, ARCH_CONVERT
) &&
2066 INT_GET(ddq
->d_icount
, ARCH_CONVERT
) >=
2067 INT_GET(ddq
->d_ino_softlimit
, ARCH_CONVERT
)) {
2068 if (!ddq
->d_itimer
) {
2069 if (flags
& XFS_QMOPT_DOWARN
)
2071 "%s : Dquot ID 0x%x (0x%p) "
2072 "INODE TIMER NOT STARTED",
2074 INT_GET(ddq
->d_id
, ARCH_CONVERT
), ddq
);
2078 if (INT_GET(ddq
->d_rtb_softlimit
, ARCH_CONVERT
) &&
2079 INT_GET(ddq
->d_rtbcount
, ARCH_CONVERT
) >=
2080 INT_GET(ddq
->d_rtb_softlimit
, ARCH_CONVERT
)) {
2081 if (!ddq
->d_rtbtimer
) {
2082 if (flags
& XFS_QMOPT_DOWARN
)
2084 "%s : Dquot ID 0x%x (0x%p) "
2085 "RTBLK TIMER NOT STARTED",
2087 INT_GET(ddq
->d_id
, ARCH_CONVERT
), ddq
);
2093 if (!errs
|| !(flags
& XFS_QMOPT_DQREPAIR
))
2096 if (flags
& XFS_QMOPT_DOWARN
)
2097 cmn_err(CE_NOTE
, "Re-initializing dquot ID 0x%x", id
);
2100 * Typically, a repair is only requested by quotacheck.
2103 ASSERT(flags
& XFS_QMOPT_DQREPAIR
);
2104 memset(d
, 0, sizeof(xfs_dqblk_t
));
2105 INT_SET(d
->dd_diskdq
.d_magic
, ARCH_CONVERT
, XFS_DQUOT_MAGIC
);
2106 INT_SET(d
->dd_diskdq
.d_version
, ARCH_CONVERT
, XFS_DQUOT_VERSION
);
2107 INT_SET(d
->dd_diskdq
.d_id
, ARCH_CONVERT
, id
);
2108 INT_SET(d
->dd_diskdq
.d_flags
, ARCH_CONVERT
, type
);
2114 * Perform a dquot buffer recovery.
2115 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2116 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2117 * Else, treat it as a regular buffer and do recovery.
2120 xlog_recover_do_dquot_buffer(
2123 xlog_recover_item_t
*item
,
2125 xfs_buf_log_format_t
*buf_f
)
2130 * Filesystems are required to send in quota flags at mount time.
2132 if (mp
->m_qflags
== 0) {
2137 if (buf_f
->blf_flags
& XFS_BLI_UDQUOT_BUF
)
2138 type
|= XFS_DQ_USER
;
2139 if (buf_f
->blf_flags
& XFS_BLI_PDQUOT_BUF
)
2140 type
|= XFS_DQ_PROJ
;
2141 if (buf_f
->blf_flags
& XFS_BLI_GDQUOT_BUF
)
2142 type
|= XFS_DQ_GROUP
;
2144 * This type of quotas was turned off, so ignore this buffer
2146 if (log
->l_quotaoffs_flag
& type
)
2149 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2153 * This routine replays a modification made to a buffer at runtime.
2154 * There are actually two types of buffer, regular and inode, which
2155 * are handled differently. Inode buffers are handled differently
2156 * in that we only recover a specific set of data from them, namely
2157 * the inode di_next_unlinked fields. This is because all other inode
2158 * data is actually logged via inode records and any data we replay
2159 * here which overlaps that may be stale.
2161 * When meta-data buffers are freed at run time we log a buffer item
2162 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2163 * of the buffer in the log should not be replayed at recovery time.
2164 * This is so that if the blocks covered by the buffer are reused for
2165 * file data before we crash we don't end up replaying old, freed
2166 * meta-data into a user's file.
2168 * To handle the cancellation of buffer log items, we make two passes
2169 * over the log during recovery. During the first we build a table of
2170 * those buffers which have been cancelled, and during the second we
2171 * only replay those buffers which do not have corresponding cancel
2172 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2173 * for more details on the implementation of the table of cancel records.
2176 xlog_recover_do_buffer_trans(
2178 xlog_recover_item_t
*item
,
2181 xfs_buf_log_format_t
*buf_f
;
2182 xfs_buf_log_format_v1_t
*obuf_f
;
2191 buf_f
= (xfs_buf_log_format_t
*)item
->ri_buf
[0].i_addr
;
2193 if (pass
== XLOG_RECOVER_PASS1
) {
2195 * In this pass we're only looking for buf items
2196 * with the XFS_BLI_CANCEL bit set.
2198 xlog_recover_do_buffer_pass1(log
, buf_f
);
2202 * In this pass we want to recover all the buffers
2203 * which have not been cancelled and are not
2204 * cancellation buffers themselves. The routine
2205 * we call here will tell us whether or not to
2206 * continue with the replay of this buffer.
2208 cancel
= xlog_recover_do_buffer_pass2(log
, buf_f
);
2213 switch (buf_f
->blf_type
) {
2215 blkno
= buf_f
->blf_blkno
;
2216 len
= buf_f
->blf_len
;
2217 flags
= buf_f
->blf_flags
;
2219 case XFS_LI_6_1_BUF
:
2220 case XFS_LI_5_3_BUF
:
2221 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
2222 blkno
= obuf_f
->blf_blkno
;
2223 len
= obuf_f
->blf_len
;
2224 flags
= obuf_f
->blf_flags
;
2227 xfs_fs_cmn_err(CE_ALERT
, log
->l_mp
,
2228 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2229 buf_f
->blf_type
, log
->l_mp
->m_logname
?
2230 log
->l_mp
->m_logname
: "internal");
2231 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2232 XFS_ERRLEVEL_LOW
, log
->l_mp
);
2233 return XFS_ERROR(EFSCORRUPTED
);
2237 if (flags
& XFS_BLI_INODE_BUF
) {
2238 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, blkno
, len
,
2241 bp
= xfs_buf_read(mp
->m_ddev_targp
, blkno
, len
, 0);
2243 if (XFS_BUF_ISERROR(bp
)) {
2244 xfs_ioerror_alert("xlog_recover_do..(read#1)", log
->l_mp
,
2246 error
= XFS_BUF_GETERROR(bp
);
2252 if (flags
& XFS_BLI_INODE_BUF
) {
2253 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2255 (XFS_BLI_UDQUOT_BUF
|XFS_BLI_PDQUOT_BUF
|XFS_BLI_GDQUOT_BUF
)) {
2256 xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2258 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2261 return XFS_ERROR(error
);
2264 * Perform delayed write on the buffer. Asynchronous writes will be
2265 * slower when taking into account all the buffers to be flushed.
2267 * Also make sure that only inode buffers with good sizes stay in
2268 * the buffer cache. The kernel moves inodes in buffers of 1 block
2269 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2270 * buffers in the log can be a different size if the log was generated
2271 * by an older kernel using unclustered inode buffers or a newer kernel
2272 * running with a different inode cluster size. Regardless, if the
2273 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2274 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2275 * the buffer out of the buffer cache so that the buffer won't
2276 * overlap with future reads of those inodes.
