2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_error.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_inode_item.h"
40 #include "xfs_alloc.h"
41 #include "xfs_ialloc.h"
42 #include "xfs_log_priv.h"
43 #include "xfs_buf_item.h"
44 #include "xfs_log_recover.h"
45 #include "xfs_extfree_item.h"
46 #include "xfs_trans_priv.h"
47 #include "xfs_quota.h"
49 #include "xfs_utils.h"
51 STATIC
int xlog_find_zeroed(xlog_t
*, xfs_daddr_t
*);
52 STATIC
int xlog_clear_stale_blocks(xlog_t
*, xfs_lsn_t
);
53 STATIC
void xlog_recover_insert_item_backq(xlog_recover_item_t
**q
,
54 xlog_recover_item_t
*item
);
56 STATIC
void xlog_recover_check_summary(xlog_t
*);
57 STATIC
void xlog_recover_check_ail(xfs_mount_t
*, xfs_log_item_t
*, int);
59 #define xlog_recover_check_summary(log)
60 #define xlog_recover_check_ail(mp, lip, gen)
65 * Sector aligned buffer routines for buffer create/read/write/access
68 #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
69 ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
70 ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
71 #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
78 ASSERT(num_bblks
> 0);
80 if (log
->l_sectbb_log
) {
82 num_bblks
+= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, 1);
83 num_bblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, num_bblks
);
85 return xfs_buf_get_noaddr(BBTOB(num_bblks
), log
->l_mp
->m_logdev_targp
);
97 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
108 if (log
->l_sectbb_log
) {
109 blk_no
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, blk_no
);
110 nbblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, nbblks
);
114 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
117 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
120 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
121 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
123 xfsbdstrat(log
->l_mp
, bp
);
124 error
= xfs_iowait(bp
);
126 xfs_ioerror_alert("xlog_bread", log
->l_mp
,
127 bp
, XFS_BUF_ADDR(bp
));
132 * Write out the buffer at the given block for the given number of blocks.
133 * The buffer is kept locked across the write and is returned locked.
134 * This can only be used for synchronous log writes.
145 if (log
->l_sectbb_log
) {
146 blk_no
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, blk_no
);
147 nbblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, nbblks
);
151 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
153 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
154 XFS_BUF_ZEROFLAGS(bp
);
157 XFS_BUF_PSEMA(bp
, PRIBIO
);
158 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
159 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
161 if ((error
= xfs_bwrite(log
->l_mp
, bp
)))
162 xfs_ioerror_alert("xlog_bwrite", log
->l_mp
,
163 bp
, XFS_BUF_ADDR(bp
));
176 if (!log
->l_sectbb_log
)
177 return XFS_BUF_PTR(bp
);
179 ptr
= XFS_BUF_PTR(bp
) + BBTOB((int)blk_no
& log
->l_sectbb_mask
);
180 ASSERT(XFS_BUF_SIZE(bp
) >=
181 BBTOB(nbblks
+ (blk_no
& log
->l_sectbb_mask
)));
187 * dump debug superblock and log record information
190 xlog_header_check_dump(
192 xlog_rec_header_t
*head
)
196 cmn_err(CE_DEBUG
, "%s: SB : uuid = ", __func__
);
197 for (b
= 0; b
< 16; b
++)
198 cmn_err(CE_DEBUG
, "%02x", ((uchar_t
*)&mp
->m_sb
.sb_uuid
)[b
]);
199 cmn_err(CE_DEBUG
, ", fmt = %d\n", XLOG_FMT
);
200 cmn_err(CE_DEBUG
, " log : uuid = ");
201 for (b
= 0; b
< 16; b
++)
202 cmn_err(CE_DEBUG
, "%02x",((uchar_t
*)&head
->h_fs_uuid
)[b
]);
203 cmn_err(CE_DEBUG
, ", fmt = %d\n", be32_to_cpu(head
->h_fmt
));
206 #define xlog_header_check_dump(mp, head)
210 * check log record header for recovery
213 xlog_header_check_recover(
215 xlog_rec_header_t
*head
)
217 ASSERT(be32_to_cpu(head
->h_magicno
) == XLOG_HEADER_MAGIC_NUM
);
220 * IRIX doesn't write the h_fmt field and leaves it zeroed
221 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
222 * a dirty log created in IRIX.
224 if (unlikely(be32_to_cpu(head
->h_fmt
) != XLOG_FMT
)) {
226 "XFS: dirty log written in incompatible format - can't recover");
227 xlog_header_check_dump(mp
, head
);
228 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
229 XFS_ERRLEVEL_HIGH
, mp
);
230 return XFS_ERROR(EFSCORRUPTED
);
231 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
233 "XFS: dirty log entry has mismatched uuid - can't recover");
234 xlog_header_check_dump(mp
, head
);
235 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
236 XFS_ERRLEVEL_HIGH
, mp
);
237 return XFS_ERROR(EFSCORRUPTED
);
243 * read the head block of the log and check the header
246 xlog_header_check_mount(
248 xlog_rec_header_t
*head
)
250 ASSERT(be32_to_cpu(head
->h_magicno
) == XLOG_HEADER_MAGIC_NUM
);
252 if (uuid_is_nil(&head
->h_fs_uuid
)) {
254 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
255 * h_fs_uuid is nil, we assume this log was last mounted
256 * by IRIX and continue.
258 xlog_warn("XFS: nil uuid in log - IRIX style log");
259 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
260 xlog_warn("XFS: log has mismatched uuid - can't recover");
261 xlog_header_check_dump(mp
, head
);
262 XFS_ERROR_REPORT("xlog_header_check_mount",
263 XFS_ERRLEVEL_HIGH
, mp
);
264 return XFS_ERROR(EFSCORRUPTED
);
275 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *));
277 if (XFS_BUF_GETERROR(bp
)) {
279 * We're not going to bother about retrying
280 * this during recovery. One strike!
282 mp
= XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*);
283 xfs_ioerror_alert("xlog_recover_iodone",
284 mp
, bp
, XFS_BUF_ADDR(bp
));
285 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
287 XFS_BUF_SET_FSPRIVATE(bp
, NULL
);
288 XFS_BUF_CLR_IODONE_FUNC(bp
);
293 * This routine finds (to an approximation) the first block in the physical
294 * log which contains the given cycle. It uses a binary search algorithm.
295 * Note that the algorithm can not be perfect because the disk will not
296 * necessarily be perfect.
299 xlog_find_cycle_start(
302 xfs_daddr_t first_blk
,
303 xfs_daddr_t
*last_blk
,
311 mid_blk
= BLK_AVG(first_blk
, *last_blk
);
312 while (mid_blk
!= first_blk
&& mid_blk
!= *last_blk
) {
313 if ((error
= xlog_bread(log
, mid_blk
, 1, bp
)))
315 offset
= xlog_align(log
, mid_blk
, 1, bp
);
316 mid_cycle
= xlog_get_cycle(offset
);
317 if (mid_cycle
== cycle
) {
319 /* last_half_cycle == mid_cycle */
322 /* first_half_cycle == mid_cycle */
324 mid_blk
= BLK_AVG(first_blk
, *last_blk
);
326 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == *last_blk
) ||
327 (mid_blk
== *last_blk
&& mid_blk
-1 == first_blk
));
333 * Check that the range of blocks does not contain the cycle number
334 * given. The scan needs to occur from front to back and the ptr into the
335 * region must be updated since a later routine will need to perform another
336 * test. If the region is completely good, we end up returning the same
339 * Set blkno to -1 if we encounter no errors. This is an invalid block number
340 * since we don't ever expect logs to get this large.
343 xlog_find_verify_cycle(
345 xfs_daddr_t start_blk
,
347 uint stop_on_cycle_no
,
348 xfs_daddr_t
*new_blk
)
354 xfs_caddr_t buf
= NULL
;
357 bufblks
= 1 << ffs(nbblks
);
359 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
360 /* can't get enough memory to do everything in one big buffer */
362 if (bufblks
<= log
->l_sectbb_log
)
366 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
369 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
371 if ((error
= xlog_bread(log
, i
, bcount
, bp
)))
374 buf
= xlog_align(log
, i
, bcount
, bp
);
375 for (j
= 0; j
< bcount
; j
++) {
376 cycle
= xlog_get_cycle(buf
);
377 if (cycle
== stop_on_cycle_no
) {
394 * Potentially backup over partial log record write.
396 * In the typical case, last_blk is the number of the block directly after
397 * a good log record. Therefore, we subtract one to get the block number
398 * of the last block in the given buffer. extra_bblks contains the number
399 * of blocks we would have read on a previous read. This happens when the
400 * last log record is split over the end of the physical log.
402 * extra_bblks is the number of blocks potentially verified on a previous
403 * call to this routine.
406 xlog_find_verify_log_record(
408 xfs_daddr_t start_blk
,
409 xfs_daddr_t
*last_blk
,
414 xfs_caddr_t offset
= NULL
;
415 xlog_rec_header_t
*head
= NULL
;
418 int num_blks
= *last_blk
- start_blk
;
421 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
423 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
424 if (!(bp
= xlog_get_bp(log
, 1)))
428 if ((error
= xlog_bread(log
, start_blk
, num_blks
, bp
)))
430 offset
= xlog_align(log
, start_blk
, num_blks
, bp
);
431 offset
+= ((num_blks
- 1) << BBSHIFT
);
434 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
436 /* valid log record not found */
438 "XFS: Log inconsistent (didn't find previous header)");
440 error
= XFS_ERROR(EIO
);
445 if ((error
= xlog_bread(log
, i
, 1, bp
)))
447 offset
= xlog_align(log
, i
, 1, bp
);
450 head
= (xlog_rec_header_t
*)offset
;
452 if (XLOG_HEADER_MAGIC_NUM
== be32_to_cpu(head
->h_magicno
))
460 * We hit the beginning of the physical log & still no header. Return
461 * to caller. If caller can handle a return of -1, then this routine
462 * will be called again for the end of the physical log.
470 * We have the final block of the good log (the first block
471 * of the log record _before_ the head. So we check the uuid.
473 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
477 * We may have found a log record header before we expected one.
478 * last_blk will be the 1st block # with a given cycle #. We may end
479 * up reading an entire log record. In this case, we don't want to
480 * reset last_blk. Only when last_blk points in the middle of a log
481 * record do we update last_blk.
483 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
484 uint h_size
= be32_to_cpu(head
->h_size
);
486 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
487 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
493 if (*last_blk
- i
+ extra_bblks
!=
494 BTOBB(be32_to_cpu(head
->h_len
)) + xhdrs
)
503 * Head is defined to be the point of the log where the next log write
504 * write could go. This means that incomplete LR writes at the end are
505 * eliminated when calculating the head. We aren't guaranteed that previous
506 * LR have complete transactions. We only know that a cycle number of
507 * current cycle number -1 won't be present in the log if we start writing
508 * from our current block number.
510 * last_blk contains the block number of the first block with a given
513 * Return: zero if normal, non-zero if error.
518 xfs_daddr_t
*return_head_blk
)
522 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
524 uint first_half_cycle
, last_half_cycle
;
526 int error
, log_bbnum
= log
->l_logBBsize
;
528 /* Is the end of the log device zeroed? */
529 if ((error
= xlog_find_zeroed(log
, &first_blk
)) == -1) {
530 *return_head_blk
= first_blk
;
532 /* Is the whole lot zeroed? */
534 /* Linux XFS shouldn't generate totally zeroed logs -
535 * mkfs etc write a dummy unmount record to a fresh
536 * log so we can store the uuid in there
538 xlog_warn("XFS: totally zeroed log");
543 xlog_warn("XFS: empty log check failed");
547 first_blk
= 0; /* get cycle # of 1st block */
548 bp
= xlog_get_bp(log
, 1);
551 if ((error
= xlog_bread(log
, 0, 1, bp
)))
553 offset
= xlog_align(log
, 0, 1, bp
);
554 first_half_cycle
= xlog_get_cycle(offset
);
556 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
557 if ((error
= xlog_bread(log
, last_blk
, 1, bp
)))
559 offset
= xlog_align(log
, last_blk
, 1, bp
);
560 last_half_cycle
= xlog_get_cycle(offset
);
561 ASSERT(last_half_cycle
!= 0);
564 * If the 1st half cycle number is equal to the last half cycle number,
565 * then the entire log is stamped with the same cycle number. In this
566 * case, head_blk can't be set to zero (which makes sense). The below
567 * math doesn't work out properly with head_blk equal to zero. Instead,
568 * we set it to log_bbnum which is an invalid block number, but this
569 * value makes the math correct. If head_blk doesn't changed through
570 * all the tests below, *head_blk is set to zero at the very end rather
571 * than log_bbnum. In a sense, log_bbnum and zero are the same block
572 * in a circular file.
574 if (first_half_cycle
== last_half_cycle
) {
576 * In this case we believe that the entire log should have
577 * cycle number last_half_cycle. We need to scan backwards
578 * from the end verifying that there are no holes still
579 * containing last_half_cycle - 1. If we find such a hole,
580 * then the start of that hole will be the new head. The
581 * simple case looks like
582 * x | x ... | x - 1 | x
583 * Another case that fits this picture would be
584 * x | x + 1 | x ... | x
585 * In this case the head really is somewhere at the end of the
586 * log, as one of the latest writes at the beginning was
589 * x | x + 1 | x ... | x - 1 | x
590 * This is really the combination of the above two cases, and
591 * the head has to end up at the start of the x-1 hole at the
594 * In the 256k log case, we will read from the beginning to the
595 * end of the log and search for cycle numbers equal to x-1.