2278 if (XFS_DINODE_MAGIC
==
2279 INT_GET(*((__uint16_t
*)(xfs_buf_offset(bp
, 0))), ARCH_CONVERT
) &&
2280 (XFS_BUF_COUNT(bp
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2281 (__uint32_t
)XFS_INODE_CLUSTER_SIZE(log
->l_mp
)))) {
2283 error
= xfs_bwrite(mp
, bp
);
2285 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
||
2286 XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*) == mp
);
2287 XFS_BUF_SET_FSPRIVATE(bp
, mp
);
2288 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2289 xfs_bdwrite(mp
, bp
);
2296 xlog_recover_do_inode_trans(
2298 xlog_recover_item_t
*item
,
2301 xfs_inode_log_format_t
*in_f
;
2313 xfs_dinode_core_t
*dicp
;
2315 if (pass
== XLOG_RECOVER_PASS1
) {
2319 in_f
= (xfs_inode_log_format_t
*)item
->ri_buf
[0].i_addr
;
2320 ino
= in_f
->ilf_ino
;
2322 if (ITEM_TYPE(item
) == XFS_LI_INODE
) {
2323 imap
.im_blkno
= (xfs_daddr_t
)in_f
->ilf_blkno
;
2324 imap
.im_len
= in_f
->ilf_len
;
2325 imap
.im_boffset
= in_f
->ilf_boffset
;
2328 * It's an old inode format record. We don't know where
2329 * its cluster is located on disk, and we can't allow
2330 * xfs_imap() to figure it out because the inode btrees
2331 * are not ready to be used. Therefore do not pass the
2332 * XFS_IMAP_LOOKUP flag to xfs_imap(). This will give
2333 * us only the single block in which the inode lives
2334 * rather than its cluster, so we must make sure to
2335 * invalidate the buffer when we write it out below.
2338 xfs_imap(log
->l_mp
, NULL
, ino
, &imap
, 0);
2342 * Inode buffers can be freed, look out for it,
2343 * and do not replay the inode.
2345 if (xlog_check_buffer_cancelled(log
, imap
.im_blkno
, imap
.im_len
, 0))
2348 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, imap
.im_blkno
, imap
.im_len
,
2350 if (XFS_BUF_ISERROR(bp
)) {
2351 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp
,
2353 error
= XFS_BUF_GETERROR(bp
);
2358 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2359 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
2362 * Make sure the place we're flushing out to really looks
2365 if (unlikely(INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) != XFS_DINODE_MAGIC
)) {
2367 xfs_fs_cmn_err(CE_ALERT
, mp
,
2368 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2370 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2371 XFS_ERRLEVEL_LOW
, mp
);
2372 return XFS_ERROR(EFSCORRUPTED
);
2374 dicp
= (xfs_dinode_core_t
*)(item
->ri_buf
[1].i_addr
);
2375 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2377 xfs_fs_cmn_err(CE_ALERT
, mp
,
2378 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2380 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2381 XFS_ERRLEVEL_LOW
, mp
);
2382 return XFS_ERROR(EFSCORRUPTED
);
2385 /* Skip replay when the on disk inode is newer than the log one */
2386 if (dicp
->di_flushiter
<
2387 INT_GET(dip
->di_core
.di_flushiter
, ARCH_CONVERT
)) {
2389 * Deal with the wrap case, DI_MAX_FLUSH is less
2390 * than smaller numbers
2392 if ((INT_GET(dip
->di_core
.di_flushiter
, ARCH_CONVERT
)
2394 (dicp
->di_flushiter
< (DI_MAX_FLUSH
>>1))) {
2401 /* Take the opportunity to reset the flush iteration count */
2402 dicp
->di_flushiter
= 0;
2404 if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFREG
)) {
2405 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2406 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2407 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2408 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2410 xfs_fs_cmn_err(CE_ALERT
, mp
,
2411 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2412 item
, dip
, bp
, ino
);
2413 return XFS_ERROR(EFSCORRUPTED
);
2415 } else if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFDIR
)) {
2416 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2417 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2418 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2419 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2420 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2422 xfs_fs_cmn_err(CE_ALERT
, mp
,
2423 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2424 item
, dip
, bp
, ino
);
2425 return XFS_ERROR(EFSCORRUPTED
);
2428 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2429 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2430 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2432 xfs_fs_cmn_err(CE_ALERT
, mp
,
2433 "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",
2435 dicp
->di_nextents
+ dicp
->di_anextents
,
2437 return XFS_ERROR(EFSCORRUPTED
);
2439 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2440 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2441 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2443 xfs_fs_cmn_err(CE_ALERT
, mp
,
2444 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2445 item
, dip
, bp
, ino
, dicp
->di_forkoff
);
2446 return XFS_ERROR(EFSCORRUPTED
);
2448 if (unlikely(item
->ri_buf
[1].i_len
> sizeof(xfs_dinode_core_t
))) {
2449 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2450 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2452 xfs_fs_cmn_err(CE_ALERT
, mp
,
2453 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2454 item
->ri_buf
[1].i_len
, item
);
2455 return XFS_ERROR(EFSCORRUPTED
);
2458 /* The core is in in-core format */
2459 xfs_xlate_dinode_core((xfs_caddr_t
)&dip
->di_core
,
2460 (xfs_dinode_core_t
*)item
->ri_buf
[1].i_addr
, -1);
2462 /* the rest is in on-disk format */
2463 if (item
->ri_buf
[1].i_len
> sizeof(xfs_dinode_core_t
)) {
2464 memcpy((xfs_caddr_t
) dip
+ sizeof(xfs_dinode_core_t
),
2465 item
->ri_buf
[1].i_addr
+ sizeof(xfs_dinode_core_t
),
2466 item
->ri_buf
[1].i_len
- sizeof(xfs_dinode_core_t
));
2469 fields
= in_f
->ilf_fields
;
2470 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
2472 INT_SET(dip
->di_u
.di_dev
, ARCH_CONVERT
, in_f
->ilf_u
.ilfu_rdev
);
2476 dip
->di_u
.di_muuid
= in_f
->ilf_u
.ilfu_uuid
;
2480 if (in_f
->ilf_size
== 2)
2481 goto write_inode_buffer
;
2482 len
= item
->ri_buf
[2].i_len
;
2483 src
= item
->ri_buf
[2].i_addr
;
2484 ASSERT(in_f
->ilf_size
<= 4);
2485 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
2486 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
2487 (len
== in_f
->ilf_dsize
));
2489 switch (fields
& XFS_ILOG_DFORK
) {
2490 case XFS_ILOG_DDATA
:
2492 memcpy(&dip
->di_u
, src
, len
);
2495 case XFS_ILOG_DBROOT
:
2496 xfs_bmbt_to_bmdr((xfs_bmbt_block_t
*)src
, len
,
2497 &(dip
->di_u
.di_bmbt
),
2498 XFS_DFORK_DSIZE(dip
, mp
));
2503 * There are no data fork flags set.
2505 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
2510 * If we logged any attribute data, recover it. There may or
2511 * may not have been any other non-core data logged in this
2514 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2515 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
2520 len
= item
->ri_buf
[attr_index
].i_len
;
2521 src
= item
->ri_buf
[attr_index
].i_addr
;
2522 ASSERT(len
== in_f
->ilf_asize
);
2524 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2525 case XFS_ILOG_ADATA
:
2527 dest
= XFS_DFORK_APTR(dip
);
2528 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
2529 memcpy(dest
, src
, len
);
2532 case XFS_ILOG_ABROOT
:
2533 dest
= XFS_DFORK_APTR(dip
);
2534 xfs_bmbt_to_bmdr((xfs_bmbt_block_t
*)src
, len
,
2535 (xfs_bmdr_block_t
*)dest
,
2536 XFS_DFORK_ASIZE(dip
, mp
));
2540 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2543 return XFS_ERROR(EIO
);
2548 if (ITEM_TYPE(item
) == XFS_LI_INODE
) {
2549 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
||
2550 XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*) == mp
);
2551 XFS_BUF_SET_FSPRIVATE(bp
, mp
);
2552 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2553 xfs_bdwrite(mp
, bp
);
2556 error
= xfs_bwrite(mp
, bp
);
2563 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2564 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2568 xlog_recover_do_quotaoff_trans(
2570 xlog_recover_item_t
*item
,
2573 xfs_qoff_logformat_t
*qoff_f
;
2575 if (pass
== XLOG_RECOVER_PASS2
) {
2579 qoff_f
= (xfs_qoff_logformat_t
*)item
->ri_buf
[0].i_addr
;
2583 * The logitem format's flag tells us if this was user quotaoff,
2584 * group quotaoff or both.
2586 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2587 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2588 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2589 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2595 * Recover a dquot record
2598 xlog_recover_do_dquot_trans(
2600 xlog_recover_item_t
*item
,
2605 struct xfs_disk_dquot
*ddq
, *recddq
;
2607 xfs_dq_logformat_t
*dq_f
;
2610 if (pass
== XLOG_RECOVER_PASS1
) {
2616 * Filesystems are required to send in quota flags at mount time.
2618 if (mp
->m_qflags
== 0)
2621 recddq
= (xfs_disk_dquot_t
*)item
->ri_buf
[1].i_addr
;
2624 * This type of quotas was turned off, so ignore this record.
2626 type
= INT_GET(recddq
->d_flags
, ARCH_CONVERT
) &
2627 (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
2629 if (log
->l_quotaoffs_flag
& type
)
2633 * At this point we know that quota was _not_ turned off.
2634 * Since the mount flags are not indicating to us otherwise, this
2635 * must mean that quota is on, and the dquot needs to be replayed.
2636 * Remember that we may not have fully recovered the superblock yet,
2637 * so we can't do the usual trick of looking at the SB quota bits.
2639 * The other possibility, of course, is that the quota subsystem was
2640 * removed since the last mount - ENOSYS.
2642 dq_f
= (xfs_dq_logformat_t
*)item
->ri_buf
[0].i_addr
;
2644 if ((error
= xfs_qm_dqcheck(recddq
,
2646 0, XFS_QMOPT_DOWARN
,
2647 "xlog_recover_do_dquot_trans (log copy)"))) {
2648 return XFS_ERROR(EIO
);
2650 ASSERT(dq_f
->qlf_len
== 1);
2652 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
2654 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
),
2657 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp
,
2658 bp
, dq_f
->qlf_blkno
);
2662 ddq
= (xfs_disk_dquot_t
*)xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
2665 * At least the magic num portion should be on disk because this
2666 * was among a chunk of dquots created earlier, and we did some
2667 * minimal initialization then.
2669 if (xfs_qm_dqcheck(ddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2670 "xlog_recover_do_dquot_trans")) {
2672 return XFS_ERROR(EIO
);
2675 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
2677 ASSERT(dq_f
->qlf_size
== 2);
2678 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
||
2679 XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*) == mp
);
2680 XFS_BUF_SET_FSPRIVATE(bp
, mp
);
2681 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2682 xfs_bdwrite(mp
, bp
);
2688 * This routine is called to create an in-core extent free intent
2689 * item from the efi format structure which was logged on disk.
2690 * It allocates an in-core efi, copies the extents from the format
2691 * structure into it, and adds the efi to the AIL with the given
2695 xlog_recover_do_efi_trans(
2697 xlog_recover_item_t
*item
,
2702 xfs_efi_log_item_t
*efip
;
2703 xfs_efi_log_format_t
*efi_formatp
;
2706 if (pass
== XLOG_RECOVER_PASS1
) {
2710 efi_formatp
= (xfs_efi_log_format_t
*)item
->ri_buf
[0].i_addr
;
2711 ASSERT(item
->ri_buf
[0].i_len
==
2712 (sizeof(xfs_efi_log_format_t
) +
2713 ((efi_formatp
->efi_nextents
- 1) * sizeof(xfs_extent_t
))));
2716 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
2717 memcpy((char *)&(efip
->efi_format
), (char *)efi_formatp
,
2718 sizeof(xfs_efi_log_format_t
) +
2719 ((efi_formatp
->efi_nextents
- 1) * sizeof(xfs_extent_t
)));
2720 efip
->efi_next_extent
= efi_formatp
->efi_nextents
;
2721 efip
->efi_flags
|= XFS_EFI_COMMITTED
;
2725 * xfs_trans_update_ail() drops the AIL lock.
2727 xfs_trans_update_ail(mp
, (xfs_log_item_t
*)efip
, lsn
, s
);
2732 * This routine is called when an efd format structure is found in
2733 * a committed transaction in the log. It's purpose is to cancel
2734 * the corresponding efi if it was still in the log. To do this
2735 * it searches the AIL for the efi with an id equal to that in the
2736 * efd format structure. If we find it, we remove the efi from the
2740 xlog_recover_do_efd_trans(
2742 xlog_recover_item_t
*item
,
2746 xfs_efd_log_format_t
*efd_formatp
;
2747 xfs_efi_log_item_t
*efip
= NULL
;
2748 xfs_log_item_t
*lip
;
2753 if (pass
== XLOG_RECOVER_PASS1
) {
2757 efd_formatp
= (xfs_efd_log_format_t
*)item
->ri_buf
[0].i_addr
;
2758 ASSERT(item
->ri_buf
[0].i_len
==
2759 (sizeof(xfs_efd_log_format_t
) +
2760 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_t
))));
2761 efi_id
= efd_formatp
->efd_efi_id
;
2764 * Search for the efi with the id in the efd format structure
2769 lip
= xfs_trans_first_ail(mp
, &gen
);
2770 while (lip
!= NULL
) {
2771 if (lip
->li_type
== XFS_LI_EFI
) {
2772 efip
= (xfs_efi_log_item_t
*)lip
;
2773 if (efip
->efi_format
.efi_id
== efi_id
) {
2775 * xfs_trans_delete_ail() drops the
2778 xfs_trans_delete_ail(mp
, lip
, s
);
2782 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
2786 * If we found it, then free it up. If it wasn't there, it
2787 * must have been overwritten in the log. Oh well.
2790 xfs_efi_item_free(efip
);
2797 * Perform the transaction
2799 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2800 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2803 xlog_recover_do_trans(
2805 xlog_recover_t
*trans
,
2809 xlog_recover_item_t
*item
, *first_item
;
2811 if ((error
= xlog_recover_reorder_trans(log
, trans
)))
2813 first_item
= item
= trans
->r_itemq
;
2816 * we don't need to worry about the block number being
2817 * truncated in > 1 TB buffers because in user-land,
2818 * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so
2819 * the blkno's will get through the user-mode buffer
2820 * cache properly. The only bad case is o32 kernels
2821 * where xfs_daddr_t is 32-bits but mount will warn us
2822 * off a > 1 TB filesystem before we get here.