596 * We don't worry about the x+1 blocks that we encounter,
597 * because we know that they cannot be the head since the log
600 head_blk
= log_bbnum
;
601 stop_on_cycle
= last_half_cycle
- 1;
604 * In this case we want to find the first block with cycle
605 * number matching last_half_cycle. We expect the log to be
608 * The first block with cycle number x (last_half_cycle) will
609 * be where the new head belongs. First we do a binary search
610 * for the first occurrence of last_half_cycle. The binary
611 * search may not be totally accurate, so then we scan back
612 * from there looking for occurrences of last_half_cycle before
613 * us. If that backwards scan wraps around the beginning of
614 * the log, then we look for occurrences of last_half_cycle - 1
615 * at the end of the log. The cases we're looking for look
617 * x + 1 ... | x | x + 1 | x ...
618 * ^ binary search stopped here
620 * x + 1 ... | x ... | x - 1 | x
621 * <---------> less than scan distance
623 stop_on_cycle
= last_half_cycle
;
624 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
625 &head_blk
, last_half_cycle
)))
630 * Now validate the answer. Scan back some number of maximum possible
631 * blocks and make sure each one has the expected cycle number. The
632 * maximum is determined by the total possible amount of buffering
633 * in the in-core log. The following number can be made tighter if
634 * we actually look at the block size of the filesystem.
636 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
637 if (head_blk
>= num_scan_bblks
) {
639 * We are guaranteed that the entire check can be performed
642 start_blk
= head_blk
- num_scan_bblks
;
643 if ((error
= xlog_find_verify_cycle(log
,
644 start_blk
, num_scan_bblks
,
645 stop_on_cycle
, &new_blk
)))
649 } else { /* need to read 2 parts of log */
651 * We are going to scan backwards in the log in two parts.
652 * First we scan the physical end of the log. In this part
653 * of the log, we are looking for blocks with cycle number
654 * last_half_cycle - 1.
655 * If we find one, then we know that the log starts there, as
656 * we've found a hole that didn't get written in going around
657 * the end of the physical log. The simple case for this is
658 * x + 1 ... | x ... | x - 1 | x
659 * <---------> less than scan distance
660 * If all of the blocks at the end of the log have cycle number
661 * last_half_cycle, then we check the blocks at the start of
662 * the log looking for occurrences of last_half_cycle. If we
663 * find one, then our current estimate for the location of the
664 * first occurrence of last_half_cycle is wrong and we move
665 * back to the hole we've found. This case looks like
666 * x + 1 ... | x | x + 1 | x ...
667 * ^ binary search stopped here
668 * Another case we need to handle that only occurs in 256k
670 * x + 1 ... | x ... | x+1 | x ...
671 * ^ binary search stops here
672 * In a 256k log, the scan at the end of the log will see the
673 * x + 1 blocks. We need to skip past those since that is
674 * certainly not the head of the log. By searching for
675 * last_half_cycle-1 we accomplish that.
677 start_blk
= log_bbnum
- num_scan_bblks
+ head_blk
;
678 ASSERT(head_blk
<= INT_MAX
&&
679 (xfs_daddr_t
) num_scan_bblks
- head_blk
>= 0);
680 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
681 num_scan_bblks
- (int)head_blk
,
682 (stop_on_cycle
- 1), &new_blk
)))
690 * Scan beginning of log now. The last part of the physical
691 * log is good. This scan needs to verify that it doesn't find
692 * the last_half_cycle.
695 ASSERT(head_blk
<= INT_MAX
);
696 if ((error
= xlog_find_verify_cycle(log
,
697 start_blk
, (int)head_blk
,
698 stop_on_cycle
, &new_blk
)))
706 * Now we need to make sure head_blk is not pointing to a block in
707 * the middle of a log record.
709 num_scan_bblks
= XLOG_REC_SHIFT(log
);
710 if (head_blk
>= num_scan_bblks
) {
711 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
713 /* start ptr at last block ptr before head_blk */
714 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
715 &head_blk
, 0)) == -1) {
716 error
= XFS_ERROR(EIO
);
722 ASSERT(head_blk
<= INT_MAX
);
723 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
724 &head_blk
, 0)) == -1) {
725 /* We hit the beginning of the log during our search */
726 start_blk
= log_bbnum
- num_scan_bblks
+ head_blk
;
728 ASSERT(start_blk
<= INT_MAX
&&
729 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
730 ASSERT(head_blk
<= INT_MAX
);
731 if ((error
= xlog_find_verify_log_record(log
,
733 (int)head_blk
)) == -1) {
734 error
= XFS_ERROR(EIO
);
738 if (new_blk
!= log_bbnum
)
745 if (head_blk
== log_bbnum
)
746 *return_head_blk
= 0;
748 *return_head_blk
= head_blk
;
750 * When returning here, we have a good block number. Bad block
751 * means that during a previous crash, we didn't have a clean break
752 * from cycle number N to cycle number N-1. In this case, we need
753 * to find the first block with cycle number N-1.
761 xlog_warn("XFS: failed to find log head");
766 * Find the sync block number or the tail of the log.
768 * This will be the block number of the last record to have its
769 * associated buffers synced to disk. Every log record header has
770 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
771 * to get a sync block number. The only concern is to figure out which
772 * log record header to believe.
774 * The following algorithm uses the log record header with the largest
775 * lsn. The entire log record does not need to be valid. We only care
776 * that the header is valid.
778 * We could speed up search by using current head_blk buffer, but it is not
784 xfs_daddr_t
*head_blk
,
785 xfs_daddr_t
*tail_blk
)
787 xlog_rec_header_t
*rhead
;
788 xlog_op_header_t
*op_head
;
789 xfs_caddr_t offset
= NULL
;
792 xfs_daddr_t umount_data_blk
;
793 xfs_daddr_t after_umount_blk
;
800 * Find previous log record
802 if ((error
= xlog_find_head(log
, head_blk
)))
805 bp
= xlog_get_bp(log
, 1);
808 if (*head_blk
== 0) { /* special case */
809 if ((error
= xlog_bread(log
, 0, 1, bp
)))
811 offset
= xlog_align(log
, 0, 1, bp
);
812 if (xlog_get_cycle(offset
) == 0) {
814 /* leave all other log inited values alone */
820 * Search backwards looking for log record header block
822 ASSERT(*head_blk
< INT_MAX
);
823 for (i
= (int)(*head_blk
) - 1; i
>= 0; i
--) {
824 if ((error
= xlog_bread(log
, i
, 1, bp
)))
826 offset
= xlog_align(log
, i
, 1, bp
);
827 if (XLOG_HEADER_MAGIC_NUM
== be32_to_cpu(*(__be32
*)offset
)) {
833 * If we haven't found the log record header block, start looking
834 * again from the end of the physical log. XXXmiken: There should be
835 * a check here to make sure we didn't search more than N blocks in
839 for (i
= log
->l_logBBsize
- 1; i
>= (int)(*head_blk
); i
--) {
840 if ((error
= xlog_bread(log
, i
, 1, bp
)))
842 offset
= xlog_align(log
, i
, 1, bp
);
843 if (XLOG_HEADER_MAGIC_NUM
==
844 be32_to_cpu(*(__be32
*)offset
)) {
851 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
853 return XFS_ERROR(EIO
);
856 /* find blk_no of tail of log */
857 rhead
= (xlog_rec_header_t
*)offset
;
858 *tail_blk
= BLOCK_LSN(be64_to_cpu(rhead
->h_tail_lsn
));
861 * Reset log values according to the state of the log when we
862 * crashed. In the case where head_blk == 0, we bump curr_cycle
863 * one because the next write starts a new cycle rather than
864 * continuing the cycle of the last good log record. At this
865 * point we have guaranteed that all partial log records have been
866 * accounted for. Therefore, we know that the last good log record
867 * written was complete and ended exactly on the end boundary
868 * of the physical log.
870 log
->l_prev_block
= i
;
871 log
->l_curr_block
= (int)*head_blk
;
872 log
->l_curr_cycle
= be32_to_cpu(rhead
->h_cycle
);
875 log
->l_tail_lsn
= be64_to_cpu(rhead
->h_tail_lsn
);
876 log
->l_last_sync_lsn
= be64_to_cpu(rhead
->h_lsn
);
877 log
->l_grant_reserve_cycle
= log
->l_curr_cycle
;
878 log
->l_grant_reserve_bytes
= BBTOB(log
->l_curr_block
);
879 log
->l_grant_write_cycle
= log
->l_curr_cycle
;
880 log
->l_grant_write_bytes
= BBTOB(log
->l_curr_block
);
883 * Look for unmount record. If we find it, then we know there
884 * was a clean unmount. Since 'i' could be the last block in
885 * the physical log, we convert to a log block before comparing
888 * Save the current tail lsn to use to pass to
889 * xlog_clear_stale_blocks() below. We won't want to clear the
890 * unmount record if there is one, so we pass the lsn of the
891 * unmount record rather than the block after it.
893 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
894 int h_size
= be32_to_cpu(rhead
->h_size
);
895 int h_version
= be32_to_cpu(rhead
->h_version
);
897 if ((h_version
& XLOG_VERSION_2
) &&
898 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
899 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
900 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
908 after_umount_blk
= (i
+ hblks
+ (int)
909 BTOBB(be32_to_cpu(rhead
->h_len
))) % log
->l_logBBsize
;
910 tail_lsn
= log
->l_tail_lsn
;
911 if (*head_blk
== after_umount_blk
&&
912 be32_to_cpu(rhead
->h_num_logops
) == 1) {
913 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
914 if ((error
= xlog_bread(log
, umount_data_blk
, 1, bp
))) {
917 offset
= xlog_align(log
, umount_data_blk
, 1, bp
);
918 op_head
= (xlog_op_header_t
*)offset
;
919 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
921 * Set tail and last sync so that newly written
922 * log records will point recovery to after the
923 * current unmount record.
926 xlog_assign_lsn(log
->l_curr_cycle
,
928 log
->l_last_sync_lsn
=
929 xlog_assign_lsn(log
->l_curr_cycle
,
931 *tail_blk
= after_umount_blk
;
934 * Note that the unmount was clean. If the unmount
935 * was not clean, we need to know this to rebuild the
936 * superblock counters from the perag headers if we
937 * have a filesystem using non-persistent counters.
939 log
->l_mp
->m_flags
|= XFS_MOUNT_WAS_CLEAN
;
944 * Make sure that there are no blocks in front of the head
945 * with the same cycle number as the head. This can happen
946 * because we allow multiple outstanding log writes concurrently,
947 * and the later writes might make it out before earlier ones.
949 * We use the lsn from before modifying it so that we'll never
950 * overwrite the unmount record after a clean unmount.
952 * Do this only if we are going to recover the filesystem
954 * NOTE: This used to say "if (!readonly)"
955 * However on Linux, we can & do recover a read-only filesystem.
956 * We only skip recovery if NORECOVERY is specified on mount,
957 * in which case we would not be here.
959 * But... if the -device- itself is readonly, just skip this.
960 * We can't recover this device anyway, so it won't matter.
962 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
)) {
963 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
971 xlog_warn("XFS: failed to locate log tail");
976 * Is the log zeroed at all?
978 * The last binary search should be changed to perform an X block read
979 * once X becomes small enough. You can then search linearly through
980 * the X blocks. This will cut down on the number of reads we need to do.
982 * If the log is partially zeroed, this routine will pass back the blkno
983 * of the first block with cycle number 0. It won't have a complete LR
987 * 0 => the log is completely written to
988 * -1 => use *blk_no as the first block of the log
989 * >0 => error has occurred
998 uint first_cycle
, last_cycle
;
999 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1000 xfs_daddr_t num_scan_bblks
;
1001 int error
, log_bbnum
= log
->l_logBBsize
;
1005 /* check totally zeroed log */
1006 bp
= xlog_get_bp(log
, 1);
1009 if ((error
= xlog_bread(log
, 0, 1, bp
)))
1011 offset
= xlog_align(log
, 0, 1, bp
);
1012 first_cycle
= xlog_get_cycle(offset
);
1013 if (first_cycle
== 0) { /* completely zeroed log */
1019 /* check partially zeroed log */
1020 if ((error
= xlog_bread(log
, log_bbnum
-1, 1, bp
)))
1022 offset
= xlog_align(log
, log_bbnum
-1, 1, bp
);
1023 last_cycle
= xlog_get_cycle(offset
);
1024 if (last_cycle
!= 0) { /* log completely written to */
1027 } else if (first_cycle
!= 1) {
1029 * If the cycle of the last block is zero, the cycle of
1030 * the first block must be 1. If it's not, maybe we're
1031 * not looking at a log... Bail out.
1033 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1034 return XFS_ERROR(EINVAL
);
1037 /* we have a partially zeroed log */
1038 last_blk
= log_bbnum
-1;
1039 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1043 * Validate the answer. Because there is no way to guarantee that
1044 * the entire log is made up of log records which are the same size,
1045 * we scan over the defined maximum blocks. At this point, the maximum
1046 * is not chosen to mean anything special. XXXmiken
1048 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1049 ASSERT(num_scan_bblks
<= INT_MAX
);
1051 if (last_blk
< num_scan_bblks
)
1052 num_scan_bblks
= last_blk
;
1053 start_blk
= last_blk
- num_scan_bblks
;
1056 * We search for any instances of cycle number 0 that occur before
1057 * our current estimate of the head. What we're trying to detect is
1058 * 1 ... | 0 | 1 | 0...
1059 * ^ binary search ends here
1061 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1062 (int)num_scan_bblks
, 0, &new_blk
)))
1068 * Potentially backup over partial log record write. We don't need
1069 * to search the end of the log because we know it is zero.
1071 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
1072 &last_blk
, 0)) == -1) {
1073 error
= XFS_ERROR(EIO
);
1087 * These are simple subroutines used by xlog_clear_stale_blocks() below
1088 * to initialize a buffer full of empty log record headers and write
1089 * them into the log.
1100 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1102 memset(buf
, 0, BBSIZE
);
1103 recp
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1104 recp
->h_cycle
= cpu_to_be32(cycle
);
1105 recp
->h_version
= cpu_to_be32(
1106 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1107 recp
->h_lsn
= cpu_to_be64(xlog_assign_lsn(cycle
, block
));
1108 recp
->h_tail_lsn
= cpu_to_be64(xlog_assign_lsn(tail_cycle
, tail_block
));
1109 recp
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1110 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1114 xlog_write_log_records(
1125 int sectbb
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, 1);
1126 int end_block
= start_block
+ blocks
;
1131 bufblks
= 1 << ffs(blocks
);
1132 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1134 if (bufblks
<= log
->l_sectbb_log
)
1138 /* We may need to do a read at the start to fill in part of
1139 * the buffer in the starting sector not covered by the first
1142 balign
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, start_block
);
1143 if (balign
!= start_block
) {
1144 if ((error
= xlog_bread(log
, start_block
, 1, bp
))) {
1148 j
= start_block
- balign
;
1151 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1152 int bcount
, endcount
;
1154 bcount
= min(bufblks
, end_block
- start_block
);
1155 endcount
= bcount
- j
;
1157 /* We may need to do a read at the end to fill in part of
1158 * the buffer in the final sector not covered by the write.
1159 * If this is the same sector as the above read, skip it.
1161 ealign
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, end_block
);
1162 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1163 offset
= XFS_BUF_PTR(bp
);
1164 balign
= BBTOB(ealign
- start_block
);
1165 XFS_BUF_SET_PTR(bp
, offset
+ balign
, BBTOB(sectbb
));
1166 if ((error
= xlog_bread(log
, ealign
, sectbb
, bp
)))
1168 XFS_BUF_SET_PTR(bp
, offset
, bufblks
);
1171 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1172 for (; j
< endcount
; j
++) {
1173 xlog_add_record(log
, offset
, cycle
, i
+j
,
1174 tail_cycle
, tail_block
);
1177 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1180 start_block
+= endcount
;
1188 * This routine is called to blow away any incomplete log writes out
1189 * in front of the log head. We do this so that we won't become confused
1190 * if we come up, write only a little bit more, and then crash again.
1191 * If we leave the partial log records out there, this situation could
1192 * cause us to think those partial writes are valid blocks since they
1193 * have the current cycle number. We get rid of them by overwriting them
1194 * with empty log records with the old cycle number rather than the
1197 * The tail lsn is passed in rather than taken from
1198 * the log so that we will not write over the unmount record after a
1199 * clean unmount in a 512 block log. Doing so would leave the log without
1200 * any valid log records in it until a new one was written. If we crashed
1201 * during that time we would not be able to recover.
1204 xlog_clear_stale_blocks(
1208 int tail_cycle
, head_cycle
;
1209 int tail_block
, head_block
;
1210 int tail_distance
, max_distance
;
1214 tail_cycle
= CYCLE_LSN(tail_lsn
);
1215 tail_block
= BLOCK_LSN(tail_lsn
);
1216 head_cycle
= log
->l_curr_cycle
;
1217 head_block
= log
->l_curr_block
;
1220 * Figure out the distance between the new head of the log
1221 * and the tail. We want to write over any blocks beyond the
1222 * head that we may have written just before the crash, but
1223 * we don't want to overwrite the tail of the log.
1225 if (head_cycle
== tail_cycle
) {
1227 * The tail is behind the head in the physical log,
1228 * so the distance from the head to the tail is the
1229 * distance from the head to the end of the log plus
1230 * the distance from the beginning of the log to the
1233 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1234 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1235 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1236 return XFS_ERROR(EFSCORRUPTED
);
1238 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1241 * The head is behind the tail in the physical log,
1242 * so the distance from the head to the tail is just
1243 * the tail block minus the head block.
1245 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1246 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1247 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1248 return XFS_ERROR(EFSCORRUPTED
);
1250 tail_distance
= tail_block
- head_block
;
1254 * If the head is right up against the tail, we can't clear
1257 if (tail_distance
<= 0) {
1258 ASSERT(tail_distance
== 0);
1262 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1264 * Take the smaller of the maximum amount of outstanding I/O
1265 * we could have and the distance to the tail to clear out.
1266 * We take the smaller so that we don't overwrite the tail and
1267 * we don't waste all day writing from the head to the tail
1270 max_distance
= MIN(max_distance
, tail_distance
);
1272 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1274 * We can stomp all the blocks we need to without
1275 * wrapping around the end of the log. Just do it
1276 * in a single write. Use the cycle number of the
1277 * current cycle minus one so that the log will look like:
1280 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1281 head_block
, max_distance
, tail_cycle
,
1287 * We need to wrap around the end of the physical log in
1288 * order to clear all the blocks. Do it in two separate
1289 * I/Os. The first write should be from the head to the
1290 * end of the physical log, and it should use the current
1291 * cycle number minus one just like above.
1293 distance
= log
->l_logBBsize
- head_block
;
1294 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1295 head_block
, distance
, tail_cycle
,
1302 * Now write the blocks at the start of the physical log.
1303 * This writes the remainder of the blocks we want to clear.
1304 * It uses the current cycle number since we're now on the
1305 * same cycle as the head so that we get:
1306 * n ... n ... | n - 1 ...
1307 * ^^^^^ blocks we're writing
1309 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1310 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1311 tail_cycle
, tail_block
);
1319 /******************************************************************************
1321 * Log recover routines
1323 ******************************************************************************
1326 STATIC xlog_recover_t
*
1327 xlog_recover_find_tid(
1331 xlog_recover_t
*p
= q
;
1334 if (p
->r_log_tid
== tid
)
1342 xlog_recover_put_hashq(
1344 xlog_recover_t
*trans
)
1351 xlog_recover_add_item(
1352 xlog_recover_item_t
**itemq
)
1354 xlog_recover_item_t
*item
;
1356 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
1357 xlog_recover_insert_item_backq(itemq
, item
);
1361 xlog_recover_add_to_cont_trans(
1362 xlog_recover_t
*trans
,
1366 xlog_recover_item_t
*item
;
1367 xfs_caddr_t ptr
, old_ptr
;
1370 item
= trans
->r_itemq
;
1372 /* finish copying rest of trans header */
1373 xlog_recover_add_item(&trans
->r_itemq
);
1374 ptr
= (xfs_caddr_t
) &trans
->r_theader
+
1375 sizeof(xfs_trans_header_t
) - len
;
1376 memcpy(ptr
, dp
, len
); /* d, s, l */
1379 item
= item
->ri_prev
;
1381 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
1382 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
1384 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, 0u);
1385 memcpy(&ptr
[old_len
], dp
, len
); /* d, s, l */
1386 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
1387 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
1392 * The next region to add is the start of a new region. It could be
1393 * a whole region or it could be the first part of a new region. Because
1394 * of this, the assumption here is that the type and size fields of all
1395 * format structures fit into the first 32 bits of the structure.
1397 * This works because all regions must be 32 bit aligned. Therefore, we
1398 * either have both fields or we have neither field. In the case we have
1399 * neither field, the data part of the region is zero length. We only have
1400 * a log_op_header and can throw away the header since a new one will appear
1401 * later. If we have at least 4 bytes, then we can determine how many regions
1402 * will appear in the current log item.
1405 xlog_recover_add_to_trans(
1406 xlog_recover_t
*trans
,
1410 xfs_inode_log_format_t
*in_f
; /* any will do */
1411 xlog_recover_item_t
*item
;
1416 item
= trans
->r_itemq
;
1418 ASSERT(*(uint
*)dp
== XFS_TRANS_HEADER_MAGIC
);
1419 if (len
== sizeof(xfs_trans_header_t
))
1420 xlog_recover_add_item(&trans
->r_itemq
);
1421 memcpy(&trans
->r_theader
, dp
, len
); /* d, s, l */
1425 ptr
= kmem_alloc(len
, KM_SLEEP
);
1426 memcpy(ptr
, dp
, len
);
1427 in_f
= (xfs_inode_log_format_t
*)ptr
;
1429 if (item
->ri_prev
->ri_total
!= 0 &&
1430 item
->ri_prev
->ri_total
== item
->ri_prev
->ri_cnt
) {
1431 xlog_recover_add_item(&trans
->r_itemq
);
1433 item
= trans
->r_itemq
;
1434 item
= item
->ri_prev
;
1436 if (item
->ri_total
== 0) { /* first region to be added */
1437 item
->ri_total
= in_f
->ilf_size
;
1438 ASSERT(item
->ri_total
<= XLOG_MAX_REGIONS_IN_ITEM
);
1439 item
->ri_buf
= kmem_zalloc((item
->ri_total
*
1440 sizeof(xfs_log_iovec_t
)), KM_SLEEP
);
1442 ASSERT(item
->ri_total
> item
->ri_cnt
);
1443 /* Description region is ri_buf[0] */
1444 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
1445 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
1451 xlog_recover_new_tid(
1456 xlog_recover_t
*trans
;
1458 trans
= kmem_zalloc(sizeof(xlog_recover_t
), KM_SLEEP
);
1459 trans
->r_log_tid
= tid
;
1461 xlog_recover_put_hashq(q
, trans
);
1465 xlog_recover_unlink_tid(
1467 xlog_recover_t
*trans
)
1472 ASSERT(trans
!= NULL
);
1478 if (tp
->r_next
== trans
) {
1486 "XFS: xlog_recover_unlink_tid: trans not found");
1488 return XFS_ERROR(EIO
);
1490 tp
->r_next
= tp
->r_next
->r_next
;
1496 xlog_recover_insert_item_backq(
1497 xlog_recover_item_t
**q
,
1498 xlog_recover_item_t
*item
)
1501 item
->ri_prev
= item
->ri_next
= item
;
1505 item
->ri_prev
= (*q
)->ri_prev
;
1506 (*q
)->ri_prev
= item
;
1507 item
->ri_prev
->ri_next
= item
;
1512 xlog_recover_insert_item_frontq(
1513 xlog_recover_item_t
**q
,
1514 xlog_recover_item_t
*item
)
1516 xlog_recover_insert_item_backq(q
, item
);
1521 xlog_recover_reorder_trans(
1522 xlog_recover_t
*trans
)
1524 xlog_recover_item_t
*first_item
, *itemq
, *itemq_next
;
1525 xfs_buf_log_format_t
*buf_f
;
1528 first_item
= itemq
= trans
->r_itemq
;
1529 trans
->r_itemq
= NULL
;
1531 itemq_next
= itemq
->ri_next
;
1532 buf_f
= (xfs_buf_log_format_t
*)itemq
->ri_buf
[0].i_addr
;
1534 switch (ITEM_TYPE(itemq
)) {
1536 flags
= buf_f
->blf_flags
;
1537 if (!(flags
& XFS_BLI_CANCEL
)) {
1538 xlog_recover_insert_item_frontq(&trans
->r_itemq
,
1544 case XFS_LI_QUOTAOFF
:
1547 xlog_recover_insert_item_backq(&trans
->r_itemq
, itemq
);
1551 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1553 return XFS_ERROR(EIO
);
1556 } while (first_item
!= itemq
);
1561 * Build up the table of buf cancel records so that we don't replay
1562 * cancelled data in the second pass. For buffer records that are
1563 * not cancel records, there is nothing to do here so we just return.
1565 * If we get a cancel record which is already in the table, this indicates
1566 * that the buffer was cancelled multiple times. In order to ensure
1567 * that during pass 2 we keep the record in the table until we reach its
1568 * last occurrence in the log, we keep a reference count in the cancel
1569 * record in the table to tell us how many times we expect to see this
1570 * record during the second pass.
1573 xlog_recover_do_buffer_pass1(
1575 xfs_buf_log_format_t
*buf_f
)
1577 xfs_buf_cancel_t
*bcp
;
1578 xfs_buf_cancel_t
*nextp
;
1579 xfs_buf_cancel_t
*prevp
;
1580 xfs_buf_cancel_t
**bucket
;
1581 xfs_daddr_t blkno
= 0;
1585 switch (buf_f
->blf_type
) {
1587 blkno
= buf_f
->blf_blkno
;
1588 len
= buf_f
->blf_len
;
1589 flags
= buf_f
->blf_flags
;
1594 * If this isn't a cancel buffer item, then just return.
1596 if (!(flags
& XFS_BLI_CANCEL
))
1600 * Insert an xfs_buf_cancel record into the hash table of
1601 * them. If there is already an identical record, bump
1602 * its reference count.