2824 if ((ITEM_TYPE(item
) == XFS_LI_BUF
) ||
2825 (ITEM_TYPE(item
) == XFS_LI_6_1_BUF
) ||
2826 (ITEM_TYPE(item
) == XFS_LI_5_3_BUF
)) {
2827 if ((error
= xlog_recover_do_buffer_trans(log
, item
,
2830 } else if ((ITEM_TYPE(item
) == XFS_LI_INODE
) ||
2831 (ITEM_TYPE(item
) == XFS_LI_6_1_INODE
) ||
2832 (ITEM_TYPE(item
) == XFS_LI_5_3_INODE
)) {
2833 if ((error
= xlog_recover_do_inode_trans(log
, item
,
2836 } else if (ITEM_TYPE(item
) == XFS_LI_EFI
) {
2837 xlog_recover_do_efi_trans(log
, item
, trans
->r_lsn
,
2839 } else if (ITEM_TYPE(item
) == XFS_LI_EFD
) {
2840 xlog_recover_do_efd_trans(log
, item
, pass
);
2841 } else if (ITEM_TYPE(item
) == XFS_LI_DQUOT
) {
2842 if ((error
= xlog_recover_do_dquot_trans(log
, item
,
2845 } else if ((ITEM_TYPE(item
) == XFS_LI_QUOTAOFF
)) {
2846 if ((error
= xlog_recover_do_quotaoff_trans(log
, item
,
2850 xlog_warn("XFS: xlog_recover_do_trans");
2852 error
= XFS_ERROR(EIO
);
2855 item
= item
->ri_next
;
2856 } while (first_item
!= item
);
2862 * Free up any resources allocated by the transaction
2864 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2867 xlog_recover_free_trans(
2868 xlog_recover_t
*trans
)
2870 xlog_recover_item_t
*first_item
, *item
, *free_item
;
2873 item
= first_item
= trans
->r_itemq
;
2876 item
= item
->ri_next
;
2877 /* Free the regions in the item. */
2878 for (i
= 0; i
< free_item
->ri_cnt
; i
++) {
2879 kmem_free(free_item
->ri_buf
[i
].i_addr
,
2880 free_item
->ri_buf
[i
].i_len
);
2882 /* Free the item itself */
2883 kmem_free(free_item
->ri_buf
,
2884 (free_item
->ri_total
* sizeof(xfs_log_iovec_t
)));
2885 kmem_free(free_item
, sizeof(xlog_recover_item_t
));
2886 } while (first_item
!= item
);
2887 /* Free the transaction recover structure */
2888 kmem_free(trans
, sizeof(xlog_recover_t
));
2892 xlog_recover_commit_trans(
2895 xlog_recover_t
*trans
,
2900 if ((error
= xlog_recover_unlink_tid(q
, trans
)))
2902 if ((error
= xlog_recover_do_trans(log
, trans
, pass
)))
2904 xlog_recover_free_trans(trans
); /* no error */
2909 xlog_recover_unmount_trans(
2910 xlog_recover_t
*trans
)
2912 /* Do nothing now */
2913 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2918 * There are two valid states of the r_state field. 0 indicates that the
2919 * transaction structure is in a normal state. We have either seen the
2920 * start of the transaction or the last operation we added was not a partial
2921 * operation. If the last operation we added to the transaction was a
2922 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2924 * NOTE: skip LRs with 0 data length.
2927 xlog_recover_process_data(
2929 xlog_recover_t
*rhash
[],
2930 xlog_rec_header_t
*rhead
,
2936 xlog_op_header_t
*ohead
;
2937 xlog_recover_t
*trans
;
2943 lp
= dp
+ INT_GET(rhead
->h_len
, ARCH_CONVERT
);
2944 num_logops
= INT_GET(rhead
->h_num_logops
, ARCH_CONVERT
);
2946 /* check the log format matches our own - else we can't recover */
2947 if (xlog_header_check_recover(log
->l_mp
, rhead
))
2948 return (XFS_ERROR(EIO
));
2950 while ((dp
< lp
) && num_logops
) {
2951 ASSERT(dp
+ sizeof(xlog_op_header_t
) <= lp
);
2952 ohead
= (xlog_op_header_t
*)dp
;
2953 dp
+= sizeof(xlog_op_header_t
);
2954 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
2955 ohead
->oh_clientid
!= XFS_LOG
) {
2957 "XFS: xlog_recover_process_data: bad clientid");
2959 return (XFS_ERROR(EIO
));
2961 tid
= INT_GET(ohead
->oh_tid
, ARCH_CONVERT
);
2962 hash
= XLOG_RHASH(tid
);
2963 trans
= xlog_recover_find_tid(rhash
[hash
], tid
);
2964 if (trans
== NULL
) { /* not found; add new tid */
2965 if (ohead
->oh_flags
& XLOG_START_TRANS
)
2966 xlog_recover_new_tid(&rhash
[hash
], tid
,
2967 INT_GET(rhead
->h_lsn
, ARCH_CONVERT
));
2969 ASSERT(dp
+INT_GET(ohead
->oh_len
, ARCH_CONVERT
) <= lp
);
2970 flags
= ohead
->oh_flags
& ~XLOG_END_TRANS
;
2971 if (flags
& XLOG_WAS_CONT_TRANS
)
2972 flags
&= ~XLOG_CONTINUE_TRANS
;
2974 case XLOG_COMMIT_TRANS
:
2975 error
= xlog_recover_commit_trans(log
,
2976 &rhash
[hash
], trans
, pass
);
2978 case XLOG_UNMOUNT_TRANS
:
2979 error
= xlog_recover_unmount_trans(trans
);
2981 case XLOG_WAS_CONT_TRANS
:
2982 error
= xlog_recover_add_to_cont_trans(trans
,
2983 dp
, INT_GET(ohead
->oh_len
,
2986 case XLOG_START_TRANS
:
2988 "XFS: xlog_recover_process_data: bad transaction");
2990 error
= XFS_ERROR(EIO
);
2993 case XLOG_CONTINUE_TRANS
:
2994 error
= xlog_recover_add_to_trans(trans
,
2995 dp
, INT_GET(ohead
->oh_len
,
3000 "XFS: xlog_recover_process_data: bad flag");
3002 error
= XFS_ERROR(EIO
);
3008 dp
+= INT_GET(ohead
->oh_len
, ARCH_CONVERT
);
3015 * Process an extent free intent item that was recovered from
3016 * the log. We need to free the extents that it describes.
3019 xlog_recover_process_efi(
3021 xfs_efi_log_item_t
*efip
)
3023 xfs_efd_log_item_t
*efdp
;
3027 xfs_fsblock_t startblock_fsb
;
3029 ASSERT(!(efip
->efi_flags
& XFS_EFI_RECOVERED
));
3032 * First check the validity of the extents described by the
3033 * EFI. If any are bad, then assume that all are bad and
3034 * just toss the EFI.
3036 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3037 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3038 startblock_fsb
= XFS_BB_TO_FSB(mp
,
3039 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
3040 if ((startblock_fsb
== 0) ||
3041 (extp
->ext_len
== 0) ||
3042 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
3043 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
3045 * This will pull the EFI from the AIL and
3046 * free the memory associated with it.
3048 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
3053 tp
= xfs_trans_alloc(mp
, 0);
3054 xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0, 0, 0);
3055 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
3057 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3058 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3059 xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
3060 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
3064 efip
->efi_flags
|= XFS_EFI_RECOVERED
;
3065 xfs_trans_commit(tp
, 0, NULL
);
3069 * Verify that once we've encountered something other than an EFI
3070 * in the AIL that there are no more EFIs in the AIL.
3074 xlog_recover_check_ail(
3076 xfs_log_item_t
*lip
,
3082 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
3083 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
3085 * The check will be bogus if we restart from the
3086 * beginning of the AIL, so ASSERT that we don't.
3087 * We never should since we're holding the AIL lock
3090 ASSERT(gen
== orig_gen
);
3091 } while (lip
!= NULL
);
3096 * When this is called, all of the EFIs which did not have
3097 * corresponding EFDs should be in the AIL. What we do now
3098 * is free the extents associated with each one.
3100 * Since we process the EFIs in normal transactions, they
3101 * will be removed at some point after the commit. This prevents
3102 * us from just walking down the list processing each one.
3103 * We'll use a flag in the EFI to skip those that we've already
3104 * processed and use the AIL iteration mechanism's generation
3105 * count to try to speed this up at least a bit.
3107 * When we start, we know that the EFIs are the only things in
3108 * the AIL. As we process them, however, other items are added
3109 * to the AIL. Since everything added to the AIL must come after
3110 * everything already in the AIL, we stop processing as soon as
3111 * we see something other than an EFI in the AIL.
3114 xlog_recover_process_efis(
3117 xfs_log_item_t
*lip
;
3118 xfs_efi_log_item_t
*efip
;
3126 lip
= xfs_trans_first_ail(mp
, &gen
);
3127 while (lip
!= NULL
) {
3129 * We're done when we see something other than an EFI.
3131 if (lip
->li_type
!= XFS_LI_EFI
) {
3132 xlog_recover_check_ail(mp
, lip
, gen
);
3137 * Skip EFIs that we've already processed.
3139 efip
= (xfs_efi_log_item_t
*)lip
;
3140 if (efip
->efi_flags
& XFS_EFI_RECOVERED
) {
3141 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
3146 xlog_recover_process_efi(mp
, efip
);
3148 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
3154 * This routine performs a transaction to null out a bad inode pointer
3155 * in an agi unlinked inode hash bucket.
3158 xlog_recover_clear_agi_bucket(
3160 xfs_agnumber_t agno
,
3169 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3170 xfs_trans_reserve(tp
, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp
), 0, 0, 0);
3172 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
3173 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
3174 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
3176 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3180 agi
= XFS_BUF_TO_AGI(agibp
);
3181 if (INT_GET(agi
->agi_magicnum
, ARCH_CONVERT
) != XFS_AGI_MAGIC
) {
3182 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3185 ASSERT(INT_GET(agi
->agi_magicnum
, ARCH_CONVERT
) == XFS_AGI_MAGIC
);
3187 INT_SET(agi
->agi_unlinked
[bucket
], ARCH_CONVERT
, NULLAGINO
);
3188 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3189 (sizeof(xfs_agino_t
) * bucket
);
3190 xfs_trans_log_buf(tp
, agibp
, offset
,
3191 (offset
+ sizeof(xfs_agino_t
) - 1));
3193 (void) xfs_trans_commit(tp
, 0, NULL
);
3197 * xlog_iunlink_recover
3199 * This is called during recovery to process any inodes which
3200 * we unlinked but not freed when the system crashed. These
3201 * inodes will be on the lists in the AGI blocks. What we do
3202 * here is scan all the AGIs and fully truncate and free any
3203 * inodes found on the lists. Each inode is removed from the
3204 * lists when it has been fully truncated and is freed. The
3205 * freeing of the inode and its removal from the list must be
3209 xlog_recover_process_iunlinks(
3213 xfs_agnumber_t agno
;
3228 * Prevent any DMAPI event from being sent while in this function.
3230 mp_dmevmask
= mp
->m_dmevmask
;
3233 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3235 * Find the agi for this ag.
3237 agibp
= xfs_buf_read(mp
->m_ddev_targp
,
3238 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
3239 XFS_FSS_TO_BB(mp
, 1), 0);
3240 if (XFS_BUF_ISERROR(agibp
)) {
3241 xfs_ioerror_alert("xlog_recover_process_iunlinks(#1)",
3243 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)));
3245 agi
= XFS_BUF_TO_AGI(agibp
);
3246 ASSERT(XFS_AGI_MAGIC
==
3247 INT_GET(agi
->agi_magicnum
, ARCH_CONVERT
));
3249 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3251 agino
= INT_GET(agi
->agi_unlinked
[bucket
], ARCH_CONVERT
);
3252 while (agino
!= NULLAGINO
) {
3255 * Release the agi buffer so that it can
3256 * be acquired in the normal course of the
3257 * transaction to truncate and free the inode.
3259 xfs_buf_relse(agibp
);
3261 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3262 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
, 0);
3263 ASSERT(error
|| (ip
!= NULL
));
3267 * Get the on disk inode to find the
3268 * next inode in the bucket.
3270 error
= xfs_itobp(mp
, NULL
, ip
, &dip
,
3272 ASSERT(error
|| (dip
!= NULL
));
3276 ASSERT(ip
->i_d
.di_nlink
== 0);
3278 /* setup for the next pass */
3279 agino
= INT_GET(dip
->di_next_unlinked
,
3283 * Prevent any DMAPI event from
3284 * being sent when the
3285 * reference on the inode is
3288 ip
->i_d
.di_dmevmask
= 0;
3291 * If this is a new inode, handle
3292 * it specially. Otherwise,
3293 * just drop our reference to the
3294 * inode. If there are no
3295 * other references, this will
3297 * xfs_inactive() which will
3298 * truncate the file and free
3301 if (ip
->i_d
.di_mode
== 0)
3302 xfs_iput_new(ip
, 0);
3304 VN_RELE(XFS_ITOV(ip
));
3307 * We can't read in the inode
3308 * this bucket points to, or
3309 * this inode is messed up. Just
3310 * ditch this bucket of inodes. We
3311 * will lose some inodes and space,
3312 * but at least we won't hang. Call
3313 * xlog_recover_clear_agi_bucket()
3314 * to perform a transaction to clear
3315 * the inode pointer in the bucket.
3317 xlog_recover_clear_agi_bucket(mp
, agno
,
3324 * Reacquire the agibuffer and continue around
3327 agibp
= xfs_buf_read(mp
->m_ddev_targp
,
3328 XFS_AG_DADDR(mp
, agno
,
3330 XFS_FSS_TO_BB(mp
, 1), 0);
3331 if (XFS_BUF_ISERROR(agibp
)) {
3333 "xlog_recover_process_iunlinks(#2)",
3335 XFS_AG_DADDR(mp
, agno
,
3336 XFS_AGI_DADDR(mp
)));
3338 agi
= XFS_BUF_TO_AGI(agibp
);
3339 ASSERT(XFS_AGI_MAGIC
== INT_GET(
3340 agi
->agi_magicnum
, ARCH_CONVERT
));
3345 * Release the buffer for the current agi so we can
3346 * go on to the next one.