1604 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1605 XLOG_BC_TABLE_SIZE
];
1607 * If the hash bucket is empty then just insert a new record into
1610 if (*bucket
== NULL
) {
1611 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1613 bcp
->bc_blkno
= blkno
;
1615 bcp
->bc_refcount
= 1;
1616 bcp
->bc_next
= NULL
;
1622 * The hash bucket is not empty, so search for duplicates of our
1623 * record. If we find one them just bump its refcount. If not
1624 * then add us at the end of the list.
1628 while (nextp
!= NULL
) {
1629 if (nextp
->bc_blkno
== blkno
&& nextp
->bc_len
== len
) {
1630 nextp
->bc_refcount
++;
1634 nextp
= nextp
->bc_next
;
1636 ASSERT(prevp
!= NULL
);
1637 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1639 bcp
->bc_blkno
= blkno
;
1641 bcp
->bc_refcount
= 1;
1642 bcp
->bc_next
= NULL
;
1643 prevp
->bc_next
= bcp
;
1647 * Check to see whether the buffer being recovered has a corresponding
1648 * entry in the buffer cancel record table. If it does then return 1
1649 * so that it will be cancelled, otherwise return 0. If the buffer is
1650 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1651 * the refcount on the entry in the table and remove it from the table
1652 * if this is the last reference.
1654 * We remove the cancel record from the table when we encounter its
1655 * last occurrence in the log so that if the same buffer is re-used
1656 * again after its last cancellation we actually replay the changes
1657 * made at that point.
1660 xlog_check_buffer_cancelled(
1666 xfs_buf_cancel_t
*bcp
;
1667 xfs_buf_cancel_t
*prevp
;
1668 xfs_buf_cancel_t
**bucket
;
1670 if (log
->l_buf_cancel_table
== NULL
) {
1672 * There is nothing in the table built in pass one,
1673 * so this buffer must not be cancelled.
1675 ASSERT(!(flags
& XFS_BLI_CANCEL
));
1679 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1680 XLOG_BC_TABLE_SIZE
];
1684 * There is no corresponding entry in the table built
1685 * in pass one, so this buffer has not been cancelled.
1687 ASSERT(!(flags
& XFS_BLI_CANCEL
));
1692 * Search for an entry in the buffer cancel table that
1693 * matches our buffer.
1696 while (bcp
!= NULL
) {
1697 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
) {
1699 * We've go a match, so return 1 so that the
1700 * recovery of this buffer is cancelled.
1701 * If this buffer is actually a buffer cancel
1702 * log item, then decrement the refcount on the
1703 * one in the table and remove it if this is the
1706 if (flags
& XFS_BLI_CANCEL
) {
1708 if (bcp
->bc_refcount
== 0) {
1709 if (prevp
== NULL
) {
1710 *bucket
= bcp
->bc_next
;
1712 prevp
->bc_next
= bcp
->bc_next
;
1715 sizeof(xfs_buf_cancel_t
));
1724 * We didn't find a corresponding entry in the table, so
1725 * return 0 so that the buffer is NOT cancelled.
1727 ASSERT(!(flags
& XFS_BLI_CANCEL
));
1732 xlog_recover_do_buffer_pass2(
1734 xfs_buf_log_format_t
*buf_f
)
1736 xfs_daddr_t blkno
= 0;
1740 switch (buf_f
->blf_type
) {
1742 blkno
= buf_f
->blf_blkno
;
1743 flags
= buf_f
->blf_flags
;
1744 len
= buf_f
->blf_len
;
1748 return xlog_check_buffer_cancelled(log
, blkno
, len
, flags
);
1752 * Perform recovery for a buffer full of inodes. In these buffers,
1753 * the only data which should be recovered is that which corresponds
1754 * to the di_next_unlinked pointers in the on disk inode structures.
1755 * The rest of the data for the inodes is always logged through the
1756 * inodes themselves rather than the inode buffer and is recovered
1757 * in xlog_recover_do_inode_trans().
1759 * The only time when buffers full of inodes are fully recovered is
1760 * when the buffer is full of newly allocated inodes. In this case
1761 * the buffer will not be marked as an inode buffer and so will be
1762 * sent to xlog_recover_do_reg_buffer() below during recovery.
1765 xlog_recover_do_inode_buffer(
1767 xlog_recover_item_t
*item
,
1769 xfs_buf_log_format_t
*buf_f
)
1777 int next_unlinked_offset
;
1779 xfs_agino_t
*logged_nextp
;
1780 xfs_agino_t
*buffer_nextp
;
1781 unsigned int *data_map
= NULL
;
1782 unsigned int map_size
= 0;
1784 switch (buf_f
->blf_type
) {
1786 data_map
= buf_f
->blf_data_map
;
1787 map_size
= buf_f
->blf_map_size
;
1791 * Set the variables corresponding to the current region to
1792 * 0 so that we'll initialize them on the first pass through
1800 inodes_per_buf
= XFS_BUF_COUNT(bp
) >> mp
->m_sb
.sb_inodelog
;
1801 for (i
= 0; i
< inodes_per_buf
; i
++) {
1802 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1803 offsetof(xfs_dinode_t
, di_next_unlinked
);
1805 while (next_unlinked_offset
>=
1806 (reg_buf_offset
+ reg_buf_bytes
)) {
1808 * The next di_next_unlinked field is beyond
1809 * the current logged region. Find the next
1810 * logged region that contains or is beyond
1811 * the current di_next_unlinked field.
1814 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1817 * If there are no more logged regions in the
1818 * buffer, then we're done.
1824 nbits
= xfs_contig_bits(data_map
, map_size
,
1827 reg_buf_offset
= bit
<< XFS_BLI_SHIFT
;
1828 reg_buf_bytes
= nbits
<< XFS_BLI_SHIFT
;
1833 * If the current logged region starts after the current
1834 * di_next_unlinked field, then move on to the next
1835 * di_next_unlinked field.
1837 if (next_unlinked_offset
< reg_buf_offset
) {
1841 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1842 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLI_CHUNK
) == 0);
1843 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <= XFS_BUF_COUNT(bp
));
1846 * The current logged region contains a copy of the
1847 * current di_next_unlinked field. Extract its value
1848 * and copy it to the buffer copy.
1850 logged_nextp
= (xfs_agino_t
*)
1851 ((char *)(item
->ri_buf
[item_index
].i_addr
) +
1852 (next_unlinked_offset
- reg_buf_offset
));
1853 if (unlikely(*logged_nextp
== 0)) {
1854 xfs_fs_cmn_err(CE_ALERT
, mp
,
1855 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1857 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1858 XFS_ERRLEVEL_LOW
, mp
);
1859 return XFS_ERROR(EFSCORRUPTED
);
1862 buffer_nextp
= (xfs_agino_t
*)xfs_buf_offset(bp
,
1863 next_unlinked_offset
);
1864 *buffer_nextp
= *logged_nextp
;
1871 * Perform a 'normal' buffer recovery. Each logged region of the
1872 * buffer should be copied over the corresponding region in the
1873 * given buffer. The bitmap in the buf log format structure indicates
1874 * where to place the logged data.
1878 xlog_recover_do_reg_buffer(
1879 xlog_recover_item_t
*item
,
1881 xfs_buf_log_format_t
*buf_f
)
1886 unsigned int *data_map
= NULL
;
1887 unsigned int map_size
= 0;
1890 switch (buf_f
->blf_type
) {
1892 data_map
= buf_f
->blf_data_map
;
1893 map_size
= buf_f
->blf_map_size
;
1897 i
= 1; /* 0 is the buf format structure */
1899 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1902 nbits
= xfs_contig_bits(data_map
, map_size
, bit
);
1904 ASSERT(item
->ri_buf
[i
].i_addr
!= NULL
);
1905 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLI_CHUNK
== 0);
1906 ASSERT(XFS_BUF_COUNT(bp
) >=
1907 ((uint
)bit
<< XFS_BLI_SHIFT
)+(nbits
<<XFS_BLI_SHIFT
));
1910 * Do a sanity check if this is a dquot buffer. Just checking
1911 * the first dquot in the buffer should do. XXXThis is
1912 * probably a good thing to do for other buf types also.
1915 if (buf_f
->blf_flags
&
1916 (XFS_BLI_UDQUOT_BUF
|XFS_BLI_PDQUOT_BUF
|XFS_BLI_GDQUOT_BUF
)) {
1917 error
= xfs_qm_dqcheck((xfs_disk_dquot_t
*)
1918 item
->ri_buf
[i
].i_addr
,
1919 -1, 0, XFS_QMOPT_DOWARN
,
1920 "dquot_buf_recover");
1923 memcpy(xfs_buf_offset(bp
,
1924 (uint
)bit
<< XFS_BLI_SHIFT
), /* dest */
1925 item
->ri_buf
[i
].i_addr
, /* source */
1926 nbits
<<XFS_BLI_SHIFT
); /* length */
1931 /* Shouldn't be any more regions */
1932 ASSERT(i
== item
->ri_total
);
1936 * Do some primitive error checking on ondisk dquot data structures.
1940 xfs_disk_dquot_t
*ddq
,
1942 uint type
, /* used only when IO_dorepair is true */
1946 xfs_dqblk_t
*d
= (xfs_dqblk_t
*)ddq
;
1950 * We can encounter an uninitialized dquot buffer for 2 reasons:
1951 * 1. If we crash while deleting the quotainode(s), and those blks got
1952 * used for user data. This is because we take the path of regular
1953 * file deletion; however, the size field of quotainodes is never
1954 * updated, so all the tricks that we play in itruncate_finish
1955 * don't quite matter.
1957 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1958 * But the allocation will be replayed so we'll end up with an
1959 * uninitialized quota block.
1961 * This is all fine; things are still consistent, and we haven't lost
1962 * any quota information. Just don't complain about bad dquot blks.
1964 if (be16_to_cpu(ddq
->d_magic
) != XFS_DQUOT_MAGIC
) {
1965 if (flags
& XFS_QMOPT_DOWARN
)
1967 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
1968 str
, id
, be16_to_cpu(ddq
->d_magic
), XFS_DQUOT_MAGIC
);
1971 if (ddq
->d_version
!= XFS_DQUOT_VERSION
) {
1972 if (flags
& XFS_QMOPT_DOWARN
)
1974 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
1975 str
, id
, ddq
->d_version
, XFS_DQUOT_VERSION
);
1979 if (ddq
->d_flags
!= XFS_DQ_USER
&&
1980 ddq
->d_flags
!= XFS_DQ_PROJ
&&
1981 ddq
->d_flags
!= XFS_DQ_GROUP
) {
1982 if (flags
& XFS_QMOPT_DOWARN
)
1984 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
1985 str
, id
, ddq
->d_flags
);
1989 if (id
!= -1 && id
!= be32_to_cpu(ddq
->d_id
)) {
1990 if (flags
& XFS_QMOPT_DOWARN
)
1992 "%s : ondisk-dquot 0x%p, ID mismatch: "
1993 "0x%x expected, found id 0x%x",
1994 str
, ddq
, id
, be32_to_cpu(ddq
->d_id
));
1998 if (!errs
&& ddq
->d_id
) {
1999 if (ddq
->d_blk_softlimit
&&
2000 be64_to_cpu(ddq
->d_bcount
) >=
2001 be64_to_cpu(ddq
->d_blk_softlimit
)) {
2002 if (!ddq
->d_btimer
) {
2003 if (flags
& XFS_QMOPT_DOWARN
)
2005 "%s : Dquot ID 0x%x (0x%p) "
2006 "BLK TIMER NOT STARTED",
2007 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2011 if (ddq
->d_ino_softlimit
&&
2012 be64_to_cpu(ddq
->d_icount
) >=
2013 be64_to_cpu(ddq
->d_ino_softlimit
)) {
2014 if (!ddq
->d_itimer
) {
2015 if (flags
& XFS_QMOPT_DOWARN
)
2017 "%s : Dquot ID 0x%x (0x%p) "
2018 "INODE TIMER NOT STARTED",
2019 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2023 if (ddq
->d_rtb_softlimit
&&
2024 be64_to_cpu(ddq
->d_rtbcount
) >=
2025 be64_to_cpu(ddq
->d_rtb_softlimit
)) {
2026 if (!ddq
->d_rtbtimer
) {
2027 if (flags
& XFS_QMOPT_DOWARN
)
2029 "%s : Dquot ID 0x%x (0x%p) "
2030 "RTBLK TIMER NOT STARTED",
2031 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2037 if (!errs
|| !(flags
& XFS_QMOPT_DQREPAIR
))
2040 if (flags
& XFS_QMOPT_DOWARN
)
2041 cmn_err(CE_NOTE
, "Re-initializing dquot ID 0x%x", id
);
2044 * Typically, a repair is only requested by quotacheck.
2047 ASSERT(flags
& XFS_QMOPT_DQREPAIR
);
2048 memset(d
, 0, sizeof(xfs_dqblk_t
));
2050 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
2051 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
2052 d
->dd_diskdq
.d_flags
= type
;
2053 d
->dd_diskdq
.d_id
= cpu_to_be32(id
);
2059 * Perform a dquot buffer recovery.
2060 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2061 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2062 * Else, treat it as a regular buffer and do recovery.
2065 xlog_recover_do_dquot_buffer(
2068 xlog_recover_item_t
*item
,
2070 xfs_buf_log_format_t
*buf_f
)
2075 * Filesystems are required to send in quota flags at mount time.