3348 xfs_buf_relse(agibp
);
3351 mp
->m_dmevmask
= mp_dmevmask
;
3357 xlog_pack_data_checksum(
3359 xlog_in_core_t
*iclog
,
3366 up
= (uint
*)iclog
->ic_datap
;
3367 /* divide length by 4 to get # words */
3368 for (i
= 0; i
< (size
>> 2); i
++) {
3369 chksum
^= INT_GET(*up
, ARCH_CONVERT
);
3372 INT_SET(iclog
->ic_header
.h_chksum
, ARCH_CONVERT
, chksum
);
3375 #define xlog_pack_data_checksum(log, iclog, size)
3379 * Stamp cycle number in every block
3384 xlog_in_core_t
*iclog
,
3388 int size
= iclog
->ic_offset
+ roundoff
;
3391 xlog_in_core_2_t
*xhdr
;
3393 xlog_pack_data_checksum(log
, iclog
, size
);
3395 cycle_lsn
= CYCLE_LSN_DISK(iclog
->ic_header
.h_lsn
);
3397 dp
= iclog
->ic_datap
;
3398 for (i
= 0; i
< BTOBB(size
) &&
3399 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3400 iclog
->ic_header
.h_cycle_data
[i
] = *(uint
*)dp
;
3401 *(uint
*)dp
= cycle_lsn
;
3405 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3406 xhdr
= (xlog_in_core_2_t
*)&iclog
->ic_header
;
3407 for ( ; i
< BTOBB(size
); i
++) {
3408 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3409 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3410 xhdr
[j
].hic_xheader
.xh_cycle_data
[k
] = *(uint
*)dp
;
3411 *(uint
*)dp
= cycle_lsn
;
3415 for (i
= 1; i
< log
->l_iclog_heads
; i
++) {
3416 xhdr
[i
].hic_xheader
.xh_cycle
= cycle_lsn
;
3421 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3423 xlog_unpack_data_checksum(
3424 xlog_rec_header_t
*rhead
,
3428 uint
*up
= (uint
*)dp
;
3432 /* divide length by 4 to get # words */
3433 for (i
=0; i
< INT_GET(rhead
->h_len
, ARCH_CONVERT
) >> 2; i
++) {
3434 chksum
^= INT_GET(*up
, ARCH_CONVERT
);
3437 if (chksum
!= INT_GET(rhead
->h_chksum
, ARCH_CONVERT
)) {
3438 if (rhead
->h_chksum
||
3439 ((log
->l_flags
& XLOG_CHKSUM_MISMATCH
) == 0)) {
3441 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)",
3442 INT_GET(rhead
->h_chksum
, ARCH_CONVERT
), chksum
);
3444 "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3445 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3447 "XFS: LogR this is a LogV2 filesystem");
3449 log
->l_flags
|= XLOG_CHKSUM_MISMATCH
;
3454 #define xlog_unpack_data_checksum(rhead, dp, log)
3459 xlog_rec_header_t
*rhead
,
3464 xlog_in_core_2_t
*xhdr
;
3466 for (i
= 0; i
< BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
)) &&
3467 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3468 *(uint
*)dp
= *(uint
*)&rhead
->h_cycle_data
[i
];
3472 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3473 xhdr
= (xlog_in_core_2_t
*)rhead
;
3474 for ( ; i
< BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
)); i
++) {
3475 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3476 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3477 *(uint
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
3482 xlog_unpack_data_checksum(rhead
, dp
, log
);
3486 xlog_valid_rec_header(
3488 xlog_rec_header_t
*rhead
,
3494 (INT_GET(rhead
->h_magicno
, ARCH_CONVERT
) !=
3495 XLOG_HEADER_MAGIC_NUM
))) {
3496 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3497 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3498 return XFS_ERROR(EFSCORRUPTED
);
3501 (!rhead
->h_version
||
3502 (INT_GET(rhead
->h_version
, ARCH_CONVERT
) &
3503 (~XLOG_VERSION_OKBITS
)) != 0))) {
3504 xlog_warn("XFS: %s: unrecognised log version (%d).",
3505 __FUNCTION__
, INT_GET(rhead
->h_version
, ARCH_CONVERT
));
3506 return XFS_ERROR(EIO
);
3509 /* LR body must have data or it wouldn't have been written */
3510 hlen
= INT_GET(rhead
->h_len
, ARCH_CONVERT
);
3511 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
3512 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3513 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3514 return XFS_ERROR(EFSCORRUPTED
);
3516 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
3517 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3518 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3519 return XFS_ERROR(EFSCORRUPTED
);
3525 * Read the log from tail to head and process the log records found.
3526 * Handle the two cases where the tail and head are in the same cycle
3527 * and where the active portion of the log wraps around the end of
3528 * the physical log separately. The pass parameter is passed through
3529 * to the routines called to process the data and is not looked at
3533 xlog_do_recovery_pass(
3535 xfs_daddr_t head_blk
,
3536 xfs_daddr_t tail_blk
,
3539 xlog_rec_header_t
*rhead
;
3541 xfs_caddr_t bufaddr
, offset
;
3542 xfs_buf_t
*hbp
, *dbp
;
3543 int error
= 0, h_size
;
3544 int bblks
, split_bblks
;
3545 int hblks
, split_hblks
, wrapped_hblks
;
3546 xlog_recover_t
*rhash
[XLOG_RHASH_SIZE
];
3548 ASSERT(head_blk
!= tail_blk
);
3551 * Read the header of the tail block and get the iclog buffer size from
3552 * h_size. Use this to tell how many sectors make up the log header.
3554 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3556 * When using variable length iclogs, read first sector of
3557 * iclog header and extract the header size from it. Get a
3558 * new hbp that is the correct size.
3560 hbp
= xlog_get_bp(log
, 1);
3563 if ((error
= xlog_bread(log
, tail_blk
, 1, hbp
)))
3565 offset
= xlog_align(log
, tail_blk
, 1, hbp
);
3566 rhead
= (xlog_rec_header_t
*)offset
;
3567 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
3570 h_size
= INT_GET(rhead
->h_size
, ARCH_CONVERT
);
3571 if ((INT_GET(rhead
->h_version
, ARCH_CONVERT
)
3572 & XLOG_VERSION_2
) &&
3573 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
3574 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
3575 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
3578 hbp
= xlog_get_bp(log
, hblks
);
3583 ASSERT(log
->l_sectbb_log
== 0);
3585 hbp
= xlog_get_bp(log
, 1);
3586 h_size
= XLOG_BIG_RECORD_BSIZE
;
3591 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
3597 memset(rhash
, 0, sizeof(rhash
));
3598 if (tail_blk
<= head_blk
) {
3599 for (blk_no
= tail_blk
; blk_no
< head_blk
; ) {
3600 if ((error
= xlog_bread(log
, blk_no
, hblks
, hbp
)))
3602 offset
= xlog_align(log
, blk_no
, hblks
, hbp
);
3603 rhead
= (xlog_rec_header_t
*)offset
;
3604 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3608 /* blocks in data section */
3609 bblks
= (int)BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
));
3610 error
= xlog_bread(log
, blk_no
+ hblks
, bblks
, dbp
);
3613 offset
= xlog_align(log
, blk_no
+ hblks
, bblks
, dbp
);
3614 xlog_unpack_data(rhead
, offset
, log
);
3615 if ((error
= xlog_recover_process_data(log
,
3616 rhash
, rhead
, offset
, pass
)))
3618 blk_no
+= bblks
+ hblks
;
3622 * Perform recovery around the end of the physical log.
3623 * When the head is not on the same cycle number as the tail,
3624 * we can't do a sequential recovery as above.
3627 while (blk_no
< log
->l_logBBsize
) {
3629 * Check for header wrapping around physical end-of-log
3634 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
3635 /* Read header in one read */
3636 error
= xlog_bread(log
, blk_no
, hblks
, hbp
);
3639 offset
= xlog_align(log
, blk_no
, hblks
, hbp
);
3641 /* This LR is split across physical log end */
3642 if (blk_no
!= log
->l_logBBsize
) {
3643 /* some data before physical log end */
3644 ASSERT(blk_no
<= INT_MAX
);
3645 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
3646 ASSERT(split_hblks
> 0);
3647 if ((error
= xlog_bread(log
, blk_no
,
3650 offset
= xlog_align(log
, blk_no
,
3654 * Note: this black magic still works with
3655 * large sector sizes (non-512) only because:
3656 * - we increased the buffer size originally
3657 * by 1 sector giving us enough extra space
3658 * for the second read;
3659 * - the log start is guaranteed to be sector
3661 * - we read the log end (LR header start)
3662 * _first_, then the log start (LR header end)
3663 * - order is important.
3665 bufaddr
= XFS_BUF_PTR(hbp
);
3666 XFS_BUF_SET_PTR(hbp
,
3667 bufaddr
+ BBTOB(split_hblks
),
3668 BBTOB(hblks
- split_hblks
));
3669 wrapped_hblks
= hblks
- split_hblks
;
3670 error
= xlog_bread(log
, 0, wrapped_hblks
, hbp
);
3673 XFS_BUF_SET_PTR(hbp
, bufaddr
, BBTOB(hblks
));
3675 offset
= xlog_align(log
, 0,
3676 wrapped_hblks
, hbp
);
3678 rhead
= (xlog_rec_header_t
*)offset
;
3679 error
= xlog_valid_rec_header(log
, rhead
,
3680 split_hblks
? blk_no
: 0);
3684 bblks
= (int)BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
));
3687 /* Read in data for log record */
3688 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
3689 error
= xlog_bread(log
, blk_no
, bblks
, dbp
);
3692 offset
= xlog_align(log
, blk_no
, bblks
, dbp
);
3694 /* This log record is split across the
3695 * physical end of log */
3698 if (blk_no
!= log
->l_logBBsize
) {
3699 /* some data is before the physical
3701 ASSERT(!wrapped_hblks
);
3702 ASSERT(blk_no
<= INT_MAX
);
3704 log
->l_logBBsize
- (int)blk_no
;
3705 ASSERT(split_bblks
> 0);
3706 if ((error
= xlog_bread(log
, blk_no
,
3709 offset
= xlog_align(log
, blk_no
,
3713 * Note: this black magic still works with
3714 * large sector sizes (non-512) only because:
3715 * - we increased the buffer size originally
3716 * by 1 sector giving us enough extra space
3717 * for the second read;
3718 * - the log start is guaranteed to be sector
3720 * - we read the log end (LR header start)
3721 * _first_, then the log start (LR header end)
3722 * - order is important.
3724 bufaddr
= XFS_BUF_PTR(dbp
);
3725 XFS_BUF_SET_PTR(dbp
,
3726 bufaddr
+ BBTOB(split_bblks
),
3727 BBTOB(bblks
- split_bblks
));
3728 if ((error
= xlog_bread(log
, wrapped_hblks
,
3729 bblks
- split_bblks
, dbp
)))
3731 XFS_BUF_SET_PTR(dbp
, bufaddr
, h_size
);
3733 offset
= xlog_align(log
, wrapped_hblks
,
3734 bblks
- split_bblks
, dbp
);
3736 xlog_unpack_data(rhead
, offset
, log
);
3737 if ((error
= xlog_recover_process_data(log
, rhash
,
3738 rhead
, offset
, pass
)))
3743 ASSERT(blk_no
>= log
->l_logBBsize
);
3744 blk_no
-= log
->l_logBBsize
;
3746 /* read first part of physical log */
3747 while (blk_no
< head_blk
) {
3748 if ((error
= xlog_bread(log
, blk_no
, hblks
, hbp
)))
3750 offset
= xlog_align(log
, blk_no
, hblks
, hbp
);
3751 rhead
= (xlog_rec_header_t
*)offset
;
3752 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3755 bblks
= (int)BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
));
3756 if ((error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
)))
3758 offset
= xlog_align(log
, blk_no
+hblks
, bblks
, dbp
);
3759 xlog_unpack_data(rhead
, offset
, log
);
3760 if ((error
= xlog_recover_process_data(log
, rhash
,
3761 rhead
, offset
, pass
)))
3763 blk_no
+= bblks
+ hblks
;
3775 * Do the recovery of the log. We actually do this in two phases.
3776 * The two passes are necessary in order to implement the function
3777 * of cancelling a record written into the log. The first pass
3778 * determines those things which have been cancelled, and the
3779 * second pass replays log items normally except for those which
3780 * have been cancelled. The handling of the replay and cancellations
3781 * takes place in the log item type specific routines.
3783 * The table of items which have cancel records in the log is allocated
3784 * and freed at this level, since only here do we know when all of
3785 * the log recovery has been completed.
3788 xlog_do_log_recovery(
3790 xfs_daddr_t head_blk
,
3791 xfs_daddr_t tail_blk
)
3795 ASSERT(head_blk
!= tail_blk
);
3798 * First do a pass to find all of the cancelled buf log items.
3799 * Store them in the buf_cancel_table for use in the second pass.
3801 log
->l_buf_cancel_table
=
3802 (xfs_buf_cancel_t
**)kmem_zalloc(XLOG_BC_TABLE_SIZE
*
3803 sizeof(xfs_buf_cancel_t
*),
3805 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3806 XLOG_RECOVER_PASS1
);
3808 kmem_free(log
->l_buf_cancel_table
,
3809 XLOG_BC_TABLE_SIZE
* sizeof(xfs_buf_cancel_t
*));
3810 log
->l_buf_cancel_table
= NULL
;
3814 * Then do a second pass to actually recover the items in the log.
3815 * When it is complete free the table of buf cancel items.
3817 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3818 XLOG_RECOVER_PASS2
);
3823 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
3824 ASSERT(log
->l_buf_cancel_table
[i
] == NULL
);
3828 kmem_free(log
->l_buf_cancel_table
,
3829 XLOG_BC_TABLE_SIZE
* sizeof(xfs_buf_cancel_t
*));
3830 log
->l_buf_cancel_table
= NULL
;
3836 * Do the actual recovery
3841 xfs_daddr_t head_blk
,
3842 xfs_daddr_t tail_blk
)
3849 * First replay the images in the log.
3851 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
3856 XFS_bflush(log
->l_mp
->m_ddev_targp
);
3859 * If IO errors happened during recovery, bail out.