2077 if (mp
->m_qflags
== 0) {
2082 if (buf_f
->blf_flags
& XFS_BLI_UDQUOT_BUF
)
2083 type
|= XFS_DQ_USER
;
2084 if (buf_f
->blf_flags
& XFS_BLI_PDQUOT_BUF
)
2085 type
|= XFS_DQ_PROJ
;
2086 if (buf_f
->blf_flags
& XFS_BLI_GDQUOT_BUF
)
2087 type
|= XFS_DQ_GROUP
;
2089 * This type of quotas was turned off, so ignore this buffer
2091 if (log
->l_quotaoffs_flag
& type
)
2094 xlog_recover_do_reg_buffer(item
, bp
, buf_f
);
2098 * This routine replays a modification made to a buffer at runtime.
2099 * There are actually two types of buffer, regular and inode, which
2100 * are handled differently. Inode buffers are handled differently
2101 * in that we only recover a specific set of data from them, namely
2102 * the inode di_next_unlinked fields. This is because all other inode
2103 * data is actually logged via inode records and any data we replay
2104 * here which overlaps that may be stale.
2106 * When meta-data buffers are freed at run time we log a buffer item
2107 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2108 * of the buffer in the log should not be replayed at recovery time.
2109 * This is so that if the blocks covered by the buffer are reused for
2110 * file data before we crash we don't end up replaying old, freed
2111 * meta-data into a user's file.
2113 * To handle the cancellation of buffer log items, we make two passes
2114 * over the log during recovery. During the first we build a table of
2115 * those buffers which have been cancelled, and during the second we
2116 * only replay those buffers which do not have corresponding cancel
2117 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2118 * for more details on the implementation of the table of cancel records.
2121 xlog_recover_do_buffer_trans(
2123 xlog_recover_item_t
*item
,
2126 xfs_buf_log_format_t
*buf_f
;
2135 buf_f
= (xfs_buf_log_format_t
*)item
->ri_buf
[0].i_addr
;
2137 if (pass
== XLOG_RECOVER_PASS1
) {
2139 * In this pass we're only looking for buf items
2140 * with the XFS_BLI_CANCEL bit set.
2142 xlog_recover_do_buffer_pass1(log
, buf_f
);
2146 * In this pass we want to recover all the buffers
2147 * which have not been cancelled and are not
2148 * cancellation buffers themselves. The routine
2149 * we call here will tell us whether or not to
2150 * continue with the replay of this buffer.
2152 cancel
= xlog_recover_do_buffer_pass2(log
, buf_f
);
2157 switch (buf_f
->blf_type
) {
2159 blkno
= buf_f
->blf_blkno
;
2160 len
= buf_f
->blf_len
;
2161 flags
= buf_f
->blf_flags
;
2164 xfs_fs_cmn_err(CE_ALERT
, log
->l_mp
,
2165 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2166 buf_f
->blf_type
, log
->l_mp
->m_logname
?
2167 log
->l_mp
->m_logname
: "internal");
2168 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2169 XFS_ERRLEVEL_LOW
, log
->l_mp
);
2170 return XFS_ERROR(EFSCORRUPTED
);
2174 if (flags
& XFS_BLI_INODE_BUF
) {
2175 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, blkno
, len
,
2178 bp
= xfs_buf_read(mp
->m_ddev_targp
, blkno
, len
, 0);
2180 if (XFS_BUF_ISERROR(bp
)) {
2181 xfs_ioerror_alert("xlog_recover_do..(read#1)", log
->l_mp
,
2183 error
= XFS_BUF_GETERROR(bp
);
2189 if (flags
& XFS_BLI_INODE_BUF
) {
2190 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2192 (XFS_BLI_UDQUOT_BUF
|XFS_BLI_PDQUOT_BUF
|XFS_BLI_GDQUOT_BUF
)) {
2193 xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2195 xlog_recover_do_reg_buffer(item
, bp
, buf_f
);
2198 return XFS_ERROR(error
);
2201 * Perform delayed write on the buffer. Asynchronous writes will be
2202 * slower when taking into account all the buffers to be flushed.
2204 * Also make sure that only inode buffers with good sizes stay in
2205 * the buffer cache. The kernel moves inodes in buffers of 1 block
2206 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2207 * buffers in the log can be a different size if the log was generated
2208 * by an older kernel using unclustered inode buffers or a newer kernel
2209 * running with a different inode cluster size. Regardless, if the
2210 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2211 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2212 * the buffer out of the buffer cache so that the buffer won't
2213 * overlap with future reads of those inodes.
2215 if (XFS_DINODE_MAGIC
==
2216 be16_to_cpu(*((__be16
*)xfs_buf_offset(bp
, 0))) &&
2217 (XFS_BUF_COUNT(bp
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2218 (__uint32_t
)XFS_INODE_CLUSTER_SIZE(log
->l_mp
)))) {
2220 error
= xfs_bwrite(mp
, bp
);
2222 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
||
2223 XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*) == mp
);
2224 XFS_BUF_SET_FSPRIVATE(bp
, mp
);
2225 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2226 xfs_bdwrite(mp
, bp
);
2233 xlog_recover_do_inode_trans(
2235 xlog_recover_item_t
*item
,
2238 xfs_inode_log_format_t
*in_f
;
2250 xfs_icdinode_t
*dicp
;
2253 if (pass
== XLOG_RECOVER_PASS1
) {
2257 if (item
->ri_buf
[0].i_len
== sizeof(xfs_inode_log_format_t
)) {
2258 in_f
= (xfs_inode_log_format_t
*)item
->ri_buf
[0].i_addr
;
2260 in_f
= (xfs_inode_log_format_t
*)kmem_alloc(
2261 sizeof(xfs_inode_log_format_t
), KM_SLEEP
);
2263 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
2267 ino
= in_f
->ilf_ino
;
2269 if (ITEM_TYPE(item
) == XFS_LI_INODE
) {
2270 imap
.im_blkno
= (xfs_daddr_t
)in_f
->ilf_blkno
;
2271 imap
.im_len
= in_f
->ilf_len
;
2272 imap
.im_boffset
= in_f
->ilf_boffset
;
2275 * It's an old inode format record. We don't know where
2276 * its cluster is located on disk, and we can't allow
2277 * xfs_imap() to figure it out because the inode btrees
2278 * are not ready to be used. Therefore do not pass the
2279 * XFS_IMAP_LOOKUP flag to xfs_imap(). This will give
2280 * us only the single block in which the inode lives
2281 * rather than its cluster, so we must make sure to
2282 * invalidate the buffer when we write it out below.
2285 xfs_imap(log
->l_mp
, NULL
, ino
, &imap
, 0);
2289 * Inode buffers can be freed, look out for it,
2290 * and do not replay the inode.
2292 if (xlog_check_buffer_cancelled(log
, imap
.im_blkno
, imap
.im_len
, 0)) {
2297 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, imap
.im_blkno
, imap
.im_len
,
2299 if (XFS_BUF_ISERROR(bp
)) {
2300 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp
,
2302 error
= XFS_BUF_GETERROR(bp
);
2307 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2308 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
2311 * Make sure the place we're flushing out to really looks
2314 if (unlikely(be16_to_cpu(dip
->di_core
.di_magic
) != XFS_DINODE_MAGIC
)) {
2316 xfs_fs_cmn_err(CE_ALERT
, mp
,
2317 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2319 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2320 XFS_ERRLEVEL_LOW
, mp
);
2321 error
= EFSCORRUPTED
;
2324 dicp
= (xfs_icdinode_t
*)(item
->ri_buf
[1].i_addr
);
2325 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2327 xfs_fs_cmn_err(CE_ALERT
, mp
,
2328 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2330 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2331 XFS_ERRLEVEL_LOW
, mp
);
2332 error
= EFSCORRUPTED
;
2336 /* Skip replay when the on disk inode is newer than the log one */
2337 if (dicp
->di_flushiter
< be16_to_cpu(dip
->di_core
.di_flushiter
)) {
2339 * Deal with the wrap case, DI_MAX_FLUSH is less
2340 * than smaller numbers
2342 if (be16_to_cpu(dip
->di_core
.di_flushiter
) == DI_MAX_FLUSH
&&
2343 dicp
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
2351 /* Take the opportunity to reset the flush iteration count */
2352 dicp
->di_flushiter
= 0;
2354 if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFREG
)) {
2355 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2356 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2357 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2358 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2360 xfs_fs_cmn_err(CE_ALERT
, mp
,
2361 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2362 item
, dip
, bp
, ino
);
2363 error
= EFSCORRUPTED
;
2366 } else if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFDIR
)) {
2367 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2368 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2369 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2370 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2371 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2373 xfs_fs_cmn_err(CE_ALERT
, mp
,
2374 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2375 item
, dip
, bp
, ino
);
2376 error
= EFSCORRUPTED
;
2380 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2381 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2382 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2384 xfs_fs_cmn_err(CE_ALERT
, mp
,
2385 "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",
2387 dicp
->di_nextents
+ dicp
->di_anextents
,
2389 error
= EFSCORRUPTED
;
2392 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2393 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2394 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2396 xfs_fs_cmn_err(CE_ALERT
, mp
,
2397 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2398 item
, dip
, bp
, ino
, dicp
->di_forkoff
);
2399 error
= EFSCORRUPTED
;
2402 if (unlikely(item
->ri_buf
[1].i_len
> sizeof(xfs_dinode_core_t
))) {
2403 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2404 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2406 xfs_fs_cmn_err(CE_ALERT
, mp
,
2407 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2408 item
->ri_buf
[1].i_len
, item
);
2409 error
= EFSCORRUPTED
;
2413 /* The core is in in-core format */
2414 xfs_dinode_to_disk(&dip
->di_core
,
2415 (xfs_icdinode_t
*)item
->ri_buf
[1].i_addr
);
2417 /* the rest is in on-disk format */
2418 if (item
->ri_buf
[1].i_len
> sizeof(xfs_dinode_core_t
)) {
2419 memcpy((xfs_caddr_t
) dip
+ sizeof(xfs_dinode_core_t
),
2420 item
->ri_buf
[1].i_addr
+ sizeof(xfs_dinode_core_t
),
2421 item
->ri_buf
[1].i_len
- sizeof(xfs_dinode_core_t
));
2424 fields
= in_f
->ilf_fields
;
2425 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
2427 dip
->di_u
.di_dev
= cpu_to_be32(in_f
->ilf_u
.ilfu_rdev
);
2430 dip
->di_u
.di_muuid
= in_f
->ilf_u
.ilfu_uuid
;
2434 if (in_f
->ilf_size
== 2)
2435 goto write_inode_buffer
;
2436 len
= item
->ri_buf
[2].i_len
;
2437 src
= item
->ri_buf
[2].i_addr
;
2438 ASSERT(in_f
->ilf_size
<= 4);
2439 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
2440 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
2441 (len
== in_f
->ilf_dsize
));
2443 switch (fields
& XFS_ILOG_DFORK
) {
2444 case XFS_ILOG_DDATA
:
2446 memcpy(&dip
->di_u
, src
, len
);
2449 case XFS_ILOG_DBROOT
:
2450 xfs_bmbt_to_bmdr((xfs_bmbt_block_t
*)src
, len
,
2451 &(dip
->di_u
.di_bmbt
),
2452 XFS_DFORK_DSIZE(dip
, mp
));
2457 * There are no data fork flags set.
2459 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
2464 * If we logged any attribute data, recover it. There may or
2465 * may not have been any other non-core data logged in this
2468 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2469 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
2474 len
= item
->ri_buf
[attr_index
].i_len
;
2475 src
= item
->ri_buf
[attr_index
].i_addr
;
2476 ASSERT(len
== in_f
->ilf_asize
);
2478 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2479 case XFS_ILOG_ADATA
:
2481 dest
= XFS_DFORK_APTR(dip
);
2482 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
2483 memcpy(dest
, src
, len
);
2486 case XFS_ILOG_ABROOT
:
2487 dest
= XFS_DFORK_APTR(dip
);
2488 xfs_bmbt_to_bmdr((xfs_bmbt_block_t
*)src
, len
,
2489 (xfs_bmdr_block_t
*)dest
,
2490 XFS_DFORK_ASIZE(dip
, mp
));
2494 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2503 if (ITEM_TYPE(item
) == XFS_LI_INODE
) {
2504 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
||
2505 XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*) == mp
);
2506 XFS_BUF_SET_FSPRIVATE(bp
, mp
);
2507 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2508 xfs_bdwrite(mp
, bp
);
2511 error
= xfs_bwrite(mp
, bp
);
2516 kmem_free(in_f
, sizeof(*in_f
));
2517 return XFS_ERROR(error
);
2521 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2522 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2526 xlog_recover_do_quotaoff_trans(
2528 xlog_recover_item_t
*item
,
2531 xfs_qoff_logformat_t
*qoff_f
;
2533 if (pass
== XLOG_RECOVER_PASS2
) {
2537 qoff_f
= (xfs_qoff_logformat_t
*)item
->ri_buf
[0].i_addr
;
2541 * The logitem format's flag tells us if this was user quotaoff,
2542 * group/project quotaoff or both.
2544 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2545 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2546 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
2547 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
2548 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2549 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2555 * Recover a dquot record
2558 xlog_recover_do_dquot_trans(
2560 xlog_recover_item_t
*item
,
2565 struct xfs_disk_dquot
*ddq
, *recddq
;
2567 xfs_dq_logformat_t
*dq_f
;
2570 if (pass
== XLOG_RECOVER_PASS1
) {
2576 * Filesystems are required to send in quota flags at mount time.
2578 if (mp
->m_qflags
== 0)
2581 recddq
= (xfs_disk_dquot_t
*)item
->ri_buf
[1].i_addr
;
2584 * This type of quotas was turned off, so ignore this record.
2586 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
2588 if (log
->l_quotaoffs_flag
& type
)
2592 * At this point we know that quota was _not_ turned off.
2593 * Since the mount flags are not indicating to us otherwise, this
2594 * must mean that quota is on, and the dquot needs to be replayed.
2595 * Remember that we may not have fully recovered the superblock yet,
2596 * so we can't do the usual trick of looking at the SB quota bits.
2598 * The other possibility, of course, is that the quota subsystem was
2599 * removed since the last mount - ENOSYS.
2601 dq_f
= (xfs_dq_logformat_t
*)item
->ri_buf
[0].i_addr
;
2603 if ((error
= xfs_qm_dqcheck(recddq
,
2605 0, XFS_QMOPT_DOWARN
,
2606 "xlog_recover_do_dquot_trans (log copy)"))) {
2607 return XFS_ERROR(EIO
);
2609 ASSERT(dq_f
->qlf_len
== 1);
2611 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
2613 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
),
2616 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp
,
2617 bp
, dq_f
->qlf_blkno
);
2621 ddq
= (xfs_disk_dquot_t
*)xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
2624 * At least the magic num portion should be on disk because this
2625 * was among a chunk of dquots created earlier, and we did some
2626 * minimal initialization then.
2628 if (xfs_qm_dqcheck(ddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2629 "xlog_recover_do_dquot_trans")) {
2631 return XFS_ERROR(EIO
);
2634 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
2636 ASSERT(dq_f
->qlf_size
== 2);
2637 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
||
2638 XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*) == mp
);
2639 XFS_BUF_SET_FSPRIVATE(bp
, mp
);
2640 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2641 xfs_bdwrite(mp
, bp
);
2647 * This routine is called to create an in-core extent free intent
2648 * item from the efi format structure which was logged on disk.
2649 * It allocates an in-core efi, copies the extents from the format
2650 * structure into it, and adds the efi to the AIL with the given
2654 xlog_recover_do_efi_trans(
2656 xlog_recover_item_t
*item
,
2662 xfs_efi_log_item_t
*efip
;
2663 xfs_efi_log_format_t
*efi_formatp
;
2665 if (pass
== XLOG_RECOVER_PASS1
) {
2669 efi_formatp
= (xfs_efi_log_format_t
*)item
->ri_buf
[0].i_addr
;
2672 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
2673 if ((error
= xfs_efi_copy_format(&(item
->ri_buf
[0]),
2674 &(efip
->efi_format
)))) {
2675 xfs_efi_item_free(efip
);
2678 efip
->efi_next_extent
= efi_formatp
->efi_nextents
;
2679 efip
->efi_flags
|= XFS_EFI_COMMITTED
;
2681 spin_lock(&mp
->m_ail_lock
);
2683 * xfs_trans_update_ail() drops the AIL lock.
2685 xfs_trans_update_ail(mp
, (xfs_log_item_t
*)efip
, lsn
);
2691 * This routine is called when an efd format structure is found in
2692 * a committed transaction in the log. It's purpose is to cancel
2693 * the corresponding efi if it was still in the log. To do this
2694 * it searches the AIL for the efi with an id equal to that in the
2695 * efd format structure. If we find it, we remove the efi from the
2699 xlog_recover_do_efd_trans(
2701 xlog_recover_item_t
*item
,
2705 xfs_efd_log_format_t
*efd_formatp
;
2706 xfs_efi_log_item_t
*efip
= NULL
;
2707 xfs_log_item_t
*lip
;
2711 if (pass
== XLOG_RECOVER_PASS1
) {
2715 efd_formatp
= (xfs_efd_log_format_t
*)item
->ri_buf
[0].i_addr
;
2716 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
2717 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
2718 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
2719 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
2720 efi_id
= efd_formatp
->efd_efi_id
;
2723 * Search for the efi with the id in the efd format structure
2727 spin_lock(&mp
->m_ail_lock
);
2728 lip
= xfs_trans_first_ail(mp
, &gen
);
2729 while (lip
!= NULL
) {
2730 if (lip
->li_type
== XFS_LI_EFI
) {
2731 efip
= (xfs_efi_log_item_t
*)lip
;
2732 if (efip
->efi_format
.efi_id
== efi_id
) {
2734 * xfs_trans_delete_ail() drops the
2737 xfs_trans_delete_ail(mp
, lip
);
2738 xfs_efi_item_free(efip
);
2742 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
2744 spin_unlock(&mp
->m_ail_lock
);
2748 * Perform the transaction
2750 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2751 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2754 xlog_recover_do_trans(
2756 xlog_recover_t
*trans
,
2760 xlog_recover_item_t
*item
, *first_item
;
2762 if ((error
= xlog_recover_reorder_trans(trans
)))
2764 first_item
= item
= trans
->r_itemq
;
2767 * we don't need to worry about the block number being
2768 * truncated in > 1 TB buffers because in user-land,
2769 * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so
2770 * the blknos will get through the user-mode buffer
2771 * cache properly. The only bad case is o32 kernels
2772 * where xfs_daddr_t is 32-bits but mount will warn us
2773 * off a > 1 TB filesystem before we get here.
2775 if ((ITEM_TYPE(item
) == XFS_LI_BUF
)) {
2776 if ((error
= xlog_recover_do_buffer_trans(log
, item
,
2779 } else if ((ITEM_TYPE(item
) == XFS_LI_INODE
)) {
2780 if ((error
= xlog_recover_do_inode_trans(log
, item
,
2783 } else if (ITEM_TYPE(item
) == XFS_LI_EFI
) {
2784 if ((error
= xlog_recover_do_efi_trans(log
, item
, trans
->r_lsn
,
2787 } else if (ITEM_TYPE(item
) == XFS_LI_EFD
) {
2788 xlog_recover_do_efd_trans(log
, item
, pass
);
2789 } else if (ITEM_TYPE(item
) == XFS_LI_DQUOT
) {
2790 if ((error
= xlog_recover_do_dquot_trans(log
, item
,
2793 } else if ((ITEM_TYPE(item
) == XFS_LI_QUOTAOFF
)) {
2794 if ((error
= xlog_recover_do_quotaoff_trans(log
, item
,
2798 xlog_warn("XFS: xlog_recover_do_trans");
2800 error
= XFS_ERROR(EIO
);
2803 item
= item
->ri_next
;
2804 } while (first_item
!= item
);
2810 * Free up any resources allocated by the transaction
2812 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2815 xlog_recover_free_trans(
2816 xlog_recover_t
*trans
)
2818 xlog_recover_item_t
*first_item
, *item
, *free_item
;
2821 item
= first_item
= trans
->r_itemq
;
2824 item
= item
->ri_next
;
2825 /* Free the regions in the item. */
2826 for (i
= 0; i
< free_item
->ri_cnt
; i
++) {
2827 kmem_free(free_item
->ri_buf
[i
].i_addr
,
2828 free_item
->ri_buf
[i
].i_len
);
2830 /* Free the item itself */
2831 kmem_free(free_item
->ri_buf
,
2832 (free_item
->ri_total
* sizeof(xfs_log_iovec_t
)));
2833 kmem_free(free_item
, sizeof(xlog_recover_item_t
));
2834 } while (first_item
!= item
);
2835 /* Free the transaction recover structure */
2836 kmem_free(trans
, sizeof(xlog_recover_t
));
2840 xlog_recover_commit_trans(
2843 xlog_recover_t
*trans
,
2848 if ((error
= xlog_recover_unlink_tid(q
, trans
)))
2850 if ((error
= xlog_recover_do_trans(log
, trans
, pass
)))
2852 xlog_recover_free_trans(trans
); /* no error */
2857 xlog_recover_unmount_trans(
2858 xlog_recover_t
*trans
)
2860 /* Do nothing now */
2861 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2866 * There are two valid states of the r_state field. 0 indicates that the
2867 * transaction structure is in a normal state. We have either seen the
2868 * start of the transaction or the last operation we added was not a partial
2869 * operation. If the last operation we added to the transaction was a
2870 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2872 * NOTE: skip LRs with 0 data length.
2875 xlog_recover_process_data(
2877 xlog_recover_t
*rhash
[],
2878 xlog_rec_header_t
*rhead
,
2884 xlog_op_header_t
*ohead
;
2885 xlog_recover_t
*trans
;
2891 lp
= dp
+ be32_to_cpu(rhead
->h_len
);
2892 num_logops
= be32_to_cpu(rhead
->h_num_logops
);
2894 /* check the log format matches our own - else we can't recover */
2895 if (xlog_header_check_recover(log
->l_mp
, rhead
))
2896 return (XFS_ERROR(EIO
));
2898 while ((dp
< lp
) && num_logops
) {
2899 ASSERT(dp
+ sizeof(xlog_op_header_t
) <= lp
);
2900 ohead
= (xlog_op_header_t
*)dp
;
2901 dp
+= sizeof(xlog_op_header_t
);
2902 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
2903 ohead
->oh_clientid
!= XFS_LOG
) {
2905 "XFS: xlog_recover_process_data: bad clientid");
2907 return (XFS_ERROR(EIO
));
2909 tid
= be32_to_cpu(ohead
->oh_tid
);
2910 hash
= XLOG_RHASH(tid
);
2911 trans
= xlog_recover_find_tid(rhash
[hash
], tid
);
2912 if (trans
== NULL
) { /* not found; add new tid */
2913 if (ohead
->oh_flags
& XLOG_START_TRANS
)
2914 xlog_recover_new_tid(&rhash
[hash
], tid
,
2915 be64_to_cpu(rhead
->h_lsn
));
2917 if (dp
+ be32_to_cpu(ohead
->oh_len
) > lp
) {
2919 "XFS: xlog_recover_process_data: bad length");
2921 return (XFS_ERROR(EIO
));
2923 flags
= ohead
->oh_flags
& ~XLOG_END_TRANS
;
2924 if (flags
& XLOG_WAS_CONT_TRANS
)
2925 flags
&= ~XLOG_CONTINUE_TRANS
;
2927 case XLOG_COMMIT_TRANS
:
2928 error
= xlog_recover_commit_trans(log
,
2929 &rhash
[hash
], trans
, pass
);
2931 case XLOG_UNMOUNT_TRANS
:
2932 error
= xlog_recover_unmount_trans(trans
);
2934 case XLOG_WAS_CONT_TRANS
:
2935 error
= xlog_recover_add_to_cont_trans(trans
,
2936 dp
, be32_to_cpu(ohead
->oh_len
));
2938 case XLOG_START_TRANS
:
2940 "XFS: xlog_recover_process_data: bad transaction");
2942 error
= XFS_ERROR(EIO
);
2945 case XLOG_CONTINUE_TRANS
:
2946 error
= xlog_recover_add_to_trans(trans
,
2947 dp
, be32_to_cpu(ohead
->oh_len
));
2951 "XFS: xlog_recover_process_data: bad flag");
2953 error
= XFS_ERROR(EIO
);
2959 dp
+= be32_to_cpu(ohead
->oh_len
);
2966 * Process an extent free intent item that was recovered from
2967 * the log. We need to free the extents that it describes.
2970 xlog_recover_process_efi(
2972 xfs_efi_log_item_t
*efip
)
2974 xfs_efd_log_item_t
*efdp
;
2979 xfs_fsblock_t startblock_fsb
;
2981 ASSERT(!(efip
->efi_flags
& XFS_EFI_RECOVERED
));
2984 * First check the validity of the extents described by the
2985 * EFI. If any are bad, then assume that all are bad and
2986 * just toss the EFI.
2988 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
2989 extp
= &(efip
->efi_format
.efi_extents
[i
]);
2990 startblock_fsb
= XFS_BB_TO_FSB(mp
,
2991 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
2992 if ((startblock_fsb
== 0) ||
2993 (extp
->ext_len
== 0) ||
2994 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
2995 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
2997 * This will pull the EFI from the AIL and
2998 * free the memory associated with it.
3000 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
3001 return XFS_ERROR(EIO
);
3005 tp
= xfs_trans_alloc(mp
, 0);
3006 error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0, 0, 0);
3009 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
3011 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3012 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3013 error
= xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
3016 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
3020 efip
->efi_flags
|= XFS_EFI_RECOVERED
;
3021 error
= xfs_trans_commit(tp
, 0);
3025 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3030 * Verify that once we've encountered something other than an EFI
3031 * in the AIL that there are no more EFIs in the AIL.
3035 xlog_recover_check_ail(
3037 xfs_log_item_t
*lip
,
3043 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
3044 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
3046 * The check will be bogus if we restart from the
3047 * beginning of the AIL, so ASSERT that we don't.
3048 * We never should since we're holding the AIL lock
3051 ASSERT(gen
== orig_gen
);
3052 } while (lip
!= NULL
);
3057 * When this is called, all of the EFIs which did not have
3058 * corresponding EFDs should be in the AIL. What we do now
3059 * is free the extents associated with each one.
3061 * Since we process the EFIs in normal transactions, they
3062 * will be removed at some point after the commit. This prevents
3063 * us from just walking down the list processing each one.
3064 * We'll use a flag in the EFI to skip those that we've already
3065 * processed and use the AIL iteration mechanism's generation
3066 * count to try to speed this up at least a bit.