3861 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
3866 * We now update the tail_lsn since much of the recovery has completed
3867 * and there may be space available to use. If there were no extent
3868 * or iunlinks, we can free up the entire log and set the tail_lsn to
3869 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3870 * lsn of the last known good LR on disk. If there are extent frees
3871 * or iunlinks they will have some entries in the AIL; so we look at
3872 * the AIL to determine how to set the tail_lsn.
3874 xlog_assign_tail_lsn(log
->l_mp
);
3877 * Now that we've finished replaying all buffer and inode
3878 * updates, re-read in the superblock.
3880 bp
= xfs_getsb(log
->l_mp
, 0);
3883 xfsbdstrat(log
->l_mp
, bp
);
3884 if ((error
= xfs_iowait(bp
))) {
3885 xfs_ioerror_alert("xlog_do_recover",
3886 log
->l_mp
, bp
, XFS_BUF_ADDR(bp
));
3892 /* Convert superblock from on-disk format */
3893 sbp
= &log
->l_mp
->m_sb
;
3894 xfs_xlatesb(XFS_BUF_TO_SBP(bp
), sbp
, 1, XFS_SB_ALL_BITS
);
3895 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
3896 ASSERT(XFS_SB_GOOD_VERSION(sbp
));
3899 xlog_recover_check_summary(log
);
3901 /* Normal transactions can now occur */
3902 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
3907 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3909 * Return error or zero.
3916 xfs_daddr_t head_blk
, tail_blk
;
3919 /* find the tail of the log */
3920 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
, readonly
)))
3923 if (tail_blk
!= head_blk
) {
3924 /* There used to be a comment here:
3926 * disallow recovery on read-only mounts. note -- mount
3927 * checks for ENOSPC and turns it into an intelligent
3929 * ...but this is no longer true. Now, unless you specify
3930 * NORECOVERY (in which case this function would never be
3931 * called), we just go ahead and recover. We do this all
3932 * under the vfs layer, so we can get away with it unless
3933 * the device itself is read-only, in which case we fail.
3935 if ((error
= xfs_dev_is_read_only(log
->l_mp
,
3936 "recovery required"))) {
3941 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3942 log
->l_mp
->m_fsname
, log
->l_mp
->m_logname
?
3943 log
->l_mp
->m_logname
: "internal");
3945 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
3946 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
3952 * In the first part of recovery we replay inodes and buffers and build
3953 * up the list of extent free items which need to be processed. Here
3954 * we process the extent free items and clean up the on disk unlinked
3955 * inode lists. This is separated from the first part of recovery so
3956 * that the root and real-time bitmap inodes can be read in from disk in
3957 * between the two stages. This is necessary so that we can free space
3958 * in the real-time portion of the file system.
3961 xlog_recover_finish(
3966 * Now we're ready to do the transactions needed for the
3967 * rest of recovery. Start with completing all the extent
3968 * free intent records and then process the unlinked inode
3969 * lists. At this point, we essentially run in normal mode
3970 * except that we're still performing recovery actions
3971 * rather than accepting new requests.
3973 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
3974 xlog_recover_process_efis(log
);
3976 * Sync the log to get all the EFIs out of the AIL.
3977 * This isn't absolutely necessary, but it helps in
3978 * case the unlink transactions would have problems
3979 * pushing the EFIs out of the way.
3981 xfs_log_force(log
->l_mp
, (xfs_lsn_t
)0,
3982 (XFS_LOG_FORCE
| XFS_LOG_SYNC
));
3984 if ( (mfsi_flags
& XFS_MFSI_NOUNLINK
) == 0 ) {
3985 xlog_recover_process_iunlinks(log
);
3988 xlog_recover_check_summary(log
);
3991 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3992 log
->l_mp
->m_fsname
, log
->l_mp
->m_logname
?
3993 log
->l_mp
->m_logname
: "internal");
3994 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
3997 "!Ending clean XFS mount for filesystem: %s",
3998 log
->l_mp
->m_fsname
);
4006 * Read all of the agf and agi counters and check that they
4007 * are consistent with the superblock counters.
4010 xlog_recover_check_summary(
4018 xfs_daddr_t agfdaddr
;
4019 xfs_daddr_t agidaddr
;
4021 #ifdef XFS_LOUD_RECOVERY
4024 xfs_agnumber_t agno
;
4025 __uint64_t freeblks
;
4034 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
4035 agfdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGF_DADDR(mp
));
4036 agfbp
= xfs_buf_read(mp
->m_ddev_targp
, agfdaddr
,
4037 XFS_FSS_TO_BB(mp
, 1), 0);
4038 if (XFS_BUF_ISERROR(agfbp
)) {
4039 xfs_ioerror_alert("xlog_recover_check_summary(agf)",
4040 mp
, agfbp
, agfdaddr
);
4042 agfp
= XFS_BUF_TO_AGF(agfbp
);
4043 ASSERT(XFS_AGF_MAGIC
==
4044 INT_GET(agfp
->agf_magicnum
, ARCH_CONVERT
));
4045 ASSERT(XFS_AGF_GOOD_VERSION(
4046 INT_GET(agfp
->agf_versionnum
, ARCH_CONVERT
)));
4047 ASSERT(INT_GET(agfp
->agf_seqno
, ARCH_CONVERT
) == agno
);
4049 freeblks
+= INT_GET(agfp
->agf_freeblks
, ARCH_CONVERT
) +
4050 INT_GET(agfp
->agf_flcount
, ARCH_CONVERT
);
4051 xfs_buf_relse(agfbp
);
4053 agidaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
4054 agibp
= xfs_buf_read(mp
->m_ddev_targp
, agidaddr
,
4055 XFS_FSS_TO_BB(mp
, 1), 0);
4056 if (XFS_BUF_ISERROR(agibp
)) {
4057 xfs_ioerror_alert("xlog_recover_check_summary(agi)",
4058 mp
, agibp
, agidaddr
);
4060 agip
= XFS_BUF_TO_AGI(agibp
);
4061 ASSERT(XFS_AGI_MAGIC
==
4062 INT_GET(agip
->agi_magicnum
, ARCH_CONVERT
));
4063 ASSERT(XFS_AGI_GOOD_VERSION(
4064 INT_GET(agip
->agi_versionnum
, ARCH_CONVERT
)));
4065 ASSERT(INT_GET(agip
->agi_seqno
, ARCH_CONVERT
) == agno
);
4067 itotal
+= INT_GET(agip
->agi_count
, ARCH_CONVERT
);
4068 ifree
+= INT_GET(agip
->agi_freecount
, ARCH_CONVERT
);
4069 xfs_buf_relse(agibp
);
4072 sbbp
= xfs_getsb(mp
, 0);
4073 #ifdef XFS_LOUD_RECOVERY
4075 xfs_xlatesb(XFS_BUF_TO_SBP(sbbp
), sbp
, 1, XFS_SB_ALL_BITS
);
4077 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4078 sbp
->sb_icount
, itotal
);
4080 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4081 sbp
->sb_ifree
, ifree
);
4083 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4084 sbp
->sb_fdblocks
, freeblks
);
4087 * This is turned off until I account for the allocation
4088 * btree blocks which live in free space.
4090 ASSERT(sbp
->sb_icount
== itotal
);
4091 ASSERT(sbp
->sb_ifree
== ifree
);
4092 ASSERT(sbp
->sb_fdblocks
== freeblks
);
4095 xfs_buf_relse(sbbp
);