3068 * When we start, we know that the EFIs are the only things in
3069 * the AIL. As we process them, however, other items are added
3070 * to the AIL. Since everything added to the AIL must come after
3071 * everything already in the AIL, we stop processing as soon as
3072 * we see something other than an EFI in the AIL.
3075 xlog_recover_process_efis(
3078 xfs_log_item_t
*lip
;
3079 xfs_efi_log_item_t
*efip
;
3085 spin_lock(&mp
->m_ail_lock
);
3087 lip
= xfs_trans_first_ail(mp
, &gen
);
3088 while (lip
!= NULL
) {
3090 * We're done when we see something other than an EFI.
3092 if (lip
->li_type
!= XFS_LI_EFI
) {
3093 xlog_recover_check_ail(mp
, lip
, gen
);
3098 * Skip EFIs that we've already processed.
3100 efip
= (xfs_efi_log_item_t
*)lip
;
3101 if (efip
->efi_flags
& XFS_EFI_RECOVERED
) {
3102 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
3106 spin_unlock(&mp
->m_ail_lock
);
3107 error
= xlog_recover_process_efi(mp
, efip
);
3110 spin_lock(&mp
->m_ail_lock
);
3111 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
3113 spin_unlock(&mp
->m_ail_lock
);
3118 * This routine performs a transaction to null out a bad inode pointer
3119 * in an agi unlinked inode hash bucket.
3122 xlog_recover_clear_agi_bucket(
3124 xfs_agnumber_t agno
,
3133 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3134 error
= xfs_trans_reserve(tp
, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp
), 0, 0, 0);
3136 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
3137 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
3138 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
3143 agi
= XFS_BUF_TO_AGI(agibp
);
3144 if (be32_to_cpu(agi
->agi_magicnum
) != XFS_AGI_MAGIC
)
3147 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
3148 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3149 (sizeof(xfs_agino_t
) * bucket
);
3150 xfs_trans_log_buf(tp
, agibp
, offset
,
3151 (offset
+ sizeof(xfs_agino_t
) - 1));
3153 error
= xfs_trans_commit(tp
, 0);
3159 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3161 xfs_fs_cmn_err(CE_WARN
, mp
, "xlog_recover_clear_agi_bucket: "
3162 "failed to clear agi %d. Continuing.", agno
);
3167 * xlog_iunlink_recover
3169 * This is called during recovery to process any inodes which
3170 * we unlinked but not freed when the system crashed. These
3171 * inodes will be on the lists in the AGI blocks. What we do
3172 * here is scan all the AGIs and fully truncate and free any
3173 * inodes found on the lists. Each inode is removed from the
3174 * lists when it has been fully truncated and is freed. The
3175 * freeing of the inode and its removal from the list must be
3179 xlog_recover_process_iunlinks(
3183 xfs_agnumber_t agno
;
3198 * Prevent any DMAPI event from being sent while in this function.
3200 mp_dmevmask
= mp
->m_dmevmask
;
3203 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3205 * Find the agi for this ag.
3207 agibp
= xfs_buf_read(mp
->m_ddev_targp
,
3208 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
3209 XFS_FSS_TO_BB(mp
, 1), 0);
3210 if (XFS_BUF_ISERROR(agibp
)) {
3211 xfs_ioerror_alert("xlog_recover_process_iunlinks(#1)",
3213 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)));
3215 agi
= XFS_BUF_TO_AGI(agibp
);
3216 ASSERT(XFS_AGI_MAGIC
== be32_to_cpu(agi
->agi_magicnum
));
3218 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3220 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
3221 while (agino
!= NULLAGINO
) {
3224 * Release the agi buffer so that it can
3225 * be acquired in the normal course of the
3226 * transaction to truncate and free the inode.
3228 xfs_buf_relse(agibp
);
3230 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3231 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
, 0);
3232 ASSERT(error
|| (ip
!= NULL
));
3236 * Get the on disk inode to find the
3237 * next inode in the bucket.
3239 error
= xfs_itobp(mp
, NULL
, ip
, &dip
,
3242 ASSERT(error
|| (dip
!= NULL
));
3246 ASSERT(ip
->i_d
.di_nlink
== 0);
3248 /* setup for the next pass */
3249 agino
= be32_to_cpu(
3250 dip
->di_next_unlinked
);
3253 * Prevent any DMAPI event from
3254 * being sent when the
3255 * reference on the inode is
3258 ip
->i_d
.di_dmevmask
= 0;
3261 * If this is a new inode, handle
3262 * it specially. Otherwise,
3263 * just drop our reference to the
3264 * inode. If there are no
3265 * other references, this will
3267 * xfs_inactive() which will
3268 * truncate the file and free
3271 if (ip
->i_d
.di_mode
== 0)
3272 xfs_iput_new(ip
, 0);
3277 * We can't read in the inode
3278 * this bucket points to, or
3279 * this inode is messed up. Just
3280 * ditch this bucket of inodes. We
3281 * will lose some inodes and space,
3282 * but at least we won't hang. Call
3283 * xlog_recover_clear_agi_bucket()
3284 * to perform a transaction to clear
3285 * the inode pointer in the bucket.
3287 xlog_recover_clear_agi_bucket(mp
, agno
,
3294 * Reacquire the agibuffer and continue around
3297 agibp
= xfs_buf_read(mp
->m_ddev_targp
,
3298 XFS_AG_DADDR(mp
, agno
,
3300 XFS_FSS_TO_BB(mp
, 1), 0);
3301 if (XFS_BUF_ISERROR(agibp
)) {
3303 "xlog_recover_process_iunlinks(#2)",
3305 XFS_AG_DADDR(mp
, agno
,
3306 XFS_AGI_DADDR(mp
)));
3308 agi
= XFS_BUF_TO_AGI(agibp
);
3309 ASSERT(XFS_AGI_MAGIC
== be32_to_cpu(
3310 agi
->agi_magicnum
));
3315 * Release the buffer for the current agi so we can
3316 * go on to the next one.
3318 xfs_buf_relse(agibp
);
3321 mp
->m_dmevmask
= mp_dmevmask
;
3327 xlog_pack_data_checksum(
3329 xlog_in_core_t
*iclog
,
3336 up
= (__be32
*)iclog
->ic_datap
;
3337 /* divide length by 4 to get # words */
3338 for (i
= 0; i
< (size
>> 2); i
++) {
3339 chksum
^= be32_to_cpu(*up
);
3342 iclog
->ic_header
.h_chksum
= cpu_to_be32(chksum
);
3345 #define xlog_pack_data_checksum(log, iclog, size)
3349 * Stamp cycle number in every block
3354 xlog_in_core_t
*iclog
,
3358 int size
= iclog
->ic_offset
+ roundoff
;
3361 xlog_in_core_2_t
*xhdr
;
3363 xlog_pack_data_checksum(log
, iclog
, size
);
3365 cycle_lsn
= CYCLE_LSN_DISK(iclog
->ic_header
.h_lsn
);
3367 dp
= iclog
->ic_datap
;
3368 for (i
= 0; i
< BTOBB(size
) &&
3369 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3370 iclog
->ic_header
.h_cycle_data
[i
] = *(__be32
*)dp
;
3371 *(__be32
*)dp
= cycle_lsn
;
3375 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3376 xhdr
= (xlog_in_core_2_t
*)&iclog
->ic_header
;
3377 for ( ; i
< BTOBB(size
); i
++) {
3378 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3379 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3380 xhdr
[j
].hic_xheader
.xh_cycle_data
[k
] = *(__be32
*)dp
;
3381 *(__be32
*)dp
= cycle_lsn
;
3385 for (i
= 1; i
< log
->l_iclog_heads
; i
++) {
3386 xhdr
[i
].hic_xheader
.xh_cycle
= cycle_lsn
;
3391 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3393 xlog_unpack_data_checksum(
3394 xlog_rec_header_t
*rhead
,
3398 __be32
*up
= (__be32
*)dp
;
3402 /* divide length by 4 to get # words */
3403 for (i
=0; i
< be32_to_cpu(rhead
->h_len
) >> 2; i
++) {
3404 chksum
^= be32_to_cpu(*up
);
3407 if (chksum
!= be32_to_cpu(rhead
->h_chksum
)) {
3408 if (rhead
->h_chksum
||
3409 ((log
->l_flags
& XLOG_CHKSUM_MISMATCH
) == 0)) {
3411 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
3412 be32_to_cpu(rhead
->h_chksum
), chksum
);
3414 "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3415 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3417 "XFS: LogR this is a LogV2 filesystem\n");
3419 log
->l_flags
|= XLOG_CHKSUM_MISMATCH
;
3424 #define xlog_unpack_data_checksum(rhead, dp, log)
3429 xlog_rec_header_t
*rhead
,
3434 xlog_in_core_2_t
*xhdr
;
3436 for (i
= 0; i
< BTOBB(be32_to_cpu(rhead
->h_len
)) &&
3437 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3438 *(__be32
*)dp
= *(__be32
*)&rhead
->h_cycle_data
[i
];
3442 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3443 xhdr
= (xlog_in_core_2_t
*)rhead
;
3444 for ( ; i
< BTOBB(be32_to_cpu(rhead
->h_len
)); i
++) {
3445 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3446 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3447 *(__be32
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
3452 xlog_unpack_data_checksum(rhead
, dp
, log
);
3456 xlog_valid_rec_header(
3458 xlog_rec_header_t
*rhead
,
3463 if (unlikely(be32_to_cpu(rhead
->h_magicno
) != XLOG_HEADER_MAGIC_NUM
)) {
3464 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3465 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3466 return XFS_ERROR(EFSCORRUPTED
);
3469 (!rhead
->h_version
||
3470 (be32_to_cpu(rhead
->h_version
) & (~XLOG_VERSION_OKBITS
))))) {
3471 xlog_warn("XFS: %s: unrecognised log version (%d).",
3472 __func__
, be32_to_cpu(rhead
->h_version
));
3473 return XFS_ERROR(EIO
);
3476 /* LR body must have data or it wouldn't have been written */
3477 hlen
= be32_to_cpu(rhead
->h_len
);
3478 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
3479 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3480 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3481 return XFS_ERROR(EFSCORRUPTED
);
3483 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
3484 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3485 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3486 return XFS_ERROR(EFSCORRUPTED
);
3492 * Read the log from tail to head and process the log records found.
3493 * Handle the two cases where the tail and head are in the same cycle
3494 * and where the active portion of the log wraps around the end of
3495 * the physical log separately. The pass parameter is passed through
3496 * to the routines called to process the data and is not looked at
3500 xlog_do_recovery_pass(
3502 xfs_daddr_t head_blk
,
3503 xfs_daddr_t tail_blk
,
3506 xlog_rec_header_t
*rhead
;
3508 xfs_caddr_t bufaddr
, offset
;
3509 xfs_buf_t
*hbp
, *dbp
;
3510 int error
= 0, h_size
;
3511 int bblks
, split_bblks
;
3512 int hblks
, split_hblks
, wrapped_hblks
;
3513 xlog_recover_t
*rhash
[XLOG_RHASH_SIZE
];
3515 ASSERT(head_blk
!= tail_blk
);
3518 * Read the header of the tail block and get the iclog buffer size from
3519 * h_size. Use this to tell how many sectors make up the log header.
3521 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3523 * When using variable length iclogs, read first sector of
3524 * iclog header and extract the header size from it. Get a
3525 * new hbp that is the correct size.
3527 hbp
= xlog_get_bp(log
, 1);
3530 if ((error
= xlog_bread(log
, tail_blk
, 1, hbp
)))
3532 offset
= xlog_align(log
, tail_blk
, 1, hbp
);
3533 rhead
= (xlog_rec_header_t
*)offset
;
3534 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
3537 h_size
= be32_to_cpu(rhead
->h_size
);
3538 if ((be32_to_cpu(rhead
->h_version
) & XLOG_VERSION_2
) &&
3539 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
3540 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
3541 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
3544 hbp
= xlog_get_bp(log
, hblks
);
3549 ASSERT(log
->l_sectbb_log
== 0);
3551 hbp
= xlog_get_bp(log
, 1);
3552 h_size
= XLOG_BIG_RECORD_BSIZE
;
3557 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
3563 memset(rhash
, 0, sizeof(rhash
));
3564 if (tail_blk
<= head_blk
) {
3565 for (blk_no
= tail_blk
; blk_no
< head_blk
; ) {
3566 if ((error
= xlog_bread(log
, blk_no
, hblks
, hbp
)))
3568 offset
= xlog_align(log
, blk_no
, hblks
, hbp
);
3569 rhead
= (xlog_rec_header_t
*)offset
;
3570 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3574 /* blocks in data section */
3575 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3576 error
= xlog_bread(log
, blk_no
+ hblks
, bblks
, dbp
);
3579 offset
= xlog_align(log
, blk_no
+ hblks
, bblks
, dbp
);
3580 xlog_unpack_data(rhead
, offset
, log
);
3581 if ((error
= xlog_recover_process_data(log
,
3582 rhash
, rhead
, offset
, pass
)))
3584 blk_no
+= bblks
+ hblks
;
3588 * Perform recovery around the end of the physical log.
3589 * When the head is not on the same cycle number as the tail,
3590 * we can't do a sequential recovery as above.
3593 while (blk_no
< log
->l_logBBsize
) {
3595 * Check for header wrapping around physical end-of-log
3600 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
3601 /* Read header in one read */
3602 error
= xlog_bread(log
, blk_no
, hblks
, hbp
);
3605 offset
= xlog_align(log
, blk_no
, hblks
, hbp
);
3607 /* This LR is split across physical log end */
3608 if (blk_no
!= log
->l_logBBsize
) {
3609 /* some data before physical log end */
3610 ASSERT(blk_no
<= INT_MAX
);
3611 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
3612 ASSERT(split_hblks
> 0);
3613 if ((error
= xlog_bread(log
, blk_no
,
3616 offset
= xlog_align(log
, blk_no
,
3620 * Note: this black magic still works with
3621 * large sector sizes (non-512) only because:
3622 * - we increased the buffer size originally
3623 * by 1 sector giving us enough extra space
3624 * for the second read;
3625 * - the log start is guaranteed to be sector
3627 * - we read the log end (LR header start)
3628 * _first_, then the log start (LR header end)
3629 * - order is important.
3631 bufaddr
= XFS_BUF_PTR(hbp
);
3632 XFS_BUF_SET_PTR(hbp
,
3633 bufaddr
+ BBTOB(split_hblks
),
3634 BBTOB(hblks
- split_hblks
));
3635 wrapped_hblks
= hblks
- split_hblks
;
3636 error
= xlog_bread(log
, 0, wrapped_hblks
, hbp
);
3639 XFS_BUF_SET_PTR(hbp
, bufaddr
, BBTOB(hblks
));
3641 offset
= xlog_align(log
, 0,
3642 wrapped_hblks
, hbp
);
3644 rhead
= (xlog_rec_header_t
*)offset
;
3645 error
= xlog_valid_rec_header(log
, rhead
,
3646 split_hblks
? blk_no
: 0);
3650 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3653 /* Read in data for log record */
3654 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
3655 error
= xlog_bread(log
, blk_no
, bblks
, dbp
);
3658 offset
= xlog_align(log
, blk_no
, bblks
, dbp
);
3660 /* This log record is split across the
3661 * physical end of log */
3664 if (blk_no
!= log
->l_logBBsize
) {
3665 /* some data is before the physical
3667 ASSERT(!wrapped_hblks
);
3668 ASSERT(blk_no
<= INT_MAX
);
3670 log
->l_logBBsize
- (int)blk_no
;
3671 ASSERT(split_bblks
> 0);
3672 if ((error
= xlog_bread(log
, blk_no
,
3675 offset
= xlog_align(log
, blk_no
,
3679 * Note: this black magic still works with
3680 * large sector sizes (non-512) only because:
3681 * - we increased the buffer size originally
3682 * by 1 sector giving us enough extra space
3683 * for the second read;
3684 * - the log start is guaranteed to be sector
3686 * - we read the log end (LR header start)
3687 * _first_, then the log start (LR header end)
3688 * - order is important.
3690 bufaddr
= XFS_BUF_PTR(dbp
);
3691 XFS_BUF_SET_PTR(dbp
,
3692 bufaddr
+ BBTOB(split_bblks
),
3693 BBTOB(bblks
- split_bblks
));
3694 if ((error
= xlog_bread(log
, wrapped_hblks
,
3695 bblks
- split_bblks
, dbp
)))
3697 XFS_BUF_SET_PTR(dbp
, bufaddr
, h_size
);
3699 offset
= xlog_align(log
, wrapped_hblks
,
3700 bblks
- split_bblks
, dbp
);
3702 xlog_unpack_data(rhead
, offset
, log
);
3703 if ((error
= xlog_recover_process_data(log
, rhash
,
3704 rhead
, offset
, pass
)))
3709 ASSERT(blk_no
>= log
->l_logBBsize
);
3710 blk_no
-= log
->l_logBBsize
;
3712 /* read first part of physical log */
3713 while (blk_no
< head_blk
) {
3714 if ((error
= xlog_bread(log
, blk_no
, hblks
, hbp
)))
3716 offset
= xlog_align(log
, blk_no
, hblks
, hbp
);
3717 rhead
= (xlog_rec_header_t
*)offset
;
3718 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3721 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3722 if ((error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
)))
3724 offset
= xlog_align(log
, blk_no
+hblks
, bblks
, dbp
);
3725 xlog_unpack_data(rhead
, offset
, log
);
3726 if ((error
= xlog_recover_process_data(log
, rhash
,
3727 rhead
, offset
, pass
)))
3729 blk_no
+= bblks
+ hblks
;
3741 * Do the recovery of the log. We actually do this in two phases.
3742 * The two passes are necessary in order to implement the function
3743 * of cancelling a record written into the log. The first pass
3744 * determines those things which have been cancelled, and the
3745 * second pass replays log items normally except for those which
3746 * have been cancelled. The handling of the replay and cancellations
3747 * takes place in the log item type specific routines.
3749 * The table of items which have cancel records in the log is allocated
3750 * and freed at this level, since only here do we know when all of
3751 * the log recovery has been completed.
3754 xlog_do_log_recovery(
3756 xfs_daddr_t head_blk
,
3757 xfs_daddr_t tail_blk
)
3761 ASSERT(head_blk
!= tail_blk
);
3764 * First do a pass to find all of the cancelled buf log items.
3765 * Store them in the buf_cancel_table for use in the second pass.
3767 log
->l_buf_cancel_table
=
3768 (xfs_buf_cancel_t
**)kmem_zalloc(XLOG_BC_TABLE_SIZE
*
3769 sizeof(xfs_buf_cancel_t
*),
3771 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3772 XLOG_RECOVER_PASS1
);
3774 kmem_free(log
->l_buf_cancel_table
,
3775 XLOG_BC_TABLE_SIZE
* sizeof(xfs_buf_cancel_t
*));
3776 log
->l_buf_cancel_table
= NULL
;
3780 * Then do a second pass to actually recover the items in the log.
3781 * When it is complete free the table of buf cancel items.
3783 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3784 XLOG_RECOVER_PASS2
);
3789 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
3790 ASSERT(log
->l_buf_cancel_table
[i
] == NULL
);
3794 kmem_free(log
->l_buf_cancel_table
,
3795 XLOG_BC_TABLE_SIZE
* sizeof(xfs_buf_cancel_t
*));
3796 log
->l_buf_cancel_table
= NULL
;
3802 * Do the actual recovery
3807 xfs_daddr_t head_blk
,
3808 xfs_daddr_t tail_blk
)
3815 * First replay the images in the log.
3817 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
3822 XFS_bflush(log
->l_mp
->m_ddev_targp
);
3825 * If IO errors happened during recovery, bail out.
3827 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
3832 * We now update the tail_lsn since much of the recovery has completed
3833 * and there may be space available to use. If there were no extent
3834 * or iunlinks, we can free up the entire log and set the tail_lsn to
3835 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3836 * lsn of the last known good LR on disk. If there are extent frees
3837 * or iunlinks they will have some entries in the AIL; so we look at
3838 * the AIL to determine how to set the tail_lsn.
3840 xlog_assign_tail_lsn(log
->l_mp
);
3843 * Now that we've finished replaying all buffer and inode
3844 * updates, re-read in the superblock.
3846 bp
= xfs_getsb(log
->l_mp
, 0);
3848 ASSERT(!(XFS_BUF_ISWRITE(bp
)));
3849 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp
)));
3851 XFS_BUF_UNASYNC(bp
);
3852 xfsbdstrat(log
->l_mp
, bp
);
3853 error
= xfs_iowait(bp
);
3855 xfs_ioerror_alert("xlog_do_recover",
3856 log
->l_mp
, bp
, XFS_BUF_ADDR(bp
));
3862 /* Convert superblock from on-disk format */
3863 sbp
= &log
->l_mp
->m_sb
;
3864 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
3865 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
3866 ASSERT(xfs_sb_good_version(sbp
));
3869 /* We've re-read the superblock so re-initialize per-cpu counters */
3870 xfs_icsb_reinit_counters(log
->l_mp
);
3872 xlog_recover_check_summary(log
);
3874 /* Normal transactions can now occur */
3875 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
3880 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3882 * Return error or zero.
3888 xfs_daddr_t head_blk
, tail_blk
;
3891 /* find the tail of the log */
3892 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
)))
3895 if (tail_blk
!= head_blk
) {
3896 /* There used to be a comment here:
3898 * disallow recovery on read-only mounts. note -- mount
3899 * checks for ENOSPC and turns it into an intelligent
3901 * ...but this is no longer true. Now, unless you specify
3902 * NORECOVERY (in which case this function would never be
3903 * called), we just go ahead and recover. We do this all
3904 * under the vfs layer, so we can get away with it unless
3905 * the device itself is read-only, in which case we fail.
3907 if ((error
= xfs_dev_is_read_only(log
->l_mp
, "recovery"))) {
3912 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3913 log
->l_mp
->m_fsname
, log
->l_mp
->m_logname
?
3914 log
->l_mp
->m_logname
: "internal");
3916 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
3917 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
3923 * In the first part of recovery we replay inodes and buffers and build
3924 * up the list of extent free items which need to be processed. Here
3925 * we process the extent free items and clean up the on disk unlinked
3926 * inode lists. This is separated from the first part of recovery so
3927 * that the root and real-time bitmap inodes can be read in from disk in
3928 * between the two stages. This is necessary so that we can free space
3929 * in the real-time portion of the file system.
3932 xlog_recover_finish(
3937 * Now we're ready to do the transactions needed for the
3938 * rest of recovery. Start with completing all the extent
3939 * free intent records and then process the unlinked inode
3940 * lists. At this point, we essentially run in normal mode
3941 * except that we're still performing recovery actions
3942 * rather than accepting new requests.
3944 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
3946 error
= xlog_recover_process_efis(log
);
3949 "Failed to recover EFIs on filesystem: %s",
3950 log
->l_mp
->m_fsname
);
3954 * Sync the log to get all the EFIs out of the AIL.
3955 * This isn't absolutely necessary, but it helps in
3956 * case the unlink transactions would have problems
3957 * pushing the EFIs out of the way.
3959 xfs_log_force(log
->l_mp
, (xfs_lsn_t
)0,
3960 (XFS_LOG_FORCE
| XFS_LOG_SYNC
));
3962 if ( (mfsi_flags
& XFS_MFSI_NOUNLINK
) == 0 ) {
3963 xlog_recover_process_iunlinks(log
);
3966 xlog_recover_check_summary(log
);
3969 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3970 log
->l_mp
->m_fsname
, log
->l_mp
->m_logname
?
3971 log
->l_mp
->m_logname
: "internal");
3972 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
3975 "!Ending clean XFS mount for filesystem: %s\n",
3976 log
->l_mp
->m_fsname
);
3984 * Read all of the agf and agi counters and check that they
3985 * are consistent with the superblock counters.
3988 xlog_recover_check_summary(
3996 xfs_daddr_t agfdaddr
;
3997 xfs_daddr_t agidaddr
;
3999 #ifdef XFS_LOUD_RECOVERY
4002 xfs_agnumber_t agno
;
4003 __uint64_t freeblks
;
4012 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
4013 agfdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGF_DADDR(mp
));
4014 agfbp
= xfs_buf_read(mp
->m_ddev_targp
, agfdaddr
,
4015 XFS_FSS_TO_BB(mp
, 1), 0);
4016 if (XFS_BUF_ISERROR(agfbp
)) {
4017 xfs_ioerror_alert("xlog_recover_check_summary(agf)",
4018 mp
, agfbp
, agfdaddr
);
4020 agfp
= XFS_BUF_TO_AGF(agfbp
);
4021 ASSERT(XFS_AGF_MAGIC
== be32_to_cpu(agfp
->agf_magicnum
));
4022 ASSERT(XFS_AGF_GOOD_VERSION(be32_to_cpu(agfp
->agf_versionnum
)));
4023 ASSERT(be32_to_cpu(agfp
->agf_seqno
) == agno
);
4025 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
4026 be32_to_cpu(agfp
->agf_flcount
);
4027 xfs_buf_relse(agfbp
);
4029 agidaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
4030 agibp
= xfs_buf_read(mp
->m_ddev_targp
, agidaddr
,
4031 XFS_FSS_TO_BB(mp
, 1), 0);
4032 if (XFS_BUF_ISERROR(agibp
)) {
4033 xfs_ioerror_alert("xlog_recover_check_summary(agi)",
4034 mp
, agibp
, agidaddr
);
4036 agip
= XFS_BUF_TO_AGI(agibp
);
4037 ASSERT(XFS_AGI_MAGIC
== be32_to_cpu(agip
->agi_magicnum
));
4038 ASSERT(XFS_AGI_GOOD_VERSION(be32_to_cpu(agip
->agi_versionnum
)));
4039 ASSERT(be32_to_cpu(agip
->agi_seqno
) == agno
);
4041 itotal
+= be32_to_cpu(agip
->agi_count
);
4042 ifree
+= be32_to_cpu(agip
->agi_freecount
);
4043 xfs_buf_relse(agibp
);
4046 sbbp
= xfs_getsb(mp
, 0);
4047 #ifdef XFS_LOUD_RECOVERY
4049 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(sbbp
));
4051 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4052 sbp
->sb_icount
, itotal
);
4054 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4055 sbp
->sb_ifree
, ifree
);
4057 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4058 sbp
->sb_fdblocks
, freeblks
);
4061 * This is turned off until I account for the allocation
4062 * btree blocks which live in free space.
4064 ASSERT(sbp
->sb_icount
== itotal
);
4065 ASSERT(sbp
->sb_ifree
== ifree
);
4066 ASSERT(sbp
->sb_fdblocks
== freeblks
);
4069 xfs_buf_relse(sbbp
);